ref: 7186a2dd863028faa54d2e5fc7995693488af63c
parent: 0e3f494b217bde5e1d47107cdfbb044e4d801cec
author: Zoe Liu <[email protected]>
date: Wed Jul 22 06:40:42 EDT 2015
Code refactor on InterpKernel It in essence refactors the code for both the interpolation filtering and the convolution. This change includes the moving of all the files as well as the changing of the code from vp9_ prefix to vpx_ prefix accordingly, for underneath architectures: (1) x86; (2) arm/neon; and (3) mips/msa. The work on mips/drsp2 will be done in a separate change list. Change-Id: Ic3ce7fb7f81210db7628b373c73553db68793c46
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -14,6 +14,7 @@
#include "./vpx_config.h"
#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "test/acm_random.h"
#include "test/clear_system_state.h"
#include "test/register_state_check.h"
@@ -20,6 +21,8 @@
#include "test/util.h"
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_filter.h"
+#include "vpx_dsp/vpx_dsp_common.h"
+#include "vpx_dsp/vpx_filter.h"
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
@@ -945,7 +948,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_horiz_sse2(src, src_stride, dst, dst_stride, filter_x,
+ vpx_highbd_convolve8_horiz_sse2(src, src_stride, dst, dst_stride, filter_x,
filter_x_stride, filter_y, filter_y_stride,
w, h, 8);
}
@@ -957,7 +960,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -969,7 +972,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_vert_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_vert_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -981,7 +984,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -993,7 +996,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1005,7 +1008,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1017,7 +1020,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_horiz_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_horiz_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1029,7 +1032,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1041,7 +1044,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_vert_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_vert_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1053,7 +1056,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1065,7 +1068,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1077,7 +1080,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1089,7 +1092,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_horiz_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_horiz_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1101,7 +1104,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_horiz_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1113,7 +1116,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_vert_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_vert_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1125,7 +1128,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_vert_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1137,7 +1140,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1149,7 +1152,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_sse2(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_sse2(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1162,7 +1165,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve_copy_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve_copy_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1174,7 +1177,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve_avg_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve_avg_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1186,7 +1189,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_horiz_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1198,7 +1201,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1210,7 +1213,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_vert_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_vert_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1222,7 +1225,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1234,7 +1237,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1246,7 +1249,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 8);
}
@@ -1258,7 +1261,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve_copy_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve_copy_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1270,7 +1273,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve_avg_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve_avg_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1282,7 +1285,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_horiz_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1294,7 +1297,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1306,7 +1309,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_vert_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_vert_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1318,7 +1321,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1330,7 +1333,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1342,7 +1345,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 10);
}
@@ -1354,7 +1357,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve_copy_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve_copy_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1366,7 +1369,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve_avg_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve_avg_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1378,7 +1381,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_horiz_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1390,7 +1393,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1402,7 +1405,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_vert_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_vert_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1414,7 +1417,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1426,7 +1429,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1438,7 +1441,7 @@
const int16_t *filter_y,
int filter_y_stride,
int w, int h) {- vp9_highbd_convolve8_avg_c(src, src_stride, dst, dst_stride,
+ vpx_highbd_convolve8_avg_c(src, src_stride, dst, dst_stride,
filter_x, filter_x_stride,
filter_y, filter_y_stride, w, h, 12);
}
@@ -1504,10 +1507,10 @@
#else
const ConvolveFunctions convolve8_c(
- vp9_convolve_copy_c, vp9_convolve_avg_c,
- vp9_convolve8_horiz_c, vp9_convolve8_avg_horiz_c,
- vp9_convolve8_vert_c, vp9_convolve8_avg_vert_c,
- vp9_convolve8_c, vp9_convolve8_avg_c, 0);
+ vpx_convolve_copy_c, vpx_convolve_avg_c,
+ vpx_convolve8_horiz_c, vpx_convolve8_avg_horiz_c,
+ vpx_convolve8_vert_c, vpx_convolve8_avg_vert_c,
+ vpx_convolve8_c, vpx_convolve8_avg_c, 0);
INSTANTIATE_TEST_CASE_P(C, ConvolveTest, ::testing::Values(
make_tuple(4, 4, &convolve8_c),
@@ -1585,13 +1588,13 @@
#else
const ConvolveFunctions convolve8_sse2(
#if CONFIG_USE_X86INC
- vp9_convolve_copy_sse2, vp9_convolve_avg_sse2,
+ vpx_convolve_copy_sse2, vpx_convolve_avg_sse2,
#else
- vp9_convolve_copy_c, vp9_convolve_avg_c,
+ vpx_convolve_copy_c, vpx_convolve_avg_c,
#endif // CONFIG_USE_X86INC
- vp9_convolve8_horiz_sse2, vp9_convolve8_avg_horiz_sse2,
- vp9_convolve8_vert_sse2, vp9_convolve8_avg_vert_sse2,
- vp9_convolve8_sse2, vp9_convolve8_avg_sse2, 0);
+ vpx_convolve8_horiz_sse2, vpx_convolve8_avg_horiz_sse2,
+ vpx_convolve8_vert_sse2, vpx_convolve8_avg_vert_sse2,
+ vpx_convolve8_sse2, vpx_convolve8_avg_sse2, 0);
INSTANTIATE_TEST_CASE_P(SSE2, ConvolveTest, ::testing::Values(
make_tuple(4, 4, &convolve8_sse2),
@@ -1612,10 +1615,10 @@
#if HAVE_SSSE3
const ConvolveFunctions convolve8_ssse3(
- vp9_convolve_copy_c, vp9_convolve_avg_c,
- vp9_convolve8_horiz_ssse3, vp9_convolve8_avg_horiz_ssse3,
- vp9_convolve8_vert_ssse3, vp9_convolve8_avg_vert_ssse3,
- vp9_convolve8_ssse3, vp9_convolve8_avg_ssse3, 0);
+ vpx_convolve_copy_c, vpx_convolve_avg_c,
+ vpx_convolve8_horiz_ssse3, vpx_convolve8_avg_horiz_ssse3,
+ vpx_convolve8_vert_ssse3, vpx_convolve8_avg_vert_ssse3,
+ vpx_convolve8_ssse3, vpx_convolve8_avg_ssse3, 0);
INSTANTIATE_TEST_CASE_P(SSSE3, ConvolveTest, ::testing::Values(
make_tuple(4, 4, &convolve8_ssse3),
@@ -1635,10 +1638,10 @@
#if HAVE_AVX2 && HAVE_SSSE3
const ConvolveFunctions convolve8_avx2(
- vp9_convolve_copy_c, vp9_convolve_avg_c,
- vp9_convolve8_horiz_avx2, vp9_convolve8_avg_horiz_ssse3,
- vp9_convolve8_vert_avx2, vp9_convolve8_avg_vert_ssse3,
- vp9_convolve8_avx2, vp9_convolve8_avg_ssse3, 0);
+ vpx_convolve_copy_c, vpx_convolve_avg_c,
+ vpx_convolve8_horiz_avx2, vpx_convolve8_avg_horiz_ssse3,
+ vpx_convolve8_vert_avx2, vpx_convolve8_avg_vert_ssse3,
+ vpx_convolve8_avx2, vpx_convolve8_avg_ssse3, 0);
INSTANTIATE_TEST_CASE_P(AVX2, ConvolveTest, ::testing::Values(
make_tuple(4, 4, &convolve8_avx2),
@@ -1659,16 +1662,16 @@
#if HAVE_NEON
#if HAVE_NEON_ASM
const ConvolveFunctions convolve8_neon(
- vp9_convolve_copy_neon, vp9_convolve_avg_neon,
- vp9_convolve8_horiz_neon, vp9_convolve8_avg_horiz_neon,
- vp9_convolve8_vert_neon, vp9_convolve8_avg_vert_neon,
- vp9_convolve8_neon, vp9_convolve8_avg_neon, 0);
+ vpx_convolve_copy_neon, vpx_convolve_avg_neon,
+ vpx_convolve8_horiz_neon, vpx_convolve8_avg_horiz_neon,
+ vpx_convolve8_vert_neon, vpx_convolve8_avg_vert_neon,
+ vpx_convolve8_neon, vpx_convolve8_avg_neon, 0);
#else // HAVE_NEON
const ConvolveFunctions convolve8_neon(
- vp9_convolve_copy_neon, vp9_convolve_avg_neon,
- vp9_convolve8_horiz_neon, vp9_convolve8_avg_horiz_neon,
- vp9_convolve8_vert_neon, vp9_convolve8_avg_vert_neon,
- vp9_convolve8_neon, vp9_convolve8_avg_neon, 0);
+ vpx_convolve_copy_neon, vpx_convolve_avg_neon,
+ vpx_convolve8_horiz_neon, vpx_convolve8_avg_horiz_neon,
+ vpx_convolve8_vert_neon, vpx_convolve8_avg_vert_neon,
+ vpx_convolve8_neon, vpx_convolve8_avg_neon, 0);
#endif // HAVE_NEON_ASM
INSTANTIATE_TEST_CASE_P(NEON, ConvolveTest, ::testing::Values(
@@ -1689,10 +1692,10 @@
#if HAVE_DSPR2
const ConvolveFunctions convolve8_dspr2(
- vp9_convolve_copy_dspr2, vp9_convolve_avg_dspr2,
- vp9_convolve8_horiz_dspr2, vp9_convolve8_avg_horiz_dspr2,
- vp9_convolve8_vert_dspr2, vp9_convolve8_avg_vert_dspr2,
- vp9_convolve8_dspr2, vp9_convolve8_avg_dspr2, 0);
+ vpx_convolve_copy_dspr2, vpx_convolve_avg_dspr2,
+ vpx_convolve8_horiz_dspr2, vpx_convolve8_avg_horiz_dspr2,
+ vpx_convolve8_vert_dspr2, vpx_convolve8_avg_vert_dspr2,
+ vpx_convolve8_dspr2, vpx_convolve8_avg_dspr2, 0);
INSTANTIATE_TEST_CASE_P(DSPR2, ConvolveTest, ::testing::Values(
make_tuple(4, 4, &convolve8_dspr2),
@@ -1712,10 +1715,10 @@
#if HAVE_MSA
const ConvolveFunctions convolve8_msa(
- vp9_convolve_copy_msa, vp9_convolve_avg_msa,
- vp9_convolve8_horiz_msa, vp9_convolve8_avg_horiz_msa,
- vp9_convolve8_vert_msa, vp9_convolve8_avg_vert_msa,
- vp9_convolve8_msa, vp9_convolve8_avg_msa, 0);
+ vpx_convolve_copy_msa, vpx_convolve_avg_msa,
+ vpx_convolve8_horiz_msa, vpx_convolve8_avg_horiz_msa,
+ vpx_convolve8_vert_msa, vpx_convolve8_avg_vert_msa,
+ vpx_convolve8_msa, vpx_convolve8_avg_msa, 0);
INSTANTIATE_TEST_CASE_P(MSA, ConvolveTest, ::testing::Values(
make_tuple(4, 4, &convolve8_msa),
--- a/vp9/common/arm/neon/vp9_convolve8_avg_neon.c
+++ /dev/null
@@ -1,390 +1,0 @@
-/*
- * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <stddef.h>
-#include <arm_neon.h>
-
-#include "./vpx_config.h"
-#include "vpx_ports/mem.h"
-
-void vp9_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h);
-void vp9_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h);
-
-static INLINE int32x4_t MULTIPLY_BY_Q0(
- int16x4_t dsrc0,
- int16x4_t dsrc1,
- int16x4_t dsrc2,
- int16x4_t dsrc3,
- int16x4_t dsrc4,
- int16x4_t dsrc5,
- int16x4_t dsrc6,
- int16x4_t dsrc7,
- int16x8_t q0s16) {- int32x4_t qdst;
- int16x4_t d0s16, d1s16;
-
- d0s16 = vget_low_s16(q0s16);
- d1s16 = vget_high_s16(q0s16);
-
- qdst = vmull_lane_s16(dsrc0, d0s16, 0);
- qdst = vmlal_lane_s16(qdst, dsrc1, d0s16, 1);
- qdst = vmlal_lane_s16(qdst, dsrc2, d0s16, 2);
- qdst = vmlal_lane_s16(qdst, dsrc3, d0s16, 3);
- qdst = vmlal_lane_s16(qdst, dsrc4, d1s16, 0);
- qdst = vmlal_lane_s16(qdst, dsrc5, d1s16, 1);
- qdst = vmlal_lane_s16(qdst, dsrc6, d1s16, 2);
- qdst = vmlal_lane_s16(qdst, dsrc7, d1s16, 3);
- return qdst;
-}
-
-void vp9_convolve8_avg_horiz_neon(
- uint8_t *src,
- ptrdiff_t src_stride,
- uint8_t *dst,
- ptrdiff_t dst_stride,
- const int16_t *filter_x,
- int x_step_q4,
- const int16_t *filter_y, // unused
- int y_step_q4, // unused
- int w,
- int h) {- int width;
- uint8_t *s, *d;
- uint8x8_t d2u8, d3u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8;
- uint32x2_t d2u32, d3u32, d6u32, d7u32, d28u32, d29u32, d30u32, d31u32;
- uint8x16_t q1u8, q3u8, q12u8, q13u8, q14u8, q15u8;
- int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d22s16, d23s16;
- int16x4_t d24s16, d25s16, d26s16, d27s16;
- uint16x4_t d2u16, d3u16, d4u16, d5u16, d16u16, d17u16, d18u16, d19u16;
- int16x8_t q0s16;
- uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
- int32x4_t q1s32, q2s32, q14s32, q15s32;
- uint16x8x2_t q0x2u16;
- uint8x8x2_t d0x2u8, d1x2u8;
- uint32x2x2_t d0x2u32;
- uint16x4x2_t d0x2u16, d1x2u16;
- uint32x4x2_t q0x2u32;
-
- if (x_step_q4 != 16) {- vp9_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4, w, h);
- return;
- }
-
- q0s16 = vld1q_s16(filter_x);
-
- src -= 3; // adjust for taps
- for (; h > 0; h -= 4) { // loop_horiz_v- s = src;
- d24u8 = vld1_u8(s);
- s += src_stride;
- d25u8 = vld1_u8(s);
- s += src_stride;
- d26u8 = vld1_u8(s);
- s += src_stride;
- d27u8 = vld1_u8(s);
-
- q12u8 = vcombine_u8(d24u8, d25u8);
- q13u8 = vcombine_u8(d26u8, d27u8);
-
- q0x2u16 = vtrnq_u16(vreinterpretq_u16_u8(q12u8),
- vreinterpretq_u16_u8(q13u8));
- d24u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[0]));
- d25u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[0]));
- d26u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[1]));
- d27u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[1]));
- d0x2u8 = vtrn_u8(d24u8, d25u8);
- d1x2u8 = vtrn_u8(d26u8, d27u8);
-
- __builtin_prefetch(src + src_stride * 4);
- __builtin_prefetch(src + src_stride * 5);
-
- q8u16 = vmovl_u8(d0x2u8.val[0]);
- q9u16 = vmovl_u8(d0x2u8.val[1]);
- q10u16 = vmovl_u8(d1x2u8.val[0]);
- q11u16 = vmovl_u8(d1x2u8.val[1]);
-
- src += 7;
- d16u16 = vget_low_u16(q8u16);
- d17u16 = vget_high_u16(q8u16);
- d18u16 = vget_low_u16(q9u16);
- d19u16 = vget_high_u16(q9u16);
- q8u16 = vcombine_u16(d16u16, d18u16); // vswp 17 18
- q9u16 = vcombine_u16(d17u16, d19u16);
-
- d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
- d23s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); // vmov 23 21
- for (width = w;
- width > 0;
- width -= 4, src += 4, dst += 4) { // loop_horiz- s = src;
- d28u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d29u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d31u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d30u32 = vld1_dup_u32((const uint32_t *)s);
-
- __builtin_prefetch(src + 64);
-
- d0x2u16 = vtrn_u16(vreinterpret_u16_u32(d28u32),
- vreinterpret_u16_u32(d31u32));
- d1x2u16 = vtrn_u16(vreinterpret_u16_u32(d29u32),
- vreinterpret_u16_u32(d30u32));
- d0x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[0]), // d28
- vreinterpret_u8_u16(d1x2u16.val[0])); // d29
- d1x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[1]), // d31
- vreinterpret_u8_u16(d1x2u16.val[1])); // d30
-
- __builtin_prefetch(src + 64 + src_stride);
-
- q14u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]);
- q15u8 = vcombine_u8(d1x2u8.val[1], d1x2u8.val[0]);
- q0x2u32 = vtrnq_u32(vreinterpretq_u32_u8(q14u8),
- vreinterpretq_u32_u8(q15u8));
-
- d28u8 = vreinterpret_u8_u32(vget_low_u32(q0x2u32.val[0]));
- d29u8 = vreinterpret_u8_u32(vget_high_u32(q0x2u32.val[0]));
- q12u16 = vmovl_u8(d28u8);
- q13u16 = vmovl_u8(d29u8);
-
- __builtin_prefetch(src + 64 + src_stride * 2);
-
- d = dst;
- d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 0);
- d += dst_stride;
- d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 0);
- d += dst_stride;
- d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 1);
- d += dst_stride;
- d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 1);
-
- d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
- d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
- d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
- d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
- d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
- d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
- d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
- d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
- d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
-
- q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d20s16, d22s16,
- d18s16, d19s16, d23s16, d24s16, q0s16);
- q2s32 = MULTIPLY_BY_Q0(d17s16, d20s16, d22s16, d18s16,
- d19s16, d23s16, d24s16, d26s16, q0s16);
- q14s32 = MULTIPLY_BY_Q0(d20s16, d22s16, d18s16, d19s16,
- d23s16, d24s16, d26s16, d27s16, q0s16);
- q15s32 = MULTIPLY_BY_Q0(d22s16, d18s16, d19s16, d23s16,
- d24s16, d26s16, d27s16, d25s16, q0s16);
-
- __builtin_prefetch(src + 64 + src_stride * 3);
-
- d2u16 = vqrshrun_n_s32(q1s32, 7);
- d3u16 = vqrshrun_n_s32(q2s32, 7);
- d4u16 = vqrshrun_n_s32(q14s32, 7);
- d5u16 = vqrshrun_n_s32(q15s32, 7);
-
- q1u16 = vcombine_u16(d2u16, d3u16);
- q2u16 = vcombine_u16(d4u16, d5u16);
-
- d2u8 = vqmovn_u16(q1u16);
- d3u8 = vqmovn_u16(q2u16);
-
- d0x2u16 = vtrn_u16(vreinterpret_u16_u8(d2u8),
- vreinterpret_u16_u8(d3u8));
- d0x2u32 = vtrn_u32(vreinterpret_u32_u16(d0x2u16.val[0]),
- vreinterpret_u32_u16(d0x2u16.val[1]));
- d0x2u8 = vtrn_u8(vreinterpret_u8_u32(d0x2u32.val[0]),
- vreinterpret_u8_u32(d0x2u32.val[1]));
-
- q1u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]);
- q3u8 = vreinterpretq_u8_u32(vcombine_u32(d6u32, d7u32));
-
- q1u8 = vrhaddq_u8(q1u8, q3u8);
-
- d2u32 = vreinterpret_u32_u8(vget_low_u8(q1u8));
- d3u32 = vreinterpret_u32_u8(vget_high_u8(q1u8));
-
- d = dst;
- vst1_lane_u32((uint32_t *)d, d2u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d2u32, 1);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 1);
-
- q8u16 = q9u16;
- d20s16 = d23s16;
- q11u16 = q12u16;
- q9u16 = q13u16;
- d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16));
- }
- src += src_stride * 4 - w - 7;
- dst += dst_stride * 4 - w;
- }
- return;
-}
-
-void vp9_convolve8_avg_vert_neon(
- uint8_t *src,
- ptrdiff_t src_stride,
- uint8_t *dst,
- ptrdiff_t dst_stride,
- const int16_t *filter_x, // unused
- int x_step_q4, // unused
- const int16_t *filter_y,
- int y_step_q4,
- int w,
- int h) {- int height;
- uint8_t *s, *d;
- uint8x8_t d2u8, d3u8;
- uint32x2_t d2u32, d3u32, d6u32, d7u32;
- uint32x2_t d16u32, d18u32, d20u32, d22u32, d24u32, d26u32;
- uint8x16_t q1u8, q3u8;
- int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16;
- int16x4_t d24s16, d25s16, d26s16, d27s16;
- uint16x4_t d2u16, d3u16, d4u16, d5u16;
- int16x8_t q0s16;
- uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
- int32x4_t q1s32, q2s32, q14s32, q15s32;
-
- if (y_step_q4 != 16) {- vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4, w, h);
- return;
- }
-
- src -= src_stride * 3;
- q0s16 = vld1q_s16(filter_y);
- for (; w > 0; w -= 4, src += 4, dst += 4) { // loop_vert_h- s = src;
- d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 0);
- s += src_stride;
- d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 1);
- s += src_stride;
- d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 0);
- s += src_stride;
- d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 1);
- s += src_stride;
- d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 0);
- s += src_stride;
- d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 1);
- s += src_stride;
- d22u32 = vld1_lane_u32((const uint32_t *)s, d22u32, 0);
- s += src_stride;
-
- q8u16 = vmovl_u8(vreinterpret_u8_u32(d16u32));
- q9u16 = vmovl_u8(vreinterpret_u8_u32(d18u32));
- q10u16 = vmovl_u8(vreinterpret_u8_u32(d20u32));
- q11u16 = vmovl_u8(vreinterpret_u8_u32(d22u32));
-
- d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
- d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
- d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
- d = dst;
- for (height = h; height > 0; height -= 4) { // loop_vert- d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 0);
- s += src_stride;
- d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 0);
- s += src_stride;
- d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 1);
- s += src_stride;
- d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 1);
- s += src_stride;
-
- q12u16 = vmovl_u8(vreinterpret_u8_u32(d24u32));
- q13u16 = vmovl_u8(vreinterpret_u8_u32(d26u32));
-
- d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 0);
- d += dst_stride;
- d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 1);
- d += dst_stride;
- d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 0);
- d += dst_stride;
- d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 1);
- d -= dst_stride * 3;
-
- d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
- d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
- d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
- d21s16 = vreinterpret_s16_u16(vget_high_u16(q10u16));
- d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
- d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
- d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
- d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
-
- __builtin_prefetch(s);
- __builtin_prefetch(s + src_stride);
- q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d18s16, d19s16,
- d20s16, d21s16, d22s16, d24s16, q0s16);
- __builtin_prefetch(s + src_stride * 2);
- __builtin_prefetch(s + src_stride * 3);
- q2s32 = MULTIPLY_BY_Q0(d17s16, d18s16, d19s16, d20s16,
- d21s16, d22s16, d24s16, d26s16, q0s16);
- __builtin_prefetch(d);
- __builtin_prefetch(d + dst_stride);
- q14s32 = MULTIPLY_BY_Q0(d18s16, d19s16, d20s16, d21s16,
- d22s16, d24s16, d26s16, d27s16, q0s16);
- __builtin_prefetch(d + dst_stride * 2);
- __builtin_prefetch(d + dst_stride * 3);
- q15s32 = MULTIPLY_BY_Q0(d19s16, d20s16, d21s16, d22s16,
- d24s16, d26s16, d27s16, d25s16, q0s16);
-
- d2u16 = vqrshrun_n_s32(q1s32, 7);
- d3u16 = vqrshrun_n_s32(q2s32, 7);
- d4u16 = vqrshrun_n_s32(q14s32, 7);
- d5u16 = vqrshrun_n_s32(q15s32, 7);
-
- q1u16 = vcombine_u16(d2u16, d3u16);
- q2u16 = vcombine_u16(d4u16, d5u16);
-
- d2u8 = vqmovn_u16(q1u16);
- d3u8 = vqmovn_u16(q2u16);
-
- q1u8 = vcombine_u8(d2u8, d3u8);
- q3u8 = vreinterpretq_u8_u32(vcombine_u32(d6u32, d7u32));
-
- q1u8 = vrhaddq_u8(q1u8, q3u8);
-
- d2u32 = vreinterpret_u32_u8(vget_low_u8(q1u8));
- d3u32 = vreinterpret_u32_u8(vget_high_u8(q1u8));
-
- vst1_lane_u32((uint32_t *)d, d2u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d2u32, 1);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 1);
- d += dst_stride;
-
- q8u16 = q10u16;
- d18s16 = d22s16;
- d19s16 = d24s16;
- q10u16 = q13u16;
- d22s16 = d25s16;
- }
- }
- return;
-}
--- a/vp9/common/arm/neon/vp9_convolve8_avg_neon_asm.asm
+++ /dev/null
@@ -1,302 +1,0 @@
-;
-; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
- ; These functions are only valid when:
- ; x_step_q4 == 16
- ; w%4 == 0
- ; h%4 == 0
- ; taps == 8
- ; VP9_FILTER_WEIGHT == 128
- ; VP9_FILTER_SHIFT == 7
-
- EXPORT |vp9_convolve8_avg_horiz_neon|
- EXPORT |vp9_convolve8_avg_vert_neon|
- IMPORT |vp9_convolve8_avg_horiz_c|
- IMPORT |vp9_convolve8_avg_vert_c|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
- ; Multiply and accumulate by q0
- MACRO
- MULTIPLY_BY_Q0 $dst, $src0, $src1, $src2, $src3, $src4, $src5, $src6, $src7
- vmull.s16 $dst, $src0, d0[0]
- vmlal.s16 $dst, $src1, d0[1]
- vmlal.s16 $dst, $src2, d0[2]
- vmlal.s16 $dst, $src3, d0[3]
- vmlal.s16 $dst, $src4, d1[0]
- vmlal.s16 $dst, $src5, d1[1]
- vmlal.s16 $dst, $src6, d1[2]
- vmlal.s16 $dst, $src7, d1[3]
- MEND
-
-; r0 const uint8_t *src
-; r1 int src_stride
-; r2 uint8_t *dst
-; r3 int dst_stride
-; sp[]const int16_t *filter_x
-; sp[]int x_step_q4
-; sp[]const int16_t *filter_y ; unused
-; sp[]int y_step_q4 ; unused
-; sp[]int w
-; sp[]int h
-
-|vp9_convolve8_avg_horiz_neon| PROC
- ldr r12, [sp, #4] ; x_step_q4
- cmp r12, #16
- bne vp9_convolve8_avg_horiz_c
-
- push {r4-r10, lr}-
- sub r0, r0, #3 ; adjust for taps
-
- ldr r5, [sp, #32] ; filter_x
- ldr r6, [sp, #48] ; w
- ldr r7, [sp, #52] ; h
-
- vld1.s16 {q0}, [r5] ; filter_x-
- sub r8, r1, r1, lsl #2 ; -src_stride * 3
- add r8, r8, #4 ; -src_stride * 3 + 4
-
- sub r4, r3, r3, lsl #2 ; -dst_stride * 3
- add r4, r4, #4 ; -dst_stride * 3 + 4
-
- rsb r9, r6, r1, lsl #2 ; reset src for outer loop
- sub r9, r9, #7
- rsb r12, r6, r3, lsl #2 ; reset dst for outer loop
-
- mov r10, r6 ; w loop counter
-
-vp9_convolve8_avg_loop_horiz_v
- vld1.8 {d24}, [r0], r1- vld1.8 {d25}, [r0], r1- vld1.8 {d26}, [r0], r1- vld1.8 {d27}, [r0], r8-
- vtrn.16 q12, q13
- vtrn.8 d24, d25
- vtrn.8 d26, d27
-
- pld [r0, r1, lsl #2]
-
- vmovl.u8 q8, d24
- vmovl.u8 q9, d25
- vmovl.u8 q10, d26
- vmovl.u8 q11, d27
-
- ; save a few instructions in the inner loop
- vswp d17, d18
- vmov d23, d21
-
- add r0, r0, #3
-
-vp9_convolve8_avg_loop_horiz
- add r5, r0, #64
-
- vld1.32 {d28[]}, [r0], r1- vld1.32 {d29[]}, [r0], r1- vld1.32 {d31[]}, [r0], r1- vld1.32 {d30[]}, [r0], r8-
- pld [r5]
-
- vtrn.16 d28, d31
- vtrn.16 d29, d30
- vtrn.8 d28, d29
- vtrn.8 d31, d30
-
- pld [r5, r1]
-
- ; extract to s16
- vtrn.32 q14, q15
- vmovl.u8 q12, d28
- vmovl.u8 q13, d29
-
- pld [r5, r1, lsl #1]
-
- ; slightly out of order load to match the existing data
- vld1.u32 {d6[0]}, [r2], r3- vld1.u32 {d7[0]}, [r2], r3- vld1.u32 {d6[1]}, [r2], r3- vld1.u32 {d7[1]}, [r2], r3-
- sub r2, r2, r3, lsl #2 ; reset for store
-
- ; src[] * filter_x
- MULTIPLY_BY_Q0 q1, d16, d17, d20, d22, d18, d19, d23, d24
- MULTIPLY_BY_Q0 q2, d17, d20, d22, d18, d19, d23, d24, d26
- MULTIPLY_BY_Q0 q14, d20, d22, d18, d19, d23, d24, d26, d27
- MULTIPLY_BY_Q0 q15, d22, d18, d19, d23, d24, d26, d27, d25
-
- pld [r5, -r8]
-
- ; += 64 >> 7
- vqrshrun.s32 d2, q1, #7
- vqrshrun.s32 d3, q2, #7
- vqrshrun.s32 d4, q14, #7
- vqrshrun.s32 d5, q15, #7
-
- ; saturate
- vqmovn.u16 d2, q1
- vqmovn.u16 d3, q2
-
- ; transpose
- vtrn.16 d2, d3
- vtrn.32 d2, d3
- vtrn.8 d2, d3
-
- ; average the new value and the dst value
- vrhadd.u8 q1, q1, q3
-
- vst1.u32 {d2[0]}, [r2@32], r3- vst1.u32 {d3[0]}, [r2@32], r3- vst1.u32 {d2[1]}, [r2@32], r3- vst1.u32 {d3[1]}, [r2@32], r4-
- vmov q8, q9
- vmov d20, d23
- vmov q11, q12
- vmov q9, q13
-
- subs r6, r6, #4 ; w -= 4
- bgt vp9_convolve8_avg_loop_horiz
-
- ; outer loop
- mov r6, r10 ; restore w counter
- add r0, r0, r9 ; src += src_stride * 4 - w
- add r2, r2, r12 ; dst += dst_stride * 4 - w
- subs r7, r7, #4 ; h -= 4
- bgt vp9_convolve8_avg_loop_horiz_v
-
- pop {r4-r10, pc}-
- ENDP
-
-|vp9_convolve8_avg_vert_neon| PROC
- ldr r12, [sp, #12]
- cmp r12, #16
- bne vp9_convolve8_avg_vert_c
-
- push {r4-r8, lr}-
- ; adjust for taps
- sub r0, r0, r1
- sub r0, r0, r1, lsl #1
-
- ldr r4, [sp, #32] ; filter_y
- ldr r6, [sp, #40] ; w
- ldr lr, [sp, #44] ; h
-
- vld1.s16 {q0}, [r4] ; filter_y-
- lsl r1, r1, #1
- lsl r3, r3, #1
-
-vp9_convolve8_avg_loop_vert_h
- mov r4, r0
- add r7, r0, r1, asr #1
- mov r5, r2
- add r8, r2, r3, asr #1
- mov r12, lr ; h loop counter
-
- vld1.u32 {d16[0]}, [r4], r1- vld1.u32 {d16[1]}, [r7], r1- vld1.u32 {d18[0]}, [r4], r1- vld1.u32 {d18[1]}, [r7], r1- vld1.u32 {d20[0]}, [r4], r1- vld1.u32 {d20[1]}, [r7], r1- vld1.u32 {d22[0]}, [r4], r1-
- vmovl.u8 q8, d16
- vmovl.u8 q9, d18
- vmovl.u8 q10, d20
- vmovl.u8 q11, d22
-
-vp9_convolve8_avg_loop_vert
- ; always process a 4x4 block at a time
- vld1.u32 {d24[0]}, [r7], r1- vld1.u32 {d26[0]}, [r4], r1- vld1.u32 {d26[1]}, [r7], r1- vld1.u32 {d24[1]}, [r4], r1-
- ; extract to s16
- vmovl.u8 q12, d24
- vmovl.u8 q13, d26
-
- vld1.u32 {d6[0]}, [r5@32], r3- vld1.u32 {d6[1]}, [r8@32], r3- vld1.u32 {d7[0]}, [r5@32], r3- vld1.u32 {d7[1]}, [r8@32], r3-
- pld [r7]
- pld [r4]
-
- ; src[] * filter_y
- MULTIPLY_BY_Q0 q1, d16, d17, d18, d19, d20, d21, d22, d24
-
- pld [r7, r1]
- pld [r4, r1]
-
- MULTIPLY_BY_Q0 q2, d17, d18, d19, d20, d21, d22, d24, d26
-
- pld [r5]
- pld [r8]
-
- MULTIPLY_BY_Q0 q14, d18, d19, d20, d21, d22, d24, d26, d27
-
- pld [r5, r3]
- pld [r8, r3]
-
- MULTIPLY_BY_Q0 q15, d19, d20, d21, d22, d24, d26, d27, d25
-
- ; += 64 >> 7
- vqrshrun.s32 d2, q1, #7
- vqrshrun.s32 d3, q2, #7
- vqrshrun.s32 d4, q14, #7
- vqrshrun.s32 d5, q15, #7
-
- ; saturate
- vqmovn.u16 d2, q1
- vqmovn.u16 d3, q2
-
- ; average the new value and the dst value
- vrhadd.u8 q1, q1, q3
-
- sub r5, r5, r3, lsl #1 ; reset for store
- sub r8, r8, r3, lsl #1
-
- vst1.u32 {d2[0]}, [r5@32], r3- vst1.u32 {d2[1]}, [r8@32], r3- vst1.u32 {d3[0]}, [r5@32], r3- vst1.u32 {d3[1]}, [r8@32], r3-
- vmov q8, q10
- vmov d18, d22
- vmov d19, d24
- vmov q10, q13
- vmov d22, d25
-
- subs r12, r12, #4 ; h -= 4
- bgt vp9_convolve8_avg_loop_vert
-
- ; outer loop
- add r0, r0, #4
- add r2, r2, #4
- subs r6, r6, #4 ; w -= 4
- bgt vp9_convolve8_avg_loop_vert_h
-
- pop {r4-r8, pc}-
- ENDP
- END
--- a/vp9/common/arm/neon/vp9_convolve8_neon.c
+++ /dev/null
@@ -1,357 +1,0 @@
-/*
- * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <stddef.h>
-#include <arm_neon.h>
-
-#include "./vpx_config.h"
-#include "vpx_ports/mem.h"
-
-void vp9_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h);
-void vp9_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h);
-
-static INLINE int32x4_t MULTIPLY_BY_Q0(
- int16x4_t dsrc0,
- int16x4_t dsrc1,
- int16x4_t dsrc2,
- int16x4_t dsrc3,
- int16x4_t dsrc4,
- int16x4_t dsrc5,
- int16x4_t dsrc6,
- int16x4_t dsrc7,
- int16x8_t q0s16) {- int32x4_t qdst;
- int16x4_t d0s16, d1s16;
-
- d0s16 = vget_low_s16(q0s16);
- d1s16 = vget_high_s16(q0s16);
-
- qdst = vmull_lane_s16(dsrc0, d0s16, 0);
- qdst = vmlal_lane_s16(qdst, dsrc1, d0s16, 1);
- qdst = vmlal_lane_s16(qdst, dsrc2, d0s16, 2);
- qdst = vmlal_lane_s16(qdst, dsrc3, d0s16, 3);
- qdst = vmlal_lane_s16(qdst, dsrc4, d1s16, 0);
- qdst = vmlal_lane_s16(qdst, dsrc5, d1s16, 1);
- qdst = vmlal_lane_s16(qdst, dsrc6, d1s16, 2);
- qdst = vmlal_lane_s16(qdst, dsrc7, d1s16, 3);
- return qdst;
-}
-
-void vp9_convolve8_horiz_neon(
- uint8_t *src,
- ptrdiff_t src_stride,
- uint8_t *dst,
- ptrdiff_t dst_stride,
- const int16_t *filter_x,
- int x_step_q4,
- const int16_t *filter_y, // unused
- int y_step_q4, // unused
- int w,
- int h) {- int width;
- uint8_t *s, *d, *psrc, *pdst;
- uint8x8_t d2u8, d3u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8;
- uint32x2_t d2u32, d3u32, d28u32, d29u32, d30u32, d31u32;
- uint8x16_t q12u8, q13u8, q14u8, q15u8;
- int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d22s16, d23s16;
- int16x4_t d24s16, d25s16, d26s16, d27s16;
- uint16x4_t d2u16, d3u16, d4u16, d5u16, d16u16, d17u16, d18u16, d19u16;
- int16x8_t q0s16;
- uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
- int32x4_t q1s32, q2s32, q14s32, q15s32;
- uint16x8x2_t q0x2u16;
- uint8x8x2_t d0x2u8, d1x2u8;
- uint32x2x2_t d0x2u32;
- uint16x4x2_t d0x2u16, d1x2u16;
- uint32x4x2_t q0x2u32;
-
- if (x_step_q4 != 16) {- vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4, w, h);
- return;
- }
-
- q0s16 = vld1q_s16(filter_x);
-
- src -= 3; // adjust for taps
- for (; h > 0; h -= 4,
- src += src_stride * 4,
- dst += dst_stride * 4) { // loop_horiz_v- s = src;
- d24u8 = vld1_u8(s);
- s += src_stride;
- d25u8 = vld1_u8(s);
- s += src_stride;
- d26u8 = vld1_u8(s);
- s += src_stride;
- d27u8 = vld1_u8(s);
-
- q12u8 = vcombine_u8(d24u8, d25u8);
- q13u8 = vcombine_u8(d26u8, d27u8);
-
- q0x2u16 = vtrnq_u16(vreinterpretq_u16_u8(q12u8),
- vreinterpretq_u16_u8(q13u8));
- d24u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[0]));
- d25u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[0]));
- d26u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[1]));
- d27u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[1]));
- d0x2u8 = vtrn_u8(d24u8, d25u8);
- d1x2u8 = vtrn_u8(d26u8, d27u8);
-
- __builtin_prefetch(src + src_stride * 4);
- __builtin_prefetch(src + src_stride * 5);
- __builtin_prefetch(src + src_stride * 6);
-
- q8u16 = vmovl_u8(d0x2u8.val[0]);
- q9u16 = vmovl_u8(d0x2u8.val[1]);
- q10u16 = vmovl_u8(d1x2u8.val[0]);
- q11u16 = vmovl_u8(d1x2u8.val[1]);
-
- d16u16 = vget_low_u16(q8u16);
- d17u16 = vget_high_u16(q8u16);
- d18u16 = vget_low_u16(q9u16);
- d19u16 = vget_high_u16(q9u16);
- q8u16 = vcombine_u16(d16u16, d18u16); // vswp 17 18
- q9u16 = vcombine_u16(d17u16, d19u16);
-
- d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
- d23s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); // vmov 23 21
- for (width = w, psrc = src + 7, pdst = dst;
- width > 0;
- width -= 4, psrc += 4, pdst += 4) { // loop_horiz- s = psrc;
- d28u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d29u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d31u32 = vld1_dup_u32((const uint32_t *)s);
- s += src_stride;
- d30u32 = vld1_dup_u32((const uint32_t *)s);
-
- __builtin_prefetch(psrc + 64);
-
- d0x2u16 = vtrn_u16(vreinterpret_u16_u32(d28u32),
- vreinterpret_u16_u32(d31u32));
- d1x2u16 = vtrn_u16(vreinterpret_u16_u32(d29u32),
- vreinterpret_u16_u32(d30u32));
- d0x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[0]), // d28
- vreinterpret_u8_u16(d1x2u16.val[0])); // d29
- d1x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[1]), // d31
- vreinterpret_u8_u16(d1x2u16.val[1])); // d30
-
- __builtin_prefetch(psrc + 64 + src_stride);
-
- q14u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]);
- q15u8 = vcombine_u8(d1x2u8.val[1], d1x2u8.val[0]);
- q0x2u32 = vtrnq_u32(vreinterpretq_u32_u8(q14u8),
- vreinterpretq_u32_u8(q15u8));
-
- d28u8 = vreinterpret_u8_u32(vget_low_u32(q0x2u32.val[0]));
- d29u8 = vreinterpret_u8_u32(vget_high_u32(q0x2u32.val[0]));
- q12u16 = vmovl_u8(d28u8);
- q13u16 = vmovl_u8(d29u8);
-
- __builtin_prefetch(psrc + 64 + src_stride * 2);
-
- d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
- d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
- d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
- d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
- d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
- d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
- d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
- d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
- d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
-
- q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d20s16, d22s16,
- d18s16, d19s16, d23s16, d24s16, q0s16);
- q2s32 = MULTIPLY_BY_Q0(d17s16, d20s16, d22s16, d18s16,
- d19s16, d23s16, d24s16, d26s16, q0s16);
- q14s32 = MULTIPLY_BY_Q0(d20s16, d22s16, d18s16, d19s16,
- d23s16, d24s16, d26s16, d27s16, q0s16);
- q15s32 = MULTIPLY_BY_Q0(d22s16, d18s16, d19s16, d23s16,
- d24s16, d26s16, d27s16, d25s16, q0s16);
-
- __builtin_prefetch(psrc + 60 + src_stride * 3);
-
- d2u16 = vqrshrun_n_s32(q1s32, 7);
- d3u16 = vqrshrun_n_s32(q2s32, 7);
- d4u16 = vqrshrun_n_s32(q14s32, 7);
- d5u16 = vqrshrun_n_s32(q15s32, 7);
-
- q1u16 = vcombine_u16(d2u16, d3u16);
- q2u16 = vcombine_u16(d4u16, d5u16);
-
- d2u8 = vqmovn_u16(q1u16);
- d3u8 = vqmovn_u16(q2u16);
-
- d0x2u16 = vtrn_u16(vreinterpret_u16_u8(d2u8),
- vreinterpret_u16_u8(d3u8));
- d0x2u32 = vtrn_u32(vreinterpret_u32_u16(d0x2u16.val[0]),
- vreinterpret_u32_u16(d0x2u16.val[1]));
- d0x2u8 = vtrn_u8(vreinterpret_u8_u32(d0x2u32.val[0]),
- vreinterpret_u8_u32(d0x2u32.val[1]));
-
- d2u32 = vreinterpret_u32_u8(d0x2u8.val[0]);
- d3u32 = vreinterpret_u32_u8(d0x2u8.val[1]);
-
- d = pdst;
- vst1_lane_u32((uint32_t *)d, d2u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d2u32, 1);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 1);
-
- q8u16 = q9u16;
- d20s16 = d23s16;
- q11u16 = q12u16;
- q9u16 = q13u16;
- d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16));
- }
- }
- return;
-}
-
-void vp9_convolve8_vert_neon(
- uint8_t *src,
- ptrdiff_t src_stride,
- uint8_t *dst,
- ptrdiff_t dst_stride,
- const int16_t *filter_x, // unused
- int x_step_q4, // unused
- const int16_t *filter_y,
- int y_step_q4,
- int w,
- int h) {- int height;
- uint8_t *s, *d;
- uint32x2_t d2u32, d3u32;
- uint32x2_t d16u32, d18u32, d20u32, d22u32, d24u32, d26u32;
- int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16;
- int16x4_t d24s16, d25s16, d26s16, d27s16;
- uint16x4_t d2u16, d3u16, d4u16, d5u16;
- int16x8_t q0s16;
- uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
- int32x4_t q1s32, q2s32, q14s32, q15s32;
-
- if (y_step_q4 != 16) {- vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4, w, h);
- return;
- }
-
- src -= src_stride * 3;
- q0s16 = vld1q_s16(filter_y);
- for (; w > 0; w -= 4, src += 4, dst += 4) { // loop_vert_h- s = src;
- d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 0);
- s += src_stride;
- d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 1);
- s += src_stride;
- d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 0);
- s += src_stride;
- d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 1);
- s += src_stride;
- d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 0);
- s += src_stride;
- d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 1);
- s += src_stride;
- d22u32 = vld1_lane_u32((const uint32_t *)s, d22u32, 0);
- s += src_stride;
-
- q8u16 = vmovl_u8(vreinterpret_u8_u32(d16u32));
- q9u16 = vmovl_u8(vreinterpret_u8_u32(d18u32));
- q10u16 = vmovl_u8(vreinterpret_u8_u32(d20u32));
- q11u16 = vmovl_u8(vreinterpret_u8_u32(d22u32));
-
- d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
- d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
- d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
- d = dst;
- for (height = h; height > 0; height -= 4) { // loop_vert- d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 0);
- s += src_stride;
- d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 0);
- s += src_stride;
- d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 1);
- s += src_stride;
- d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 1);
- s += src_stride;
-
- q12u16 = vmovl_u8(vreinterpret_u8_u32(d24u32));
- q13u16 = vmovl_u8(vreinterpret_u8_u32(d26u32));
-
- d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
- d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
- d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
- d21s16 = vreinterpret_s16_u16(vget_high_u16(q10u16));
- d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
- d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
- d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
- d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
-
- __builtin_prefetch(d);
- __builtin_prefetch(d + dst_stride);
- q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d18s16, d19s16,
- d20s16, d21s16, d22s16, d24s16, q0s16);
- __builtin_prefetch(d + dst_stride * 2);
- __builtin_prefetch(d + dst_stride * 3);
- q2s32 = MULTIPLY_BY_Q0(d17s16, d18s16, d19s16, d20s16,
- d21s16, d22s16, d24s16, d26s16, q0s16);
- __builtin_prefetch(s);
- __builtin_prefetch(s + src_stride);
- q14s32 = MULTIPLY_BY_Q0(d18s16, d19s16, d20s16, d21s16,
- d22s16, d24s16, d26s16, d27s16, q0s16);
- __builtin_prefetch(s + src_stride * 2);
- __builtin_prefetch(s + src_stride * 3);
- q15s32 = MULTIPLY_BY_Q0(d19s16, d20s16, d21s16, d22s16,
- d24s16, d26s16, d27s16, d25s16, q0s16);
-
- d2u16 = vqrshrun_n_s32(q1s32, 7);
- d3u16 = vqrshrun_n_s32(q2s32, 7);
- d4u16 = vqrshrun_n_s32(q14s32, 7);
- d5u16 = vqrshrun_n_s32(q15s32, 7);
-
- q1u16 = vcombine_u16(d2u16, d3u16);
- q2u16 = vcombine_u16(d4u16, d5u16);
-
- d2u32 = vreinterpret_u32_u8(vqmovn_u16(q1u16));
- d3u32 = vreinterpret_u32_u8(vqmovn_u16(q2u16));
-
- vst1_lane_u32((uint32_t *)d, d2u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d2u32, 1);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 0);
- d += dst_stride;
- vst1_lane_u32((uint32_t *)d, d3u32, 1);
- d += dst_stride;
-
- q8u16 = q10u16;
- d18s16 = d22s16;
- d19s16 = d24s16;
- q10u16 = q13u16;
- d22s16 = d25s16;
- }
- }
- return;
-}
--- a/vp9/common/arm/neon/vp9_convolve8_neon_asm.asm
+++ /dev/null
@@ -1,280 +1,0 @@
-;
-; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
- ; These functions are only valid when:
- ; x_step_q4 == 16
- ; w%4 == 0
- ; h%4 == 0
- ; taps == 8
- ; VP9_FILTER_WEIGHT == 128
- ; VP9_FILTER_SHIFT == 7
-
- EXPORT |vp9_convolve8_horiz_neon|
- EXPORT |vp9_convolve8_vert_neon|
- IMPORT |vp9_convolve8_horiz_c|
- IMPORT |vp9_convolve8_vert_c|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
- ; Multiply and accumulate by q0
- MACRO
- MULTIPLY_BY_Q0 $dst, $src0, $src1, $src2, $src3, $src4, $src5, $src6, $src7
- vmull.s16 $dst, $src0, d0[0]
- vmlal.s16 $dst, $src1, d0[1]
- vmlal.s16 $dst, $src2, d0[2]
- vmlal.s16 $dst, $src3, d0[3]
- vmlal.s16 $dst, $src4, d1[0]
- vmlal.s16 $dst, $src5, d1[1]
- vmlal.s16 $dst, $src6, d1[2]
- vmlal.s16 $dst, $src7, d1[3]
- MEND
-
-; r0 const uint8_t *src
-; r1 int src_stride
-; r2 uint8_t *dst
-; r3 int dst_stride
-; sp[]const int16_t *filter_x
-; sp[]int x_step_q4
-; sp[]const int16_t *filter_y ; unused
-; sp[]int y_step_q4 ; unused
-; sp[]int w
-; sp[]int h
-
-|vp9_convolve8_horiz_neon| PROC
- ldr r12, [sp, #4] ; x_step_q4
- cmp r12, #16
- bne vp9_convolve8_horiz_c
-
- push {r4-r10, lr}-
- sub r0, r0, #3 ; adjust for taps
-
- ldr r5, [sp, #32] ; filter_x
- ldr r6, [sp, #48] ; w
- ldr r7, [sp, #52] ; h
-
- vld1.s16 {q0}, [r5] ; filter_x-
- sub r8, r1, r1, lsl #2 ; -src_stride * 3
- add r8, r8, #4 ; -src_stride * 3 + 4
-
- sub r4, r3, r3, lsl #2 ; -dst_stride * 3
- add r4, r4, #4 ; -dst_stride * 3 + 4
-
- rsb r9, r6, r1, lsl #2 ; reset src for outer loop
- sub r9, r9, #7
- rsb r12, r6, r3, lsl #2 ; reset dst for outer loop
-
- mov r10, r6 ; w loop counter
-
-vp9_convolve8_loop_horiz_v
- vld1.8 {d24}, [r0], r1- vld1.8 {d25}, [r0], r1- vld1.8 {d26}, [r0], r1- vld1.8 {d27}, [r0], r8-
- vtrn.16 q12, q13
- vtrn.8 d24, d25
- vtrn.8 d26, d27
-
- pld [r0, r1, lsl #2]
-
- vmovl.u8 q8, d24
- vmovl.u8 q9, d25
- vmovl.u8 q10, d26
- vmovl.u8 q11, d27
-
- ; save a few instructions in the inner loop
- vswp d17, d18
- vmov d23, d21
-
- add r0, r0, #3
-
-vp9_convolve8_loop_horiz
- add r5, r0, #64
-
- vld1.32 {d28[]}, [r0], r1- vld1.32 {d29[]}, [r0], r1- vld1.32 {d31[]}, [r0], r1- vld1.32 {d30[]}, [r0], r8-
- pld [r5]
-
- vtrn.16 d28, d31
- vtrn.16 d29, d30
- vtrn.8 d28, d29
- vtrn.8 d31, d30
-
- pld [r5, r1]
-
- ; extract to s16
- vtrn.32 q14, q15
- vmovl.u8 q12, d28
- vmovl.u8 q13, d29
-
- pld [r5, r1, lsl #1]
-
- ; src[] * filter_x
- MULTIPLY_BY_Q0 q1, d16, d17, d20, d22, d18, d19, d23, d24
- MULTIPLY_BY_Q0 q2, d17, d20, d22, d18, d19, d23, d24, d26
- MULTIPLY_BY_Q0 q14, d20, d22, d18, d19, d23, d24, d26, d27
- MULTIPLY_BY_Q0 q15, d22, d18, d19, d23, d24, d26, d27, d25
-
- pld [r5, -r8]
-
- ; += 64 >> 7
- vqrshrun.s32 d2, q1, #7
- vqrshrun.s32 d3, q2, #7
- vqrshrun.s32 d4, q14, #7
- vqrshrun.s32 d5, q15, #7
-
- ; saturate
- vqmovn.u16 d2, q1
- vqmovn.u16 d3, q2
-
- ; transpose
- vtrn.16 d2, d3
- vtrn.32 d2, d3
- vtrn.8 d2, d3
-
- vst1.u32 {d2[0]}, [r2@32], r3- vst1.u32 {d3[0]}, [r2@32], r3- vst1.u32 {d2[1]}, [r2@32], r3- vst1.u32 {d3[1]}, [r2@32], r4-
- vmov q8, q9
- vmov d20, d23
- vmov q11, q12
- vmov q9, q13
-
- subs r6, r6, #4 ; w -= 4
- bgt vp9_convolve8_loop_horiz
-
- ; outer loop
- mov r6, r10 ; restore w counter
- add r0, r0, r9 ; src += src_stride * 4 - w
- add r2, r2, r12 ; dst += dst_stride * 4 - w
- subs r7, r7, #4 ; h -= 4
- bgt vp9_convolve8_loop_horiz_v
-
- pop {r4-r10, pc}-
- ENDP
-
-|vp9_convolve8_vert_neon| PROC
- ldr r12, [sp, #12]
- cmp r12, #16
- bne vp9_convolve8_vert_c
-
- push {r4-r8, lr}-
- ; adjust for taps
- sub r0, r0, r1
- sub r0, r0, r1, lsl #1
-
- ldr r4, [sp, #32] ; filter_y
- ldr r6, [sp, #40] ; w
- ldr lr, [sp, #44] ; h
-
- vld1.s16 {q0}, [r4] ; filter_y-
- lsl r1, r1, #1
- lsl r3, r3, #1
-
-vp9_convolve8_loop_vert_h
- mov r4, r0
- add r7, r0, r1, asr #1
- mov r5, r2
- add r8, r2, r3, asr #1
- mov r12, lr ; h loop counter
-
- vld1.u32 {d16[0]}, [r4], r1- vld1.u32 {d16[1]}, [r7], r1- vld1.u32 {d18[0]}, [r4], r1- vld1.u32 {d18[1]}, [r7], r1- vld1.u32 {d20[0]}, [r4], r1- vld1.u32 {d20[1]}, [r7], r1- vld1.u32 {d22[0]}, [r4], r1-
- vmovl.u8 q8, d16
- vmovl.u8 q9, d18
- vmovl.u8 q10, d20
- vmovl.u8 q11, d22
-
-vp9_convolve8_loop_vert
- ; always process a 4x4 block at a time
- vld1.u32 {d24[0]}, [r7], r1- vld1.u32 {d26[0]}, [r4], r1- vld1.u32 {d26[1]}, [r7], r1- vld1.u32 {d24[1]}, [r4], r1-
- ; extract to s16
- vmovl.u8 q12, d24
- vmovl.u8 q13, d26
-
- pld [r5]
- pld [r8]
-
- ; src[] * filter_y
- MULTIPLY_BY_Q0 q1, d16, d17, d18, d19, d20, d21, d22, d24
-
- pld [r5, r3]
- pld [r8, r3]
-
- MULTIPLY_BY_Q0 q2, d17, d18, d19, d20, d21, d22, d24, d26
-
- pld [r7]
- pld [r4]
-
- MULTIPLY_BY_Q0 q14, d18, d19, d20, d21, d22, d24, d26, d27
-
- pld [r7, r1]
- pld [r4, r1]
-
- MULTIPLY_BY_Q0 q15, d19, d20, d21, d22, d24, d26, d27, d25
-
- ; += 64 >> 7
- vqrshrun.s32 d2, q1, #7
- vqrshrun.s32 d3, q2, #7
- vqrshrun.s32 d4, q14, #7
- vqrshrun.s32 d5, q15, #7
-
- ; saturate
- vqmovn.u16 d2, q1
- vqmovn.u16 d3, q2
-
- vst1.u32 {d2[0]}, [r5@32], r3- vst1.u32 {d2[1]}, [r8@32], r3- vst1.u32 {d3[0]}, [r5@32], r3- vst1.u32 {d3[1]}, [r8@32], r3-
- vmov q8, q10
- vmov d18, d22
- vmov d19, d24
- vmov q10, q13
- vmov d22, d25
-
- subs r12, r12, #4 ; h -= 4
- bgt vp9_convolve8_loop_vert
-
- ; outer loop
- add r0, r0, #4
- add r2, r2, #4
- subs r6, r6, #4 ; w -= 4
- bgt vp9_convolve8_loop_vert_h
-
- pop {r4-r8, pc}-
- ENDP
- END
--- a/vp9/common/arm/neon/vp9_convolve_avg_neon.c
+++ /dev/null
@@ -1,145 +1,0 @@
-/*
- * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <stddef.h>
-#include <arm_neon.h>
-
-void vp9_convolve_avg_neon(
- const uint8_t *src, // r0
- ptrdiff_t src_stride, // r1
- uint8_t *dst, // r2
- ptrdiff_t dst_stride, // r3
- const int16_t *filter_x,
- int filter_x_stride,
- const int16_t *filter_y,
- int filter_y_stride,
- int w,
- int h) {- uint8_t *d;
- uint8x8_t d0u8, d1u8, d2u8, d3u8;
- uint32x2_t d0u32, d2u32;
- uint8x16_t q0u8, q1u8, q2u8, q3u8, q8u8, q9u8, q10u8, q11u8;
- (void)filter_x; (void)filter_x_stride;
- (void)filter_y; (void)filter_y_stride;
-
- d = dst;
- if (w > 32) { // avg64- for (; h > 0; h -= 1) {- q0u8 = vld1q_u8(src);
- q1u8 = vld1q_u8(src + 16);
- q2u8 = vld1q_u8(src + 32);
- q3u8 = vld1q_u8(src + 48);
- src += src_stride;
- q8u8 = vld1q_u8(d);
- q9u8 = vld1q_u8(d + 16);
- q10u8 = vld1q_u8(d + 32);
- q11u8 = vld1q_u8(d + 48);
- d += dst_stride;
-
- q0u8 = vrhaddq_u8(q0u8, q8u8);
- q1u8 = vrhaddq_u8(q1u8, q9u8);
- q2u8 = vrhaddq_u8(q2u8, q10u8);
- q3u8 = vrhaddq_u8(q3u8, q11u8);
-
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q1u8);
- vst1q_u8(dst + 32, q2u8);
- vst1q_u8(dst + 48, q3u8);
- dst += dst_stride;
- }
- } else if (w == 32) { // avg32- for (; h > 0; h -= 2) {- q0u8 = vld1q_u8(src);
- q1u8 = vld1q_u8(src + 16);
- src += src_stride;
- q2u8 = vld1q_u8(src);
- q3u8 = vld1q_u8(src + 16);
- src += src_stride;
- q8u8 = vld1q_u8(d);
- q9u8 = vld1q_u8(d + 16);
- d += dst_stride;
- q10u8 = vld1q_u8(d);
- q11u8 = vld1q_u8(d + 16);
- d += dst_stride;
-
- q0u8 = vrhaddq_u8(q0u8, q8u8);
- q1u8 = vrhaddq_u8(q1u8, q9u8);
- q2u8 = vrhaddq_u8(q2u8, q10u8);
- q3u8 = vrhaddq_u8(q3u8, q11u8);
-
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q1u8);
- dst += dst_stride;
- vst1q_u8(dst, q2u8);
- vst1q_u8(dst + 16, q3u8);
- dst += dst_stride;
- }
- } else if (w > 8) { // avg16- for (; h > 0; h -= 2) {- q0u8 = vld1q_u8(src);
- src += src_stride;
- q1u8 = vld1q_u8(src);
- src += src_stride;
- q2u8 = vld1q_u8(d);
- d += dst_stride;
- q3u8 = vld1q_u8(d);
- d += dst_stride;
-
- q0u8 = vrhaddq_u8(q0u8, q2u8);
- q1u8 = vrhaddq_u8(q1u8, q3u8);
-
- vst1q_u8(dst, q0u8);
- dst += dst_stride;
- vst1q_u8(dst, q1u8);
- dst += dst_stride;
- }
- } else if (w == 8) { // avg8- for (; h > 0; h -= 2) {- d0u8 = vld1_u8(src);
- src += src_stride;
- d1u8 = vld1_u8(src);
- src += src_stride;
- d2u8 = vld1_u8(d);
- d += dst_stride;
- d3u8 = vld1_u8(d);
- d += dst_stride;
-
- q0u8 = vcombine_u8(d0u8, d1u8);
- q1u8 = vcombine_u8(d2u8, d3u8);
- q0u8 = vrhaddq_u8(q0u8, q1u8);
-
- vst1_u8(dst, vget_low_u8(q0u8));
- dst += dst_stride;
- vst1_u8(dst, vget_high_u8(q0u8));
- dst += dst_stride;
- }
- } else { // avg4- for (; h > 0; h -= 2) {- d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 0);
- src += src_stride;
- d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 1);
- src += src_stride;
- d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 0);
- d += dst_stride;
- d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 1);
- d += dst_stride;
-
- d0u8 = vrhadd_u8(vreinterpret_u8_u32(d0u32),
- vreinterpret_u8_u32(d2u32));
-
- d0u32 = vreinterpret_u32_u8(d0u8);
- vst1_lane_u32((uint32_t *)dst, d0u32, 0);
- dst += dst_stride;
- vst1_lane_u32((uint32_t *)dst, d0u32, 1);
- dst += dst_stride;
- }
- }
- return;
-}
--- a/vp9/common/arm/neon/vp9_convolve_avg_neon_asm.asm
+++ /dev/null
@@ -1,116 +1,0 @@
-;
-; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
- EXPORT |vp9_convolve_avg_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-|vp9_convolve_avg_neon| PROC
- push {r4-r6, lr}- ldrd r4, r5, [sp, #32]
- mov r6, r2
-
- cmp r4, #32
- bgt avg64
- beq avg32
- cmp r4, #8
- bgt avg16
- beq avg8
- b avg4
-
-avg64
- sub lr, r1, #32
- sub r4, r3, #32
-avg64_h
- pld [r0, r1, lsl #1]
- vld1.8 {q0-q1}, [r0]!- vld1.8 {q2-q3}, [r0], lr- pld [r2, r3]
- vld1.8 {q8-q9}, [r6@128]!- vld1.8 {q10-q11}, [r6@128], r4- vrhadd.u8 q0, q0, q8
- vrhadd.u8 q1, q1, q9
- vrhadd.u8 q2, q2, q10
- vrhadd.u8 q3, q3, q11
- vst1.8 {q0-q1}, [r2@128]!- vst1.8 {q2-q3}, [r2@128], r4- subs r5, r5, #1
- bgt avg64_h
- pop {r4-r6, pc}-
-avg32
- vld1.8 {q0-q1}, [r0], r1- vld1.8 {q2-q3}, [r0], r1- vld1.8 {q8-q9}, [r6@128], r3- vld1.8 {q10-q11}, [r6@128], r3- pld [r0]
- vrhadd.u8 q0, q0, q8
- pld [r0, r1]
- vrhadd.u8 q1, q1, q9
- pld [r6]
- vrhadd.u8 q2, q2, q10
- pld [r6, r3]
- vrhadd.u8 q3, q3, q11
- vst1.8 {q0-q1}, [r2@128], r3- vst1.8 {q2-q3}, [r2@128], r3- subs r5, r5, #2
- bgt avg32
- pop {r4-r6, pc}-
-avg16
- vld1.8 {q0}, [r0], r1- vld1.8 {q1}, [r0], r1- vld1.8 {q2}, [r6@128], r3- vld1.8 {q3}, [r6@128], r3- pld [r0]
- pld [r0, r1]
- vrhadd.u8 q0, q0, q2
- pld [r6]
- pld [r6, r3]
- vrhadd.u8 q1, q1, q3
- vst1.8 {q0}, [r2@128], r3- vst1.8 {q1}, [r2@128], r3- subs r5, r5, #2
- bgt avg16
- pop {r4-r6, pc}-
-avg8
- vld1.8 {d0}, [r0], r1- vld1.8 {d1}, [r0], r1- vld1.8 {d2}, [r6@64], r3- vld1.8 {d3}, [r6@64], r3- pld [r0]
- pld [r0, r1]
- vrhadd.u8 q0, q0, q1
- pld [r6]
- pld [r6, r3]
- vst1.8 {d0}, [r2@64], r3- vst1.8 {d1}, [r2@64], r3- subs r5, r5, #2
- bgt avg8
- pop {r4-r6, pc}-
-avg4
- vld1.32 {d0[0]}, [r0], r1- vld1.32 {d0[1]}, [r0], r1- vld1.32 {d2[0]}, [r6@32], r3- vld1.32 {d2[1]}, [r6@32], r3- vrhadd.u8 d0, d0, d2
- vst1.32 {d0[0]}, [r2@32], r3- vst1.32 {d0[1]}, [r2@32], r3- subs r5, r5, #2
- bgt avg4
- pop {r4-r6, pc}- ENDP
-
- END
--- a/vp9/common/arm/neon/vp9_convolve_neon.c
+++ /dev/null
@@ -1,82 +1,0 @@
-/*
- * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/vp9_common.h"
-#include "vpx_ports/mem.h"
-
-void vp9_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- /* Given our constraints: w <= 64, h <= 64, taps == 8 we can reduce the
- * maximum buffer size to 64 * 64 + 7 (+ 1 to make it divisible by 4).
- */
- DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);
-
- // Account for the vertical phase needing 3 lines prior and 4 lines post
- int intermediate_height = h + 7;
-
- if (x_step_q4 != 16 || y_step_q4 != 16) {- vp9_convolve8_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
- return;
- }
-
- /* Filter starting 3 lines back. The neon implementation will ignore the
- * given height and filter a multiple of 4 lines. Since this goes in to
- * the temp buffer which has lots of extra room and is subsequently discarded
- * this is safe if somewhat less than ideal.
- */
- vp9_convolve8_horiz_neon(src - src_stride * 3, src_stride,
- temp, 64,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, intermediate_height);
-
- /* Step into the temp buffer 3 lines to get the actual frame data */
- vp9_convolve8_vert_neon(temp + 64 * 3, 64,
- dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
-}
-
-void vp9_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);
- int intermediate_height = h + 7;
-
- if (x_step_q4 != 16 || y_step_q4 != 16) {- vp9_convolve8_avg_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
- return;
- }
-
- /* This implementation has the same issues as above. In addition, we only want
- * to average the values after both passes.
- */
- vp9_convolve8_horiz_neon(src - src_stride * 3, src_stride,
- temp, 64,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, intermediate_height);
- vp9_convolve8_avg_vert_neon(temp + 64 * 3,
- 64, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
-}
--- a/vp9/common/arm/neon/vp9_copy_neon.c
+++ /dev/null
@@ -1,92 +1,0 @@
-/*
- * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <stddef.h>
-#include <arm_neon.h>
-
-void vp9_convolve_copy_neon(
- const uint8_t *src, // r0
- ptrdiff_t src_stride, // r1
- uint8_t *dst, // r2
- ptrdiff_t dst_stride, // r3
- const int16_t *filter_x,
- int filter_x_stride,
- const int16_t *filter_y,
- int filter_y_stride,
- int w,
- int h) {- uint8x8_t d0u8, d2u8;
- uint8x16_t q0u8, q1u8, q2u8, q3u8;
- (void)filter_x; (void)filter_x_stride;
- (void)filter_y; (void)filter_y_stride;
-
- if (w > 32) { // copy64- for (; h > 0; h--) {- q0u8 = vld1q_u8(src);
- q1u8 = vld1q_u8(src + 16);
- q2u8 = vld1q_u8(src + 32);
- q3u8 = vld1q_u8(src + 48);
- src += src_stride;
-
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q1u8);
- vst1q_u8(dst + 32, q2u8);
- vst1q_u8(dst + 48, q3u8);
- dst += dst_stride;
- }
- } else if (w == 32) { // copy32- for (; h > 0; h -= 2) {- q0u8 = vld1q_u8(src);
- q1u8 = vld1q_u8(src + 16);
- src += src_stride;
- q2u8 = vld1q_u8(src);
- q3u8 = vld1q_u8(src + 16);
- src += src_stride;
-
- vst1q_u8(dst, q0u8);
- vst1q_u8(dst + 16, q1u8);
- dst += dst_stride;
- vst1q_u8(dst, q2u8);
- vst1q_u8(dst + 16, q3u8);
- dst += dst_stride;
- }
- } else if (w > 8) { // copy16- for (; h > 0; h -= 2) {- q0u8 = vld1q_u8(src);
- src += src_stride;
- q1u8 = vld1q_u8(src);
- src += src_stride;
-
- vst1q_u8(dst, q0u8);
- dst += dst_stride;
- vst1q_u8(dst, q1u8);
- dst += dst_stride;
- }
- } else if (w == 8) { // copy8- for (; h > 0; h -= 2) {- d0u8 = vld1_u8(src);
- src += src_stride;
- d2u8 = vld1_u8(src);
- src += src_stride;
-
- vst1_u8(dst, d0u8);
- dst += dst_stride;
- vst1_u8(dst, d2u8);
- dst += dst_stride;
- }
- } else { // copy4- for (; h > 0; h--) {- *(uint32_t *)dst = *(const uint32_t *)src;
- src += src_stride;
- dst += dst_stride;
- }
- }
- return;
-}
--- a/vp9/common/arm/neon/vp9_copy_neon_asm.asm
+++ /dev/null
@@ -1,84 +1,0 @@
-;
-; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
- EXPORT |vp9_convolve_copy_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-|vp9_convolve_copy_neon| PROC
- push {r4-r5, lr}- ldrd r4, r5, [sp, #28]
-
- cmp r4, #32
- bgt copy64
- beq copy32
- cmp r4, #8
- bgt copy16
- beq copy8
- b copy4
-
-copy64
- sub lr, r1, #32
- sub r3, r3, #32
-copy64_h
- pld [r0, r1, lsl #1]
- vld1.8 {q0-q1}, [r0]!- vld1.8 {q2-q3}, [r0], lr- vst1.8 {q0-q1}, [r2@128]!- vst1.8 {q2-q3}, [r2@128], r3- subs r5, r5, #1
- bgt copy64_h
- pop {r4-r5, pc}-
-copy32
- pld [r0, r1, lsl #1]
- vld1.8 {q0-q1}, [r0], r1- pld [r0, r1, lsl #1]
- vld1.8 {q2-q3}, [r0], r1- vst1.8 {q0-q1}, [r2@128], r3- vst1.8 {q2-q3}, [r2@128], r3- subs r5, r5, #2
- bgt copy32
- pop {r4-r5, pc}-
-copy16
- pld [r0, r1, lsl #1]
- vld1.8 {q0}, [r0], r1- pld [r0, r1, lsl #1]
- vld1.8 {q1}, [r0], r1- vst1.8 {q0}, [r2@128], r3- vst1.8 {q1}, [r2@128], r3- subs r5, r5, #2
- bgt copy16
- pop {r4-r5, pc}-
-copy8
- pld [r0, r1, lsl #1]
- vld1.8 {d0}, [r0], r1- pld [r0, r1, lsl #1]
- vld1.8 {d2}, [r0], r1- vst1.8 {d0}, [r2@64], r3- vst1.8 {d2}, [r2@64], r3- subs r5, r5, #2
- bgt copy8
- pop {r4-r5, pc}-
-copy4
- ldr r12, [r0], r1
- str r12, [r2], r3
- subs r5, r5, #1
- bgt copy4
- pop {r4-r5, pc}- ENDP
-
- END
--- a/vp9/common/mips/msa/vp9_convolve8_avg_horiz_msa.c
+++ /dev/null
@@ -1,782 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/mips/msa/vp9_convolve_msa.h"
-
-static void common_hz_8t_and_aver_dst_4x4_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 dst0, dst1, dst2, dst3, res2, res3;
- v16u8 mask0, mask1, mask2, mask3;
- v8i16 filt, res0, res1;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, res0, res1);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- SRARI_H2_SH(res0, res1, FILTER_BITS);
- SAT_SH2_SH(res0, res1, 7);
- PCKEV_B2_UB(res0, res0, res1, res1, res2, res3);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
- XORI_B2_128_UB(res2, res3);
- AVER_UB2_UB(res2, dst0, res3, dst2, res2, res3);
- ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hz_8t_and_aver_dst_4x8_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, res0, res1, res2, res3;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v8i16 filt, vec0, vec1, vec2, vec3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- src += (4 * src_stride);
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, vec0, vec1);
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, vec2, vec3);
- SRARI_H4_SH(vec0, vec1, vec2, vec3, FILTER_BITS);
- SAT_SH4_SH(vec0, vec1, vec2, vec3, 7);
- PCKEV_B4_UB(vec0, vec0, vec1, vec1, vec2, vec2, vec3, vec3, res0, res1, res2,
- res3);
- ILVR_D2_UB(res1, res0, res3, res2, res0, res2);
- XORI_B2_128_UB(res0, res2);
- ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2, dst4,
- dst6);
- ILVR_D2_UB(dst2, dst0, dst6, dst4, dst0, dst4);
- AVER_UB2_UB(res0, dst0, res2, dst4, res0, res2);
- ST4x8_UB(res0, res2, dst, dst_stride);
-}
-
-static void common_hz_8t_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- if (4 == height) {- common_hz_8t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {- common_hz_8t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_hz_8t_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- int32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, dst0, dst1, dst2, dst3;
- v8i16 filt, out0, out1, out2, out3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- src += (4 * src_stride);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
- mask3, filt0, filt1, filt2, filt3, out0, out1,
- out2, out3);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- CONVERT_UB_AVG_ST8x4_UB(out0, out1, out2, out3, dst0, dst1, dst2, dst3,
- dst, dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void common_hz_8t_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- int32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, dst0, dst1;
- v8i16 filt, out0, out1, out2, out3;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = height >> 1; loop_cnt--;) {- LD_SB2(src, src_stride, src0, src2);
- LD_SB2(src + 8, src_stride, src1, src3);
- src += (2 * src_stride);
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8, vec12);
- VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9, vec13);
- VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
- vec14);
- VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
- vec15);
- DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
- vec9, vec10, vec11);
- DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0, vec1,
- vec2, vec3);
- DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
- vec9, vec10, vec11);
- ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
- out2, out3);
- LD_UB2(dst, dst_stride, dst0, dst1);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- PCKEV_XORI128_AVG_ST_UB(out1, out0, dst0, dst);
- dst += dst_stride;
- PCKEV_XORI128_AVG_ST_UB(out3, out2, dst1, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hz_8t_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 dst1, dst2, mask0, mask1, mask2, mask3;
- v8i16 filt, out0, out1, out2, out3;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = height; loop_cnt--;) {- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src3 = LD_SB(src + 24);
- src1 = __msa_sldi_b(src2, src0, 8);
- src += src_stride;
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8, vec12);
- VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9, vec13);
- VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
- vec14);
- VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
- vec15);
- DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
- vec9, vec10, vec11);
- DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0, vec1,
- vec2, vec3);
- DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
- vec9, vec10, vec11);
- ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- LD_UB2(dst, 16, dst1, dst2);
- PCKEV_XORI128_AVG_ST_UB(out1, out0, dst1, dst);
- PCKEV_XORI128_AVG_ST_UB(out3, out2, dst2, dst + 16);
- dst += dst_stride;
- }
-}
-
-static void common_hz_8t_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt, cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 dst1, dst2, mask0, mask1, mask2, mask3;
- v8i16 filt, out0, out1, out2, out3;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = height; loop_cnt--;) {- for (cnt = 0; cnt < 2; ++cnt) {- src0 = LD_SB(&src[cnt << 5]);
- src2 = LD_SB(&src[16 + (cnt << 5)]);
- src3 = LD_SB(&src[24 + (cnt << 5)]);
- src1 = __msa_sldi_b(src2, src0, 8);
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8,
- vec12);
- VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9,
- vec13);
- VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
- vec14);
- VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
- vec15);
- DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0,
- vec1, vec2, vec3);
- DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
- vec9, vec10, vec11);
- DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0,
- vec1, vec2, vec3);
- DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
- vec9, vec10, vec11);
- ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- LD_UB2(&dst[cnt << 5], 16, dst1, dst2);
- PCKEV_XORI128_AVG_ST_UB(out1, out0, dst1, &dst[cnt << 5]);
- PCKEV_XORI128_AVG_ST_UB(out3, out2, dst2, &dst[16 + (cnt << 5)]);
- }
-
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_and_aver_dst_4x4_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3, vec0, vec1, res0, res1;
- v8u16 vec2, vec3, const255, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- const255 = (v8u16)__msa_ldi_h(255);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, vec2, vec3);
- SRARI_H2_UH(vec2, vec3, FILTER_BITS);
- MIN_UH2_UH(vec2, vec3, const255);
- PCKEV_B2_UB(vec2, vec2, vec3, vec3, res0, res1);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
- AVER_UB2_UB(res0, dst0, res1, dst2, res0, res1);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hz_2t_and_aver_dst_4x8_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, vec0, vec1, vec2, vec3, res0, res1, res2, res3;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v8u16 vec4, vec5, vec6, vec7, const255, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- const255 = (v8u16)__msa_ldi_h(255);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
- VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
- VSHF_B2_UB(src4, src5, src6, src7, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec4, vec5,
- vec6, vec7);
- SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS);
- MIN_UH4_UH(vec4, vec5, vec6, vec7, const255);
- PCKEV_B4_UB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1, res2,
- res3);
- ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2, dst4,
- dst6);
- AVER_UB4_UB(res0, dst0, res1, dst2, res2, dst4, res3, dst6, res0, res1, res2,
- res3);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
- dst += (4 * dst_stride);
- ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hz_2t_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- if (4 == height) {- common_hz_2t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {- common_hz_2t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_hz_2t_and_aver_dst_8x4_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v8u16 vec0, vec1, vec2, vec3, const255, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- const255 = (v8u16)__msa_ldi_h(255);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
- dst, dst_stride);
-}
-
-static void common_hz_2t_and_aver_dst_8x8mult_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- v16i8 src0, src1, src2, src3, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v8u16 vec0, vec1, vec2, vec3, const255, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- const255 = (v8u16)__msa_ldi_h(255);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
- dst, dst_stride);
- dst += (4 * dst_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
- dst, dst_stride);
- dst += (4 * dst_stride);
-
- if (16 == height) {- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
- dst, dst_stride);
- dst += (4 * dst_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
- dst, dst_stride);
- }
-}
-
-static void common_hz_2t_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- if (4 == height) {- common_hz_2t_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride, filter);
- } else {- common_hz_2t_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
- filter, height);
- }
-}
-
-static void common_hz_2t_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 res0, res1, res2, res3, res4, res5, res6, res7, const255, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- const255 = (v8u16)__msa_ldi_h(255);
-
- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
- res2, res3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
- res6, res7);
- SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
- SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- MIN_UH4_UH(res0, res1, res2, res3, const255);
- MIN_UH4_UH(res4, res5, res6, res7, const255);
- PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res3, res2, dst1, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res7, res6, dst3, dst);
- dst += dst_stride;
-
- for (loop_cnt = (height >> 2) - 1; loop_cnt--;) {- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
- res2, res3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
- res6, res7);
- SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
- SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- MIN_UH4_UH(res0, res1, res2, res3, const255);
- MIN_UH4_UH(res4, res5, res6, res7, const255);
- PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res3, res2, dst1, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
- dst += dst_stride;
- PCKEV_AVG_ST_UB(res7, res6, dst3, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 res0, res1, res2, res3, res4, res5, res6, res7, const255, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- const255 = (v8u16)__msa_ldi_h(255);
-
- for (loop_cnt = (height >> 1); loop_cnt--;) {- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src3 = LD_SB(src + 24);
- src1 = __msa_sldi_b(src2, src0, 8);
- src += src_stride;
- src4 = LD_SB(src);
- src6 = LD_SB(src + 16);
- src7 = LD_SB(src + 24);
- src5 = __msa_sldi_b(src6, src4, 8);
- src += src_stride;
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
- res2, res3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
- res6, res7);
- SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
- SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
- MIN_UH4_UH(res0, res1, res2, res3, const255);
- MIN_UH4_UH(res4, res5, res6, res7, const255);
- LD_UB2(dst, 16, dst0, dst1);
- PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
- PCKEV_AVG_ST_UB(res3, res2, dst1, (dst + 16));
- dst += dst_stride;
- LD_UB2(dst, 16, dst2, dst3);
- PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
- PCKEV_AVG_ST_UB(res7, res6, dst3, (dst + 16));
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, dst0, dst1, dst2, dst3;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 out0, out1, out2, out3, out4, out5, out6, out7, const255, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- const255 = (v8u16)__msa_ldi_h(255);
-
- for (loop_cnt = height; loop_cnt--;) {- LD_SB4(src, 16, src0, src2, src4, src6);
- src7 = LD_SB(src + 56);
- SLDI_B3_SB(src2, src4, src6, src0, src2, src4, src1, src3, src5, 8);
- src += src_stride;
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
- out2, out3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
- out6, out7);
- SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
- SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
- LD_UB4(dst, 16, dst0, dst1, dst2, dst3);
- MIN_UH4_UH(out0, out1, out2, out3, const255);
- MIN_UH4_UH(out4, out5, out6, out7, const255);
- PCKEV_AVG_ST_UB(out1, out0, dst0, dst);
- PCKEV_AVG_ST_UB(out3, out2, dst1, dst + 16);
- PCKEV_AVG_ST_UB(out5, out4, dst2, dst + 32);
- PCKEV_AVG_ST_UB(out7, out6, dst3, dst + 48);
- dst += dst_stride;
- }
-}
-
-void vp9_convolve8_avg_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- int8_t cnt, filt_hor[8];
-
- if (16 != x_step_q4) {- vp9_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- if (((const int32_t *)filter_x)[1] == 0x800000) {- vp9_convolve_avg(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- for (cnt = 0; cnt < 8; ++cnt) {- filt_hor[cnt] = filter_x[cnt];
- }
-
- if (((const int32_t *)filter_x)[0] == 0) {- switch (w) {- case 4:
- common_hz_2t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- case 8:
- common_hz_2t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- case 16:
- common_hz_2t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- case 32:
- common_hz_2t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- case 64:
- common_hz_2t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- default:
- vp9_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- } else {- switch (w) {- case 4:
- common_hz_8t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- case 8:
- common_hz_8t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- case 16:
- common_hz_8t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- case 32:
- common_hz_8t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- case 64:
- common_hz_8t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- default:
- vp9_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- }
-}
--- a/vp9/common/mips/msa/vp9_convolve8_avg_msa.c
+++ /dev/null
@@ -1,679 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/mips/msa/vp9_convolve_msa.h"
-
-static void common_hv_8ht_8vt_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16u8 dst0, dst1, dst2, dst3, mask0, mask1, mask2, mask3, tmp0, tmp1;
- v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
- v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8i16 hz_out7, hz_out8, hz_out9, res0, res1, vec0, vec1, vec2, vec3, vec4;
- v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= (3 + 3 * src_stride);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- hz_out0 = HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out2 = HORIZ_8TAP_FILT(src2, src3, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out4 = HORIZ_8TAP_FILT(src4, src5, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out5 = HORIZ_8TAP_FILT(src5, src6, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- SLDI_B2_SH(hz_out2, hz_out4, hz_out0, hz_out2, hz_out1, hz_out3, 8);
-
- filt = LD_SH(filter_vert);
- SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
-
- ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- vec2 = (v8i16)__msa_ilvev_b((v16i8)hz_out5, (v16i8)hz_out4);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- XORI_B4_128_SB(src7, src8, src9, src10);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- hz_out7 = HORIZ_8TAP_FILT(src7, src8, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- hz_out6 = (v8i16)__msa_sldi_b((v16i8)hz_out7, (v16i8)hz_out5, 8);
- vec3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
- res0 = FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out9 = HORIZ_8TAP_FILT(src9, src10, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- hz_out8 = (v8i16)__msa_sldi_b((v16i8)hz_out9, (v16i8)hz_out7, 8);
- vec4 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
- res1 = FILT_8TAP_DPADD_S_H(vec1, vec2, vec3, vec4, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
-
- SRARI_H2_SH(res0, res1, FILTER_BITS);
- SAT_SH2_SH(res0, res1, 7);
- PCKEV_B2_UB(res0, res0, res1, res1, tmp0, tmp1);
- XORI_B2_128_UB(tmp0, tmp1);
- AVER_UB2_UB(tmp0, dst0, tmp1, dst2, tmp0, tmp1);
- ST4x4_UB(tmp0, tmp1, 0, 1, 0, 1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- hz_out5 = hz_out9;
- vec0 = vec2;
- vec1 = vec3;
- vec2 = vec4;
- }
-}
-
-static void common_hv_8ht_8vt_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
- v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
- v16u8 dst0, dst1, dst2, dst3, mask0, mask1, mask2, mask3;
- v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8i16 hz_out7, hz_out8, hz_out9, hz_out10, tmp0, tmp1, tmp2, tmp3;
- v8i16 out0, out1, out2, out3, out4, out5, out6, out7, out8, out9;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= (3 + 3 * src_stride);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- hz_out0 = HORIZ_8TAP_FILT(src0, src0, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out1 = HORIZ_8TAP_FILT(src1, src1, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out2 = HORIZ_8TAP_FILT(src2, src2, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out3 = HORIZ_8TAP_FILT(src3, src3, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out4 = HORIZ_8TAP_FILT(src4, src4, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out5 = HORIZ_8TAP_FILT(src5, src5, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out6 = HORIZ_8TAP_FILT(src6, src6, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
-
- filt = LD_SH(filter_vert);
- SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
-
- ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1);
- ILVEV_B2_SH(hz_out4, hz_out5, hz_out1, hz_out2, out2, out4);
- ILVEV_B2_SH(hz_out3, hz_out4, hz_out5, hz_out6, out5, out6);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- XORI_B4_128_SB(src7, src8, src9, src10);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-
- hz_out7 = HORIZ_8TAP_FILT(src7, src7, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
- tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out8 = HORIZ_8TAP_FILT(src8, src8, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out7 = (v8i16)__msa_ilvev_b((v16i8)hz_out8, (v16i8)hz_out7);
- tmp1 = FILT_8TAP_DPADD_S_H(out4, out5, out6, out7, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out9 = HORIZ_8TAP_FILT(src9, src9, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out8 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
- tmp2 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out8, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out10 = HORIZ_8TAP_FILT(src10, src10, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out9 = (v8i16)__msa_ilvev_b((v16i8)hz_out10, (v16i8)hz_out9);
- tmp3 = FILT_8TAP_DPADD_S_H(out5, out6, out7, out9, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
- CONVERT_UB_AVG_ST8x4_UB(tmp0, tmp1, tmp2, tmp3, dst0, dst1, dst2, dst3,
- dst, dst_stride);
- dst += (4 * dst_stride);
-
- hz_out6 = hz_out10;
- out0 = out2;
- out1 = out3;
- out2 = out8;
- out4 = out6;
- out5 = out7;
- out6 = out9;
- }
-}
-
-static void common_hv_8ht_8vt_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 2; multiple8_cnt--;) {- common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_8ht_8vt_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 4; multiple8_cnt--;) {- common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_8ht_8vt_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 8; multiple8_cnt--;) {- common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_4x4_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert) {- v16i8 src0, src1, src2, src3, src4, mask;
- v16u8 filt_hz, filt_vt, vec0, vec1;
- v16u8 dst0, dst1, dst2, dst3, res0, res1;
- v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, tmp0, tmp1, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- filt = LD_UH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
- hz_out4 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- hz_out1 = (v8u16)__msa_sldi_b((v16i8)hz_out2, (v16i8)hz_out0, 8);
- hz_out3 = (v8u16)__msa_pckod_d((v2i64)hz_out4, (v2i64)hz_out2);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_B2_UB(tmp0, tmp0, tmp1, tmp1, res0, res1);
- AVER_UB2_UB(res0, dst0, res1, dst2, res0, res1);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_4x8_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert) {- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, mask;
- v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3, res0, res1, res2, res3;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8u16 hz_out7, hz_out8, tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
- src8 = LD_SB(src);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
- hz_out4 = HORIZ_2TAP_FILT_UH(src4, src5, mask, filt_hz, FILTER_BITS);
- hz_out6 = HORIZ_2TAP_FILT_UH(src6, src7, mask, filt_hz, FILTER_BITS);
- hz_out8 = HORIZ_2TAP_FILT_UH(src8, src8, mask, filt_hz, FILTER_BITS);
- SLDI_B3_UH(hz_out2, hz_out4, hz_out6, hz_out0, hz_out2, hz_out4, hz_out1,
- hz_out3, hz_out5, 8);
- hz_out7 = (v8u16)__msa_pckod_d((v2i64)hz_out8, (v2i64)hz_out6);
-
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
- ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2,
- dst4, dst6);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- ILVEV_B2_UB(hz_out4, hz_out5, hz_out6, hz_out7, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt_vt, filt_vt, filt_vt, filt_vt,
- tmp0, tmp1, tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_B4_UB(tmp0, tmp0, tmp1, tmp1, tmp2, tmp2, tmp3, tmp3, res0, res1,
- res2, res3);
- AVER_UB4_UB(res0, dst0, res1, dst2, res2, dst4, res3, dst6, res0, res1,
- res2, res3);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
- dst += (4 * dst_stride);
- ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- if (4 == height) {- common_hv_2ht_2vt_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert);
- } else if (8 == height) {- common_hv_2ht_2vt_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert);
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_8x4_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert) {- v16i8 src0, src1, src2, src3, src4, mask;
- v16u8 filt_hz, filt_vt, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
- v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
- src += (5 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp0 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- vec1 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp1 = __msa_dotp_u_h(vec1, filt_vt);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- vec2 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp2 = __msa_dotp_u_h(vec2, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- vec3 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp3 = __msa_dotp_u_h(vec3, filt_vt);
-
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3,
- dst, dst_stride);
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_8x8mult_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, mask;
- v16u8 filt_hz, filt_vt, vec0, dst0, dst1, dst2, dst3;
- v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- src0 = LD_SB(src);
- src += src_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src1, src2, src3, src4);
- src += (4 * src_stride);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp0 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp1 = __msa_dotp_u_h(vec0, filt_vt);
-
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp2 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp3 = __msa_dotp_u_h(vec0, filt_vt);
-
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3,
- dst, dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- if (4 == height) {- common_hv_2ht_2vt_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert);
- } else {- common_hv_2ht_2vt_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert,
- height);
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt_hz, filt_vt, vec0, vec1, dst0, dst1, dst2, dst3;
- v8u16 hz_out0, hz_out1, hz_out2, hz_out3, tmp0, tmp1;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB2(src, 8, src0, src1);
- src += src_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out3 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
- dst += dst_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst1, dst);
- dst += dst_stride;
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- hz_out3 = HORIZ_2TAP_FILT_UH(src5, src5, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst);
- dst += dst_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src6, src6, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src7, src7, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst3, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 2; multiple8_cnt--;) {- common_hv_2ht_2vt_and_aver_dst_16w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 16;
- dst += 16;
- }
-}
-
-static void common_hv_2ht_2vt_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 4; multiple8_cnt--;) {- common_hv_2ht_2vt_and_aver_dst_16w_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- src += 16;
- dst += 16;
- }
-}
-
-void vp9_convolve8_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- int8_t cnt, filt_hor[8], filt_ver[8];
-
- if (16 != x_step_q4 || 16 != y_step_q4) {- vp9_convolve8_avg_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- if (((const int32_t *)filter_x)[1] == 0x800000 &&
- ((const int32_t *)filter_y)[1] == 0x800000) {- vp9_convolve_avg(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- for (cnt = 0; cnt < 8; ++cnt) {- filt_hor[cnt] = filter_x[cnt];
- filt_ver[cnt] = filter_y[cnt];
- }
-
- if (((const int32_t *)filter_x)[0] == 0 &&
- ((const int32_t *)filter_y)[0] == 0) {- switch (w) {- case 4:
- common_hv_2ht_2vt_and_aver_dst_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], h);
- break;
- case 8:
- common_hv_2ht_2vt_and_aver_dst_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], h);
- break;
- case 16:
- common_hv_2ht_2vt_and_aver_dst_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], h);
- break;
- case 32:
- common_hv_2ht_2vt_and_aver_dst_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], h);
- break;
- case 64:
- common_hv_2ht_2vt_and_aver_dst_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], h);
- break;
- default:
- vp9_convolve8_avg_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- } else if (((const int32_t *)filter_x)[0] == 0 ||
- ((const int32_t *)filter_y)[0] == 0) {- vp9_convolve8_avg_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- } else {- switch (w) {- case 4:
- common_hv_8ht_8vt_and_aver_dst_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, h);
- break;
- case 8:
- common_hv_8ht_8vt_and_aver_dst_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, h);
- break;
- case 16:
- common_hv_8ht_8vt_and_aver_dst_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, h);
- break;
- case 32:
- common_hv_8ht_8vt_and_aver_dst_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, h);
- break;
- case 64:
- common_hv_8ht_8vt_and_aver_dst_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, h);
- break;
- default:
- vp9_convolve8_avg_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- }
-}
--- a/vp9/common/mips/msa/vp9_convolve8_avg_vert_msa.c
+++ /dev/null
@@ -1,753 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/mips/msa/vp9_convolve_msa.h"
-
-static void common_vt_8t_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16u8 dst0, dst1, dst2, dst3, out;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776;
- v16i8 src10998, filt0, filt1, filt2, filt3;
- v8i16 filt, out10, out32;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
- ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src2110,
- src4332, src6554);
- XORI_B3_128_SB(src2110, src4332, src6554);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- ILVR_D2_SB(src87_r, src76_r, src109_r, src98_r, src8776, src10998);
- XORI_B2_128_SB(src8776, src10998);
- out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, filt0,
- filt1, filt2, filt3);
- out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, filt0,
- filt1, filt2, filt3);
- SRARI_H2_SH(out10, out32, FILTER_BITS);
- SAT_SH2_SH(out10, out32, 7);
- out = PCKEV_XORI128_UB(out10, out32);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
-
- dst0 = (v16u8)__msa_ilvr_d((v2i64)dst2, (v2i64)dst0);
- out = __msa_aver_u_b(out, dst0);
-
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
- dst += (4 * dst_stride);
-
- src2110 = src6554;
- src4332 = src8776;
- src6554 = src10998;
- src6 = src10;
- }
-}
-
-static void common_vt_8t_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16u8 dst0, dst1, dst2, dst3;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, filt0, filt1, filt2, filt3;
- v8i16 filt, out0, out1, out2, out3;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- XORI_B4_128_SB(src7, src8, src9, src10);
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- out0 = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
- filt1, filt2, filt3);
- out1 = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
- filt1, filt2, filt3);
- out2 = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
- filt1, filt2, filt3);
- out3 = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
- filt1, filt2, filt3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- CONVERT_UB_AVG_ST8x4_UB(out0, out1, out2, out3, dst0, dst1, dst2, dst3,
- dst, dst_stride);
- dst += (4 * dst_stride);
-
- src10_r = src54_r;
- src32_r = src76_r;
- src54_r = src98_r;
- src21_r = src65_r;
- src43_r = src87_r;
- src65_r = src109_r;
- src6 = src10;
- }
-}
-
-static void common_vt_8t_and_aver_dst_16w_mult_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height,
- int32_t width) {- const uint8_t *src_tmp;
- uint8_t *dst_tmp;
- uint32_t loop_cnt, cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l;
- v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l;
- v16i8 filt0, filt1, filt2, filt3;
- v16u8 dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3;
- v8i16 out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l, filt;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- for (cnt = (width >> 4); cnt--;) {- src_tmp = src;
- dst_tmp = dst;
-
- LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- src_tmp += (7 * src_stride);
-
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
- ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l, src32_l,
- src54_l, src21_l);
- ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src_tmp, src_stride, src7, src8, src9, src10);
- src_tmp += (4 * src_stride);
-
- LD_UB4(dst_tmp, dst_stride, dst0, dst1, dst2, dst3);
- XORI_B4_128_SB(src7, src8, src9, src10);
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l,
- src87_l, src98_l, src109_l);
- out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
- filt1, filt2, filt3);
- out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
- filt1, filt2, filt3);
- out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
- filt1, filt2, filt3);
- out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
- filt1, filt2, filt3);
- out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0,
- filt1, filt2, filt3);
- out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0,
- filt1, filt2, filt3);
- out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0,
- filt1, filt2, filt3);
- out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0,
- filt1, filt2, filt3);
- SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
- SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS);
- SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
- SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7);
- PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l,
- out3_r, tmp0, tmp1, tmp2, tmp3);
- XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3);
- AVER_UB4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, dst0, dst1,
- dst2, dst3);
- ST_UB4(dst0, dst1, dst2, dst3, dst_tmp, dst_stride);
- dst_tmp += (4 * dst_stride);
-
- src10_r = src54_r;
- src32_r = src76_r;
- src54_r = src98_r;
- src21_r = src65_r;
- src43_r = src87_r;
- src65_r = src109_r;
- src10_l = src54_l;
- src32_l = src76_l;
- src54_l = src98_l;
- src21_l = src65_l;
- src43_l = src87_l;
- src65_l = src109_l;
- src6 = src10;
- }
-
- src += 16;
- dst += 16;
- }
-}
-
-static void common_vt_8t_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
- filter, height, 16);
-}
-
-static void common_vt_8t_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
- filter, height, 32);
-}
-
-static void common_vt_8t_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
- filter, height, 64);
-}
-
-static void common_vt_2t_and_aver_dst_4x4_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, src4;
- v16u8 dst0, dst1, dst2, dst3, out, filt0, src2110, src4332;
- v16i8 src10_r, src32_r, src21_r, src43_r;
- v8i16 filt;
- v8u16 tmp0, tmp1;
-
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- src4 = LD_SB(src);
- src += src_stride;
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
- dst0 = (v16u8)__msa_ilvr_d((v2i64)dst1, (v2i64)dst0);
- ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
- src32_r, src43_r);
- ILVR_D2_UB(src21_r, src10_r, src43_r, src32_r, src2110, src4332);
- DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
-
- out = (v16u8)__msa_pckev_b((v16i8)tmp1, (v16i8)tmp0);
- out = __msa_aver_u_b(out, dst0);
-
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
-}
-
-static void common_vt_2t_and_aver_dst_4x8_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src87_r;
- v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r, src65_r;
- v16u8 src2110, src4332, src6554, src8776, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
- src8 = LD_SB(src);
-
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
- ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst1,
- dst2, dst3);
- ILVR_D2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
- ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
- src32_r, src43_r);
- ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r,
- src76_r, src87_r);
- ILVR_D4_UB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r,
- src87_r, src76_r, src2110, src4332, src6554, src8776);
- DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0,
- tmp0, tmp1, tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, src2110, src4332);
- AVER_UB2_UB(src2110, dst0, src4332, dst1, src2110, src4332);
- ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
- dst += (4 * dst_stride);
- ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst, dst_stride);
-}
-
-static void common_vt_2t_and_aver_dst_4w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- if (4 == height) {- common_vt_2t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {- common_vt_2t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_vt_2t_and_aver_dst_8x4_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter) {- v16u8 src0, src1, src2, src3, src4;
- v16u8 dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1);
- ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3,
- dst, dst_stride);
-}
-
-static void common_vt_2t_and_aver_dst_8x8mult_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
- v16u8 dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- src0 = LD_UB(src);
- src += src_stride;
-
- for (loop_cnt = (height >> 3); loop_cnt--;) {- LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8);
- src += (8 * src_stride);
- LD_UB8(dst, dst_stride, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8);
-
- ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3, vec0, vec1,
- vec2, vec3);
- ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7, vec4, vec5,
- vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_AVG_ST8x4_UB(tmp0, dst1, tmp1, dst2, tmp2, dst3, tmp3, dst4,
- dst, dst_stride);
- dst += (4 * dst_stride);
-
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_AVG_ST8x4_UB(tmp0, dst5, tmp1, dst6, tmp2, dst7, tmp3, dst8,
- dst, dst_stride);
- dst += (4 * dst_stride);
-
- src0 = src8;
- }
-}
-
-static void common_vt_2t_and_aver_dst_8w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- if (4 == height) {- common_vt_2t_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride, filter);
- } else {- common_vt_2t_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
- filter, height);
- }
-}
-
-static void common_vt_2t_and_aver_dst_16w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, dst0, dst1, dst2, dst3, filt0;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 tmp0, tmp1, tmp2, tmp3, filt;
-
- /* rearranging filter_y */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- src0 = LD_UB(src);
- src += src_stride;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_UB4(src, src_stride, src1, src2, src3, src4);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
- dst += dst_stride;
-
- ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
- ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst);
- dst += dst_stride;
-
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst);
- dst += dst_stride;
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst);
- dst += dst_stride;
-
- src0 = src4;
- }
-}
-
-static void common_vt_2t_and_aver_dst_32w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3, filt;
-
- /* rearranging filter_y */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_UB2(src, 16, src0, src5);
- src += src_stride;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_UB4(src, src_stride, src1, src2, src3, src4);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
-
- LD_UB4(src + 16, src_stride, src6, src7, src8, src9);
- LD_UB4(dst + 16, dst_stride, dst4, dst5, dst6, dst7);
- src += (4 * src_stride);
-
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride);
-
- ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
- ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst + 2 * dst_stride);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst + 3 * dst_stride);
-
- ILVR_B2_UB(src6, src5, src7, src6, vec0, vec2);
- ILVL_B2_UB(src6, src5, src7, src6, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 16);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 16 + dst_stride);
-
- ILVR_B2_UB(src8, src7, src9, src8, vec4, vec6);
- ILVL_B2_UB(src8, src7, src9, src8, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst6, dst + 16 + 2 * dst_stride);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst7, dst + 16 + 3 * dst_stride);
- dst += (4 * dst_stride);
-
- src0 = src4;
- src5 = src9;
- }
-}
-
-static void common_vt_2t_and_aver_dst_64w_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter,
- int32_t height) {- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5;
- v16u8 src6, src7, src8, src9, src10, src11, filt0;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
- v8u16 filt;
-
- /* rearranging filter_y */
- filt = LD_UH(filter);
- filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_UB4(src, 16, src0, src3, src6, src9);
- src += src_stride;
-
- for (loop_cnt = (height >> 1); loop_cnt--;) {- LD_UB2(src, src_stride, src1, src2);
- LD_UB2(dst, dst_stride, dst0, dst1);
- LD_UB2(src + 16, src_stride, src4, src5);
- LD_UB2(dst + 16, dst_stride, dst2, dst3);
- LD_UB2(src + 32, src_stride, src7, src8);
- LD_UB2(dst + 32, dst_stride, dst4, dst5);
- LD_UB2(src + 48, src_stride, src10, src11);
- LD_UB2(dst + 48, dst_stride, dst6, dst7);
- src += (2 * src_stride);
-
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride);
-
- ILVR_B2_UB(src4, src3, src5, src4, vec4, vec6);
- ILVL_B2_UB(src4, src3, src5, src4, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
- SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
- SAT_UH2_UH(tmp4, tmp5, 7);
- PCKEV_AVG_ST_UB(tmp5, tmp4, dst2, dst + 16);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
- SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
- SAT_UH2_UH(tmp6, tmp7, 7);
- PCKEV_AVG_ST_UB(tmp7, tmp6, dst3, dst + 16 + dst_stride);
-
- ILVR_B2_UB(src7, src6, src8, src7, vec0, vec2);
- ILVL_B2_UB(src7, src6, src8, src7, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 32);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 32 + dst_stride);
-
- ILVR_B2_UB(src10, src9, src11, src10, vec4, vec6);
- ILVL_B2_UB(src10, src9, src11, src10, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
- SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
- SAT_UH2_UH(tmp4, tmp5, 7);
- PCKEV_AVG_ST_UB(tmp5, tmp4, dst6, (dst + 48));
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
- SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
- SAT_UH2_UH(tmp6, tmp7, 7);
- PCKEV_AVG_ST_UB(tmp7, tmp6, dst7, dst + 48 + dst_stride);
- dst += (2 * dst_stride);
-
- src0 = src2;
- src3 = src5;
- src6 = src8;
- src9 = src11;
- }
-}
-
-void vp9_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- int8_t cnt, filt_ver[8];
-
- if (16 != y_step_q4) {- vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- if (((const int32_t *)filter_y)[1] == 0x800000) {- vp9_convolve_avg(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- for (cnt = 0; cnt < 8; ++cnt) {- filt_ver[cnt] = filter_y[cnt];
- }
-
- if (((const int32_t *)filter_y)[0] == 0) {- switch (w) {- case 4:
- common_vt_2t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- case 8:
- common_vt_2t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- case 16:
- common_vt_2t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- case 32:
- common_vt_2t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- case 64:
- common_vt_2t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- default:
- vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- } else {- switch (w) {- case 4:
- common_vt_8t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- case 8:
- common_vt_8t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- case 16:
- common_vt_8t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
-
- break;
- case 32:
- common_vt_8t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- case 64:
- common_vt_8t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- default:
- vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- }
-}
--- a/vp9/common/mips/msa/vp9_convolve8_horiz_msa.c
+++ /dev/null
@@ -1,742 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/mips/msa/vp9_convolve_msa.h"
-
-static void common_hz_8t_4x4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16u8 mask0, mask1, mask2, mask3, out;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v8i16 filt, out0, out1;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, out0, out1);
- SRARI_H2_SH(out0, out1, FILTER_BITS);
- SAT_SH2_SH(out0, out1, 7);
- out = PCKEV_XORI128_UB(out0, out1);
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
-}
-
-static void common_hz_8t_4x8_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16i8 filt0, filt1, filt2, filt3;
- v16i8 src0, src1, src2, src3;
- v16u8 mask0, mask1, mask2, mask3, out;
- v8i16 filt, out0, out1, out2, out3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- src += (4 * src_stride);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, out0, out1);
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- out = PCKEV_XORI128_UB(out0, out1);
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
- dst += (4 * dst_stride);
- out = PCKEV_XORI128_UB(out2, out3);
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
-}
-
-static void common_hz_8t_4w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- if (4 == height) {- common_hz_8t_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {- common_hz_8t_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_hz_8t_8x4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, tmp0, tmp1;
- v8i16 filt, out0, out1, out2, out3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
- filt0, filt1, filt2, filt3, out0, out1, out2,
- out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- tmp0 = PCKEV_XORI128_UB(out0, out1);
- tmp1 = PCKEV_XORI128_UB(out2, out3);
- ST8x4_UB(tmp0, tmp1, dst, dst_stride);
-}
-
-static void common_hz_8t_8x8mult_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, tmp0, tmp1;
- v8i16 filt, out0, out1, out2, out3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src0, src1, src2, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- src += (4 * src_stride);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
- mask3, filt0, filt1, filt2, filt3, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- tmp0 = PCKEV_XORI128_UB(out0, out1);
- tmp1 = PCKEV_XORI128_UB(out2, out3);
- ST8x4_UB(tmp0, tmp1, dst, dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void common_hz_8t_8w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- if (4 == height) {- common_hz_8t_8x4_msa(src, src_stride, dst, dst_stride, filter);
- } else {- common_hz_8t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height);
- }
-}
-
-static void common_hz_8t_16w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, out;
- v8i16 filt, out0, out1, out2, out3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = (height >> 1); loop_cnt--;) {- LD_SB2(src, src_stride, src0, src2);
- LD_SB2(src + 8, src_stride, src1, src3);
- XORI_B4_128_SB(src0, src1, src2, src3);
- src += (2 * src_stride);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
- mask3, filt0, filt1, filt2, filt3, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- out = PCKEV_XORI128_UB(out0, out1);
- ST_UB(out, dst);
- dst += dst_stride;
- out = PCKEV_XORI128_UB(out2, out3);
- ST_UB(out, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hz_8t_32w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, out;
- v8i16 filt, out0, out1, out2, out3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = (height >> 1); loop_cnt--;) {- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src3 = LD_SB(src + 24);
- src1 = __msa_sldi_b(src2, src0, 8);
- src += src_stride;
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
- mask3, filt0, filt1, filt2, filt3, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
-
- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src3 = LD_SB(src + 24);
- src1 = __msa_sldi_b(src2, src0, 8);
- src += src_stride;
-
- out = PCKEV_XORI128_UB(out0, out1);
- ST_UB(out, dst);
- out = PCKEV_XORI128_UB(out2, out3);
- ST_UB(out, dst + 16);
- dst += dst_stride;
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
- mask3, filt0, filt1, filt2, filt3, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- out = PCKEV_XORI128_UB(out0, out1);
- ST_UB(out, dst);
- out = PCKEV_XORI128_UB(out2, out3);
- ST_UB(out, dst + 16);
- dst += dst_stride;
- }
-}
-
-static void common_hz_8t_64w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- int32_t loop_cnt;
- v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
- v16u8 mask0, mask1, mask2, mask3, out;
- v8i16 filt, out0, out1, out2, out3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= 3;
-
- /* rearranging filter */
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- for (loop_cnt = height; loop_cnt--;) {- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src3 = LD_SB(src + 24);
- src1 = __msa_sldi_b(src2, src0, 8);
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
- mask3, filt0, filt1, filt2, filt3, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- out = PCKEV_XORI128_UB(out0, out1);
- ST_UB(out, dst);
- out = PCKEV_XORI128_UB(out2, out3);
- ST_UB(out, dst + 16);
-
- src0 = LD_SB(src + 32);
- src2 = LD_SB(src + 48);
- src3 = LD_SB(src + 56);
- src1 = __msa_sldi_b(src2, src0, 8);
- src += src_stride;
-
- XORI_B4_128_SB(src0, src1, src2, src3);
- HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
- mask3, filt0, filt1, filt2, filt3, out0, out1,
- out2, out3);
- SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
- SAT_SH4_SH(out0, out1, out2, out3, 7);
- out = PCKEV_XORI128_UB(out0, out1);
- ST_UB(out, dst + 32);
- out = PCKEV_XORI128_UB(out2, out3);
- ST_UB(out, dst + 48);
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_4x4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, mask;
- v16u8 filt0, vec0, vec1, res0, res1;
- v8u16 vec2, vec3, filt, const255;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
-
- const255 = (v8u16) __msa_ldi_h(255);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, vec2, vec3);
- SRARI_H2_UH(vec2, vec3, FILTER_BITS);
- MIN_UH2_UH(vec2, vec3, const255);
- PCKEV_B2_UB(vec2, vec2, vec3, vec3, res0, res1);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hz_2t_4x8_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16u8 vec0, vec1, vec2, vec3, filt0;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16i8 res0, res1, res2, res3;
- v8u16 vec4, vec5, vec6, vec7, filt, const255;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
-
- const255 = (v8u16) __msa_ldi_h(255);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
- VSHF_B2_UB(src4, src5, src6, src7, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec4, vec5,
- vec6, vec7);
- SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS);
- MIN_UH4_UH(vec4, vec5, vec6, vec7, const255);
- PCKEV_B4_SB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1,
- res2, res3);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
- dst += (4 * dst_stride);
- ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hz_2t_4w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- if (4 == height) {- common_hz_2t_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {- common_hz_2t_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_hz_2t_8x4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16u8 filt0;
- v16i8 src0, src1, src2, src3, mask;
- v8u16 vec0, vec1, vec2, vec3, const255, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
-
- const255 = (v8u16) __msa_ldi_h(255);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- PCKEV_B2_SB(vec1, vec0, vec3, vec2, src0, src1);
- ST8x4_UB(src0, src1, dst, dst_stride);
-}
-
-static void common_hz_2t_8x8mult_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- v16u8 filt0;
- v16i8 src0, src1, src2, src3, mask, out0, out1;
- v8u16 vec0, vec1, vec2, vec3, filt, const255;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
-
- const255 = (v8u16) __msa_ldi_h(255);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
-
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- if (16 == height) {- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- LD_SB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
-
- VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
- vec2, vec3);
- SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
- MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
- PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
- ST8x4_UB(out0, out1, dst + 4 * dst_stride, dst_stride);
- }
-}
-
-static void common_hz_2t_8w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- if (4 == height) {- common_hz_2t_8x4_msa(src, src_stride, dst, dst_stride, filter);
- } else {- common_hz_2t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height);
- }
-}
-
-static void common_hz_2t_16w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- loop_cnt = (height >> 2) - 1;
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
-
- const255 = (v8u16) __msa_ldi_h(255);
-
- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
- out2, out3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
- out6, out7);
- SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
- SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
- MIN_UH4_UH(out0, out1, out2, out3, const255);
- MIN_UH4_UH(out4, out5, out6, out7, const255);
- PCKEV_ST_SB(out0, out1, dst);
- dst += dst_stride;
- PCKEV_ST_SB(out2, out3, dst);
- dst += dst_stride;
- PCKEV_ST_SB(out4, out5, dst);
- dst += dst_stride;
- PCKEV_ST_SB(out6, out7, dst);
- dst += dst_stride;
-
- for (; loop_cnt--;) {- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
- out2, out3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
- out6, out7);
- SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
- SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
- MIN_UH4_UH(out0, out1, out2, out3, const255);
- MIN_UH4_UH(out4, out5, out6, out7, const255);
- PCKEV_ST_SB(out0, out1, dst);
- dst += dst_stride;
- PCKEV_ST_SB(out2, out3, dst);
- dst += dst_stride;
- PCKEV_ST_SB(out4, out5, dst);
- dst += dst_stride;
- PCKEV_ST_SB(out6, out7, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_32w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
-
- const255 = (v8u16) __msa_ldi_h(255);
-
- for (loop_cnt = height >> 1; loop_cnt--;) {- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src3 = LD_SB(src + 24);
- src1 = __msa_sldi_b(src2, src0, 8);
- src += src_stride;
- src4 = LD_SB(src);
- src6 = LD_SB(src + 16);
- src7 = LD_SB(src + 24);
- src5 = __msa_sldi_b(src6, src4, 8);
- src += src_stride;
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
- out2, out3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
- out6, out7);
- SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
- SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
- MIN_UH4_UH(out0, out1, out2, out3, const255);
- MIN_UH4_UH(out4, out5, out6, out7, const255);
- PCKEV_ST_SB(out0, out1, dst);
- PCKEV_ST_SB(out2, out3, dst + 16);
- dst += dst_stride;
- PCKEV_ST_SB(out4, out5, dst);
- PCKEV_ST_SB(out6, out7, dst + 16);
- dst += dst_stride;
- }
-}
-
-static void common_hz_2t_64w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_UH(filter);
- filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
-
- const255 = (v8u16) __msa_ldi_h(255);
-
- for (loop_cnt = height; loop_cnt--;) {- src0 = LD_SB(src);
- src2 = LD_SB(src + 16);
- src4 = LD_SB(src + 32);
- src6 = LD_SB(src + 48);
- src7 = LD_SB(src + 56);
- SLDI_B3_SB(src2, src4, src6, src0, src2, src4, src1, src3, src5, 8);
- src += src_stride;
-
- VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
- VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
- VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
- VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
- out2, out3);
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
- out6, out7);
- SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
- SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
- MIN_UH4_UH(out0, out1, out2, out3, const255);
- MIN_UH4_UH(out4, out5, out6, out7, const255);
- PCKEV_ST_SB(out0, out1, dst);
- PCKEV_ST_SB(out2, out3, dst + 16);
- PCKEV_ST_SB(out4, out5, dst + 32);
- PCKEV_ST_SB(out6, out7, dst + 48);
- dst += dst_stride;
- }
-}
-
-void vp9_convolve8_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- int8_t cnt, filt_hor[8];
-
- if (16 != x_step_q4) {- vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- if (((const int32_t *)filter_x)[1] == 0x800000) {- vp9_convolve_copy(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- for (cnt = 0; cnt < 8; ++cnt) {- filt_hor[cnt] = filter_x[cnt];
- }
-
- if (((const int32_t *)filter_x)[0] == 0) {- switch (w) {- case 4:
- common_hz_2t_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- case 8:
- common_hz_2t_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- case 16:
- common_hz_2t_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- case 32:
- common_hz_2t_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- case 64:
- common_hz_2t_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], h);
- break;
- default:
- vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- } else {- switch (w) {- case 4:
- common_hz_8t_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- case 8:
- common_hz_8t_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- case 16:
- common_hz_8t_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- case 32:
- common_hz_8t_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- case 64:
- common_hz_8t_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, h);
- break;
- default:
- vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- }
-}
--- a/vp9/common/mips/msa/vp9_convolve8_msa.c
+++ /dev/null
@@ -1,654 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/mips/msa/vp9_convolve_msa.h"
-
-const uint8_t mc_filt_mask_arr[16 * 3] = {- /* 8 width cases */
- 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
- /* 4 width cases */
- 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20,
- /* 4 width cases */
- 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28
-};
-
-static void common_hv_8ht_8vt_4w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
- v16u8 mask0, mask1, mask2, mask3, out;
- v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8i16 hz_out7, hz_out8, hz_out9, tmp0, tmp1, out0, out1, out2, out3, out4;
- v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
-
- mask0 = LD_UB(&mc_filt_mask_arr[16]);
- src -= (3 + 3 * src_stride);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- hz_out0 = HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out2 = HORIZ_8TAP_FILT(src2, src3, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out4 = HORIZ_8TAP_FILT(src4, src5, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out5 = HORIZ_8TAP_FILT(src5, src6, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- SLDI_B2_SH(hz_out2, hz_out4, hz_out0, hz_out2, hz_out1, hz_out3, 8);
-
- filt = LD_SH(filter_vert);
- SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
-
- ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1);
- out2 = (v8i16)__msa_ilvev_b((v16i8)hz_out5, (v16i8)hz_out4);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- XORI_B4_128_SB(src7, src8, src9, src10);
- src += (4 * src_stride);
-
- hz_out7 = HORIZ_8TAP_FILT(src7, src8, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- hz_out6 = (v8i16)__msa_sldi_b((v16i8)hz_out7, (v16i8)hz_out5, 8);
- out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
- tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out9 = HORIZ_8TAP_FILT(src9, src10, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- hz_out8 = (v8i16)__msa_sldi_b((v16i8)hz_out9, (v16i8)hz_out7, 8);
- out4 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
- tmp1 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out4, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
- SRARI_H2_SH(tmp0, tmp1, FILTER_BITS);
- SAT_SH2_SH(tmp0, tmp1, 7);
- out = PCKEV_XORI128_UB(tmp0, tmp1);
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
- dst += (4 * dst_stride);
-
- hz_out5 = hz_out9;
- out0 = out2;
- out1 = out3;
- out2 = out4;
- }
-}
-
-static void common_hv_8ht_8vt_8w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
- v16u8 mask0, mask1, mask2, mask3, vec0, vec1;
- v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
- v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8i16 hz_out7, hz_out8, hz_out9, hz_out10, tmp0, tmp1, tmp2, tmp3;
- v8i16 out0, out1, out2, out3, out4, out5, out6, out7, out8, out9;
-
- mask0 = LD_UB(&mc_filt_mask_arr[0]);
- src -= (3 + 3 * src_stride);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
-
- mask1 = mask0 + 2;
- mask2 = mask0 + 4;
- mask3 = mask0 + 6;
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- hz_out0 = HORIZ_8TAP_FILT(src0, src0, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out1 = HORIZ_8TAP_FILT(src1, src1, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out2 = HORIZ_8TAP_FILT(src2, src2, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out3 = HORIZ_8TAP_FILT(src3, src3, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out4 = HORIZ_8TAP_FILT(src4, src4, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out5 = HORIZ_8TAP_FILT(src5, src5, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
- hz_out6 = HORIZ_8TAP_FILT(src6, src6, mask0, mask1, mask2, mask3, filt_hz0,
- filt_hz1, filt_hz2, filt_hz3);
-
- filt = LD_SH(filter_vert);
- SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
-
- ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1);
- ILVEV_B2_SH(hz_out4, hz_out5, hz_out1, hz_out2, out2, out4);
- ILVEV_B2_SH(hz_out3, hz_out4, hz_out5, hz_out6, out5, out6);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- src += (4 * src_stride);
-
- XORI_B4_128_SB(src7, src8, src9, src10);
-
- hz_out7 = HORIZ_8TAP_FILT(src7, src7, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
- tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out8 = HORIZ_8TAP_FILT(src8, src8, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out7 = (v8i16)__msa_ilvev_b((v16i8)hz_out8, (v16i8)hz_out7);
- tmp1 = FILT_8TAP_DPADD_S_H(out4, out5, out6, out7, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out9 = HORIZ_8TAP_FILT(src9, src9, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out8 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
- tmp2 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out8, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
-
- hz_out10 = HORIZ_8TAP_FILT(src10, src10, mask0, mask1, mask2, mask3,
- filt_hz0, filt_hz1, filt_hz2, filt_hz3);
- out9 = (v8i16)__msa_ilvev_b((v16i8)hz_out10, (v16i8)hz_out9);
- tmp3 = FILT_8TAP_DPADD_S_H(out5, out6, out7, out9, filt_vt0, filt_vt1,
- filt_vt2, filt_vt3);
- SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
- vec0 = PCKEV_XORI128_UB(tmp0, tmp1);
- vec1 = PCKEV_XORI128_UB(tmp2, tmp3);
- ST8x4_UB(vec0, vec1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- hz_out6 = hz_out10;
- out0 = out2;
- out1 = out3;
- out2 = out8;
- out4 = out6;
- out5 = out7;
- out6 = out9;
- }
-}
-
-static void common_hv_8ht_8vt_16w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 2; multiple8_cnt--;) {- common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz,
- filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_8ht_8vt_32w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 4; multiple8_cnt--;) {- common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz,
- filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_8ht_8vt_64w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 8; multiple8_cnt--;) {- common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz,
- filter_vert, height);
- src += 8;
- dst += 8;
- }
-}
-
-static void common_hv_2ht_2vt_4x4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert) {- v16i8 src0, src1, src2, src3, src4, mask;
- v16u8 filt_vt, filt_hz, vec0, vec1, res0, res1;
- v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, filt, tmp0, tmp1;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- filt = LD_UH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
- hz_out4 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- hz_out1 = (v8u16)__msa_sldi_b((v16i8)hz_out2, (v16i8)hz_out0, 8);
- hz_out3 = (v8u16)__msa_pckod_d((v2i64)hz_out4, (v2i64)hz_out2);
-
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_B2_UB(tmp0, tmp0, tmp1, tmp1, res0, res1);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hv_2ht_2vt_4x8_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert) {- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, mask;
- v16i8 res0, res1, res2, res3;
- v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3;
- v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
- v8u16 hz_out7, hz_out8, vec4, vec5, vec6, vec7, filt;
-
- mask = LD_SB(&mc_filt_mask_arr[16]);
-
- /* rearranging filter */
- filt = LD_UH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- filt = LD_UH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
- src8 = LD_SB(src);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
- hz_out4 = HORIZ_2TAP_FILT_UH(src4, src5, mask, filt_hz, FILTER_BITS);
- hz_out6 = HORIZ_2TAP_FILT_UH(src6, src7, mask, filt_hz, FILTER_BITS);
- hz_out8 = HORIZ_2TAP_FILT_UH(src8, src8, mask, filt_hz, FILTER_BITS);
- SLDI_B3_UH(hz_out2, hz_out4, hz_out6, hz_out0, hz_out2, hz_out4, hz_out1,
- hz_out3, hz_out5, 8);
- hz_out7 = (v8u16)__msa_pckod_d((v2i64)hz_out8, (v2i64)hz_out6);
-
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- ILVEV_B2_UB(hz_out4, hz_out5, hz_out6, hz_out7, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt_vt, filt_vt, filt_vt, filt_vt,
- vec4, vec5, vec6, vec7);
- SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS);
- SAT_UH4_UH(vec4, vec5, vec6, vec7, 7);
- PCKEV_B4_SB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1,
- res2, res3);
- ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
- dst += (4 * dst_stride);
- ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
-}
-
-static void common_hv_2ht_2vt_4w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- if (4 == height) {- common_hv_2ht_2vt_4x4_msa(src, src_stride, dst, dst_stride, filter_horiz,
- filter_vert);
- } else if (8 == height) {- common_hv_2ht_2vt_4x8_msa(src, src_stride, dst, dst_stride, filter_horiz,
- filter_vert);
- }
-}
-
-static void common_hv_2ht_2vt_8x4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert) {- v16i8 src0, src1, src2, src3, src4, mask, out0, out1;
- v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3;
- v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp0 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- vec1 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp1 = __msa_dotp_u_h(vec1, filt_vt);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- vec2 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp2 = __msa_dotp_u_h(vec2, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- vec3 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp3 = __msa_dotp_u_h(vec3, filt_vt);
-
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
-}
-
-static void common_hv_2ht_2vt_8x8mult_msa(const uint8_t *src,
- int32_t src_stride,
- uint8_t *dst,
- int32_t dst_stride,
- int8_t *filter_horiz,
- int8_t *filter_vert,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, mask, out0, out1;
- v16u8 filt_hz, filt_vt, vec0;
- v8u16 hz_out0, hz_out1, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- src0 = LD_SB(src);
- src += src_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
-
- for (loop_cnt = (height >> 3); loop_cnt--;) {- LD_SB4(src, src_stride, src1, src2, src3, src4);
- src += (4 * src_stride);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp1 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp2 = __msa_dotp_u_h(vec0, filt_vt);
-
- SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
- SAT_UH2_UH(tmp1, tmp2, 7);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp3 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- LD_SB4(src, src_stride, src1, src2, src3, src4);
- src += (4 * src_stride);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp4 = __msa_dotp_u_h(vec0, filt_vt);
-
- SRARI_H2_UH(tmp3, tmp4, FILTER_BITS);
- SAT_UH2_UH(tmp3, tmp4, 7);
- PCKEV_B2_SB(tmp2, tmp1, tmp4, tmp3, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp5 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp6 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
- tmp7 = __msa_dotp_u_h(vec0, filt_vt);
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
- tmp8 = __msa_dotp_u_h(vec0, filt_vt);
-
- SRARI_H4_UH(tmp5, tmp6, tmp7, tmp8, FILTER_BITS);
- SAT_UH4_UH(tmp5, tmp6, tmp7, tmp8, 7);
- PCKEV_B2_SB(tmp6, tmp5, tmp8, tmp7, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void common_hv_2ht_2vt_8w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- if (4 == height) {- common_hv_2ht_2vt_8x4_msa(src, src_stride, dst, dst_stride, filter_horiz,
- filter_vert);
- } else {- common_hv_2ht_2vt_8x8mult_msa(src, src_stride, dst, dst_stride,
- filter_horiz, filter_vert, height);
- }
-}
-
-static void common_hv_2ht_2vt_16w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
- v16u8 filt_hz, filt_vt, vec0, vec1;
- v8u16 tmp1, tmp2, hz_out0, hz_out1, hz_out2, hz_out3;
- v8i16 filt;
-
- mask = LD_SB(&mc_filt_mask_arr[0]);
-
- /* rearranging filter */
- filt = LD_SH(filter_horiz);
- filt_hz = (v16u8)__msa_splati_h(filt, 0);
-
- filt = LD_SH(filter_vert);
- filt_vt = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB2(src, 8, src0, src1);
- src += src_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src0, src2, src4, src6);
- LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
- hz_out3 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
- SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
- SAT_UH2_UH(tmp1, tmp2, 7);
- PCKEV_ST_SB(tmp1, tmp2, dst);
- dst += dst_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
- SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
- SAT_UH2_UH(tmp1, tmp2, 7);
- PCKEV_ST_SB(tmp1, tmp2, dst);
- dst += dst_stride;
-
- hz_out1 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
- hz_out3 = HORIZ_2TAP_FILT_UH(src5, src5, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
- SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
- SAT_UH2_UH(tmp1, tmp2, 7);
- PCKEV_ST_SB(tmp1, tmp2, dst);
- dst += dst_stride;
-
- hz_out0 = HORIZ_2TAP_FILT_UH(src6, src6, mask, filt_hz, FILTER_BITS);
- hz_out2 = HORIZ_2TAP_FILT_UH(src7, src7, mask, filt_hz, FILTER_BITS);
- ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
- DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
- SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
- SAT_UH2_UH(tmp1, tmp2, 7);
- PCKEV_ST_SB(tmp1, tmp2, dst);
- dst += dst_stride;
- }
-}
-
-static void common_hv_2ht_2vt_32w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 2; multiple8_cnt--;) {- common_hv_2ht_2vt_16w_msa(src, src_stride, dst, dst_stride, filter_horiz,
- filter_vert, height);
- src += 16;
- dst += 16;
- }
-}
-
-static void common_hv_2ht_2vt_64w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter_horiz, int8_t *filter_vert,
- int32_t height) {- int32_t multiple8_cnt;
- for (multiple8_cnt = 4; multiple8_cnt--;) {- common_hv_2ht_2vt_16w_msa(src, src_stride, dst, dst_stride, filter_horiz,
- filter_vert, height);
- src += 16;
- dst += 16;
- }
-}
-
-void vp9_convolve8_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int32_t x_step_q4,
- const int16_t *filter_y, int32_t y_step_q4,
- int32_t w, int32_t h) {- int8_t cnt, filt_hor[8], filt_ver[8];
-
- if (16 != x_step_q4 || 16 != y_step_q4) {- vp9_convolve8_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- if (((const int32_t *)filter_x)[1] == 0x800000 &&
- ((const int32_t *)filter_y)[1] == 0x800000) {- vp9_convolve_copy(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- for (cnt = 0; cnt < 8; ++cnt) {- filt_hor[cnt] = filter_x[cnt];
- filt_ver[cnt] = filter_y[cnt];
- }
-
- if (((const int32_t *)filter_x)[0] == 0 &&
- ((const int32_t *)filter_y)[0] == 0) {- switch (w) {- case 4:
- common_hv_2ht_2vt_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], (int32_t)h);
- break;
- case 8:
- common_hv_2ht_2vt_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], (int32_t)h);
- break;
- case 16:
- common_hv_2ht_2vt_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], (int32_t)h);
- break;
- case 32:
- common_hv_2ht_2vt_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], (int32_t)h);
- break;
- case 64:
- common_hv_2ht_2vt_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_hor[3], &filt_ver[3], (int32_t)h);
- break;
- default:
- vp9_convolve8_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- } else if (((const int32_t *)filter_x)[0] == 0 ||
- ((const int32_t *)filter_y)[0] == 0) {- vp9_convolve8_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- } else {- switch (w) {- case 4:
- common_hv_8ht_8vt_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, (int32_t)h);
- break;
- case 8:
- common_hv_8ht_8vt_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, (int32_t)h);
- break;
- case 16:
- common_hv_8ht_8vt_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, (int32_t)h);
- break;
- case 32:
- common_hv_8ht_8vt_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, (int32_t)h);
- break;
- case 64:
- common_hv_8ht_8vt_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_hor, filt_ver, (int32_t)h);
- break;
- default:
- vp9_convolve8_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- }
-}
--- a/vp9/common/mips/msa/vp9_convolve8_vert_msa.c
+++ /dev/null
@@ -1,745 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "vp9/common/mips/msa/vp9_convolve_msa.h"
-
-static void common_vt_8t_4w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776;
- v16i8 src10998, filt0, filt1, filt2, filt3;
- v16u8 out;
- v8i16 filt, out10, out32;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
-
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
- ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src2110,
- src4332, src6554);
- XORI_B3_128_SB(src2110, src4332, src6554);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- src += (4 * src_stride);
-
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- ILVR_D2_SB(src87_r, src76_r, src109_r, src98_r, src8776, src10998);
- XORI_B2_128_SB(src8776, src10998);
- out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, filt0,
- filt1, filt2, filt3);
- out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, filt0,
- filt1, filt2, filt3);
- SRARI_H2_SH(out10, out32, FILTER_BITS);
- SAT_SH2_SH(out10, out32, 7);
- out = PCKEV_XORI128_UB(out10, out32);
- ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
- dst += (4 * dst_stride);
-
- src2110 = src6554;
- src4332 = src8776;
- src6554 = src10998;
- src6 = src10;
- }
-}
-
-static void common_vt_8t_8w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, filt0, filt1, filt2, filt3;
- v16u8 tmp0, tmp1;
- v8i16 filt, out0_r, out1_r, out2_r, out3_r;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- XORI_B4_128_SB(src7, src8, src9, src10);
- src += (4 * src_stride);
-
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
- filt1, filt2, filt3);
- out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
- filt1, filt2, filt3);
- out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
- filt1, filt2, filt3);
- out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
- filt1, filt2, filt3);
- SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
- SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
- tmp0 = PCKEV_XORI128_UB(out0_r, out1_r);
- tmp1 = PCKEV_XORI128_UB(out2_r, out3_r);
- ST8x4_UB(tmp0, tmp1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- src10_r = src54_r;
- src32_r = src76_r;
- src54_r = src98_r;
- src21_r = src65_r;
- src43_r = src87_r;
- src65_r = src109_r;
- src6 = src10;
- }
-}
-
-static void common_vt_8t_16w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 filt0, filt1, filt2, filt3;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l;
- v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l;
- v16u8 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt, out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- src += (7 * src_stride);
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
- src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
- ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l, src32_l,
- src54_l, src21_l);
- ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src, src_stride, src7, src8, src9, src10);
- XORI_B4_128_SB(src7, src8, src9, src10);
- src += (4 * src_stride);
-
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l,
- src87_l, src98_l, src109_l);
- out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
- filt1, filt2, filt3);
- out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
- filt1, filt2, filt3);
- out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
- filt1, filt2, filt3);
- out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
- filt1, filt2, filt3);
- out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0,
- filt1, filt2, filt3);
- out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0,
- filt1, filt2, filt3);
- out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0,
- filt1, filt2, filt3);
- out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0,
- filt1, filt2, filt3);
- SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
- SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS);
- SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
- SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7);
- PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l, out3_r,
- tmp0, tmp1, tmp2, tmp3);
- XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3);
- ST_UB4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride);
- dst += (4 * dst_stride);
-
- src10_r = src54_r;
- src32_r = src76_r;
- src54_r = src98_r;
- src21_r = src65_r;
- src43_r = src87_r;
- src65_r = src109_r;
- src10_l = src54_l;
- src32_l = src76_l;
- src54_l = src98_l;
- src21_l = src65_l;
- src43_l = src87_l;
- src65_l = src109_l;
- src6 = src10;
- }
-}
-
-static void common_vt_8t_16w_mult_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height,
- int32_t width) {- const uint8_t *src_tmp;
- uint8_t *dst_tmp;
- uint32_t loop_cnt, cnt;
- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16i8 filt0, filt1, filt2, filt3;
- v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l;
- v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l;
- v16u8 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt, out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l;
-
- src -= (3 * src_stride);
-
- filt = LD_SH(filter);
- SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
-
- for (cnt = (width >> 4); cnt--;) {- src_tmp = src;
- dst_tmp = dst;
-
- LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
- XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
- src_tmp += (7 * src_stride);
- ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r,
- src32_r, src54_r, src21_r);
- ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
- ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l,
- src32_l, src54_l, src21_l);
- ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_SB4(src_tmp, src_stride, src7, src8, src9, src10);
- XORI_B4_128_SB(src7, src8, src9, src10);
- src_tmp += (4 * src_stride);
- ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
- src87_r, src98_r, src109_r);
- ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l,
- src87_l, src98_l, src109_l);
- out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
- filt1, filt2, filt3);
- out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
- filt1, filt2, filt3);
- out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
- filt1, filt2, filt3);
- out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
- filt1, filt2, filt3);
- out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0,
- filt1, filt2, filt3);
- out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0,
- filt1, filt2, filt3);
- out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0,
- filt1, filt2, filt3);
- out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0,
- filt1, filt2, filt3);
- SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
- SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS);
- SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
- SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7);
- PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l,
- out3_r, tmp0, tmp1, tmp2, tmp3);
- XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3);
- ST_UB4(tmp0, tmp1, tmp2, tmp3, dst_tmp, dst_stride);
- dst_tmp += (4 * dst_stride);
-
- src10_r = src54_r;
- src32_r = src76_r;
- src54_r = src98_r;
- src21_r = src65_r;
- src43_r = src87_r;
- src65_r = src109_r;
- src10_l = src54_l;
- src32_l = src76_l;
- src54_l = src98_l;
- src21_l = src65_l;
- src43_l = src87_l;
- src65_l = src109_l;
- src6 = src10;
- }
-
- src += 16;
- dst += 16;
- }
-}
-
-static void common_vt_8t_32w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, filter, height,
- 32);
-}
-
-static void common_vt_8t_64w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, filter, height,
- 64);
-}
-
-static void common_vt_2t_4x4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, src4;
- v16i8 src10_r, src32_r, src21_r, src43_r, src2110, src4332;
- v16u8 filt0;
- v8i16 filt;
- v8u16 tmp0, tmp1;
-
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
- src += (5 * src_stride);
-
- ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
- src32_r, src43_r);
- ILVR_D2_SB(src21_r, src10_r, src43_r, src32_r, src2110, src4332);
- DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- src2110 = __msa_pckev_b((v16i8)tmp1, (v16i8)tmp0);
- ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
-}
-
-static void common_vt_2t_4x8_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
- v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r;
- v16i8 src65_r, src87_r, src2110, src4332, src6554, src8776;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v16u8 filt0;
- v8i16 filt;
-
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
-
- src8 = LD_SB(src);
- src += src_stride;
-
- ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
- src32_r, src43_r);
- ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r,
- src76_r, src87_r);
- ILVR_D4_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r,
- src87_r, src76_r, src2110, src4332, src6554, src8776);
- DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0,
- tmp0, tmp1, tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, src2110, src4332);
- ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
- ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst + 4 * dst_stride, dst_stride);
-}
-
-static void common_vt_2t_4w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- if (4 == height) {- common_vt_2t_4x4_msa(src, src_stride, dst, dst_stride, filter);
- } else if (8 == height) {- common_vt_2t_4x8_msa(src, src_stride, dst, dst_stride, filter);
- }
-}
-
-static void common_vt_2t_8x4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter) {- v16u8 src0, src1, src2, src3, src4, vec0, vec1, vec2, vec3, filt0;
- v16i8 out0, out1;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1);
- ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
-}
-
-static void common_vt_2t_8x8mult_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
- v16i8 out0, out1;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- src0 = LD_UB(src);
- src += src_stride;
-
- for (loop_cnt = (height >> 3); loop_cnt--;) {- LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8);
- src += (8 * src_stride);
-
- ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3, vec0, vec1,
- vec2, vec3);
- ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7, vec4, vec5,
- vec6, vec7);
- DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, tmp0, tmp1,
- tmp2, tmp3);
- SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
- SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
- PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
- ST8x4_UB(out0, out1, dst, dst_stride);
- dst += (4 * dst_stride);
-
- src0 = src8;
- }
-}
-
-static void common_vt_2t_8w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- if (4 == height) {- common_vt_2t_8x4_msa(src, src_stride, dst, dst_stride, filter);
- } else {- common_vt_2t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height);
- }
-}
-
-static void common_vt_2t_16w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- src0 = LD_UB(src);
- src += src_stride;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_UB4(src, src_stride, src1, src2, src3, src4);
- src += (4 * src_stride);
-
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_ST_SB(tmp0, tmp1, dst);
- dst += dst_stride;
-
- ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
- ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_ST_SB(tmp2, tmp3, dst);
- dst += dst_stride;
-
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_ST_SB(tmp0, tmp1, dst);
- dst += dst_stride;
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_ST_SB(tmp2, tmp3, dst);
- dst += dst_stride;
-
- src0 = src4;
- }
-}
-
-static void common_vt_2t_32w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
- v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- src0 = LD_UB(src);
- src5 = LD_UB(src + 16);
- src += src_stride;
-
- for (loop_cnt = (height >> 2); loop_cnt--;) {- LD_UB4(src, src_stride, src1, src2, src3, src4);
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
-
- LD_UB4(src + 16, src_stride, src6, src7, src8, src9);
- src += (4 * src_stride);
-
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_ST_SB(tmp0, tmp1, dst);
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_ST_SB(tmp2, tmp3, dst + dst_stride);
-
- ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
- ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_ST_SB(tmp0, tmp1, dst + 2 * dst_stride);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_ST_SB(tmp2, tmp3, dst + 3 * dst_stride);
-
- ILVR_B2_UB(src6, src5, src7, src6, vec0, vec2);
- ILVL_B2_UB(src6, src5, src7, src6, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_ST_SB(tmp0, tmp1, dst + 16);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_ST_SB(tmp2, tmp3, dst + 16 + dst_stride);
-
- ILVR_B2_UB(src8, src7, src9, src8, vec4, vec6);
- ILVL_B2_UB(src8, src7, src9, src8, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_ST_SB(tmp0, tmp1, dst + 16 + 2 * dst_stride);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_ST_SB(tmp2, tmp3, dst + 16 + 3 * dst_stride);
- dst += (4 * dst_stride);
-
- src0 = src4;
- src5 = src9;
- }
-}
-
-static void common_vt_2t_64w_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int8_t *filter, int32_t height) {- uint32_t loop_cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
- v16u8 src11, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
- v8u16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
- v8i16 filt;
-
- /* rearranging filter_y */
- filt = LD_SH(filter);
- filt0 = (v16u8)__msa_splati_h(filt, 0);
-
- LD_UB4(src, 16, src0, src3, src6, src9);
- src += src_stride;
-
- for (loop_cnt = (height >> 1); loop_cnt--;) {- LD_UB2(src, src_stride, src1, src2);
- LD_UB2(src + 16, src_stride, src4, src5);
- LD_UB2(src + 32, src_stride, src7, src8);
- LD_UB2(src + 48, src_stride, src10, src11);
- src += (2 * src_stride);
-
- ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
- ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_ST_SB(tmp0, tmp1, dst);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_ST_SB(tmp2, tmp3, dst + dst_stride);
-
- ILVR_B2_UB(src4, src3, src5, src4, vec4, vec6);
- ILVL_B2_UB(src4, src3, src5, src4, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
- SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
- SAT_UH2_UH(tmp4, tmp5, 7);
- PCKEV_ST_SB(tmp4, tmp5, dst + 16);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
- SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
- SAT_UH2_UH(tmp6, tmp7, 7);
- PCKEV_ST_SB(tmp6, tmp7, dst + 16 + dst_stride);
-
- ILVR_B2_UB(src7, src6, src8, src7, vec0, vec2);
- ILVL_B2_UB(src7, src6, src8, src7, vec1, vec3);
- DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
- SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
- SAT_UH2_UH(tmp0, tmp1, 7);
- PCKEV_ST_SB(tmp0, tmp1, dst + 32);
-
- DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
- SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
- SAT_UH2_UH(tmp2, tmp3, 7);
- PCKEV_ST_SB(tmp2, tmp3, dst + 32 + dst_stride);
-
- ILVR_B2_UB(src10, src9, src11, src10, vec4, vec6);
- ILVL_B2_UB(src10, src9, src11, src10, vec5, vec7);
- DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
- SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
- SAT_UH2_UH(tmp4, tmp5, 7);
- PCKEV_ST_SB(tmp4, tmp5, dst + 48);
-
- DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
- SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
- SAT_UH2_UH(tmp6, tmp7, 7);
- PCKEV_ST_SB(tmp6, tmp7, dst + 48 + dst_stride);
- dst += (2 * dst_stride);
-
- src0 = src2;
- src3 = src5;
- src6 = src8;
- src9 = src11;
- }
-}
-
-void vp9_convolve8_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- int8_t cnt, filt_ver[8];
-
- if (16 != y_step_q4) {- vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- if (((const int32_t *)filter_y)[1] == 0x800000) {- vp9_convolve_copy(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- return;
- }
-
- for (cnt = 8; cnt--;) {- filt_ver[cnt] = filter_y[cnt];
- }
-
- if (((const int32_t *)filter_y)[0] == 0) {- switch (w) {- case 4:
- common_vt_2t_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- case 8:
- common_vt_2t_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- case 16:
- common_vt_2t_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- case 32:
- common_vt_2t_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- case 64:
- common_vt_2t_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- &filt_ver[3], h);
- break;
- default:
- vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- } else {- switch (w) {- case 4:
- common_vt_8t_4w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- case 8:
- common_vt_8t_8w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- case 16:
- common_vt_8t_16w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- case 32:
- common_vt_8t_32w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- case 64:
- common_vt_8t_64w_msa(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filt_ver, h);
- break;
- default:
- vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
- filter_x, x_step_q4, filter_y, y_step_q4,
- w, h);
- break;
- }
- }
-}
--- a/vp9/common/mips/msa/vp9_convolve_avg_msa.c
+++ /dev/null
@@ -1,232 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "vpx_dsp/mips/macros_msa.h"
-
-static void avg_width4_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- int32_t cnt;
- uint32_t out0, out1, out2, out3;
- v16u8 src0, src1, src2, src3;
- v16u8 dst0, dst1, dst2, dst3;
-
- if (0 == (height % 4)) {- for (cnt = (height / 4); cnt--;) {- LD_UB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-
- AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
- dst0, dst1, dst2, dst3);
-
- out0 = __msa_copy_u_w((v4i32)dst0, 0);
- out1 = __msa_copy_u_w((v4i32)dst1, 0);
- out2 = __msa_copy_u_w((v4i32)dst2, 0);
- out3 = __msa_copy_u_w((v4i32)dst3, 0);
- SW4(out0, out1, out2, out3, dst, dst_stride);
- dst += (4 * dst_stride);
- }
- } else if (0 == (height % 2)) {- for (cnt = (height / 2); cnt--;) {- LD_UB2(src, src_stride, src0, src1);
- src += (2 * src_stride);
-
- LD_UB2(dst, dst_stride, dst0, dst1);
-
- AVER_UB2_UB(src0, dst0, src1, dst1, dst0, dst1);
-
- out0 = __msa_copy_u_w((v4i32)dst0, 0);
- out1 = __msa_copy_u_w((v4i32)dst1, 0);
- SW(out0, dst);
- dst += dst_stride;
- SW(out1, dst);
- dst += dst_stride;
- }
- }
-}
-
-static void avg_width8_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- int32_t cnt;
- uint64_t out0, out1, out2, out3;
- v16u8 src0, src1, src2, src3;
- v16u8 dst0, dst1, dst2, dst3;
-
- for (cnt = (height / 4); cnt--;) {- LD_UB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
- LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
-
- AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
- dst0, dst1, dst2, dst3);
-
- out0 = __msa_copy_u_d((v2i64)dst0, 0);
- out1 = __msa_copy_u_d((v2i64)dst1, 0);
- out2 = __msa_copy_u_d((v2i64)dst2, 0);
- out3 = __msa_copy_u_d((v2i64)dst3, 0);
- SD4(out0, out1, out2, out3, dst, dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void avg_width16_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- int32_t cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
-
- for (cnt = (height / 8); cnt--;) {- LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
- LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
-
- AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
- dst0, dst1, dst2, dst3);
- AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7,
- dst4, dst5, dst6, dst7);
- ST_UB8(dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst, dst_stride);
- dst += (8 * dst_stride);
- }
-}
-
-static void avg_width32_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- int32_t cnt;
- uint8_t *dst_dup = dst;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
- v16u8 src8, src9, src10, src11, src12, src13, src14, src15;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
-
- for (cnt = (height / 8); cnt--;) {- LD_UB4(src, src_stride, src0, src2, src4, src6);
- LD_UB4(src + 16, src_stride, src1, src3, src5, src7);
- src += (4 * src_stride);
- LD_UB4(dst_dup, dst_stride, dst0, dst2, dst4, dst6);
- LD_UB4(dst_dup + 16, dst_stride, dst1, dst3, dst5, dst7);
- dst_dup += (4 * dst_stride);
- LD_UB4(src, src_stride, src8, src10, src12, src14);
- LD_UB4(src + 16, src_stride, src9, src11, src13, src15);
- src += (4 * src_stride);
- LD_UB4(dst_dup, dst_stride, dst8, dst10, dst12, dst14);
- LD_UB4(dst_dup + 16, dst_stride, dst9, dst11, dst13, dst15);
- dst_dup += (4 * dst_stride);
-
- AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
- dst0, dst1, dst2, dst3);
- AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7,
- dst4, dst5, dst6, dst7);
- AVER_UB4_UB(src8, dst8, src9, dst9, src10, dst10, src11, dst11,
- dst8, dst9, dst10, dst11);
- AVER_UB4_UB(src12, dst12, src13, dst13, src14, dst14, src15, dst15,
- dst12, dst13, dst14, dst15);
-
- ST_UB4(dst0, dst2, dst4, dst6, dst, dst_stride);
- ST_UB4(dst1, dst3, dst5, dst7, dst + 16, dst_stride);
- dst += (4 * dst_stride);
- ST_UB4(dst8, dst10, dst12, dst14, dst, dst_stride);
- ST_UB4(dst9, dst11, dst13, dst15, dst + 16, dst_stride);
- dst += (4 * dst_stride);
- }
-}
-
-static void avg_width64_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- int32_t cnt;
- uint8_t *dst_dup = dst;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
- v16u8 src8, src9, src10, src11, src12, src13, src14, src15;
- v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
- v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
-
- for (cnt = (height / 4); cnt--;) {- LD_UB4(src, 16, src0, src1, src2, src3);
- src += src_stride;
- LD_UB4(src, 16, src4, src5, src6, src7);
- src += src_stride;
- LD_UB4(src, 16, src8, src9, src10, src11);
- src += src_stride;
- LD_UB4(src, 16, src12, src13, src14, src15);
- src += src_stride;
-
- LD_UB4(dst_dup, 16, dst0, dst1, dst2, dst3);
- dst_dup += dst_stride;
- LD_UB4(dst_dup, 16, dst4, dst5, dst6, dst7);
- dst_dup += dst_stride;
- LD_UB4(dst_dup, 16, dst8, dst9, dst10, dst11);
- dst_dup += dst_stride;
- LD_UB4(dst_dup, 16, dst12, dst13, dst14, dst15);
- dst_dup += dst_stride;
-
- AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
- dst0, dst1, dst2, dst3);
- AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7,
- dst4, dst5, dst6, dst7);
- AVER_UB4_UB(src8, dst8, src9, dst9, src10, dst10, src11, dst11,
- dst8, dst9, dst10, dst11);
- AVER_UB4_UB(src12, dst12, src13, dst13, src14, dst14, src15, dst15,
- dst12, dst13, dst14, dst15);
-
- ST_UB4(dst0, dst1, dst2, dst3, dst, 16);
- dst += dst_stride;
- ST_UB4(dst4, dst5, dst6, dst7, dst, 16);
- dst += dst_stride;
- ST_UB4(dst8, dst9, dst10, dst11, dst, 16);
- dst += dst_stride;
- ST_UB4(dst12, dst13, dst14, dst15, dst, 16);
- dst += dst_stride;
- }
-}
-
-void vp9_convolve_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int32_t filter_x_stride,
- const int16_t *filter_y, int32_t filter_y_stride,
- int32_t w, int32_t h) {- (void)filter_x;
- (void)filter_y;
- (void)filter_x_stride;
- (void)filter_y_stride;
-
- switch (w) {- case 4: {- avg_width4_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- case 8: {- avg_width8_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- case 16: {- avg_width16_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- case 32: {- avg_width32_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- case 64: {- avg_width64_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- default: {- int32_t lp, cnt;
- for (cnt = h; cnt--;) {- for (lp = 0; lp < w; ++lp) {- dst[lp] = (((dst[lp] + src[lp]) + 1) >> 1);
- }
- src += src_stride;
- dst += dst_stride;
- }
- break;
- }
- }
-}
--- a/vp9/common/mips/msa/vp9_convolve_copy_msa.c
+++ /dev/null
@@ -1,247 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <string.h>
-#include "vpx_dsp/mips/macros_msa.h"
-
-static void copy_width8_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- int32_t cnt;
- uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
-
- if (0 == height % 12) {- for (cnt = (height / 12); cnt--;) {- LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
-
- out0 = __msa_copy_u_d((v2i64)src0, 0);
- out1 = __msa_copy_u_d((v2i64)src1, 0);
- out2 = __msa_copy_u_d((v2i64)src2, 0);
- out3 = __msa_copy_u_d((v2i64)src3, 0);
- out4 = __msa_copy_u_d((v2i64)src4, 0);
- out5 = __msa_copy_u_d((v2i64)src5, 0);
- out6 = __msa_copy_u_d((v2i64)src6, 0);
- out7 = __msa_copy_u_d((v2i64)src7, 0);
-
- SD4(out0, out1, out2, out3, dst, dst_stride);
- dst += (4 * dst_stride);
- SD4(out4, out5, out6, out7, dst, dst_stride);
- dst += (4 * dst_stride);
-
- LD_UB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- out0 = __msa_copy_u_d((v2i64)src0, 0);
- out1 = __msa_copy_u_d((v2i64)src1, 0);
- out2 = __msa_copy_u_d((v2i64)src2, 0);
- out3 = __msa_copy_u_d((v2i64)src3, 0);
- SD4(out0, out1, out2, out3, dst, dst_stride);
- dst += (4 * dst_stride);
- }
- } else if (0 == height % 8) {- for (cnt = height >> 3; cnt--;) {- LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
-
- out0 = __msa_copy_u_d((v2i64)src0, 0);
- out1 = __msa_copy_u_d((v2i64)src1, 0);
- out2 = __msa_copy_u_d((v2i64)src2, 0);
- out3 = __msa_copy_u_d((v2i64)src3, 0);
- out4 = __msa_copy_u_d((v2i64)src4, 0);
- out5 = __msa_copy_u_d((v2i64)src5, 0);
- out6 = __msa_copy_u_d((v2i64)src6, 0);
- out7 = __msa_copy_u_d((v2i64)src7, 0);
-
- SD4(out0, out1, out2, out3, dst, dst_stride);
- dst += (4 * dst_stride);
- SD4(out4, out5, out6, out7, dst, dst_stride);
- dst += (4 * dst_stride);
- }
- } else if (0 == height % 4) {- for (cnt = (height / 4); cnt--;) {- LD_UB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
- out0 = __msa_copy_u_d((v2i64)src0, 0);
- out1 = __msa_copy_u_d((v2i64)src1, 0);
- out2 = __msa_copy_u_d((v2i64)src2, 0);
- out3 = __msa_copy_u_d((v2i64)src3, 0);
-
- SD4(out0, out1, out2, out3, dst, dst_stride);
- dst += (4 * dst_stride);
- }
- } else if (0 == height % 2) {- for (cnt = (height / 2); cnt--;) {- LD_UB2(src, src_stride, src0, src1);
- src += (2 * src_stride);
- out0 = __msa_copy_u_d((v2i64)src0, 0);
- out1 = __msa_copy_u_d((v2i64)src1, 0);
-
- SD(out0, dst);
- dst += dst_stride;
- SD(out1, dst);
- dst += dst_stride;
- }
- }
-}
-
-static void copy_16multx8mult_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride,
- int32_t height, int32_t width) {- int32_t cnt, loop_cnt;
- const uint8_t *src_tmp;
- uint8_t *dst_tmp;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
-
- for (cnt = (width >> 4); cnt--;) {- src_tmp = src;
- dst_tmp = dst;
-
- for (loop_cnt = (height >> 3); loop_cnt--;) {- LD_UB8(src_tmp, src_stride,
- src0, src1, src2, src3, src4, src5, src6, src7);
- src_tmp += (8 * src_stride);
-
- ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7,
- dst_tmp, dst_stride);
- dst_tmp += (8 * dst_stride);
- }
-
- src += 16;
- dst += 16;
- }
-}
-
-static void copy_width16_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- int32_t cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
-
- if (0 == height % 12) {- for (cnt = (height / 12); cnt--;) {- LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
- src += (8 * src_stride);
- ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, dst, dst_stride);
- dst += (8 * dst_stride);
-
- LD_UB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
- ST_UB4(src0, src1, src2, src3, dst, dst_stride);
- dst += (4 * dst_stride);
- }
- } else if (0 == height % 8) {- copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 16);
- } else if (0 == height % 4) {- for (cnt = (height >> 2); cnt--;) {- LD_UB4(src, src_stride, src0, src1, src2, src3);
- src += (4 * src_stride);
-
- ST_UB4(src0, src1, src2, src3, dst, dst_stride);
- dst += (4 * dst_stride);
- }
- }
-}
-
-static void copy_width32_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- int32_t cnt;
- v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
-
- if (0 == height % 12) {- for (cnt = (height / 12); cnt--;) {- LD_UB4(src, src_stride, src0, src1, src2, src3);
- LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
- src += (4 * src_stride);
- ST_UB4(src0, src1, src2, src3, dst, dst_stride);
- ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
- dst += (4 * dst_stride);
-
- LD_UB4(src, src_stride, src0, src1, src2, src3);
- LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
- src += (4 * src_stride);
- ST_UB4(src0, src1, src2, src3, dst, dst_stride);
- ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
- dst += (4 * dst_stride);
-
- LD_UB4(src, src_stride, src0, src1, src2, src3);
- LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
- src += (4 * src_stride);
- ST_UB4(src0, src1, src2, src3, dst, dst_stride);
- ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
- dst += (4 * dst_stride);
- }
- } else if (0 == height % 8) {- copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 32);
- } else if (0 == height % 4) {- for (cnt = (height >> 2); cnt--;) {- LD_UB4(src, src_stride, src0, src1, src2, src3);
- LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
- src += (4 * src_stride);
- ST_UB4(src0, src1, src2, src3, dst, dst_stride);
- ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
- dst += (4 * dst_stride);
- }
- }
-}
-
-static void copy_width64_msa(const uint8_t *src, int32_t src_stride,
- uint8_t *dst, int32_t dst_stride, int32_t height) {- copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 64);
-}
-
-void vp9_convolve_copy_msa(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int32_t filter_x_stride,
- const int16_t *filter_y, int32_t filter_y_stride,
- int32_t w, int32_t h) {- (void)filter_x;
- (void)filter_y;
- (void)filter_x_stride;
- (void)filter_y_stride;
-
- switch (w) {- case 4: {- uint32_t cnt, tmp;
- /* 1 word storage */
- for (cnt = h; cnt--;) {- tmp = LW(src);
- SW(tmp, dst);
- src += src_stride;
- dst += dst_stride;
- }
- break;
- }
- case 8: {- copy_width8_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- case 16: {- copy_width16_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- case 32: {- copy_width32_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- case 64: {- copy_width64_msa(src, src_stride, dst, dst_stride, h);
- break;
- }
- default: {- uint32_t cnt;
- for (cnt = h; cnt--;) {- memcpy(dst, src, w);
- src += src_stride;
- dst += dst_stride;
- }
- break;
- }
- }
-}
--- a/vp9/common/mips/msa/vp9_convolve_msa.h
+++ /dev/null
@@ -1,119 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#ifndef VP9_COMMON_MIPS_MSA_VP9_CONVOLVE_MSA_H_
-#define VP9_COMMON_MIPS_MSA_VP9_CONVOLVE_MSA_H_
-
-#include "vp9/common/vp9_filter.h"
-#include "vpx_dsp/mips/macros_msa.h"
-
-extern const uint8_t mc_filt_mask_arr[16 * 3];
-
-#define FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, \
- filt0, filt1, filt2, filt3) ({ \- v8i16 tmp0, tmp1; \
- \
- tmp0 = __msa_dotp_s_h((v16i8)vec0, (v16i8)filt0); \
- tmp0 = __msa_dpadd_s_h(tmp0, (v16i8)vec1, (v16i8)filt1); \
- tmp1 = __msa_dotp_s_h((v16i8)vec2, (v16i8)filt2); \
- tmp1 = __msa_dpadd_s_h(tmp1, (v16i8)vec3, (v16i8)filt3); \
- tmp0 = __msa_adds_s_h(tmp0, tmp1); \
- \
- tmp0; \
-})
-
-#define HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, \
- filt_h0, filt_h1, filt_h2, filt_h3) ({ \- v16i8 vec0_m, vec1_m, vec2_m, vec3_m; \
- v8i16 hz_out_m; \
- \
- VSHF_B4_SB(src0, src1, mask0, mask1, mask2, mask3, \
- vec0_m, vec1_m, vec2_m, vec3_m); \
- hz_out_m = FILT_8TAP_DPADD_S_H(vec0_m, vec1_m, vec2_m, vec3_m, \
- filt_h0, filt_h1, filt_h2, filt_h3); \
- \
- hz_out_m = __msa_srari_h(hz_out_m, FILTER_BITS); \
- hz_out_m = __msa_sat_s_h(hz_out_m, 7); \
- \
- hz_out_m; \
-})
-
-#define HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, \
- mask0, mask1, mask2, mask3, \
- filt0, filt1, filt2, filt3, \
- out0, out1) { \- v16i8 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \
- v8i16 res0_m, res1_m, res2_m, res3_m; \
- \
- VSHF_B2_SB(src0, src1, src2, src3, mask0, mask0, vec0_m, vec1_m); \
- DOTP_SB2_SH(vec0_m, vec1_m, filt0, filt0, res0_m, res1_m); \
- VSHF_B2_SB(src0, src1, src2, src3, mask1, mask1, vec2_m, vec3_m); \
- DPADD_SB2_SH(vec2_m, vec3_m, filt1, filt1, res0_m, res1_m); \
- VSHF_B2_SB(src0, src1, src2, src3, mask2, mask2, vec4_m, vec5_m); \
- DOTP_SB2_SH(vec4_m, vec5_m, filt2, filt2, res2_m, res3_m); \
- VSHF_B2_SB(src0, src1, src2, src3, mask3, mask3, vec6_m, vec7_m); \
- DPADD_SB2_SH(vec6_m, vec7_m, filt3, filt3, res2_m, res3_m); \
- ADDS_SH2_SH(res0_m, res2_m, res1_m, res3_m, out0, out1); \
-}
-
-#define HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, \
- mask0, mask1, mask2, mask3, \
- filt0, filt1, filt2, filt3, \
- out0, out1, out2, out3) { \- v16i8 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \
- v8i16 res0_m, res1_m, res2_m, res3_m, res4_m, res5_m, res6_m, res7_m; \
- \
- VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0_m, vec1_m); \
- VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2_m, vec3_m); \
- DOTP_SB4_SH(vec0_m, vec1_m, vec2_m, vec3_m, filt0, filt0, filt0, filt0, \
- res0_m, res1_m, res2_m, res3_m); \
- VSHF_B2_SB(src0, src0, src1, src1, mask2, mask2, vec0_m, vec1_m); \
- VSHF_B2_SB(src2, src2, src3, src3, mask2, mask2, vec2_m, vec3_m); \
- DOTP_SB4_SH(vec0_m, vec1_m, vec2_m, vec3_m, filt2, filt2, filt2, filt2, \
- res4_m, res5_m, res6_m, res7_m); \
- VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec4_m, vec5_m); \
- VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec6_m, vec7_m); \
- DPADD_SB4_SH(vec4_m, vec5_m, vec6_m, vec7_m, filt1, filt1, filt1, filt1, \
- res0_m, res1_m, res2_m, res3_m); \
- VSHF_B2_SB(src0, src0, src1, src1, mask3, mask3, vec4_m, vec5_m); \
- VSHF_B2_SB(src2, src2, src3, src3, mask3, mask3, vec6_m, vec7_m); \
- DPADD_SB4_SH(vec4_m, vec5_m, vec6_m, vec7_m, filt3, filt3, filt3, filt3, \
- res4_m, res5_m, res6_m, res7_m); \
- ADDS_SH4_SH(res0_m, res4_m, res1_m, res5_m, res2_m, res6_m, res3_m, \
- res7_m, out0, out1, out2, out3); \
-}
-
-#define PCKEV_XORI128_AVG_ST_UB(in0, in1, dst, pdst) { \- v16u8 tmp_m; \
- \
- tmp_m = PCKEV_XORI128_UB(in1, in0); \
- tmp_m = __msa_aver_u_b(tmp_m, (v16u8)dst); \
- ST_UB(tmp_m, (pdst)); \
-}
-
-#define PCKEV_AVG_ST_UB(in0, in1, dst, pdst) { \- v16u8 tmp_m; \
- \
- tmp_m = (v16u8)__msa_pckev_b((v16i8)in0, (v16i8)in1); \
- tmp_m = __msa_aver_u_b(tmp_m, (v16u8)dst); \
- ST_UB(tmp_m, (pdst)); \
-}
-
-#define PCKEV_AVG_ST8x4_UB(in1, dst0, in2, dst1, in3, dst2, in4, dst3, \
- pdst, stride) { \- v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
- uint8_t *pdst_m = (uint8_t *)(pdst); \
- \
- PCKEV_B2_UB(in2, in1, in4, in3, tmp0_m, tmp1_m); \
- PCKEV_D2_UB(dst1, dst0, dst3, dst2, tmp2_m, tmp3_m); \
- AVER_UB2_UB(tmp0_m, tmp2_m, tmp1_m, tmp3_m, tmp0_m, tmp1_m); \
- ST8x4_UB(tmp0_m, tmp1_m, pdst_m, stride); \
-}
-#endif /* VP9_COMMON_MIPS_MSA_VP9_CONVOLVE_MSA_H_ */
--- a/vp9/common/vp9_convolve.c
+++ /dev/null
@@ -1,557 +1,0 @@
-/*
- * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include <assert.h>
-
-#include "./vpx_config.h"
-#include "./vp9_rtcd.h"
-#include "vp9/common/vp9_common.h"
-#include "vp9/common/vp9_convolve.h"
-#include "vp9/common/vp9_filter.h"
-#include "vpx/vpx_integer.h"
-#include "vpx_ports/mem.h"
-
-static void convolve_horiz(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const InterpKernel *x_filters,
- int x0_q4, int x_step_q4, int w, int h) {- int x, y;
- src -= SUBPEL_TAPS / 2 - 1;
- for (y = 0; y < h; ++y) {- int x_q4 = x0_q4;
- for (x = 0; x < w; ++x) {- const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
- const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
- int k, sum = 0;
- for (k = 0; k < SUBPEL_TAPS; ++k)
- sum += src_x[k] * x_filter[k];
- dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
- x_q4 += x_step_q4;
- }
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const InterpKernel *x_filters,
- int x0_q4, int x_step_q4, int w, int h) {- int x, y;
- src -= SUBPEL_TAPS / 2 - 1;
- for (y = 0; y < h; ++y) {- int x_q4 = x0_q4;
- for (x = 0; x < w; ++x) {- const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
- const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
- int k, sum = 0;
- for (k = 0; k < SUBPEL_TAPS; ++k)
- sum += src_x[k] * x_filter[k];
- dst[x] = ROUND_POWER_OF_TWO(dst[x] +
- clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
- x_q4 += x_step_q4;
- }
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const InterpKernel *y_filters,
- int y0_q4, int y_step_q4, int w, int h) {- int x, y;
- src -= src_stride * (SUBPEL_TAPS / 2 - 1);
-
- for (x = 0; x < w; ++x) {- int y_q4 = y0_q4;
- for (y = 0; y < h; ++y) {- const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
- const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
- int k, sum = 0;
- for (k = 0; k < SUBPEL_TAPS; ++k)
- sum += src_y[k * src_stride] * y_filter[k];
- dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
- y_q4 += y_step_q4;
- }
- ++src;
- ++dst;
- }
-}
-
-static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const InterpKernel *y_filters,
- int y0_q4, int y_step_q4, int w, int h) {- int x, y;
- src -= src_stride * (SUBPEL_TAPS / 2 - 1);
-
- for (x = 0; x < w; ++x) {- int y_q4 = y0_q4;
- for (y = 0; y < h; ++y) {- const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
- const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
- int k, sum = 0;
- for (k = 0; k < SUBPEL_TAPS; ++k)
- sum += src_y[k * src_stride] * y_filter[k];
- dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] +
- clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
- y_q4 += y_step_q4;
- }
- ++src;
- ++dst;
- }
-}
-
-static void convolve(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const InterpKernel *const x_filters,
- int x0_q4, int x_step_q4,
- const InterpKernel *const y_filters,
- int y0_q4, int y_step_q4,
- int w, int h) {- // Note: Fixed size intermediate buffer, temp, places limits on parameters.
- // 2d filtering proceeds in 2 steps:
- // (1) Interpolate horizontally into an intermediate buffer, temp.
- // (2) Interpolate temp vertically to derive the sub-pixel result.
- // Deriving the maximum number of rows in the temp buffer (135):
- // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
- // --Largest block size is 64x64 pixels.
- // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
- // original frame (in 1/16th pixel units).
- // --Must round-up because block may be located at sub-pixel position.
- // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
- // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
- uint8_t temp[135 * 64];
- int intermediate_height =
- (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
-
- assert(w <= 64);
- assert(h <= 64);
- assert(y_step_q4 <= 32);
- assert(x_step_q4 <= 32);
-
- convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp, 64,
- x_filters, x0_q4, x_step_q4, w, intermediate_height);
- convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride,
- y_filters, y0_q4, y_step_q4, w, h);
-}
-
-static const InterpKernel *get_filter_base(const int16_t *filter) {- // NOTE: This assumes that the filter table is 256-byte aligned.
- // TODO(agrange) Modify to make independent of table alignment.
- return (const InterpKernel *)(((intptr_t)filter) & ~((intptr_t)0xFF));
-}
-
-static int get_filter_offset(const int16_t *f, const InterpKernel *base) {- return (int)((const InterpKernel *)(intptr_t)f - base);
-}
-
-void vp9_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- const InterpKernel *const filters_x = get_filter_base(filter_x);
- const int x0_q4 = get_filter_offset(filter_x, filters_x);
-
- (void)filter_y;
- (void)y_step_q4;
-
- convolve_horiz(src, src_stride, dst, dst_stride, filters_x,
- x0_q4, x_step_q4, w, h);
-}
-
-void vp9_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- const InterpKernel *const filters_x = get_filter_base(filter_x);
- const int x0_q4 = get_filter_offset(filter_x, filters_x);
-
- (void)filter_y;
- (void)y_step_q4;
-
- convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x,
- x0_q4, x_step_q4, w, h);
-}
-
-void vp9_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- const InterpKernel *const filters_y = get_filter_base(filter_y);
- const int y0_q4 = get_filter_offset(filter_y, filters_y);
-
- (void)filter_x;
- (void)x_step_q4;
-
- convolve_vert(src, src_stride, dst, dst_stride, filters_y,
- y0_q4, y_step_q4, w, h);
-}
-
-void vp9_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- const InterpKernel *const filters_y = get_filter_base(filter_y);
- const int y0_q4 = get_filter_offset(filter_y, filters_y);
-
- (void)filter_x;
- (void)x_step_q4;
-
- convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y,
- y0_q4, y_step_q4, w, h);
-}
-
-void vp9_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- const InterpKernel *const filters_x = get_filter_base(filter_x);
- const int x0_q4 = get_filter_offset(filter_x, filters_x);
-
- const InterpKernel *const filters_y = get_filter_base(filter_y);
- const int y0_q4 = get_filter_offset(filter_y, filters_y);
-
- convolve(src, src_stride, dst, dst_stride,
- filters_x, x0_q4, x_step_q4,
- filters_y, y0_q4, y_step_q4, w, h);
-}
-
-void vp9_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h) {- /* Fixed size intermediate buffer places limits on parameters. */
- DECLARE_ALIGNED(16, uint8_t, temp[64 * 64]);
- assert(w <= 64);
- assert(h <= 64);
-
- vp9_convolve8_c(src, src_stride, temp, 64,
- filter_x, x_step_q4, filter_y, y_step_q4, w, h);
- vp9_convolve_avg_c(temp, 64, dst, dst_stride, NULL, 0, NULL, 0, w, h);
-}
-
-void vp9_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int filter_x_stride,
- const int16_t *filter_y, int filter_y_stride,
- int w, int h) {- int r;
-
- (void)filter_x; (void)filter_x_stride;
- (void)filter_y; (void)filter_y_stride;
-
- for (r = h; r > 0; --r) {- memcpy(dst, src, w);
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-void vp9_convolve_avg_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int filter_x_stride,
- const int16_t *filter_y, int filter_y_stride,
- int w, int h) {- int x, y;
-
- (void)filter_x; (void)filter_x_stride;
- (void)filter_y; (void)filter_y_stride;
-
- for (y = 0; y < h; ++y) {- for (x = 0; x < w; ++x)
- dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
-
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static void highbd_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
- uint8_t *dst8, ptrdiff_t dst_stride,
- const InterpKernel *x_filters,
- int x0_q4, int x_step_q4,
- int w, int h, int bd) {- int x, y;
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
- src -= SUBPEL_TAPS / 2 - 1;
- for (y = 0; y < h; ++y) {- int x_q4 = x0_q4;
- for (x = 0; x < w; ++x) {- const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
- const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
- int k, sum = 0;
- for (k = 0; k < SUBPEL_TAPS; ++k)
- sum += src_x[k] * x_filter[k];
- dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
- x_q4 += x_step_q4;
- }
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void highbd_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride,
- uint8_t *dst8, ptrdiff_t dst_stride,
- const InterpKernel *x_filters,
- int x0_q4, int x_step_q4,
- int w, int h, int bd) {- int x, y;
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
- src -= SUBPEL_TAPS / 2 - 1;
- for (y = 0; y < h; ++y) {- int x_q4 = x0_q4;
- for (x = 0; x < w; ++x) {- const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
- const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
- int k, sum = 0;
- for (k = 0; k < SUBPEL_TAPS; ++k)
- sum += src_x[k] * x_filter[k];
- dst[x] = ROUND_POWER_OF_TWO(dst[x] +
- clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
- x_q4 += x_step_q4;
- }
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-static void highbd_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride,
- uint8_t *dst8, ptrdiff_t dst_stride,
- const InterpKernel *y_filters,
- int y0_q4, int y_step_q4, int w, int h,
- int bd) {- int x, y;
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
- src -= src_stride * (SUBPEL_TAPS / 2 - 1);
- for (x = 0; x < w; ++x) {- int y_q4 = y0_q4;
- for (y = 0; y < h; ++y) {- const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
- const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
- int k, sum = 0;
- for (k = 0; k < SUBPEL_TAPS; ++k)
- sum += src_y[k * src_stride] * y_filter[k];
- dst[y * dst_stride] = clip_pixel_highbd(
- ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
- y_q4 += y_step_q4;
- }
- ++src;
- ++dst;
- }
-}
-
-static void highbd_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride,
- uint8_t *dst8, ptrdiff_t dst_stride,
- const InterpKernel *y_filters,
- int y0_q4, int y_step_q4, int w, int h,
- int bd) {- int x, y;
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
- src -= src_stride * (SUBPEL_TAPS / 2 - 1);
- for (x = 0; x < w; ++x) {- int y_q4 = y0_q4;
- for (y = 0; y < h; ++y) {- const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
- const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
- int k, sum = 0;
- for (k = 0; k < SUBPEL_TAPS; ++k)
- sum += src_y[k * src_stride] * y_filter[k];
- dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] +
- clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
- y_q4 += y_step_q4;
- }
- ++src;
- ++dst;
- }
-}
-
-static void highbd_convolve(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const InterpKernel *const x_filters,
- int x0_q4, int x_step_q4,
- const InterpKernel *const y_filters,
- int y0_q4, int y_step_q4,
- int w, int h, int bd) {- // Note: Fixed size intermediate buffer, temp, places limits on parameters.
- // 2d filtering proceeds in 2 steps:
- // (1) Interpolate horizontally into an intermediate buffer, temp.
- // (2) Interpolate temp vertically to derive the sub-pixel result.
- // Deriving the maximum number of rows in the temp buffer (135):
- // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
- // --Largest block size is 64x64 pixels.
- // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
- // original frame (in 1/16th pixel units).
- // --Must round-up because block may be located at sub-pixel position.
- // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
- // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
- uint16_t temp[64 * 135];
- int intermediate_height =
- (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
-
- assert(w <= 64);
- assert(h <= 64);
- assert(y_step_q4 <= 32);
- assert(x_step_q4 <= 32);
-
- highbd_convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1),
- src_stride, CONVERT_TO_BYTEPTR(temp), 64,
- x_filters, x0_q4, x_step_q4, w,
- intermediate_height, bd);
- highbd_convolve_vert(CONVERT_TO_BYTEPTR(temp) + 64 * (SUBPEL_TAPS / 2 - 1),
- 64, dst, dst_stride, y_filters, y0_q4, y_step_q4,
- w, h, bd);
-}
-
-
-void vp9_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd) {- const InterpKernel *const filters_x = get_filter_base(filter_x);
- const int x0_q4 = get_filter_offset(filter_x, filters_x);
- (void)filter_y;
- (void)y_step_q4;
-
- highbd_convolve_horiz(src, src_stride, dst, dst_stride, filters_x,
- x0_q4, x_step_q4, w, h, bd);
-}
-
-void vp9_highbd_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd) {- const InterpKernel *const filters_x = get_filter_base(filter_x);
- const int x0_q4 = get_filter_offset(filter_x, filters_x);
- (void)filter_y;
- (void)y_step_q4;
-
- highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x,
- x0_q4, x_step_q4, w, h, bd);
-}
-
-void vp9_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd) {- const InterpKernel *const filters_y = get_filter_base(filter_y);
- const int y0_q4 = get_filter_offset(filter_y, filters_y);
- (void)filter_x;
- (void)x_step_q4;
-
- highbd_convolve_vert(src, src_stride, dst, dst_stride, filters_y,
- y0_q4, y_step_q4, w, h, bd);
-}
-
-void vp9_highbd_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd) {- const InterpKernel *const filters_y = get_filter_base(filter_y);
- const int y0_q4 = get_filter_offset(filter_y, filters_y);
- (void)filter_x;
- (void)x_step_q4;
-
- highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y,
- y0_q4, y_step_q4, w, h, bd);
-}
-
-void vp9_highbd_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd) {- const InterpKernel *const filters_x = get_filter_base(filter_x);
- const int x0_q4 = get_filter_offset(filter_x, filters_x);
-
- const InterpKernel *const filters_y = get_filter_base(filter_y);
- const int y0_q4 = get_filter_offset(filter_y, filters_y);
-
- highbd_convolve(src, src_stride, dst, dst_stride,
- filters_x, x0_q4, x_step_q4,
- filters_y, y0_q4, y_step_q4, w, h, bd);
-}
-
-void vp9_highbd_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd) {- // Fixed size intermediate buffer places limits on parameters.
- DECLARE_ALIGNED(16, uint16_t, temp[64 * 64]);
- assert(w <= 64);
- assert(h <= 64);
-
- vp9_highbd_convolve8_c(src, src_stride, CONVERT_TO_BYTEPTR(temp), 64,
- filter_x, x_step_q4, filter_y, y_step_q4, w, h, bd);
- vp9_highbd_convolve_avg_c(CONVERT_TO_BYTEPTR(temp), 64, dst, dst_stride,
- NULL, 0, NULL, 0, w, h, bd);
-}
-
-void vp9_highbd_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride,
- uint8_t *dst8, ptrdiff_t dst_stride,
- const int16_t *filter_x, int filter_x_stride,
- const int16_t *filter_y, int filter_y_stride,
- int w, int h, int bd) {- int r;
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
- (void)filter_x;
- (void)filter_y;
- (void)filter_x_stride;
- (void)filter_y_stride;
- (void)bd;
-
- for (r = h; r > 0; --r) {- memcpy(dst, src, w * sizeof(uint16_t));
- src += src_stride;
- dst += dst_stride;
- }
-}
-
-void vp9_highbd_convolve_avg_c(const uint8_t *src8, ptrdiff_t src_stride,
- uint8_t *dst8, ptrdiff_t dst_stride,
- const int16_t *filter_x, int filter_x_stride,
- const int16_t *filter_y, int filter_y_stride,
- int w, int h, int bd) {- int x, y;
- uint16_t *src = CONVERT_TO_SHORTPTR(src8);
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
- (void)filter_x;
- (void)filter_y;
- (void)filter_x_stride;
- (void)filter_y_stride;
- (void)bd;
-
- for (y = 0; y < h; ++y) {- for (x = 0; x < w; ++x) {- dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
- }
- src += src_stride;
- dst += dst_stride;
- }
-}
-#endif
--- a/vp9/common/vp9_convolve.h
+++ /dev/null
@@ -1,38 +1,0 @@
-/*
- * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-#ifndef VP9_COMMON_VP9_CONVOLVE_H_
-#define VP9_COMMON_VP9_CONVOLVE_H_
-
-#include "./vpx_config.h"
-#include "vpx/vpx_integer.h"
-
-#ifdef __cplusplus
-extern "C" {-#endif
-
-typedef void (*convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h);
-
-#if CONFIG_VP9_HIGHBITDEPTH
-typedef void (*highbd_convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
- uint8_t *dst, ptrdiff_t dst_stride,
- const int16_t *filter_x, int x_step_q4,
- const int16_t *filter_y, int y_step_q4,
- int w, int h, int bd);
-#endif
-
-#ifdef __cplusplus
-} // extern "C"
-#endif
-
-#endif // VP9_COMMON_VP9_CONVOLVE_H_
--- a/vp9/common/vp9_entropymode.h
+++ b/vp9/common/vp9_entropymode.h
@@ -11,9 +11,10 @@
#ifndef VP9_COMMON_VP9_ENTROPYMODE_H_
#define VP9_COMMON_VP9_ENTROPYMODE_H_
-#include "vp9/common/vp9_filter.h"
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymv.h"
+#include "vp9/common/vp9_filter.h"
+#include "vpx_dsp/vpx_filter.h"
#ifdef __cplusplus
extern "C" {--- a/vp9/common/vp9_filter.h
+++ b/vp9/common/vp9_filter.h
@@ -13,6 +13,7 @@
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
+#include "vpx_dsp/vpx_filter.h"
#include "vpx_ports/mem.h"
@@ -20,13 +21,6 @@
extern "C" {#endif
-#define FILTER_BITS 7
-
-#define SUBPEL_BITS 4
-#define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1)
-#define SUBPEL_SHIFTS (1 << SUBPEL_BITS)
-#define SUBPEL_TAPS 8
-
#define EIGHTTAP 0
#define EIGHTTAP_SMOOTH 1
#define EIGHTTAP_SHARP 2
@@ -36,9 +30,8 @@
// 8-tap, 8-tap-smooth, 8-tap-sharp, and switching between the three.
#define SWITCHABLE_FILTER_CONTEXTS (SWITCHABLE_FILTERS + 1)
#define SWITCHABLE 4 /* should be the last one */
-typedef uint8_t INTERP_FILTER;
-typedef int16_t InterpKernel[SUBPEL_TAPS];
+typedef uint8_t INTERP_FILTER;
extern const InterpKernel *vp9_filter_kernels[4];
--- a/vp9/common/vp9_idct.h
+++ b/vp9/common/vp9_idct.h
@@ -15,6 +15,9 @@
#include "./vpx_config.h"
#include "vpx_dsp/txfm_common.h"
+#if CONFIG_VP9_HIGHBITDEPTH
+#include "vpx_dsp/vpx_dsp_common.h"
+#endif // CONFIG_VP9_HIGHBITDEPTH
#include "vpx_ports/mem.h"
#include "vp9/common/vp9_common.h"
#include "vp9/common/vp9_enums.h"
--- a/vp9/common/vp9_reconinter.c
+++ b/vp9/common/vp9_reconinter.c
@@ -16,7 +16,6 @@
#include "vpx/vpx_integer.h"
#include "vp9/common/vp9_blockd.h"
-#include "vp9/common/vp9_filter.h"
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_reconintra.h"
--- a/vp9/common/vp9_reconinter.h
+++ b/vp9/common/vp9_reconinter.h
@@ -11,8 +11,10 @@
#ifndef VP9_COMMON_VP9_RECONINTER_H_
#define VP9_COMMON_VP9_RECONINTER_H_
-#include "vpx/vpx_integer.h"
+#include "vp9/common/vp9_filter.h"
#include "vp9/common/vp9_onyxc_int.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_dsp/vpx_filter.h"
#ifdef __cplusplus
extern "C" {--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -11,6 +11,9 @@
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
+#if CONFIG_VP9_HIGHBITDEPTH
+#include "vpx_dsp/vpx_dsp_common.h"
+#endif // CONFIG_VP9_HIGHBITDEPTH
#include "vpx_mem/vpx_mem.h"
#include "vpx_ports/mem.h"
#include "vpx_ports/vpx_once.h"
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -54,12 +54,6 @@
$avx2_x86_64 = 'avx2';
}
-# optimizations which depend on multiple features
-$avx2_ssse3 = '';
-if ((vpx_config("HAVE_AVX2") eq "yes") && (vpx_config("HAVE_SSSE3") eq "yes")) {- $avx2_ssse3 = 'avx2';
-}
-
#
# post proc
#
@@ -86,33 +80,6 @@
add_proto qw/void vp9_filter_by_weight8x8/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
specialize qw/vp9_filter_by_weight8x8 sse2 msa/;
}
-
-#
-# Sub Pixel Filters
-#
-add_proto qw/void vp9_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve_copy neon dspr2 msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve_avg neon dspr2 msa/, "$sse2_x86inc";
-
-add_proto qw/void vp9_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8 sse2 ssse3 neon dspr2 msa/, "$avx2_ssse3";
-
-add_proto qw/void vp9_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_horiz sse2 ssse3 neon dspr2 msa/, "$avx2_ssse3";
-
-add_proto qw/void vp9_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_vert sse2 ssse3 neon dspr2 msa/, "$avx2_ssse3";
-
-add_proto qw/void vp9_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_avg sse2 ssse3 neon dspr2 msa/;
-
-add_proto qw/void vp9_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_avg_horiz sse2 ssse3 neon dspr2 msa/;
-
-add_proto qw/void vp9_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_avg_vert sse2 ssse3 neon dspr2 msa/;
#
# dct
--- a/vp9/common/vp9_scale.c
+++ b/vp9/common/vp9_scale.c
@@ -8,9 +8,10 @@
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "vp9/common/vp9_filter.h"
#include "vp9/common/vp9_scale.h"
+#include "vpx_dsp/vpx_filter.h"
static INLINE int scaled_x(int val, const struct scale_factors *sf) {return (int)((int64_t)val * sf->x_scale_fp >> REF_SCALE_SHIFT);
@@ -81,85 +82,85 @@
if (sf->x_step_q4 == 16) { if (sf->y_step_q4 == 16) {// No scaling in either direction.
- sf->predict[0][0][0] = vp9_convolve_copy;
- sf->predict[0][0][1] = vp9_convolve_avg;
- sf->predict[0][1][0] = vp9_convolve8_vert;
- sf->predict[0][1][1] = vp9_convolve8_avg_vert;
- sf->predict[1][0][0] = vp9_convolve8_horiz;
- sf->predict[1][0][1] = vp9_convolve8_avg_horiz;
+ sf->predict[0][0][0] = vpx_convolve_copy;
+ sf->predict[0][0][1] = vpx_convolve_avg;
+ sf->predict[0][1][0] = vpx_convolve8_vert;
+ sf->predict[0][1][1] = vpx_convolve8_avg_vert;
+ sf->predict[1][0][0] = vpx_convolve8_horiz;
+ sf->predict[1][0][1] = vpx_convolve8_avg_horiz;
} else {// No scaling in x direction. Must always scale in the y direction.
- sf->predict[0][0][0] = vp9_convolve8_vert;
- sf->predict[0][0][1] = vp9_convolve8_avg_vert;
- sf->predict[0][1][0] = vp9_convolve8_vert;
- sf->predict[0][1][1] = vp9_convolve8_avg_vert;
- sf->predict[1][0][0] = vp9_convolve8;
- sf->predict[1][0][1] = vp9_convolve8_avg;
+ sf->predict[0][0][0] = vpx_convolve8_vert;
+ sf->predict[0][0][1] = vpx_convolve8_avg_vert;
+ sf->predict[0][1][0] = vpx_convolve8_vert;
+ sf->predict[0][1][1] = vpx_convolve8_avg_vert;
+ sf->predict[1][0][0] = vpx_convolve8;
+ sf->predict[1][0][1] = vpx_convolve8_avg;
}
} else { if (sf->y_step_q4 == 16) {// No scaling in the y direction. Must always scale in the x direction.
- sf->predict[0][0][0] = vp9_convolve8_horiz;
- sf->predict[0][0][1] = vp9_convolve8_avg_horiz;
- sf->predict[0][1][0] = vp9_convolve8;
- sf->predict[0][1][1] = vp9_convolve8_avg;
- sf->predict[1][0][0] = vp9_convolve8_horiz;
- sf->predict[1][0][1] = vp9_convolve8_avg_horiz;
+ sf->predict[0][0][0] = vpx_convolve8_horiz;
+ sf->predict[0][0][1] = vpx_convolve8_avg_horiz;
+ sf->predict[0][1][0] = vpx_convolve8;
+ sf->predict[0][1][1] = vpx_convolve8_avg;
+ sf->predict[1][0][0] = vpx_convolve8_horiz;
+ sf->predict[1][0][1] = vpx_convolve8_avg_horiz;
} else {// Must always scale in both directions.
- sf->predict[0][0][0] = vp9_convolve8;
- sf->predict[0][0][1] = vp9_convolve8_avg;
- sf->predict[0][1][0] = vp9_convolve8;
- sf->predict[0][1][1] = vp9_convolve8_avg;
- sf->predict[1][0][0] = vp9_convolve8;
- sf->predict[1][0][1] = vp9_convolve8_avg;
+ sf->predict[0][0][0] = vpx_convolve8;
+ sf->predict[0][0][1] = vpx_convolve8_avg;
+ sf->predict[0][1][0] = vpx_convolve8;
+ sf->predict[0][1][1] = vpx_convolve8_avg;
+ sf->predict[1][0][0] = vpx_convolve8;
+ sf->predict[1][0][1] = vpx_convolve8_avg;
}
}
// 2D subpel motion always gets filtered in both directions
- sf->predict[1][1][0] = vp9_convolve8;
- sf->predict[1][1][1] = vp9_convolve8_avg;
+ sf->predict[1][1][0] = vpx_convolve8;
+ sf->predict[1][1][1] = vpx_convolve8_avg;
#if CONFIG_VP9_HIGHBITDEPTH
if (use_highbd) { if (sf->x_step_q4 == 16) { if (sf->y_step_q4 == 16) {// No scaling in either direction.
- sf->highbd_predict[0][0][0] = vp9_highbd_convolve_copy;
- sf->highbd_predict[0][0][1] = vp9_highbd_convolve_avg;
- sf->highbd_predict[0][1][0] = vp9_highbd_convolve8_vert;
- sf->highbd_predict[0][1][1] = vp9_highbd_convolve8_avg_vert;
- sf->highbd_predict[1][0][0] = vp9_highbd_convolve8_horiz;
- sf->highbd_predict[1][0][1] = vp9_highbd_convolve8_avg_horiz;
+ sf->highbd_predict[0][0][0] = vpx_highbd_convolve_copy;
+ sf->highbd_predict[0][0][1] = vpx_highbd_convolve_avg;
+ sf->highbd_predict[0][1][0] = vpx_highbd_convolve8_vert;
+ sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg_vert;
+ sf->highbd_predict[1][0][0] = vpx_highbd_convolve8_horiz;
+ sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg_horiz;
} else {// No scaling in x direction. Must always scale in the y direction.
- sf->highbd_predict[0][0][0] = vp9_highbd_convolve8_vert;
- sf->highbd_predict[0][0][1] = vp9_highbd_convolve8_avg_vert;
- sf->highbd_predict[0][1][0] = vp9_highbd_convolve8_vert;
- sf->highbd_predict[0][1][1] = vp9_highbd_convolve8_avg_vert;
- sf->highbd_predict[1][0][0] = vp9_highbd_convolve8;
- sf->highbd_predict[1][0][1] = vp9_highbd_convolve8_avg;
+ sf->highbd_predict[0][0][0] = vpx_highbd_convolve8_vert;
+ sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg_vert;
+ sf->highbd_predict[0][1][0] = vpx_highbd_convolve8_vert;
+ sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg_vert;
+ sf->highbd_predict[1][0][0] = vpx_highbd_convolve8;
+ sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg;
}
} else { if (sf->y_step_q4 == 16) {// No scaling in the y direction. Must always scale in the x direction.
- sf->highbd_predict[0][0][0] = vp9_highbd_convolve8_horiz;
- sf->highbd_predict[0][0][1] = vp9_highbd_convolve8_avg_horiz;
- sf->highbd_predict[0][1][0] = vp9_highbd_convolve8;
- sf->highbd_predict[0][1][1] = vp9_highbd_convolve8_avg;
- sf->highbd_predict[1][0][0] = vp9_highbd_convolve8_horiz;
- sf->highbd_predict[1][0][1] = vp9_highbd_convolve8_avg_horiz;
+ sf->highbd_predict[0][0][0] = vpx_highbd_convolve8_horiz;
+ sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg_horiz;
+ sf->highbd_predict[0][1][0] = vpx_highbd_convolve8;
+ sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg;
+ sf->highbd_predict[1][0][0] = vpx_highbd_convolve8_horiz;
+ sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg_horiz;
} else {// Must always scale in both directions.
- sf->highbd_predict[0][0][0] = vp9_highbd_convolve8;
- sf->highbd_predict[0][0][1] = vp9_highbd_convolve8_avg;
- sf->highbd_predict[0][1][0] = vp9_highbd_convolve8;
- sf->highbd_predict[0][1][1] = vp9_highbd_convolve8_avg;
- sf->highbd_predict[1][0][0] = vp9_highbd_convolve8;
- sf->highbd_predict[1][0][1] = vp9_highbd_convolve8_avg;
+ sf->highbd_predict[0][0][0] = vpx_highbd_convolve8;
+ sf->highbd_predict[0][0][1] = vpx_highbd_convolve8_avg;
+ sf->highbd_predict[0][1][0] = vpx_highbd_convolve8;
+ sf->highbd_predict[0][1][1] = vpx_highbd_convolve8_avg;
+ sf->highbd_predict[1][0][0] = vpx_highbd_convolve8;
+ sf->highbd_predict[1][0][1] = vpx_highbd_convolve8_avg;
}
}
// 2D subpel motion always gets filtered in both directions.
- sf->highbd_predict[1][1][0] = vp9_highbd_convolve8;
- sf->highbd_predict[1][1][1] = vp9_highbd_convolve8_avg;
+ sf->highbd_predict[1][1][0] = vpx_highbd_convolve8;
+ sf->highbd_predict[1][1][1] = vpx_highbd_convolve8_avg;
}
#endif
}
--- a/vp9/common/vp9_scale.h
+++ b/vp9/common/vp9_scale.h
@@ -12,7 +12,7 @@
#define VP9_COMMON_VP9_SCALE_H_
#include "vp9/common/vp9_mv.h"
-#include "vp9/common/vp9_convolve.h"
+#include "vpx_dsp/vpx_convolve.h"
#ifdef __cplusplus
extern "C" {--- a/vp9/common/x86/convolve.h
+++ /dev/null
@@ -1,296 +1,0 @@
-/*
- * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-#ifndef VP9_COMMON_X86_CONVOLVE_H_
-#define VP9_COMMON_X86_CONVOLVE_H_
-
-#include <assert.h>
-
-#include "./vpx_config.h"
-#include "vpx/vpx_integer.h"
-#include "vpx_ports/mem.h"
-
-typedef void filter8_1dfunction (
- const uint8_t *src_ptr,
- ptrdiff_t src_pitch,
- uint8_t *output_ptr,
- ptrdiff_t out_pitch,
- uint32_t output_height,
- const int16_t *filter
-);
-
-#define FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
- void vp9_convolve8_##name##_##opt(const uint8_t *src, ptrdiff_t src_stride, \
- uint8_t *dst, ptrdiff_t dst_stride, \
- const int16_t *filter_x, int x_step_q4, \
- const int16_t *filter_y, int y_step_q4, \
- int w, int h) { \- if (step_q4 == 16 && filter[3] != 128) { \- if (filter[0] || filter[1] || filter[2]) { \- while (w >= 16) { \- vp9_filter_block1d16_##dir##8_##avg##opt(src_start, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter); \
- src += 16; \
- dst += 16; \
- w -= 16; \
- } \
- while (w >= 8) { \- vp9_filter_block1d8_##dir##8_##avg##opt(src_start, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter); \
- src += 8; \
- dst += 8; \
- w -= 8; \
- } \
- while (w >= 4) { \- vp9_filter_block1d4_##dir##8_##avg##opt(src_start, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter); \
- src += 4; \
- dst += 4; \
- w -= 4; \
- } \
- } else { \- while (w >= 16) { \- vp9_filter_block1d16_##dir##2_##avg##opt(src, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter); \
- src += 16; \
- dst += 16; \
- w -= 16; \
- } \
- while (w >= 8) { \- vp9_filter_block1d8_##dir##2_##avg##opt(src, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter); \
- src += 8; \
- dst += 8; \
- w -= 8; \
- } \
- while (w >= 4) { \- vp9_filter_block1d4_##dir##2_##avg##opt(src, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter); \
- src += 4; \
- dst += 4; \
- w -= 4; \
- } \
- } \
- } \
- if (w) { \- vp9_convolve8_##name##_c(src, src_stride, dst, dst_stride, \
- filter_x, x_step_q4, filter_y, y_step_q4, \
- w, h); \
- } \
-}
-
-#define FUN_CONV_2D(avg, opt) \
-void vp9_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \
- uint8_t *dst, ptrdiff_t dst_stride, \
- const int16_t *filter_x, int x_step_q4, \
- const int16_t *filter_y, int y_step_q4, \
- int w, int h) { \- assert(w <= 64); \
- assert(h <= 64); \
- if (x_step_q4 == 16 && y_step_q4 == 16) { \- if (filter_x[0] || filter_x[1] || filter_x[2] || filter_x[3] == 128 || \
- filter_y[0] || filter_y[1] || filter_y[2] || filter_y[3] == 128) { \- DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71]); \
- vp9_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \
- filter_x, x_step_q4, filter_y, y_step_q4, \
- w, h + 7); \
- vp9_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \
- filter_x, x_step_q4, filter_y, \
- y_step_q4, w, h); \
- } else { \- DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 65]); \
- vp9_convolve8_horiz_##opt(src, src_stride, fdata2, 64, \
- filter_x, x_step_q4, filter_y, y_step_q4, \
- w, h + 1); \
- vp9_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, \
- filter_x, x_step_q4, filter_y, \
- y_step_q4, w, h); \
- } \
- } else { \- vp9_convolve8_##avg##c(src, src_stride, dst, dst_stride, \
- filter_x, x_step_q4, filter_y, y_step_q4, w, h); \
- } \
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-
-typedef void highbd_filter8_1dfunction (
- const uint16_t *src_ptr,
- const ptrdiff_t src_pitch,
- uint16_t *output_ptr,
- ptrdiff_t out_pitch,
- unsigned int output_height,
- const int16_t *filter,
- int bd
-);
-
-#define HIGH_FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
- void vp9_highbd_convolve8_##name##_##opt(const uint8_t *src8, \
- ptrdiff_t src_stride, \
- uint8_t *dst8, \
- ptrdiff_t dst_stride, \
- const int16_t *filter_x, \
- int x_step_q4, \
- const int16_t *filter_y, \
- int y_step_q4, \
- int w, int h, int bd) { \- if (step_q4 == 16 && filter[3] != 128) { \- uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
- uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
- if (filter[0] || filter[1] || filter[2]) { \- while (w >= 16) { \- vp9_highbd_filter_block1d16_##dir##8_##avg##opt(src_start, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter, \
- bd); \
- src += 16; \
- dst += 16; \
- w -= 16; \
- } \
- while (w >= 8) { \- vp9_highbd_filter_block1d8_##dir##8_##avg##opt(src_start, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter, \
- bd); \
- src += 8; \
- dst += 8; \
- w -= 8; \
- } \
- while (w >= 4) { \- vp9_highbd_filter_block1d4_##dir##8_##avg##opt(src_start, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter, \
- bd); \
- src += 4; \
- dst += 4; \
- w -= 4; \
- } \
- } else { \- while (w >= 16) { \- vp9_highbd_filter_block1d16_##dir##2_##avg##opt(src, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter, \
- bd); \
- src += 16; \
- dst += 16; \
- w -= 16; \
- } \
- while (w >= 8) { \- vp9_highbd_filter_block1d8_##dir##2_##avg##opt(src, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter, \
- bd); \
- src += 8; \
- dst += 8; \
- w -= 8; \
- } \
- while (w >= 4) { \- vp9_highbd_filter_block1d4_##dir##2_##avg##opt(src, \
- src_stride, \
- dst, \
- dst_stride, \
- h, \
- filter, \
- bd); \
- src += 4; \
- dst += 4; \
- w -= 4; \
- } \
- } \
- } \
- if (w) { \- vp9_highbd_convolve8_##name##_c(src8, src_stride, dst8, dst_stride, \
- filter_x, x_step_q4, filter_y, y_step_q4, \
- w, h, bd); \
- } \
-}
-
-#define HIGH_FUN_CONV_2D(avg, opt) \
-void vp9_highbd_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \
- uint8_t *dst, ptrdiff_t dst_stride, \
- const int16_t *filter_x, int x_step_q4, \
- const int16_t *filter_y, int y_step_q4, \
- int w, int h, int bd) { \- assert(w <= 64); \
- assert(h <= 64); \
- if (x_step_q4 == 16 && y_step_q4 == 16) { \- if (filter_x[0] || filter_x[1] || filter_x[2] || filter_x[3] == 128 || \
- filter_y[0] || filter_y[1] || filter_y[2] || filter_y[3] == 128) { \- DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 71]); \
- vp9_highbd_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \
- CONVERT_TO_BYTEPTR(fdata2), 64, \
- filter_x, x_step_q4, \
- filter_y, y_step_q4, \
- w, h + 7, bd); \
- vp9_highbd_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2) + 192, \
- 64, dst, dst_stride, \
- filter_x, x_step_q4, \
- filter_y, y_step_q4, \
- w, h, bd); \
- } else { \- DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 65]); \
- vp9_highbd_convolve8_horiz_##opt(src, src_stride, \
- CONVERT_TO_BYTEPTR(fdata2), 64, \
- filter_x, x_step_q4, \
- filter_y, y_step_q4, \
- w, h + 1, bd); \
- vp9_highbd_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2), 64, \
- dst, dst_stride, \
- filter_x, x_step_q4, \
- filter_y, y_step_q4, \
- w, h, bd); \
- } \
- } else { \- vp9_highbd_convolve8_##avg##c(src, src_stride, dst, dst_stride, \
- filter_x, x_step_q4, filter_y, y_step_q4, w, \
- h, bd); \
- } \
-}
-#endif // CONFIG_VP9_HIGHBITDEPTH
-
-#endif // VP9_COMMON_X86_CONVOLVE_H_
--- a/vp9/common/x86/vp9_asm_stubs.c
+++ /dev/null
@@ -1,162 +1,0 @@
-/*
- * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-#include "./vp9_rtcd.h"
-#include "./vpx_config.h"
-#include "vp9/common/x86/convolve.h"
-
-#if HAVE_SSE2
-filter8_1dfunction vp9_filter_block1d16_v8_sse2;
-filter8_1dfunction vp9_filter_block1d16_h8_sse2;
-filter8_1dfunction vp9_filter_block1d8_v8_sse2;
-filter8_1dfunction vp9_filter_block1d8_h8_sse2;
-filter8_1dfunction vp9_filter_block1d4_v8_sse2;
-filter8_1dfunction vp9_filter_block1d4_h8_sse2;
-filter8_1dfunction vp9_filter_block1d16_v8_avg_sse2;
-filter8_1dfunction vp9_filter_block1d16_h8_avg_sse2;
-filter8_1dfunction vp9_filter_block1d8_v8_avg_sse2;
-filter8_1dfunction vp9_filter_block1d8_h8_avg_sse2;
-filter8_1dfunction vp9_filter_block1d4_v8_avg_sse2;
-filter8_1dfunction vp9_filter_block1d4_h8_avg_sse2;
-
-filter8_1dfunction vp9_filter_block1d16_v2_sse2;
-filter8_1dfunction vp9_filter_block1d16_h2_sse2;
-filter8_1dfunction vp9_filter_block1d8_v2_sse2;
-filter8_1dfunction vp9_filter_block1d8_h2_sse2;
-filter8_1dfunction vp9_filter_block1d4_v2_sse2;
-filter8_1dfunction vp9_filter_block1d4_h2_sse2;
-filter8_1dfunction vp9_filter_block1d16_v2_avg_sse2;
-filter8_1dfunction vp9_filter_block1d16_h2_avg_sse2;
-filter8_1dfunction vp9_filter_block1d8_v2_avg_sse2;
-filter8_1dfunction vp9_filter_block1d8_h2_avg_sse2;
-filter8_1dfunction vp9_filter_block1d4_v2_avg_sse2;
-filter8_1dfunction vp9_filter_block1d4_h2_avg_sse2;
-
-// void vp9_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_avg_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_avg_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
-FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
-FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
-FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, sse2);
-
-// void vp9_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-FUN_CONV_2D(, sse2);
-FUN_CONV_2D(avg_ , sse2);
-
-#if CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64
-highbd_filter8_1dfunction vp9_highbd_filter_block1d16_v8_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d16_h8_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d8_v8_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d8_h8_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d4_v8_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d4_h8_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d16_v8_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d16_h8_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d8_v8_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d8_h8_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d4_v8_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d4_h8_avg_sse2;
-
-highbd_filter8_1dfunction vp9_highbd_filter_block1d16_v2_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d16_h2_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d8_v2_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d8_h2_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d4_v2_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d4_h2_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d16_v2_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d16_h2_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d8_v2_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d8_h2_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d4_v2_avg_sse2;
-highbd_filter8_1dfunction vp9_highbd_filter_block1d4_h2_avg_sse2;
-
-// void vp9_highbd_convolve8_horiz_sse2(const uint8_t *src,
-// ptrdiff_t src_stride,
-// uint8_t *dst,
-// ptrdiff_t dst_stride,
-// const int16_t *filter_x,
-// int x_step_q4,
-// const int16_t *filter_y,
-// int y_step_q4,
-// int w, int h, int bd);
-// void vp9_highbd_convolve8_vert_sse2(const uint8_t *src,
-// ptrdiff_t src_stride,
-// uint8_t *dst,
-// ptrdiff_t dst_stride,
-// const int16_t *filter_x,
-// int x_step_q4,
-// const int16_t *filter_y,
-// int y_step_q4,
-// int w, int h, int bd);
-// void vp9_highbd_convolve8_avg_horiz_sse2(const uint8_t *src,
-// ptrdiff_t src_stride,
-// uint8_t *dst,
-// ptrdiff_t dst_stride,
-// const int16_t *filter_x,
-// int x_step_q4,
-// const int16_t *filter_y,
-// int y_step_q4,
-// int w, int h, int bd);
-// void vp9_highbd_convolve8_avg_vert_sse2(const uint8_t *src,
-// ptrdiff_t src_stride,
-// uint8_t *dst,
-// ptrdiff_t dst_stride,
-// const int16_t *filter_x,
-// int x_step_q4,
-// const int16_t *filter_y,
-// int y_step_q4,
-// int w, int h, int bd);
-HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
-HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
-HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
-HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
- sse2);
-
-// void vp9_highbd_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h, int bd);
-// void vp9_highbd_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h, int bd);
-HIGH_FUN_CONV_2D(, sse2);
-HIGH_FUN_CONV_2D(avg_ , sse2);
-#endif // CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64
-#endif // HAVE_SSE2
--- a/vp9/common/x86/vp9_copy_sse2.asm
+++ /dev/null
@@ -1,156 +1,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-%include "third_party/x86inc/x86inc.asm"
-
-SECTION .text
-
-%macro convolve_fn 1
-INIT_XMM sse2
-cglobal convolve_%1, 4, 7, 4, src, src_stride, dst, dst_stride, \
- fx, fxs, fy, fys, w, h
- mov r4d, dword wm
- cmp r4d, 4
- je .w4
- cmp r4d, 8
- je .w8
- cmp r4d, 16
- je .w16
- cmp r4d, 32
- je .w32
-
- mov r4d, dword hm
-.loop64:
- movu m0, [srcq]
- movu m1, [srcq+16]
- movu m2, [srcq+32]
- movu m3, [srcq+48]
- add srcq, src_strideq
-%ifidn %1, avg
- pavgb m0, [dstq]
- pavgb m1, [dstq+16]
- pavgb m2, [dstq+32]
- pavgb m3, [dstq+48]
-%endif
- mova [dstq ], m0
- mova [dstq+16], m1
- mova [dstq+32], m2
- mova [dstq+48], m3
- add dstq, dst_strideq
- dec r4d
- jnz .loop64
- RET
-
-.w32:
- mov r4d, dword hm
-.loop32:
- movu m0, [srcq]
- movu m1, [srcq+16]
- movu m2, [srcq+src_strideq]
- movu m3, [srcq+src_strideq+16]
- lea srcq, [srcq+src_strideq*2]
-%ifidn %1, avg
- pavgb m0, [dstq]
- pavgb m1, [dstq +16]
- pavgb m2, [dstq+dst_strideq]
- pavgb m3, [dstq+dst_strideq+16]
-%endif
- mova [dstq ], m0
- mova [dstq +16], m1
- mova [dstq+dst_strideq ], m2
- mova [dstq+dst_strideq+16], m3
- lea dstq, [dstq+dst_strideq*2]
- sub r4d, 2
- jnz .loop32
- RET
-
-.w16:
- mov r4d, dword hm
- lea r5q, [src_strideq*3]
- lea r6q, [dst_strideq*3]
-.loop16:
- movu m0, [srcq]
- movu m1, [srcq+src_strideq]
- movu m2, [srcq+src_strideq*2]
- movu m3, [srcq+r5q]
- lea srcq, [srcq+src_strideq*4]
-%ifidn %1, avg
- pavgb m0, [dstq]
- pavgb m1, [dstq+dst_strideq]
- pavgb m2, [dstq+dst_strideq*2]
- pavgb m3, [dstq+r6q]
-%endif
- mova [dstq ], m0
- mova [dstq+dst_strideq ], m1
- mova [dstq+dst_strideq*2], m2
- mova [dstq+r6q ], m3
- lea dstq, [dstq+dst_strideq*4]
- sub r4d, 4
- jnz .loop16
- RET
-
-INIT_MMX sse
-.w8:
- mov r4d, dword hm
- lea r5q, [src_strideq*3]
- lea r6q, [dst_strideq*3]
-.loop8:
- movu m0, [srcq]
- movu m1, [srcq+src_strideq]
- movu m2, [srcq+src_strideq*2]
- movu m3, [srcq+r5q]
- lea srcq, [srcq+src_strideq*4]
-%ifidn %1, avg
- pavgb m0, [dstq]
- pavgb m1, [dstq+dst_strideq]
- pavgb m2, [dstq+dst_strideq*2]
- pavgb m3, [dstq+r6q]
-%endif
- mova [dstq ], m0
- mova [dstq+dst_strideq ], m1
- mova [dstq+dst_strideq*2], m2
- mova [dstq+r6q ], m3
- lea dstq, [dstq+dst_strideq*4]
- sub r4d, 4
- jnz .loop8
- RET
-
-.w4:
- mov r4d, dword hm
- lea r5q, [src_strideq*3]
- lea r6q, [dst_strideq*3]
-.loop4:
- movh m0, [srcq]
- movh m1, [srcq+src_strideq]
- movh m2, [srcq+src_strideq*2]
- movh m3, [srcq+r5q]
- lea srcq, [srcq+src_strideq*4]
-%ifidn %1, avg
- movh m4, [dstq]
- movh m5, [dstq+dst_strideq]
- movh m6, [dstq+dst_strideq*2]
- movh m7, [dstq+r6q]
- pavgb m0, m4
- pavgb m1, m5
- pavgb m2, m6
- pavgb m3, m7
-%endif
- movh [dstq ], m0
- movh [dstq+dst_strideq ], m1
- movh [dstq+dst_strideq*2], m2
- movh [dstq+r6q ], m3
- lea dstq, [dstq+dst_strideq*4]
- sub r4d, 4
- jnz .loop4
- RET
-%endmacro
-
-convolve_fn copy
-convolve_fn avg
--- a/vp9/common/x86/vp9_high_subpixel_8t_sse2.asm
+++ /dev/null
@@ -1,962 +1,0 @@
-;
-; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-
-;Note: tap3 and tap4 have to be applied and added after other taps to avoid
-;overflow.
-
-%macro HIGH_GET_FILTERS_4 0
- mov rdx, arg(5) ;filter ptr
- mov rcx, 0x00000040
-
- movdqa xmm7, [rdx] ;load filters
- pshuflw xmm0, xmm7, 0b ;k0
- pshuflw xmm1, xmm7, 01010101b ;k1
- pshuflw xmm2, xmm7, 10101010b ;k2
- pshuflw xmm3, xmm7, 11111111b ;k3
- psrldq xmm7, 8
- pshuflw xmm4, xmm7, 0b ;k4
- pshuflw xmm5, xmm7, 01010101b ;k5
- pshuflw xmm6, xmm7, 10101010b ;k6
- pshuflw xmm7, xmm7, 11111111b ;k7
-
- punpcklwd xmm0, xmm6
- punpcklwd xmm2, xmm5
- punpcklwd xmm3, xmm4
- punpcklwd xmm1, xmm7
-
- movdqa k0k6, xmm0
- movdqa k2k5, xmm2
- movdqa k3k4, xmm3
- movdqa k1k7, xmm1
-
- movq xmm6, rcx
- pshufd xmm6, xmm6, 0
- movdqa krd, xmm6
-
- ;Compute max and min values of a pixel
- mov rdx, 0x00010001
- movsxd rcx, DWORD PTR arg(6) ;bps
- movq xmm0, rdx
- movq xmm1, rcx
- pshufd xmm0, xmm0, 0b
- movdqa xmm2, xmm0
- psllw xmm0, xmm1
- psubw xmm0, xmm2
- pxor xmm1, xmm1
- movdqa max, xmm0 ;max value (for clamping)
- movdqa min, xmm1 ;min value (for clamping)
-
-%endm
-
-%macro HIGH_APPLY_FILTER_4 1
- punpcklwd xmm0, xmm6 ;two row in one register
- punpcklwd xmm1, xmm7
- punpcklwd xmm2, xmm5
- punpcklwd xmm3, xmm4
-
- pmaddwd xmm0, k0k6 ;multiply the filter factors
- pmaddwd xmm1, k1k7
- pmaddwd xmm2, k2k5
- pmaddwd xmm3, k3k4
-
- paddd xmm0, xmm1 ;sum
- paddd xmm0, xmm2
- paddd xmm0, xmm3
-
- paddd xmm0, krd ;rounding
- psrad xmm0, 7 ;shift
- packssdw xmm0, xmm0 ;pack to word
-
- ;clamp the values
- pminsw xmm0, max
- pmaxsw xmm0, min
-
-%if %1
- movq xmm1, [rdi]
- pavgw xmm0, xmm1
-%endif
- movq [rdi], xmm0
-%endm
-
-%macro HIGH_GET_FILTERS 0
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x00000040
-
- movdqa xmm7, [rdx] ;load filters
- pshuflw xmm0, xmm7, 0b ;k0
- pshuflw xmm1, xmm7, 01010101b ;k1
- pshuflw xmm2, xmm7, 10101010b ;k2
- pshuflw xmm3, xmm7, 11111111b ;k3
- pshufhw xmm4, xmm7, 0b ;k4
- pshufhw xmm5, xmm7, 01010101b ;k5
- pshufhw xmm6, xmm7, 10101010b ;k6
- pshufhw xmm7, xmm7, 11111111b ;k7
- punpcklqdq xmm2, xmm2
- punpcklqdq xmm3, xmm3
- punpcklwd xmm0, xmm1
- punpckhwd xmm6, xmm7
- punpckhwd xmm2, xmm5
- punpckhwd xmm3, xmm4
-
- movdqa k0k1, xmm0 ;store filter factors on stack
- movdqa k6k7, xmm6
- movdqa k2k5, xmm2
- movdqa k3k4, xmm3
-
- movq xmm6, rcx
- pshufd xmm6, xmm6, 0
- movdqa krd, xmm6 ;rounding
-
- ;Compute max and min values of a pixel
- mov rdx, 0x00010001
- movsxd rcx, DWORD PTR arg(6) ;bps
- movq xmm0, rdx
- movq xmm1, rcx
- pshufd xmm0, xmm0, 0b
- movdqa xmm2, xmm0
- psllw xmm0, xmm1
- psubw xmm0, xmm2
- pxor xmm1, xmm1
- movdqa max, xmm0 ;max value (for clamping)
- movdqa min, xmm1 ;min value (for clamping)
-%endm
-
-%macro LOAD_VERT_8 1
- movdqu xmm0, [rsi + %1] ;0
- movdqu xmm1, [rsi + rax + %1] ;1
- movdqu xmm6, [rsi + rdx * 2 + %1] ;6
- lea rsi, [rsi + rax]
- movdqu xmm7, [rsi + rdx * 2 + %1] ;7
- movdqu xmm2, [rsi + rax + %1] ;2
- movdqu xmm3, [rsi + rax * 2 + %1] ;3
- movdqu xmm4, [rsi + rdx + %1] ;4
- movdqu xmm5, [rsi + rax * 4 + %1] ;5
-%endm
-
-%macro HIGH_APPLY_FILTER_8 2
- movdqu temp, xmm4
- movdqa xmm4, xmm0
- punpcklwd xmm0, xmm1
- punpckhwd xmm4, xmm1
- movdqa xmm1, xmm6
- punpcklwd xmm6, xmm7
- punpckhwd xmm1, xmm7
- movdqa xmm7, xmm2
- punpcklwd xmm2, xmm5
- punpckhwd xmm7, xmm5
-
- movdqu xmm5, temp
- movdqu temp, xmm4
- movdqa xmm4, xmm3
- punpcklwd xmm3, xmm5
- punpckhwd xmm4, xmm5
- movdqu xmm5, temp
-
- pmaddwd xmm0, k0k1
- pmaddwd xmm5, k0k1
- pmaddwd xmm6, k6k7
- pmaddwd xmm1, k6k7
- pmaddwd xmm2, k2k5
- pmaddwd xmm7, k2k5
- pmaddwd xmm3, k3k4
- pmaddwd xmm4, k3k4
-
- paddd xmm0, xmm6
- paddd xmm0, xmm2
- paddd xmm0, xmm3
- paddd xmm5, xmm1
- paddd xmm5, xmm7
- paddd xmm5, xmm4
-
- paddd xmm0, krd ;rounding
- paddd xmm5, krd
- psrad xmm0, 7 ;shift
- psrad xmm5, 7
- packssdw xmm0, xmm5 ;pack back to word
-
- ;clamp the values
- pminsw xmm0, max
- pmaxsw xmm0, min
-
-%if %1
- movdqu xmm1, [rdi + %2]
- pavgw xmm0, xmm1
-%endif
- movdqu [rdi + %2], xmm0
-%endm
-
-;void vp9_filter_block1d4_v8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_highbd_filter_block1d4_v8_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d4_v8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 7
- %define k0k6 [rsp + 16 * 0]
- %define k2k5 [rsp + 16 * 1]
- %define k3k4 [rsp + 16 * 2]
- %define k1k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define max [rsp + 16 * 5]
- %define min [rsp + 16 * 6]
-
- HIGH_GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movq xmm0, [rsi] ;load src: row 0
- movq xmm1, [rsi + rax] ;1
- movq xmm6, [rsi + rdx * 2] ;6
- lea rsi, [rsi + rax]
- movq xmm7, [rsi + rdx * 2] ;7
- movq xmm2, [rsi + rax] ;2
- movq xmm3, [rsi + rax * 2] ;3
- movq xmm4, [rsi + rdx] ;4
- movq xmm5, [rsi + rax * 4] ;5
-
- HIGH_APPLY_FILTER_4 0
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 7
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d8_v8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_highbd_filter_block1d8_v8_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d8_v8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- LOAD_VERT_8 0
- HIGH_APPLY_FILTER_8 0, 0
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d16_v8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_highbd_filter_block1d16_v8_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d16_v8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- LOAD_VERT_8 0
- HIGH_APPLY_FILTER_8 0, 0
- sub rsi, rax
-
- LOAD_VERT_8 16
- HIGH_APPLY_FILTER_8 0, 16
- add rdi, rbx
-
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d4_v8_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d4_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 7
- %define k0k6 [rsp + 16 * 0]
- %define k2k5 [rsp + 16 * 1]
- %define k3k4 [rsp + 16 * 2]
- %define k1k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define max [rsp + 16 * 5]
- %define min [rsp + 16 * 6]
-
- HIGH_GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movq xmm0, [rsi] ;load src: row 0
- movq xmm1, [rsi + rax] ;1
- movq xmm6, [rsi + rdx * 2] ;6
- lea rsi, [rsi + rax]
- movq xmm7, [rsi + rdx * 2] ;7
- movq xmm2, [rsi + rax] ;2
- movq xmm3, [rsi + rax * 2] ;3
- movq xmm4, [rsi + rdx] ;4
- movq xmm5, [rsi + rax * 4] ;5
-
- HIGH_APPLY_FILTER_4 1
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 7
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d8_v8_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d8_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-.loop:
- LOAD_VERT_8 0
- HIGH_APPLY_FILTER_8 1, 0
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d16_v8_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d16_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rbx, [rbx + rbx]
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-.loop:
- LOAD_VERT_8 0
- HIGH_APPLY_FILTER_8 1, 0
- sub rsi, rax
-
- LOAD_VERT_8 16
- HIGH_APPLY_FILTER_8 1, 16
- add rdi, rbx
-
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d4_h8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_highbd_filter_block1d4_h8_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d4_h8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 7
- %define k0k6 [rsp + 16 * 0]
- %define k2k5 [rsp + 16 * 1]
- %define k3k4 [rsp + 16 * 2]
- %define k1k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define max [rsp + 16 * 5]
- %define min [rsp + 16 * 6]
-
- HIGH_GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm4, [rsi + 2]
- movdqa xmm1, xmm0
- movdqa xmm6, xmm4
- movdqa xmm7, xmm4
- movdqa xmm2, xmm0
- movdqa xmm3, xmm0
- movdqa xmm5, xmm4
-
- psrldq xmm1, 2
- psrldq xmm6, 4
- psrldq xmm7, 6
- psrldq xmm2, 4
- psrldq xmm3, 6
- psrldq xmm5, 2
-
- HIGH_APPLY_FILTER_4 0
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 7
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d8_h8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_highbd_filter_block1d8_h8_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d8_h8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm1, [rsi - 4]
- movdqu xmm2, [rsi - 2]
- movdqu xmm3, [rsi]
- movdqu xmm4, [rsi + 2]
- movdqu xmm5, [rsi + 4]
- movdqu xmm6, [rsi + 6]
- movdqu xmm7, [rsi + 8]
-
- HIGH_APPLY_FILTER_8 0, 0
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d16_h8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_highbd_filter_block1d16_h8_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d16_h8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm1, [rsi - 4]
- movdqu xmm2, [rsi - 2]
- movdqu xmm3, [rsi]
- movdqu xmm4, [rsi + 2]
- movdqu xmm5, [rsi + 4]
- movdqu xmm6, [rsi + 6]
- movdqu xmm7, [rsi + 8]
-
- HIGH_APPLY_FILTER_8 0, 0
-
- movdqu xmm0, [rsi + 10] ;load src
- movdqu xmm1, [rsi + 12]
- movdqu xmm2, [rsi + 14]
- movdqu xmm3, [rsi + 16]
- movdqu xmm4, [rsi + 18]
- movdqu xmm5, [rsi + 20]
- movdqu xmm6, [rsi + 22]
- movdqu xmm7, [rsi + 24]
-
- HIGH_APPLY_FILTER_8 0, 16
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d4_h8_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d4_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 7
- %define k0k6 [rsp + 16 * 0]
- %define k2k5 [rsp + 16 * 1]
- %define k3k4 [rsp + 16 * 2]
- %define k1k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define max [rsp + 16 * 5]
- %define min [rsp + 16 * 6]
-
- HIGH_GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm4, [rsi + 2]
- movdqa xmm1, xmm0
- movdqa xmm6, xmm4
- movdqa xmm7, xmm4
- movdqa xmm2, xmm0
- movdqa xmm3, xmm0
- movdqa xmm5, xmm4
-
- psrldq xmm1, 2
- psrldq xmm6, 4
- psrldq xmm7, 6
- psrldq xmm2, 4
- psrldq xmm3, 6
- psrldq xmm5, 2
-
- HIGH_APPLY_FILTER_4 1
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 7
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d8_h8_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d8_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm1, [rsi - 4]
- movdqu xmm2, [rsi - 2]
- movdqu xmm3, [rsi]
- movdqu xmm4, [rsi + 2]
- movdqu xmm5, [rsi + 4]
- movdqu xmm6, [rsi + 6]
- movdqu xmm7, [rsi + 8]
-
- HIGH_APPLY_FILTER_8 1, 0
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d16_h8_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d16_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 8
- %define k0k1 [rsp + 16 * 0]
- %define k6k7 [rsp + 16 * 1]
- %define k2k5 [rsp + 16 * 2]
- %define k3k4 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define temp [rsp + 16 * 5]
- %define max [rsp + 16 * 6]
- %define min [rsp + 16 * 7]
-
- HIGH_GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- lea rax, [rax + rax] ;bytes per line
- lea rdx, [rdx + rdx]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 6] ;load src
- movdqu xmm1, [rsi - 4]
- movdqu xmm2, [rsi - 2]
- movdqu xmm3, [rsi]
- movdqu xmm4, [rsi + 2]
- movdqu xmm5, [rsi + 4]
- movdqu xmm6, [rsi + 6]
- movdqu xmm7, [rsi + 8]
-
- HIGH_APPLY_FILTER_8 1, 0
-
- movdqu xmm0, [rsi + 10] ;load src
- movdqu xmm1, [rsi + 12]
- movdqu xmm2, [rsi + 14]
- movdqu xmm3, [rsi + 16]
- movdqu xmm4, [rsi + 18]
- movdqu xmm5, [rsi + 20]
- movdqu xmm6, [rsi + 22]
- movdqu xmm7, [rsi + 24]
-
- HIGH_APPLY_FILTER_8 1, 16
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 8
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
--- a/vp9/common/x86/vp9_high_subpixel_bilinear_sse2.asm
+++ /dev/null
@@ -1,494 +1,0 @@
-;
-; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-%include "vpx_ports/x86_abi_support.asm"
-
-%macro HIGH_GET_PARAM_4 0
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x00000040
-
- movdqa xmm3, [rdx] ;load filters
- pshuflw xmm4, xmm3, 11111111b ;k3
- psrldq xmm3, 8
- pshuflw xmm3, xmm3, 0b ;k4
- punpcklwd xmm4, xmm3 ;k3k4
-
- movq xmm3, rcx ;rounding
- pshufd xmm3, xmm3, 0
-
- mov rdx, 0x00010001
- movsxd rcx, DWORD PTR arg(6) ;bps
- movq xmm5, rdx
- movq xmm2, rcx
- pshufd xmm5, xmm5, 0b
- movdqa xmm1, xmm5
- psllw xmm5, xmm2
- psubw xmm5, xmm1 ;max value (for clamping)
- pxor xmm2, xmm2 ;min value (for clamping)
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-%endm
-
-%macro HIGH_APPLY_FILTER_4 1
-
- punpcklwd xmm0, xmm1 ;two row in one register
- pmaddwd xmm0, xmm4 ;multiply the filter factors
-
- paddd xmm0, xmm3 ;rounding
- psrad xmm0, 7 ;shift
- packssdw xmm0, xmm0 ;pack to word
-
- ;clamp the values
- pminsw xmm0, xmm5
- pmaxsw xmm0, xmm2
-
-%if %1
- movq xmm1, [rdi]
- pavgw xmm0, xmm1
-%endif
-
- movq [rdi], xmm0
- lea rsi, [rsi + 2*rax]
- lea rdi, [rdi + 2*rdx]
- dec rcx
-%endm
-
-%if ARCH_X86_64
-%macro HIGH_GET_PARAM 0
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x00000040
-
- movdqa xmm6, [rdx] ;load filters
-
- pshuflw xmm7, xmm6, 11111111b ;k3
- pshufhw xmm6, xmm6, 0b ;k4
- psrldq xmm6, 8
- punpcklwd xmm7, xmm6 ;k3k4k3k4k3k4k3k4
-
- movq xmm4, rcx ;rounding
- pshufd xmm4, xmm4, 0
-
- mov rdx, 0x00010001
- movsxd rcx, DWORD PTR arg(6) ;bps
- movq xmm8, rdx
- movq xmm5, rcx
- pshufd xmm8, xmm8, 0b
- movdqa xmm1, xmm8
- psllw xmm8, xmm5
- psubw xmm8, xmm1 ;max value (for clamping)
- pxor xmm5, xmm5 ;min value (for clamping)
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-%endm
-
-%macro HIGH_APPLY_FILTER_8 1
- movdqa xmm6, xmm0
- punpckhwd xmm6, xmm1
- punpcklwd xmm0, xmm1
- pmaddwd xmm6, xmm7
- pmaddwd xmm0, xmm7
-
- paddd xmm6, xmm4 ;rounding
- paddd xmm0, xmm4 ;rounding
- psrad xmm6, 7 ;shift
- psrad xmm0, 7 ;shift
- packssdw xmm0, xmm6 ;pack back to word
-
- ;clamp the values
- pminsw xmm0, xmm8
- pmaxsw xmm0, xmm5
-
-%if %1
- movdqu xmm1, [rdi]
- pavgw xmm0, xmm1
-%endif
- movdqu [rdi], xmm0 ;store the result
-
- lea rsi, [rsi + 2*rax]
- lea rdi, [rdi + 2*rdx]
- dec rcx
-%endm
-
-%macro HIGH_APPLY_FILTER_16 1
- movdqa xmm9, xmm0
- movdqa xmm6, xmm2
- punpckhwd xmm9, xmm1
- punpckhwd xmm6, xmm3
- punpcklwd xmm0, xmm1
- punpcklwd xmm2, xmm3
-
- pmaddwd xmm9, xmm7
- pmaddwd xmm6, xmm7
- pmaddwd xmm0, xmm7
- pmaddwd xmm2, xmm7
-
- paddd xmm9, xmm4 ;rounding
- paddd xmm6, xmm4
- paddd xmm0, xmm4
- paddd xmm2, xmm4
-
- psrad xmm9, 7 ;shift
- psrad xmm6, 7
- psrad xmm0, 7
- psrad xmm2, 7
-
- packssdw xmm0, xmm9 ;pack back to word
- packssdw xmm2, xmm6 ;pack back to word
-
- ;clamp the values
- pminsw xmm0, xmm8
- pmaxsw xmm0, xmm5
- pminsw xmm2, xmm8
- pmaxsw xmm2, xmm5
-
-%if %1
- movdqu xmm1, [rdi]
- movdqu xmm3, [rdi + 16]
- pavgw xmm0, xmm1
- pavgw xmm2, xmm3
-%endif
- movdqu [rdi], xmm0 ;store the result
- movdqu [rdi + 16], xmm2 ;store the result
-
- lea rsi, [rsi + 2*rax]
- lea rdi, [rdi + 2*rdx]
- dec rcx
-%endm
-%endif
-
-global sym(vp9_highbd_filter_block1d4_v2_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d4_v2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM_4
-.loop:
- movq xmm0, [rsi] ;load src
- movq xmm1, [rsi + 2*rax]
-
- HIGH_APPLY_FILTER_4 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-%if ARCH_X86_64
-global sym(vp9_highbd_filter_block1d8_v2_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d8_v2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 8
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + 2*rax] ;1
-
- HIGH_APPLY_FILTER_8 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d16_v2_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d16_v2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 9
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm2, [rsi + 16]
- movdqu xmm1, [rsi + 2*rax] ;1
- movdqu xmm3, [rsi + 2*rax + 16]
-
- HIGH_APPLY_FILTER_16 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-%endif
-
-global sym(vp9_highbd_filter_block1d4_v2_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d4_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM_4
-.loop:
- movq xmm0, [rsi] ;load src
- movq xmm1, [rsi + 2*rax]
-
- HIGH_APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-%if ARCH_X86_64
-global sym(vp9_highbd_filter_block1d8_v2_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d8_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 8
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + 2*rax] ;1
-
- HIGH_APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d16_v2_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d16_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 9
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + 2*rax] ;1
- movdqu xmm2, [rsi + 16]
- movdqu xmm3, [rsi + 2*rax + 16]
-
- HIGH_APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-%endif
-
-global sym(vp9_highbd_filter_block1d4_h2_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d4_h2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 2
-
- HIGH_APPLY_FILTER_4 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-%if ARCH_X86_64
-global sym(vp9_highbd_filter_block1d8_h2_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d8_h2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 8
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 2]
-
- HIGH_APPLY_FILTER_8 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d16_h2_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d16_h2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 9
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 2]
- movdqu xmm2, [rsi + 16]
- movdqu xmm3, [rsi + 18]
-
- HIGH_APPLY_FILTER_16 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-%endif
-
-global sym(vp9_highbd_filter_block1d4_h2_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d4_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 2
-
- HIGH_APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-%if ARCH_X86_64
-global sym(vp9_highbd_filter_block1d8_h2_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d8_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 8
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 2]
-
- HIGH_APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_highbd_filter_block1d16_h2_avg_sse2) PRIVATE
-sym(vp9_highbd_filter_block1d16_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 7
- SAVE_XMM 9
- push rsi
- push rdi
- ; end prolog
-
- HIGH_GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 2]
- movdqu xmm2, [rsi + 16]
- movdqu xmm3, [rsi + 18]
-
- HIGH_APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-%endif
--- a/vp9/common/x86/vp9_subpixel_8t_intrin_avx2.c
+++ /dev/null
@@ -1,602 +1,0 @@
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-// Due to a header conflict between math.h and intrinsics includes with ceil()
-// in certain configurations under vs9 this include needs to precede
-// immintrin.h.
-#include "./vp9_rtcd.h"
-
-#include <immintrin.h>
-
-#include "vp9/common/x86/convolve.h"
-#include "vpx_ports/mem.h"
-
-// filters for 16_h8 and 16_v8
-DECLARE_ALIGNED(32, static const uint8_t, filt1_global_avx2[32]) = {- 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
- 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
-};
-
-DECLARE_ALIGNED(32, static const uint8_t, filt2_global_avx2[32]) = {- 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10,
- 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
-};
-
-DECLARE_ALIGNED(32, static const uint8_t, filt3_global_avx2[32]) = {- 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12,
- 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
-};
-
-DECLARE_ALIGNED(32, static const uint8_t, filt4_global_avx2[32]) = {- 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14,
- 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
-};
-
-#if defined(__clang__)
-# if __clang_major__ < 3 || (__clang_major__ == 3 && __clang_minor__ <= 3) || \
- (defined(__APPLE__) && __clang_major__ == 5 && __clang_minor__ == 0)
-# define MM256_BROADCASTSI128_SI256(x) \
- _mm_broadcastsi128_si256((__m128i const *)&(x))
-# else // clang > 3.3, and not 5.0 on macosx.
-# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
-# endif // clang <= 3.3
-#elif defined(__GNUC__)
-# if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ <= 6)
-# define MM256_BROADCASTSI128_SI256(x) \
- _mm_broadcastsi128_si256((__m128i const *)&(x))
-# elif __GNUC__ == 4 && __GNUC_MINOR__ == 7
-# define MM256_BROADCASTSI128_SI256(x) _mm_broadcastsi128_si256(x)
-# else // gcc > 4.7
-# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
-# endif // gcc <= 4.6
-#else // !(gcc || clang)
-# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
-#endif // __clang__
-
-static void vp9_filter_block1d16_h8_avx2(const uint8_t *src_ptr,
- ptrdiff_t src_pixels_per_line,
- uint8_t *output_ptr,
- ptrdiff_t output_pitch,
- uint32_t output_height,
- const int16_t *filter) {- __m128i filtersReg;
- __m256i addFilterReg64, filt1Reg, filt2Reg, filt3Reg, filt4Reg;
- __m256i firstFilters, secondFilters, thirdFilters, forthFilters;
- __m256i srcRegFilt32b1_1, srcRegFilt32b2_1, srcRegFilt32b2, srcRegFilt32b3;
- __m256i srcReg32b1, srcReg32b2, filtersReg32;
- unsigned int i;
- ptrdiff_t src_stride, dst_stride;
-
- // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
- addFilterReg64 = _mm256_set1_epi32((int)0x0400040u);
- filtersReg = _mm_loadu_si128((const __m128i *)filter);
- // converting the 16 bit (short) to 8 bit (byte) and have the same data
- // in both lanes of 128 bit register.
- filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
- // have the same data in both lanes of a 256 bit register
- filtersReg32 = MM256_BROADCASTSI128_SI256(filtersReg);
-
- // duplicate only the first 16 bits (first and second byte)
- // across 256 bit register
- firstFilters = _mm256_shuffle_epi8(filtersReg32,
- _mm256_set1_epi16(0x100u));
- // duplicate only the second 16 bits (third and forth byte)
- // across 256 bit register
- secondFilters = _mm256_shuffle_epi8(filtersReg32,
- _mm256_set1_epi16(0x302u));
- // duplicate only the third 16 bits (fifth and sixth byte)
- // across 256 bit register
- thirdFilters = _mm256_shuffle_epi8(filtersReg32,
- _mm256_set1_epi16(0x504u));
- // duplicate only the forth 16 bits (seventh and eighth byte)
- // across 256 bit register
- forthFilters = _mm256_shuffle_epi8(filtersReg32,
- _mm256_set1_epi16(0x706u));
-
- filt1Reg = _mm256_load_si256((__m256i const *)filt1_global_avx2);
- filt2Reg = _mm256_load_si256((__m256i const *)filt2_global_avx2);
- filt3Reg = _mm256_load_si256((__m256i const *)filt3_global_avx2);
- filt4Reg = _mm256_load_si256((__m256i const *)filt4_global_avx2);
-
- // multiple the size of the source and destination stride by two
- src_stride = src_pixels_per_line << 1;
- dst_stride = output_pitch << 1;
- for (i = output_height; i > 1; i-=2) {- // load the 2 strides of source
- srcReg32b1 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr - 3)));
- srcReg32b1 = _mm256_inserti128_si256(srcReg32b1,
- _mm_loadu_si128((const __m128i *)
- (src_ptr+src_pixels_per_line-3)), 1);
-
- // filter the source buffer
- srcRegFilt32b1_1= _mm256_shuffle_epi8(srcReg32b1, filt1Reg);
- srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt4Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt32b1_1 = _mm256_maddubs_epi16(srcRegFilt32b1_1, firstFilters);
- srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, forthFilters);
-
- // add and saturate the results together
- srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, srcRegFilt32b2);
-
- // filter the source buffer
- srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b1, filt2Reg);
- srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt3Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, secondFilters);
- srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, thirdFilters);
-
- // add and saturate the results together
- srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1,
- _mm256_min_epi16(srcRegFilt32b3, srcRegFilt32b2));
-
- // reading 2 strides of the next 16 bytes
- // (part of it was being read by earlier read)
- srcReg32b2 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + 5)));
- srcReg32b2 = _mm256_inserti128_si256(srcReg32b2,
- _mm_loadu_si128((const __m128i *)
- (src_ptr+src_pixels_per_line+5)), 1);
-
- // add and saturate the results together
- srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1,
- _mm256_max_epi16(srcRegFilt32b3, srcRegFilt32b2));
-
- // filter the source buffer
- srcRegFilt32b2_1 = _mm256_shuffle_epi8(srcReg32b2, filt1Reg);
- srcRegFilt32b2 = _mm256_shuffle_epi8(srcReg32b2, filt4Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt32b2_1 = _mm256_maddubs_epi16(srcRegFilt32b2_1, firstFilters);
- srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, forthFilters);
-
- // add and saturate the results together
- srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, srcRegFilt32b2);
-
- // filter the source buffer
- srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b2, filt2Reg);
- srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b2, filt3Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, secondFilters);
- srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, thirdFilters);
-
- // add and saturate the results together
- srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1,
- _mm256_min_epi16(srcRegFilt32b3, srcRegFilt32b2));
- srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1,
- _mm256_max_epi16(srcRegFilt32b3, srcRegFilt32b2));
-
-
- srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, addFilterReg64);
-
- srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, addFilterReg64);
-
- // shift by 7 bit each 16 bit
- srcRegFilt32b1_1 = _mm256_srai_epi16(srcRegFilt32b1_1, 7);
- srcRegFilt32b2_1 = _mm256_srai_epi16(srcRegFilt32b2_1, 7);
-
- // shrink to 8 bit each 16 bits, the first lane contain the first
- // convolve result and the second lane contain the second convolve
- // result
- srcRegFilt32b1_1 = _mm256_packus_epi16(srcRegFilt32b1_1,
- srcRegFilt32b2_1);
-
- src_ptr+=src_stride;
-
- // save 16 bytes
- _mm_store_si128((__m128i*)output_ptr,
- _mm256_castsi256_si128(srcRegFilt32b1_1));
-
- // save the next 16 bits
- _mm_store_si128((__m128i*)(output_ptr+output_pitch),
- _mm256_extractf128_si256(srcRegFilt32b1_1, 1));
- output_ptr+=dst_stride;
- }
-
- // if the number of strides is odd.
- // process only 16 bytes
- if (i > 0) {- __m128i srcReg1, srcReg2, srcRegFilt1_1, srcRegFilt2_1;
- __m128i srcRegFilt2, srcRegFilt3;
-
- srcReg1 = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
-
- // filter the source buffer
- srcRegFilt1_1 = _mm_shuffle_epi8(srcReg1,
- _mm256_castsi256_si128(filt1Reg));
- srcRegFilt2 = _mm_shuffle_epi8(srcReg1,
- _mm256_castsi256_si128(filt4Reg));
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt1_1 = _mm_maddubs_epi16(srcRegFilt1_1,
- _mm256_castsi256_si128(firstFilters));
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
- _mm256_castsi256_si128(forthFilters));
-
- // add and saturate the results together
- srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, srcRegFilt2);
-
- // filter the source buffer
- srcRegFilt3= _mm_shuffle_epi8(srcReg1,
- _mm256_castsi256_si128(filt2Reg));
- srcRegFilt2= _mm_shuffle_epi8(srcReg1,
- _mm256_castsi256_si128(filt3Reg));
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3,
- _mm256_castsi256_si128(secondFilters));
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
- _mm256_castsi256_si128(thirdFilters));
-
- // add and saturate the results together
- srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
- _mm_min_epi16(srcRegFilt3, srcRegFilt2));
-
- // reading the next 16 bytes
- // (part of it was being read by earlier read)
- srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + 5));
-
- // add and saturate the results together
- srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
- _mm_max_epi16(srcRegFilt3, srcRegFilt2));
-
- // filter the source buffer
- srcRegFilt2_1 = _mm_shuffle_epi8(srcReg2,
- _mm256_castsi256_si128(filt1Reg));
- srcRegFilt2 = _mm_shuffle_epi8(srcReg2,
- _mm256_castsi256_si128(filt4Reg));
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt2_1 = _mm_maddubs_epi16(srcRegFilt2_1,
- _mm256_castsi256_si128(firstFilters));
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
- _mm256_castsi256_si128(forthFilters));
-
- // add and saturate the results together
- srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, srcRegFilt2);
-
- // filter the source buffer
- srcRegFilt3 = _mm_shuffle_epi8(srcReg2,
- _mm256_castsi256_si128(filt2Reg));
- srcRegFilt2 = _mm_shuffle_epi8(srcReg2,
- _mm256_castsi256_si128(filt3Reg));
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3,
- _mm256_castsi256_si128(secondFilters));
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
- _mm256_castsi256_si128(thirdFilters));
-
- // add and saturate the results together
- srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
- _mm_min_epi16(srcRegFilt3, srcRegFilt2));
- srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
- _mm_max_epi16(srcRegFilt3, srcRegFilt2));
-
-
- srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
- _mm256_castsi256_si128(addFilterReg64));
-
- srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
- _mm256_castsi256_si128(addFilterReg64));
-
- // shift by 7 bit each 16 bit
- srcRegFilt1_1 = _mm_srai_epi16(srcRegFilt1_1, 7);
- srcRegFilt2_1 = _mm_srai_epi16(srcRegFilt2_1, 7);
-
- // shrink to 8 bit each 16 bits, the first lane contain the first
- // convolve result and the second lane contain the second convolve
- // result
- srcRegFilt1_1 = _mm_packus_epi16(srcRegFilt1_1, srcRegFilt2_1);
-
- // save 16 bytes
- _mm_store_si128((__m128i*)output_ptr, srcRegFilt1_1);
- }
-}
-
-static void vp9_filter_block1d16_v8_avx2(const uint8_t *src_ptr,
- ptrdiff_t src_pitch,
- uint8_t *output_ptr,
- ptrdiff_t out_pitch,
- uint32_t output_height,
- const int16_t *filter) {- __m128i filtersReg;
- __m256i addFilterReg64;
- __m256i srcReg32b1, srcReg32b2, srcReg32b3, srcReg32b4, srcReg32b5;
- __m256i srcReg32b6, srcReg32b7, srcReg32b8, srcReg32b9, srcReg32b10;
- __m256i srcReg32b11, srcReg32b12, filtersReg32;
- __m256i firstFilters, secondFilters, thirdFilters, forthFilters;
- unsigned int i;
- ptrdiff_t src_stride, dst_stride;
-
- // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
- addFilterReg64 = _mm256_set1_epi32((int)0x0400040u);
- filtersReg = _mm_loadu_si128((const __m128i *)filter);
- // converting the 16 bit (short) to 8 bit (byte) and have the
- // same data in both lanes of 128 bit register.
- filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
- // have the same data in both lanes of a 256 bit register
- filtersReg32 = MM256_BROADCASTSI128_SI256(filtersReg);
-
- // duplicate only the first 16 bits (first and second byte)
- // across 256 bit register
- firstFilters = _mm256_shuffle_epi8(filtersReg32,
- _mm256_set1_epi16(0x100u));
- // duplicate only the second 16 bits (third and forth byte)
- // across 256 bit register
- secondFilters = _mm256_shuffle_epi8(filtersReg32,
- _mm256_set1_epi16(0x302u));
- // duplicate only the third 16 bits (fifth and sixth byte)
- // across 256 bit register
- thirdFilters = _mm256_shuffle_epi8(filtersReg32,
- _mm256_set1_epi16(0x504u));
- // duplicate only the forth 16 bits (seventh and eighth byte)
- // across 256 bit register
- forthFilters = _mm256_shuffle_epi8(filtersReg32,
- _mm256_set1_epi16(0x706u));
-
- // multiple the size of the source and destination stride by two
- src_stride = src_pitch << 1;
- dst_stride = out_pitch << 1;
-
- // load 16 bytes 7 times in stride of src_pitch
- srcReg32b1 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr)));
- srcReg32b2 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch)));
- srcReg32b3 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2)));
- srcReg32b4 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3)));
- srcReg32b5 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4)));
- srcReg32b6 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5)));
- srcReg32b7 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6)));
-
- // have each consecutive loads on the same 256 register
- srcReg32b1 = _mm256_inserti128_si256(srcReg32b1,
- _mm256_castsi256_si128(srcReg32b2), 1);
- srcReg32b2 = _mm256_inserti128_si256(srcReg32b2,
- _mm256_castsi256_si128(srcReg32b3), 1);
- srcReg32b3 = _mm256_inserti128_si256(srcReg32b3,
- _mm256_castsi256_si128(srcReg32b4), 1);
- srcReg32b4 = _mm256_inserti128_si256(srcReg32b4,
- _mm256_castsi256_si128(srcReg32b5), 1);
- srcReg32b5 = _mm256_inserti128_si256(srcReg32b5,
- _mm256_castsi256_si128(srcReg32b6), 1);
- srcReg32b6 = _mm256_inserti128_si256(srcReg32b6,
- _mm256_castsi256_si128(srcReg32b7), 1);
-
- // merge every two consecutive registers except the last one
- srcReg32b10 = _mm256_unpacklo_epi8(srcReg32b1, srcReg32b2);
- srcReg32b1 = _mm256_unpackhi_epi8(srcReg32b1, srcReg32b2);
-
- // save
- srcReg32b11 = _mm256_unpacklo_epi8(srcReg32b3, srcReg32b4);
-
- // save
- srcReg32b3 = _mm256_unpackhi_epi8(srcReg32b3, srcReg32b4);
-
- // save
- srcReg32b2 = _mm256_unpacklo_epi8(srcReg32b5, srcReg32b6);
-
- // save
- srcReg32b5 = _mm256_unpackhi_epi8(srcReg32b5, srcReg32b6);
-
-
- for (i = output_height; i > 1; i-=2) {- // load the last 2 loads of 16 bytes and have every two
- // consecutive loads in the same 256 bit register
- srcReg32b8 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7)));
- srcReg32b7 = _mm256_inserti128_si256(srcReg32b7,
- _mm256_castsi256_si128(srcReg32b8), 1);
- srcReg32b9 = _mm256_castsi128_si256(
- _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 8)));
- srcReg32b8 = _mm256_inserti128_si256(srcReg32b8,
- _mm256_castsi256_si128(srcReg32b9), 1);
-
- // merge every two consecutive registers
- // save
- srcReg32b4 = _mm256_unpacklo_epi8(srcReg32b7, srcReg32b8);
- srcReg32b7 = _mm256_unpackhi_epi8(srcReg32b7, srcReg32b8);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcReg32b10 = _mm256_maddubs_epi16(srcReg32b10, firstFilters);
- srcReg32b6 = _mm256_maddubs_epi16(srcReg32b4, forthFilters);
-
- // add and saturate the results together
- srcReg32b10 = _mm256_adds_epi16(srcReg32b10, srcReg32b6);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcReg32b8 = _mm256_maddubs_epi16(srcReg32b11, secondFilters);
- srcReg32b12 = _mm256_maddubs_epi16(srcReg32b2, thirdFilters);
-
- // add and saturate the results together
- srcReg32b10 = _mm256_adds_epi16(srcReg32b10,
- _mm256_min_epi16(srcReg32b8, srcReg32b12));
- srcReg32b10 = _mm256_adds_epi16(srcReg32b10,
- _mm256_max_epi16(srcReg32b8, srcReg32b12));
-
- // multiply 2 adjacent elements with the filter and add the result
- srcReg32b1 = _mm256_maddubs_epi16(srcReg32b1, firstFilters);
- srcReg32b6 = _mm256_maddubs_epi16(srcReg32b7, forthFilters);
-
- srcReg32b1 = _mm256_adds_epi16(srcReg32b1, srcReg32b6);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcReg32b8 = _mm256_maddubs_epi16(srcReg32b3, secondFilters);
- srcReg32b12 = _mm256_maddubs_epi16(srcReg32b5, thirdFilters);
-
- // add and saturate the results together
- srcReg32b1 = _mm256_adds_epi16(srcReg32b1,
- _mm256_min_epi16(srcReg32b8, srcReg32b12));
- srcReg32b1 = _mm256_adds_epi16(srcReg32b1,
- _mm256_max_epi16(srcReg32b8, srcReg32b12));
-
- srcReg32b10 = _mm256_adds_epi16(srcReg32b10, addFilterReg64);
- srcReg32b1 = _mm256_adds_epi16(srcReg32b1, addFilterReg64);
-
- // shift by 7 bit each 16 bit
- srcReg32b10 = _mm256_srai_epi16(srcReg32b10, 7);
- srcReg32b1 = _mm256_srai_epi16(srcReg32b1, 7);
-
- // shrink to 8 bit each 16 bits, the first lane contain the first
- // convolve result and the second lane contain the second convolve
- // result
- srcReg32b1 = _mm256_packus_epi16(srcReg32b10, srcReg32b1);
-
- src_ptr+=src_stride;
-
- // save 16 bytes
- _mm_store_si128((__m128i*)output_ptr,
- _mm256_castsi256_si128(srcReg32b1));
-
- // save the next 16 bits
- _mm_store_si128((__m128i*)(output_ptr+out_pitch),
- _mm256_extractf128_si256(srcReg32b1, 1));
-
- output_ptr+=dst_stride;
-
- // save part of the registers for next strides
- srcReg32b10 = srcReg32b11;
- srcReg32b1 = srcReg32b3;
- srcReg32b11 = srcReg32b2;
- srcReg32b3 = srcReg32b5;
- srcReg32b2 = srcReg32b4;
- srcReg32b5 = srcReg32b7;
- srcReg32b7 = srcReg32b9;
- }
- if (i > 0) {- __m128i srcRegFilt1, srcRegFilt3, srcRegFilt4, srcRegFilt5;
- __m128i srcRegFilt6, srcRegFilt7, srcRegFilt8;
- // load the last 16 bytes
- srcRegFilt8 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7));
-
- // merge the last 2 results together
- srcRegFilt4 = _mm_unpacklo_epi8(
- _mm256_castsi256_si128(srcReg32b7), srcRegFilt8);
- srcRegFilt7 = _mm_unpackhi_epi8(
- _mm256_castsi256_si128(srcReg32b7), srcRegFilt8);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt1 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b10),
- _mm256_castsi256_si128(firstFilters));
- srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4,
- _mm256_castsi256_si128(forthFilters));
- srcRegFilt3 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b1),
- _mm256_castsi256_si128(firstFilters));
- srcRegFilt7 = _mm_maddubs_epi16(srcRegFilt7,
- _mm256_castsi256_si128(forthFilters));
-
- // add and saturate the results together
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
- srcRegFilt3 = _mm_adds_epi16(srcRegFilt3, srcRegFilt7);
-
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt4 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b11),
- _mm256_castsi256_si128(secondFilters));
- srcRegFilt5 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b3),
- _mm256_castsi256_si128(secondFilters));
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt6 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b2),
- _mm256_castsi256_si128(thirdFilters));
- srcRegFilt7 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b5),
- _mm256_castsi256_si128(thirdFilters));
-
- // add and saturate the results together
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
- _mm_min_epi16(srcRegFilt4, srcRegFilt6));
- srcRegFilt3 = _mm_adds_epi16(srcRegFilt3,
- _mm_min_epi16(srcRegFilt5, srcRegFilt7));
-
- // add and saturate the results together
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
- _mm_max_epi16(srcRegFilt4, srcRegFilt6));
- srcRegFilt3 = _mm_adds_epi16(srcRegFilt3,
- _mm_max_epi16(srcRegFilt5, srcRegFilt7));
-
-
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
- _mm256_castsi256_si128(addFilterReg64));
- srcRegFilt3 = _mm_adds_epi16(srcRegFilt3,
- _mm256_castsi256_si128(addFilterReg64));
-
- // shift by 7 bit each 16 bit
- srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
- srcRegFilt3 = _mm_srai_epi16(srcRegFilt3, 7);
-
- // shrink to 8 bit each 16 bits, the first lane contain the first
- // convolve result and the second lane contain the second convolve
- // result
- srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt3);
-
- // save 16 bytes
- _mm_store_si128((__m128i*)output_ptr, srcRegFilt1);
- }
-}
-
-#if HAVE_AVX2 && HAVE_SSSE3
-filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
-#if ARCH_X86_64
-filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h8_intrin_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h8_intrin_ssse3;
-#define vp9_filter_block1d8_v8_avx2 vp9_filter_block1d8_v8_intrin_ssse3
-#define vp9_filter_block1d8_h8_avx2 vp9_filter_block1d8_h8_intrin_ssse3
-#define vp9_filter_block1d4_h8_avx2 vp9_filter_block1d4_h8_intrin_ssse3
-#else // ARCH_X86
-filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
-#define vp9_filter_block1d8_v8_avx2 vp9_filter_block1d8_v8_ssse3
-#define vp9_filter_block1d8_h8_avx2 vp9_filter_block1d8_h8_ssse3
-#define vp9_filter_block1d4_h8_avx2 vp9_filter_block1d4_h8_ssse3
-#endif // ARCH_X86_64
-filter8_1dfunction vp9_filter_block1d16_v2_ssse3;
-filter8_1dfunction vp9_filter_block1d16_h2_ssse3;
-filter8_1dfunction vp9_filter_block1d8_v2_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h2_ssse3;
-filter8_1dfunction vp9_filter_block1d4_v2_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h2_ssse3;
-#define vp9_filter_block1d4_v8_avx2 vp9_filter_block1d4_v8_ssse3
-#define vp9_filter_block1d16_v2_avx2 vp9_filter_block1d16_v2_ssse3
-#define vp9_filter_block1d16_h2_avx2 vp9_filter_block1d16_h2_ssse3
-#define vp9_filter_block1d8_v2_avx2 vp9_filter_block1d8_v2_ssse3
-#define vp9_filter_block1d8_h2_avx2 vp9_filter_block1d8_h2_ssse3
-#define vp9_filter_block1d4_v2_avx2 vp9_filter_block1d4_v2_ssse3
-#define vp9_filter_block1d4_h2_avx2 vp9_filter_block1d4_h2_ssse3
-// void vp9_convolve8_horiz_avx2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_vert_avx2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , avx2);
-FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , avx2);
-
-// void vp9_convolve8_avx2(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-FUN_CONV_2D(, avx2);
-#endif // HAVE_AX2 && HAVE_SSSE3
--- a/vp9/common/x86/vp9_subpixel_8t_intrin_ssse3.c
+++ /dev/null
@@ -1,601 +1,0 @@
-/*
- * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
- *
- * Use of this source code is governed by a BSD-style license
- * that can be found in the LICENSE file in the root of the source
- * tree. An additional intellectual property rights grant can be found
- * in the file PATENTS. All contributing project authors may
- * be found in the AUTHORS file in the root of the source tree.
- */
-
-// Due to a header conflict between math.h and intrinsics includes with ceil()
-// in certain configurations under vs9 this include needs to precede
-// tmmintrin.h.
-#include "./vp9_rtcd.h"
-
-#include <tmmintrin.h>
-
-#include "vp9/common/x86/convolve.h"
-#include "vpx_ports/mem.h"
-#include "vpx_ports/emmintrin_compat.h"
-
-// filters only for the 4_h8 convolution
-DECLARE_ALIGNED(16, static const uint8_t, filt1_4_h8[16]) = {- 0, 1, 1, 2, 2, 3, 3, 4, 2, 3, 3, 4, 4, 5, 5, 6
-};
-
-DECLARE_ALIGNED(16, static const uint8_t, filt2_4_h8[16]) = {- 4, 5, 5, 6, 6, 7, 7, 8, 6, 7, 7, 8, 8, 9, 9, 10
-};
-
-// filters for 8_h8 and 16_h8
-DECLARE_ALIGNED(16, static const uint8_t, filt1_global[16]) = {- 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
-};
-
-DECLARE_ALIGNED(16, static const uint8_t, filt2_global[16]) = {- 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
-};
-
-DECLARE_ALIGNED(16, static const uint8_t, filt3_global[16]) = {- 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
-};
-
-DECLARE_ALIGNED(16, static const uint8_t, filt4_global[16]) = {- 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
-};
-
-// These are reused by the avx2 intrinsics.
-filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h8_intrin_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h8_intrin_ssse3;
-
-void vp9_filter_block1d4_h8_intrin_ssse3(const uint8_t *src_ptr,
- ptrdiff_t src_pixels_per_line,
- uint8_t *output_ptr,
- ptrdiff_t output_pitch,
- uint32_t output_height,
- const int16_t *filter) {- __m128i firstFilters, secondFilters, shuffle1, shuffle2;
- __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4;
- __m128i addFilterReg64, filtersReg, srcReg, minReg;
- unsigned int i;
-
- // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
- addFilterReg64 =_mm_set1_epi32((int)0x0400040u);
- filtersReg = _mm_loadu_si128((const __m128i *)filter);
- // converting the 16 bit (short) to 8 bit (byte) and have the same data
- // in both lanes of 128 bit register.
- filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
-
- // duplicate only the first 16 bits in the filter into the first lane
- firstFilters = _mm_shufflelo_epi16(filtersReg, 0);
- // duplicate only the third 16 bit in the filter into the first lane
- secondFilters = _mm_shufflelo_epi16(filtersReg, 0xAAu);
- // duplicate only the seconds 16 bits in the filter into the second lane
- // firstFilters: k0 k1 k0 k1 k0 k1 k0 k1 k2 k3 k2 k3 k2 k3 k2 k3
- firstFilters = _mm_shufflehi_epi16(firstFilters, 0x55u);
- // duplicate only the forth 16 bits in the filter into the second lane
- // secondFilters: k4 k5 k4 k5 k4 k5 k4 k5 k6 k7 k6 k7 k6 k7 k6 k7
- secondFilters = _mm_shufflehi_epi16(secondFilters, 0xFFu);
-
- // loading the local filters
- shuffle1 =_mm_load_si128((__m128i const *)filt1_4_h8);
- shuffle2 = _mm_load_si128((__m128i const *)filt2_4_h8);
-
- for (i = 0; i < output_height; i++) {- srcReg = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
-
- // filter the source buffer
- srcRegFilt1= _mm_shuffle_epi8(srcReg, shuffle1);
- srcRegFilt2= _mm_shuffle_epi8(srcReg, shuffle2);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters);
-
- // extract the higher half of the lane
- srcRegFilt3 = _mm_srli_si128(srcRegFilt1, 8);
- srcRegFilt4 = _mm_srli_si128(srcRegFilt2, 8);
-
- minReg = _mm_min_epi16(srcRegFilt3, srcRegFilt2);
-
- // add and saturate all the results together
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
- srcRegFilt3 = _mm_max_epi16(srcRegFilt3, srcRegFilt2);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt3);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
-
- // shift by 7 bit each 16 bits
- srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
-
- // shrink to 8 bit each 16 bits
- srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
- src_ptr+=src_pixels_per_line;
-
- // save only 4 bytes
- *((int*)&output_ptr[0])= _mm_cvtsi128_si32(srcRegFilt1);
-
- output_ptr+=output_pitch;
- }
-}
-
-void vp9_filter_block1d8_h8_intrin_ssse3(const uint8_t *src_ptr,
- ptrdiff_t src_pixels_per_line,
- uint8_t *output_ptr,
- ptrdiff_t output_pitch,
- uint32_t output_height,
- const int16_t *filter) {- __m128i firstFilters, secondFilters, thirdFilters, forthFilters, srcReg;
- __m128i filt1Reg, filt2Reg, filt3Reg, filt4Reg;
- __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4;
- __m128i addFilterReg64, filtersReg, minReg;
- unsigned int i;
-
- // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
- addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
- filtersReg = _mm_loadu_si128((const __m128i *)filter);
- // converting the 16 bit (short) to 8 bit (byte) and have the same data
- // in both lanes of 128 bit register.
- filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
-
- // duplicate only the first 16 bits (first and second byte)
- // across 128 bit register
- firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
- // duplicate only the second 16 bits (third and forth byte)
- // across 128 bit register
- secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
- // duplicate only the third 16 bits (fifth and sixth byte)
- // across 128 bit register
- thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
- // duplicate only the forth 16 bits (seventh and eighth byte)
- // across 128 bit register
- forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
-
- filt1Reg = _mm_load_si128((__m128i const *)filt1_global);
- filt2Reg = _mm_load_si128((__m128i const *)filt2_global);
- filt3Reg = _mm_load_si128((__m128i const *)filt3_global);
- filt4Reg = _mm_load_si128((__m128i const *)filt4_global);
-
- for (i = 0; i < output_height; i++) {- srcReg = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
-
- // filter the source buffer
- srcRegFilt1= _mm_shuffle_epi8(srcReg, filt1Reg);
- srcRegFilt2= _mm_shuffle_epi8(srcReg, filt2Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters);
-
- // filter the source buffer
- srcRegFilt3= _mm_shuffle_epi8(srcReg, filt3Reg);
- srcRegFilt4= _mm_shuffle_epi8(srcReg, filt4Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, thirdFilters);
- srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4, forthFilters);
-
- // add and saturate all the results together
- minReg = _mm_min_epi16(srcRegFilt2, srcRegFilt3);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
-
- srcRegFilt2= _mm_max_epi16(srcRegFilt2, srcRegFilt3);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
-
- // shift by 7 bit each 16 bits
- srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
-
- // shrink to 8 bit each 16 bits
- srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
-
- src_ptr+=src_pixels_per_line;
-
- // save only 8 bytes
- _mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1);
-
- output_ptr+=output_pitch;
- }
-}
-
-static void vp9_filter_block1d16_h8_intrin_ssse3(const uint8_t *src_ptr,
- ptrdiff_t src_pixels_per_line,
- uint8_t *output_ptr,
- ptrdiff_t output_pitch,
- uint32_t output_height,
- const int16_t *filter) {- __m128i addFilterReg64, filtersReg, srcReg1, srcReg2;
- __m128i filt1Reg, filt2Reg, filt3Reg, filt4Reg;
- __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
- __m128i srcRegFilt1_1, srcRegFilt2_1, srcRegFilt2, srcRegFilt3;
- unsigned int i;
-
- // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
- addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
- filtersReg = _mm_loadu_si128((const __m128i *)filter);
- // converting the 16 bit (short) to 8 bit (byte) and have the same data
- // in both lanes of 128 bit register.
- filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
-
- // duplicate only the first 16 bits (first and second byte)
- // across 128 bit register
- firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
- // duplicate only the second 16 bits (third and forth byte)
- // across 128 bit register
- secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
- // duplicate only the third 16 bits (fifth and sixth byte)
- // across 128 bit register
- thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
- // duplicate only the forth 16 bits (seventh and eighth byte)
- // across 128 bit register
- forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
-
- filt1Reg = _mm_load_si128((__m128i const *)filt1_global);
- filt2Reg = _mm_load_si128((__m128i const *)filt2_global);
- filt3Reg = _mm_load_si128((__m128i const *)filt3_global);
- filt4Reg = _mm_load_si128((__m128i const *)filt4_global);
-
- for (i = 0; i < output_height; i++) {- srcReg1 = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
-
- // filter the source buffer
- srcRegFilt1_1= _mm_shuffle_epi8(srcReg1, filt1Reg);
- srcRegFilt2= _mm_shuffle_epi8(srcReg1, filt4Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt1_1 = _mm_maddubs_epi16(srcRegFilt1_1, firstFilters);
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, forthFilters);
-
- // add and saturate the results together
- srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, srcRegFilt2);
-
- // filter the source buffer
- srcRegFilt3= _mm_shuffle_epi8(srcReg1, filt2Reg);
- srcRegFilt2= _mm_shuffle_epi8(srcReg1, filt3Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
-
- // add and saturate the results together
- srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
- _mm_min_epi16(srcRegFilt3, srcRegFilt2));
-
- // reading the next 16 bytes.
- // (part of it was being read by earlier read)
- srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + 5));
-
- // add and saturate the results together
- srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
- _mm_max_epi16(srcRegFilt3, srcRegFilt2));
-
- // filter the source buffer
- srcRegFilt2_1= _mm_shuffle_epi8(srcReg2, filt1Reg);
- srcRegFilt2= _mm_shuffle_epi8(srcReg2, filt4Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt2_1 = _mm_maddubs_epi16(srcRegFilt2_1, firstFilters);
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, forthFilters);
-
- // add and saturate the results together
- srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, srcRegFilt2);
-
- // filter the source buffer
- srcRegFilt3= _mm_shuffle_epi8(srcReg2, filt2Reg);
- srcRegFilt2= _mm_shuffle_epi8(srcReg2, filt3Reg);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
-
- // add and saturate the results together
- srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
- _mm_min_epi16(srcRegFilt3, srcRegFilt2));
- srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
- _mm_max_epi16(srcRegFilt3, srcRegFilt2));
-
- srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, addFilterReg64);
- srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, addFilterReg64);
-
- // shift by 7 bit each 16 bit
- srcRegFilt1_1 = _mm_srai_epi16(srcRegFilt1_1, 7);
- srcRegFilt2_1 = _mm_srai_epi16(srcRegFilt2_1, 7);
-
- // shrink to 8 bit each 16 bits, the first lane contain the first
- // convolve result and the second lane contain the second convolve
- // result
- srcRegFilt1_1 = _mm_packus_epi16(srcRegFilt1_1, srcRegFilt2_1);
-
- src_ptr+=src_pixels_per_line;
-
- // save 16 bytes
- _mm_store_si128((__m128i*)output_ptr, srcRegFilt1_1);
-
- output_ptr+=output_pitch;
- }
-}
-
-void vp9_filter_block1d8_v8_intrin_ssse3(const uint8_t *src_ptr,
- ptrdiff_t src_pitch,
- uint8_t *output_ptr,
- ptrdiff_t out_pitch,
- uint32_t output_height,
- const int16_t *filter) {- __m128i addFilterReg64, filtersReg, minReg;
- __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
- __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt5;
- __m128i srcReg1, srcReg2, srcReg3, srcReg4, srcReg5, srcReg6, srcReg7;
- __m128i srcReg8;
- unsigned int i;
-
- // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
- addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
- filtersReg = _mm_loadu_si128((const __m128i *)filter);
- // converting the 16 bit (short) to 8 bit (byte) and have the same data
- // in both lanes of 128 bit register.
- filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
-
- // duplicate only the first 16 bits in the filter
- firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
- // duplicate only the second 16 bits in the filter
- secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
- // duplicate only the third 16 bits in the filter
- thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
- // duplicate only the forth 16 bits in the filter
- forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
-
- // load the first 7 rows of 8 bytes
- srcReg1 = _mm_loadl_epi64((const __m128i *)src_ptr);
- srcReg2 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch));
- srcReg3 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
- srcReg4 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
- srcReg5 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4));
- srcReg6 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5));
- srcReg7 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6));
-
- for (i = 0; i < output_height; i++) {- // load the last 8 bytes
- srcReg8 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 7));
-
- // merge the result together
- srcRegFilt1 = _mm_unpacklo_epi8(srcReg1, srcReg2);
- srcRegFilt3 = _mm_unpacklo_epi8(srcReg3, srcReg4);
-
- // merge the result together
- srcRegFilt2 = _mm_unpacklo_epi8(srcReg5, srcReg6);
- srcRegFilt5 = _mm_unpacklo_epi8(srcReg7, srcReg8);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
- srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
- srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
- srcRegFilt5 = _mm_maddubs_epi16(srcRegFilt5, forthFilters);
-
- // add and saturate the results together
- minReg = _mm_min_epi16(srcRegFilt2, srcRegFilt3);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt5);
- srcRegFilt2 = _mm_max_epi16(srcRegFilt2, srcRegFilt3);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
-
- // shift by 7 bit each 16 bit
- srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
-
- // shrink to 8 bit each 16 bits
- srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
-
- src_ptr+=src_pitch;
-
- // shift down a row
- srcReg1 = srcReg2;
- srcReg2 = srcReg3;
- srcReg3 = srcReg4;
- srcReg4 = srcReg5;
- srcReg5 = srcReg6;
- srcReg6 = srcReg7;
- srcReg7 = srcReg8;
-
- // save only 8 bytes convolve result
- _mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1);
-
- output_ptr+=out_pitch;
- }
-}
-
-static void vp9_filter_block1d16_v8_intrin_ssse3(const uint8_t *src_ptr,
- ptrdiff_t src_pitch,
- uint8_t *output_ptr,
- ptrdiff_t out_pitch,
- uint32_t output_height,
- const int16_t *filter) {- __m128i addFilterReg64, filtersReg, srcRegFilt1, srcRegFilt3;
- __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
- __m128i srcRegFilt5, srcRegFilt6, srcRegFilt7, srcRegFilt8;
- __m128i srcReg1, srcReg2, srcReg3, srcReg4, srcReg5, srcReg6, srcReg7;
- __m128i srcReg8;
- unsigned int i;
-
- // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
- addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
- filtersReg = _mm_loadu_si128((const __m128i *)filter);
- // converting the 16 bit (short) to 8 bit (byte) and have the same data
- // in both lanes of 128 bit register.
- filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
-
- // duplicate only the first 16 bits in the filter
- firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
- // duplicate only the second 16 bits in the filter
- secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
- // duplicate only the third 16 bits in the filter
- thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
- // duplicate only the forth 16 bits in the filter
- forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
-
- // load the first 7 rows of 16 bytes
- srcReg1 = _mm_loadu_si128((const __m128i *)(src_ptr));
- srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch));
- srcReg3 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
- srcReg4 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
- srcReg5 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4));
- srcReg6 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5));
- srcReg7 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6));
-
- for (i = 0; i < output_height; i++) {- // load the last 16 bytes
- srcReg8 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7));
-
- // merge the result together
- srcRegFilt5 = _mm_unpacklo_epi8(srcReg1, srcReg2);
- srcRegFilt6 = _mm_unpacklo_epi8(srcReg7, srcReg8);
- srcRegFilt1 = _mm_unpackhi_epi8(srcReg1, srcReg2);
- srcRegFilt3 = _mm_unpackhi_epi8(srcReg7, srcReg8);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt5 = _mm_maddubs_epi16(srcRegFilt5, firstFilters);
- srcRegFilt6 = _mm_maddubs_epi16(srcRegFilt6, forthFilters);
- srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
- srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, forthFilters);
-
- // add and saturate the results together
- srcRegFilt5 = _mm_adds_epi16(srcRegFilt5, srcRegFilt6);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt3);
-
- // merge the result together
- srcRegFilt3 = _mm_unpacklo_epi8(srcReg3, srcReg4);
- srcRegFilt6 = _mm_unpackhi_epi8(srcReg3, srcReg4);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
- srcRegFilt6 = _mm_maddubs_epi16(srcRegFilt6, secondFilters);
-
- // merge the result together
- srcRegFilt7 = _mm_unpacklo_epi8(srcReg5, srcReg6);
- srcRegFilt8 = _mm_unpackhi_epi8(srcReg5, srcReg6);
-
- // multiply 2 adjacent elements with the filter and add the result
- srcRegFilt7 = _mm_maddubs_epi16(srcRegFilt7, thirdFilters);
- srcRegFilt8 = _mm_maddubs_epi16(srcRegFilt8, thirdFilters);
-
- // add and saturate the results together
- srcRegFilt5 = _mm_adds_epi16(srcRegFilt5,
- _mm_min_epi16(srcRegFilt3, srcRegFilt7));
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
- _mm_min_epi16(srcRegFilt6, srcRegFilt8));
-
- // add and saturate the results together
- srcRegFilt5 = _mm_adds_epi16(srcRegFilt5,
- _mm_max_epi16(srcRegFilt3, srcRegFilt7));
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
- _mm_max_epi16(srcRegFilt6, srcRegFilt8));
- srcRegFilt5 = _mm_adds_epi16(srcRegFilt5, addFilterReg64);
- srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
-
- // shift by 7 bit each 16 bit
- srcRegFilt5 = _mm_srai_epi16(srcRegFilt5, 7);
- srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
-
- // shrink to 8 bit each 16 bits, the first lane contain the first
- // convolve result and the second lane contain the second convolve
- // result
- srcRegFilt1 = _mm_packus_epi16(srcRegFilt5, srcRegFilt1);
-
- src_ptr+=src_pitch;
-
- // shift down a row
- srcReg1 = srcReg2;
- srcReg2 = srcReg3;
- srcReg3 = srcReg4;
- srcReg4 = srcReg5;
- srcReg5 = srcReg6;
- srcReg6 = srcReg7;
- srcReg7 = srcReg8;
-
- // save 16 bytes convolve result
- _mm_store_si128((__m128i*)output_ptr, srcRegFilt1);
-
- output_ptr+=out_pitch;
- }
-}
-
-#if ARCH_X86_64
-filter8_1dfunction vp9_filter_block1d16_v8_intrin_ssse3;
-filter8_1dfunction vp9_filter_block1d16_h8_intrin_ssse3;
-filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h8_intrin_ssse3;
-filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h8_intrin_ssse3;
-#define vp9_filter_block1d16_v8_ssse3 vp9_filter_block1d16_v8_intrin_ssse3
-#define vp9_filter_block1d16_h8_ssse3 vp9_filter_block1d16_h8_intrin_ssse3
-#define vp9_filter_block1d8_v8_ssse3 vp9_filter_block1d8_v8_intrin_ssse3
-#define vp9_filter_block1d8_h8_ssse3 vp9_filter_block1d8_h8_intrin_ssse3
-#define vp9_filter_block1d4_h8_ssse3 vp9_filter_block1d4_h8_intrin_ssse3
-#else // ARCH_X86
-filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
-filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
-filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
-#endif // ARCH_X86_64
-filter8_1dfunction vp9_filter_block1d16_v8_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d16_h8_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d8_v8_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h8_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d4_v8_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h8_avg_ssse3;
-
-filter8_1dfunction vp9_filter_block1d16_v2_ssse3;
-filter8_1dfunction vp9_filter_block1d16_h2_ssse3;
-filter8_1dfunction vp9_filter_block1d8_v2_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h2_ssse3;
-filter8_1dfunction vp9_filter_block1d4_v2_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h2_ssse3;
-filter8_1dfunction vp9_filter_block1d16_v2_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d16_h2_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d8_v2_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d8_h2_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d4_v2_avg_ssse3;
-filter8_1dfunction vp9_filter_block1d4_h2_avg_ssse3;
-
-// void vp9_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_avg_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , ssse3);
-FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , ssse3);
-FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, ssse3);
-FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
- ssse3);
-
-// void vp9_convolve8_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-// void vp9_convolve8_avg_ssse3(const uint8_t *src, ptrdiff_t src_stride,
-// uint8_t *dst, ptrdiff_t dst_stride,
-// const int16_t *filter_x, int x_step_q4,
-// const int16_t *filter_y, int y_step_q4,
-// int w, int h);
-FUN_CONV_2D(, ssse3);
-FUN_CONV_2D(avg_ , ssse3);
--- a/vp9/common/x86/vp9_subpixel_8t_sse2.asm
+++ /dev/null
@@ -1,987 +1,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-
-;Note: tap3 and tap4 have to be applied and added after other taps to avoid
-;overflow.
-
-%macro GET_FILTERS_4 0
- mov rdx, arg(5) ;filter ptr
- mov rcx, 0x0400040
-
- movdqa xmm7, [rdx] ;load filters
- pshuflw xmm0, xmm7, 0b ;k0
- pshuflw xmm1, xmm7, 01010101b ;k1
- pshuflw xmm2, xmm7, 10101010b ;k2
- pshuflw xmm3, xmm7, 11111111b ;k3
- psrldq xmm7, 8
- pshuflw xmm4, xmm7, 0b ;k4
- pshuflw xmm5, xmm7, 01010101b ;k5
- pshuflw xmm6, xmm7, 10101010b ;k6
- pshuflw xmm7, xmm7, 11111111b ;k7
-
- punpcklqdq xmm0, xmm1
- punpcklqdq xmm2, xmm3
- punpcklqdq xmm5, xmm4
- punpcklqdq xmm6, xmm7
-
- movdqa k0k1, xmm0
- movdqa k2k3, xmm2
- movdqa k5k4, xmm5
- movdqa k6k7, xmm6
-
- movq xmm6, rcx
- pshufd xmm6, xmm6, 0
- movdqa krd, xmm6
-
- pxor xmm7, xmm7
- movdqa zero, xmm7
-%endm
-
-%macro APPLY_FILTER_4 1
- punpckldq xmm0, xmm1 ;two row in one register
- punpckldq xmm6, xmm7
- punpckldq xmm2, xmm3
- punpckldq xmm5, xmm4
-
- punpcklbw xmm0, zero ;unpack to word
- punpcklbw xmm6, zero
- punpcklbw xmm2, zero
- punpcklbw xmm5, zero
-
- pmullw xmm0, k0k1 ;multiply the filter factors
- pmullw xmm6, k6k7
- pmullw xmm2, k2k3
- pmullw xmm5, k5k4
-
- paddsw xmm0, xmm6 ;sum
- movdqa xmm1, xmm0
- psrldq xmm1, 8
- paddsw xmm0, xmm1
- paddsw xmm0, xmm2
- psrldq xmm2, 8
- paddsw xmm0, xmm5
- psrldq xmm5, 8
- paddsw xmm0, xmm2
- paddsw xmm0, xmm5
-
- paddsw xmm0, krd ;rounding
- psraw xmm0, 7 ;shift
- packuswb xmm0, xmm0 ;pack to byte
-
-%if %1
- movd xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movd [rdi], xmm0
-%endm
-
-%macro GET_FILTERS 0
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm7, [rdx] ;load filters
- pshuflw xmm0, xmm7, 0b ;k0
- pshuflw xmm1, xmm7, 01010101b ;k1
- pshuflw xmm2, xmm7, 10101010b ;k2
- pshuflw xmm3, xmm7, 11111111b ;k3
- pshufhw xmm4, xmm7, 0b ;k4
- pshufhw xmm5, xmm7, 01010101b ;k5
- pshufhw xmm6, xmm7, 10101010b ;k6
- pshufhw xmm7, xmm7, 11111111b ;k7
-
- punpcklwd xmm0, xmm0
- punpcklwd xmm1, xmm1
- punpcklwd xmm2, xmm2
- punpcklwd xmm3, xmm3
- punpckhwd xmm4, xmm4
- punpckhwd xmm5, xmm5
- punpckhwd xmm6, xmm6
- punpckhwd xmm7, xmm7
-
- movdqa k0, xmm0 ;store filter factors on stack
- movdqa k1, xmm1
- movdqa k2, xmm2
- movdqa k3, xmm3
- movdqa k4, xmm4
- movdqa k5, xmm5
- movdqa k6, xmm6
- movdqa k7, xmm7
-
- movq xmm6, rcx
- pshufd xmm6, xmm6, 0
- movdqa krd, xmm6 ;rounding
-
- pxor xmm7, xmm7
- movdqa zero, xmm7
-%endm
-
-%macro LOAD_VERT_8 1
- movq xmm0, [rsi + %1] ;0
- movq xmm1, [rsi + rax + %1] ;1
- movq xmm6, [rsi + rdx * 2 + %1] ;6
- lea rsi, [rsi + rax]
- movq xmm7, [rsi + rdx * 2 + %1] ;7
- movq xmm2, [rsi + rax + %1] ;2
- movq xmm3, [rsi + rax * 2 + %1] ;3
- movq xmm4, [rsi + rdx + %1] ;4
- movq xmm5, [rsi + rax * 4 + %1] ;5
-%endm
-
-%macro APPLY_FILTER_8 2
- punpcklbw xmm0, zero
- punpcklbw xmm1, zero
- punpcklbw xmm6, zero
- punpcklbw xmm7, zero
- punpcklbw xmm2, zero
- punpcklbw xmm5, zero
- punpcklbw xmm3, zero
- punpcklbw xmm4, zero
-
- pmullw xmm0, k0
- pmullw xmm1, k1
- pmullw xmm6, k6
- pmullw xmm7, k7
- pmullw xmm2, k2
- pmullw xmm5, k5
- pmullw xmm3, k3
- pmullw xmm4, k4
-
- paddsw xmm0, xmm1
- paddsw xmm0, xmm6
- paddsw xmm0, xmm7
- paddsw xmm0, xmm2
- paddsw xmm0, xmm5
- paddsw xmm0, xmm3
- paddsw xmm0, xmm4
-
- paddsw xmm0, krd ;rounding
- psraw xmm0, 7 ;shift
- packuswb xmm0, xmm0 ;pack back to byte
-%if %1
- movq xmm1, [rdi + %2]
- pavgb xmm0, xmm1
-%endif
- movq [rdi + %2], xmm0
-%endm
-
-;void vp9_filter_block1d4_v8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d4_v8_sse2) PRIVATE
-sym(vp9_filter_block1d4_v8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 6
- %define k0k1 [rsp + 16 * 0]
- %define k2k3 [rsp + 16 * 1]
- %define k5k4 [rsp + 16 * 2]
- %define k6k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define zero [rsp + 16 * 5]
-
- GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movd xmm0, [rsi] ;load src: row 0
- movd xmm1, [rsi + rax] ;1
- movd xmm6, [rsi + rdx * 2] ;6
- lea rsi, [rsi + rax]
- movd xmm7, [rsi + rdx * 2] ;7
- movd xmm2, [rsi + rax] ;2
- movd xmm3, [rsi + rax * 2] ;3
- movd xmm4, [rsi + rdx] ;4
- movd xmm5, [rsi + rax * 4] ;5
-
- APPLY_FILTER_4 0
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 6
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d8_v8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d8_v8_sse2) PRIVATE
-sym(vp9_filter_block1d8_v8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- LOAD_VERT_8 0
- APPLY_FILTER_8 0, 0
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d16_v8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d16_v8_sse2) PRIVATE
-sym(vp9_filter_block1d16_v8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- LOAD_VERT_8 0
- APPLY_FILTER_8 0, 0
- sub rsi, rax
-
- LOAD_VERT_8 8
- APPLY_FILTER_8 0, 8
- add rdi, rbx
-
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_v8_avg_sse2) PRIVATE
-sym(vp9_filter_block1d4_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 6
- %define k0k1 [rsp + 16 * 0]
- %define k2k3 [rsp + 16 * 1]
- %define k5k4 [rsp + 16 * 2]
- %define k6k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define zero [rsp + 16 * 5]
-
- GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movd xmm0, [rsi] ;load src: row 0
- movd xmm1, [rsi + rax] ;1
- movd xmm6, [rsi + rdx * 2] ;6
- lea rsi, [rsi + rax]
- movd xmm7, [rsi + rdx * 2] ;7
- movd xmm2, [rsi + rax] ;2
- movd xmm3, [rsi + rax * 2] ;3
- movd xmm4, [rsi + rdx] ;4
- movd xmm5, [rsi + rax * 4] ;5
-
- APPLY_FILTER_4 1
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 6
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_v8_avg_sse2) PRIVATE
-sym(vp9_filter_block1d8_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-.loop:
- LOAD_VERT_8 0
- APPLY_FILTER_8 1, 0
-
- lea rdi, [rdi + rbx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_v8_avg_sse2) PRIVATE
-sym(vp9_filter_block1d16_v8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rbx, DWORD PTR arg(3) ;out_pitch
- lea rdx, [rax + rax * 2]
- movsxd rcx, DWORD PTR arg(4) ;output_height
-.loop:
- LOAD_VERT_8 0
- APPLY_FILTER_8 1, 0
- sub rsi, rax
-
- LOAD_VERT_8 8
- APPLY_FILTER_8 1, 8
- add rdi, rbx
-
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d4_h8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d4_h8_sse2) PRIVATE
-sym(vp9_filter_block1d4_h8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 6
- %define k0k1 [rsp + 16 * 0]
- %define k2k3 [rsp + 16 * 1]
- %define k5k4 [rsp + 16 * 2]
- %define k6k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define zero [rsp + 16 * 5]
-
- GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm3, xmm0
- movdqa xmm5, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm3, 3
- psrldq xmm5, 5
- psrldq xmm4, 4
-
- APPLY_FILTER_4 0
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 6
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d8_h8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d8_h8_sse2) PRIVATE
-sym(vp9_filter_block1d8_h8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 0, 0
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d16_h8_sse2
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d16_h8_sse2) PRIVATE
-sym(vp9_filter_block1d16_h8_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 0, 0
-
- movdqu xmm0, [rsi + 5] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 0, 8
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_h8_avg_sse2) PRIVATE
-sym(vp9_filter_block1d4_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 6
- %define k0k1 [rsp + 16 * 0]
- %define k2k3 [rsp + 16 * 1]
- %define k5k4 [rsp + 16 * 2]
- %define k6k7 [rsp + 16 * 3]
- %define krd [rsp + 16 * 4]
- %define zero [rsp + 16 * 5]
-
- GET_FILTERS_4
-
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm3, xmm0
- movdqa xmm5, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm3, 3
- psrldq xmm5, 5
- psrldq xmm4, 4
-
- APPLY_FILTER_4 1
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 6
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_h8_avg_sse2) PRIVATE
-sym(vp9_filter_block1d8_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 1, 0
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_h8_avg_sse2) PRIVATE
-sym(vp9_filter_block1d16_h8_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 10
- %define k0 [rsp + 16 * 0]
- %define k1 [rsp + 16 * 1]
- %define k2 [rsp + 16 * 2]
- %define k3 [rsp + 16 * 3]
- %define k4 [rsp + 16 * 4]
- %define k5 [rsp + 16 * 5]
- %define k6 [rsp + 16 * 6]
- %define k7 [rsp + 16 * 7]
- %define krd [rsp + 16 * 8]
- %define zero [rsp + 16 * 9]
-
- GET_FILTERS
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-
-.loop:
- movdqu xmm0, [rsi - 3] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 1, 0
-
- movdqu xmm0, [rsi + 5] ;load src
-
- movdqa xmm1, xmm0
- movdqa xmm6, xmm0
- movdqa xmm7, xmm0
- movdqa xmm2, xmm0
- movdqa xmm5, xmm0
- movdqa xmm3, xmm0
- movdqa xmm4, xmm0
-
- psrldq xmm1, 1
- psrldq xmm6, 6
- psrldq xmm7, 7
- psrldq xmm2, 2
- psrldq xmm5, 5
- psrldq xmm3, 3
- psrldq xmm4, 4
-
- APPLY_FILTER_8 1, 8
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-
- add rsp, 16 * 10
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
--- a/vp9/common/x86/vp9_subpixel_8t_ssse3.asm
+++ /dev/null
@@ -1,1071 +1,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-
-%macro VERTx4 1
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm4, [rdx] ;load filters
- movq xmm5, rcx
- packsswb xmm4, xmm4
- pshuflw xmm0, xmm4, 0b ;k0_k1
- pshuflw xmm1, xmm4, 01010101b ;k2_k3
- pshuflw xmm2, xmm4, 10101010b ;k4_k5
- pshuflw xmm3, xmm4, 11111111b ;k6_k7
-
- punpcklqdq xmm0, xmm0
- punpcklqdq xmm1, xmm1
- punpcklqdq xmm2, xmm2
- punpcklqdq xmm3, xmm3
-
- movdqa k0k1, xmm0
- movdqa k2k3, xmm1
- pshufd xmm5, xmm5, 0
- movdqa k4k5, xmm2
- movdqa k6k7, xmm3
- movdqa krd, xmm5
-
- movsxd rdx, DWORD PTR arg(1) ;pixels_per_line
-
-%if ABI_IS_32BIT=0
- movsxd r8, DWORD PTR arg(3) ;out_pitch
-%endif
- mov rax, rsi
- movsxd rcx, DWORD PTR arg(4) ;output_height
- add rax, rdx
-
- lea rbx, [rdx + rdx*4]
- add rbx, rdx ;pitch * 6
-
-.loop:
- movd xmm0, [rsi] ;A
- movd xmm1, [rsi + rdx] ;B
- movd xmm2, [rsi + rdx * 2] ;C
- movd xmm3, [rax + rdx * 2] ;D
- movd xmm4, [rsi + rdx * 4] ;E
- movd xmm5, [rax + rdx * 4] ;F
-
- punpcklbw xmm0, xmm1 ;A B
- punpcklbw xmm2, xmm3 ;C D
- punpcklbw xmm4, xmm5 ;E F
-
- movd xmm6, [rsi + rbx] ;G
- movd xmm7, [rax + rbx] ;H
-
- pmaddubsw xmm0, k0k1
- pmaddubsw xmm2, k2k3
- punpcklbw xmm6, xmm7 ;G H
- pmaddubsw xmm4, k4k5
- pmaddubsw xmm6, k6k7
-
- movdqa xmm1, xmm2
- paddsw xmm0, xmm6
- pmaxsw xmm2, xmm4
- pminsw xmm4, xmm1
- paddsw xmm0, xmm4
- paddsw xmm0, xmm2
-
- paddsw xmm0, krd
- psraw xmm0, 7
- packuswb xmm0, xmm0
-
- add rsi, rdx
- add rax, rdx
-%if %1
- movd xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movd [rdi], xmm0
-
-%if ABI_IS_32BIT
- add rdi, DWORD PTR arg(3) ;out_pitch
-%else
- add rdi, r8
-%endif
- dec rcx
- jnz .loop
-%endm
-
-%macro VERTx8 1
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm4, [rdx] ;load filters
- movq xmm5, rcx
- packsswb xmm4, xmm4
- pshuflw xmm0, xmm4, 0b ;k0_k1
- pshuflw xmm1, xmm4, 01010101b ;k2_k3
- pshuflw xmm2, xmm4, 10101010b ;k4_k5
- pshuflw xmm3, xmm4, 11111111b ;k6_k7
-
- punpcklqdq xmm0, xmm0
- punpcklqdq xmm1, xmm1
- punpcklqdq xmm2, xmm2
- punpcklqdq xmm3, xmm3
-
- movdqa k0k1, xmm0
- movdqa k2k3, xmm1
- pshufd xmm5, xmm5, 0
- movdqa k4k5, xmm2
- movdqa k6k7, xmm3
- movdqa krd, xmm5
-
- movsxd rdx, DWORD PTR arg(1) ;pixels_per_line
-
-%if ABI_IS_32BIT=0
- movsxd r8, DWORD PTR arg(3) ;out_pitch
-%endif
- mov rax, rsi
- movsxd rcx, DWORD PTR arg(4) ;output_height
- add rax, rdx
-
- lea rbx, [rdx + rdx*4]
- add rbx, rdx ;pitch * 6
-
-.loop:
- movq xmm0, [rsi] ;A
- movq xmm1, [rsi + rdx] ;B
- movq xmm2, [rsi + rdx * 2] ;C
- movq xmm3, [rax + rdx * 2] ;D
- movq xmm4, [rsi + rdx * 4] ;E
- movq xmm5, [rax + rdx * 4] ;F
-
- punpcklbw xmm0, xmm1 ;A B
- punpcklbw xmm2, xmm3 ;C D
- punpcklbw xmm4, xmm5 ;E F
-
- movq xmm6, [rsi + rbx] ;G
- movq xmm7, [rax + rbx] ;H
-
- pmaddubsw xmm0, k0k1
- pmaddubsw xmm2, k2k3
- punpcklbw xmm6, xmm7 ;G H
- pmaddubsw xmm4, k4k5
- pmaddubsw xmm6, k6k7
-
- paddsw xmm0, xmm6
- movdqa xmm1, xmm2
- pmaxsw xmm2, xmm4
- pminsw xmm4, xmm1
- paddsw xmm0, xmm4
- paddsw xmm0, xmm2
-
- paddsw xmm0, krd
- psraw xmm0, 7
- packuswb xmm0, xmm0
-
- add rsi, rdx
- add rax, rdx
-%if %1
- movq xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movq [rdi], xmm0
-
-%if ABI_IS_32BIT
- add rdi, DWORD PTR arg(3) ;out_pitch
-%else
- add rdi, r8
-%endif
- dec rcx
- jnz .loop
-%endm
-
-
-%macro VERTx16 1
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm4, [rdx] ;load filters
- movq xmm5, rcx
- packsswb xmm4, xmm4
- pshuflw xmm0, xmm4, 0b ;k0_k1
- pshuflw xmm1, xmm4, 01010101b ;k2_k3
- pshuflw xmm2, xmm4, 10101010b ;k4_k5
- pshuflw xmm3, xmm4, 11111111b ;k6_k7
-
- punpcklqdq xmm0, xmm0
- punpcklqdq xmm1, xmm1
- punpcklqdq xmm2, xmm2
- punpcklqdq xmm3, xmm3
-
- movdqa k0k1, xmm0
- movdqa k2k3, xmm1
- pshufd xmm5, xmm5, 0
- movdqa k4k5, xmm2
- movdqa k6k7, xmm3
- movdqa krd, xmm5
-
- movsxd rdx, DWORD PTR arg(1) ;pixels_per_line
-
-%if ABI_IS_32BIT=0
- movsxd r8, DWORD PTR arg(3) ;out_pitch
-%endif
- mov rax, rsi
- movsxd rcx, DWORD PTR arg(4) ;output_height
- add rax, rdx
-
- lea rbx, [rdx + rdx*4]
- add rbx, rdx ;pitch * 6
-
-.loop:
- movq xmm0, [rsi] ;A
- movq xmm1, [rsi + rdx] ;B
- movq xmm2, [rsi + rdx * 2] ;C
- movq xmm3, [rax + rdx * 2] ;D
- movq xmm4, [rsi + rdx * 4] ;E
- movq xmm5, [rax + rdx * 4] ;F
-
- punpcklbw xmm0, xmm1 ;A B
- punpcklbw xmm2, xmm3 ;C D
- punpcklbw xmm4, xmm5 ;E F
-
- movq xmm6, [rsi + rbx] ;G
- movq xmm7, [rax + rbx] ;H
-
- pmaddubsw xmm0, k0k1
- pmaddubsw xmm2, k2k3
- punpcklbw xmm6, xmm7 ;G H
- pmaddubsw xmm4, k4k5
- pmaddubsw xmm6, k6k7
-
- paddsw xmm0, xmm6
- movdqa xmm1, xmm2
- pmaxsw xmm2, xmm4
- pminsw xmm4, xmm1
- paddsw xmm0, xmm4
- paddsw xmm0, xmm2
-
- paddsw xmm0, krd
- psraw xmm0, 7
- packuswb xmm0, xmm0
-%if %1
- movq xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movq [rdi], xmm0
-
- movq xmm0, [rsi + 8] ;A
- movq xmm1, [rsi + rdx + 8] ;B
- movq xmm2, [rsi + rdx * 2 + 8] ;C
- movq xmm3, [rax + rdx * 2 + 8] ;D
- movq xmm4, [rsi + rdx * 4 + 8] ;E
- movq xmm5, [rax + rdx * 4 + 8] ;F
-
- punpcklbw xmm0, xmm1 ;A B
- punpcklbw xmm2, xmm3 ;C D
- punpcklbw xmm4, xmm5 ;E F
-
- movq xmm6, [rsi + rbx + 8] ;G
- movq xmm7, [rax + rbx + 8] ;H
- punpcklbw xmm6, xmm7 ;G H
-
- pmaddubsw xmm0, k0k1
- pmaddubsw xmm2, k2k3
- pmaddubsw xmm4, k4k5
- pmaddubsw xmm6, k6k7
-
- paddsw xmm0, xmm6
- movdqa xmm1, xmm2
- pmaxsw xmm2, xmm4
- pminsw xmm4, xmm1
- paddsw xmm0, xmm4
- paddsw xmm0, xmm2
-
- paddsw xmm0, krd
- psraw xmm0, 7
- packuswb xmm0, xmm0
-
- add rsi, rdx
- add rax, rdx
-%if %1
- movq xmm1, [rdi+8]
- pavgb xmm0, xmm1
-%endif
-
- movq [rdi+8], xmm0
-
-%if ABI_IS_32BIT
- add rdi, DWORD PTR arg(3) ;out_pitch
-%else
- add rdi, r8
-%endif
- dec rcx
- jnz .loop
-%endm
-
-;void vp9_filter_block1d8_v8_ssse3
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d4_v8_ssse3) PRIVATE
-sym(vp9_filter_block1d4_v8_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- VERTx4 0
-
- add rsp, 16*5
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d8_v8_ssse3
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d8_v8_ssse3) PRIVATE
-sym(vp9_filter_block1d8_v8_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- VERTx8 0
-
- add rsp, 16*5
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d16_v8_ssse3
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pitch,
-; unsigned char *output_ptr,
-; unsigned int out_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d16_v8_ssse3) PRIVATE
-sym(vp9_filter_block1d16_v8_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- VERTx16 0
-
- add rsp, 16*5
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-
-global sym(vp9_filter_block1d4_v8_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d4_v8_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- VERTx4 1
-
- add rsp, 16*5
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_v8_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d8_v8_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- VERTx8 1
-
- add rsp, 16*5
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_v8_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d16_v8_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- push rbx
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- VERTx16 1
-
- add rsp, 16*5
- pop rsp
- pop rbx
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-%macro HORIZx4_ROW 2
- movdqa %2, %1
- pshufb %1, [GLOBAL(shuf_t0t1)]
- pshufb %2, [GLOBAL(shuf_t2t3)]
- pmaddubsw %1, k0k1k4k5
- pmaddubsw %2, k2k3k6k7
-
- movdqa xmm4, %1
- movdqa xmm5, %2
- psrldq %1, 8
- psrldq %2, 8
- movdqa xmm6, xmm5
-
- paddsw xmm4, %2
- pmaxsw xmm5, %1
- pminsw %1, xmm6
- paddsw %1, xmm4
- paddsw %1, xmm5
-
- paddsw %1, krd
- psraw %1, 7
- packuswb %1, %1
-%endm
-
-%macro HORIZx4 1
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm4, [rdx] ;load filters
- movq xmm5, rcx
- packsswb xmm4, xmm4
- pshuflw xmm6, xmm4, 0b ;k0_k1
- pshufhw xmm6, xmm6, 10101010b ;k0_k1_k4_k5
- pshuflw xmm7, xmm4, 01010101b ;k2_k3
- pshufhw xmm7, xmm7, 11111111b ;k2_k3_k6_k7
- pshufd xmm5, xmm5, 0 ;rounding
-
- movdqa k0k1k4k5, xmm6
- movdqa k2k3k6k7, xmm7
- movdqa krd, xmm5
-
- movsxd rax, dword ptr arg(1) ;src_pixels_per_line
- movsxd rdx, dword ptr arg(3) ;output_pitch
- movsxd rcx, dword ptr arg(4) ;output_height
- shr rcx, 1
-.loop:
- ;Do two rows once
- movq xmm0, [rsi - 3] ;load src
- movq xmm1, [rsi + 5]
- movq xmm2, [rsi + rax - 3]
- movq xmm3, [rsi + rax + 5]
- punpcklqdq xmm0, xmm1
- punpcklqdq xmm2, xmm3
-
- HORIZx4_ROW xmm0, xmm1
- HORIZx4_ROW xmm2, xmm3
-%if %1
- movd xmm1, [rdi]
- pavgb xmm0, xmm1
- movd xmm3, [rdi + rdx]
- pavgb xmm2, xmm3
-%endif
- movd [rdi], xmm0
- movd [rdi +rdx], xmm2
-
- lea rsi, [rsi + rax]
- prefetcht0 [rsi + 4 * rax - 3]
- lea rsi, [rsi + rax]
- lea rdi, [rdi + 2 * rdx]
- prefetcht0 [rsi + 2 * rax - 3]
-
- dec rcx
- jnz .loop
-
- ; Do last row if output_height is odd
- movsxd rcx, dword ptr arg(4) ;output_height
- and rcx, 1
- je .done
-
- movq xmm0, [rsi - 3] ; load src
- movq xmm1, [rsi + 5]
- punpcklqdq xmm0, xmm1
-
- HORIZx4_ROW xmm0, xmm1
-%if %1
- movd xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movd [rdi], xmm0
-.done
-%endm
-
-%macro HORIZx8_ROW 4
- movdqa %2, %1
- movdqa %3, %1
- movdqa %4, %1
-
- pshufb %1, [GLOBAL(shuf_t0t1)]
- pshufb %2, [GLOBAL(shuf_t2t3)]
- pshufb %3, [GLOBAL(shuf_t4t5)]
- pshufb %4, [GLOBAL(shuf_t6t7)]
-
- pmaddubsw %1, k0k1
- pmaddubsw %2, k2k3
- pmaddubsw %3, k4k5
- pmaddubsw %4, k6k7
-
- paddsw %1, %4
- movdqa %4, %2
- pmaxsw %2, %3
- pminsw %3, %4
- paddsw %1, %3
- paddsw %1, %2
-
- paddsw %1, krd
- psraw %1, 7
- packuswb %1, %1
-%endm
-
-%macro HORIZx8 1
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm4, [rdx] ;load filters
- movq xmm5, rcx
- packsswb xmm4, xmm4
- pshuflw xmm0, xmm4, 0b ;k0_k1
- pshuflw xmm1, xmm4, 01010101b ;k2_k3
- pshuflw xmm2, xmm4, 10101010b ;k4_k5
- pshuflw xmm3, xmm4, 11111111b ;k6_k7
-
- punpcklqdq xmm0, xmm0
- punpcklqdq xmm1, xmm1
- punpcklqdq xmm2, xmm2
- punpcklqdq xmm3, xmm3
-
- movdqa k0k1, xmm0
- movdqa k2k3, xmm1
- pshufd xmm5, xmm5, 0
- movdqa k4k5, xmm2
- movdqa k6k7, xmm3
- movdqa krd, xmm5
-
- movsxd rax, dword ptr arg(1) ;src_pixels_per_line
- movsxd rdx, dword ptr arg(3) ;output_pitch
- movsxd rcx, dword ptr arg(4) ;output_height
- shr rcx, 1
-
-.loop:
- movq xmm0, [rsi - 3] ;load src
- movq xmm3, [rsi + 5]
- movq xmm4, [rsi + rax - 3]
- movq xmm7, [rsi + rax + 5]
- punpcklqdq xmm0, xmm3
- punpcklqdq xmm4, xmm7
-
- HORIZx8_ROW xmm0, xmm1, xmm2, xmm3
- HORIZx8_ROW xmm4, xmm5, xmm6, xmm7
-%if %1
- movq xmm1, [rdi]
- movq xmm2, [rdi + rdx]
- pavgb xmm0, xmm1
- pavgb xmm4, xmm2
-%endif
- movq [rdi], xmm0
- movq [rdi + rdx], xmm4
-
- lea rsi, [rsi + rax]
- prefetcht0 [rsi + 4 * rax - 3]
- lea rsi, [rsi + rax]
- lea rdi, [rdi + 2 * rdx]
- prefetcht0 [rsi + 2 * rax - 3]
- dec rcx
- jnz .loop
-
- ;Do last row if output_height is odd
- movsxd rcx, dword ptr arg(4) ;output_height
- and rcx, 1
- je .done
-
- movq xmm0, [rsi - 3]
- movq xmm3, [rsi + 5]
- punpcklqdq xmm0, xmm3
-
- HORIZx8_ROW xmm0, xmm1, xmm2, xmm3
-%if %1
- movq xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movq [rdi], xmm0
-.done
-%endm
-
-%macro HORIZx16 1
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm4, [rdx] ;load filters
- movq xmm5, rcx
- packsswb xmm4, xmm4
- pshuflw xmm0, xmm4, 0b ;k0_k1
- pshuflw xmm1, xmm4, 01010101b ;k2_k3
- pshuflw xmm2, xmm4, 10101010b ;k4_k5
- pshuflw xmm3, xmm4, 11111111b ;k6_k7
-
- punpcklqdq xmm0, xmm0
- punpcklqdq xmm1, xmm1
- punpcklqdq xmm2, xmm2
- punpcklqdq xmm3, xmm3
-
- movdqa k0k1, xmm0
- movdqa k2k3, xmm1
- pshufd xmm5, xmm5, 0
- movdqa k4k5, xmm2
- movdqa k6k7, xmm3
- movdqa krd, xmm5
-
- movsxd rax, dword ptr arg(1) ;src_pixels_per_line
- movsxd rdx, dword ptr arg(3) ;output_pitch
- movsxd rcx, dword ptr arg(4) ;output_height
-
-.loop:
- prefetcht0 [rsi + 2 * rax -3]
-
- movq xmm0, [rsi - 3] ;load src data
- movq xmm4, [rsi + 5]
- movq xmm6, [rsi + 13]
- punpcklqdq xmm0, xmm4
- punpcklqdq xmm4, xmm6
-
- movdqa xmm7, xmm0
-
- punpcklbw xmm7, xmm7
- punpckhbw xmm0, xmm0
- movdqa xmm1, xmm0
- movdqa xmm2, xmm0
- movdqa xmm3, xmm0
-
- palignr xmm0, xmm7, 1
- palignr xmm1, xmm7, 5
- pmaddubsw xmm0, k0k1
- palignr xmm2, xmm7, 9
- pmaddubsw xmm1, k2k3
- palignr xmm3, xmm7, 13
-
- pmaddubsw xmm2, k4k5
- pmaddubsw xmm3, k6k7
- paddsw xmm0, xmm3
-
- movdqa xmm3, xmm4
- punpcklbw xmm3, xmm3
- punpckhbw xmm4, xmm4
-
- movdqa xmm5, xmm4
- movdqa xmm6, xmm4
- movdqa xmm7, xmm4
-
- palignr xmm4, xmm3, 1
- palignr xmm5, xmm3, 5
- palignr xmm6, xmm3, 9
- palignr xmm7, xmm3, 13
-
- movdqa xmm3, xmm1
- pmaddubsw xmm4, k0k1
- pmaxsw xmm1, xmm2
- pmaddubsw xmm5, k2k3
- pminsw xmm2, xmm3
- pmaddubsw xmm6, k4k5
- paddsw xmm0, xmm2
- pmaddubsw xmm7, k6k7
- paddsw xmm0, xmm1
-
- paddsw xmm4, xmm7
- movdqa xmm7, xmm5
- pmaxsw xmm5, xmm6
- pminsw xmm6, xmm7
- paddsw xmm4, xmm6
- paddsw xmm4, xmm5
-
- paddsw xmm0, krd
- paddsw xmm4, krd
- psraw xmm0, 7
- psraw xmm4, 7
- packuswb xmm0, xmm0
- packuswb xmm4, xmm4
- punpcklqdq xmm0, xmm4
-%if %1
- movdqa xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
-
- lea rsi, [rsi + rax]
- movdqa [rdi], xmm0
-
- lea rdi, [rdi + rdx]
- dec rcx
- jnz .loop
-%endm
-
-;void vp9_filter_block1d4_h8_ssse3
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d4_h8_ssse3) PRIVATE
-sym(vp9_filter_block1d4_h8_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 3
- %define k0k1k4k5 [rsp + 16 * 0]
- %define k2k3k6k7 [rsp + 16 * 1]
- %define krd [rsp + 16 * 2]
-
- HORIZx4 0
-
- add rsp, 16 * 3
- pop rsp
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d8_h8_ssse3
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d8_h8_ssse3) PRIVATE
-sym(vp9_filter_block1d8_h8_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- HORIZx8 0
-
- add rsp, 16*5
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-;void vp9_filter_block1d16_h8_ssse3
-;(
-; unsigned char *src_ptr,
-; unsigned int src_pixels_per_line,
-; unsigned char *output_ptr,
-; unsigned int output_pitch,
-; unsigned int output_height,
-; short *filter
-;)
-global sym(vp9_filter_block1d16_h8_ssse3) PRIVATE
-sym(vp9_filter_block1d16_h8_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- HORIZx16 0
-
- add rsp, 16*5
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_h8_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d4_h8_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16 * 3
- %define k0k1k4k5 [rsp + 16 * 0]
- %define k2k3k6k7 [rsp + 16 * 1]
- %define krd [rsp + 16 * 2]
-
- HORIZx4 1
-
- add rsp, 16 * 3
- pop rsp
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_h8_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d8_h8_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- HORIZx8 1
-
- add rsp, 16*5
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_h8_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d16_h8_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 16*5
- %define k0k1 [rsp + 16*0]
- %define k2k3 [rsp + 16*1]
- %define k4k5 [rsp + 16*2]
- %define k6k7 [rsp + 16*3]
- %define krd [rsp + 16*4]
-
- HORIZx16 1
-
- add rsp, 16*5
- pop rsp
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-SECTION_RODATA
-align 16
-shuf_t0t1:
- db 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
-align 16
-shuf_t2t3:
- db 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
-align 16
-shuf_t4t5:
- db 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
-align 16
-shuf_t6t7:
- db 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
--- a/vp9/common/x86/vp9_subpixel_bilinear_sse2.asm
+++ /dev/null
@@ -1,448 +1,0 @@
-;
-; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-%include "vpx_ports/x86_abi_support.asm"
-
-%macro GET_PARAM_4 0
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm3, [rdx] ;load filters
- pshuflw xmm4, xmm3, 11111111b ;k3
- psrldq xmm3, 8
- pshuflw xmm3, xmm3, 0b ;k4
- punpcklqdq xmm4, xmm3 ;k3k4
-
- movq xmm3, rcx ;rounding
- pshufd xmm3, xmm3, 0
-
- pxor xmm2, xmm2
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-%endm
-
-%macro APPLY_FILTER_4 1
-
- punpckldq xmm0, xmm1 ;two row in one register
- punpcklbw xmm0, xmm2 ;unpack to word
- pmullw xmm0, xmm4 ;multiply the filter factors
-
- movdqa xmm1, xmm0
- psrldq xmm1, 8
- paddsw xmm0, xmm1
-
- paddsw xmm0, xmm3 ;rounding
- psraw xmm0, 7 ;shift
- packuswb xmm0, xmm0 ;pack to byte
-
-%if %1
- movd xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
-
- movd [rdi], xmm0
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
-%endm
-
-%macro GET_PARAM 0
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm7, [rdx] ;load filters
-
- pshuflw xmm6, xmm7, 11111111b ;k3
- pshufhw xmm7, xmm7, 0b ;k4
- punpcklwd xmm6, xmm6
- punpckhwd xmm7, xmm7
-
- movq xmm4, rcx ;rounding
- pshufd xmm4, xmm4, 0
-
- pxor xmm5, xmm5
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-%endm
-
-%macro APPLY_FILTER_8 1
- punpcklbw xmm0, xmm5
- punpcklbw xmm1, xmm5
-
- pmullw xmm0, xmm6
- pmullw xmm1, xmm7
- paddsw xmm0, xmm1
- paddsw xmm0, xmm4 ;rounding
- psraw xmm0, 7 ;shift
- packuswb xmm0, xmm0 ;pack back to byte
-%if %1
- movq xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movq [rdi], xmm0 ;store the result
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
-%endm
-
-%macro APPLY_FILTER_16 1
- punpcklbw xmm0, xmm5
- punpcklbw xmm1, xmm5
- punpckhbw xmm2, xmm5
- punpckhbw xmm3, xmm5
-
- pmullw xmm0, xmm6
- pmullw xmm1, xmm7
- pmullw xmm2, xmm6
- pmullw xmm3, xmm7
-
- paddsw xmm0, xmm1
- paddsw xmm2, xmm3
-
- paddsw xmm0, xmm4 ;rounding
- paddsw xmm2, xmm4
- psraw xmm0, 7 ;shift
- psraw xmm2, 7
- packuswb xmm0, xmm2 ;pack back to byte
-%if %1
- movdqu xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movdqu [rdi], xmm0 ;store the result
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
-%endm
-
-global sym(vp9_filter_block1d4_v2_sse2) PRIVATE
-sym(vp9_filter_block1d4_v2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movd xmm0, [rsi] ;load src
- movd xmm1, [rsi + rax]
-
- APPLY_FILTER_4 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_v2_sse2) PRIVATE
-sym(vp9_filter_block1d8_v2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movq xmm0, [rsi] ;0
- movq xmm1, [rsi + rax] ;1
-
- APPLY_FILTER_8 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_v2_sse2) PRIVATE
-sym(vp9_filter_block1d16_v2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + rax] ;1
- movdqa xmm2, xmm0
- movdqa xmm3, xmm1
-
- APPLY_FILTER_16 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_v2_avg_sse2) PRIVATE
-sym(vp9_filter_block1d4_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movd xmm0, [rsi] ;load src
- movd xmm1, [rsi + rax]
-
- APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_v2_avg_sse2) PRIVATE
-sym(vp9_filter_block1d8_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movq xmm0, [rsi] ;0
- movq xmm1, [rsi + rax] ;1
-
- APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_v2_avg_sse2) PRIVATE
-sym(vp9_filter_block1d16_v2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + rax] ;1
- movdqa xmm2, xmm0
- movdqa xmm3, xmm1
-
- APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_h2_sse2) PRIVATE
-sym(vp9_filter_block1d4_h2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_4 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_h2_sse2) PRIVATE
-sym(vp9_filter_block1d8_h2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_8 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_h2_sse2) PRIVATE
-sym(vp9_filter_block1d16_h2_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 1]
- movdqa xmm2, xmm0
- movdqa xmm3, xmm1
-
- APPLY_FILTER_16 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_h2_avg_sse2) PRIVATE
-sym(vp9_filter_block1d4_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_h2_avg_sse2) PRIVATE
-sym(vp9_filter_block1d8_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_h2_avg_sse2) PRIVATE
-sym(vp9_filter_block1d16_h2_avg_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 1]
- movdqa xmm2, xmm0
- movdqa xmm3, xmm1
-
- APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
--- a/vp9/common/x86/vp9_subpixel_bilinear_ssse3.asm
+++ /dev/null
@@ -1,422 +1,0 @@
-;
-; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license
-; that can be found in the LICENSE file in the root of the source
-; tree. An additional intellectual property rights grant can be found
-; in the file PATENTS. All contributing project authors may
-; be found in the AUTHORS file in the root of the source tree.
-;
-
-%include "vpx_ports/x86_abi_support.asm"
-
-%macro GET_PARAM_4 0
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm3, [rdx] ;load filters
- psrldq xmm3, 6
- packsswb xmm3, xmm3
- pshuflw xmm3, xmm3, 0b ;k3_k4
-
- movq xmm2, rcx ;rounding
- pshufd xmm2, xmm2, 0
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-%endm
-
-%macro APPLY_FILTER_4 1
- punpcklbw xmm0, xmm1
- pmaddubsw xmm0, xmm3
-
- paddsw xmm0, xmm2 ;rounding
- psraw xmm0, 7 ;shift
- packuswb xmm0, xmm0 ;pack to byte
-
-%if %1
- movd xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movd [rdi], xmm0
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
-%endm
-
-%macro GET_PARAM 0
- mov rdx, arg(5) ;filter ptr
- mov rsi, arg(0) ;src_ptr
- mov rdi, arg(2) ;output_ptr
- mov rcx, 0x0400040
-
- movdqa xmm7, [rdx] ;load filters
- psrldq xmm7, 6
- packsswb xmm7, xmm7
- pshuflw xmm7, xmm7, 0b ;k3_k4
- punpcklwd xmm7, xmm7
-
- movq xmm6, rcx ;rounding
- pshufd xmm6, xmm6, 0
-
- movsxd rax, DWORD PTR arg(1) ;pixels_per_line
- movsxd rdx, DWORD PTR arg(3) ;out_pitch
- movsxd rcx, DWORD PTR arg(4) ;output_height
-%endm
-
-%macro APPLY_FILTER_8 1
- punpcklbw xmm0, xmm1
- pmaddubsw xmm0, xmm7
-
- paddsw xmm0, xmm6 ;rounding
- psraw xmm0, 7 ;shift
- packuswb xmm0, xmm0 ;pack back to byte
-
-%if %1
- movq xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movq [rdi], xmm0 ;store the result
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
-%endm
-
-%macro APPLY_FILTER_16 1
- punpcklbw xmm0, xmm1
- punpckhbw xmm2, xmm1
- pmaddubsw xmm0, xmm7
- pmaddubsw xmm2, xmm7
-
- paddsw xmm0, xmm6 ;rounding
- paddsw xmm2, xmm6
- psraw xmm0, 7 ;shift
- psraw xmm2, 7
- packuswb xmm0, xmm2 ;pack back to byte
-
-%if %1
- movdqu xmm1, [rdi]
- pavgb xmm0, xmm1
-%endif
- movdqu [rdi], xmm0 ;store the result
-
- lea rsi, [rsi + rax]
- lea rdi, [rdi + rdx]
- dec rcx
-%endm
-
-global sym(vp9_filter_block1d4_v2_ssse3) PRIVATE
-sym(vp9_filter_block1d4_v2_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movd xmm0, [rsi] ;load src
- movd xmm1, [rsi + rax]
-
- APPLY_FILTER_4 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_v2_ssse3) PRIVATE
-sym(vp9_filter_block1d8_v2_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movq xmm0, [rsi] ;0
- movq xmm1, [rsi + rax] ;1
-
- APPLY_FILTER_8 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_v2_ssse3) PRIVATE
-sym(vp9_filter_block1d16_v2_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + rax] ;1
- movdqa xmm2, xmm0
-
- APPLY_FILTER_16 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_v2_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d4_v2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movd xmm0, [rsi] ;load src
- movd xmm1, [rsi + rax]
-
- APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_v2_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d8_v2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movq xmm0, [rsi] ;0
- movq xmm1, [rsi + rax] ;1
-
- APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_v2_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d16_v2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;0
- movdqu xmm1, [rsi + rax] ;1
- movdqa xmm2, xmm0
-
- APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_h2_ssse3) PRIVATE
-sym(vp9_filter_block1d4_h2_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_4 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_h2_ssse3) PRIVATE
-sym(vp9_filter_block1d8_h2_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_8 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_h2_ssse3) PRIVATE
-sym(vp9_filter_block1d16_h2_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 1]
- movdqa xmm2, xmm0
-
- APPLY_FILTER_16 0
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d4_h2_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d4_h2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM_4
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_4 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d8_h2_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d8_h2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqa xmm1, xmm0
- psrldq xmm1, 1
-
- APPLY_FILTER_8 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-global sym(vp9_filter_block1d16_h2_avg_ssse3) PRIVATE
-sym(vp9_filter_block1d16_h2_avg_ssse3):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 6
- SAVE_XMM 7
- push rsi
- push rdi
- ; end prolog
-
- GET_PARAM
-.loop:
- movdqu xmm0, [rsi] ;load src
- movdqu xmm1, [rsi + 1]
- movdqa xmm2, xmm0
-
- APPLY_FILTER_16 1
- jnz .loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -12,6 +12,7 @@
#include <stdlib.h> // qsort()
#include "./vp9_rtcd.h"
+#include "./vpx_dsp_rtcd.h"
#include "./vpx_scale_rtcd.h"
#include "vpx_dsp/bitreader_buffer.h"
--- a/vp9/encoder/vp9_blockiness.c
+++ b/vp9/encoder/vp9_blockiness.c
@@ -8,12 +8,14 @@
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "./vpx_config.h"
#include "./vp9_rtcd.h"
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
#include "vp9/common/vp9_common.h"
-#include "vp9/common/vp9_convolve.h"
#include "vp9/common/vp9_filter.h"
#include "vpx/vpx_integer.h"
+#include "vpx_dsp/vpx_convolve.h"
+#include "vpx_dsp/vpx_filter.h"
#include "vpx_ports/mem.h"
static int horizontal_filter(const uint8_t *s) {--- a/vp9/encoder/vp9_denoiser.c
+++ b/vp9/encoder/vp9_denoiser.c
@@ -10,6 +10,7 @@
#include <assert.h>
#include <limits.h>
+#include "./vpx_dsp_rtcd.h"
#include "vpx_scale/yv12config.h"
#include "vpx/vpx_integer.h"
#include "vp9/common/vp9_reconinter.h"
@@ -336,12 +337,12 @@
}
if (decision == FILTER_BLOCK) {- vp9_convolve_copy(avg_start, avg.y_stride, src.buf, src.stride,
+ vpx_convolve_copy(avg_start, avg.y_stride, src.buf, src.stride,
NULL, 0, NULL, 0,
num_4x4_blocks_wide_lookup[bs] << 2,
num_4x4_blocks_high_lookup[bs] << 2);
} else { // COPY_BLOCK- vp9_convolve_copy(src.buf, src.stride, avg_start, avg.y_stride,
+ vpx_convolve_copy(src.buf, src.stride, avg_start, avg.y_stride,
NULL, 0, NULL, 0,
num_4x4_blocks_wide_lookup[bs] << 2,
num_4x4_blocks_high_lookup[bs] << 2);
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -12,11 +12,12 @@
#include <stdio.h>
#include <limits.h>
-#include "./vpx_config.h"
#include "./vp9_rtcd.h"
+#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
#include "./vpx_scale_rtcd.h"
#include "vpx/internal/vpx_psnr.h"
+#include "vpx_dsp/vpx_filter.h"
#include "vpx_ports/mem.h"
#include "vpx_ports/vpx_timer.h"
#include "vpx_scale/vpx_scale.h"
@@ -2580,18 +2581,18 @@
#if CONFIG_VP9_HIGHBITDEPTH
if (src->flags & YV12_FLAG_HIGHBITDEPTH) {- vp9_highbd_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
+ vpx_highbd_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h,
16 / factor, 16 / factor, bd);
} else {- vp9_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
+ vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h,
16 / factor, 16 / factor);
}
#else
- vp9_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
+ vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h,
16 / factor, 16 / factor);
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -1504,15 +1504,15 @@
this_mode_pred = &tmp[get_pred_buffer(tmp, 3)];
#if CONFIG_VP9_HIGHBITDEPTH
if (cm->use_highbitdepth)
- vp9_highbd_convolve_copy(best_pred->data, best_pred->stride,
+ vpx_highbd_convolve_copy(best_pred->data, best_pred->stride,
this_mode_pred->data, this_mode_pred->stride,
NULL, 0, NULL, 0, bw, bh, xd->bd);
else
- vp9_convolve_copy(best_pred->data, best_pred->stride,
+ vpx_convolve_copy(best_pred->data, best_pred->stride,
this_mode_pred->data, this_mode_pred->stride,
NULL, 0, NULL, 0, bw, bh);
#else
- vp9_convolve_copy(best_pred->data, best_pred->stride,
+ vpx_convolve_copy(best_pred->data, best_pred->stride,
this_mode_pred->data, this_mode_pred->stride,
NULL, 0, NULL, 0, bw, bh);
#endif // CONFIG_VP9_HIGHBITDEPTH
@@ -1577,15 +1577,15 @@
if (best_pred->data != orig_dst.buf && is_inter_mode(mbmi->mode)) {#if CONFIG_VP9_HIGHBITDEPTH
if (cm->use_highbitdepth)
- vp9_highbd_convolve_copy(best_pred->data, best_pred->stride,
+ vpx_highbd_convolve_copy(best_pred->data, best_pred->stride,
pd->dst.buf, pd->dst.stride, NULL, 0,
NULL, 0, bw, bh, xd->bd);
else
- vp9_convolve_copy(best_pred->data, best_pred->stride,
+ vpx_convolve_copy(best_pred->data, best_pred->stride,
pd->dst.buf, pd->dst.stride, NULL, 0,
NULL, 0, bw, bh);
#else
- vp9_convolve_copy(best_pred->data, best_pred->stride,
+ vpx_convolve_copy(best_pred->data, best_pred->stride,
pd->dst.buf, pd->dst.stride, NULL, 0,
NULL, 0, bw, bh);
#endif // CONFIG_VP9_HIGHBITDEPTH
--- a/vp9/encoder/vp9_resize.c
+++ b/vp9/encoder/vp9_resize.c
@@ -15,6 +15,9 @@
#include <stdlib.h>
#include <string.h>
+#if CONFIG_VP9_HIGHBITDEPTH
+#include "vpx_dsp/vpx_dsp_common.h"
+#endif // CONFIG_VP9_HIGHBITDEPTH
#include "vpx_ports/mem.h"
#include "vp9/common/vp9_common.h"
#include "vp9/encoder/vp9_resize.h"
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -13,14 +13,10 @@
VP9_COMMON_SRCS-yes += common/vp9_ppflags.h
VP9_COMMON_SRCS-yes += common/vp9_alloccommon.c
VP9_COMMON_SRCS-yes += common/vp9_blockd.c
-VP9_COMMON_SRCS-yes += common/vp9_convolve.c
-VP9_COMMON_SRCS-yes += common/vp9_convolve.h
VP9_COMMON_SRCS-yes += common/vp9_debugmodes.c
VP9_COMMON_SRCS-yes += common/vp9_entropy.c
VP9_COMMON_SRCS-yes += common/vp9_entropymode.c
VP9_COMMON_SRCS-yes += common/vp9_entropymv.c
-VP9_COMMON_SRCS-yes += common/vp9_filter.c
-VP9_COMMON_SRCS-yes += common/vp9_filter.h
VP9_COMMON_SRCS-yes += common/vp9_frame_buffers.c
VP9_COMMON_SRCS-yes += common/vp9_frame_buffers.h
VP9_COMMON_SRCS-yes += common/vp9_idct.c
@@ -31,6 +27,8 @@
VP9_COMMON_SRCS-yes += common/vp9_entropymode.h
VP9_COMMON_SRCS-yes += common/vp9_entropymv.h
VP9_COMMON_SRCS-yes += common/vp9_enums.h
+VP9_COMMON_SRCS-yes += common/vp9_filter.h
+VP9_COMMON_SRCS-yes += common/vp9_filter.c
VP9_COMMON_SRCS-yes += common/vp9_idct.h
VP9_COMMON_SRCS-yes += common/vp9_loopfilter.h
VP9_COMMON_SRCS-yes += common/vp9_thread_common.h
@@ -64,33 +62,16 @@
VP9_COMMON_SRCS-yes += common/vp9_scan.c
VP9_COMMON_SRCS-yes += common/vp9_scan.h
-VP9_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/convolve.h
-VP9_COMMON_SRCS-$(ARCH_X86)$(ARCH_X86_64) += common/x86/vp9_asm_stubs.c
VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.h
VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.c
VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_mfqe.h
VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_mfqe.c
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_sse2.asm
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_subpixel_8t_sse2.asm
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_subpixel_bilinear_sse2.asm
-VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_ssse3.asm
-VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_bilinear_ssse3.asm
-VP9_COMMON_SRCS-$(HAVE_AVX2) += common/x86/vp9_subpixel_8t_intrin_avx2.c
-VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_intrin_ssse3.c
ifeq ($(CONFIG_VP9_POSTPROC),yes)
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_mfqe_sse2.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_postproc_sse2.asm
endif
-ifeq ($(CONFIG_USE_X86INC),yes)
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_copy_sse2.asm
-endif
-
-ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_subpixel_8t_sse2.asm
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_subpixel_bilinear_sse2.asm
-endif
-
# common (c)
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_common_dspr2.h
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve2_avg_dspr2.c
@@ -113,15 +94,6 @@
endif
# common (msa)
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_avg_horiz_msa.c
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_avg_msa.c
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_avg_vert_msa.c
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_horiz_msa.c
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_msa.c
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_vert_msa.c
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_avg_msa.c
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_copy_msa.c
-VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_msa.h
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct4x4_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct8x8_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct16x16_msa.c
@@ -151,11 +123,6 @@
# neon with assembly and intrinsics implementations. If both are available
# prefer assembly.
ifeq ($(HAVE_NEON_ASM), yes)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_avg_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_avg_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_1_add_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_add_neon_asm$(ASM)
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_neon.c
@@ -167,11 +134,6 @@
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_add_neon_asm$(ASM)
else
ifeq ($(HAVE_NEON), yes)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_avg_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_avg_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_1_add_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_add_neon.c
VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_neon.c
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve8_avg_neon.c
@@ -1,0 +1,393 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+
+void vpx_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h);
+void vpx_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h);
+
+static INLINE int32x4_t MULTIPLY_BY_Q0(
+ int16x4_t dsrc0,
+ int16x4_t dsrc1,
+ int16x4_t dsrc2,
+ int16x4_t dsrc3,
+ int16x4_t dsrc4,
+ int16x4_t dsrc5,
+ int16x4_t dsrc6,
+ int16x4_t dsrc7,
+ int16x8_t q0s16) {+ int32x4_t qdst;
+ int16x4_t d0s16, d1s16;
+
+ d0s16 = vget_low_s16(q0s16);
+ d1s16 = vget_high_s16(q0s16);
+
+ qdst = vmull_lane_s16(dsrc0, d0s16, 0);
+ qdst = vmlal_lane_s16(qdst, dsrc1, d0s16, 1);
+ qdst = vmlal_lane_s16(qdst, dsrc2, d0s16, 2);
+ qdst = vmlal_lane_s16(qdst, dsrc3, d0s16, 3);
+ qdst = vmlal_lane_s16(qdst, dsrc4, d1s16, 0);
+ qdst = vmlal_lane_s16(qdst, dsrc5, d1s16, 1);
+ qdst = vmlal_lane_s16(qdst, dsrc6, d1s16, 2);
+ qdst = vmlal_lane_s16(qdst, dsrc7, d1s16, 3);
+ return qdst;
+}
+
+void vpx_convolve8_avg_horiz_neon(
+ const uint8_t *src,
+ ptrdiff_t src_stride,
+ uint8_t *dst,
+ ptrdiff_t dst_stride,
+ const int16_t *filter_x,
+ int x_step_q4,
+ const int16_t *filter_y, // unused
+ int y_step_q4, // unused
+ int w,
+ int h) {+ int width;
+ const uint8_t *s;
+ uint8_t *d;
+ uint8x8_t d2u8, d3u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8;
+ uint32x2_t d2u32, d3u32, d6u32, d7u32, d28u32, d29u32, d30u32, d31u32;
+ uint8x16_t q1u8, q3u8, q12u8, q13u8, q14u8, q15u8;
+ int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d22s16, d23s16;
+ int16x4_t d24s16, d25s16, d26s16, d27s16;
+ uint16x4_t d2u16, d3u16, d4u16, d5u16, d16u16, d17u16, d18u16, d19u16;
+ int16x8_t q0s16;
+ uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
+ int32x4_t q1s32, q2s32, q14s32, q15s32;
+ uint16x8x2_t q0x2u16;
+ uint8x8x2_t d0x2u8, d1x2u8;
+ uint32x2x2_t d0x2u32;
+ uint16x4x2_t d0x2u16, d1x2u16;
+ uint32x4x2_t q0x2u32;
+
+ if (x_step_q4 != 16) {+ vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4, w, h);
+ return;
+}
+
+ q0s16 = vld1q_s16(filter_x);
+
+ src -= 3; // adjust for taps
+ for (; h > 0; h -= 4) { // loop_horiz_v+ s = src;
+ d24u8 = vld1_u8(s);
+ s += src_stride;
+ d25u8 = vld1_u8(s);
+ s += src_stride;
+ d26u8 = vld1_u8(s);
+ s += src_stride;
+ d27u8 = vld1_u8(s);
+
+ q12u8 = vcombine_u8(d24u8, d25u8);
+ q13u8 = vcombine_u8(d26u8, d27u8);
+
+ q0x2u16 = vtrnq_u16(vreinterpretq_u16_u8(q12u8),
+ vreinterpretq_u16_u8(q13u8));
+ d24u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[0]));
+ d25u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[0]));
+ d26u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[1]));
+ d27u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[1]));
+ d0x2u8 = vtrn_u8(d24u8, d25u8);
+ d1x2u8 = vtrn_u8(d26u8, d27u8);
+
+ __builtin_prefetch(src + src_stride * 4);
+ __builtin_prefetch(src + src_stride * 5);
+
+ q8u16 = vmovl_u8(d0x2u8.val[0]);
+ q9u16 = vmovl_u8(d0x2u8.val[1]);
+ q10u16 = vmovl_u8(d1x2u8.val[0]);
+ q11u16 = vmovl_u8(d1x2u8.val[1]);
+
+ src += 7;
+ d16u16 = vget_low_u16(q8u16);
+ d17u16 = vget_high_u16(q8u16);
+ d18u16 = vget_low_u16(q9u16);
+ d19u16 = vget_high_u16(q9u16);
+ q8u16 = vcombine_u16(d16u16, d18u16); // vswp 17 18
+ q9u16 = vcombine_u16(d17u16, d19u16);
+
+ d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
+ d23s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); // vmov 23 21
+ for (width = w;
+ width > 0;
+ width -= 4, src += 4, dst += 4) { // loop_horiz+ s = src;
+ d28u32 = vld1_dup_u32((const uint32_t *)s);
+ s += src_stride;
+ d29u32 = vld1_dup_u32((const uint32_t *)s);
+ s += src_stride;
+ d31u32 = vld1_dup_u32((const uint32_t *)s);
+ s += src_stride;
+ d30u32 = vld1_dup_u32((const uint32_t *)s);
+
+ __builtin_prefetch(src + 64);
+
+ d0x2u16 = vtrn_u16(vreinterpret_u16_u32(d28u32),
+ vreinterpret_u16_u32(d31u32));
+ d1x2u16 = vtrn_u16(vreinterpret_u16_u32(d29u32),
+ vreinterpret_u16_u32(d30u32));
+ d0x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[0]), // d28
+ vreinterpret_u8_u16(d1x2u16.val[0])); // d29
+ d1x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[1]), // d31
+ vreinterpret_u8_u16(d1x2u16.val[1])); // d30
+
+ __builtin_prefetch(src + 64 + src_stride);
+
+ q14u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]);
+ q15u8 = vcombine_u8(d1x2u8.val[1], d1x2u8.val[0]);
+ q0x2u32 = vtrnq_u32(vreinterpretq_u32_u8(q14u8),
+ vreinterpretq_u32_u8(q15u8));
+
+ d28u8 = vreinterpret_u8_u32(vget_low_u32(q0x2u32.val[0]));
+ d29u8 = vreinterpret_u8_u32(vget_high_u32(q0x2u32.val[0]));
+ q12u16 = vmovl_u8(d28u8);
+ q13u16 = vmovl_u8(d29u8);
+
+ __builtin_prefetch(src + 64 + src_stride * 2);
+
+ d = dst;
+ d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 0);
+ d += dst_stride;
+ d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 0);
+ d += dst_stride;
+ d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 1);
+ d += dst_stride;
+ d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 1);
+
+ d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
+ d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
+ d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
+ d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
+ d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
+ d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
+ d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
+ d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
+ d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
+
+ q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d20s16, d22s16,
+ d18s16, d19s16, d23s16, d24s16, q0s16);
+ q2s32 = MULTIPLY_BY_Q0(d17s16, d20s16, d22s16, d18s16,
+ d19s16, d23s16, d24s16, d26s16, q0s16);
+ q14s32 = MULTIPLY_BY_Q0(d20s16, d22s16, d18s16, d19s16,
+ d23s16, d24s16, d26s16, d27s16, q0s16);
+ q15s32 = MULTIPLY_BY_Q0(d22s16, d18s16, d19s16, d23s16,
+ d24s16, d26s16, d27s16, d25s16, q0s16);
+
+ __builtin_prefetch(src + 64 + src_stride * 3);
+
+ d2u16 = vqrshrun_n_s32(q1s32, 7);
+ d3u16 = vqrshrun_n_s32(q2s32, 7);
+ d4u16 = vqrshrun_n_s32(q14s32, 7);
+ d5u16 = vqrshrun_n_s32(q15s32, 7);
+
+ q1u16 = vcombine_u16(d2u16, d3u16);
+ q2u16 = vcombine_u16(d4u16, d5u16);
+
+ d2u8 = vqmovn_u16(q1u16);
+ d3u8 = vqmovn_u16(q2u16);
+
+ d0x2u16 = vtrn_u16(vreinterpret_u16_u8(d2u8),
+ vreinterpret_u16_u8(d3u8));
+ d0x2u32 = vtrn_u32(vreinterpret_u32_u16(d0x2u16.val[0]),
+ vreinterpret_u32_u16(d0x2u16.val[1]));
+ d0x2u8 = vtrn_u8(vreinterpret_u8_u32(d0x2u32.val[0]),
+ vreinterpret_u8_u32(d0x2u32.val[1]));
+
+ q1u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]);
+ q3u8 = vreinterpretq_u8_u32(vcombine_u32(d6u32, d7u32));
+
+ q1u8 = vrhaddq_u8(q1u8, q3u8);
+
+ d2u32 = vreinterpret_u32_u8(vget_low_u8(q1u8));
+ d3u32 = vreinterpret_u32_u8(vget_high_u8(q1u8));
+
+ d = dst;
+ vst1_lane_u32((uint32_t *)d, d2u32, 0);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d3u32, 0);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d2u32, 1);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d3u32, 1);
+
+ q8u16 = q9u16;
+ d20s16 = d23s16;
+ q11u16 = q12u16;
+ q9u16 = q13u16;
+ d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16));
+ }
+ src += src_stride * 4 - w - 7;
+ dst += dst_stride * 4 - w;
+ }
+ return;
+}
+
+void vpx_convolve8_avg_vert_neon(
+ const uint8_t *src,
+ ptrdiff_t src_stride,
+ uint8_t *dst,
+ ptrdiff_t dst_stride,
+ const int16_t *filter_x, // unused
+ int x_step_q4, // unused
+ const int16_t *filter_y,
+ int y_step_q4,
+ int w,
+ int h) {+ int height;
+ const uint8_t *s;
+ uint8_t *d;
+ uint8x8_t d2u8, d3u8;
+ uint32x2_t d2u32, d3u32, d6u32, d7u32;
+ uint32x2_t d16u32, d18u32, d20u32, d22u32, d24u32, d26u32;
+ uint8x16_t q1u8, q3u8;
+ int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16;
+ int16x4_t d24s16, d25s16, d26s16, d27s16;
+ uint16x4_t d2u16, d3u16, d4u16, d5u16;
+ int16x8_t q0s16;
+ uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
+ int32x4_t q1s32, q2s32, q14s32, q15s32;
+
+ if (y_step_q4 != 16) {+ vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4, w, h);
+ return;
+ }
+
+ src -= src_stride * 3;
+ q0s16 = vld1q_s16(filter_y);
+ for (; w > 0; w -= 4, src += 4, dst += 4) { // loop_vert_h+ s = src;
+ d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 0);
+ s += src_stride;
+ d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 1);
+ s += src_stride;
+ d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 0);
+ s += src_stride;
+ d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 1);
+ s += src_stride;
+ d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 0);
+ s += src_stride;
+ d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 1);
+ s += src_stride;
+ d22u32 = vld1_lane_u32((const uint32_t *)s, d22u32, 0);
+ s += src_stride;
+
+ q8u16 = vmovl_u8(vreinterpret_u8_u32(d16u32));
+ q9u16 = vmovl_u8(vreinterpret_u8_u32(d18u32));
+ q10u16 = vmovl_u8(vreinterpret_u8_u32(d20u32));
+ q11u16 = vmovl_u8(vreinterpret_u8_u32(d22u32));
+
+ d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
+ d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
+ d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
+ d = dst;
+ for (height = h; height > 0; height -= 4) { // loop_vert+ d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 0);
+ s += src_stride;
+ d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 0);
+ s += src_stride;
+ d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 1);
+ s += src_stride;
+ d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 1);
+ s += src_stride;
+
+ q12u16 = vmovl_u8(vreinterpret_u8_u32(d24u32));
+ q13u16 = vmovl_u8(vreinterpret_u8_u32(d26u32));
+
+ d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 0);
+ d += dst_stride;
+ d6u32 = vld1_lane_u32((const uint32_t *)d, d6u32, 1);
+ d += dst_stride;
+ d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 0);
+ d += dst_stride;
+ d7u32 = vld1_lane_u32((const uint32_t *)d, d7u32, 1);
+ d -= dst_stride * 3;
+
+ d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
+ d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
+ d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
+ d21s16 = vreinterpret_s16_u16(vget_high_u16(q10u16));
+ d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
+ d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
+ d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
+ d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
+
+ __builtin_prefetch(s);
+ __builtin_prefetch(s + src_stride);
+ q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d18s16, d19s16,
+ d20s16, d21s16, d22s16, d24s16, q0s16);
+ __builtin_prefetch(s + src_stride * 2);
+ __builtin_prefetch(s + src_stride * 3);
+ q2s32 = MULTIPLY_BY_Q0(d17s16, d18s16, d19s16, d20s16,
+ d21s16, d22s16, d24s16, d26s16, q0s16);
+ __builtin_prefetch(d);
+ __builtin_prefetch(d + dst_stride);
+ q14s32 = MULTIPLY_BY_Q0(d18s16, d19s16, d20s16, d21s16,
+ d22s16, d24s16, d26s16, d27s16, q0s16);
+ __builtin_prefetch(d + dst_stride * 2);
+ __builtin_prefetch(d + dst_stride * 3);
+ q15s32 = MULTIPLY_BY_Q0(d19s16, d20s16, d21s16, d22s16,
+ d24s16, d26s16, d27s16, d25s16, q0s16);
+
+ d2u16 = vqrshrun_n_s32(q1s32, 7);
+ d3u16 = vqrshrun_n_s32(q2s32, 7);
+ d4u16 = vqrshrun_n_s32(q14s32, 7);
+ d5u16 = vqrshrun_n_s32(q15s32, 7);
+
+ q1u16 = vcombine_u16(d2u16, d3u16);
+ q2u16 = vcombine_u16(d4u16, d5u16);
+
+ d2u8 = vqmovn_u16(q1u16);
+ d3u8 = vqmovn_u16(q2u16);
+
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ q3u8 = vreinterpretq_u8_u32(vcombine_u32(d6u32, d7u32));
+
+ q1u8 = vrhaddq_u8(q1u8, q3u8);
+
+ d2u32 = vreinterpret_u32_u8(vget_low_u8(q1u8));
+ d3u32 = vreinterpret_u32_u8(vget_high_u8(q1u8));
+
+ vst1_lane_u32((uint32_t *)d, d2u32, 0);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d2u32, 1);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d3u32, 0);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d3u32, 1);
+ d += dst_stride;
+
+ q8u16 = q10u16;
+ d18s16 = d22s16;
+ d19s16 = d24s16;
+ q10u16 = q13u16;
+ d22s16 = d25s16;
+ }
+ }
+ return;
+}
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve8_avg_neon_asm.asm
@@ -1,0 +1,302 @@
+;
+; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+
+ ; These functions are only valid when:
+ ; x_step_q4 == 16
+ ; w%4 == 0
+ ; h%4 == 0
+ ; taps == 8
+ ; VP9_FILTER_WEIGHT == 128
+ ; VP9_FILTER_SHIFT == 7
+
+ EXPORT |vpx_convolve8_avg_horiz_neon|
+ EXPORT |vpx_convolve8_avg_vert_neon|
+ IMPORT |vpx_convolve8_avg_horiz_c|
+ IMPORT |vpx_convolve8_avg_vert_c|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+ ; Multiply and accumulate by q0
+ MACRO
+ MULTIPLY_BY_Q0 $dst, $src0, $src1, $src2, $src3, $src4, $src5, $src6, $src7
+ vmull.s16 $dst, $src0, d0[0]
+ vmlal.s16 $dst, $src1, d0[1]
+ vmlal.s16 $dst, $src2, d0[2]
+ vmlal.s16 $dst, $src3, d0[3]
+ vmlal.s16 $dst, $src4, d1[0]
+ vmlal.s16 $dst, $src5, d1[1]
+ vmlal.s16 $dst, $src6, d1[2]
+ vmlal.s16 $dst, $src7, d1[3]
+ MEND
+
+; r0 const uint8_t *src
+; r1 int src_stride
+; r2 uint8_t *dst
+; r3 int dst_stride
+; sp[]const int16_t *filter_x
+; sp[]int x_step_q4
+; sp[]const int16_t *filter_y ; unused
+; sp[]int y_step_q4 ; unused
+; sp[]int w
+; sp[]int h
+
+|vpx_convolve8_avg_horiz_neon| PROC
+ ldr r12, [sp, #4] ; x_step_q4
+ cmp r12, #16
+ bne vpx_convolve8_avg_horiz_c
+
+ push {r4-r10, lr}+
+ sub r0, r0, #3 ; adjust for taps
+
+ ldr r5, [sp, #32] ; filter_x
+ ldr r6, [sp, #48] ; w
+ ldr r7, [sp, #52] ; h
+
+ vld1.s16 {q0}, [r5] ; filter_x+
+ sub r8, r1, r1, lsl #2 ; -src_stride * 3
+ add r8, r8, #4 ; -src_stride * 3 + 4
+
+ sub r4, r3, r3, lsl #2 ; -dst_stride * 3
+ add r4, r4, #4 ; -dst_stride * 3 + 4
+
+ rsb r9, r6, r1, lsl #2 ; reset src for outer loop
+ sub r9, r9, #7
+ rsb r12, r6, r3, lsl #2 ; reset dst for outer loop
+
+ mov r10, r6 ; w loop counter
+
+vpx_convolve8_avg_loop_horiz_v
+ vld1.8 {d24}, [r0], r1+ vld1.8 {d25}, [r0], r1+ vld1.8 {d26}, [r0], r1+ vld1.8 {d27}, [r0], r8+
+ vtrn.16 q12, q13
+ vtrn.8 d24, d25
+ vtrn.8 d26, d27
+
+ pld [r0, r1, lsl #2]
+
+ vmovl.u8 q8, d24
+ vmovl.u8 q9, d25
+ vmovl.u8 q10, d26
+ vmovl.u8 q11, d27
+
+ ; save a few instructions in the inner loop
+ vswp d17, d18
+ vmov d23, d21
+
+ add r0, r0, #3
+
+vpx_convolve8_avg_loop_horiz
+ add r5, r0, #64
+
+ vld1.32 {d28[]}, [r0], r1+ vld1.32 {d29[]}, [r0], r1+ vld1.32 {d31[]}, [r0], r1+ vld1.32 {d30[]}, [r0], r8+
+ pld [r5]
+
+ vtrn.16 d28, d31
+ vtrn.16 d29, d30
+ vtrn.8 d28, d29
+ vtrn.8 d31, d30
+
+ pld [r5, r1]
+
+ ; extract to s16
+ vtrn.32 q14, q15
+ vmovl.u8 q12, d28
+ vmovl.u8 q13, d29
+
+ pld [r5, r1, lsl #1]
+
+ ; slightly out of order load to match the existing data
+ vld1.u32 {d6[0]}, [r2], r3+ vld1.u32 {d7[0]}, [r2], r3+ vld1.u32 {d6[1]}, [r2], r3+ vld1.u32 {d7[1]}, [r2], r3+
+ sub r2, r2, r3, lsl #2 ; reset for store
+
+ ; src[] * filter_x
+ MULTIPLY_BY_Q0 q1, d16, d17, d20, d22, d18, d19, d23, d24
+ MULTIPLY_BY_Q0 q2, d17, d20, d22, d18, d19, d23, d24, d26
+ MULTIPLY_BY_Q0 q14, d20, d22, d18, d19, d23, d24, d26, d27
+ MULTIPLY_BY_Q0 q15, d22, d18, d19, d23, d24, d26, d27, d25
+
+ pld [r5, -r8]
+
+ ; += 64 >> 7
+ vqrshrun.s32 d2, q1, #7
+ vqrshrun.s32 d3, q2, #7
+ vqrshrun.s32 d4, q14, #7
+ vqrshrun.s32 d5, q15, #7
+
+ ; saturate
+ vqmovn.u16 d2, q1
+ vqmovn.u16 d3, q2
+
+ ; transpose
+ vtrn.16 d2, d3
+ vtrn.32 d2, d3
+ vtrn.8 d2, d3
+
+ ; average the new value and the dst value
+ vrhadd.u8 q1, q1, q3
+
+ vst1.u32 {d2[0]}, [r2@32], r3+ vst1.u32 {d3[0]}, [r2@32], r3+ vst1.u32 {d2[1]}, [r2@32], r3+ vst1.u32 {d3[1]}, [r2@32], r4+
+ vmov q8, q9
+ vmov d20, d23
+ vmov q11, q12
+ vmov q9, q13
+
+ subs r6, r6, #4 ; w -= 4
+ bgt vpx_convolve8_avg_loop_horiz
+
+ ; outer loop
+ mov r6, r10 ; restore w counter
+ add r0, r0, r9 ; src += src_stride * 4 - w
+ add r2, r2, r12 ; dst += dst_stride * 4 - w
+ subs r7, r7, #4 ; h -= 4
+ bgt vpx_convolve8_avg_loop_horiz_v
+
+ pop {r4-r10, pc}+
+ ENDP
+
+|vpx_convolve8_avg_vert_neon| PROC
+ ldr r12, [sp, #12]
+ cmp r12, #16
+ bne vpx_convolve8_avg_vert_c
+
+ push {r4-r8, lr}+
+ ; adjust for taps
+ sub r0, r0, r1
+ sub r0, r0, r1, lsl #1
+
+ ldr r4, [sp, #32] ; filter_y
+ ldr r6, [sp, #40] ; w
+ ldr lr, [sp, #44] ; h
+
+ vld1.s16 {q0}, [r4] ; filter_y+
+ lsl r1, r1, #1
+ lsl r3, r3, #1
+
+vpx_convolve8_avg_loop_vert_h
+ mov r4, r0
+ add r7, r0, r1, asr #1
+ mov r5, r2
+ add r8, r2, r3, asr #1
+ mov r12, lr ; h loop counter
+
+ vld1.u32 {d16[0]}, [r4], r1+ vld1.u32 {d16[1]}, [r7], r1+ vld1.u32 {d18[0]}, [r4], r1+ vld1.u32 {d18[1]}, [r7], r1+ vld1.u32 {d20[0]}, [r4], r1+ vld1.u32 {d20[1]}, [r7], r1+ vld1.u32 {d22[0]}, [r4], r1+
+ vmovl.u8 q8, d16
+ vmovl.u8 q9, d18
+ vmovl.u8 q10, d20
+ vmovl.u8 q11, d22
+
+vpx_convolve8_avg_loop_vert
+ ; always process a 4x4 block at a time
+ vld1.u32 {d24[0]}, [r7], r1+ vld1.u32 {d26[0]}, [r4], r1+ vld1.u32 {d26[1]}, [r7], r1+ vld1.u32 {d24[1]}, [r4], r1+
+ ; extract to s16
+ vmovl.u8 q12, d24
+ vmovl.u8 q13, d26
+
+ vld1.u32 {d6[0]}, [r5@32], r3+ vld1.u32 {d6[1]}, [r8@32], r3+ vld1.u32 {d7[0]}, [r5@32], r3+ vld1.u32 {d7[1]}, [r8@32], r3+
+ pld [r7]
+ pld [r4]
+
+ ; src[] * filter_y
+ MULTIPLY_BY_Q0 q1, d16, d17, d18, d19, d20, d21, d22, d24
+
+ pld [r7, r1]
+ pld [r4, r1]
+
+ MULTIPLY_BY_Q0 q2, d17, d18, d19, d20, d21, d22, d24, d26
+
+ pld [r5]
+ pld [r8]
+
+ MULTIPLY_BY_Q0 q14, d18, d19, d20, d21, d22, d24, d26, d27
+
+ pld [r5, r3]
+ pld [r8, r3]
+
+ MULTIPLY_BY_Q0 q15, d19, d20, d21, d22, d24, d26, d27, d25
+
+ ; += 64 >> 7
+ vqrshrun.s32 d2, q1, #7
+ vqrshrun.s32 d3, q2, #7
+ vqrshrun.s32 d4, q14, #7
+ vqrshrun.s32 d5, q15, #7
+
+ ; saturate
+ vqmovn.u16 d2, q1
+ vqmovn.u16 d3, q2
+
+ ; average the new value and the dst value
+ vrhadd.u8 q1, q1, q3
+
+ sub r5, r5, r3, lsl #1 ; reset for store
+ sub r8, r8, r3, lsl #1
+
+ vst1.u32 {d2[0]}, [r5@32], r3+ vst1.u32 {d2[1]}, [r8@32], r3+ vst1.u32 {d3[0]}, [r5@32], r3+ vst1.u32 {d3[1]}, [r8@32], r3+
+ vmov q8, q10
+ vmov d18, d22
+ vmov d19, d24
+ vmov q10, q13
+ vmov d22, d25
+
+ subs r12, r12, #4 ; h -= 4
+ bgt vpx_convolve8_avg_loop_vert
+
+ ; outer loop
+ add r0, r0, #4
+ add r2, r2, #4
+ subs r6, r6, #4 ; w -= 4
+ bgt vpx_convolve8_avg_loop_vert_h
+
+ pop {r4-r8, pc}+
+ ENDP
+ END
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve8_neon.c
@@ -1,0 +1,360 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+
+void vpx_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h);
+void vpx_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h);
+
+static INLINE int32x4_t MULTIPLY_BY_Q0(
+ int16x4_t dsrc0,
+ int16x4_t dsrc1,
+ int16x4_t dsrc2,
+ int16x4_t dsrc3,
+ int16x4_t dsrc4,
+ int16x4_t dsrc5,
+ int16x4_t dsrc6,
+ int16x4_t dsrc7,
+ int16x8_t q0s16) {+ int32x4_t qdst;
+ int16x4_t d0s16, d1s16;
+
+ d0s16 = vget_low_s16(q0s16);
+ d1s16 = vget_high_s16(q0s16);
+
+ qdst = vmull_lane_s16(dsrc0, d0s16, 0);
+ qdst = vmlal_lane_s16(qdst, dsrc1, d0s16, 1);
+ qdst = vmlal_lane_s16(qdst, dsrc2, d0s16, 2);
+ qdst = vmlal_lane_s16(qdst, dsrc3, d0s16, 3);
+ qdst = vmlal_lane_s16(qdst, dsrc4, d1s16, 0);
+ qdst = vmlal_lane_s16(qdst, dsrc5, d1s16, 1);
+ qdst = vmlal_lane_s16(qdst, dsrc6, d1s16, 2);
+ qdst = vmlal_lane_s16(qdst, dsrc7, d1s16, 3);
+ return qdst;
+}
+
+void vpx_convolve8_horiz_neon(
+ const uint8_t *src,
+ ptrdiff_t src_stride,
+ uint8_t *dst,
+ ptrdiff_t dst_stride,
+ const int16_t *filter_x,
+ int x_step_q4,
+ const int16_t *filter_y, // unused
+ int y_step_q4, // unused
+ int w,
+ int h) {+ int width;
+ const uint8_t *s, *psrc;
+ uint8_t *d, *pdst;
+ uint8x8_t d2u8, d3u8, d24u8, d25u8, d26u8, d27u8, d28u8, d29u8;
+ uint32x2_t d2u32, d3u32, d28u32, d29u32, d30u32, d31u32;
+ uint8x16_t q12u8, q13u8, q14u8, q15u8;
+ int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d22s16, d23s16;
+ int16x4_t d24s16, d25s16, d26s16, d27s16;
+ uint16x4_t d2u16, d3u16, d4u16, d5u16, d16u16, d17u16, d18u16, d19u16;
+ int16x8_t q0s16;
+ uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
+ int32x4_t q1s32, q2s32, q14s32, q15s32;
+ uint16x8x2_t q0x2u16;
+ uint8x8x2_t d0x2u8, d1x2u8;
+ uint32x2x2_t d0x2u32;
+ uint16x4x2_t d0x2u16, d1x2u16;
+ uint32x4x2_t q0x2u32;
+
+ if (x_step_q4 != 16) {+ vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4, w, h);
+ return;
+ }
+
+ q0s16 = vld1q_s16(filter_x);
+
+ src -= 3; // adjust for taps
+ for (; h > 0; h -= 4,
+ src += src_stride * 4,
+ dst += dst_stride * 4) { // loop_horiz_v+ s = src;
+ d24u8 = vld1_u8(s);
+ s += src_stride;
+ d25u8 = vld1_u8(s);
+ s += src_stride;
+ d26u8 = vld1_u8(s);
+ s += src_stride;
+ d27u8 = vld1_u8(s);
+
+ q12u8 = vcombine_u8(d24u8, d25u8);
+ q13u8 = vcombine_u8(d26u8, d27u8);
+
+ q0x2u16 = vtrnq_u16(vreinterpretq_u16_u8(q12u8),
+ vreinterpretq_u16_u8(q13u8));
+ d24u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[0]));
+ d25u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[0]));
+ d26u8 = vreinterpret_u8_u16(vget_low_u16(q0x2u16.val[1]));
+ d27u8 = vreinterpret_u8_u16(vget_high_u16(q0x2u16.val[1]));
+ d0x2u8 = vtrn_u8(d24u8, d25u8);
+ d1x2u8 = vtrn_u8(d26u8, d27u8);
+
+ __builtin_prefetch(src + src_stride * 4);
+ __builtin_prefetch(src + src_stride * 5);
+ __builtin_prefetch(src + src_stride * 6);
+
+ q8u16 = vmovl_u8(d0x2u8.val[0]);
+ q9u16 = vmovl_u8(d0x2u8.val[1]);
+ q10u16 = vmovl_u8(d1x2u8.val[0]);
+ q11u16 = vmovl_u8(d1x2u8.val[1]);
+
+ d16u16 = vget_low_u16(q8u16);
+ d17u16 = vget_high_u16(q8u16);
+ d18u16 = vget_low_u16(q9u16);
+ d19u16 = vget_high_u16(q9u16);
+ q8u16 = vcombine_u16(d16u16, d18u16); // vswp 17 18
+ q9u16 = vcombine_u16(d17u16, d19u16);
+
+ d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
+ d23s16 = vreinterpret_s16_u16(vget_high_u16(q10u16)); // vmov 23 21
+ for (width = w, psrc = src + 7, pdst = dst;
+ width > 0;
+ width -= 4, psrc += 4, pdst += 4) { // loop_horiz+ s = psrc;
+ d28u32 = vld1_dup_u32((const uint32_t *)s);
+ s += src_stride;
+ d29u32 = vld1_dup_u32((const uint32_t *)s);
+ s += src_stride;
+ d31u32 = vld1_dup_u32((const uint32_t *)s);
+ s += src_stride;
+ d30u32 = vld1_dup_u32((const uint32_t *)s);
+
+ __builtin_prefetch(psrc + 64);
+
+ d0x2u16 = vtrn_u16(vreinterpret_u16_u32(d28u32),
+ vreinterpret_u16_u32(d31u32));
+ d1x2u16 = vtrn_u16(vreinterpret_u16_u32(d29u32),
+ vreinterpret_u16_u32(d30u32));
+ d0x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[0]), // d28
+ vreinterpret_u8_u16(d1x2u16.val[0])); // d29
+ d1x2u8 = vtrn_u8(vreinterpret_u8_u16(d0x2u16.val[1]), // d31
+ vreinterpret_u8_u16(d1x2u16.val[1])); // d30
+
+ __builtin_prefetch(psrc + 64 + src_stride);
+
+ q14u8 = vcombine_u8(d0x2u8.val[0], d0x2u8.val[1]);
+ q15u8 = vcombine_u8(d1x2u8.val[1], d1x2u8.val[0]);
+ q0x2u32 = vtrnq_u32(vreinterpretq_u32_u8(q14u8),
+ vreinterpretq_u32_u8(q15u8));
+
+ d28u8 = vreinterpret_u8_u32(vget_low_u32(q0x2u32.val[0]));
+ d29u8 = vreinterpret_u8_u32(vget_high_u32(q0x2u32.val[0]));
+ q12u16 = vmovl_u8(d28u8);
+ q13u16 = vmovl_u8(d29u8);
+
+ __builtin_prefetch(psrc + 64 + src_stride * 2);
+
+ d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
+ d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
+ d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
+ d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
+ d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
+ d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
+ d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
+ d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
+ d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
+
+ q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d20s16, d22s16,
+ d18s16, d19s16, d23s16, d24s16, q0s16);
+ q2s32 = MULTIPLY_BY_Q0(d17s16, d20s16, d22s16, d18s16,
+ d19s16, d23s16, d24s16, d26s16, q0s16);
+ q14s32 = MULTIPLY_BY_Q0(d20s16, d22s16, d18s16, d19s16,
+ d23s16, d24s16, d26s16, d27s16, q0s16);
+ q15s32 = MULTIPLY_BY_Q0(d22s16, d18s16, d19s16, d23s16,
+ d24s16, d26s16, d27s16, d25s16, q0s16);
+
+ __builtin_prefetch(psrc + 60 + src_stride * 3);
+
+ d2u16 = vqrshrun_n_s32(q1s32, 7);
+ d3u16 = vqrshrun_n_s32(q2s32, 7);
+ d4u16 = vqrshrun_n_s32(q14s32, 7);
+ d5u16 = vqrshrun_n_s32(q15s32, 7);
+
+ q1u16 = vcombine_u16(d2u16, d3u16);
+ q2u16 = vcombine_u16(d4u16, d5u16);
+
+ d2u8 = vqmovn_u16(q1u16);
+ d3u8 = vqmovn_u16(q2u16);
+
+ d0x2u16 = vtrn_u16(vreinterpret_u16_u8(d2u8),
+ vreinterpret_u16_u8(d3u8));
+ d0x2u32 = vtrn_u32(vreinterpret_u32_u16(d0x2u16.val[0]),
+ vreinterpret_u32_u16(d0x2u16.val[1]));
+ d0x2u8 = vtrn_u8(vreinterpret_u8_u32(d0x2u32.val[0]),
+ vreinterpret_u8_u32(d0x2u32.val[1]));
+
+ d2u32 = vreinterpret_u32_u8(d0x2u8.val[0]);
+ d3u32 = vreinterpret_u32_u8(d0x2u8.val[1]);
+
+ d = pdst;
+ vst1_lane_u32((uint32_t *)d, d2u32, 0);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d3u32, 0);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d2u32, 1);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d3u32, 1);
+
+ q8u16 = q9u16;
+ d20s16 = d23s16;
+ q11u16 = q12u16;
+ q9u16 = q13u16;
+ d23s16 = vreinterpret_s16_u16(vget_high_u16(q11u16));
+ }
+ }
+ return;
+}
+
+void vpx_convolve8_vert_neon(
+ const uint8_t *src,
+ ptrdiff_t src_stride,
+ uint8_t *dst,
+ ptrdiff_t dst_stride,
+ const int16_t *filter_x, // unused
+ int x_step_q4, // unused
+ const int16_t *filter_y,
+ int y_step_q4,
+ int w,
+ int h) {+ int height;
+ const uint8_t *s;
+ uint8_t *d;
+ uint32x2_t d2u32, d3u32;
+ uint32x2_t d16u32, d18u32, d20u32, d22u32, d24u32, d26u32;
+ int16x4_t d16s16, d17s16, d18s16, d19s16, d20s16, d21s16, d22s16;
+ int16x4_t d24s16, d25s16, d26s16, d27s16;
+ uint16x4_t d2u16, d3u16, d4u16, d5u16;
+ int16x8_t q0s16;
+ uint16x8_t q1u16, q2u16, q8u16, q9u16, q10u16, q11u16, q12u16, q13u16;
+ int32x4_t q1s32, q2s32, q14s32, q15s32;
+
+ if (y_step_q4 != 16) {+ vpx_convolve8_vert_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4, w, h);
+ return;
+ }
+
+ src -= src_stride * 3;
+ q0s16 = vld1q_s16(filter_y);
+ for (; w > 0; w -= 4, src += 4, dst += 4) { // loop_vert_h+ s = src;
+ d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 0);
+ s += src_stride;
+ d16u32 = vld1_lane_u32((const uint32_t *)s, d16u32, 1);
+ s += src_stride;
+ d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 0);
+ s += src_stride;
+ d18u32 = vld1_lane_u32((const uint32_t *)s, d18u32, 1);
+ s += src_stride;
+ d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 0);
+ s += src_stride;
+ d20u32 = vld1_lane_u32((const uint32_t *)s, d20u32, 1);
+ s += src_stride;
+ d22u32 = vld1_lane_u32((const uint32_t *)s, d22u32, 0);
+ s += src_stride;
+
+ q8u16 = vmovl_u8(vreinterpret_u8_u32(d16u32));
+ q9u16 = vmovl_u8(vreinterpret_u8_u32(d18u32));
+ q10u16 = vmovl_u8(vreinterpret_u8_u32(d20u32));
+ q11u16 = vmovl_u8(vreinterpret_u8_u32(d22u32));
+
+ d18s16 = vreinterpret_s16_u16(vget_low_u16(q9u16));
+ d19s16 = vreinterpret_s16_u16(vget_high_u16(q9u16));
+ d22s16 = vreinterpret_s16_u16(vget_low_u16(q11u16));
+ d = dst;
+ for (height = h; height > 0; height -= 4) { // loop_vert+ d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 0);
+ s += src_stride;
+ d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 0);
+ s += src_stride;
+ d26u32 = vld1_lane_u32((const uint32_t *)s, d26u32, 1);
+ s += src_stride;
+ d24u32 = vld1_lane_u32((const uint32_t *)s, d24u32, 1);
+ s += src_stride;
+
+ q12u16 = vmovl_u8(vreinterpret_u8_u32(d24u32));
+ q13u16 = vmovl_u8(vreinterpret_u8_u32(d26u32));
+
+ d16s16 = vreinterpret_s16_u16(vget_low_u16(q8u16));
+ d17s16 = vreinterpret_s16_u16(vget_high_u16(q8u16));
+ d20s16 = vreinterpret_s16_u16(vget_low_u16(q10u16));
+ d21s16 = vreinterpret_s16_u16(vget_high_u16(q10u16));
+ d24s16 = vreinterpret_s16_u16(vget_low_u16(q12u16));
+ d25s16 = vreinterpret_s16_u16(vget_high_u16(q12u16));
+ d26s16 = vreinterpret_s16_u16(vget_low_u16(q13u16));
+ d27s16 = vreinterpret_s16_u16(vget_high_u16(q13u16));
+
+ __builtin_prefetch(d);
+ __builtin_prefetch(d + dst_stride);
+ q1s32 = MULTIPLY_BY_Q0(d16s16, d17s16, d18s16, d19s16,
+ d20s16, d21s16, d22s16, d24s16, q0s16);
+ __builtin_prefetch(d + dst_stride * 2);
+ __builtin_prefetch(d + dst_stride * 3);
+ q2s32 = MULTIPLY_BY_Q0(d17s16, d18s16, d19s16, d20s16,
+ d21s16, d22s16, d24s16, d26s16, q0s16);
+ __builtin_prefetch(s);
+ __builtin_prefetch(s + src_stride);
+ q14s32 = MULTIPLY_BY_Q0(d18s16, d19s16, d20s16, d21s16,
+ d22s16, d24s16, d26s16, d27s16, q0s16);
+ __builtin_prefetch(s + src_stride * 2);
+ __builtin_prefetch(s + src_stride * 3);
+ q15s32 = MULTIPLY_BY_Q0(d19s16, d20s16, d21s16, d22s16,
+ d24s16, d26s16, d27s16, d25s16, q0s16);
+
+ d2u16 = vqrshrun_n_s32(q1s32, 7);
+ d3u16 = vqrshrun_n_s32(q2s32, 7);
+ d4u16 = vqrshrun_n_s32(q14s32, 7);
+ d5u16 = vqrshrun_n_s32(q15s32, 7);
+
+ q1u16 = vcombine_u16(d2u16, d3u16);
+ q2u16 = vcombine_u16(d4u16, d5u16);
+
+ d2u32 = vreinterpret_u32_u8(vqmovn_u16(q1u16));
+ d3u32 = vreinterpret_u32_u8(vqmovn_u16(q2u16));
+
+ vst1_lane_u32((uint32_t *)d, d2u32, 0);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d2u32, 1);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d3u32, 0);
+ d += dst_stride;
+ vst1_lane_u32((uint32_t *)d, d3u32, 1);
+ d += dst_stride;
+
+ q8u16 = q10u16;
+ d18s16 = d22s16;
+ d19s16 = d24s16;
+ q10u16 = q13u16;
+ d22s16 = d25s16;
+ }
+ }
+ return;
+}
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve8_neon_asm.asm
@@ -1,0 +1,280 @@
+;
+; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+
+ ; These functions are only valid when:
+ ; x_step_q4 == 16
+ ; w%4 == 0
+ ; h%4 == 0
+ ; taps == 8
+ ; VP9_FILTER_WEIGHT == 128
+ ; VP9_FILTER_SHIFT == 7
+
+ EXPORT |vpx_convolve8_horiz_neon|
+ EXPORT |vpx_convolve8_vert_neon|
+ IMPORT |vpx_convolve8_horiz_c|
+ IMPORT |vpx_convolve8_vert_c|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+ ; Multiply and accumulate by q0
+ MACRO
+ MULTIPLY_BY_Q0 $dst, $src0, $src1, $src2, $src3, $src4, $src5, $src6, $src7
+ vmull.s16 $dst, $src0, d0[0]
+ vmlal.s16 $dst, $src1, d0[1]
+ vmlal.s16 $dst, $src2, d0[2]
+ vmlal.s16 $dst, $src3, d0[3]
+ vmlal.s16 $dst, $src4, d1[0]
+ vmlal.s16 $dst, $src5, d1[1]
+ vmlal.s16 $dst, $src6, d1[2]
+ vmlal.s16 $dst, $src7, d1[3]
+ MEND
+
+; r0 const uint8_t *src
+; r1 int src_stride
+; r2 uint8_t *dst
+; r3 int dst_stride
+; sp[]const int16_t *filter_x
+; sp[]int x_step_q4
+; sp[]const int16_t *filter_y ; unused
+; sp[]int y_step_q4 ; unused
+; sp[]int w
+; sp[]int h
+
+|vpx_convolve8_horiz_neon| PROC
+ ldr r12, [sp, #4] ; x_step_q4
+ cmp r12, #16
+ bne vpx_convolve8_horiz_c
+
+ push {r4-r10, lr}+
+ sub r0, r0, #3 ; adjust for taps
+
+ ldr r5, [sp, #32] ; filter_x
+ ldr r6, [sp, #48] ; w
+ ldr r7, [sp, #52] ; h
+
+ vld1.s16 {q0}, [r5] ; filter_x+
+ sub r8, r1, r1, lsl #2 ; -src_stride * 3
+ add r8, r8, #4 ; -src_stride * 3 + 4
+
+ sub r4, r3, r3, lsl #2 ; -dst_stride * 3
+ add r4, r4, #4 ; -dst_stride * 3 + 4
+
+ rsb r9, r6, r1, lsl #2 ; reset src for outer loop
+ sub r9, r9, #7
+ rsb r12, r6, r3, lsl #2 ; reset dst for outer loop
+
+ mov r10, r6 ; w loop counter
+
+vpx_convolve8_loop_horiz_v
+ vld1.8 {d24}, [r0], r1+ vld1.8 {d25}, [r0], r1+ vld1.8 {d26}, [r0], r1+ vld1.8 {d27}, [r0], r8+
+ vtrn.16 q12, q13
+ vtrn.8 d24, d25
+ vtrn.8 d26, d27
+
+ pld [r0, r1, lsl #2]
+
+ vmovl.u8 q8, d24
+ vmovl.u8 q9, d25
+ vmovl.u8 q10, d26
+ vmovl.u8 q11, d27
+
+ ; save a few instructions in the inner loop
+ vswp d17, d18
+ vmov d23, d21
+
+ add r0, r0, #3
+
+vpx_convolve8_loop_horiz
+ add r5, r0, #64
+
+ vld1.32 {d28[]}, [r0], r1+ vld1.32 {d29[]}, [r0], r1+ vld1.32 {d31[]}, [r0], r1+ vld1.32 {d30[]}, [r0], r8+
+ pld [r5]
+
+ vtrn.16 d28, d31
+ vtrn.16 d29, d30
+ vtrn.8 d28, d29
+ vtrn.8 d31, d30
+
+ pld [r5, r1]
+
+ ; extract to s16
+ vtrn.32 q14, q15
+ vmovl.u8 q12, d28
+ vmovl.u8 q13, d29
+
+ pld [r5, r1, lsl #1]
+
+ ; src[] * filter_x
+ MULTIPLY_BY_Q0 q1, d16, d17, d20, d22, d18, d19, d23, d24
+ MULTIPLY_BY_Q0 q2, d17, d20, d22, d18, d19, d23, d24, d26
+ MULTIPLY_BY_Q0 q14, d20, d22, d18, d19, d23, d24, d26, d27
+ MULTIPLY_BY_Q0 q15, d22, d18, d19, d23, d24, d26, d27, d25
+
+ pld [r5, -r8]
+
+ ; += 64 >> 7
+ vqrshrun.s32 d2, q1, #7
+ vqrshrun.s32 d3, q2, #7
+ vqrshrun.s32 d4, q14, #7
+ vqrshrun.s32 d5, q15, #7
+
+ ; saturate
+ vqmovn.u16 d2, q1
+ vqmovn.u16 d3, q2
+
+ ; transpose
+ vtrn.16 d2, d3
+ vtrn.32 d2, d3
+ vtrn.8 d2, d3
+
+ vst1.u32 {d2[0]}, [r2@32], r3+ vst1.u32 {d3[0]}, [r2@32], r3+ vst1.u32 {d2[1]}, [r2@32], r3+ vst1.u32 {d3[1]}, [r2@32], r4+
+ vmov q8, q9
+ vmov d20, d23
+ vmov q11, q12
+ vmov q9, q13
+
+ subs r6, r6, #4 ; w -= 4
+ bgt vpx_convolve8_loop_horiz
+
+ ; outer loop
+ mov r6, r10 ; restore w counter
+ add r0, r0, r9 ; src += src_stride * 4 - w
+ add r2, r2, r12 ; dst += dst_stride * 4 - w
+ subs r7, r7, #4 ; h -= 4
+ bgt vpx_convolve8_loop_horiz_v
+
+ pop {r4-r10, pc}+
+ ENDP
+
+|vpx_convolve8_vert_neon| PROC
+ ldr r12, [sp, #12]
+ cmp r12, #16
+ bne vpx_convolve8_vert_c
+
+ push {r4-r8, lr}+
+ ; adjust for taps
+ sub r0, r0, r1
+ sub r0, r0, r1, lsl #1
+
+ ldr r4, [sp, #32] ; filter_y
+ ldr r6, [sp, #40] ; w
+ ldr lr, [sp, #44] ; h
+
+ vld1.s16 {q0}, [r4] ; filter_y+
+ lsl r1, r1, #1
+ lsl r3, r3, #1
+
+vpx_convolve8_loop_vert_h
+ mov r4, r0
+ add r7, r0, r1, asr #1
+ mov r5, r2
+ add r8, r2, r3, asr #1
+ mov r12, lr ; h loop counter
+
+ vld1.u32 {d16[0]}, [r4], r1+ vld1.u32 {d16[1]}, [r7], r1+ vld1.u32 {d18[0]}, [r4], r1+ vld1.u32 {d18[1]}, [r7], r1+ vld1.u32 {d20[0]}, [r4], r1+ vld1.u32 {d20[1]}, [r7], r1+ vld1.u32 {d22[0]}, [r4], r1+
+ vmovl.u8 q8, d16
+ vmovl.u8 q9, d18
+ vmovl.u8 q10, d20
+ vmovl.u8 q11, d22
+
+vpx_convolve8_loop_vert
+ ; always process a 4x4 block at a time
+ vld1.u32 {d24[0]}, [r7], r1+ vld1.u32 {d26[0]}, [r4], r1+ vld1.u32 {d26[1]}, [r7], r1+ vld1.u32 {d24[1]}, [r4], r1+
+ ; extract to s16
+ vmovl.u8 q12, d24
+ vmovl.u8 q13, d26
+
+ pld [r5]
+ pld [r8]
+
+ ; src[] * filter_y
+ MULTIPLY_BY_Q0 q1, d16, d17, d18, d19, d20, d21, d22, d24
+
+ pld [r5, r3]
+ pld [r8, r3]
+
+ MULTIPLY_BY_Q0 q2, d17, d18, d19, d20, d21, d22, d24, d26
+
+ pld [r7]
+ pld [r4]
+
+ MULTIPLY_BY_Q0 q14, d18, d19, d20, d21, d22, d24, d26, d27
+
+ pld [r7, r1]
+ pld [r4, r1]
+
+ MULTIPLY_BY_Q0 q15, d19, d20, d21, d22, d24, d26, d27, d25
+
+ ; += 64 >> 7
+ vqrshrun.s32 d2, q1, #7
+ vqrshrun.s32 d3, q2, #7
+ vqrshrun.s32 d4, q14, #7
+ vqrshrun.s32 d5, q15, #7
+
+ ; saturate
+ vqmovn.u16 d2, q1
+ vqmovn.u16 d3, q2
+
+ vst1.u32 {d2[0]}, [r5@32], r3+ vst1.u32 {d2[1]}, [r8@32], r3+ vst1.u32 {d3[0]}, [r5@32], r3+ vst1.u32 {d3[1]}, [r8@32], r3+
+ vmov q8, q10
+ vmov d18, d22
+ vmov d19, d24
+ vmov q10, q13
+ vmov d22, d25
+
+ subs r12, r12, #4 ; h -= 4
+ bgt vpx_convolve8_loop_vert
+
+ ; outer loop
+ add r0, r0, #4
+ add r2, r2, #4
+ subs r6, r6, #4 ; w -= 4
+ bgt vpx_convolve8_loop_vert_h
+
+ pop {r4-r8, pc}+
+ ENDP
+ END
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve_avg_neon.c
@@ -1,0 +1,147 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+
+void vpx_convolve_avg_neon(
+ const uint8_t *src, // r0
+ ptrdiff_t src_stride, // r1
+ uint8_t *dst, // r2
+ ptrdiff_t dst_stride, // r3
+ const int16_t *filter_x,
+ int filter_x_stride,
+ const int16_t *filter_y,
+ int filter_y_stride,
+ int w,
+ int h) {+ uint8_t *d;
+ uint8x8_t d0u8, d1u8, d2u8, d3u8;
+ uint32x2_t d0u32, d2u32;
+ uint8x16_t q0u8, q1u8, q2u8, q3u8, q8u8, q9u8, q10u8, q11u8;
+ (void)filter_x; (void)filter_x_stride;
+ (void)filter_y; (void)filter_y_stride;
+
+ d = dst;
+ if (w > 32) { // avg64+ for (; h > 0; h -= 1) {+ q0u8 = vld1q_u8(src);
+ q1u8 = vld1q_u8(src + 16);
+ q2u8 = vld1q_u8(src + 32);
+ q3u8 = vld1q_u8(src + 48);
+ src += src_stride;
+ q8u8 = vld1q_u8(d);
+ q9u8 = vld1q_u8(d + 16);
+ q10u8 = vld1q_u8(d + 32);
+ q11u8 = vld1q_u8(d + 48);
+ d += dst_stride;
+
+ q0u8 = vrhaddq_u8(q0u8, q8u8);
+ q1u8 = vrhaddq_u8(q1u8, q9u8);
+ q2u8 = vrhaddq_u8(q2u8, q10u8);
+ q3u8 = vrhaddq_u8(q3u8, q11u8);
+
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q1u8);
+ vst1q_u8(dst + 32, q2u8);
+ vst1q_u8(dst + 48, q3u8);
+ dst += dst_stride;
+ }
+ } else if (w == 32) { // avg32+ for (; h > 0; h -= 2) {+ q0u8 = vld1q_u8(src);
+ q1u8 = vld1q_u8(src + 16);
+ src += src_stride;
+ q2u8 = vld1q_u8(src);
+ q3u8 = vld1q_u8(src + 16);
+ src += src_stride;
+ q8u8 = vld1q_u8(d);
+ q9u8 = vld1q_u8(d + 16);
+ d += dst_stride;
+ q10u8 = vld1q_u8(d);
+ q11u8 = vld1q_u8(d + 16);
+ d += dst_stride;
+
+ q0u8 = vrhaddq_u8(q0u8, q8u8);
+ q1u8 = vrhaddq_u8(q1u8, q9u8);
+ q2u8 = vrhaddq_u8(q2u8, q10u8);
+ q3u8 = vrhaddq_u8(q3u8, q11u8);
+
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q1u8);
+ dst += dst_stride;
+ vst1q_u8(dst, q2u8);
+ vst1q_u8(dst + 16, q3u8);
+ dst += dst_stride;
+ }
+ } else if (w > 8) { // avg16+ for (; h > 0; h -= 2) {+ q0u8 = vld1q_u8(src);
+ src += src_stride;
+ q1u8 = vld1q_u8(src);
+ src += src_stride;
+ q2u8 = vld1q_u8(d);
+ d += dst_stride;
+ q3u8 = vld1q_u8(d);
+ d += dst_stride;
+
+ q0u8 = vrhaddq_u8(q0u8, q2u8);
+ q1u8 = vrhaddq_u8(q1u8, q3u8);
+
+ vst1q_u8(dst, q0u8);
+ dst += dst_stride;
+ vst1q_u8(dst, q1u8);
+ dst += dst_stride;
+ }
+ } else if (w == 8) { // avg8+ for (; h > 0; h -= 2) {+ d0u8 = vld1_u8(src);
+ src += src_stride;
+ d1u8 = vld1_u8(src);
+ src += src_stride;
+ d2u8 = vld1_u8(d);
+ d += dst_stride;
+ d3u8 = vld1_u8(d);
+ d += dst_stride;
+
+ q0u8 = vcombine_u8(d0u8, d1u8);
+ q1u8 = vcombine_u8(d2u8, d3u8);
+ q0u8 = vrhaddq_u8(q0u8, q1u8);
+
+ vst1_u8(dst, vget_low_u8(q0u8));
+ dst += dst_stride;
+ vst1_u8(dst, vget_high_u8(q0u8));
+ dst += dst_stride;
+ }
+ } else { // avg4+ for (; h > 0; h -= 2) {+ d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 0);
+ src += src_stride;
+ d0u32 = vld1_lane_u32((const uint32_t *)src, d0u32, 1);
+ src += src_stride;
+ d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 0);
+ d += dst_stride;
+ d2u32 = vld1_lane_u32((const uint32_t *)d, d2u32, 1);
+ d += dst_stride;
+
+ d0u8 = vrhadd_u8(vreinterpret_u8_u32(d0u32),
+ vreinterpret_u8_u32(d2u32));
+
+ d0u32 = vreinterpret_u32_u8(d0u8);
+ vst1_lane_u32((uint32_t *)dst, d0u32, 0);
+ dst += dst_stride;
+ vst1_lane_u32((uint32_t *)dst, d0u32, 1);
+ dst += dst_stride;
+ }
+ }
+ return;
+}
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve_avg_neon_asm.asm
@@ -1,0 +1,116 @@
+;
+; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+ EXPORT |vpx_convolve_avg_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+|vpx_convolve_avg_neon| PROC
+ push {r4-r6, lr}+ ldrd r4, r5, [sp, #32]
+ mov r6, r2
+
+ cmp r4, #32
+ bgt avg64
+ beq avg32
+ cmp r4, #8
+ bgt avg16
+ beq avg8
+ b avg4
+
+avg64
+ sub lr, r1, #32
+ sub r4, r3, #32
+avg64_h
+ pld [r0, r1, lsl #1]
+ vld1.8 {q0-q1}, [r0]!+ vld1.8 {q2-q3}, [r0], lr+ pld [r2, r3]
+ vld1.8 {q8-q9}, [r6@128]!+ vld1.8 {q10-q11}, [r6@128], r4+ vrhadd.u8 q0, q0, q8
+ vrhadd.u8 q1, q1, q9
+ vrhadd.u8 q2, q2, q10
+ vrhadd.u8 q3, q3, q11
+ vst1.8 {q0-q1}, [r2@128]!+ vst1.8 {q2-q3}, [r2@128], r4+ subs r5, r5, #1
+ bgt avg64_h
+ pop {r4-r6, pc}+
+avg32
+ vld1.8 {q0-q1}, [r0], r1+ vld1.8 {q2-q3}, [r0], r1+ vld1.8 {q8-q9}, [r6@128], r3+ vld1.8 {q10-q11}, [r6@128], r3+ pld [r0]
+ vrhadd.u8 q0, q0, q8
+ pld [r0, r1]
+ vrhadd.u8 q1, q1, q9
+ pld [r6]
+ vrhadd.u8 q2, q2, q10
+ pld [r6, r3]
+ vrhadd.u8 q3, q3, q11
+ vst1.8 {q0-q1}, [r2@128], r3+ vst1.8 {q2-q3}, [r2@128], r3+ subs r5, r5, #2
+ bgt avg32
+ pop {r4-r6, pc}+
+avg16
+ vld1.8 {q0}, [r0], r1+ vld1.8 {q1}, [r0], r1+ vld1.8 {q2}, [r6@128], r3+ vld1.8 {q3}, [r6@128], r3+ pld [r0]
+ pld [r0, r1]
+ vrhadd.u8 q0, q0, q2
+ pld [r6]
+ pld [r6, r3]
+ vrhadd.u8 q1, q1, q3
+ vst1.8 {q0}, [r2@128], r3+ vst1.8 {q1}, [r2@128], r3+ subs r5, r5, #2
+ bgt avg16
+ pop {r4-r6, pc}+
+avg8
+ vld1.8 {d0}, [r0], r1+ vld1.8 {d1}, [r0], r1+ vld1.8 {d2}, [r6@64], r3+ vld1.8 {d3}, [r6@64], r3+ pld [r0]
+ pld [r0, r1]
+ vrhadd.u8 q0, q0, q1
+ pld [r6]
+ pld [r6, r3]
+ vst1.8 {d0}, [r2@64], r3+ vst1.8 {d1}, [r2@64], r3+ subs r5, r5, #2
+ bgt avg8
+ pop {r4-r6, pc}+
+avg4
+ vld1.32 {d0[0]}, [r0], r1+ vld1.32 {d0[1]}, [r0], r1+ vld1.32 {d2[0]}, [r6@32], r3+ vld1.32 {d2[1]}, [r6@32], r3+ vrhadd.u8 d0, d0, d2
+ vst1.32 {d0[0]}, [r2@32], r3+ vst1.32 {d0[1]}, [r2@32], r3+ subs r5, r5, #2
+ bgt avg4
+ pop {r4-r6, pc}+ ENDP
+
+ END
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve_copy_neon.c
@@ -1,0 +1,94 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+
+void vpx_convolve_copy_neon(
+ const uint8_t *src, // r0
+ ptrdiff_t src_stride, // r1
+ uint8_t *dst, // r2
+ ptrdiff_t dst_stride, // r3
+ const int16_t *filter_x,
+ int filter_x_stride,
+ const int16_t *filter_y,
+ int filter_y_stride,
+ int w,
+ int h) {+ uint8x8_t d0u8, d2u8;
+ uint8x16_t q0u8, q1u8, q2u8, q3u8;
+ (void)filter_x; (void)filter_x_stride;
+ (void)filter_y; (void)filter_y_stride;
+
+ if (w > 32) { // copy64+ for (; h > 0; h--) {+ q0u8 = vld1q_u8(src);
+ q1u8 = vld1q_u8(src + 16);
+ q2u8 = vld1q_u8(src + 32);
+ q3u8 = vld1q_u8(src + 48);
+ src += src_stride;
+
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q1u8);
+ vst1q_u8(dst + 32, q2u8);
+ vst1q_u8(dst + 48, q3u8);
+ dst += dst_stride;
+ }
+ } else if (w == 32) { // copy32+ for (; h > 0; h -= 2) {+ q0u8 = vld1q_u8(src);
+ q1u8 = vld1q_u8(src + 16);
+ src += src_stride;
+ q2u8 = vld1q_u8(src);
+ q3u8 = vld1q_u8(src + 16);
+ src += src_stride;
+
+ vst1q_u8(dst, q0u8);
+ vst1q_u8(dst + 16, q1u8);
+ dst += dst_stride;
+ vst1q_u8(dst, q2u8);
+ vst1q_u8(dst + 16, q3u8);
+ dst += dst_stride;
+ }
+ } else if (w > 8) { // copy16+ for (; h > 0; h -= 2) {+ q0u8 = vld1q_u8(src);
+ src += src_stride;
+ q1u8 = vld1q_u8(src);
+ src += src_stride;
+
+ vst1q_u8(dst, q0u8);
+ dst += dst_stride;
+ vst1q_u8(dst, q1u8);
+ dst += dst_stride;
+ }
+ } else if (w == 8) { // copy8+ for (; h > 0; h -= 2) {+ d0u8 = vld1_u8(src);
+ src += src_stride;
+ d2u8 = vld1_u8(src);
+ src += src_stride;
+
+ vst1_u8(dst, d0u8);
+ dst += dst_stride;
+ vst1_u8(dst, d2u8);
+ dst += dst_stride;
+ }
+ } else { // copy4+ for (; h > 0; h--) {+ *(uint32_t *)dst = *(const uint32_t *)src;
+ src += src_stride;
+ dst += dst_stride;
+ }
+ }
+ return;
+}
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve_copy_neon_asm.asm
@@ -1,0 +1,84 @@
+;
+; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+ EXPORT |vpx_convolve_copy_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+|vpx_convolve_copy_neon| PROC
+ push {r4-r5, lr}+ ldrd r4, r5, [sp, #28]
+
+ cmp r4, #32
+ bgt copy64
+ beq copy32
+ cmp r4, #8
+ bgt copy16
+ beq copy8
+ b copy4
+
+copy64
+ sub lr, r1, #32
+ sub r3, r3, #32
+copy64_h
+ pld [r0, r1, lsl #1]
+ vld1.8 {q0-q1}, [r0]!+ vld1.8 {q2-q3}, [r0], lr+ vst1.8 {q0-q1}, [r2@128]!+ vst1.8 {q2-q3}, [r2@128], r3+ subs r5, r5, #1
+ bgt copy64_h
+ pop {r4-r5, pc}+
+copy32
+ pld [r0, r1, lsl #1]
+ vld1.8 {q0-q1}, [r0], r1+ pld [r0, r1, lsl #1]
+ vld1.8 {q2-q3}, [r0], r1+ vst1.8 {q0-q1}, [r2@128], r3+ vst1.8 {q2-q3}, [r2@128], r3+ subs r5, r5, #2
+ bgt copy32
+ pop {r4-r5, pc}+
+copy16
+ pld [r0, r1, lsl #1]
+ vld1.8 {q0}, [r0], r1+ pld [r0, r1, lsl #1]
+ vld1.8 {q1}, [r0], r1+ vst1.8 {q0}, [r2@128], r3+ vst1.8 {q1}, [r2@128], r3+ subs r5, r5, #2
+ bgt copy16
+ pop {r4-r5, pc}+
+copy8
+ pld [r0, r1, lsl #1]
+ vld1.8 {d0}, [r0], r1+ pld [r0, r1, lsl #1]
+ vld1.8 {d2}, [r0], r1+ vst1.8 {d0}, [r2@64], r3+ vst1.8 {d2}, [r2@64], r3+ subs r5, r5, #2
+ bgt copy8
+ pop {r4-r5, pc}+
+copy4
+ ldr r12, [r0], r1
+ str r12, [r2], r3
+ subs r5, r5, #1
+ bgt copy4
+ pop {r4-r5, pc}+ ENDP
+
+ END
--- /dev/null
+++ b/vpx_dsp/arm/vpx_convolve_neon.c
@@ -1,0 +1,82 @@
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/vpx_dsp_common.h"
+#include "vpx_ports/mem.h"
+
+void vpx_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ /* Given our constraints: w <= 64, h <= 64, taps == 8 we can reduce the
+ * maximum buffer size to 64 * 64 + 7 (+ 1 to make it divisible by 4).
+ */
+ DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);
+
+ // Account for the vertical phase needing 3 lines prior and 4 lines post
+ int intermediate_height = h + 7;
+
+ if (x_step_q4 != 16 || y_step_q4 != 16) {+ vpx_convolve8_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ /* Filter starting 3 lines back. The neon implementation will ignore the
+ * given height and filter a multiple of 4 lines. Since this goes in to
+ * the temp buffer which has lots of extra room and is subsequently discarded
+ * this is safe if somewhat less than ideal.
+ */
+ vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride,
+ temp, 64,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, intermediate_height);
+
+ /* Step into the temp buffer 3 lines to get the actual frame data */
+ vpx_convolve8_vert_neon(temp + 64 * 3, 64,
+ dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+}
+
+void vpx_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ DECLARE_ALIGNED(8, uint8_t, temp[64 * 72]);
+ int intermediate_height = h + 7;
+
+ if (x_step_q4 != 16 || y_step_q4 != 16) {+ vpx_convolve8_avg_c(src, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ /* This implementation has the same issues as above. In addition, we only want
+ * to average the values after both passes.
+ */
+ vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride,
+ temp, 64,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, intermediate_height);
+ vpx_convolve8_avg_vert_neon(temp + 64 * 3,
+ 64, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+}
--- a/vpx_dsp/loopfilter.c
+++ b/vpx_dsp/loopfilter.c
@@ -8,6 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <stdlib.h>
+
#include "./vpx_config.h"
#include "vpx_dsp/vpx_dsp_common.h"
#include "vpx_ports/mem.h"
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve8_avg_horiz_msa.c
@@ -1,0 +1,782 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/mips/vpx_convolve_msa.h"
+
+static void common_hz_8t_and_aver_dst_4x4_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 dst0, dst1, dst2, dst3, res2, res3;
+ v16u8 mask0, mask1, mask2, mask3;
+ v8i16 filt, res0, res1;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[16]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
+ filt0, filt1, filt2, filt3, res0, res1);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ SRARI_H2_SH(res0, res1, FILTER_BITS);
+ SAT_SH2_SH(res0, res1, 7);
+ PCKEV_B2_UB(res0, res0, res1, res1, res2, res3);
+ ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
+ XORI_B2_128_UB(res2, res3);
+ AVER_UB2_UB(res2, dst0, res3, dst2, res2, res3);
+ ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hz_8t_and_aver_dst_4x8_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 mask0, mask1, mask2, mask3, res0, res1, res2, res3;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v8i16 filt, vec0, vec1, vec2, vec3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[16]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ src += (4 * src_stride);
+ LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+ HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
+ filt0, filt1, filt2, filt3, vec0, vec1);
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
+ filt0, filt1, filt2, filt3, vec2, vec3);
+ SRARI_H4_SH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ SAT_SH4_SH(vec0, vec1, vec2, vec3, 7);
+ PCKEV_B4_UB(vec0, vec0, vec1, vec1, vec2, vec2, vec3, vec3, res0, res1, res2,
+ res3);
+ ILVR_D2_UB(res1, res0, res3, res2, res0, res2);
+ XORI_B2_128_UB(res0, res2);
+ ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2, dst4,
+ dst6);
+ ILVR_D2_UB(dst2, dst0, dst6, dst4, dst0, dst4);
+ AVER_UB2_UB(res0, dst0, res2, dst4, res0, res2);
+ ST4x8_UB(res0, res2, dst, dst_stride);
+}
+
+static void common_hz_8t_and_aver_dst_4w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ if (4 == height) {+ common_hz_8t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else if (8 == height) {+ common_hz_8t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
+ }
+}
+
+static void common_hz_8t_and_aver_dst_8w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ int32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 mask0, mask1, mask2, mask3, dst0, dst1, dst2, dst3;
+ v8i16 filt, out0, out1, out2, out3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ src += (4 * src_stride);
+ HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
+ mask3, filt0, filt1, filt2, filt3, out0, out1,
+ out2, out3);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ CONVERT_UB_AVG_ST8x4_UB(out0, out1, out2, out3, dst0, dst1, dst2, dst3,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void common_hz_8t_and_aver_dst_16w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ int32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 mask0, mask1, mask2, mask3, dst0, dst1;
+ v8i16 filt, out0, out1, out2, out3;
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ for (loop_cnt = height >> 1; loop_cnt--;) {+ LD_SB2(src, src_stride, src0, src2);
+ LD_SB2(src + 8, src_stride, src1, src3);
+ src += (2 * src_stride);
+
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8, vec12);
+ VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9, vec13);
+ VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
+ vec14);
+ VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
+ vec15);
+ DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
+ vec9, vec10, vec11);
+ DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0, vec1,
+ vec2, vec3);
+ DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
+ vec9, vec10, vec11);
+ ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
+ out2, out3);
+ LD_UB2(dst, dst_stride, dst0, dst1);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ PCKEV_XORI128_AVG_ST_UB(out1, out0, dst0, dst);
+ dst += dst_stride;
+ PCKEV_XORI128_AVG_ST_UB(out3, out2, dst1, dst);
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_8t_and_aver_dst_32w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 dst1, dst2, mask0, mask1, mask2, mask3;
+ v8i16 filt, out0, out1, out2, out3;
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ for (loop_cnt = height; loop_cnt--;) {+ src0 = LD_SB(src);
+ src2 = LD_SB(src + 16);
+ src3 = LD_SB(src + 24);
+ src1 = __msa_sldi_b(src2, src0, 8);
+ src += src_stride;
+
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8, vec12);
+ VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9, vec13);
+ VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
+ vec14);
+ VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
+ vec15);
+ DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
+ vec9, vec10, vec11);
+ DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0, vec1,
+ vec2, vec3);
+ DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
+ vec9, vec10, vec11);
+ ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
+ out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ LD_UB2(dst, 16, dst1, dst2);
+ PCKEV_XORI128_AVG_ST_UB(out1, out0, dst1, dst);
+ PCKEV_XORI128_AVG_ST_UB(out3, out2, dst2, dst + 16);
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_8t_and_aver_dst_64w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt, cnt;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 dst1, dst2, mask0, mask1, mask2, mask3;
+ v8i16 filt, out0, out1, out2, out3;
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8i16 vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ for (loop_cnt = height; loop_cnt--;) {+ for (cnt = 0; cnt < 2; ++cnt) {+ src0 = LD_SB(&src[cnt << 5]);
+ src2 = LD_SB(&src[16 + (cnt << 5)]);
+ src3 = LD_SB(&src[24 + (cnt << 5)]);
+ src1 = __msa_sldi_b(src2, src0, 8);
+
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ VSHF_B4_SH(src0, src0, mask0, mask1, mask2, mask3, vec0, vec4, vec8,
+ vec12);
+ VSHF_B4_SH(src1, src1, mask0, mask1, mask2, mask3, vec1, vec5, vec9,
+ vec13);
+ VSHF_B4_SH(src2, src2, mask0, mask1, mask2, mask3, vec2, vec6, vec10,
+ vec14);
+ VSHF_B4_SH(src3, src3, mask0, mask1, mask2, mask3, vec3, vec7, vec11,
+ vec15);
+ DOTP_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0,
+ vec1, vec2, vec3);
+ DOTP_SB4_SH(vec8, vec9, vec10, vec11, filt2, filt2, filt2, filt2, vec8,
+ vec9, vec10, vec11);
+ DPADD_SB4_SH(vec4, vec5, vec6, vec7, filt1, filt1, filt1, filt1, vec0,
+ vec1, vec2, vec3);
+ DPADD_SB4_SH(vec12, vec13, vec14, vec15, filt3, filt3, filt3, filt3, vec8,
+ vec9, vec10, vec11);
+ ADDS_SH4_SH(vec0, vec8, vec1, vec9, vec2, vec10, vec3, vec11, out0, out1,
+ out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ LD_UB2(&dst[cnt << 5], 16, dst1, dst2);
+ PCKEV_XORI128_AVG_ST_UB(out1, out0, dst1, &dst[cnt << 5]);
+ PCKEV_XORI128_AVG_ST_UB(out3, out2, dst2, &dst[16 + (cnt << 5)]);
+ }
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_2t_and_aver_dst_4x4_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, mask;
+ v16u8 filt0, dst0, dst1, dst2, dst3, vec0, vec1, res0, res1;
+ v8u16 vec2, vec3, const255, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[16]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ const255 = (v8u16)__msa_ldi_h(255);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, vec2, vec3);
+ SRARI_H2_UH(vec2, vec3, FILTER_BITS);
+ MIN_UH2_UH(vec2, vec3, const255);
+ PCKEV_B2_UB(vec2, vec2, vec3, vec3, res0, res1);
+ ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
+ AVER_UB2_UB(res0, dst0, res1, dst2, res0, res1);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hz_2t_and_aver_dst_4x8_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt0, vec0, vec1, vec2, vec3, res0, res1, res2, res3;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v8u16 vec4, vec5, vec6, vec7, const255, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[16]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ const255 = (v8u16)__msa_ldi_h(255);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+ VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src4, src5, src6, src7, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec4, vec5,
+ vec6, vec7);
+ SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS);
+ MIN_UH4_UH(vec4, vec5, vec6, vec7, const255);
+ PCKEV_B4_UB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1, res2,
+ res3);
+ ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2, dst4,
+ dst6);
+ AVER_UB4_UB(res0, dst0, res1, dst2, res2, dst4, res3, dst6, res0, res1, res2,
+ res3);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hz_2t_and_aver_dst_4w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ if (4 == height) {+ common_hz_2t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else if (8 == height) {+ common_hz_2t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
+ }
+}
+
+static void common_hz_2t_and_aver_dst_8x4_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, mask;
+ v16u8 filt0, dst0, dst1, dst2, dst3;
+ v8u16 vec0, vec1, vec2, vec3, const255, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ const255 = (v8u16)__msa_ldi_h(255);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
+ dst, dst_stride);
+}
+
+static void common_hz_2t_and_aver_dst_8x8mult_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ v16i8 src0, src1, src2, src3, mask;
+ v16u8 filt0, dst0, dst1, dst2, dst3;
+ v8u16 vec0, vec1, vec2, vec3, const255, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ const255 = (v8u16)__msa_ldi_h(255);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+ PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ if (16 == height) {+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ PCKEV_AVG_ST8x4_UB(vec0, dst0, vec1, dst1, vec2, dst2, vec3, dst3,
+ dst, dst_stride);
+ }
+}
+
+static void common_hz_2t_and_aver_dst_8w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ if (4 == height) {+ common_hz_2t_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else {+ common_hz_2t_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
+ filter, height);
+ }
+}
+
+static void common_hz_2t_and_aver_dst_16w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt0, dst0, dst1, dst2, dst3;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 res0, res1, res2, res3, res4, res5, res6, res7, const255, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ const255 = (v8u16)__msa_ldi_h(255);
+
+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
+ res2, res3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
+ res6, res7);
+ SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
+ SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ MIN_UH4_UH(res0, res1, res2, res3, const255);
+ MIN_UH4_UH(res4, res5, res6, res7, const255);
+ PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
+ dst += dst_stride;
+ PCKEV_AVG_ST_UB(res3, res2, dst1, dst);
+ dst += dst_stride;
+ PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
+ dst += dst_stride;
+ PCKEV_AVG_ST_UB(res7, res6, dst3, dst);
+ dst += dst_stride;
+
+ for (loop_cnt = (height >> 2) - 1; loop_cnt--;) {+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
+ res2, res3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
+ res6, res7);
+ SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
+ SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ MIN_UH4_UH(res0, res1, res2, res3, const255);
+ MIN_UH4_UH(res4, res5, res6, res7, const255);
+ PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
+ dst += dst_stride;
+ PCKEV_AVG_ST_UB(res3, res2, dst1, dst);
+ dst += dst_stride;
+ PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
+ dst += dst_stride;
+ PCKEV_AVG_ST_UB(res7, res6, dst3, dst);
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_2t_and_aver_dst_32w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt0, dst0, dst1, dst2, dst3;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 res0, res1, res2, res3, res4, res5, res6, res7, const255, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ const255 = (v8u16)__msa_ldi_h(255);
+
+ for (loop_cnt = (height >> 1); loop_cnt--;) {+ src0 = LD_SB(src);
+ src2 = LD_SB(src + 16);
+ src3 = LD_SB(src + 24);
+ src1 = __msa_sldi_b(src2, src0, 8);
+ src += src_stride;
+ src4 = LD_SB(src);
+ src6 = LD_SB(src + 16);
+ src7 = LD_SB(src + 24);
+ src5 = __msa_sldi_b(src6, src4, 8);
+ src += src_stride;
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, res0, res1,
+ res2, res3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, res4, res5,
+ res6, res7);
+ SRARI_H4_UH(res0, res1, res2, res3, FILTER_BITS);
+ SRARI_H4_UH(res4, res5, res6, res7, FILTER_BITS);
+ MIN_UH4_UH(res0, res1, res2, res3, const255);
+ MIN_UH4_UH(res4, res5, res6, res7, const255);
+ LD_UB2(dst, 16, dst0, dst1);
+ PCKEV_AVG_ST_UB(res1, res0, dst0, dst);
+ PCKEV_AVG_ST_UB(res3, res2, dst1, (dst + 16));
+ dst += dst_stride;
+ LD_UB2(dst, 16, dst2, dst3);
+ PCKEV_AVG_ST_UB(res5, res4, dst2, dst);
+ PCKEV_AVG_ST_UB(res7, res6, dst3, (dst + 16));
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_2t_and_aver_dst_64w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt0, dst0, dst1, dst2, dst3;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 out0, out1, out2, out3, out4, out5, out6, out7, const255, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ const255 = (v8u16)__msa_ldi_h(255);
+
+ for (loop_cnt = height; loop_cnt--;) {+ LD_SB4(src, 16, src0, src2, src4, src6);
+ src7 = LD_SB(src + 56);
+ SLDI_B3_SB(src2, src4, src6, src0, src2, src4, src1, src3, src5, 8);
+ src += src_stride;
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
+ out2, out3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
+ out6, out7);
+ SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
+ SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
+ LD_UB4(dst, 16, dst0, dst1, dst2, dst3);
+ MIN_UH4_UH(out0, out1, out2, out3, const255);
+ MIN_UH4_UH(out4, out5, out6, out7, const255);
+ PCKEV_AVG_ST_UB(out1, out0, dst0, dst);
+ PCKEV_AVG_ST_UB(out3, out2, dst1, dst + 16);
+ PCKEV_AVG_ST_UB(out5, out4, dst2, dst + 32);
+ PCKEV_AVG_ST_UB(out7, out6, dst3, dst + 48);
+ dst += dst_stride;
+ }
+}
+
+void vpx_convolve8_avg_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ int8_t cnt, filt_hor[8];
+
+ if (16 != x_step_q4) {+ vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ if (((const int32_t *)filter_x)[1] == 0x800000) {+ vpx_convolve_avg(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ for (cnt = 0; cnt < 8; ++cnt) {+ filt_hor[cnt] = filter_x[cnt];
+ }
+
+ if (((const int32_t *)filter_x)[0] == 0) {+ switch (w) {+ case 4:
+ common_hz_2t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ case 8:
+ common_hz_2t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ case 16:
+ common_hz_2t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ case 32:
+ common_hz_2t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ case 64:
+ common_hz_2t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ default:
+ vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {+ switch (w) {+ case 4:
+ common_hz_8t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ case 8:
+ common_hz_8t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ case 16:
+ common_hz_8t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ case 32:
+ common_hz_8t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ case 64:
+ common_hz_8t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ default:
+ vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve8_avg_msa.c
@@ -1,0 +1,679 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/mips/vpx_convolve_msa.h"
+
+static void common_hv_8ht_8vt_and_aver_dst_4w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16u8 dst0, dst1, dst2, dst3, mask0, mask1, mask2, mask3, tmp0, tmp1;
+ v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
+ v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
+ v8i16 hz_out7, hz_out8, hz_out9, res0, res1, vec0, vec1, vec2, vec3, vec4;
+ v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[16]);
+ src -= (3 + 3 * src_stride);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+
+ hz_out0 = HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out2 = HORIZ_8TAP_FILT(src2, src3, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out4 = HORIZ_8TAP_FILT(src4, src5, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out5 = HORIZ_8TAP_FILT(src5, src6, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ SLDI_B2_SH(hz_out2, hz_out4, hz_out0, hz_out2, hz_out1, hz_out3, 8);
+
+ filt = LD_SH(filter_vert);
+ SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
+
+ ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ vec2 = (v8i16)__msa_ilvev_b((v16i8)hz_out5, (v16i8)hz_out4);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ XORI_B4_128_SB(src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ hz_out7 = HORIZ_8TAP_FILT(src7, src8, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ hz_out6 = (v8i16)__msa_sldi_b((v16i8)hz_out7, (v16i8)hz_out5, 8);
+ vec3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
+ res0 = FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ hz_out9 = HORIZ_8TAP_FILT(src9, src10, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ hz_out8 = (v8i16)__msa_sldi_b((v16i8)hz_out9, (v16i8)hz_out7, 8);
+ vec4 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
+ res1 = FILT_8TAP_DPADD_S_H(vec1, vec2, vec3, vec4, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+ ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
+
+ SRARI_H2_SH(res0, res1, FILTER_BITS);
+ SAT_SH2_SH(res0, res1, 7);
+ PCKEV_B2_UB(res0, res0, res1, res1, tmp0, tmp1);
+ XORI_B2_128_UB(tmp0, tmp1);
+ AVER_UB2_UB(tmp0, dst0, tmp1, dst2, tmp0, tmp1);
+ ST4x4_UB(tmp0, tmp1, 0, 1, 0, 1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ hz_out5 = hz_out9;
+ vec0 = vec2;
+ vec1 = vec3;
+ vec2 = vec4;
+ }
+}
+
+static void common_hv_8ht_8vt_and_aver_dst_8w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
+ v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
+ v16u8 dst0, dst1, dst2, dst3, mask0, mask1, mask2, mask3;
+ v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
+ v8i16 hz_out7, hz_out8, hz_out9, hz_out10, tmp0, tmp1, tmp2, tmp3;
+ v8i16 out0, out1, out2, out3, out4, out5, out6, out7, out8, out9;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= (3 + 3 * src_stride);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ hz_out0 = HORIZ_8TAP_FILT(src0, src0, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out1 = HORIZ_8TAP_FILT(src1, src1, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out2 = HORIZ_8TAP_FILT(src2, src2, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out3 = HORIZ_8TAP_FILT(src3, src3, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out4 = HORIZ_8TAP_FILT(src4, src4, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out5 = HORIZ_8TAP_FILT(src5, src5, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out6 = HORIZ_8TAP_FILT(src6, src6, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+
+ filt = LD_SH(filter_vert);
+ SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
+
+ ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1);
+ ILVEV_B2_SH(hz_out4, hz_out5, hz_out1, hz_out2, out2, out4);
+ ILVEV_B2_SH(hz_out3, hz_out4, hz_out5, hz_out6, out5, out6);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ XORI_B4_128_SB(src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+
+ hz_out7 = HORIZ_8TAP_FILT(src7, src7, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
+ tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ hz_out8 = HORIZ_8TAP_FILT(src8, src8, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ out7 = (v8i16)__msa_ilvev_b((v16i8)hz_out8, (v16i8)hz_out7);
+ tmp1 = FILT_8TAP_DPADD_S_H(out4, out5, out6, out7, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ hz_out9 = HORIZ_8TAP_FILT(src9, src9, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ out8 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
+ tmp2 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out8, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ hz_out10 = HORIZ_8TAP_FILT(src10, src10, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ out9 = (v8i16)__msa_ilvev_b((v16i8)hz_out10, (v16i8)hz_out9);
+ tmp3 = FILT_8TAP_DPADD_S_H(out5, out6, out7, out9, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+ CONVERT_UB_AVG_ST8x4_UB(tmp0, tmp1, tmp2, tmp3, dst0, dst1, dst2, dst3,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ hz_out6 = hz_out10;
+ out0 = out2;
+ out1 = out3;
+ out2 = out8;
+ out4 = out6;
+ out5 = out7;
+ out6 = out9;
+ }
+}
+
+static void common_hv_8ht_8vt_and_aver_dst_16w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 2; multiple8_cnt--;) {+ common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert, height);
+ src += 8;
+ dst += 8;
+ }
+}
+
+static void common_hv_8ht_8vt_and_aver_dst_32w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 4; multiple8_cnt--;) {+ common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert, height);
+ src += 8;
+ dst += 8;
+ }
+}
+
+static void common_hv_8ht_8vt_and_aver_dst_64w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 8; multiple8_cnt--;) {+ common_hv_8ht_8vt_and_aver_dst_8w_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert, height);
+ src += 8;
+ dst += 8;
+ }
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_4x4_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert) {+ v16i8 src0, src1, src2, src3, src4, mask;
+ v16u8 filt_hz, filt_vt, vec0, vec1;
+ v16u8 dst0, dst1, dst2, dst3, res0, res1;
+ v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, tmp0, tmp1, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[16]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ filt = LD_UH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
+ hz_out4 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ hz_out1 = (v8u16)__msa_sldi_b((v16i8)hz_out2, (v16i8)hz_out0, 8);
+ hz_out3 = (v8u16)__msa_pckod_d((v2i64)hz_out4, (v2i64)hz_out2);
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_B2_UB(tmp0, tmp0, tmp1, tmp1, res0, res1);
+ AVER_UB2_UB(res0, dst0, res1, dst2, res0, res1);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_4x8_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert) {+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, mask;
+ v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3, res0, res1, res2, res3;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
+ v8u16 hz_out7, hz_out8, tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[16]);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+ src8 = LD_SB(src);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
+ hz_out4 = HORIZ_2TAP_FILT_UH(src4, src5, mask, filt_hz, FILTER_BITS);
+ hz_out6 = HORIZ_2TAP_FILT_UH(src6, src7, mask, filt_hz, FILTER_BITS);
+ hz_out8 = HORIZ_2TAP_FILT_UH(src8, src8, mask, filt_hz, FILTER_BITS);
+ SLDI_B3_UH(hz_out2, hz_out4, hz_out6, hz_out0, hz_out2, hz_out4, hz_out1,
+ hz_out3, hz_out5, 8);
+ hz_out7 = (v8u16)__msa_pckod_d((v2i64)hz_out8, (v2i64)hz_out6);
+
+ LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+ ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst2,
+ dst4, dst6);
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ ILVEV_B2_UB(hz_out4, hz_out5, hz_out6, hz_out7, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt_vt, filt_vt, filt_vt, filt_vt,
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_B4_UB(tmp0, tmp0, tmp1, tmp1, tmp2, tmp2, tmp3, tmp3, res0, res1,
+ res2, res3);
+ AVER_UB4_UB(res0, dst0, res1, dst2, res2, dst4, res3, dst6, res0, res1,
+ res2, res3);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_4w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ if (4 == height) {+ common_hv_2ht_2vt_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert);
+ } else if (8 == height) {+ common_hv_2ht_2vt_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert);
+ }
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_8x4_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert) {+ v16i8 src0, src1, src2, src3, src4, mask;
+ v16u8 filt_hz, filt_vt, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
+ v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+ src += (5 * src_stride);
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
+ hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp0 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
+ vec1 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp1 = __msa_dotp_u_h(vec1, filt_vt);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
+ vec2 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp2 = __msa_dotp_u_h(vec2, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ vec3 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp3 = __msa_dotp_u_h(vec3, filt_vt);
+
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3,
+ dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_8x8mult_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, mask;
+ v16u8 filt_hz, filt_vt, vec0, dst0, dst1, dst2, dst3;
+ v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_SB(src);
+ src += src_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src1, src2, src3, src4);
+ src += (4 * src_stride);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp0 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp1 = __msa_dotp_u_h(vec0, filt_vt);
+
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp2 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp3 = __msa_dotp_u_h(vec0, filt_vt);
+
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_8w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ if (4 == height) {+ common_hv_2ht_2vt_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert);
+ } else {+ common_hv_2ht_2vt_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert,
+ height);
+ }
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_16w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt_hz, filt_vt, vec0, vec1, dst0, dst1, dst2, dst3;
+ v8u16 hz_out0, hz_out1, hz_out2, hz_out3, tmp0, tmp1;
+ v8i16 filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB2(src, 8, src0, src1);
+ src += src_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
+ hz_out3 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
+ dst += dst_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
+ ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst1, dst);
+ dst += dst_stride;
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ hz_out3 = HORIZ_2TAP_FILT_UH(src5, src5, mask, filt_hz, FILTER_BITS);
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst);
+ dst += dst_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src6, src6, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src7, src7, mask, filt_hz, FILTER_BITS);
+ ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst3, dst);
+ dst += dst_stride;
+ }
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_32w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 2; multiple8_cnt--;) {+ common_hv_2ht_2vt_and_aver_dst_16w_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert, height);
+ src += 16;
+ dst += 16;
+ }
+}
+
+static void common_hv_2ht_2vt_and_aver_dst_64w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 4; multiple8_cnt--;) {+ common_hv_2ht_2vt_and_aver_dst_16w_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert, height);
+ src += 16;
+ dst += 16;
+ }
+}
+
+void vpx_convolve8_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ int8_t cnt, filt_hor[8], filt_ver[8];
+
+ if (16 != x_step_q4 || 16 != y_step_q4) {+ vpx_convolve8_avg_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ if (((const int32_t *)filter_x)[1] == 0x800000 &&
+ ((const int32_t *)filter_y)[1] == 0x800000) {+ vpx_convolve_avg(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ for (cnt = 0; cnt < 8; ++cnt) {+ filt_hor[cnt] = filter_x[cnt];
+ filt_ver[cnt] = filter_y[cnt];
+ }
+
+ if (((const int32_t *)filter_x)[0] == 0 &&
+ ((const int32_t *)filter_y)[0] == 0) {+ switch (w) {+ case 4:
+ common_hv_2ht_2vt_and_aver_dst_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], h);
+ break;
+ case 8:
+ common_hv_2ht_2vt_and_aver_dst_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], h);
+ break;
+ case 16:
+ common_hv_2ht_2vt_and_aver_dst_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], h);
+ break;
+ case 32:
+ common_hv_2ht_2vt_and_aver_dst_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], h);
+ break;
+ case 64:
+ common_hv_2ht_2vt_and_aver_dst_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], h);
+ break;
+ default:
+ vpx_convolve8_avg_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else if (((const int32_t *)filter_x)[0] == 0 ||
+ ((const int32_t *)filter_y)[0] == 0) {+ vpx_convolve8_avg_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ } else {+ switch (w) {+ case 4:
+ common_hv_8ht_8vt_and_aver_dst_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, h);
+ break;
+ case 8:
+ common_hv_8ht_8vt_and_aver_dst_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, h);
+ break;
+ case 16:
+ common_hv_8ht_8vt_and_aver_dst_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, h);
+ break;
+ case 32:
+ common_hv_8ht_8vt_and_aver_dst_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, h);
+ break;
+ case 64:
+ common_hv_8ht_8vt_and_aver_dst_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, h);
+ break;
+ default:
+ vpx_convolve8_avg_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve8_avg_vert_msa.c
@@ -1,0 +1,753 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/mips/vpx_convolve_msa.h"
+
+static void common_vt_8t_and_aver_dst_4w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16u8 dst0, dst1, dst2, dst3, out;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776;
+ v16i8 src10998, filt0, filt1, filt2, filt3;
+ v8i16 filt, out10, out32;
+
+ src -= (3 * src_stride);
+
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+
+ ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
+ src54_r, src21_r);
+ ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+ ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src2110,
+ src4332, src6554);
+ XORI_B3_128_SB(src2110, src4332, src6554);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
+ src87_r, src98_r, src109_r);
+ ILVR_D2_SB(src87_r, src76_r, src109_r, src98_r, src8776, src10998);
+ XORI_B2_128_SB(src8776, src10998);
+ out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, filt0,
+ filt1, filt2, filt3);
+ out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, filt0,
+ filt1, filt2, filt3);
+ SRARI_H2_SH(out10, out32, FILTER_BITS);
+ SAT_SH2_SH(out10, out32, 7);
+ out = PCKEV_XORI128_UB(out10, out32);
+ ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2);
+
+ dst0 = (v16u8)__msa_ilvr_d((v2i64)dst2, (v2i64)dst0);
+ out = __msa_aver_u_b(out, dst0);
+
+ ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ src2110 = src6554;
+ src4332 = src8776;
+ src6554 = src10998;
+ src6 = src10;
+ }
+}
+
+static void common_vt_8t_and_aver_dst_8w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16u8 dst0, dst1, dst2, dst3;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src109_r, filt0, filt1, filt2, filt3;
+ v8i16 filt, out0, out1, out2, out3;
+
+ src -= (3 * src_stride);
+
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
+ src54_r, src21_r);
+ ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ XORI_B4_128_SB(src7, src8, src9, src10);
+ ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
+ src87_r, src98_r, src109_r);
+ out0 = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
+ filt1, filt2, filt3);
+ out1 = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
+ filt1, filt2, filt3);
+ out2 = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
+ filt1, filt2, filt3);
+ out3 = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
+ filt1, filt2, filt3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ CONVERT_UB_AVG_ST8x4_UB(out0, out1, out2, out3, dst0, dst1, dst2, dst3,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ src10_r = src54_r;
+ src32_r = src76_r;
+ src54_r = src98_r;
+ src21_r = src65_r;
+ src43_r = src87_r;
+ src65_r = src109_r;
+ src6 = src10;
+ }
+}
+
+static void common_vt_8t_and_aver_dst_16w_mult_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height,
+ int32_t width) {+ const uint8_t *src_tmp;
+ uint8_t *dst_tmp;
+ uint32_t loop_cnt, cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l;
+ v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l;
+ v16i8 filt0, filt1, filt2, filt3;
+ v16u8 dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3;
+ v8i16 out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l, filt;
+
+ src -= (3 * src_stride);
+
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ for (cnt = (width >> 4); cnt--;) {+ src_tmp = src;
+ dst_tmp = dst;
+
+ LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ src_tmp += (7 * src_stride);
+
+ ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
+ src54_r, src21_r);
+ ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+ ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l, src32_l,
+ src54_l, src21_l);
+ ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src_tmp, src_stride, src7, src8, src9, src10);
+ src_tmp += (4 * src_stride);
+
+ LD_UB4(dst_tmp, dst_stride, dst0, dst1, dst2, dst3);
+ XORI_B4_128_SB(src7, src8, src9, src10);
+ ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
+ src87_r, src98_r, src109_r);
+ ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l,
+ src87_l, src98_l, src109_l);
+ out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
+ filt1, filt2, filt3);
+ out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
+ filt1, filt2, filt3);
+ out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
+ filt1, filt2, filt3);
+ out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
+ filt1, filt2, filt3);
+ out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0,
+ filt1, filt2, filt3);
+ out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0,
+ filt1, filt2, filt3);
+ out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0,
+ filt1, filt2, filt3);
+ out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0,
+ filt1, filt2, filt3);
+ SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
+ SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS);
+ SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
+ SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7);
+ PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l,
+ out3_r, tmp0, tmp1, tmp2, tmp3);
+ XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3);
+ AVER_UB4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, dst0, dst1,
+ dst2, dst3);
+ ST_UB4(dst0, dst1, dst2, dst3, dst_tmp, dst_stride);
+ dst_tmp += (4 * dst_stride);
+
+ src10_r = src54_r;
+ src32_r = src76_r;
+ src54_r = src98_r;
+ src21_r = src65_r;
+ src43_r = src87_r;
+ src65_r = src109_r;
+ src10_l = src54_l;
+ src32_l = src76_l;
+ src54_l = src98_l;
+ src21_l = src65_l;
+ src43_l = src87_l;
+ src65_l = src109_l;
+ src6 = src10;
+ }
+
+ src += 16;
+ dst += 16;
+ }
+}
+
+static void common_vt_8t_and_aver_dst_16w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
+ filter, height, 16);
+}
+
+static void common_vt_8t_and_aver_dst_32w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
+ filter, height, 32);
+}
+
+static void common_vt_8t_and_aver_dst_64w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride,
+ filter, height, 64);
+}
+
+static void common_vt_2t_and_aver_dst_4x4_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, src4;
+ v16u8 dst0, dst1, dst2, dst3, out, filt0, src2110, src4332;
+ v16i8 src10_r, src32_r, src21_r, src43_r;
+ v8i16 filt;
+ v8u16 tmp0, tmp1;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ src4 = LD_SB(src);
+ src += src_stride;
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
+ dst0 = (v16u8)__msa_ilvr_d((v2i64)dst1, (v2i64)dst0);
+ ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
+ src32_r, src43_r);
+ ILVR_D2_UB(src21_r, src10_r, src43_r, src32_r, src2110, src4332);
+ DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+
+ out = (v16u8)__msa_pckev_b((v16i8)tmp1, (v16i8)tmp0);
+ out = __msa_aver_u_b(out, dst0);
+
+ ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
+}
+
+static void common_vt_2t_and_aver_dst_4x8_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter) {+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src87_r;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r, src65_r;
+ v16u8 src2110, src4332, src6554, src8776, filt0;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+ src8 = LD_SB(src);
+
+ LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+ ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst1,
+ dst2, dst3);
+ ILVR_D2_UB(dst1, dst0, dst3, dst2, dst0, dst1);
+ ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
+ src32_r, src43_r);
+ ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r,
+ src76_r, src87_r);
+ ILVR_D4_UB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r,
+ src87_r, src76_r, src2110, src4332, src6554, src8776);
+ DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0,
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, src2110, src4332);
+ AVER_UB2_UB(src2110, dst0, src4332, dst1, src2110, src4332);
+ ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst, dst_stride);
+}
+
+static void common_vt_2t_and_aver_dst_4w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ if (4 == height) {+ common_vt_2t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else if (8 == height) {+ common_vt_2t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter);
+ }
+}
+
+static void common_vt_2t_and_aver_dst_8x4_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter) {+ v16u8 src0, src1, src2, src3, src4;
+ v16u8 dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3, filt0;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ /* rearranging filter_y */
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1);
+ ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
+ tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3,
+ dst, dst_stride);
+}
+
+static void common_vt_2t_and_aver_dst_8x8mult_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+ v16u8 dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ /* rearranging filter_y */
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_UB(src);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 3); loop_cnt--;) {+ LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8);
+ src += (8 * src_stride);
+ LD_UB8(dst, dst_stride, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8);
+
+ ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3, vec0, vec1,
+ vec2, vec3);
+ ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7, vec4, vec5,
+ vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
+ tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_AVG_ST8x4_UB(tmp0, dst1, tmp1, dst2, tmp2, dst3, tmp3, dst4,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, tmp0, tmp1,
+ tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_AVG_ST8x4_UB(tmp0, dst5, tmp1, dst6, tmp2, dst7, tmp3, dst8,
+ dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ src0 = src8;
+ }
+}
+
+static void common_vt_2t_and_aver_dst_8w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ if (4 == height) {+ common_vt_2t_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else {+ common_vt_2t_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride,
+ filter, height);
+ }
+}
+
+static void common_vt_2t_and_aver_dst_16w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4, dst0, dst1, dst2, dst3, filt0;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 tmp0, tmp1, tmp2, tmp3, filt;
+
+ /* rearranging filter_y */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ src0 = LD_UB(src);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_UB4(src, src_stride, src1, src2, src3, src4);
+ src += (4 * src_stride);
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
+ ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
+ dst += dst_stride;
+
+ ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
+ ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst);
+ dst += dst_stride;
+
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst);
+ dst += dst_stride;
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst);
+ dst += dst_stride;
+
+ src0 = src4;
+ }
+}
+
+static void common_vt_2t_and_aver_dst_32w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
+ v8u16 tmp0, tmp1, tmp2, tmp3, filt;
+
+ /* rearranging filter_y */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_UB2(src, 16, src0, src5);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_UB4(src, src_stride, src1, src2, src3, src4);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
+ ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
+
+ LD_UB4(src + 16, src_stride, src6, src7, src8, src9);
+ LD_UB4(dst + 16, dst_stride, dst4, dst5, dst6, dst7);
+ src += (4 * src_stride);
+
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
+
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride);
+
+ ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
+ ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst + 2 * dst_stride);
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst + 3 * dst_stride);
+
+ ILVR_B2_UB(src6, src5, src7, src6, vec0, vec2);
+ ILVL_B2_UB(src6, src5, src7, src6, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 16);
+
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 16 + dst_stride);
+
+ ILVR_B2_UB(src8, src7, src9, src8, vec4, vec6);
+ ILVL_B2_UB(src8, src7, src9, src8, vec5, vec7);
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst6, dst + 16 + 2 * dst_stride);
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_AVG_ST_UB(tmp3, tmp2, dst7, dst + 16 + 3 * dst_stride);
+ dst += (4 * dst_stride);
+
+ src0 = src4;
+ src5 = src9;
+ }
+}
+
+static void common_vt_2t_and_aver_dst_64w_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4, src5;
+ v16u8 src6, src7, src8, src9, src10, src11, filt0;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+ v8u16 filt;
+
+ /* rearranging filter_y */
+ filt = LD_UH(filter);
+ filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_UB4(src, 16, src0, src3, src6, src9);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 1); loop_cnt--;) {+ LD_UB2(src, src_stride, src1, src2);
+ LD_UB2(dst, dst_stride, dst0, dst1);
+ LD_UB2(src + 16, src_stride, src4, src5);
+ LD_UB2(dst + 16, dst_stride, dst2, dst3);
+ LD_UB2(src + 32, src_stride, src7, src8);
+ LD_UB2(dst + 32, dst_stride, dst4, dst5);
+ LD_UB2(src + 48, src_stride, src10, src11);
+ LD_UB2(dst + 48, dst_stride, dst6, dst7);
+ src += (2 * src_stride);
+
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
+ ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst);
+
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride);
+
+ ILVR_B2_UB(src4, src3, src5, src4, vec4, vec6);
+ ILVL_B2_UB(src4, src3, src5, src4, vec5, vec7);
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
+ SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
+ SAT_UH2_UH(tmp4, tmp5, 7);
+ PCKEV_AVG_ST_UB(tmp5, tmp4, dst2, dst + 16);
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
+ SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
+ SAT_UH2_UH(tmp6, tmp7, 7);
+ PCKEV_AVG_ST_UB(tmp7, tmp6, dst3, dst + 16 + dst_stride);
+
+ ILVR_B2_UB(src7, src6, src8, src7, vec0, vec2);
+ ILVL_B2_UB(src7, src6, src8, src7, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 32);
+
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 32 + dst_stride);
+
+ ILVR_B2_UB(src10, src9, src11, src10, vec4, vec6);
+ ILVL_B2_UB(src10, src9, src11, src10, vec5, vec7);
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
+ SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
+ SAT_UH2_UH(tmp4, tmp5, 7);
+ PCKEV_AVG_ST_UB(tmp5, tmp4, dst6, (dst + 48));
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
+ SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
+ SAT_UH2_UH(tmp6, tmp7, 7);
+ PCKEV_AVG_ST_UB(tmp7, tmp6, dst7, dst + 48 + dst_stride);
+ dst += (2 * dst_stride);
+
+ src0 = src2;
+ src3 = src5;
+ src6 = src8;
+ src9 = src11;
+ }
+}
+
+void vpx_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ int8_t cnt, filt_ver[8];
+
+ if (16 != y_step_q4) {+ vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ if (((const int32_t *)filter_y)[1] == 0x800000) {+ vpx_convolve_avg(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ for (cnt = 0; cnt < 8; ++cnt) {+ filt_ver[cnt] = filter_y[cnt];
+ }
+
+ if (((const int32_t *)filter_y)[0] == 0) {+ switch (w) {+ case 4:
+ common_vt_2t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ case 8:
+ common_vt_2t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ case 16:
+ common_vt_2t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ case 32:
+ common_vt_2t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ case 64:
+ common_vt_2t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ default:
+ vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {+ switch (w) {+ case 4:
+ common_vt_8t_and_aver_dst_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ case 8:
+ common_vt_8t_and_aver_dst_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ case 16:
+ common_vt_8t_and_aver_dst_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+
+ break;
+ case 32:
+ common_vt_8t_and_aver_dst_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ case 64:
+ common_vt_8t_and_aver_dst_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ default:
+ vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve8_horiz_msa.c
@@ -1,0 +1,742 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/mips/vpx_convolve_msa.h"
+
+static void common_hz_8t_4x4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16u8 mask0, mask1, mask2, mask3, out;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v8i16 filt, out0, out1;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[16]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
+ filt0, filt1, filt2, filt3, out0, out1);
+ SRARI_H2_SH(out0, out1, FILTER_BITS);
+ SAT_SH2_SH(out0, out1, 7);
+ out = PCKEV_XORI128_UB(out0, out1);
+ ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
+}
+
+static void common_hz_8t_4x8_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16i8 filt0, filt1, filt2, filt3;
+ v16i8 src0, src1, src2, src3;
+ v16u8 mask0, mask1, mask2, mask3, out;
+ v8i16 filt, out0, out1, out2, out3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[16]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ src += (4 * src_stride);
+ HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
+ filt0, filt1, filt2, filt3, out0, out1);
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
+ filt0, filt1, filt2, filt3, out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ out = PCKEV_XORI128_UB(out0, out1);
+ ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ out = PCKEV_XORI128_UB(out2, out3);
+ ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
+}
+
+static void common_hz_8t_4w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ if (4 == height) {+ common_hz_8t_4x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else if (8 == height) {+ common_hz_8t_4x8_msa(src, src_stride, dst, dst_stride, filter);
+ }
+}
+
+static void common_hz_8t_8x4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 mask0, mask1, mask2, mask3, tmp0, tmp1;
+ v8i16 filt, out0, out1, out2, out3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3,
+ filt0, filt1, filt2, filt3, out0, out1, out2,
+ out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ tmp0 = PCKEV_XORI128_UB(out0, out1);
+ tmp1 = PCKEV_XORI128_UB(out2, out3);
+ ST8x4_UB(tmp0, tmp1, dst, dst_stride);
+}
+
+static void common_hz_8t_8x8mult_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 mask0, mask1, mask2, mask3, tmp0, tmp1;
+ v8i16 filt, out0, out1, out2, out3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ src += (4 * src_stride);
+ HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
+ mask3, filt0, filt1, filt2, filt3, out0, out1,
+ out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ tmp0 = PCKEV_XORI128_UB(out0, out1);
+ tmp1 = PCKEV_XORI128_UB(out2, out3);
+ ST8x4_UB(tmp0, tmp1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void common_hz_8t_8w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ if (4 == height) {+ common_hz_8t_8x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else {+ common_hz_8t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height);
+ }
+}
+
+static void common_hz_8t_16w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 mask0, mask1, mask2, mask3, out;
+ v8i16 filt, out0, out1, out2, out3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ for (loop_cnt = (height >> 1); loop_cnt--;) {+ LD_SB2(src, src_stride, src0, src2);
+ LD_SB2(src + 8, src_stride, src1, src3);
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ src += (2 * src_stride);
+ HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
+ mask3, filt0, filt1, filt2, filt3, out0, out1,
+ out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ out = PCKEV_XORI128_UB(out0, out1);
+ ST_UB(out, dst);
+ dst += dst_stride;
+ out = PCKEV_XORI128_UB(out2, out3);
+ ST_UB(out, dst);
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_8t_32w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 mask0, mask1, mask2, mask3, out;
+ v8i16 filt, out0, out1, out2, out3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ for (loop_cnt = (height >> 1); loop_cnt--;) {+ src0 = LD_SB(src);
+ src2 = LD_SB(src + 16);
+ src3 = LD_SB(src + 24);
+ src1 = __msa_sldi_b(src2, src0, 8);
+ src += src_stride;
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
+ mask3, filt0, filt1, filt2, filt3, out0, out1,
+ out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+
+ src0 = LD_SB(src);
+ src2 = LD_SB(src + 16);
+ src3 = LD_SB(src + 24);
+ src1 = __msa_sldi_b(src2, src0, 8);
+ src += src_stride;
+
+ out = PCKEV_XORI128_UB(out0, out1);
+ ST_UB(out, dst);
+ out = PCKEV_XORI128_UB(out2, out3);
+ ST_UB(out, dst + 16);
+ dst += dst_stride;
+
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
+ mask3, filt0, filt1, filt2, filt3, out0, out1,
+ out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ out = PCKEV_XORI128_UB(out0, out1);
+ ST_UB(out, dst);
+ out = PCKEV_XORI128_UB(out2, out3);
+ ST_UB(out, dst + 16);
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_8t_64w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ int32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, filt0, filt1, filt2, filt3;
+ v16u8 mask0, mask1, mask2, mask3, out;
+ v8i16 filt, out0, out1, out2, out3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= 3;
+
+ /* rearranging filter */
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ for (loop_cnt = height; loop_cnt--;) {+ src0 = LD_SB(src);
+ src2 = LD_SB(src + 16);
+ src3 = LD_SB(src + 24);
+ src1 = __msa_sldi_b(src2, src0, 8);
+
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
+ mask3, filt0, filt1, filt2, filt3, out0, out1,
+ out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ out = PCKEV_XORI128_UB(out0, out1);
+ ST_UB(out, dst);
+ out = PCKEV_XORI128_UB(out2, out3);
+ ST_UB(out, dst + 16);
+
+ src0 = LD_SB(src + 32);
+ src2 = LD_SB(src + 48);
+ src3 = LD_SB(src + 56);
+ src1 = __msa_sldi_b(src2, src0, 8);
+ src += src_stride;
+
+ XORI_B4_128_SB(src0, src1, src2, src3);
+ HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2,
+ mask3, filt0, filt1, filt2, filt3, out0, out1,
+ out2, out3);
+ SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS);
+ SAT_SH4_SH(out0, out1, out2, out3, 7);
+ out = PCKEV_XORI128_UB(out0, out1);
+ ST_UB(out, dst + 32);
+ out = PCKEV_XORI128_UB(out2, out3);
+ ST_UB(out, dst + 48);
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_2t_4x4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, mask;
+ v16u8 filt0, vec0, vec1, res0, res1;
+ v8u16 vec2, vec3, filt, const255;
+
+ mask = LD_SB(&mc_filt_mask_arr[16]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
+
+ const255 = (v8u16) __msa_ldi_h(255);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, vec2, vec3);
+ SRARI_H2_UH(vec2, vec3, FILTER_BITS);
+ MIN_UH2_UH(vec2, vec3, const255);
+ PCKEV_B2_UB(vec2, vec2, vec3, vec3, res0, res1);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hz_2t_4x8_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16u8 vec0, vec1, vec2, vec3, filt0;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16i8 res0, res1, res2, res3;
+ v8u16 vec4, vec5, vec6, vec7, filt, const255;
+
+ mask = LD_SB(&mc_filt_mask_arr[16]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
+
+ const255 = (v8u16) __msa_ldi_h(255);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ VSHF_B2_UB(src0, src1, src2, src3, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src4, src5, src6, src7, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec4, vec5,
+ vec6, vec7);
+ SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS);
+ MIN_UH4_UH(vec4, vec5, vec6, vec7, const255);
+ PCKEV_B4_SB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1,
+ res2, res3);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hz_2t_4w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ if (4 == height) {+ common_hz_2t_4x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else if (8 == height) {+ common_hz_2t_4x8_msa(src, src_stride, dst, dst_stride, filter);
+ }
+}
+
+static void common_hz_2t_8x4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16u8 filt0;
+ v16i8 src0, src1, src2, src3, mask;
+ v8u16 vec0, vec1, vec2, vec3, const255, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
+
+ const255 = (v8u16) __msa_ldi_h(255);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, src0, src1);
+ ST8x4_UB(src0, src1, dst, dst_stride);
+}
+
+static void common_hz_2t_8x8mult_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ v16u8 filt0;
+ v16i8 src0, src1, src2, src3, mask, out0, out1;
+ v8u16 vec0, vec1, vec2, vec3, filt, const255;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
+
+ const255 = (v8u16) __msa_ldi_h(255);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ if (16 == height) {+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ LD_SB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+
+ VSHF_B2_UH(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UH(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, vec0, vec1,
+ vec2, vec3);
+ SRARI_H4_UH(vec0, vec1, vec2, vec3, FILTER_BITS);
+ MIN_UH4_UH(vec0, vec1, vec2, vec3, const255);
+ PCKEV_B2_SB(vec1, vec0, vec3, vec2, out0, out1);
+ ST8x4_UB(out0, out1, dst + 4 * dst_stride, dst_stride);
+ }
+}
+
+static void common_hz_2t_8w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ if (4 == height) {+ common_hz_2t_8x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else {+ common_hz_2t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height);
+ }
+}
+
+static void common_hz_2t_16w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ loop_cnt = (height >> 2) - 1;
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
+
+ const255 = (v8u16) __msa_ldi_h(255);
+
+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
+ out2, out3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
+ out6, out7);
+ SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
+ SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
+ MIN_UH4_UH(out0, out1, out2, out3, const255);
+ MIN_UH4_UH(out4, out5, out6, out7, const255);
+ PCKEV_ST_SB(out0, out1, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out2, out3, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out4, out5, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out6, out7, dst);
+ dst += dst_stride;
+
+ for (; loop_cnt--;) {+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
+ out2, out3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
+ out6, out7);
+ SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
+ SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
+ MIN_UH4_UH(out0, out1, out2, out3, const255);
+ MIN_UH4_UH(out4, out5, out6, out7, const255);
+ PCKEV_ST_SB(out0, out1, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out2, out3, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out4, out5, dst);
+ dst += dst_stride;
+ PCKEV_ST_SB(out6, out7, dst);
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_2t_32w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
+
+ const255 = (v8u16) __msa_ldi_h(255);
+
+ for (loop_cnt = height >> 1; loop_cnt--;) {+ src0 = LD_SB(src);
+ src2 = LD_SB(src + 16);
+ src3 = LD_SB(src + 24);
+ src1 = __msa_sldi_b(src2, src0, 8);
+ src += src_stride;
+ src4 = LD_SB(src);
+ src6 = LD_SB(src + 16);
+ src7 = LD_SB(src + 24);
+ src5 = __msa_sldi_b(src6, src4, 8);
+ src += src_stride;
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
+ out2, out3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
+ out6, out7);
+ SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
+ SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
+ MIN_UH4_UH(out0, out1, out2, out3, const255);
+ MIN_UH4_UH(out4, out5, out6, out7, const255);
+ PCKEV_ST_SB(out0, out1, dst);
+ PCKEV_ST_SB(out2, out3, dst + 16);
+ dst += dst_stride;
+ PCKEV_ST_SB(out4, out5, dst);
+ PCKEV_ST_SB(out6, out7, dst + 16);
+ dst += dst_stride;
+ }
+}
+
+static void common_hz_2t_64w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt0, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8u16 out0, out1, out2, out3, out4, out5, out6, out7, filt, const255;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter);
+ filt0 = (v16u8) __msa_splati_h((v8i16) filt, 0);
+
+ const255 = (v8u16) __msa_ldi_h(255);
+
+ for (loop_cnt = height; loop_cnt--;) {+ src0 = LD_SB(src);
+ src2 = LD_SB(src + 16);
+ src4 = LD_SB(src + 32);
+ src6 = LD_SB(src + 48);
+ src7 = LD_SB(src + 56);
+ SLDI_B3_SB(src2, src4, src6, src0, src2, src4, src1, src3, src5, 8);
+ src += src_stride;
+
+ VSHF_B2_UB(src0, src0, src1, src1, mask, mask, vec0, vec1);
+ VSHF_B2_UB(src2, src2, src3, src3, mask, mask, vec2, vec3);
+ VSHF_B2_UB(src4, src4, src5, src5, mask, mask, vec4, vec5);
+ VSHF_B2_UB(src6, src6, src7, src7, mask, mask, vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, out0, out1,
+ out2, out3);
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, out4, out5,
+ out6, out7);
+ SRARI_H4_UH(out0, out1, out2, out3, FILTER_BITS);
+ SRARI_H4_UH(out4, out5, out6, out7, FILTER_BITS);
+ MIN_UH4_UH(out0, out1, out2, out3, const255);
+ MIN_UH4_UH(out4, out5, out6, out7, const255);
+ PCKEV_ST_SB(out0, out1, dst);
+ PCKEV_ST_SB(out2, out3, dst + 16);
+ PCKEV_ST_SB(out4, out5, dst + 32);
+ PCKEV_ST_SB(out6, out7, dst + 48);
+ dst += dst_stride;
+ }
+}
+
+void vpx_convolve8_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ int8_t cnt, filt_hor[8];
+
+ if (16 != x_step_q4) {+ vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ if (((const int32_t *)filter_x)[1] == 0x800000) {+ vpx_convolve_copy(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ for (cnt = 0; cnt < 8; ++cnt) {+ filt_hor[cnt] = filter_x[cnt];
+ }
+
+ if (((const int32_t *)filter_x)[0] == 0) {+ switch (w) {+ case 4:
+ common_hz_2t_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ case 8:
+ common_hz_2t_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ case 16:
+ common_hz_2t_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ case 32:
+ common_hz_2t_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ case 64:
+ common_hz_2t_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], h);
+ break;
+ default:
+ vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {+ switch (w) {+ case 4:
+ common_hz_8t_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ case 8:
+ common_hz_8t_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ case 16:
+ common_hz_8t_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ case 32:
+ common_hz_8t_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ case 64:
+ common_hz_8t_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, h);
+ break;
+ default:
+ vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve8_msa.c
@@ -1,0 +1,654 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/mips/vpx_convolve_msa.h"
+
+const uint8_t mc_filt_mask_arr[16 * 3] = {+ /* 8 width cases */
+ 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
+ /* 4 width cases */
+ 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20,
+ /* 4 width cases */
+ 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28
+};
+
+static void common_hv_8ht_8vt_4w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
+ v16u8 mask0, mask1, mask2, mask3, out;
+ v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
+ v8i16 hz_out7, hz_out8, hz_out9, tmp0, tmp1, out0, out1, out2, out3, out4;
+ v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[16]);
+ src -= (3 + 3 * src_stride);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+
+ hz_out0 = HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out2 = HORIZ_8TAP_FILT(src2, src3, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out4 = HORIZ_8TAP_FILT(src4, src5, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out5 = HORIZ_8TAP_FILT(src5, src6, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ SLDI_B2_SH(hz_out2, hz_out4, hz_out0, hz_out2, hz_out1, hz_out3, 8);
+
+ filt = LD_SH(filter_vert);
+ SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
+
+ ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1);
+ out2 = (v8i16)__msa_ilvev_b((v16i8)hz_out5, (v16i8)hz_out4);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ XORI_B4_128_SB(src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ hz_out7 = HORIZ_8TAP_FILT(src7, src8, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ hz_out6 = (v8i16)__msa_sldi_b((v16i8)hz_out7, (v16i8)hz_out5, 8);
+ out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
+ tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ hz_out9 = HORIZ_8TAP_FILT(src9, src10, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ hz_out8 = (v8i16)__msa_sldi_b((v16i8)hz_out9, (v16i8)hz_out7, 8);
+ out4 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
+ tmp1 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out4, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+ SRARI_H2_SH(tmp0, tmp1, FILTER_BITS);
+ SAT_SH2_SH(tmp0, tmp1, 7);
+ out = PCKEV_XORI128_UB(tmp0, tmp1);
+ ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ hz_out5 = hz_out9;
+ out0 = out2;
+ out1 = out3;
+ out2 = out4;
+ }
+}
+
+static void common_hv_8ht_8vt_8w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 filt_hz0, filt_hz1, filt_hz2, filt_hz3;
+ v16u8 mask0, mask1, mask2, mask3, vec0, vec1;
+ v8i16 filt, filt_vt0, filt_vt1, filt_vt2, filt_vt3;
+ v8i16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
+ v8i16 hz_out7, hz_out8, hz_out9, hz_out10, tmp0, tmp1, tmp2, tmp3;
+ v8i16 out0, out1, out2, out3, out4, out5, out6, out7, out8, out9;
+
+ mask0 = LD_UB(&mc_filt_mask_arr[0]);
+ src -= (3 + 3 * src_stride);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+
+ mask1 = mask0 + 2;
+ mask2 = mask0 + 4;
+ mask3 = mask0 + 6;
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ hz_out0 = HORIZ_8TAP_FILT(src0, src0, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out1 = HORIZ_8TAP_FILT(src1, src1, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out2 = HORIZ_8TAP_FILT(src2, src2, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out3 = HORIZ_8TAP_FILT(src3, src3, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out4 = HORIZ_8TAP_FILT(src4, src4, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out5 = HORIZ_8TAP_FILT(src5, src5, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+ hz_out6 = HORIZ_8TAP_FILT(src6, src6, mask0, mask1, mask2, mask3, filt_hz0,
+ filt_hz1, filt_hz2, filt_hz3);
+
+ filt = LD_SH(filter_vert);
+ SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
+
+ ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1);
+ ILVEV_B2_SH(hz_out4, hz_out5, hz_out1, hz_out2, out2, out4);
+ ILVEV_B2_SH(hz_out3, hz_out4, hz_out5, hz_out6, out5, out6);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ XORI_B4_128_SB(src7, src8, src9, src10);
+
+ hz_out7 = HORIZ_8TAP_FILT(src7, src7, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
+ tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ hz_out8 = HORIZ_8TAP_FILT(src8, src8, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ out7 = (v8i16)__msa_ilvev_b((v16i8)hz_out8, (v16i8)hz_out7);
+ tmp1 = FILT_8TAP_DPADD_S_H(out4, out5, out6, out7, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ hz_out9 = HORIZ_8TAP_FILT(src9, src9, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ out8 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
+ tmp2 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out8, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+
+ hz_out10 = HORIZ_8TAP_FILT(src10, src10, mask0, mask1, mask2, mask3,
+ filt_hz0, filt_hz1, filt_hz2, filt_hz3);
+ out9 = (v8i16)__msa_ilvev_b((v16i8)hz_out10, (v16i8)hz_out9);
+ tmp3 = FILT_8TAP_DPADD_S_H(out5, out6, out7, out9, filt_vt0, filt_vt1,
+ filt_vt2, filt_vt3);
+ SRARI_H4_SH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_SH4_SH(tmp0, tmp1, tmp2, tmp3, 7);
+ vec0 = PCKEV_XORI128_UB(tmp0, tmp1);
+ vec1 = PCKEV_XORI128_UB(tmp2, tmp3);
+ ST8x4_UB(vec0, vec1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ hz_out6 = hz_out10;
+ out0 = out2;
+ out1 = out3;
+ out2 = out8;
+ out4 = out6;
+ out5 = out7;
+ out6 = out9;
+ }
+}
+
+static void common_hv_8ht_8vt_16w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 2; multiple8_cnt--;) {+ common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz,
+ filter_vert, height);
+ src += 8;
+ dst += 8;
+ }
+}
+
+static void common_hv_8ht_8vt_32w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 4; multiple8_cnt--;) {+ common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz,
+ filter_vert, height);
+ src += 8;
+ dst += 8;
+ }
+}
+
+static void common_hv_8ht_8vt_64w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 8; multiple8_cnt--;) {+ common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz,
+ filter_vert, height);
+ src += 8;
+ dst += 8;
+ }
+}
+
+static void common_hv_2ht_2vt_4x4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert) {+ v16i8 src0, src1, src2, src3, src4, mask;
+ v16u8 filt_vt, filt_hz, vec0, vec1, res0, res1;
+ v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, filt, tmp0, tmp1;
+
+ mask = LD_SB(&mc_filt_mask_arr[16]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ filt = LD_UH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
+ hz_out4 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ hz_out1 = (v8u16)__msa_sldi_b((v16i8)hz_out2, (v16i8)hz_out0, 8);
+ hz_out3 = (v8u16)__msa_pckod_d((v2i64)hz_out4, (v2i64)hz_out2);
+
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_B2_UB(tmp0, tmp0, tmp1, tmp1, res0, res1);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_4x8_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert) {+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, mask;
+ v16i8 res0, res1, res2, res3;
+ v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3;
+ v8u16 hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6;
+ v8u16 hz_out7, hz_out8, vec4, vec5, vec6, vec7, filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[16]);
+
+ /* rearranging filter */
+ filt = LD_UH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ filt = LD_UH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h((v8i16)filt, 0);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+ src8 = LD_SB(src);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src1, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src2, src3, mask, filt_hz, FILTER_BITS);
+ hz_out4 = HORIZ_2TAP_FILT_UH(src4, src5, mask, filt_hz, FILTER_BITS);
+ hz_out6 = HORIZ_2TAP_FILT_UH(src6, src7, mask, filt_hz, FILTER_BITS);
+ hz_out8 = HORIZ_2TAP_FILT_UH(src8, src8, mask, filt_hz, FILTER_BITS);
+ SLDI_B3_UH(hz_out2, hz_out4, hz_out6, hz_out0, hz_out2, hz_out4, hz_out1,
+ hz_out3, hz_out5, 8);
+ hz_out7 = (v8u16)__msa_pckod_d((v2i64)hz_out8, (v2i64)hz_out6);
+
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ ILVEV_B2_UB(hz_out4, hz_out5, hz_out6, hz_out7, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt_vt, filt_vt, filt_vt, filt_vt,
+ vec4, vec5, vec6, vec7);
+ SRARI_H4_UH(vec4, vec5, vec6, vec7, FILTER_BITS);
+ SAT_UH4_UH(vec4, vec5, vec6, vec7, 7);
+ PCKEV_B4_SB(vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7, res0, res1,
+ res2, res3);
+ ST4x4_UB(res0, res1, 0, 1, 0, 1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ ST4x4_UB(res2, res3, 0, 1, 0, 1, dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_4w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ if (4 == height) {+ common_hv_2ht_2vt_4x4_msa(src, src_stride, dst, dst_stride, filter_horiz,
+ filter_vert);
+ } else if (8 == height) {+ common_hv_2ht_2vt_4x8_msa(src, src_stride, dst, dst_stride, filter_horiz,
+ filter_vert);
+ }
+}
+
+static void common_hv_2ht_2vt_8x4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert) {+ v16i8 src0, src1, src2, src3, src4, mask, out0, out1;
+ v16u8 filt_hz, filt_vt, vec0, vec1, vec2, vec3;
+ v8u16 hz_out0, hz_out1, tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
+ hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp0 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
+ vec1 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp1 = __msa_dotp_u_h(vec1, filt_vt);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
+ vec2 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp2 = __msa_dotp_u_h(vec2, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ vec3 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp3 = __msa_dotp_u_h(vec3, filt_vt);
+
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+}
+
+static void common_hv_2ht_2vt_8x8mult_msa(const uint8_t *src,
+ int32_t src_stride,
+ uint8_t *dst,
+ int32_t dst_stride,
+ int8_t *filter_horiz,
+ int8_t *filter_vert,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, mask, out0, out1;
+ v16u8 filt_hz, filt_vt, vec0;
+ v8u16 hz_out0, hz_out1, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
+ v8i16 filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_SB(src);
+ src += src_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
+
+ for (loop_cnt = (height >> 3); loop_cnt--;) {+ LD_SB4(src, src_stride, src1, src2, src3, src4);
+ src += (4 * src_stride);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp1 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp2 = __msa_dotp_u_h(vec0, filt_vt);
+
+ SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
+ SAT_UH2_UH(tmp1, tmp2, 7);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp3 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ LD_SB4(src, src_stride, src1, src2, src3, src4);
+ src += (4 * src_stride);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp4 = __msa_dotp_u_h(vec0, filt_vt);
+
+ SRARI_H2_UH(tmp3, tmp4, FILTER_BITS);
+ SAT_UH2_UH(tmp3, tmp4, 7);
+ PCKEV_B2_SB(tmp2, tmp1, tmp4, tmp3, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp5 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp6 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out1, (v16i8)hz_out0);
+ tmp7 = __msa_dotp_u_h(vec0, filt_vt);
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ vec0 = (v16u8)__msa_ilvev_b((v16i8)hz_out0, (v16i8)hz_out1);
+ tmp8 = __msa_dotp_u_h(vec0, filt_vt);
+
+ SRARI_H4_UH(tmp5, tmp6, tmp7, tmp8, FILTER_BITS);
+ SAT_UH4_UH(tmp5, tmp6, tmp7, tmp8, 7);
+ PCKEV_B2_SB(tmp6, tmp5, tmp8, tmp7, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void common_hv_2ht_2vt_8w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ if (4 == height) {+ common_hv_2ht_2vt_8x4_msa(src, src_stride, dst, dst_stride, filter_horiz,
+ filter_vert);
+ } else {+ common_hv_2ht_2vt_8x8mult_msa(src, src_stride, dst, dst_stride,
+ filter_horiz, filter_vert, height);
+ }
+}
+
+static void common_hv_2ht_2vt_16w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask;
+ v16u8 filt_hz, filt_vt, vec0, vec1;
+ v8u16 tmp1, tmp2, hz_out0, hz_out1, hz_out2, hz_out3;
+ v8i16 filt;
+
+ mask = LD_SB(&mc_filt_mask_arr[0]);
+
+ /* rearranging filter */
+ filt = LD_SH(filter_horiz);
+ filt_hz = (v16u8)__msa_splati_h(filt, 0);
+
+ filt = LD_SH(filter_vert);
+ filt_vt = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB2(src, 8, src0, src1);
+ src += src_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src0, src2, src4, src6);
+ LD_SB4(src + 8, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src0, src0, mask, filt_hz, FILTER_BITS);
+ hz_out3 = HORIZ_2TAP_FILT_UH(src1, src1, mask, filt_hz, FILTER_BITS);
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
+ SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
+ SAT_UH2_UH(tmp1, tmp2, 7);
+ PCKEV_ST_SB(tmp1, tmp2, dst);
+ dst += dst_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src2, src2, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src3, src3, mask, filt_hz, FILTER_BITS);
+ ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
+ SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
+ SAT_UH2_UH(tmp1, tmp2, 7);
+ PCKEV_ST_SB(tmp1, tmp2, dst);
+ dst += dst_stride;
+
+ hz_out1 = HORIZ_2TAP_FILT_UH(src4, src4, mask, filt_hz, FILTER_BITS);
+ hz_out3 = HORIZ_2TAP_FILT_UH(src5, src5, mask, filt_hz, FILTER_BITS);
+ ILVEV_B2_UB(hz_out0, hz_out1, hz_out2, hz_out3, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
+ SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
+ SAT_UH2_UH(tmp1, tmp2, 7);
+ PCKEV_ST_SB(tmp1, tmp2, dst);
+ dst += dst_stride;
+
+ hz_out0 = HORIZ_2TAP_FILT_UH(src6, src6, mask, filt_hz, FILTER_BITS);
+ hz_out2 = HORIZ_2TAP_FILT_UH(src7, src7, mask, filt_hz, FILTER_BITS);
+ ILVEV_B2_UB(hz_out1, hz_out0, hz_out3, hz_out2, vec0, vec1);
+ DOTP_UB2_UH(vec0, vec1, filt_vt, filt_vt, tmp1, tmp2);
+ SRARI_H2_UH(tmp1, tmp2, FILTER_BITS);
+ SAT_UH2_UH(tmp1, tmp2, 7);
+ PCKEV_ST_SB(tmp1, tmp2, dst);
+ dst += dst_stride;
+ }
+}
+
+static void common_hv_2ht_2vt_32w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 2; multiple8_cnt--;) {+ common_hv_2ht_2vt_16w_msa(src, src_stride, dst, dst_stride, filter_horiz,
+ filter_vert, height);
+ src += 16;
+ dst += 16;
+ }
+}
+
+static void common_hv_2ht_2vt_64w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter_horiz, int8_t *filter_vert,
+ int32_t height) {+ int32_t multiple8_cnt;
+ for (multiple8_cnt = 4; multiple8_cnt--;) {+ common_hv_2ht_2vt_16w_msa(src, src_stride, dst, dst_stride, filter_horiz,
+ filter_vert, height);
+ src += 16;
+ dst += 16;
+ }
+}
+
+void vpx_convolve8_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int32_t x_step_q4,
+ const int16_t *filter_y, int32_t y_step_q4,
+ int32_t w, int32_t h) {+ int8_t cnt, filt_hor[8], filt_ver[8];
+
+ if (16 != x_step_q4 || 16 != y_step_q4) {+ vpx_convolve8_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ if (((const int32_t *)filter_x)[1] == 0x800000 &&
+ ((const int32_t *)filter_y)[1] == 0x800000) {+ vpx_convolve_copy(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ for (cnt = 0; cnt < 8; ++cnt) {+ filt_hor[cnt] = filter_x[cnt];
+ filt_ver[cnt] = filter_y[cnt];
+ }
+
+ if (((const int32_t *)filter_x)[0] == 0 &&
+ ((const int32_t *)filter_y)[0] == 0) {+ switch (w) {+ case 4:
+ common_hv_2ht_2vt_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], (int32_t)h);
+ break;
+ case 8:
+ common_hv_2ht_2vt_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], (int32_t)h);
+ break;
+ case 16:
+ common_hv_2ht_2vt_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], (int32_t)h);
+ break;
+ case 32:
+ common_hv_2ht_2vt_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], (int32_t)h);
+ break;
+ case 64:
+ common_hv_2ht_2vt_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_hor[3], &filt_ver[3], (int32_t)h);
+ break;
+ default:
+ vpx_convolve8_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else if (((const int32_t *)filter_x)[0] == 0 ||
+ ((const int32_t *)filter_y)[0] == 0) {+ vpx_convolve8_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ } else {+ switch (w) {+ case 4:
+ common_hv_8ht_8vt_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, (int32_t)h);
+ break;
+ case 8:
+ common_hv_8ht_8vt_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, (int32_t)h);
+ break;
+ case 16:
+ common_hv_8ht_8vt_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, (int32_t)h);
+ break;
+ case 32:
+ common_hv_8ht_8vt_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, (int32_t)h);
+ break;
+ case 64:
+ common_hv_8ht_8vt_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_hor, filt_ver, (int32_t)h);
+ break;
+ default:
+ vpx_convolve8_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve8_vert_msa.c
@@ -1,0 +1,745 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/mips/vpx_convolve_msa.h"
+
+static void common_vt_8t_4w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776;
+ v16i8 src10998, filt0, filt1, filt2, filt3;
+ v16u8 out;
+ v8i16 filt, out10, out32;
+
+ src -= (3 * src_stride);
+
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+
+ ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
+ src54_r, src21_r);
+ ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+ ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src2110,
+ src4332, src6554);
+ XORI_B3_128_SB(src2110, src4332, src6554);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
+ src87_r, src98_r, src109_r);
+ ILVR_D2_SB(src87_r, src76_r, src109_r, src98_r, src8776, src10998);
+ XORI_B2_128_SB(src8776, src10998);
+ out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, filt0,
+ filt1, filt2, filt3);
+ out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, filt0,
+ filt1, filt2, filt3);
+ SRARI_H2_SH(out10, out32, FILTER_BITS);
+ SAT_SH2_SH(out10, out32, 7);
+ out = PCKEV_XORI128_UB(out10, out32);
+ ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ src2110 = src6554;
+ src4332 = src8776;
+ src6554 = src10998;
+ src6 = src10;
+ }
+}
+
+static void common_vt_8t_8w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src109_r, filt0, filt1, filt2, filt3;
+ v16u8 tmp0, tmp1;
+ v8i16 filt, out0_r, out1_r, out2_r, out3_r;
+
+ src -= (3 * src_stride);
+
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+ ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
+ src54_r, src21_r);
+ ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ XORI_B4_128_SB(src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
+ src87_r, src98_r, src109_r);
+ out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
+ filt1, filt2, filt3);
+ out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
+ filt1, filt2, filt3);
+ out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
+ filt1, filt2, filt3);
+ out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
+ filt1, filt2, filt3);
+ SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
+ SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
+ tmp0 = PCKEV_XORI128_UB(out0_r, out1_r);
+ tmp1 = PCKEV_XORI128_UB(out2_r, out3_r);
+ ST8x4_UB(tmp0, tmp1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ src10_r = src54_r;
+ src32_r = src76_r;
+ src54_r = src98_r;
+ src21_r = src65_r;
+ src43_r = src87_r;
+ src65_r = src109_r;
+ src6 = src10;
+ }
+}
+
+static void common_vt_8t_16w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 filt0, filt1, filt2, filt3;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l;
+ v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l;
+ v16u8 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt, out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l;
+
+ src -= (3 * src_stride);
+
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ src += (7 * src_stride);
+ ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r,
+ src54_r, src21_r);
+ ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+ ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l, src32_l,
+ src54_l, src21_l);
+ ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src, src_stride, src7, src8, src9, src10);
+ XORI_B4_128_SB(src7, src8, src9, src10);
+ src += (4 * src_stride);
+
+ ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
+ src87_r, src98_r, src109_r);
+ ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l,
+ src87_l, src98_l, src109_l);
+ out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
+ filt1, filt2, filt3);
+ out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
+ filt1, filt2, filt3);
+ out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
+ filt1, filt2, filt3);
+ out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
+ filt1, filt2, filt3);
+ out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0,
+ filt1, filt2, filt3);
+ out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0,
+ filt1, filt2, filt3);
+ out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0,
+ filt1, filt2, filt3);
+ out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0,
+ filt1, filt2, filt3);
+ SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
+ SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS);
+ SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
+ SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7);
+ PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l, out3_r,
+ tmp0, tmp1, tmp2, tmp3);
+ XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3);
+ ST_UB4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ src10_r = src54_r;
+ src32_r = src76_r;
+ src54_r = src98_r;
+ src21_r = src65_r;
+ src43_r = src87_r;
+ src65_r = src109_r;
+ src10_l = src54_l;
+ src32_l = src76_l;
+ src54_l = src98_l;
+ src21_l = src65_l;
+ src43_l = src87_l;
+ src65_l = src109_l;
+ src6 = src10;
+ }
+}
+
+static void common_vt_8t_16w_mult_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height,
+ int32_t width) {+ const uint8_t *src_tmp;
+ uint8_t *dst_tmp;
+ uint32_t loop_cnt, cnt;
+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16i8 filt0, filt1, filt2, filt3;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l;
+ v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l;
+ v16u8 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt, out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l;
+
+ src -= (3 * src_stride);
+
+ filt = LD_SH(filter);
+ SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3);
+
+ for (cnt = (width >> 4); cnt--;) {+ src_tmp = src;
+ dst_tmp = dst;
+
+ LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6);
+ XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6);
+ src_tmp += (7 * src_stride);
+ ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r,
+ src32_r, src54_r, src21_r);
+ ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r);
+ ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l,
+ src32_l, src54_l, src21_l);
+ ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l);
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_SB4(src_tmp, src_stride, src7, src8, src9, src10);
+ XORI_B4_128_SB(src7, src8, src9, src10);
+ src_tmp += (4 * src_stride);
+ ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r,
+ src87_r, src98_r, src109_r);
+ ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l,
+ src87_l, src98_l, src109_l);
+ out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0,
+ filt1, filt2, filt3);
+ out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0,
+ filt1, filt2, filt3);
+ out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0,
+ filt1, filt2, filt3);
+ out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0,
+ filt1, filt2, filt3);
+ out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0,
+ filt1, filt2, filt3);
+ out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0,
+ filt1, filt2, filt3);
+ out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0,
+ filt1, filt2, filt3);
+ out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0,
+ filt1, filt2, filt3);
+ SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS);
+ SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS);
+ SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7);
+ SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7);
+ PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l,
+ out3_r, tmp0, tmp1, tmp2, tmp3);
+ XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3);
+ ST_UB4(tmp0, tmp1, tmp2, tmp3, dst_tmp, dst_stride);
+ dst_tmp += (4 * dst_stride);
+
+ src10_r = src54_r;
+ src32_r = src76_r;
+ src54_r = src98_r;
+ src21_r = src65_r;
+ src43_r = src87_r;
+ src65_r = src109_r;
+ src10_l = src54_l;
+ src32_l = src76_l;
+ src54_l = src98_l;
+ src21_l = src65_l;
+ src43_l = src87_l;
+ src65_l = src109_l;
+ src6 = src10;
+ }
+
+ src += 16;
+ dst += 16;
+ }
+}
+
+static void common_vt_8t_32w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, filter, height,
+ 32);
+}
+
+static void common_vt_8t_64w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ common_vt_8t_16w_mult_msa(src, src_stride, dst, dst_stride, filter, height,
+ 64);
+}
+
+static void common_vt_2t_4x4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, src4;
+ v16i8 src10_r, src32_r, src21_r, src43_r, src2110, src4332;
+ v16u8 filt0;
+ v8i16 filt;
+ v8u16 tmp0, tmp1;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB5(src, src_stride, src0, src1, src2, src3, src4);
+ src += (5 * src_stride);
+
+ ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
+ src32_r, src43_r);
+ ILVR_D2_SB(src21_r, src10_r, src43_r, src32_r, src2110, src4332);
+ DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ src2110 = __msa_pckev_b((v16i8)tmp1, (v16i8)tmp0);
+ ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
+}
+
+static void common_vt_2t_4x8_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+ v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r;
+ v16i8 src65_r, src87_r, src2110, src4332, src6554, src8776;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v16u8 filt0;
+ v8i16 filt;
+
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+
+ src8 = LD_SB(src);
+ src += src_stride;
+
+ ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r,
+ src32_r, src43_r);
+ ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r,
+ src76_r, src87_r);
+ ILVR_D4_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r,
+ src87_r, src76_r, src2110, src4332, src6554, src8776);
+ DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0,
+ tmp0, tmp1, tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, src2110, src4332);
+ ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride);
+ ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst + 4 * dst_stride, dst_stride);
+}
+
+static void common_vt_2t_4w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ if (4 == height) {+ common_vt_2t_4x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else if (8 == height) {+ common_vt_2t_4x8_msa(src, src_stride, dst, dst_stride, filter);
+ }
+}
+
+static void common_vt_2t_8x4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter) {+ v16u8 src0, src1, src2, src3, src4, vec0, vec1, vec2, vec3, filt0;
+ v16i8 out0, out1;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ /* rearranging filter_y */
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_UB5(src, src_stride, src0, src1, src2, src3, src4);
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1);
+ ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
+ tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+}
+
+static void common_vt_2t_8x8mult_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
+ v16i8 out0, out1;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ /* rearranging filter_y */
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_UB(src);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 3); loop_cnt--;) {+ LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8);
+ src += (8 * src_stride);
+
+ ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3, vec0, vec1,
+ vec2, vec3);
+ ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7, vec4, vec5,
+ vec6, vec7);
+ DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1,
+ tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, tmp0, tmp1,
+ tmp2, tmp3);
+ SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS);
+ SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7);
+ PCKEV_B2_SB(tmp1, tmp0, tmp3, tmp2, out0, out1);
+ ST8x4_UB(out0, out1, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ src0 = src8;
+ }
+}
+
+static void common_vt_2t_8w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ if (4 == height) {+ common_vt_2t_8x4_msa(src, src_stride, dst, dst_stride, filter);
+ } else {+ common_vt_2t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height);
+ }
+}
+
+static void common_vt_2t_16w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ /* rearranging filter_y */
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_UB(src);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_UB4(src, src_stride, src1, src2, src3, src4);
+ src += (4 * src_stride);
+
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
+ ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_ST_SB(tmp0, tmp1, dst);
+ dst += dst_stride;
+
+ ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
+ ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_ST_SB(tmp2, tmp3, dst);
+ dst += dst_stride;
+
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_ST_SB(tmp0, tmp1, dst);
+ dst += dst_stride;
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_ST_SB(tmp2, tmp3, dst);
+ dst += dst_stride;
+
+ src0 = src4;
+ }
+}
+
+static void common_vt_2t_32w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9;
+ v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
+ v8u16 tmp0, tmp1, tmp2, tmp3;
+ v8i16 filt;
+
+ /* rearranging filter_y */
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ src0 = LD_UB(src);
+ src5 = LD_UB(src + 16);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 2); loop_cnt--;) {+ LD_UB4(src, src_stride, src1, src2, src3, src4);
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
+ ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
+
+ LD_UB4(src + 16, src_stride, src6, src7, src8, src9);
+ src += (4 * src_stride);
+
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_ST_SB(tmp0, tmp1, dst);
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_ST_SB(tmp2, tmp3, dst + dst_stride);
+
+ ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6);
+ ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7);
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_ST_SB(tmp0, tmp1, dst + 2 * dst_stride);
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_ST_SB(tmp2, tmp3, dst + 3 * dst_stride);
+
+ ILVR_B2_UB(src6, src5, src7, src6, vec0, vec2);
+ ILVL_B2_UB(src6, src5, src7, src6, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_ST_SB(tmp0, tmp1, dst + 16);
+
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_ST_SB(tmp2, tmp3, dst + 16 + dst_stride);
+
+ ILVR_B2_UB(src8, src7, src9, src8, vec4, vec6);
+ ILVL_B2_UB(src8, src7, src9, src8, vec5, vec7);
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_ST_SB(tmp0, tmp1, dst + 16 + 2 * dst_stride);
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_ST_SB(tmp2, tmp3, dst + 16 + 3 * dst_stride);
+ dst += (4 * dst_stride);
+
+ src0 = src4;
+ src5 = src9;
+ }
+}
+
+static void common_vt_2t_64w_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int8_t *filter, int32_t height) {+ uint32_t loop_cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;
+ v16u8 src11, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0;
+ v8u16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+ v8i16 filt;
+
+ /* rearranging filter_y */
+ filt = LD_SH(filter);
+ filt0 = (v16u8)__msa_splati_h(filt, 0);
+
+ LD_UB4(src, 16, src0, src3, src6, src9);
+ src += src_stride;
+
+ for (loop_cnt = (height >> 1); loop_cnt--;) {+ LD_UB2(src, src_stride, src1, src2);
+ LD_UB2(src + 16, src_stride, src4, src5);
+ LD_UB2(src + 32, src_stride, src7, src8);
+ LD_UB2(src + 48, src_stride, src10, src11);
+ src += (2 * src_stride);
+
+ ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2);
+ ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_ST_SB(tmp0, tmp1, dst);
+
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_ST_SB(tmp2, tmp3, dst + dst_stride);
+
+ ILVR_B2_UB(src4, src3, src5, src4, vec4, vec6);
+ ILVL_B2_UB(src4, src3, src5, src4, vec5, vec7);
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
+ SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
+ SAT_UH2_UH(tmp4, tmp5, 7);
+ PCKEV_ST_SB(tmp4, tmp5, dst + 16);
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
+ SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
+ SAT_UH2_UH(tmp6, tmp7, 7);
+ PCKEV_ST_SB(tmp6, tmp7, dst + 16 + dst_stride);
+
+ ILVR_B2_UB(src7, src6, src8, src7, vec0, vec2);
+ ILVL_B2_UB(src7, src6, src8, src7, vec1, vec3);
+ DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1);
+ SRARI_H2_UH(tmp0, tmp1, FILTER_BITS);
+ SAT_UH2_UH(tmp0, tmp1, 7);
+ PCKEV_ST_SB(tmp0, tmp1, dst + 32);
+
+ DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3);
+ SRARI_H2_UH(tmp2, tmp3, FILTER_BITS);
+ SAT_UH2_UH(tmp2, tmp3, 7);
+ PCKEV_ST_SB(tmp2, tmp3, dst + 32 + dst_stride);
+
+ ILVR_B2_UB(src10, src9, src11, src10, vec4, vec6);
+ ILVL_B2_UB(src10, src9, src11, src10, vec5, vec7);
+ DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5);
+ SRARI_H2_UH(tmp4, tmp5, FILTER_BITS);
+ SAT_UH2_UH(tmp4, tmp5, 7);
+ PCKEV_ST_SB(tmp4, tmp5, dst + 48);
+
+ DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7);
+ SRARI_H2_UH(tmp6, tmp7, FILTER_BITS);
+ SAT_UH2_UH(tmp6, tmp7, 7);
+ PCKEV_ST_SB(tmp6, tmp7, dst + 48 + dst_stride);
+ dst += (2 * dst_stride);
+
+ src0 = src2;
+ src3 = src5;
+ src6 = src8;
+ src9 = src11;
+ }
+}
+
+void vpx_convolve8_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ int8_t cnt, filt_ver[8];
+
+ if (16 != y_step_q4) {+ vpx_convolve8_vert_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ if (((const int32_t *)filter_y)[1] == 0x800000) {+ vpx_convolve_copy(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ return;
+ }
+
+ for (cnt = 8; cnt--;) {+ filt_ver[cnt] = filter_y[cnt];
+ }
+
+ if (((const int32_t *)filter_y)[0] == 0) {+ switch (w) {+ case 4:
+ common_vt_2t_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ case 8:
+ common_vt_2t_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ case 16:
+ common_vt_2t_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ case 32:
+ common_vt_2t_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ case 64:
+ common_vt_2t_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ &filt_ver[3], h);
+ break;
+ default:
+ vpx_convolve8_vert_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ } else {+ switch (w) {+ case 4:
+ common_vt_8t_4w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ case 8:
+ common_vt_8t_8w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ case 16:
+ common_vt_8t_16w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ case 32:
+ common_vt_8t_32w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ case 64:
+ common_vt_8t_64w_msa(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filt_ver, h);
+ break;
+ default:
+ vpx_convolve8_vert_c(src, src_stride, dst, dst_stride,
+ filter_x, x_step_q4, filter_y, y_step_q4,
+ w, h);
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve_avg_msa.c
@@ -1,0 +1,232 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "vpx_dsp/mips/macros_msa.h"
+
+static void avg_width4_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint32_t out0, out1, out2, out3;
+ v16u8 src0, src1, src2, src3;
+ v16u8 dst0, dst1, dst2, dst3;
+
+ if (0 == (height % 4)) {+ for (cnt = (height / 4); cnt--;) {+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+
+ AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
+ dst0, dst1, dst2, dst3);
+
+ out0 = __msa_copy_u_w((v4i32)dst0, 0);
+ out1 = __msa_copy_u_w((v4i32)dst1, 0);
+ out2 = __msa_copy_u_w((v4i32)dst2, 0);
+ out3 = __msa_copy_u_w((v4i32)dst3, 0);
+ SW4(out0, out1, out2, out3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+ } else if (0 == (height % 2)) {+ for (cnt = (height / 2); cnt--;) {+ LD_UB2(src, src_stride, src0, src1);
+ src += (2 * src_stride);
+
+ LD_UB2(dst, dst_stride, dst0, dst1);
+
+ AVER_UB2_UB(src0, dst0, src1, dst1, dst0, dst1);
+
+ out0 = __msa_copy_u_w((v4i32)dst0, 0);
+ out1 = __msa_copy_u_w((v4i32)dst1, 0);
+ SW(out0, dst);
+ dst += dst_stride;
+ SW(out1, dst);
+ dst += dst_stride;
+ }
+ }
+}
+
+static void avg_width8_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint64_t out0, out1, out2, out3;
+ v16u8 src0, src1, src2, src3;
+ v16u8 dst0, dst1, dst2, dst3;
+
+ for (cnt = (height / 4); cnt--;) {+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+ LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3);
+
+ AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
+ dst0, dst1, dst2, dst3);
+
+ out0 = __msa_copy_u_d((v2i64)dst0, 0);
+ out1 = __msa_copy_u_d((v2i64)dst1, 0);
+ out2 = __msa_copy_u_d((v2i64)dst2, 0);
+ out3 = __msa_copy_u_d((v2i64)dst3, 0);
+ SD4(out0, out1, out2, out3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void avg_width16_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+
+ for (cnt = (height / 8); cnt--;) {+ LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+ LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7);
+
+ AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
+ dst0, dst1, dst2, dst3);
+ AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7,
+ dst4, dst5, dst6, dst7);
+ ST_UB8(dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst, dst_stride);
+ dst += (8 * dst_stride);
+ }
+}
+
+static void avg_width32_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint8_t *dst_dup = dst;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+ v16u8 src8, src9, src10, src11, src12, src13, src14, src15;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
+
+ for (cnt = (height / 8); cnt--;) {+ LD_UB4(src, src_stride, src0, src2, src4, src6);
+ LD_UB4(src + 16, src_stride, src1, src3, src5, src7);
+ src += (4 * src_stride);
+ LD_UB4(dst_dup, dst_stride, dst0, dst2, dst4, dst6);
+ LD_UB4(dst_dup + 16, dst_stride, dst1, dst3, dst5, dst7);
+ dst_dup += (4 * dst_stride);
+ LD_UB4(src, src_stride, src8, src10, src12, src14);
+ LD_UB4(src + 16, src_stride, src9, src11, src13, src15);
+ src += (4 * src_stride);
+ LD_UB4(dst_dup, dst_stride, dst8, dst10, dst12, dst14);
+ LD_UB4(dst_dup + 16, dst_stride, dst9, dst11, dst13, dst15);
+ dst_dup += (4 * dst_stride);
+
+ AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
+ dst0, dst1, dst2, dst3);
+ AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7,
+ dst4, dst5, dst6, dst7);
+ AVER_UB4_UB(src8, dst8, src9, dst9, src10, dst10, src11, dst11,
+ dst8, dst9, dst10, dst11);
+ AVER_UB4_UB(src12, dst12, src13, dst13, src14, dst14, src15, dst15,
+ dst12, dst13, dst14, dst15);
+
+ ST_UB4(dst0, dst2, dst4, dst6, dst, dst_stride);
+ ST_UB4(dst1, dst3, dst5, dst7, dst + 16, dst_stride);
+ dst += (4 * dst_stride);
+ ST_UB4(dst8, dst10, dst12, dst14, dst, dst_stride);
+ ST_UB4(dst9, dst11, dst13, dst15, dst + 16, dst_stride);
+ dst += (4 * dst_stride);
+ }
+}
+
+static void avg_width64_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint8_t *dst_dup = dst;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+ v16u8 src8, src9, src10, src11, src12, src13, src14, src15;
+ v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
+ v16u8 dst8, dst9, dst10, dst11, dst12, dst13, dst14, dst15;
+
+ for (cnt = (height / 4); cnt--;) {+ LD_UB4(src, 16, src0, src1, src2, src3);
+ src += src_stride;
+ LD_UB4(src, 16, src4, src5, src6, src7);
+ src += src_stride;
+ LD_UB4(src, 16, src8, src9, src10, src11);
+ src += src_stride;
+ LD_UB4(src, 16, src12, src13, src14, src15);
+ src += src_stride;
+
+ LD_UB4(dst_dup, 16, dst0, dst1, dst2, dst3);
+ dst_dup += dst_stride;
+ LD_UB4(dst_dup, 16, dst4, dst5, dst6, dst7);
+ dst_dup += dst_stride;
+ LD_UB4(dst_dup, 16, dst8, dst9, dst10, dst11);
+ dst_dup += dst_stride;
+ LD_UB4(dst_dup, 16, dst12, dst13, dst14, dst15);
+ dst_dup += dst_stride;
+
+ AVER_UB4_UB(src0, dst0, src1, dst1, src2, dst2, src3, dst3,
+ dst0, dst1, dst2, dst3);
+ AVER_UB4_UB(src4, dst4, src5, dst5, src6, dst6, src7, dst7,
+ dst4, dst5, dst6, dst7);
+ AVER_UB4_UB(src8, dst8, src9, dst9, src10, dst10, src11, dst11,
+ dst8, dst9, dst10, dst11);
+ AVER_UB4_UB(src12, dst12, src13, dst13, src14, dst14, src15, dst15,
+ dst12, dst13, dst14, dst15);
+
+ ST_UB4(dst0, dst1, dst2, dst3, dst, 16);
+ dst += dst_stride;
+ ST_UB4(dst4, dst5, dst6, dst7, dst, 16);
+ dst += dst_stride;
+ ST_UB4(dst8, dst9, dst10, dst11, dst, 16);
+ dst += dst_stride;
+ ST_UB4(dst12, dst13, dst14, dst15, dst, 16);
+ dst += dst_stride;
+ }
+}
+
+void vpx_convolve_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int32_t filter_x_stride,
+ const int16_t *filter_y, int32_t filter_y_stride,
+ int32_t w, int32_t h) {+ (void)filter_x;
+ (void)filter_y;
+ (void)filter_x_stride;
+ (void)filter_y_stride;
+
+ switch (w) {+ case 4: {+ avg_width4_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 8: {+ avg_width8_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 16: {+ avg_width16_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 32: {+ avg_width32_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 64: {+ avg_width64_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ default: {+ int32_t lp, cnt;
+ for (cnt = h; cnt--;) {+ for (lp = 0; lp < w; ++lp) {+ dst[lp] = (((dst[lp] + src[lp]) + 1) >> 1);
+ }
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve_copy_msa.c
@@ -1,0 +1,247 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <string.h>
+#include "vpx_dsp/mips/macros_msa.h"
+
+static void copy_width8_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ if (0 == height % 12) {+ for (cnt = (height / 12); cnt--;) {+ LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+ out2 = __msa_copy_u_d((v2i64)src2, 0);
+ out3 = __msa_copy_u_d((v2i64)src3, 0);
+ out4 = __msa_copy_u_d((v2i64)src4, 0);
+ out5 = __msa_copy_u_d((v2i64)src5, 0);
+ out6 = __msa_copy_u_d((v2i64)src6, 0);
+ out7 = __msa_copy_u_d((v2i64)src7, 0);
+
+ SD4(out0, out1, out2, out3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ SD4(out4, out5, out6, out7, dst, dst_stride);
+ dst += (4 * dst_stride);
+
+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+ out2 = __msa_copy_u_d((v2i64)src2, 0);
+ out3 = __msa_copy_u_d((v2i64)src3, 0);
+ SD4(out0, out1, out2, out3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+ } else if (0 == height % 8) {+ for (cnt = height >> 3; cnt--;) {+ LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+ out2 = __msa_copy_u_d((v2i64)src2, 0);
+ out3 = __msa_copy_u_d((v2i64)src3, 0);
+ out4 = __msa_copy_u_d((v2i64)src4, 0);
+ out5 = __msa_copy_u_d((v2i64)src5, 0);
+ out6 = __msa_copy_u_d((v2i64)src6, 0);
+ out7 = __msa_copy_u_d((v2i64)src7, 0);
+
+ SD4(out0, out1, out2, out3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ SD4(out4, out5, out6, out7, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+ } else if (0 == height % 4) {+ for (cnt = (height / 4); cnt--;) {+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+ out2 = __msa_copy_u_d((v2i64)src2, 0);
+ out3 = __msa_copy_u_d((v2i64)src3, 0);
+
+ SD4(out0, out1, out2, out3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+ } else if (0 == height % 2) {+ for (cnt = (height / 2); cnt--;) {+ LD_UB2(src, src_stride, src0, src1);
+ src += (2 * src_stride);
+ out0 = __msa_copy_u_d((v2i64)src0, 0);
+ out1 = __msa_copy_u_d((v2i64)src1, 0);
+
+ SD(out0, dst);
+ dst += dst_stride;
+ SD(out1, dst);
+ dst += dst_stride;
+ }
+ }
+}
+
+static void copy_16multx8mult_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride,
+ int32_t height, int32_t width) {+ int32_t cnt, loop_cnt;
+ const uint8_t *src_tmp;
+ uint8_t *dst_tmp;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ for (cnt = (width >> 4); cnt--;) {+ src_tmp = src;
+ dst_tmp = dst;
+
+ for (loop_cnt = (height >> 3); loop_cnt--;) {+ LD_UB8(src_tmp, src_stride,
+ src0, src1, src2, src3, src4, src5, src6, src7);
+ src_tmp += (8 * src_stride);
+
+ ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7,
+ dst_tmp, dst_stride);
+ dst_tmp += (8 * dst_stride);
+ }
+
+ src += 16;
+ dst += 16;
+ }
+}
+
+static void copy_width16_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ if (0 == height % 12) {+ for (cnt = (height / 12); cnt--;) {+ LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+ src += (8 * src_stride);
+ ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, dst, dst_stride);
+ dst += (8 * dst_stride);
+
+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+ ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+ } else if (0 == height % 8) {+ copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 16);
+ } else if (0 == height % 4) {+ for (cnt = (height >> 2); cnt--;) {+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ src += (4 * src_stride);
+
+ ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+ dst += (4 * dst_stride);
+ }
+ }
+}
+
+static void copy_width32_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ int32_t cnt;
+ v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+
+ if (0 == height % 12) {+ for (cnt = (height / 12); cnt--;) {+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
+ src += (4 * src_stride);
+ ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+ ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
+ dst += (4 * dst_stride);
+
+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
+ src += (4 * src_stride);
+ ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+ ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
+ dst += (4 * dst_stride);
+
+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
+ src += (4 * src_stride);
+ ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+ ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
+ dst += (4 * dst_stride);
+ }
+ } else if (0 == height % 8) {+ copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 32);
+ } else if (0 == height % 4) {+ for (cnt = (height >> 2); cnt--;) {+ LD_UB4(src, src_stride, src0, src1, src2, src3);
+ LD_UB4(src + 16, src_stride, src4, src5, src6, src7);
+ src += (4 * src_stride);
+ ST_UB4(src0, src1, src2, src3, dst, dst_stride);
+ ST_UB4(src4, src5, src6, src7, dst + 16, dst_stride);
+ dst += (4 * dst_stride);
+ }
+ }
+}
+
+static void copy_width64_msa(const uint8_t *src, int32_t src_stride,
+ uint8_t *dst, int32_t dst_stride, int32_t height) {+ copy_16multx8mult_msa(src, src_stride, dst, dst_stride, height, 64);
+}
+
+void vpx_convolve_copy_msa(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int32_t filter_x_stride,
+ const int16_t *filter_y, int32_t filter_y_stride,
+ int32_t w, int32_t h) {+ (void)filter_x;
+ (void)filter_y;
+ (void)filter_x_stride;
+ (void)filter_y_stride;
+
+ switch (w) {+ case 4: {+ uint32_t cnt, tmp;
+ /* 1 word storage */
+ for (cnt = h; cnt--;) {+ tmp = LW(src);
+ SW(tmp, dst);
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+ case 8: {+ copy_width8_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 16: {+ copy_width16_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 32: {+ copy_width32_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 64: {+ copy_width64_msa(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ default: {+ uint32_t cnt;
+ for (cnt = h; cnt--;) {+ memcpy(dst, src, w);
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+ }
+}
--- /dev/null
+++ b/vpx_dsp/mips/vpx_convolve_msa.h
@@ -1,0 +1,119 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VPX_DSP_MIPS_VPX_CONVOLVE_MSA_H_
+#define VPX_DSP_MIPS_VPX_CONVOLVE_MSA_H_
+
+#include "vpx_dsp/mips/macros_msa.h"
+#include "vpx_dsp/vpx_filter.h"
+
+extern const uint8_t mc_filt_mask_arr[16 * 3];
+
+#define FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, \
+ filt0, filt1, filt2, filt3) ({ \+ v8i16 tmp0, tmp1; \
+ \
+ tmp0 = __msa_dotp_s_h((v16i8)vec0, (v16i8)filt0); \
+ tmp0 = __msa_dpadd_s_h(tmp0, (v16i8)vec1, (v16i8)filt1); \
+ tmp1 = __msa_dotp_s_h((v16i8)vec2, (v16i8)filt2); \
+ tmp1 = __msa_dpadd_s_h(tmp1, (v16i8)vec3, (v16i8)filt3); \
+ tmp0 = __msa_adds_s_h(tmp0, tmp1); \
+ \
+ tmp0; \
+})
+
+#define HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, \
+ filt_h0, filt_h1, filt_h2, filt_h3) ({ \+ v16i8 vec0_m, vec1_m, vec2_m, vec3_m; \
+ v8i16 hz_out_m; \
+ \
+ VSHF_B4_SB(src0, src1, mask0, mask1, mask2, mask3, \
+ vec0_m, vec1_m, vec2_m, vec3_m); \
+ hz_out_m = FILT_8TAP_DPADD_S_H(vec0_m, vec1_m, vec2_m, vec3_m, \
+ filt_h0, filt_h1, filt_h2, filt_h3); \
+ \
+ hz_out_m = __msa_srari_h(hz_out_m, FILTER_BITS); \
+ hz_out_m = __msa_sat_s_h(hz_out_m, 7); \
+ \
+ hz_out_m; \
+})
+
+#define HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, \
+ mask0, mask1, mask2, mask3, \
+ filt0, filt1, filt2, filt3, \
+ out0, out1) { \+ v16i8 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \
+ v8i16 res0_m, res1_m, res2_m, res3_m; \
+ \
+ VSHF_B2_SB(src0, src1, src2, src3, mask0, mask0, vec0_m, vec1_m); \
+ DOTP_SB2_SH(vec0_m, vec1_m, filt0, filt0, res0_m, res1_m); \
+ VSHF_B2_SB(src0, src1, src2, src3, mask1, mask1, vec2_m, vec3_m); \
+ DPADD_SB2_SH(vec2_m, vec3_m, filt1, filt1, res0_m, res1_m); \
+ VSHF_B2_SB(src0, src1, src2, src3, mask2, mask2, vec4_m, vec5_m); \
+ DOTP_SB2_SH(vec4_m, vec5_m, filt2, filt2, res2_m, res3_m); \
+ VSHF_B2_SB(src0, src1, src2, src3, mask3, mask3, vec6_m, vec7_m); \
+ DPADD_SB2_SH(vec6_m, vec7_m, filt3, filt3, res2_m, res3_m); \
+ ADDS_SH2_SH(res0_m, res2_m, res1_m, res3_m, out0, out1); \
+}
+
+#define HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, \
+ mask0, mask1, mask2, mask3, \
+ filt0, filt1, filt2, filt3, \
+ out0, out1, out2, out3) { \+ v16i8 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \
+ v8i16 res0_m, res1_m, res2_m, res3_m, res4_m, res5_m, res6_m, res7_m; \
+ \
+ VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0_m, vec1_m); \
+ VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2_m, vec3_m); \
+ DOTP_SB4_SH(vec0_m, vec1_m, vec2_m, vec3_m, filt0, filt0, filt0, filt0, \
+ res0_m, res1_m, res2_m, res3_m); \
+ VSHF_B2_SB(src0, src0, src1, src1, mask2, mask2, vec0_m, vec1_m); \
+ VSHF_B2_SB(src2, src2, src3, src3, mask2, mask2, vec2_m, vec3_m); \
+ DOTP_SB4_SH(vec0_m, vec1_m, vec2_m, vec3_m, filt2, filt2, filt2, filt2, \
+ res4_m, res5_m, res6_m, res7_m); \
+ VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec4_m, vec5_m); \
+ VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec6_m, vec7_m); \
+ DPADD_SB4_SH(vec4_m, vec5_m, vec6_m, vec7_m, filt1, filt1, filt1, filt1, \
+ res0_m, res1_m, res2_m, res3_m); \
+ VSHF_B2_SB(src0, src0, src1, src1, mask3, mask3, vec4_m, vec5_m); \
+ VSHF_B2_SB(src2, src2, src3, src3, mask3, mask3, vec6_m, vec7_m); \
+ DPADD_SB4_SH(vec4_m, vec5_m, vec6_m, vec7_m, filt3, filt3, filt3, filt3, \
+ res4_m, res5_m, res6_m, res7_m); \
+ ADDS_SH4_SH(res0_m, res4_m, res1_m, res5_m, res2_m, res6_m, res3_m, \
+ res7_m, out0, out1, out2, out3); \
+}
+
+#define PCKEV_XORI128_AVG_ST_UB(in0, in1, dst, pdst) { \+ v16u8 tmp_m; \
+ \
+ tmp_m = PCKEV_XORI128_UB(in1, in0); \
+ tmp_m = __msa_aver_u_b(tmp_m, (v16u8)dst); \
+ ST_UB(tmp_m, (pdst)); \
+}
+
+#define PCKEV_AVG_ST_UB(in0, in1, dst, pdst) { \+ v16u8 tmp_m; \
+ \
+ tmp_m = (v16u8)__msa_pckev_b((v16i8)in0, (v16i8)in1); \
+ tmp_m = __msa_aver_u_b(tmp_m, (v16u8)dst); \
+ ST_UB(tmp_m, (pdst)); \
+}
+
+#define PCKEV_AVG_ST8x4_UB(in1, dst0, in2, dst1, in3, dst2, in4, dst3, \
+ pdst, stride) { \+ v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
+ uint8_t *pdst_m = (uint8_t *)(pdst); \
+ \
+ PCKEV_B2_UB(in2, in1, in4, in3, tmp0_m, tmp1_m); \
+ PCKEV_D2_UB(dst1, dst0, dst3, dst2, tmp2_m, tmp3_m); \
+ AVER_UB2_UB(tmp0_m, tmp2_m, tmp1_m, tmp3_m, tmp0_m, tmp1_m); \
+ ST8x4_UB(tmp0_m, tmp1_m, pdst_m, stride); \
+}
+#endif /* VPX_DSP_MIPS_VPX_CONVOLVE_MSA_H_ */
--- /dev/null
+++ b/vpx_dsp/vpx_convolve.c
@@ -1,0 +1,558 @@
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <string.h>
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_dsp/vpx_convolve.h"
+#include "vpx_dsp/vpx_dsp_common.h"
+#include "vpx_dsp/vpx_filter.h"
+#include "vpx_ports/mem.h"
+
+static void convolve_horiz(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const InterpKernel *x_filters,
+ int x0_q4, int x_step_q4, int w, int h) {+ int x, y;
+ src -= SUBPEL_TAPS / 2 - 1;
+ for (y = 0; y < h; ++y) {+ int x_q4 = x0_q4;
+ for (x = 0; x < w; ++x) {+ const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
+ const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
+ int k, sum = 0;
+ for (k = 0; k < SUBPEL_TAPS; ++k)
+ sum += src_x[k] * x_filter[k];
+ dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
+ x_q4 += x_step_q4;
+ }
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const InterpKernel *x_filters,
+ int x0_q4, int x_step_q4, int w, int h) {+ int x, y;
+ src -= SUBPEL_TAPS / 2 - 1;
+ for (y = 0; y < h; ++y) {+ int x_q4 = x0_q4;
+ for (x = 0; x < w; ++x) {+ const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
+ const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
+ int k, sum = 0;
+ for (k = 0; k < SUBPEL_TAPS; ++k)
+ sum += src_x[k] * x_filter[k];
+ dst[x] = ROUND_POWER_OF_TWO(dst[x] +
+ clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
+ x_q4 += x_step_q4;
+ }
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const InterpKernel *y_filters,
+ int y0_q4, int y_step_q4, int w, int h) {+ int x, y;
+ src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+
+ for (x = 0; x < w; ++x) {+ int y_q4 = y0_q4;
+ for (y = 0; y < h; ++y) {+ const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
+ const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
+ int k, sum = 0;
+ for (k = 0; k < SUBPEL_TAPS; ++k)
+ sum += src_y[k * src_stride] * y_filter[k];
+ dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
+ y_q4 += y_step_q4;
+ }
+ ++src;
+ ++dst;
+ }
+}
+
+static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const InterpKernel *y_filters,
+ int y0_q4, int y_step_q4, int w, int h) {+ int x, y;
+ src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+
+ for (x = 0; x < w; ++x) {+ int y_q4 = y0_q4;
+ for (y = 0; y < h; ++y) {+ const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
+ const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
+ int k, sum = 0;
+ for (k = 0; k < SUBPEL_TAPS; ++k)
+ sum += src_y[k * src_stride] * y_filter[k];
+ dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] +
+ clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
+ y_q4 += y_step_q4;
+ }
+ ++src;
+ ++dst;
+ }
+}
+
+static void convolve(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const InterpKernel *const x_filters,
+ int x0_q4, int x_step_q4,
+ const InterpKernel *const y_filters,
+ int y0_q4, int y_step_q4,
+ int w, int h) {+ // Note: Fixed size intermediate buffer, temp, places limits on parameters.
+ // 2d filtering proceeds in 2 steps:
+ // (1) Interpolate horizontally into an intermediate buffer, temp.
+ // (2) Interpolate temp vertically to derive the sub-pixel result.
+ // Deriving the maximum number of rows in the temp buffer (135):
+ // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
+ // --Largest block size is 64x64 pixels.
+ // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
+ // original frame (in 1/16th pixel units).
+ // --Must round-up because block may be located at sub-pixel position.
+ // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
+ // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
+ uint8_t temp[135 * 64];
+ int intermediate_height =
+ (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
+
+ assert(w <= 64);
+ assert(h <= 64);
+ assert(y_step_q4 <= 32);
+ assert(x_step_q4 <= 32);
+
+ convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp, 64,
+ x_filters, x0_q4, x_step_q4, w, intermediate_height);
+ convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride,
+ y_filters, y0_q4, y_step_q4, w, h);
+}
+
+static const InterpKernel *get_filter_base(const int16_t *filter) {+ // NOTE: This assumes that the filter table is 256-byte aligned.
+ // TODO(agrange) Modify to make independent of table alignment.
+ return (const InterpKernel *)(((intptr_t)filter) & ~((intptr_t)0xFF));
+}
+
+static int get_filter_offset(const int16_t *f, const InterpKernel *base) {+ return (int)((const InterpKernel *)(intptr_t)f - base);
+}
+
+void vpx_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ const InterpKernel *const filters_x = get_filter_base(filter_x);
+ const int x0_q4 = get_filter_offset(filter_x, filters_x);
+
+ (void)filter_y;
+ (void)y_step_q4;
+
+ convolve_horiz(src, src_stride, dst, dst_stride, filters_x,
+ x0_q4, x_step_q4, w, h);
+}
+
+void vpx_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ const InterpKernel *const filters_x = get_filter_base(filter_x);
+ const int x0_q4 = get_filter_offset(filter_x, filters_x);
+
+ (void)filter_y;
+ (void)y_step_q4;
+
+ convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x,
+ x0_q4, x_step_q4, w, h);
+}
+
+void vpx_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ const InterpKernel *const filters_y = get_filter_base(filter_y);
+ const int y0_q4 = get_filter_offset(filter_y, filters_y);
+
+ (void)filter_x;
+ (void)x_step_q4;
+
+ convolve_vert(src, src_stride, dst, dst_stride, filters_y,
+ y0_q4, y_step_q4, w, h);
+}
+
+void vpx_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ const InterpKernel *const filters_y = get_filter_base(filter_y);
+ const int y0_q4 = get_filter_offset(filter_y, filters_y);
+
+ (void)filter_x;
+ (void)x_step_q4;
+
+ convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y,
+ y0_q4, y_step_q4, w, h);
+}
+
+void vpx_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ const InterpKernel *const filters_x = get_filter_base(filter_x);
+ const int x0_q4 = get_filter_offset(filter_x, filters_x);
+
+ const InterpKernel *const filters_y = get_filter_base(filter_y);
+ const int y0_q4 = get_filter_offset(filter_y, filters_y);
+
+ convolve(src, src_stride, dst, dst_stride,
+ filters_x, x0_q4, x_step_q4,
+ filters_y, y0_q4, y_step_q4, w, h);
+}
+
+void vpx_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h) {+ /* Fixed size intermediate buffer places limits on parameters. */
+ DECLARE_ALIGNED(16, uint8_t, temp[64 * 64]);
+ assert(w <= 64);
+ assert(h <= 64);
+
+ vpx_convolve8_c(src, src_stride, temp, 64,
+ filter_x, x_step_q4, filter_y, y_step_q4, w, h);
+ vpx_convolve_avg_c(temp, 64, dst, dst_stride, NULL, 0, NULL, 0, w, h);
+}
+
+void vpx_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int filter_x_stride,
+ const int16_t *filter_y, int filter_y_stride,
+ int w, int h) {+ int r;
+
+ (void)filter_x; (void)filter_x_stride;
+ (void)filter_y; (void)filter_y_stride;
+
+ for (r = h; r > 0; --r) {+ memcpy(dst, src, w);
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+void vpx_convolve_avg_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int filter_x_stride,
+ const int16_t *filter_y, int filter_y_stride,
+ int w, int h) {+ int x, y;
+
+ (void)filter_x; (void)filter_x_stride;
+ (void)filter_y; (void)filter_y_stride;
+
+ for (y = 0; y < h; ++y) {+ for (x = 0; x < w; ++x)
+ dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
+
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static void highbd_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
+ uint8_t *dst8, ptrdiff_t dst_stride,
+ const InterpKernel *x_filters,
+ int x0_q4, int x_step_q4,
+ int w, int h, int bd) {+ int x, y;
+ uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+ uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+ src -= SUBPEL_TAPS / 2 - 1;
+ for (y = 0; y < h; ++y) {+ int x_q4 = x0_q4;
+ for (x = 0; x < w; ++x) {+ const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
+ const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
+ int k, sum = 0;
+ for (k = 0; k < SUBPEL_TAPS; ++k)
+ sum += src_x[k] * x_filter[k];
+ dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+ x_q4 += x_step_q4;
+ }
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void highbd_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride,
+ uint8_t *dst8, ptrdiff_t dst_stride,
+ const InterpKernel *x_filters,
+ int x0_q4, int x_step_q4,
+ int w, int h, int bd) {+ int x, y;
+ uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+ uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+ src -= SUBPEL_TAPS / 2 - 1;
+ for (y = 0; y < h; ++y) {+ int x_q4 = x0_q4;
+ for (x = 0; x < w; ++x) {+ const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
+ const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
+ int k, sum = 0;
+ for (k = 0; k < SUBPEL_TAPS; ++k)
+ sum += src_x[k] * x_filter[k];
+ dst[x] = ROUND_POWER_OF_TWO(dst[x] +
+ clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
+ x_q4 += x_step_q4;
+ }
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static void highbd_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride,
+ uint8_t *dst8, ptrdiff_t dst_stride,
+ const InterpKernel *y_filters,
+ int y0_q4, int y_step_q4, int w, int h,
+ int bd) {+ int x, y;
+ uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+ uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+ src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+ for (x = 0; x < w; ++x) {+ int y_q4 = y0_q4;
+ for (y = 0; y < h; ++y) {+ const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
+ const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
+ int k, sum = 0;
+ for (k = 0; k < SUBPEL_TAPS; ++k)
+ sum += src_y[k * src_stride] * y_filter[k];
+ dst[y * dst_stride] = clip_pixel_highbd(
+ ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
+ y_q4 += y_step_q4;
+ }
+ ++src;
+ ++dst;
+ }
+}
+
+static void highbd_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride,
+ uint8_t *dst8, ptrdiff_t dst_stride,
+ const InterpKernel *y_filters,
+ int y0_q4, int y_step_q4, int w, int h,
+ int bd) {+ int x, y;
+ uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+ uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+ src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+ for (x = 0; x < w; ++x) {+ int y_q4 = y0_q4;
+ for (y = 0; y < h; ++y) {+ const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
+ const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
+ int k, sum = 0;
+ for (k = 0; k < SUBPEL_TAPS; ++k)
+ sum += src_y[k * src_stride] * y_filter[k];
+ dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] +
+ clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
+ y_q4 += y_step_q4;
+ }
+ ++src;
+ ++dst;
+ }
+}
+
+static void highbd_convolve(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const InterpKernel *const x_filters,
+ int x0_q4, int x_step_q4,
+ const InterpKernel *const y_filters,
+ int y0_q4, int y_step_q4,
+ int w, int h, int bd) {+ // Note: Fixed size intermediate buffer, temp, places limits on parameters.
+ // 2d filtering proceeds in 2 steps:
+ // (1) Interpolate horizontally into an intermediate buffer, temp.
+ // (2) Interpolate temp vertically to derive the sub-pixel result.
+ // Deriving the maximum number of rows in the temp buffer (135):
+ // --Smallest scaling factor is x1/2 ==> y_step_q4 = 32 (Normative).
+ // --Largest block size is 64x64 pixels.
+ // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
+ // original frame (in 1/16th pixel units).
+ // --Must round-up because block may be located at sub-pixel position.
+ // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
+ // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
+ uint16_t temp[64 * 135];
+ int intermediate_height =
+ (((h - 1) * y_step_q4 + y0_q4) >> SUBPEL_BITS) + SUBPEL_TAPS;
+
+ assert(w <= 64);
+ assert(h <= 64);
+ assert(y_step_q4 <= 32);
+ assert(x_step_q4 <= 32);
+
+ highbd_convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1),
+ src_stride, CONVERT_TO_BYTEPTR(temp), 64,
+ x_filters, x0_q4, x_step_q4, w,
+ intermediate_height, bd);
+ highbd_convolve_vert(CONVERT_TO_BYTEPTR(temp) + 64 * (SUBPEL_TAPS / 2 - 1),
+ 64, dst, dst_stride, y_filters, y0_q4, y_step_q4,
+ w, h, bd);
+}
+
+
+void vpx_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h, int bd) {+ const InterpKernel *const filters_x = get_filter_base(filter_x);
+ const int x0_q4 = get_filter_offset(filter_x, filters_x);
+ (void)filter_y;
+ (void)y_step_q4;
+
+ highbd_convolve_horiz(src, src_stride, dst, dst_stride, filters_x,
+ x0_q4, x_step_q4, w, h, bd);
+}
+
+void vpx_highbd_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h, int bd) {+ const InterpKernel *const filters_x = get_filter_base(filter_x);
+ const int x0_q4 = get_filter_offset(filter_x, filters_x);
+ (void)filter_y;
+ (void)y_step_q4;
+
+ highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x,
+ x0_q4, x_step_q4, w, h, bd);
+}
+
+void vpx_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h, int bd) {+ const InterpKernel *const filters_y = get_filter_base(filter_y);
+ const int y0_q4 = get_filter_offset(filter_y, filters_y);
+ (void)filter_x;
+ (void)x_step_q4;
+
+ highbd_convolve_vert(src, src_stride, dst, dst_stride, filters_y,
+ y0_q4, y_step_q4, w, h, bd);
+}
+
+void vpx_highbd_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h, int bd) {+ const InterpKernel *const filters_y = get_filter_base(filter_y);
+ const int y0_q4 = get_filter_offset(filter_y, filters_y);
+ (void)filter_x;
+ (void)x_step_q4;
+
+ highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y,
+ y0_q4, y_step_q4, w, h, bd);
+}
+
+void vpx_highbd_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h, int bd) {+ const InterpKernel *const filters_x = get_filter_base(filter_x);
+ const int x0_q4 = get_filter_offset(filter_x, filters_x);
+
+ const InterpKernel *const filters_y = get_filter_base(filter_y);
+ const int y0_q4 = get_filter_offset(filter_y, filters_y);
+
+ highbd_convolve(src, src_stride, dst, dst_stride,
+ filters_x, x0_q4, x_step_q4,
+ filters_y, y0_q4, y_step_q4, w, h, bd);
+}
+
+void vpx_highbd_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h, int bd) {+ // Fixed size intermediate buffer places limits on parameters.
+ DECLARE_ALIGNED(16, uint16_t, temp[64 * 64]);
+ assert(w <= 64);
+ assert(h <= 64);
+
+ vpx_highbd_convolve8_c(src, src_stride, CONVERT_TO_BYTEPTR(temp), 64,
+ filter_x, x_step_q4, filter_y, y_step_q4, w, h, bd);
+ vpx_highbd_convolve_avg_c(CONVERT_TO_BYTEPTR(temp), 64, dst, dst_stride,
+ NULL, 0, NULL, 0, w, h, bd);
+}
+
+void vpx_highbd_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride,
+ uint8_t *dst8, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int filter_x_stride,
+ const int16_t *filter_y, int filter_y_stride,
+ int w, int h, int bd) {+ int r;
+ uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+ uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+ (void)filter_x;
+ (void)filter_y;
+ (void)filter_x_stride;
+ (void)filter_y_stride;
+ (void)bd;
+
+ for (r = h; r > 0; --r) {+ memcpy(dst, src, w * sizeof(uint16_t));
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+void vpx_highbd_convolve_avg_c(const uint8_t *src8, ptrdiff_t src_stride,
+ uint8_t *dst8, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int filter_x_stride,
+ const int16_t *filter_y, int filter_y_stride,
+ int w, int h, int bd) {+ int x, y;
+ uint16_t *src = CONVERT_TO_SHORTPTR(src8);
+ uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+ (void)filter_x;
+ (void)filter_y;
+ (void)filter_x_stride;
+ (void)filter_y_stride;
+ (void)bd;
+
+ for (y = 0; y < h; ++y) {+ for (x = 0; x < w; ++x) {+ dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
+ }
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+#endif
--- /dev/null
+++ b/vpx_dsp/vpx_convolve.h
@@ -1,0 +1,38 @@
+/*
+ * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef VPX_DSP_VPX_CONVOLVE_H_
+#define VPX_DSP_VPX_CONVOLVE_H_
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+
+#ifdef __cplusplus
+extern "C" {+#endif
+
+typedef void (*convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h);
+
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef void (*highbd_convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int x_step_q4,
+ const int16_t *filter_y, int y_step_q4,
+ int w, int h, int bd);
+#endif
+
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
+#endif // VPX_DSP_VPX_CONVOLVE_H_
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -54,6 +54,54 @@
DSP_SRCS-$(HAVE_DSPR2) += mips/intrapred16_dspr2.c
endif # CONFIG_VP9
+# interpolation filters
+DSP_SRCS-yes += vpx_convolve.c
+DSP_SRCS-yes += vpx_convolve.h
+DSP_SRCS-yes += vpx_filter.h
+
+DSP_SRCS-$(ARCH_X86)$(ARCH_X86_64) += x86/convolve.h
+DSP_SRCS-$(ARCH_X86)$(ARCH_X86_64) += x86/vpx_asm_stubs.c
+DSP_SRCS-$(HAVE_SSE2) += x86/vpx_subpixel_8t_sse2.asm
+DSP_SRCS-$(HAVE_SSE2) += x86/vpx_subpixel_bilinear_sse2.asm
+DSP_SRCS-$(HAVE_SSSE3) += x86/vpx_subpixel_8t_ssse3.asm
+DSP_SRCS-$(HAVE_SSSE3) += x86/vpx_subpixel_bilinear_ssse3.asm
+DSP_SRCS-$(HAVE_AVX2) += x86/vpx_subpixel_8t_intrin_avx2.c
+DSP_SRCS-$(HAVE_SSSE3) += x86/vpx_subpixel_8t_intrin_ssse3.c
+ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
+DSP_SRCS-$(HAVE_SSE2) += x86/vpx_high_subpixel_8t_sse2.asm
+DSP_SRCS-$(HAVE_SSE2) += x86/vpx_high_subpixel_bilinear_sse2.asm
+endif
+ifeq ($(CONFIG_USE_X86INC),yes)
+DSP_SRCS-$(HAVE_SSE2) += x86/vpx_convolve_copy_sse2.asm
+endif
+
+ifeq ($(HAVE_NEON_ASM),yes)
+DSP_SRCS-yes += arm/vpx_convolve_copy_neon_asm$(ASM)
+DSP_SRCS-yes += arm/vpx_convolve8_avg_neon_asm$(ASM)
+DSP_SRCS-yes += arm/vpx_convolve8_neon_asm$(ASM)
+DSP_SRCS-yes += arm/vpx_convolve_avg_neon_asm$(ASM)
+DSP_SRCS-yes += arm/vpx_convolve_neon.c
+else
+ifeq ($(HAVE_NEON),yes)
+DSP_SRCS-yes += arm/vpx_convolve_copy_neon.c
+DSP_SRCS-yes += arm/vpx_convolve8_avg_neon.c
+DSP_SRCS-yes += arm/vpx_convolve8_neon.c
+DSP_SRCS-yes += arm/vpx_convolve_avg_neon.c
+DSP_SRCS-yes += arm/vpx_convolve_neon.c
+endif # HAVE_NEON
+endif # HAVE_NEON_ASM
+
+# common (msa)
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve8_avg_horiz_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve8_avg_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve8_avg_vert_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve8_horiz_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve8_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve8_vert_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve_avg_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve_copy_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/vpx_convolve_msa.h
+
# loop filters
DSP_SRCS-yes += loopfilter.c
--- a/vpx_dsp/vpx_dsp_common.h
+++ b/vpx_dsp/vpx_dsp_common.h
@@ -11,8 +11,6 @@
#ifndef VPX_DSP_COMMON_H_
#define VPX_DSP_COMMON_H_
-#include <stdlib.h>
-
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#include "vpx_ports/mem.h"
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -34,6 +34,12 @@
}
}
+# optimizations which depend on multiple features
+$avx2_ssse3 = '';
+if ((vpx_config("HAVE_AVX2") eq "yes") && (vpx_config("HAVE_SSSE3") eq "yes")) {+ $avx2_ssse3 = 'avx2';
+}
+
# functions that are 64 bit only.
$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
if ($opts{arch} eq "x86_64") {@@ -364,6 +370,62 @@
specialize qw/vpx_highbd_dc_128_predictor_32x32/;
} # CONFIG_VP9_HIGHBITDEPTH
} # CONFIG_VP9
+
+#
+# Sub Pixel Filters
+#
+add_proto qw/void vpx_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+specialize qw/vpx_convolve_copy neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vpx_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+specialize qw/vpx_convolve_avg neon msa/, "$sse2_x86inc";
+
+add_proto qw/void vpx_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+specialize qw/vpx_convolve8 sse2 ssse3 neon msa/, "$avx2_ssse3";
+
+add_proto qw/void vpx_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+specialize qw/vpx_convolve8_horiz sse2 ssse3 neon msa/, "$avx2_ssse3";
+
+add_proto qw/void vpx_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+specialize qw/vpx_convolve8_vert sse2 ssse3 neon msa/, "$avx2_ssse3";
+
+add_proto qw/void vpx_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+specialize qw/vpx_convolve8_avg sse2 ssse3 neon msa/;
+
+add_proto qw/void vpx_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+specialize qw/vpx_convolve8_avg_horiz sse2 ssse3 neon msa/;
+
+add_proto qw/void vpx_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
+specialize qw/vpx_convolve8_avg_vert sse2 ssse3 neon msa/;
+
+if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {+ #
+ # Sub Pixel Filters
+ #
+ add_proto qw/void vpx_highbd_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+ specialize qw/vpx_highbd_convolve_copy/;
+
+ add_proto qw/void vpx_highbd_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+ specialize qw/vpx_highbd_convolve_avg/;
+
+ add_proto qw/void vpx_highbd_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+ specialize qw/vpx_highbd_convolve8/, "$sse2_x86_64";
+
+ add_proto qw/void vpx_highbd_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+ specialize qw/vpx_highbd_convolve8_horiz/, "$sse2_x86_64";
+
+ add_proto qw/void vpx_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+ specialize qw/vpx_highbd_convolve8_vert/, "$sse2_x86_64";
+
+ add_proto qw/void vpx_highbd_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+ specialize qw/vpx_highbd_convolve8_avg/, "$sse2_x86_64";
+
+ add_proto qw/void vpx_highbd_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+ specialize qw/vpx_highbd_convolve8_avg_horiz/, "$sse2_x86_64";
+
+ add_proto qw/void vpx_highbd_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
+ specialize qw/vpx_highbd_convolve8_avg_vert/, "$sse2_x86_64";
+} # CONFIG_VP9_HIGHBITDEPTH
#
# Loopfilter
--- /dev/null
+++ b/vpx_dsp/vpx_filter.h
@@ -1,0 +1,34 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VPX_DSP_VPX_FILTER_H_
+#define VPX_DSP_VPX_FILTER_H_
+
+#include "vpx/vpx_integer.h"
+
+
+#ifdef __cplusplus
+extern "C" {+#endif
+
+#define FILTER_BITS 7
+
+#define SUBPEL_BITS 4
+#define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1)
+#define SUBPEL_SHIFTS (1 << SUBPEL_BITS)
+#define SUBPEL_TAPS 8
+
+typedef int16_t InterpKernel[SUBPEL_TAPS];
+
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
+#endif // VPX_DSP_VPX_FILTER_H_
--- /dev/null
+++ b/vpx_dsp/x86/convolve.h
@@ -1,0 +1,296 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+#ifndef VPX_DSP_X86_CONVOLVE_H_
+#define VPX_DSP_X86_CONVOLVE_H_
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "vpx/vpx_integer.h"
+#include "vpx_ports/mem.h"
+
+typedef void filter8_1dfunction (
+ const uint8_t *src_ptr,
+ ptrdiff_t src_pitch,
+ uint8_t *output_ptr,
+ ptrdiff_t out_pitch,
+ uint32_t output_height,
+ const int16_t *filter
+);
+
+#define FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
+ void vpx_convolve8_##name##_##opt(const uint8_t *src, ptrdiff_t src_stride, \
+ uint8_t *dst, ptrdiff_t dst_stride, \
+ const int16_t *filter_x, int x_step_q4, \
+ const int16_t *filter_y, int y_step_q4, \
+ int w, int h) { \+ if (step_q4 == 16 && filter[3] != 128) { \+ if (filter[0] || filter[1] || filter[2]) { \+ while (w >= 16) { \+ vpx_filter_block1d16_##dir##8_##avg##opt(src_start, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter); \
+ src += 16; \
+ dst += 16; \
+ w -= 16; \
+ } \
+ while (w >= 8) { \+ vpx_filter_block1d8_##dir##8_##avg##opt(src_start, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter); \
+ src += 8; \
+ dst += 8; \
+ w -= 8; \
+ } \
+ while (w >= 4) { \+ vpx_filter_block1d4_##dir##8_##avg##opt(src_start, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter); \
+ src += 4; \
+ dst += 4; \
+ w -= 4; \
+ } \
+ } else { \+ while (w >= 16) { \+ vpx_filter_block1d16_##dir##2_##avg##opt(src, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter); \
+ src += 16; \
+ dst += 16; \
+ w -= 16; \
+ } \
+ while (w >= 8) { \+ vpx_filter_block1d8_##dir##2_##avg##opt(src, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter); \
+ src += 8; \
+ dst += 8; \
+ w -= 8; \
+ } \
+ while (w >= 4) { \+ vpx_filter_block1d4_##dir##2_##avg##opt(src, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter); \
+ src += 4; \
+ dst += 4; \
+ w -= 4; \
+ } \
+ } \
+ } \
+ if (w) { \+ vpx_convolve8_##name##_c(src, src_stride, dst, dst_stride, \
+ filter_x, x_step_q4, filter_y, y_step_q4, \
+ w, h); \
+ } \
+}
+
+#define FUN_CONV_2D(avg, opt) \
+void vpx_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \
+ uint8_t *dst, ptrdiff_t dst_stride, \
+ const int16_t *filter_x, int x_step_q4, \
+ const int16_t *filter_y, int y_step_q4, \
+ int w, int h) { \+ assert(w <= 64); \
+ assert(h <= 64); \
+ if (x_step_q4 == 16 && y_step_q4 == 16) { \+ if (filter_x[0] || filter_x[1] || filter_x[2] || filter_x[3] == 128 || \
+ filter_y[0] || filter_y[1] || filter_y[2] || filter_y[3] == 128) { \+ DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71]); \
+ vpx_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \
+ filter_x, x_step_q4, filter_y, y_step_q4, \
+ w, h + 7); \
+ vpx_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \
+ filter_x, x_step_q4, filter_y, \
+ y_step_q4, w, h); \
+ } else { \+ DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 65]); \
+ vpx_convolve8_horiz_##opt(src, src_stride, fdata2, 64, \
+ filter_x, x_step_q4, filter_y, y_step_q4, \
+ w, h + 1); \
+ vpx_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, \
+ filter_x, x_step_q4, filter_y, \
+ y_step_q4, w, h); \
+ } \
+ } else { \+ vpx_convolve8_##avg##c(src, src_stride, dst, dst_stride, \
+ filter_x, x_step_q4, filter_y, y_step_q4, w, h); \
+ } \
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+
+typedef void highbd_filter8_1dfunction (
+ const uint16_t *src_ptr,
+ const ptrdiff_t src_pitch,
+ uint16_t *output_ptr,
+ ptrdiff_t out_pitch,
+ unsigned int output_height,
+ const int16_t *filter,
+ int bd
+);
+
+#define HIGH_FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
+ void vpx_highbd_convolve8_##name##_##opt(const uint8_t *src8, \
+ ptrdiff_t src_stride, \
+ uint8_t *dst8, \
+ ptrdiff_t dst_stride, \
+ const int16_t *filter_x, \
+ int x_step_q4, \
+ const int16_t *filter_y, \
+ int y_step_q4, \
+ int w, int h, int bd) { \+ if (step_q4 == 16 && filter[3] != 128) { \+ uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
+ uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
+ if (filter[0] || filter[1] || filter[2]) { \+ while (w >= 16) { \+ vpx_highbd_filter_block1d16_##dir##8_##avg##opt(src_start, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter, \
+ bd); \
+ src += 16; \
+ dst += 16; \
+ w -= 16; \
+ } \
+ while (w >= 8) { \+ vpx_highbd_filter_block1d8_##dir##8_##avg##opt(src_start, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter, \
+ bd); \
+ src += 8; \
+ dst += 8; \
+ w -= 8; \
+ } \
+ while (w >= 4) { \+ vpx_highbd_filter_block1d4_##dir##8_##avg##opt(src_start, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter, \
+ bd); \
+ src += 4; \
+ dst += 4; \
+ w -= 4; \
+ } \
+ } else { \+ while (w >= 16) { \+ vpx_highbd_filter_block1d16_##dir##2_##avg##opt(src, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter, \
+ bd); \
+ src += 16; \
+ dst += 16; \
+ w -= 16; \
+ } \
+ while (w >= 8) { \+ vpx_highbd_filter_block1d8_##dir##2_##avg##opt(src, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter, \
+ bd); \
+ src += 8; \
+ dst += 8; \
+ w -= 8; \
+ } \
+ while (w >= 4) { \+ vpx_highbd_filter_block1d4_##dir##2_##avg##opt(src, \
+ src_stride, \
+ dst, \
+ dst_stride, \
+ h, \
+ filter, \
+ bd); \
+ src += 4; \
+ dst += 4; \
+ w -= 4; \
+ } \
+ } \
+ } \
+ if (w) { \+ vpx_highbd_convolve8_##name##_c(src8, src_stride, dst8, dst_stride, \
+ filter_x, x_step_q4, filter_y, y_step_q4, \
+ w, h, bd); \
+ } \
+}
+
+#define HIGH_FUN_CONV_2D(avg, opt) \
+void vpx_highbd_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \
+ uint8_t *dst, ptrdiff_t dst_stride, \
+ const int16_t *filter_x, int x_step_q4, \
+ const int16_t *filter_y, int y_step_q4, \
+ int w, int h, int bd) { \+ assert(w <= 64); \
+ assert(h <= 64); \
+ if (x_step_q4 == 16 && y_step_q4 == 16) { \+ if (filter_x[0] || filter_x[1] || filter_x[2] || filter_x[3] == 128 || \
+ filter_y[0] || filter_y[1] || filter_y[2] || filter_y[3] == 128) { \+ DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 71]); \
+ vpx_highbd_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \
+ CONVERT_TO_BYTEPTR(fdata2), 64, \
+ filter_x, x_step_q4, \
+ filter_y, y_step_q4, \
+ w, h + 7, bd); \
+ vpx_highbd_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2) + 192, \
+ 64, dst, dst_stride, \
+ filter_x, x_step_q4, \
+ filter_y, y_step_q4, \
+ w, h, bd); \
+ } else { \+ DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 65]); \
+ vpx_highbd_convolve8_horiz_##opt(src, src_stride, \
+ CONVERT_TO_BYTEPTR(fdata2), 64, \
+ filter_x, x_step_q4, \
+ filter_y, y_step_q4, \
+ w, h + 1, bd); \
+ vpx_highbd_convolve8_##avg##vert_##opt(CONVERT_TO_BYTEPTR(fdata2), 64, \
+ dst, dst_stride, \
+ filter_x, x_step_q4, \
+ filter_y, y_step_q4, \
+ w, h, bd); \
+ } \
+ } else { \+ vpx_highbd_convolve8_##avg##c(src, src_stride, dst, dst_stride, \
+ filter_x, x_step_q4, filter_y, y_step_q4, w, \
+ h, bd); \
+ } \
+}
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
+#endif // VPX_DSP_X86_CONVOLVE_H_
--- /dev/null
+++ b/vpx_dsp/x86/vpx_asm_stubs.c
@@ -1,0 +1,162 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/x86/convolve.h"
+
+#if HAVE_SSE2
+filter8_1dfunction vpx_filter_block1d16_v8_sse2;
+filter8_1dfunction vpx_filter_block1d16_h8_sse2;
+filter8_1dfunction vpx_filter_block1d8_v8_sse2;
+filter8_1dfunction vpx_filter_block1d8_h8_sse2;
+filter8_1dfunction vpx_filter_block1d4_v8_sse2;
+filter8_1dfunction vpx_filter_block1d4_h8_sse2;
+filter8_1dfunction vpx_filter_block1d16_v8_avg_sse2;
+filter8_1dfunction vpx_filter_block1d16_h8_avg_sse2;
+filter8_1dfunction vpx_filter_block1d8_v8_avg_sse2;
+filter8_1dfunction vpx_filter_block1d8_h8_avg_sse2;
+filter8_1dfunction vpx_filter_block1d4_v8_avg_sse2;
+filter8_1dfunction vpx_filter_block1d4_h8_avg_sse2;
+
+filter8_1dfunction vpx_filter_block1d16_v2_sse2;
+filter8_1dfunction vpx_filter_block1d16_h2_sse2;
+filter8_1dfunction vpx_filter_block1d8_v2_sse2;
+filter8_1dfunction vpx_filter_block1d8_h2_sse2;
+filter8_1dfunction vpx_filter_block1d4_v2_sse2;
+filter8_1dfunction vpx_filter_block1d4_h2_sse2;
+filter8_1dfunction vpx_filter_block1d16_v2_avg_sse2;
+filter8_1dfunction vpx_filter_block1d16_h2_avg_sse2;
+filter8_1dfunction vpx_filter_block1d8_v2_avg_sse2;
+filter8_1dfunction vpx_filter_block1d8_h2_avg_sse2;
+filter8_1dfunction vpx_filter_block1d4_v2_avg_sse2;
+filter8_1dfunction vpx_filter_block1d4_h2_avg_sse2;
+
+// void vpx_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_avg_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_avg_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
+FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
+FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
+FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, sse2);
+
+// void vpx_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+FUN_CONV_2D(, sse2);
+FUN_CONV_2D(avg_ , sse2);
+
+#if CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64
+highbd_filter8_1dfunction vpx_highbd_filter_block1d16_v8_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d16_h8_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d8_v8_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d8_h8_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v8_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h8_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d16_v8_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d16_h8_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d8_v8_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d8_h8_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v8_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h8_avg_sse2;
+
+highbd_filter8_1dfunction vpx_highbd_filter_block1d16_v2_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d16_h2_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d8_v2_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d8_h2_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v2_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h2_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d16_v2_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d16_h2_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d8_v2_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d8_h2_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v2_avg_sse2;
+highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h2_avg_sse2;
+
+// void vpx_highbd_convolve8_horiz_sse2(const uint8_t *src,
+// ptrdiff_t src_stride,
+// uint8_t *dst,
+// ptrdiff_t dst_stride,
+// const int16_t *filter_x,
+// int x_step_q4,
+// const int16_t *filter_y,
+// int y_step_q4,
+// int w, int h, int bd);
+// void vpx_highbd_convolve8_vert_sse2(const uint8_t *src,
+// ptrdiff_t src_stride,
+// uint8_t *dst,
+// ptrdiff_t dst_stride,
+// const int16_t *filter_x,
+// int x_step_q4,
+// const int16_t *filter_y,
+// int y_step_q4,
+// int w, int h, int bd);
+// void vpx_highbd_convolve8_avg_horiz_sse2(const uint8_t *src,
+// ptrdiff_t src_stride,
+// uint8_t *dst,
+// ptrdiff_t dst_stride,
+// const int16_t *filter_x,
+// int x_step_q4,
+// const int16_t *filter_y,
+// int y_step_q4,
+// int w, int h, int bd);
+// void vpx_highbd_convolve8_avg_vert_sse2(const uint8_t *src,
+// ptrdiff_t src_stride,
+// uint8_t *dst,
+// ptrdiff_t dst_stride,
+// const int16_t *filter_x,
+// int x_step_q4,
+// const int16_t *filter_y,
+// int y_step_q4,
+// int w, int h, int bd);
+HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
+HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
+HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
+HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
+ sse2);
+
+// void vpx_highbd_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h, int bd);
+// void vpx_highbd_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h, int bd);
+HIGH_FUN_CONV_2D(, sse2);
+HIGH_FUN_CONV_2D(avg_ , sse2);
+#endif // CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64
+#endif // HAVE_SSE2
--- /dev/null
+++ b/vpx_dsp/x86/vpx_convolve_copy_sse2.asm
@@ -1,0 +1,158 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+%define program_name vpx
+
+%include "third_party/x86inc/x86inc.asm"
+
+SECTION .text
+
+%macro convolve_fn 1
+INIT_XMM sse2
+cglobal convolve_%1, 4, 7, 4, src, src_stride, dst, dst_stride, \
+ fx, fxs, fy, fys, w, h
+ mov r4d, dword wm
+ cmp r4d, 4
+ je .w4
+ cmp r4d, 8
+ je .w8
+ cmp r4d, 16
+ je .w16
+ cmp r4d, 32
+ je .w32
+
+ mov r4d, dword hm
+.loop64:
+ movu m0, [srcq]
+ movu m1, [srcq+16]
+ movu m2, [srcq+32]
+ movu m3, [srcq+48]
+ add srcq, src_strideq
+%ifidn %1, avg
+ pavgb m0, [dstq]
+ pavgb m1, [dstq+16]
+ pavgb m2, [dstq+32]
+ pavgb m3, [dstq+48]
+%endif
+ mova [dstq ], m0
+ mova [dstq+16], m1
+ mova [dstq+32], m2
+ mova [dstq+48], m3
+ add dstq, dst_strideq
+ dec r4d
+ jnz .loop64
+ RET
+
+.w32:
+ mov r4d, dword hm
+.loop32:
+ movu m0, [srcq]
+ movu m1, [srcq+16]
+ movu m2, [srcq+src_strideq]
+ movu m3, [srcq+src_strideq+16]
+ lea srcq, [srcq+src_strideq*2]
+%ifidn %1, avg
+ pavgb m0, [dstq]
+ pavgb m1, [dstq +16]
+ pavgb m2, [dstq+dst_strideq]
+ pavgb m3, [dstq+dst_strideq+16]
+%endif
+ mova [dstq ], m0
+ mova [dstq +16], m1
+ mova [dstq+dst_strideq ], m2
+ mova [dstq+dst_strideq+16], m3
+ lea dstq, [dstq+dst_strideq*2]
+ sub r4d, 2
+ jnz .loop32
+ RET
+
+.w16:
+ mov r4d, dword hm
+ lea r5q, [src_strideq*3]
+ lea r6q, [dst_strideq*3]
+.loop16:
+ movu m0, [srcq]
+ movu m1, [srcq+src_strideq]
+ movu m2, [srcq+src_strideq*2]
+ movu m3, [srcq+r5q]
+ lea srcq, [srcq+src_strideq*4]
+%ifidn %1, avg
+ pavgb m0, [dstq]
+ pavgb m1, [dstq+dst_strideq]
+ pavgb m2, [dstq+dst_strideq*2]
+ pavgb m3, [dstq+r6q]
+%endif
+ mova [dstq ], m0
+ mova [dstq+dst_strideq ], m1
+ mova [dstq+dst_strideq*2], m2
+ mova [dstq+r6q ], m3
+ lea dstq, [dstq+dst_strideq*4]
+ sub r4d, 4
+ jnz .loop16
+ RET
+
+INIT_MMX sse
+.w8:
+ mov r4d, dword hm
+ lea r5q, [src_strideq*3]
+ lea r6q, [dst_strideq*3]
+.loop8:
+ movu m0, [srcq]
+ movu m1, [srcq+src_strideq]
+ movu m2, [srcq+src_strideq*2]
+ movu m3, [srcq+r5q]
+ lea srcq, [srcq+src_strideq*4]
+%ifidn %1, avg
+ pavgb m0, [dstq]
+ pavgb m1, [dstq+dst_strideq]
+ pavgb m2, [dstq+dst_strideq*2]
+ pavgb m3, [dstq+r6q]
+%endif
+ mova [dstq ], m0
+ mova [dstq+dst_strideq ], m1
+ mova [dstq+dst_strideq*2], m2
+ mova [dstq+r6q ], m3
+ lea dstq, [dstq+dst_strideq*4]
+ sub r4d, 4
+ jnz .loop8
+ RET
+
+.w4:
+ mov r4d, dword hm
+ lea r5q, [src_strideq*3]
+ lea r6q, [dst_strideq*3]
+.loop4:
+ movh m0, [srcq]
+ movh m1, [srcq+src_strideq]
+ movh m2, [srcq+src_strideq*2]
+ movh m3, [srcq+r5q]
+ lea srcq, [srcq+src_strideq*4]
+%ifidn %1, avg
+ movh m4, [dstq]
+ movh m5, [dstq+dst_strideq]
+ movh m6, [dstq+dst_strideq*2]
+ movh m7, [dstq+r6q]
+ pavgb m0, m4
+ pavgb m1, m5
+ pavgb m2, m6
+ pavgb m3, m7
+%endif
+ movh [dstq ], m0
+ movh [dstq+dst_strideq ], m1
+ movh [dstq+dst_strideq*2], m2
+ movh [dstq+r6q ], m3
+ lea dstq, [dstq+dst_strideq*4]
+ sub r4d, 4
+ jnz .loop4
+ RET
+%endmacro
+
+convolve_fn copy
+convolve_fn avg
--- /dev/null
+++ b/vpx_dsp/x86/vpx_high_subpixel_8t_sse2.asm
@@ -1,0 +1,962 @@
+;
+; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;Note: tap3 and tap4 have to be applied and added after other taps to avoid
+;overflow.
+
+%macro HIGH_GET_FILTERS_4 0
+ mov rdx, arg(5) ;filter ptr
+ mov rcx, 0x00000040
+
+ movdqa xmm7, [rdx] ;load filters
+ pshuflw xmm0, xmm7, 0b ;k0
+ pshuflw xmm1, xmm7, 01010101b ;k1
+ pshuflw xmm2, xmm7, 10101010b ;k2
+ pshuflw xmm3, xmm7, 11111111b ;k3
+ psrldq xmm7, 8
+ pshuflw xmm4, xmm7, 0b ;k4
+ pshuflw xmm5, xmm7, 01010101b ;k5
+ pshuflw xmm6, xmm7, 10101010b ;k6
+ pshuflw xmm7, xmm7, 11111111b ;k7
+
+ punpcklwd xmm0, xmm6
+ punpcklwd xmm2, xmm5
+ punpcklwd xmm3, xmm4
+ punpcklwd xmm1, xmm7
+
+ movdqa k0k6, xmm0
+ movdqa k2k5, xmm2
+ movdqa k3k4, xmm3
+ movdqa k1k7, xmm1
+
+ movq xmm6, rcx
+ pshufd xmm6, xmm6, 0
+ movdqa krd, xmm6
+
+ ;Compute max and min values of a pixel
+ mov rdx, 0x00010001
+ movsxd rcx, DWORD PTR arg(6) ;bps
+ movq xmm0, rdx
+ movq xmm1, rcx
+ pshufd xmm0, xmm0, 0b
+ movdqa xmm2, xmm0
+ psllw xmm0, xmm1
+ psubw xmm0, xmm2
+ pxor xmm1, xmm1
+ movdqa max, xmm0 ;max value (for clamping)
+ movdqa min, xmm1 ;min value (for clamping)
+
+%endm
+
+%macro HIGH_APPLY_FILTER_4 1
+ punpcklwd xmm0, xmm6 ;two row in one register
+ punpcklwd xmm1, xmm7
+ punpcklwd xmm2, xmm5
+ punpcklwd xmm3, xmm4
+
+ pmaddwd xmm0, k0k6 ;multiply the filter factors
+ pmaddwd xmm1, k1k7
+ pmaddwd xmm2, k2k5
+ pmaddwd xmm3, k3k4
+
+ paddd xmm0, xmm1 ;sum
+ paddd xmm0, xmm2
+ paddd xmm0, xmm3
+
+ paddd xmm0, krd ;rounding
+ psrad xmm0, 7 ;shift
+ packssdw xmm0, xmm0 ;pack to word
+
+ ;clamp the values
+ pminsw xmm0, max
+ pmaxsw xmm0, min
+
+%if %1
+ movq xmm1, [rdi]
+ pavgw xmm0, xmm1
+%endif
+ movq [rdi], xmm0
+%endm
+
+%macro HIGH_GET_FILTERS 0
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x00000040
+
+ movdqa xmm7, [rdx] ;load filters
+ pshuflw xmm0, xmm7, 0b ;k0
+ pshuflw xmm1, xmm7, 01010101b ;k1
+ pshuflw xmm2, xmm7, 10101010b ;k2
+ pshuflw xmm3, xmm7, 11111111b ;k3
+ pshufhw xmm4, xmm7, 0b ;k4
+ pshufhw xmm5, xmm7, 01010101b ;k5
+ pshufhw xmm6, xmm7, 10101010b ;k6
+ pshufhw xmm7, xmm7, 11111111b ;k7
+ punpcklqdq xmm2, xmm2
+ punpcklqdq xmm3, xmm3
+ punpcklwd xmm0, xmm1
+ punpckhwd xmm6, xmm7
+ punpckhwd xmm2, xmm5
+ punpckhwd xmm3, xmm4
+
+ movdqa k0k1, xmm0 ;store filter factors on stack
+ movdqa k6k7, xmm6
+ movdqa k2k5, xmm2
+ movdqa k3k4, xmm3
+
+ movq xmm6, rcx
+ pshufd xmm6, xmm6, 0
+ movdqa krd, xmm6 ;rounding
+
+ ;Compute max and min values of a pixel
+ mov rdx, 0x00010001
+ movsxd rcx, DWORD PTR arg(6) ;bps
+ movq xmm0, rdx
+ movq xmm1, rcx
+ pshufd xmm0, xmm0, 0b
+ movdqa xmm2, xmm0
+ psllw xmm0, xmm1
+ psubw xmm0, xmm2
+ pxor xmm1, xmm1
+ movdqa max, xmm0 ;max value (for clamping)
+ movdqa min, xmm1 ;min value (for clamping)
+%endm
+
+%macro LOAD_VERT_8 1
+ movdqu xmm0, [rsi + %1] ;0
+ movdqu xmm1, [rsi + rax + %1] ;1
+ movdqu xmm6, [rsi + rdx * 2 + %1] ;6
+ lea rsi, [rsi + rax]
+ movdqu xmm7, [rsi + rdx * 2 + %1] ;7
+ movdqu xmm2, [rsi + rax + %1] ;2
+ movdqu xmm3, [rsi + rax * 2 + %1] ;3
+ movdqu xmm4, [rsi + rdx + %1] ;4
+ movdqu xmm5, [rsi + rax * 4 + %1] ;5
+%endm
+
+%macro HIGH_APPLY_FILTER_8 2
+ movdqu temp, xmm4
+ movdqa xmm4, xmm0
+ punpcklwd xmm0, xmm1
+ punpckhwd xmm4, xmm1
+ movdqa xmm1, xmm6
+ punpcklwd xmm6, xmm7
+ punpckhwd xmm1, xmm7
+ movdqa xmm7, xmm2
+ punpcklwd xmm2, xmm5
+ punpckhwd xmm7, xmm5
+
+ movdqu xmm5, temp
+ movdqu temp, xmm4
+ movdqa xmm4, xmm3
+ punpcklwd xmm3, xmm5
+ punpckhwd xmm4, xmm5
+ movdqu xmm5, temp
+
+ pmaddwd xmm0, k0k1
+ pmaddwd xmm5, k0k1
+ pmaddwd xmm6, k6k7
+ pmaddwd xmm1, k6k7
+ pmaddwd xmm2, k2k5
+ pmaddwd xmm7, k2k5
+ pmaddwd xmm3, k3k4
+ pmaddwd xmm4, k3k4
+
+ paddd xmm0, xmm6
+ paddd xmm0, xmm2
+ paddd xmm0, xmm3
+ paddd xmm5, xmm1
+ paddd xmm5, xmm7
+ paddd xmm5, xmm4
+
+ paddd xmm0, krd ;rounding
+ paddd xmm5, krd
+ psrad xmm0, 7 ;shift
+ psrad xmm5, 7
+ packssdw xmm0, xmm5 ;pack back to word
+
+ ;clamp the values
+ pminsw xmm0, max
+ pmaxsw xmm0, min
+
+%if %1
+ movdqu xmm1, [rdi + %2]
+ pavgw xmm0, xmm1
+%endif
+ movdqu [rdi + %2], xmm0
+%endm
+
+;void vpx_filter_block1d4_v8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_highbd_filter_block1d4_v8_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d4_v8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 7
+ %define k0k6 [rsp + 16 * 0]
+ %define k2k5 [rsp + 16 * 1]
+ %define k3k4 [rsp + 16 * 2]
+ %define k1k7 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define max [rsp + 16 * 5]
+ %define min [rsp + 16 * 6]
+
+ HIGH_GET_FILTERS_4
+
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rbx, [rbx + rbx]
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movq xmm0, [rsi] ;load src: row 0
+ movq xmm1, [rsi + rax] ;1
+ movq xmm6, [rsi + rdx * 2] ;6
+ lea rsi, [rsi + rax]
+ movq xmm7, [rsi + rdx * 2] ;7
+ movq xmm2, [rsi + rax] ;2
+ movq xmm3, [rsi + rax * 2] ;3
+ movq xmm4, [rsi + rdx] ;4
+ movq xmm5, [rsi + rax * 4] ;5
+
+ HIGH_APPLY_FILTER_4 0
+
+ lea rdi, [rdi + rbx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 7
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d8_v8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_highbd_filter_block1d8_v8_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d8_v8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 8
+ %define k0k1 [rsp + 16 * 0]
+ %define k6k7 [rsp + 16 * 1]
+ %define k2k5 [rsp + 16 * 2]
+ %define k3k4 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define temp [rsp + 16 * 5]
+ %define max [rsp + 16 * 6]
+ %define min [rsp + 16 * 7]
+
+ HIGH_GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rbx, [rbx + rbx]
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ LOAD_VERT_8 0
+ HIGH_APPLY_FILTER_8 0, 0
+
+ lea rdi, [rdi + rbx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 8
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d16_v8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_highbd_filter_block1d16_v8_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d16_v8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 8
+ %define k0k1 [rsp + 16 * 0]
+ %define k6k7 [rsp + 16 * 1]
+ %define k2k5 [rsp + 16 * 2]
+ %define k3k4 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define temp [rsp + 16 * 5]
+ %define max [rsp + 16 * 6]
+ %define min [rsp + 16 * 7]
+
+ HIGH_GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rbx, [rbx + rbx]
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ LOAD_VERT_8 0
+ HIGH_APPLY_FILTER_8 0, 0
+ sub rsi, rax
+
+ LOAD_VERT_8 16
+ HIGH_APPLY_FILTER_8 0, 16
+ add rdi, rbx
+
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 8
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d4_v8_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d4_v8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 7
+ %define k0k6 [rsp + 16 * 0]
+ %define k2k5 [rsp + 16 * 1]
+ %define k3k4 [rsp + 16 * 2]
+ %define k1k7 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define max [rsp + 16 * 5]
+ %define min [rsp + 16 * 6]
+
+ HIGH_GET_FILTERS_4
+
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rbx, [rbx + rbx]
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movq xmm0, [rsi] ;load src: row 0
+ movq xmm1, [rsi + rax] ;1
+ movq xmm6, [rsi + rdx * 2] ;6
+ lea rsi, [rsi + rax]
+ movq xmm7, [rsi + rdx * 2] ;7
+ movq xmm2, [rsi + rax] ;2
+ movq xmm3, [rsi + rax * 2] ;3
+ movq xmm4, [rsi + rdx] ;4
+ movq xmm5, [rsi + rax * 4] ;5
+
+ HIGH_APPLY_FILTER_4 1
+
+ lea rdi, [rdi + rbx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 7
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d8_v8_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d8_v8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 8
+ %define k0k1 [rsp + 16 * 0]
+ %define k6k7 [rsp + 16 * 1]
+ %define k2k5 [rsp + 16 * 2]
+ %define k3k4 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define temp [rsp + 16 * 5]
+ %define max [rsp + 16 * 6]
+ %define min [rsp + 16 * 7]
+
+ HIGH_GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rbx, [rbx + rbx]
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+.loop:
+ LOAD_VERT_8 0
+ HIGH_APPLY_FILTER_8 1, 0
+
+ lea rdi, [rdi + rbx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 8
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d16_v8_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d16_v8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 8
+ %define k0k1 [rsp + 16 * 0]
+ %define k6k7 [rsp + 16 * 1]
+ %define k2k5 [rsp + 16 * 2]
+ %define k3k4 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define temp [rsp + 16 * 5]
+ %define max [rsp + 16 * 6]
+ %define min [rsp + 16 * 7]
+
+ HIGH_GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rbx, [rbx + rbx]
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+.loop:
+ LOAD_VERT_8 0
+ HIGH_APPLY_FILTER_8 1, 0
+ sub rsi, rax
+
+ LOAD_VERT_8 16
+ HIGH_APPLY_FILTER_8 1, 16
+ add rdi, rbx
+
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 8
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d4_h8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_highbd_filter_block1d4_h8_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d4_h8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 7
+ %define k0k6 [rsp + 16 * 0]
+ %define k2k5 [rsp + 16 * 1]
+ %define k3k4 [rsp + 16 * 2]
+ %define k1k7 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define max [rsp + 16 * 5]
+ %define min [rsp + 16 * 6]
+
+ HIGH_GET_FILTERS_4
+
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rdx, [rdx + rdx]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 6] ;load src
+ movdqu xmm4, [rsi + 2]
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm4
+ movdqa xmm7, xmm4
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm5, xmm4
+
+ psrldq xmm1, 2
+ psrldq xmm6, 4
+ psrldq xmm7, 6
+ psrldq xmm2, 4
+ psrldq xmm3, 6
+ psrldq xmm5, 2
+
+ HIGH_APPLY_FILTER_4 0
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 7
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d8_h8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_highbd_filter_block1d8_h8_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d8_h8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 8
+ %define k0k1 [rsp + 16 * 0]
+ %define k6k7 [rsp + 16 * 1]
+ %define k2k5 [rsp + 16 * 2]
+ %define k3k4 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define temp [rsp + 16 * 5]
+ %define max [rsp + 16 * 6]
+ %define min [rsp + 16 * 7]
+
+ HIGH_GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rdx, [rdx + rdx]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 6] ;load src
+ movdqu xmm1, [rsi - 4]
+ movdqu xmm2, [rsi - 2]
+ movdqu xmm3, [rsi]
+ movdqu xmm4, [rsi + 2]
+ movdqu xmm5, [rsi + 4]
+ movdqu xmm6, [rsi + 6]
+ movdqu xmm7, [rsi + 8]
+
+ HIGH_APPLY_FILTER_8 0, 0
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 8
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d16_h8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_highbd_filter_block1d16_h8_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d16_h8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 8
+ %define k0k1 [rsp + 16 * 0]
+ %define k6k7 [rsp + 16 * 1]
+ %define k2k5 [rsp + 16 * 2]
+ %define k3k4 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define temp [rsp + 16 * 5]
+ %define max [rsp + 16 * 6]
+ %define min [rsp + 16 * 7]
+
+ HIGH_GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rdx, [rdx + rdx]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 6] ;load src
+ movdqu xmm1, [rsi - 4]
+ movdqu xmm2, [rsi - 2]
+ movdqu xmm3, [rsi]
+ movdqu xmm4, [rsi + 2]
+ movdqu xmm5, [rsi + 4]
+ movdqu xmm6, [rsi + 6]
+ movdqu xmm7, [rsi + 8]
+
+ HIGH_APPLY_FILTER_8 0, 0
+
+ movdqu xmm0, [rsi + 10] ;load src
+ movdqu xmm1, [rsi + 12]
+ movdqu xmm2, [rsi + 14]
+ movdqu xmm3, [rsi + 16]
+ movdqu xmm4, [rsi + 18]
+ movdqu xmm5, [rsi + 20]
+ movdqu xmm6, [rsi + 22]
+ movdqu xmm7, [rsi + 24]
+
+ HIGH_APPLY_FILTER_8 0, 16
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 8
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d4_h8_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d4_h8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 7
+ %define k0k6 [rsp + 16 * 0]
+ %define k2k5 [rsp + 16 * 1]
+ %define k3k4 [rsp + 16 * 2]
+ %define k1k7 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define max [rsp + 16 * 5]
+ %define min [rsp + 16 * 6]
+
+ HIGH_GET_FILTERS_4
+
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rdx, [rdx + rdx]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 6] ;load src
+ movdqu xmm4, [rsi + 2]
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm4
+ movdqa xmm7, xmm4
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm5, xmm4
+
+ psrldq xmm1, 2
+ psrldq xmm6, 4
+ psrldq xmm7, 6
+ psrldq xmm2, 4
+ psrldq xmm3, 6
+ psrldq xmm5, 2
+
+ HIGH_APPLY_FILTER_4 1
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 7
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d8_h8_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d8_h8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 8
+ %define k0k1 [rsp + 16 * 0]
+ %define k6k7 [rsp + 16 * 1]
+ %define k2k5 [rsp + 16 * 2]
+ %define k3k4 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define temp [rsp + 16 * 5]
+ %define max [rsp + 16 * 6]
+ %define min [rsp + 16 * 7]
+
+ HIGH_GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rdx, [rdx + rdx]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 6] ;load src
+ movdqu xmm1, [rsi - 4]
+ movdqu xmm2, [rsi - 2]
+ movdqu xmm3, [rsi]
+ movdqu xmm4, [rsi + 2]
+ movdqu xmm5, [rsi + 4]
+ movdqu xmm6, [rsi + 6]
+ movdqu xmm7, [rsi + 8]
+
+ HIGH_APPLY_FILTER_8 1, 0
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 8
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d16_h8_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d16_h8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 8
+ %define k0k1 [rsp + 16 * 0]
+ %define k6k7 [rsp + 16 * 1]
+ %define k2k5 [rsp + 16 * 2]
+ %define k3k4 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define temp [rsp + 16 * 5]
+ %define max [rsp + 16 * 6]
+ %define min [rsp + 16 * 7]
+
+ HIGH_GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ lea rax, [rax + rax] ;bytes per line
+ lea rdx, [rdx + rdx]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 6] ;load src
+ movdqu xmm1, [rsi - 4]
+ movdqu xmm2, [rsi - 2]
+ movdqu xmm3, [rsi]
+ movdqu xmm4, [rsi + 2]
+ movdqu xmm5, [rsi + 4]
+ movdqu xmm6, [rsi + 6]
+ movdqu xmm7, [rsi + 8]
+
+ HIGH_APPLY_FILTER_8 1, 0
+
+ movdqu xmm0, [rsi + 10] ;load src
+ movdqu xmm1, [rsi + 12]
+ movdqu xmm2, [rsi + 14]
+ movdqu xmm3, [rsi + 16]
+ movdqu xmm4, [rsi + 18]
+ movdqu xmm5, [rsi + 20]
+ movdqu xmm6, [rsi + 22]
+ movdqu xmm7, [rsi + 24]
+
+ HIGH_APPLY_FILTER_8 1, 16
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 8
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
--- /dev/null
+++ b/vpx_dsp/x86/vpx_high_subpixel_bilinear_sse2.asm
@@ -1,0 +1,494 @@
+;
+; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro HIGH_GET_PARAM_4 0
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x00000040
+
+ movdqa xmm3, [rdx] ;load filters
+ pshuflw xmm4, xmm3, 11111111b ;k3
+ psrldq xmm3, 8
+ pshuflw xmm3, xmm3, 0b ;k4
+ punpcklwd xmm4, xmm3 ;k3k4
+
+ movq xmm3, rcx ;rounding
+ pshufd xmm3, xmm3, 0
+
+ mov rdx, 0x00010001
+ movsxd rcx, DWORD PTR arg(6) ;bps
+ movq xmm5, rdx
+ movq xmm2, rcx
+ pshufd xmm5, xmm5, 0b
+ movdqa xmm1, xmm5
+ psllw xmm5, xmm2
+ psubw xmm5, xmm1 ;max value (for clamping)
+ pxor xmm2, xmm2 ;min value (for clamping)
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+%endm
+
+%macro HIGH_APPLY_FILTER_4 1
+
+ punpcklwd xmm0, xmm1 ;two row in one register
+ pmaddwd xmm0, xmm4 ;multiply the filter factors
+
+ paddd xmm0, xmm3 ;rounding
+ psrad xmm0, 7 ;shift
+ packssdw xmm0, xmm0 ;pack to word
+
+ ;clamp the values
+ pminsw xmm0, xmm5
+ pmaxsw xmm0, xmm2
+
+%if %1
+ movq xmm1, [rdi]
+ pavgw xmm0, xmm1
+%endif
+
+ movq [rdi], xmm0
+ lea rsi, [rsi + 2*rax]
+ lea rdi, [rdi + 2*rdx]
+ dec rcx
+%endm
+
+%if ARCH_X86_64
+%macro HIGH_GET_PARAM 0
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x00000040
+
+ movdqa xmm6, [rdx] ;load filters
+
+ pshuflw xmm7, xmm6, 11111111b ;k3
+ pshufhw xmm6, xmm6, 0b ;k4
+ psrldq xmm6, 8
+ punpcklwd xmm7, xmm6 ;k3k4k3k4k3k4k3k4
+
+ movq xmm4, rcx ;rounding
+ pshufd xmm4, xmm4, 0
+
+ mov rdx, 0x00010001
+ movsxd rcx, DWORD PTR arg(6) ;bps
+ movq xmm8, rdx
+ movq xmm5, rcx
+ pshufd xmm8, xmm8, 0b
+ movdqa xmm1, xmm8
+ psllw xmm8, xmm5
+ psubw xmm8, xmm1 ;max value (for clamping)
+ pxor xmm5, xmm5 ;min value (for clamping)
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+%endm
+
+%macro HIGH_APPLY_FILTER_8 1
+ movdqa xmm6, xmm0
+ punpckhwd xmm6, xmm1
+ punpcklwd xmm0, xmm1
+ pmaddwd xmm6, xmm7
+ pmaddwd xmm0, xmm7
+
+ paddd xmm6, xmm4 ;rounding
+ paddd xmm0, xmm4 ;rounding
+ psrad xmm6, 7 ;shift
+ psrad xmm0, 7 ;shift
+ packssdw xmm0, xmm6 ;pack back to word
+
+ ;clamp the values
+ pminsw xmm0, xmm8
+ pmaxsw xmm0, xmm5
+
+%if %1
+ movdqu xmm1, [rdi]
+ pavgw xmm0, xmm1
+%endif
+ movdqu [rdi], xmm0 ;store the result
+
+ lea rsi, [rsi + 2*rax]
+ lea rdi, [rdi + 2*rdx]
+ dec rcx
+%endm
+
+%macro HIGH_APPLY_FILTER_16 1
+ movdqa xmm9, xmm0
+ movdqa xmm6, xmm2
+ punpckhwd xmm9, xmm1
+ punpckhwd xmm6, xmm3
+ punpcklwd xmm0, xmm1
+ punpcklwd xmm2, xmm3
+
+ pmaddwd xmm9, xmm7
+ pmaddwd xmm6, xmm7
+ pmaddwd xmm0, xmm7
+ pmaddwd xmm2, xmm7
+
+ paddd xmm9, xmm4 ;rounding
+ paddd xmm6, xmm4
+ paddd xmm0, xmm4
+ paddd xmm2, xmm4
+
+ psrad xmm9, 7 ;shift
+ psrad xmm6, 7
+ psrad xmm0, 7
+ psrad xmm2, 7
+
+ packssdw xmm0, xmm9 ;pack back to word
+ packssdw xmm2, xmm6 ;pack back to word
+
+ ;clamp the values
+ pminsw xmm0, xmm8
+ pmaxsw xmm0, xmm5
+ pminsw xmm2, xmm8
+ pmaxsw xmm2, xmm5
+
+%if %1
+ movdqu xmm1, [rdi]
+ movdqu xmm3, [rdi + 16]
+ pavgw xmm0, xmm1
+ pavgw xmm2, xmm3
+%endif
+ movdqu [rdi], xmm0 ;store the result
+ movdqu [rdi + 16], xmm2 ;store the result
+
+ lea rsi, [rsi + 2*rax]
+ lea rdi, [rdi + 2*rdx]
+ dec rcx
+%endm
+%endif
+
+global sym(vpx_highbd_filter_block1d4_v2_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d4_v2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM_4
+.loop:
+ movq xmm0, [rsi] ;load src
+ movq xmm1, [rsi + 2*rax]
+
+ HIGH_APPLY_FILTER_4 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+%if ARCH_X86_64
+global sym(vpx_highbd_filter_block1d8_v2_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d8_v2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 8
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;0
+ movdqu xmm1, [rsi + 2*rax] ;1
+
+ HIGH_APPLY_FILTER_8 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d16_v2_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d16_v2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 9
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;0
+ movdqu xmm2, [rsi + 16]
+ movdqu xmm1, [rsi + 2*rax] ;1
+ movdqu xmm3, [rsi + 2*rax + 16]
+
+ HIGH_APPLY_FILTER_16 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+%endif
+
+global sym(vpx_highbd_filter_block1d4_v2_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d4_v2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM_4
+.loop:
+ movq xmm0, [rsi] ;load src
+ movq xmm1, [rsi + 2*rax]
+
+ HIGH_APPLY_FILTER_4 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+%if ARCH_X86_64
+global sym(vpx_highbd_filter_block1d8_v2_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d8_v2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 8
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;0
+ movdqu xmm1, [rsi + 2*rax] ;1
+
+ HIGH_APPLY_FILTER_8 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d16_v2_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d16_v2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 9
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;0
+ movdqu xmm1, [rsi + 2*rax] ;1
+ movdqu xmm2, [rsi + 16]
+ movdqu xmm3, [rsi + 2*rax + 16]
+
+ HIGH_APPLY_FILTER_16 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+%endif
+
+global sym(vpx_highbd_filter_block1d4_h2_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d4_h2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM_4
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 2
+
+ HIGH_APPLY_FILTER_4 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+%if ARCH_X86_64
+global sym(vpx_highbd_filter_block1d8_h2_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d8_h2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 8
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqu xmm1, [rsi + 2]
+
+ HIGH_APPLY_FILTER_8 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d16_h2_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d16_h2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 9
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqu xmm1, [rsi + 2]
+ movdqu xmm2, [rsi + 16]
+ movdqu xmm3, [rsi + 18]
+
+ HIGH_APPLY_FILTER_16 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+%endif
+
+global sym(vpx_highbd_filter_block1d4_h2_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d4_h2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM_4
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 2
+
+ HIGH_APPLY_FILTER_4 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+%if ARCH_X86_64
+global sym(vpx_highbd_filter_block1d8_h2_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d8_h2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 8
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqu xmm1, [rsi + 2]
+
+ HIGH_APPLY_FILTER_8 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_highbd_filter_block1d16_h2_avg_sse2) PRIVATE
+sym(vpx_highbd_filter_block1d16_h2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 7
+ SAVE_XMM 9
+ push rsi
+ push rdi
+ ; end prolog
+
+ HIGH_GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqu xmm1, [rsi + 2]
+ movdqu xmm2, [rsi + 16]
+ movdqu xmm3, [rsi + 18]
+
+ HIGH_APPLY_FILTER_16 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+%endif
--- /dev/null
+++ b/vpx_dsp/x86/vpx_subpixel_8t_intrin_avx2.c
@@ -1,0 +1,602 @@
+/*
+ * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+// Due to a header conflict between math.h and intrinsics includes with ceil()
+// in certain configurations under vs9 this include needs to precede
+// immintrin.h.
+
+#include <immintrin.h>
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/x86/convolve.h"
+#include "vpx_ports/mem.h"
+
+// filters for 16_h8 and 16_v8
+DECLARE_ALIGNED(32, static const uint8_t, filt1_global_avx2[32]) = {+ 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8,
+ 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
+};
+
+DECLARE_ALIGNED(32, static const uint8_t, filt2_global_avx2[32]) = {+ 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10,
+ 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
+};
+
+DECLARE_ALIGNED(32, static const uint8_t, filt3_global_avx2[32]) = {+ 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12,
+ 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
+};
+
+DECLARE_ALIGNED(32, static const uint8_t, filt4_global_avx2[32]) = {+ 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14,
+ 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
+};
+
+#if defined(__clang__)
+# if __clang_major__ < 3 || (__clang_major__ == 3 && __clang_minor__ <= 3) || \
+ (defined(__APPLE__) && __clang_major__ == 5 && __clang_minor__ == 0)
+# define MM256_BROADCASTSI128_SI256(x) \
+ _mm_broadcastsi128_si256((__m128i const *)&(x))
+# else // clang > 3.3, and not 5.0 on macosx.
+# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
+# endif // clang <= 3.3
+#elif defined(__GNUC__)
+# if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ <= 6)
+# define MM256_BROADCASTSI128_SI256(x) \
+ _mm_broadcastsi128_si256((__m128i const *)&(x))
+# elif __GNUC__ == 4 && __GNUC_MINOR__ == 7
+# define MM256_BROADCASTSI128_SI256(x) _mm_broadcastsi128_si256(x)
+# else // gcc > 4.7
+# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
+# endif // gcc <= 4.6
+#else // !(gcc || clang)
+# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
+#endif // __clang__
+
+static void vpx_filter_block1d16_h8_avx2(const uint8_t *src_ptr,
+ ptrdiff_t src_pixels_per_line,
+ uint8_t *output_ptr,
+ ptrdiff_t output_pitch,
+ uint32_t output_height,
+ const int16_t *filter) {+ __m128i filtersReg;
+ __m256i addFilterReg64, filt1Reg, filt2Reg, filt3Reg, filt4Reg;
+ __m256i firstFilters, secondFilters, thirdFilters, forthFilters;
+ __m256i srcRegFilt32b1_1, srcRegFilt32b2_1, srcRegFilt32b2, srcRegFilt32b3;
+ __m256i srcReg32b1, srcReg32b2, filtersReg32;
+ unsigned int i;
+ ptrdiff_t src_stride, dst_stride;
+
+ // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+ addFilterReg64 = _mm256_set1_epi32((int)0x0400040u);
+ filtersReg = _mm_loadu_si128((const __m128i *)filter);
+ // converting the 16 bit (short) to 8 bit (byte) and have the same data
+ // in both lanes of 128 bit register.
+ filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+ // have the same data in both lanes of a 256 bit register
+ filtersReg32 = MM256_BROADCASTSI128_SI256(filtersReg);
+
+ // duplicate only the first 16 bits (first and second byte)
+ // across 256 bit register
+ firstFilters = _mm256_shuffle_epi8(filtersReg32,
+ _mm256_set1_epi16(0x100u));
+ // duplicate only the second 16 bits (third and forth byte)
+ // across 256 bit register
+ secondFilters = _mm256_shuffle_epi8(filtersReg32,
+ _mm256_set1_epi16(0x302u));
+ // duplicate only the third 16 bits (fifth and sixth byte)
+ // across 256 bit register
+ thirdFilters = _mm256_shuffle_epi8(filtersReg32,
+ _mm256_set1_epi16(0x504u));
+ // duplicate only the forth 16 bits (seventh and eighth byte)
+ // across 256 bit register
+ forthFilters = _mm256_shuffle_epi8(filtersReg32,
+ _mm256_set1_epi16(0x706u));
+
+ filt1Reg = _mm256_load_si256((__m256i const *)filt1_global_avx2);
+ filt2Reg = _mm256_load_si256((__m256i const *)filt2_global_avx2);
+ filt3Reg = _mm256_load_si256((__m256i const *)filt3_global_avx2);
+ filt4Reg = _mm256_load_si256((__m256i const *)filt4_global_avx2);
+
+ // multiple the size of the source and destination stride by two
+ src_stride = src_pixels_per_line << 1;
+ dst_stride = output_pitch << 1;
+ for (i = output_height; i > 1; i-=2) {+ // load the 2 strides of source
+ srcReg32b1 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr - 3)));
+ srcReg32b1 = _mm256_inserti128_si256(srcReg32b1,
+ _mm_loadu_si128((const __m128i *)
+ (src_ptr+src_pixels_per_line-3)), 1);
+
+ // filter the source buffer
+ srcRegFilt32b1_1= _mm256_shuffle_epi8(srcReg32b1, filt1Reg);
+ srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt4Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt32b1_1 = _mm256_maddubs_epi16(srcRegFilt32b1_1, firstFilters);
+ srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, forthFilters);
+
+ // add and saturate the results together
+ srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, srcRegFilt32b2);
+
+ // filter the source buffer
+ srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b1, filt2Reg);
+ srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b1, filt3Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, secondFilters);
+ srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, thirdFilters);
+
+ // add and saturate the results together
+ srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1,
+ _mm256_min_epi16(srcRegFilt32b3, srcRegFilt32b2));
+
+ // reading 2 strides of the next 16 bytes
+ // (part of it was being read by earlier read)
+ srcReg32b2 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + 5)));
+ srcReg32b2 = _mm256_inserti128_si256(srcReg32b2,
+ _mm_loadu_si128((const __m128i *)
+ (src_ptr+src_pixels_per_line+5)), 1);
+
+ // add and saturate the results together
+ srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1,
+ _mm256_max_epi16(srcRegFilt32b3, srcRegFilt32b2));
+
+ // filter the source buffer
+ srcRegFilt32b2_1 = _mm256_shuffle_epi8(srcReg32b2, filt1Reg);
+ srcRegFilt32b2 = _mm256_shuffle_epi8(srcReg32b2, filt4Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt32b2_1 = _mm256_maddubs_epi16(srcRegFilt32b2_1, firstFilters);
+ srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, forthFilters);
+
+ // add and saturate the results together
+ srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, srcRegFilt32b2);
+
+ // filter the source buffer
+ srcRegFilt32b3= _mm256_shuffle_epi8(srcReg32b2, filt2Reg);
+ srcRegFilt32b2= _mm256_shuffle_epi8(srcReg32b2, filt3Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt32b3 = _mm256_maddubs_epi16(srcRegFilt32b3, secondFilters);
+ srcRegFilt32b2 = _mm256_maddubs_epi16(srcRegFilt32b2, thirdFilters);
+
+ // add and saturate the results together
+ srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1,
+ _mm256_min_epi16(srcRegFilt32b3, srcRegFilt32b2));
+ srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1,
+ _mm256_max_epi16(srcRegFilt32b3, srcRegFilt32b2));
+
+
+ srcRegFilt32b1_1 = _mm256_adds_epi16(srcRegFilt32b1_1, addFilterReg64);
+
+ srcRegFilt32b2_1 = _mm256_adds_epi16(srcRegFilt32b2_1, addFilterReg64);
+
+ // shift by 7 bit each 16 bit
+ srcRegFilt32b1_1 = _mm256_srai_epi16(srcRegFilt32b1_1, 7);
+ srcRegFilt32b2_1 = _mm256_srai_epi16(srcRegFilt32b2_1, 7);
+
+ // shrink to 8 bit each 16 bits, the first lane contain the first
+ // convolve result and the second lane contain the second convolve
+ // result
+ srcRegFilt32b1_1 = _mm256_packus_epi16(srcRegFilt32b1_1,
+ srcRegFilt32b2_1);
+
+ src_ptr+=src_stride;
+
+ // save 16 bytes
+ _mm_store_si128((__m128i*)output_ptr,
+ _mm256_castsi256_si128(srcRegFilt32b1_1));
+
+ // save the next 16 bits
+ _mm_store_si128((__m128i*)(output_ptr+output_pitch),
+ _mm256_extractf128_si256(srcRegFilt32b1_1, 1));
+ output_ptr+=dst_stride;
+ }
+
+ // if the number of strides is odd.
+ // process only 16 bytes
+ if (i > 0) {+ __m128i srcReg1, srcReg2, srcRegFilt1_1, srcRegFilt2_1;
+ __m128i srcRegFilt2, srcRegFilt3;
+
+ srcReg1 = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
+
+ // filter the source buffer
+ srcRegFilt1_1 = _mm_shuffle_epi8(srcReg1,
+ _mm256_castsi256_si128(filt1Reg));
+ srcRegFilt2 = _mm_shuffle_epi8(srcReg1,
+ _mm256_castsi256_si128(filt4Reg));
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt1_1 = _mm_maddubs_epi16(srcRegFilt1_1,
+ _mm256_castsi256_si128(firstFilters));
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
+ _mm256_castsi256_si128(forthFilters));
+
+ // add and saturate the results together
+ srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, srcRegFilt2);
+
+ // filter the source buffer
+ srcRegFilt3= _mm_shuffle_epi8(srcReg1,
+ _mm256_castsi256_si128(filt2Reg));
+ srcRegFilt2= _mm_shuffle_epi8(srcReg1,
+ _mm256_castsi256_si128(filt3Reg));
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3,
+ _mm256_castsi256_si128(secondFilters));
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
+ _mm256_castsi256_si128(thirdFilters));
+
+ // add and saturate the results together
+ srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
+ _mm_min_epi16(srcRegFilt3, srcRegFilt2));
+
+ // reading the next 16 bytes
+ // (part of it was being read by earlier read)
+ srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + 5));
+
+ // add and saturate the results together
+ srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
+ _mm_max_epi16(srcRegFilt3, srcRegFilt2));
+
+ // filter the source buffer
+ srcRegFilt2_1 = _mm_shuffle_epi8(srcReg2,
+ _mm256_castsi256_si128(filt1Reg));
+ srcRegFilt2 = _mm_shuffle_epi8(srcReg2,
+ _mm256_castsi256_si128(filt4Reg));
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt2_1 = _mm_maddubs_epi16(srcRegFilt2_1,
+ _mm256_castsi256_si128(firstFilters));
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
+ _mm256_castsi256_si128(forthFilters));
+
+ // add and saturate the results together
+ srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, srcRegFilt2);
+
+ // filter the source buffer
+ srcRegFilt3 = _mm_shuffle_epi8(srcReg2,
+ _mm256_castsi256_si128(filt2Reg));
+ srcRegFilt2 = _mm_shuffle_epi8(srcReg2,
+ _mm256_castsi256_si128(filt3Reg));
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3,
+ _mm256_castsi256_si128(secondFilters));
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2,
+ _mm256_castsi256_si128(thirdFilters));
+
+ // add and saturate the results together
+ srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
+ _mm_min_epi16(srcRegFilt3, srcRegFilt2));
+ srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
+ _mm_max_epi16(srcRegFilt3, srcRegFilt2));
+
+
+ srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
+ _mm256_castsi256_si128(addFilterReg64));
+
+ srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
+ _mm256_castsi256_si128(addFilterReg64));
+
+ // shift by 7 bit each 16 bit
+ srcRegFilt1_1 = _mm_srai_epi16(srcRegFilt1_1, 7);
+ srcRegFilt2_1 = _mm_srai_epi16(srcRegFilt2_1, 7);
+
+ // shrink to 8 bit each 16 bits, the first lane contain the first
+ // convolve result and the second lane contain the second convolve
+ // result
+ srcRegFilt1_1 = _mm_packus_epi16(srcRegFilt1_1, srcRegFilt2_1);
+
+ // save 16 bytes
+ _mm_store_si128((__m128i*)output_ptr, srcRegFilt1_1);
+ }
+}
+
+static void vpx_filter_block1d16_v8_avx2(const uint8_t *src_ptr,
+ ptrdiff_t src_pitch,
+ uint8_t *output_ptr,
+ ptrdiff_t out_pitch,
+ uint32_t output_height,
+ const int16_t *filter) {+ __m128i filtersReg;
+ __m256i addFilterReg64;
+ __m256i srcReg32b1, srcReg32b2, srcReg32b3, srcReg32b4, srcReg32b5;
+ __m256i srcReg32b6, srcReg32b7, srcReg32b8, srcReg32b9, srcReg32b10;
+ __m256i srcReg32b11, srcReg32b12, filtersReg32;
+ __m256i firstFilters, secondFilters, thirdFilters, forthFilters;
+ unsigned int i;
+ ptrdiff_t src_stride, dst_stride;
+
+ // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+ addFilterReg64 = _mm256_set1_epi32((int)0x0400040u);
+ filtersReg = _mm_loadu_si128((const __m128i *)filter);
+ // converting the 16 bit (short) to 8 bit (byte) and have the
+ // same data in both lanes of 128 bit register.
+ filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+ // have the same data in both lanes of a 256 bit register
+ filtersReg32 = MM256_BROADCASTSI128_SI256(filtersReg);
+
+ // duplicate only the first 16 bits (first and second byte)
+ // across 256 bit register
+ firstFilters = _mm256_shuffle_epi8(filtersReg32,
+ _mm256_set1_epi16(0x100u));
+ // duplicate only the second 16 bits (third and forth byte)
+ // across 256 bit register
+ secondFilters = _mm256_shuffle_epi8(filtersReg32,
+ _mm256_set1_epi16(0x302u));
+ // duplicate only the third 16 bits (fifth and sixth byte)
+ // across 256 bit register
+ thirdFilters = _mm256_shuffle_epi8(filtersReg32,
+ _mm256_set1_epi16(0x504u));
+ // duplicate only the forth 16 bits (seventh and eighth byte)
+ // across 256 bit register
+ forthFilters = _mm256_shuffle_epi8(filtersReg32,
+ _mm256_set1_epi16(0x706u));
+
+ // multiple the size of the source and destination stride by two
+ src_stride = src_pitch << 1;
+ dst_stride = out_pitch << 1;
+
+ // load 16 bytes 7 times in stride of src_pitch
+ srcReg32b1 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr)));
+ srcReg32b2 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch)));
+ srcReg32b3 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2)));
+ srcReg32b4 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3)));
+ srcReg32b5 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4)));
+ srcReg32b6 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5)));
+ srcReg32b7 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6)));
+
+ // have each consecutive loads on the same 256 register
+ srcReg32b1 = _mm256_inserti128_si256(srcReg32b1,
+ _mm256_castsi256_si128(srcReg32b2), 1);
+ srcReg32b2 = _mm256_inserti128_si256(srcReg32b2,
+ _mm256_castsi256_si128(srcReg32b3), 1);
+ srcReg32b3 = _mm256_inserti128_si256(srcReg32b3,
+ _mm256_castsi256_si128(srcReg32b4), 1);
+ srcReg32b4 = _mm256_inserti128_si256(srcReg32b4,
+ _mm256_castsi256_si128(srcReg32b5), 1);
+ srcReg32b5 = _mm256_inserti128_si256(srcReg32b5,
+ _mm256_castsi256_si128(srcReg32b6), 1);
+ srcReg32b6 = _mm256_inserti128_si256(srcReg32b6,
+ _mm256_castsi256_si128(srcReg32b7), 1);
+
+ // merge every two consecutive registers except the last one
+ srcReg32b10 = _mm256_unpacklo_epi8(srcReg32b1, srcReg32b2);
+ srcReg32b1 = _mm256_unpackhi_epi8(srcReg32b1, srcReg32b2);
+
+ // save
+ srcReg32b11 = _mm256_unpacklo_epi8(srcReg32b3, srcReg32b4);
+
+ // save
+ srcReg32b3 = _mm256_unpackhi_epi8(srcReg32b3, srcReg32b4);
+
+ // save
+ srcReg32b2 = _mm256_unpacklo_epi8(srcReg32b5, srcReg32b6);
+
+ // save
+ srcReg32b5 = _mm256_unpackhi_epi8(srcReg32b5, srcReg32b6);
+
+
+ for (i = output_height; i > 1; i-=2) {+ // load the last 2 loads of 16 bytes and have every two
+ // consecutive loads in the same 256 bit register
+ srcReg32b8 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7)));
+ srcReg32b7 = _mm256_inserti128_si256(srcReg32b7,
+ _mm256_castsi256_si128(srcReg32b8), 1);
+ srcReg32b9 = _mm256_castsi128_si256(
+ _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 8)));
+ srcReg32b8 = _mm256_inserti128_si256(srcReg32b8,
+ _mm256_castsi256_si128(srcReg32b9), 1);
+
+ // merge every two consecutive registers
+ // save
+ srcReg32b4 = _mm256_unpacklo_epi8(srcReg32b7, srcReg32b8);
+ srcReg32b7 = _mm256_unpackhi_epi8(srcReg32b7, srcReg32b8);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcReg32b10 = _mm256_maddubs_epi16(srcReg32b10, firstFilters);
+ srcReg32b6 = _mm256_maddubs_epi16(srcReg32b4, forthFilters);
+
+ // add and saturate the results together
+ srcReg32b10 = _mm256_adds_epi16(srcReg32b10, srcReg32b6);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcReg32b8 = _mm256_maddubs_epi16(srcReg32b11, secondFilters);
+ srcReg32b12 = _mm256_maddubs_epi16(srcReg32b2, thirdFilters);
+
+ // add and saturate the results together
+ srcReg32b10 = _mm256_adds_epi16(srcReg32b10,
+ _mm256_min_epi16(srcReg32b8, srcReg32b12));
+ srcReg32b10 = _mm256_adds_epi16(srcReg32b10,
+ _mm256_max_epi16(srcReg32b8, srcReg32b12));
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcReg32b1 = _mm256_maddubs_epi16(srcReg32b1, firstFilters);
+ srcReg32b6 = _mm256_maddubs_epi16(srcReg32b7, forthFilters);
+
+ srcReg32b1 = _mm256_adds_epi16(srcReg32b1, srcReg32b6);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcReg32b8 = _mm256_maddubs_epi16(srcReg32b3, secondFilters);
+ srcReg32b12 = _mm256_maddubs_epi16(srcReg32b5, thirdFilters);
+
+ // add and saturate the results together
+ srcReg32b1 = _mm256_adds_epi16(srcReg32b1,
+ _mm256_min_epi16(srcReg32b8, srcReg32b12));
+ srcReg32b1 = _mm256_adds_epi16(srcReg32b1,
+ _mm256_max_epi16(srcReg32b8, srcReg32b12));
+
+ srcReg32b10 = _mm256_adds_epi16(srcReg32b10, addFilterReg64);
+ srcReg32b1 = _mm256_adds_epi16(srcReg32b1, addFilterReg64);
+
+ // shift by 7 bit each 16 bit
+ srcReg32b10 = _mm256_srai_epi16(srcReg32b10, 7);
+ srcReg32b1 = _mm256_srai_epi16(srcReg32b1, 7);
+
+ // shrink to 8 bit each 16 bits, the first lane contain the first
+ // convolve result and the second lane contain the second convolve
+ // result
+ srcReg32b1 = _mm256_packus_epi16(srcReg32b10, srcReg32b1);
+
+ src_ptr+=src_stride;
+
+ // save 16 bytes
+ _mm_store_si128((__m128i*)output_ptr,
+ _mm256_castsi256_si128(srcReg32b1));
+
+ // save the next 16 bits
+ _mm_store_si128((__m128i*)(output_ptr+out_pitch),
+ _mm256_extractf128_si256(srcReg32b1, 1));
+
+ output_ptr+=dst_stride;
+
+ // save part of the registers for next strides
+ srcReg32b10 = srcReg32b11;
+ srcReg32b1 = srcReg32b3;
+ srcReg32b11 = srcReg32b2;
+ srcReg32b3 = srcReg32b5;
+ srcReg32b2 = srcReg32b4;
+ srcReg32b5 = srcReg32b7;
+ srcReg32b7 = srcReg32b9;
+ }
+ if (i > 0) {+ __m128i srcRegFilt1, srcRegFilt3, srcRegFilt4, srcRegFilt5;
+ __m128i srcRegFilt6, srcRegFilt7, srcRegFilt8;
+ // load the last 16 bytes
+ srcRegFilt8 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7));
+
+ // merge the last 2 results together
+ srcRegFilt4 = _mm_unpacklo_epi8(
+ _mm256_castsi256_si128(srcReg32b7), srcRegFilt8);
+ srcRegFilt7 = _mm_unpackhi_epi8(
+ _mm256_castsi256_si128(srcReg32b7), srcRegFilt8);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt1 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b10),
+ _mm256_castsi256_si128(firstFilters));
+ srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4,
+ _mm256_castsi256_si128(forthFilters));
+ srcRegFilt3 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b1),
+ _mm256_castsi256_si128(firstFilters));
+ srcRegFilt7 = _mm_maddubs_epi16(srcRegFilt7,
+ _mm256_castsi256_si128(forthFilters));
+
+ // add and saturate the results together
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
+ srcRegFilt3 = _mm_adds_epi16(srcRegFilt3, srcRegFilt7);
+
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt4 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b11),
+ _mm256_castsi256_si128(secondFilters));
+ srcRegFilt5 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b3),
+ _mm256_castsi256_si128(secondFilters));
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt6 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b2),
+ _mm256_castsi256_si128(thirdFilters));
+ srcRegFilt7 = _mm_maddubs_epi16(_mm256_castsi256_si128(srcReg32b5),
+ _mm256_castsi256_si128(thirdFilters));
+
+ // add and saturate the results together
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
+ _mm_min_epi16(srcRegFilt4, srcRegFilt6));
+ srcRegFilt3 = _mm_adds_epi16(srcRegFilt3,
+ _mm_min_epi16(srcRegFilt5, srcRegFilt7));
+
+ // add and saturate the results together
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
+ _mm_max_epi16(srcRegFilt4, srcRegFilt6));
+ srcRegFilt3 = _mm_adds_epi16(srcRegFilt3,
+ _mm_max_epi16(srcRegFilt5, srcRegFilt7));
+
+
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
+ _mm256_castsi256_si128(addFilterReg64));
+ srcRegFilt3 = _mm_adds_epi16(srcRegFilt3,
+ _mm256_castsi256_si128(addFilterReg64));
+
+ // shift by 7 bit each 16 bit
+ srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+ srcRegFilt3 = _mm_srai_epi16(srcRegFilt3, 7);
+
+ // shrink to 8 bit each 16 bits, the first lane contain the first
+ // convolve result and the second lane contain the second convolve
+ // result
+ srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt3);
+
+ // save 16 bytes
+ _mm_store_si128((__m128i*)output_ptr, srcRegFilt1);
+ }
+}
+
+#if HAVE_AVX2 && HAVE_SSSE3
+filter8_1dfunction vpx_filter_block1d4_v8_ssse3;
+#if ARCH_X86_64
+filter8_1dfunction vpx_filter_block1d8_v8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h8_intrin_ssse3;
+#define vpx_filter_block1d8_v8_avx2 vpx_filter_block1d8_v8_intrin_ssse3
+#define vpx_filter_block1d8_h8_avx2 vpx_filter_block1d8_h8_intrin_ssse3
+#define vpx_filter_block1d4_h8_avx2 vpx_filter_block1d4_h8_intrin_ssse3
+#else // ARCH_X86
+filter8_1dfunction vpx_filter_block1d8_v8_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h8_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h8_ssse3;
+#define vpx_filter_block1d8_v8_avx2 vpx_filter_block1d8_v8_ssse3
+#define vpx_filter_block1d8_h8_avx2 vpx_filter_block1d8_h8_ssse3
+#define vpx_filter_block1d4_h8_avx2 vpx_filter_block1d4_h8_ssse3
+#endif // ARCH_X86_64
+filter8_1dfunction vpx_filter_block1d16_v2_ssse3;
+filter8_1dfunction vpx_filter_block1d16_h2_ssse3;
+filter8_1dfunction vpx_filter_block1d8_v2_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h2_ssse3;
+filter8_1dfunction vpx_filter_block1d4_v2_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h2_ssse3;
+#define vpx_filter_block1d4_v8_avx2 vpx_filter_block1d4_v8_ssse3
+#define vpx_filter_block1d16_v2_avx2 vpx_filter_block1d16_v2_ssse3
+#define vpx_filter_block1d16_h2_avx2 vpx_filter_block1d16_h2_ssse3
+#define vpx_filter_block1d8_v2_avx2 vpx_filter_block1d8_v2_ssse3
+#define vpx_filter_block1d8_h2_avx2 vpx_filter_block1d8_h2_ssse3
+#define vpx_filter_block1d4_v2_avx2 vpx_filter_block1d4_v2_ssse3
+#define vpx_filter_block1d4_h2_avx2 vpx_filter_block1d4_h2_ssse3
+// void vpx_convolve8_horiz_avx2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_vert_avx2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , avx2);
+FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , avx2);
+
+// void vpx_convolve8_avx2(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+FUN_CONV_2D(, avx2);
+#endif // HAVE_AX2 && HAVE_SSSE3
--- /dev/null
+++ b/vpx_dsp/x86/vpx_subpixel_8t_intrin_ssse3.c
@@ -1,0 +1,601 @@
+/*
+ * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+// Due to a header conflict between math.h and intrinsics includes with ceil()
+// in certain configurations under vs9 this include needs to precede
+// tmmintrin.h.
+
+#include <tmmintrin.h>
+
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/x86/convolve.h"
+#include "vpx_ports/mem.h"
+#include "vpx_ports/emmintrin_compat.h"
+
+// filters only for the 4_h8 convolution
+DECLARE_ALIGNED(16, static const uint8_t, filt1_4_h8[16]) = {+ 0, 1, 1, 2, 2, 3, 3, 4, 2, 3, 3, 4, 4, 5, 5, 6
+};
+
+DECLARE_ALIGNED(16, static const uint8_t, filt2_4_h8[16]) = {+ 4, 5, 5, 6, 6, 7, 7, 8, 6, 7, 7, 8, 8, 9, 9, 10
+};
+
+// filters for 8_h8 and 16_h8
+DECLARE_ALIGNED(16, static const uint8_t, filt1_global[16]) = {+ 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
+};
+
+DECLARE_ALIGNED(16, static const uint8_t, filt2_global[16]) = {+ 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
+};
+
+DECLARE_ALIGNED(16, static const uint8_t, filt3_global[16]) = {+ 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
+};
+
+DECLARE_ALIGNED(16, static const uint8_t, filt4_global[16]) = {+ 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
+};
+
+// These are reused by the avx2 intrinsics.
+filter8_1dfunction vpx_filter_block1d8_v8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h8_intrin_ssse3;
+
+void vpx_filter_block1d4_h8_intrin_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_pixels_per_line,
+ uint8_t *output_ptr,
+ ptrdiff_t output_pitch,
+ uint32_t output_height,
+ const int16_t *filter) {+ __m128i firstFilters, secondFilters, shuffle1, shuffle2;
+ __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4;
+ __m128i addFilterReg64, filtersReg, srcReg, minReg;
+ unsigned int i;
+
+ // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+ addFilterReg64 =_mm_set1_epi32((int)0x0400040u);
+ filtersReg = _mm_loadu_si128((const __m128i *)filter);
+ // converting the 16 bit (short) to 8 bit (byte) and have the same data
+ // in both lanes of 128 bit register.
+ filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+ // duplicate only the first 16 bits in the filter into the first lane
+ firstFilters = _mm_shufflelo_epi16(filtersReg, 0);
+ // duplicate only the third 16 bit in the filter into the first lane
+ secondFilters = _mm_shufflelo_epi16(filtersReg, 0xAAu);
+ // duplicate only the seconds 16 bits in the filter into the second lane
+ // firstFilters: k0 k1 k0 k1 k0 k1 k0 k1 k2 k3 k2 k3 k2 k3 k2 k3
+ firstFilters = _mm_shufflehi_epi16(firstFilters, 0x55u);
+ // duplicate only the forth 16 bits in the filter into the second lane
+ // secondFilters: k4 k5 k4 k5 k4 k5 k4 k5 k6 k7 k6 k7 k6 k7 k6 k7
+ secondFilters = _mm_shufflehi_epi16(secondFilters, 0xFFu);
+
+ // loading the local filters
+ shuffle1 =_mm_load_si128((__m128i const *)filt1_4_h8);
+ shuffle2 = _mm_load_si128((__m128i const *)filt2_4_h8);
+
+ for (i = 0; i < output_height; i++) {+ srcReg = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
+
+ // filter the source buffer
+ srcRegFilt1= _mm_shuffle_epi8(srcReg, shuffle1);
+ srcRegFilt2= _mm_shuffle_epi8(srcReg, shuffle2);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters);
+
+ // extract the higher half of the lane
+ srcRegFilt3 = _mm_srli_si128(srcRegFilt1, 8);
+ srcRegFilt4 = _mm_srli_si128(srcRegFilt2, 8);
+
+ minReg = _mm_min_epi16(srcRegFilt3, srcRegFilt2);
+
+ // add and saturate all the results together
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
+ srcRegFilt3 = _mm_max_epi16(srcRegFilt3, srcRegFilt2);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt3);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+ // shift by 7 bit each 16 bits
+ srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+ // shrink to 8 bit each 16 bits
+ srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+ src_ptr+=src_pixels_per_line;
+
+ // save only 4 bytes
+ *((int*)&output_ptr[0])= _mm_cvtsi128_si32(srcRegFilt1);
+
+ output_ptr+=output_pitch;
+ }
+}
+
+void vpx_filter_block1d8_h8_intrin_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_pixels_per_line,
+ uint8_t *output_ptr,
+ ptrdiff_t output_pitch,
+ uint32_t output_height,
+ const int16_t *filter) {+ __m128i firstFilters, secondFilters, thirdFilters, forthFilters, srcReg;
+ __m128i filt1Reg, filt2Reg, filt3Reg, filt4Reg;
+ __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt4;
+ __m128i addFilterReg64, filtersReg, minReg;
+ unsigned int i;
+
+ // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+ addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+ filtersReg = _mm_loadu_si128((const __m128i *)filter);
+ // converting the 16 bit (short) to 8 bit (byte) and have the same data
+ // in both lanes of 128 bit register.
+ filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+ // duplicate only the first 16 bits (first and second byte)
+ // across 128 bit register
+ firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
+ // duplicate only the second 16 bits (third and forth byte)
+ // across 128 bit register
+ secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+ // duplicate only the third 16 bits (fifth and sixth byte)
+ // across 128 bit register
+ thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+ // duplicate only the forth 16 bits (seventh and eighth byte)
+ // across 128 bit register
+ forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+
+ filt1Reg = _mm_load_si128((__m128i const *)filt1_global);
+ filt2Reg = _mm_load_si128((__m128i const *)filt2_global);
+ filt3Reg = _mm_load_si128((__m128i const *)filt3_global);
+ filt4Reg = _mm_load_si128((__m128i const *)filt4_global);
+
+ for (i = 0; i < output_height; i++) {+ srcReg = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
+
+ // filter the source buffer
+ srcRegFilt1= _mm_shuffle_epi8(srcReg, filt1Reg);
+ srcRegFilt2= _mm_shuffle_epi8(srcReg, filt2Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, secondFilters);
+
+ // filter the source buffer
+ srcRegFilt3= _mm_shuffle_epi8(srcReg, filt3Reg);
+ srcRegFilt4= _mm_shuffle_epi8(srcReg, filt4Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, thirdFilters);
+ srcRegFilt4 = _mm_maddubs_epi16(srcRegFilt4, forthFilters);
+
+ // add and saturate all the results together
+ minReg = _mm_min_epi16(srcRegFilt2, srcRegFilt3);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt4);
+
+ srcRegFilt2= _mm_max_epi16(srcRegFilt2, srcRegFilt3);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+ // shift by 7 bit each 16 bits
+ srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+ // shrink to 8 bit each 16 bits
+ srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+
+ src_ptr+=src_pixels_per_line;
+
+ // save only 8 bytes
+ _mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1);
+
+ output_ptr+=output_pitch;
+ }
+}
+
+static void vpx_filter_block1d16_h8_intrin_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_pixels_per_line,
+ uint8_t *output_ptr,
+ ptrdiff_t output_pitch,
+ uint32_t output_height,
+ const int16_t *filter) {+ __m128i addFilterReg64, filtersReg, srcReg1, srcReg2;
+ __m128i filt1Reg, filt2Reg, filt3Reg, filt4Reg;
+ __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
+ __m128i srcRegFilt1_1, srcRegFilt2_1, srcRegFilt2, srcRegFilt3;
+ unsigned int i;
+
+ // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+ addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+ filtersReg = _mm_loadu_si128((const __m128i *)filter);
+ // converting the 16 bit (short) to 8 bit (byte) and have the same data
+ // in both lanes of 128 bit register.
+ filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+ // duplicate only the first 16 bits (first and second byte)
+ // across 128 bit register
+ firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
+ // duplicate only the second 16 bits (third and forth byte)
+ // across 128 bit register
+ secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+ // duplicate only the third 16 bits (fifth and sixth byte)
+ // across 128 bit register
+ thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+ // duplicate only the forth 16 bits (seventh and eighth byte)
+ // across 128 bit register
+ forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+
+ filt1Reg = _mm_load_si128((__m128i const *)filt1_global);
+ filt2Reg = _mm_load_si128((__m128i const *)filt2_global);
+ filt3Reg = _mm_load_si128((__m128i const *)filt3_global);
+ filt4Reg = _mm_load_si128((__m128i const *)filt4_global);
+
+ for (i = 0; i < output_height; i++) {+ srcReg1 = _mm_loadu_si128((const __m128i *)(src_ptr - 3));
+
+ // filter the source buffer
+ srcRegFilt1_1= _mm_shuffle_epi8(srcReg1, filt1Reg);
+ srcRegFilt2= _mm_shuffle_epi8(srcReg1, filt4Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt1_1 = _mm_maddubs_epi16(srcRegFilt1_1, firstFilters);
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, forthFilters);
+
+ // add and saturate the results together
+ srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, srcRegFilt2);
+
+ // filter the source buffer
+ srcRegFilt3= _mm_shuffle_epi8(srcReg1, filt2Reg);
+ srcRegFilt2= _mm_shuffle_epi8(srcReg1, filt3Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
+
+ // add and saturate the results together
+ srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
+ _mm_min_epi16(srcRegFilt3, srcRegFilt2));
+
+ // reading the next 16 bytes.
+ // (part of it was being read by earlier read)
+ srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + 5));
+
+ // add and saturate the results together
+ srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1,
+ _mm_max_epi16(srcRegFilt3, srcRegFilt2));
+
+ // filter the source buffer
+ srcRegFilt2_1= _mm_shuffle_epi8(srcReg2, filt1Reg);
+ srcRegFilt2= _mm_shuffle_epi8(srcReg2, filt4Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt2_1 = _mm_maddubs_epi16(srcRegFilt2_1, firstFilters);
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, forthFilters);
+
+ // add and saturate the results together
+ srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, srcRegFilt2);
+
+ // filter the source buffer
+ srcRegFilt3= _mm_shuffle_epi8(srcReg2, filt2Reg);
+ srcRegFilt2= _mm_shuffle_epi8(srcReg2, filt3Reg);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
+
+ // add and saturate the results together
+ srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
+ _mm_min_epi16(srcRegFilt3, srcRegFilt2));
+ srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1,
+ _mm_max_epi16(srcRegFilt3, srcRegFilt2));
+
+ srcRegFilt1_1 = _mm_adds_epi16(srcRegFilt1_1, addFilterReg64);
+ srcRegFilt2_1 = _mm_adds_epi16(srcRegFilt2_1, addFilterReg64);
+
+ // shift by 7 bit each 16 bit
+ srcRegFilt1_1 = _mm_srai_epi16(srcRegFilt1_1, 7);
+ srcRegFilt2_1 = _mm_srai_epi16(srcRegFilt2_1, 7);
+
+ // shrink to 8 bit each 16 bits, the first lane contain the first
+ // convolve result and the second lane contain the second convolve
+ // result
+ srcRegFilt1_1 = _mm_packus_epi16(srcRegFilt1_1, srcRegFilt2_1);
+
+ src_ptr+=src_pixels_per_line;
+
+ // save 16 bytes
+ _mm_store_si128((__m128i*)output_ptr, srcRegFilt1_1);
+
+ output_ptr+=output_pitch;
+ }
+}
+
+void vpx_filter_block1d8_v8_intrin_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_pitch,
+ uint8_t *output_ptr,
+ ptrdiff_t out_pitch,
+ uint32_t output_height,
+ const int16_t *filter) {+ __m128i addFilterReg64, filtersReg, minReg;
+ __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
+ __m128i srcRegFilt1, srcRegFilt2, srcRegFilt3, srcRegFilt5;
+ __m128i srcReg1, srcReg2, srcReg3, srcReg4, srcReg5, srcReg6, srcReg7;
+ __m128i srcReg8;
+ unsigned int i;
+
+ // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+ addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+ filtersReg = _mm_loadu_si128((const __m128i *)filter);
+ // converting the 16 bit (short) to 8 bit (byte) and have the same data
+ // in both lanes of 128 bit register.
+ filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+ // duplicate only the first 16 bits in the filter
+ firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
+ // duplicate only the second 16 bits in the filter
+ secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+ // duplicate only the third 16 bits in the filter
+ thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+ // duplicate only the forth 16 bits in the filter
+ forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+
+ // load the first 7 rows of 8 bytes
+ srcReg1 = _mm_loadl_epi64((const __m128i *)src_ptr);
+ srcReg2 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch));
+ srcReg3 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 2));
+ srcReg4 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 3));
+ srcReg5 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 4));
+ srcReg6 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 5));
+ srcReg7 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 6));
+
+ for (i = 0; i < output_height; i++) {+ // load the last 8 bytes
+ srcReg8 = _mm_loadl_epi64((const __m128i *)(src_ptr + src_pitch * 7));
+
+ // merge the result together
+ srcRegFilt1 = _mm_unpacklo_epi8(srcReg1, srcReg2);
+ srcRegFilt3 = _mm_unpacklo_epi8(srcReg3, srcReg4);
+
+ // merge the result together
+ srcRegFilt2 = _mm_unpacklo_epi8(srcReg5, srcReg6);
+ srcRegFilt5 = _mm_unpacklo_epi8(srcReg7, srcReg8);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
+ srcRegFilt2 = _mm_maddubs_epi16(srcRegFilt2, thirdFilters);
+ srcRegFilt5 = _mm_maddubs_epi16(srcRegFilt5, forthFilters);
+
+ // add and saturate the results together
+ minReg = _mm_min_epi16(srcRegFilt2, srcRegFilt3);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt5);
+ srcRegFilt2 = _mm_max_epi16(srcRegFilt2, srcRegFilt3);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, minReg);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt2);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+ // shift by 7 bit each 16 bit
+ srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+ // shrink to 8 bit each 16 bits
+ srcRegFilt1 = _mm_packus_epi16(srcRegFilt1, srcRegFilt1);
+
+ src_ptr+=src_pitch;
+
+ // shift down a row
+ srcReg1 = srcReg2;
+ srcReg2 = srcReg3;
+ srcReg3 = srcReg4;
+ srcReg4 = srcReg5;
+ srcReg5 = srcReg6;
+ srcReg6 = srcReg7;
+ srcReg7 = srcReg8;
+
+ // save only 8 bytes convolve result
+ _mm_storel_epi64((__m128i*)&output_ptr[0], srcRegFilt1);
+
+ output_ptr+=out_pitch;
+ }
+}
+
+static void vpx_filter_block1d16_v8_intrin_ssse3(const uint8_t *src_ptr,
+ ptrdiff_t src_pitch,
+ uint8_t *output_ptr,
+ ptrdiff_t out_pitch,
+ uint32_t output_height,
+ const int16_t *filter) {+ __m128i addFilterReg64, filtersReg, srcRegFilt1, srcRegFilt3;
+ __m128i firstFilters, secondFilters, thirdFilters, forthFilters;
+ __m128i srcRegFilt5, srcRegFilt6, srcRegFilt7, srcRegFilt8;
+ __m128i srcReg1, srcReg2, srcReg3, srcReg4, srcReg5, srcReg6, srcReg7;
+ __m128i srcReg8;
+ unsigned int i;
+
+ // create a register with 0,64,0,64,0,64,0,64,0,64,0,64,0,64,0,64
+ addFilterReg64 = _mm_set1_epi32((int)0x0400040u);
+ filtersReg = _mm_loadu_si128((const __m128i *)filter);
+ // converting the 16 bit (short) to 8 bit (byte) and have the same data
+ // in both lanes of 128 bit register.
+ filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+
+ // duplicate only the first 16 bits in the filter
+ firstFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x100u));
+ // duplicate only the second 16 bits in the filter
+ secondFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x302u));
+ // duplicate only the third 16 bits in the filter
+ thirdFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x504u));
+ // duplicate only the forth 16 bits in the filter
+ forthFilters = _mm_shuffle_epi8(filtersReg, _mm_set1_epi16(0x706u));
+
+ // load the first 7 rows of 16 bytes
+ srcReg1 = _mm_loadu_si128((const __m128i *)(src_ptr));
+ srcReg2 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch));
+ srcReg3 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 2));
+ srcReg4 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 3));
+ srcReg5 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 4));
+ srcReg6 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 5));
+ srcReg7 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 6));
+
+ for (i = 0; i < output_height; i++) {+ // load the last 16 bytes
+ srcReg8 = _mm_loadu_si128((const __m128i *)(src_ptr + src_pitch * 7));
+
+ // merge the result together
+ srcRegFilt5 = _mm_unpacklo_epi8(srcReg1, srcReg2);
+ srcRegFilt6 = _mm_unpacklo_epi8(srcReg7, srcReg8);
+ srcRegFilt1 = _mm_unpackhi_epi8(srcReg1, srcReg2);
+ srcRegFilt3 = _mm_unpackhi_epi8(srcReg7, srcReg8);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt5 = _mm_maddubs_epi16(srcRegFilt5, firstFilters);
+ srcRegFilt6 = _mm_maddubs_epi16(srcRegFilt6, forthFilters);
+ srcRegFilt1 = _mm_maddubs_epi16(srcRegFilt1, firstFilters);
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, forthFilters);
+
+ // add and saturate the results together
+ srcRegFilt5 = _mm_adds_epi16(srcRegFilt5, srcRegFilt6);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, srcRegFilt3);
+
+ // merge the result together
+ srcRegFilt3 = _mm_unpacklo_epi8(srcReg3, srcReg4);
+ srcRegFilt6 = _mm_unpackhi_epi8(srcReg3, srcReg4);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt3 = _mm_maddubs_epi16(srcRegFilt3, secondFilters);
+ srcRegFilt6 = _mm_maddubs_epi16(srcRegFilt6, secondFilters);
+
+ // merge the result together
+ srcRegFilt7 = _mm_unpacklo_epi8(srcReg5, srcReg6);
+ srcRegFilt8 = _mm_unpackhi_epi8(srcReg5, srcReg6);
+
+ // multiply 2 adjacent elements with the filter and add the result
+ srcRegFilt7 = _mm_maddubs_epi16(srcRegFilt7, thirdFilters);
+ srcRegFilt8 = _mm_maddubs_epi16(srcRegFilt8, thirdFilters);
+
+ // add and saturate the results together
+ srcRegFilt5 = _mm_adds_epi16(srcRegFilt5,
+ _mm_min_epi16(srcRegFilt3, srcRegFilt7));
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
+ _mm_min_epi16(srcRegFilt6, srcRegFilt8));
+
+ // add and saturate the results together
+ srcRegFilt5 = _mm_adds_epi16(srcRegFilt5,
+ _mm_max_epi16(srcRegFilt3, srcRegFilt7));
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1,
+ _mm_max_epi16(srcRegFilt6, srcRegFilt8));
+ srcRegFilt5 = _mm_adds_epi16(srcRegFilt5, addFilterReg64);
+ srcRegFilt1 = _mm_adds_epi16(srcRegFilt1, addFilterReg64);
+
+ // shift by 7 bit each 16 bit
+ srcRegFilt5 = _mm_srai_epi16(srcRegFilt5, 7);
+ srcRegFilt1 = _mm_srai_epi16(srcRegFilt1, 7);
+
+ // shrink to 8 bit each 16 bits, the first lane contain the first
+ // convolve result and the second lane contain the second convolve
+ // result
+ srcRegFilt1 = _mm_packus_epi16(srcRegFilt5, srcRegFilt1);
+
+ src_ptr+=src_pitch;
+
+ // shift down a row
+ srcReg1 = srcReg2;
+ srcReg2 = srcReg3;
+ srcReg3 = srcReg4;
+ srcReg4 = srcReg5;
+ srcReg5 = srcReg6;
+ srcReg6 = srcReg7;
+ srcReg7 = srcReg8;
+
+ // save 16 bytes convolve result
+ _mm_store_si128((__m128i*)output_ptr, srcRegFilt1);
+
+ output_ptr+=out_pitch;
+ }
+}
+
+#if ARCH_X86_64
+filter8_1dfunction vpx_filter_block1d16_v8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d16_h8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d8_v8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h8_intrin_ssse3;
+filter8_1dfunction vpx_filter_block1d4_v8_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h8_intrin_ssse3;
+#define vpx_filter_block1d16_v8_ssse3 vpx_filter_block1d16_v8_intrin_ssse3
+#define vpx_filter_block1d16_h8_ssse3 vpx_filter_block1d16_h8_intrin_ssse3
+#define vpx_filter_block1d8_v8_ssse3 vpx_filter_block1d8_v8_intrin_ssse3
+#define vpx_filter_block1d8_h8_ssse3 vpx_filter_block1d8_h8_intrin_ssse3
+#define vpx_filter_block1d4_h8_ssse3 vpx_filter_block1d4_h8_intrin_ssse3
+#else // ARCH_X86
+filter8_1dfunction vpx_filter_block1d16_v8_ssse3;
+filter8_1dfunction vpx_filter_block1d16_h8_ssse3;
+filter8_1dfunction vpx_filter_block1d8_v8_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h8_ssse3;
+filter8_1dfunction vpx_filter_block1d4_v8_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h8_ssse3;
+#endif // ARCH_X86_64
+filter8_1dfunction vpx_filter_block1d16_v8_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d16_h8_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d8_v8_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h8_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d4_v8_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h8_avg_ssse3;
+
+filter8_1dfunction vpx_filter_block1d16_v2_ssse3;
+filter8_1dfunction vpx_filter_block1d16_h2_ssse3;
+filter8_1dfunction vpx_filter_block1d8_v2_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h2_ssse3;
+filter8_1dfunction vpx_filter_block1d4_v2_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h2_ssse3;
+filter8_1dfunction vpx_filter_block1d16_v2_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d16_h2_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d8_v2_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d8_h2_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d4_v2_avg_ssse3;
+filter8_1dfunction vpx_filter_block1d4_h2_avg_ssse3;
+
+// void vpx_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_avg_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , ssse3);
+FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , ssse3);
+FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, ssse3);
+FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
+ ssse3);
+
+// void vpx_convolve8_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+// void vpx_convolve8_avg_ssse3(const uint8_t *src, ptrdiff_t src_stride,
+// uint8_t *dst, ptrdiff_t dst_stride,
+// const int16_t *filter_x, int x_step_q4,
+// const int16_t *filter_y, int y_step_q4,
+// int w, int h);
+FUN_CONV_2D(, ssse3);
+FUN_CONV_2D(avg_ , ssse3);
--- /dev/null
+++ b/vpx_dsp/x86/vpx_subpixel_8t_sse2.asm
@@ -1,0 +1,987 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;Note: tap3 and tap4 have to be applied and added after other taps to avoid
+;overflow.
+
+%macro GET_FILTERS_4 0
+ mov rdx, arg(5) ;filter ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm7, [rdx] ;load filters
+ pshuflw xmm0, xmm7, 0b ;k0
+ pshuflw xmm1, xmm7, 01010101b ;k1
+ pshuflw xmm2, xmm7, 10101010b ;k2
+ pshuflw xmm3, xmm7, 11111111b ;k3
+ psrldq xmm7, 8
+ pshuflw xmm4, xmm7, 0b ;k4
+ pshuflw xmm5, xmm7, 01010101b ;k5
+ pshuflw xmm6, xmm7, 10101010b ;k6
+ pshuflw xmm7, xmm7, 11111111b ;k7
+
+ punpcklqdq xmm0, xmm1
+ punpcklqdq xmm2, xmm3
+ punpcklqdq xmm5, xmm4
+ punpcklqdq xmm6, xmm7
+
+ movdqa k0k1, xmm0
+ movdqa k2k3, xmm2
+ movdqa k5k4, xmm5
+ movdqa k6k7, xmm6
+
+ movq xmm6, rcx
+ pshufd xmm6, xmm6, 0
+ movdqa krd, xmm6
+
+ pxor xmm7, xmm7
+ movdqa zero, xmm7
+%endm
+
+%macro APPLY_FILTER_4 1
+ punpckldq xmm0, xmm1 ;two row in one register
+ punpckldq xmm6, xmm7
+ punpckldq xmm2, xmm3
+ punpckldq xmm5, xmm4
+
+ punpcklbw xmm0, zero ;unpack to word
+ punpcklbw xmm6, zero
+ punpcklbw xmm2, zero
+ punpcklbw xmm5, zero
+
+ pmullw xmm0, k0k1 ;multiply the filter factors
+ pmullw xmm6, k6k7
+ pmullw xmm2, k2k3
+ pmullw xmm5, k5k4
+
+ paddsw xmm0, xmm6 ;sum
+ movdqa xmm1, xmm0
+ psrldq xmm1, 8
+ paddsw xmm0, xmm1
+ paddsw xmm0, xmm2
+ psrldq xmm2, 8
+ paddsw xmm0, xmm5
+ psrldq xmm5, 8
+ paddsw xmm0, xmm2
+ paddsw xmm0, xmm5
+
+ paddsw xmm0, krd ;rounding
+ psraw xmm0, 7 ;shift
+ packuswb xmm0, xmm0 ;pack to byte
+
+%if %1
+ movd xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movd [rdi], xmm0
+%endm
+
+%macro GET_FILTERS 0
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm7, [rdx] ;load filters
+ pshuflw xmm0, xmm7, 0b ;k0
+ pshuflw xmm1, xmm7, 01010101b ;k1
+ pshuflw xmm2, xmm7, 10101010b ;k2
+ pshuflw xmm3, xmm7, 11111111b ;k3
+ pshufhw xmm4, xmm7, 0b ;k4
+ pshufhw xmm5, xmm7, 01010101b ;k5
+ pshufhw xmm6, xmm7, 10101010b ;k6
+ pshufhw xmm7, xmm7, 11111111b ;k7
+
+ punpcklwd xmm0, xmm0
+ punpcklwd xmm1, xmm1
+ punpcklwd xmm2, xmm2
+ punpcklwd xmm3, xmm3
+ punpckhwd xmm4, xmm4
+ punpckhwd xmm5, xmm5
+ punpckhwd xmm6, xmm6
+ punpckhwd xmm7, xmm7
+
+ movdqa k0, xmm0 ;store filter factors on stack
+ movdqa k1, xmm1
+ movdqa k2, xmm2
+ movdqa k3, xmm3
+ movdqa k4, xmm4
+ movdqa k5, xmm5
+ movdqa k6, xmm6
+ movdqa k7, xmm7
+
+ movq xmm6, rcx
+ pshufd xmm6, xmm6, 0
+ movdqa krd, xmm6 ;rounding
+
+ pxor xmm7, xmm7
+ movdqa zero, xmm7
+%endm
+
+%macro LOAD_VERT_8 1
+ movq xmm0, [rsi + %1] ;0
+ movq xmm1, [rsi + rax + %1] ;1
+ movq xmm6, [rsi + rdx * 2 + %1] ;6
+ lea rsi, [rsi + rax]
+ movq xmm7, [rsi + rdx * 2 + %1] ;7
+ movq xmm2, [rsi + rax + %1] ;2
+ movq xmm3, [rsi + rax * 2 + %1] ;3
+ movq xmm4, [rsi + rdx + %1] ;4
+ movq xmm5, [rsi + rax * 4 + %1] ;5
+%endm
+
+%macro APPLY_FILTER_8 2
+ punpcklbw xmm0, zero
+ punpcklbw xmm1, zero
+ punpcklbw xmm6, zero
+ punpcklbw xmm7, zero
+ punpcklbw xmm2, zero
+ punpcklbw xmm5, zero
+ punpcklbw xmm3, zero
+ punpcklbw xmm4, zero
+
+ pmullw xmm0, k0
+ pmullw xmm1, k1
+ pmullw xmm6, k6
+ pmullw xmm7, k7
+ pmullw xmm2, k2
+ pmullw xmm5, k5
+ pmullw xmm3, k3
+ pmullw xmm4, k4
+
+ paddsw xmm0, xmm1
+ paddsw xmm0, xmm6
+ paddsw xmm0, xmm7
+ paddsw xmm0, xmm2
+ paddsw xmm0, xmm5
+ paddsw xmm0, xmm3
+ paddsw xmm0, xmm4
+
+ paddsw xmm0, krd ;rounding
+ psraw xmm0, 7 ;shift
+ packuswb xmm0, xmm0 ;pack back to byte
+%if %1
+ movq xmm1, [rdi + %2]
+ pavgb xmm0, xmm1
+%endif
+ movq [rdi + %2], xmm0
+%endm
+
+;void vpx_filter_block1d4_v8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d4_v8_sse2) PRIVATE
+sym(vpx_filter_block1d4_v8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 6
+ %define k0k1 [rsp + 16 * 0]
+ %define k2k3 [rsp + 16 * 1]
+ %define k5k4 [rsp + 16 * 2]
+ %define k6k7 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define zero [rsp + 16 * 5]
+
+ GET_FILTERS_4
+
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movd xmm0, [rsi] ;load src: row 0
+ movd xmm1, [rsi + rax] ;1
+ movd xmm6, [rsi + rdx * 2] ;6
+ lea rsi, [rsi + rax]
+ movd xmm7, [rsi + rdx * 2] ;7
+ movd xmm2, [rsi + rax] ;2
+ movd xmm3, [rsi + rax * 2] ;3
+ movd xmm4, [rsi + rdx] ;4
+ movd xmm5, [rsi + rax * 4] ;5
+
+ APPLY_FILTER_4 0
+
+ lea rdi, [rdi + rbx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 6
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d8_v8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d8_v8_sse2) PRIVATE
+sym(vpx_filter_block1d8_v8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 10
+ %define k0 [rsp + 16 * 0]
+ %define k1 [rsp + 16 * 1]
+ %define k2 [rsp + 16 * 2]
+ %define k3 [rsp + 16 * 3]
+ %define k4 [rsp + 16 * 4]
+ %define k5 [rsp + 16 * 5]
+ %define k6 [rsp + 16 * 6]
+ %define k7 [rsp + 16 * 7]
+ %define krd [rsp + 16 * 8]
+ %define zero [rsp + 16 * 9]
+
+ GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ LOAD_VERT_8 0
+ APPLY_FILTER_8 0, 0
+
+ lea rdi, [rdi + rbx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 10
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d16_v8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d16_v8_sse2) PRIVATE
+sym(vpx_filter_block1d16_v8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 10
+ %define k0 [rsp + 16 * 0]
+ %define k1 [rsp + 16 * 1]
+ %define k2 [rsp + 16 * 2]
+ %define k3 [rsp + 16 * 3]
+ %define k4 [rsp + 16 * 4]
+ %define k5 [rsp + 16 * 5]
+ %define k6 [rsp + 16 * 6]
+ %define k7 [rsp + 16 * 7]
+ %define krd [rsp + 16 * 8]
+ %define zero [rsp + 16 * 9]
+
+ GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ LOAD_VERT_8 0
+ APPLY_FILTER_8 0, 0
+ sub rsi, rax
+
+ LOAD_VERT_8 8
+ APPLY_FILTER_8 0, 8
+ add rdi, rbx
+
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 10
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_v8_avg_sse2) PRIVATE
+sym(vpx_filter_block1d4_v8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 6
+ %define k0k1 [rsp + 16 * 0]
+ %define k2k3 [rsp + 16 * 1]
+ %define k5k4 [rsp + 16 * 2]
+ %define k6k7 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define zero [rsp + 16 * 5]
+
+ GET_FILTERS_4
+
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movd xmm0, [rsi] ;load src: row 0
+ movd xmm1, [rsi + rax] ;1
+ movd xmm6, [rsi + rdx * 2] ;6
+ lea rsi, [rsi + rax]
+ movd xmm7, [rsi + rdx * 2] ;7
+ movd xmm2, [rsi + rax] ;2
+ movd xmm3, [rsi + rax * 2] ;3
+ movd xmm4, [rsi + rdx] ;4
+ movd xmm5, [rsi + rax * 4] ;5
+
+ APPLY_FILTER_4 1
+
+ lea rdi, [rdi + rbx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 6
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_v8_avg_sse2) PRIVATE
+sym(vpx_filter_block1d8_v8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 10
+ %define k0 [rsp + 16 * 0]
+ %define k1 [rsp + 16 * 1]
+ %define k2 [rsp + 16 * 2]
+ %define k3 [rsp + 16 * 3]
+ %define k4 [rsp + 16 * 4]
+ %define k5 [rsp + 16 * 5]
+ %define k6 [rsp + 16 * 6]
+ %define k7 [rsp + 16 * 7]
+ %define krd [rsp + 16 * 8]
+ %define zero [rsp + 16 * 9]
+
+ GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+.loop:
+ LOAD_VERT_8 0
+ APPLY_FILTER_8 1, 0
+
+ lea rdi, [rdi + rbx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 10
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_v8_avg_sse2) PRIVATE
+sym(vpx_filter_block1d16_v8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 10
+ %define k0 [rsp + 16 * 0]
+ %define k1 [rsp + 16 * 1]
+ %define k2 [rsp + 16 * 2]
+ %define k3 [rsp + 16 * 3]
+ %define k4 [rsp + 16 * 4]
+ %define k5 [rsp + 16 * 5]
+ %define k6 [rsp + 16 * 6]
+ %define k7 [rsp + 16 * 7]
+ %define krd [rsp + 16 * 8]
+ %define zero [rsp + 16 * 9]
+
+ GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rbx, DWORD PTR arg(3) ;out_pitch
+ lea rdx, [rax + rax * 2]
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+.loop:
+ LOAD_VERT_8 0
+ APPLY_FILTER_8 1, 0
+ sub rsi, rax
+
+ LOAD_VERT_8 8
+ APPLY_FILTER_8 1, 8
+ add rdi, rbx
+
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 10
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d4_h8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d4_h8_sse2) PRIVATE
+sym(vpx_filter_block1d4_h8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 6
+ %define k0k1 [rsp + 16 * 0]
+ %define k2k3 [rsp + 16 * 1]
+ %define k5k4 [rsp + 16 * 2]
+ %define k6k7 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define zero [rsp + 16 * 5]
+
+ GET_FILTERS_4
+
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 3] ;load src
+
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm0
+ movdqa xmm7, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm5, xmm0
+ movdqa xmm4, xmm0
+
+ psrldq xmm1, 1
+ psrldq xmm6, 6
+ psrldq xmm7, 7
+ psrldq xmm2, 2
+ psrldq xmm3, 3
+ psrldq xmm5, 5
+ psrldq xmm4, 4
+
+ APPLY_FILTER_4 0
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 6
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d8_h8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d8_h8_sse2) PRIVATE
+sym(vpx_filter_block1d8_h8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 10
+ %define k0 [rsp + 16 * 0]
+ %define k1 [rsp + 16 * 1]
+ %define k2 [rsp + 16 * 2]
+ %define k3 [rsp + 16 * 3]
+ %define k4 [rsp + 16 * 4]
+ %define k5 [rsp + 16 * 5]
+ %define k6 [rsp + 16 * 6]
+ %define k7 [rsp + 16 * 7]
+ %define krd [rsp + 16 * 8]
+ %define zero [rsp + 16 * 9]
+
+ GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 3] ;load src
+
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm0
+ movdqa xmm7, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm5, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm4, xmm0
+
+ psrldq xmm1, 1
+ psrldq xmm6, 6
+ psrldq xmm7, 7
+ psrldq xmm2, 2
+ psrldq xmm5, 5
+ psrldq xmm3, 3
+ psrldq xmm4, 4
+
+ APPLY_FILTER_8 0, 0
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 10
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d16_h8_sse2
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d16_h8_sse2) PRIVATE
+sym(vpx_filter_block1d16_h8_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 10
+ %define k0 [rsp + 16 * 0]
+ %define k1 [rsp + 16 * 1]
+ %define k2 [rsp + 16 * 2]
+ %define k3 [rsp + 16 * 3]
+ %define k4 [rsp + 16 * 4]
+ %define k5 [rsp + 16 * 5]
+ %define k6 [rsp + 16 * 6]
+ %define k7 [rsp + 16 * 7]
+ %define krd [rsp + 16 * 8]
+ %define zero [rsp + 16 * 9]
+
+ GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 3] ;load src
+
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm0
+ movdqa xmm7, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm5, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm4, xmm0
+
+ psrldq xmm1, 1
+ psrldq xmm6, 6
+ psrldq xmm7, 7
+ psrldq xmm2, 2
+ psrldq xmm5, 5
+ psrldq xmm3, 3
+ psrldq xmm4, 4
+
+ APPLY_FILTER_8 0, 0
+
+ movdqu xmm0, [rsi + 5] ;load src
+
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm0
+ movdqa xmm7, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm5, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm4, xmm0
+
+ psrldq xmm1, 1
+ psrldq xmm6, 6
+ psrldq xmm7, 7
+ psrldq xmm2, 2
+ psrldq xmm5, 5
+ psrldq xmm3, 3
+ psrldq xmm4, 4
+
+ APPLY_FILTER_8 0, 8
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 10
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_h8_avg_sse2) PRIVATE
+sym(vpx_filter_block1d4_h8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 6
+ %define k0k1 [rsp + 16 * 0]
+ %define k2k3 [rsp + 16 * 1]
+ %define k5k4 [rsp + 16 * 2]
+ %define k6k7 [rsp + 16 * 3]
+ %define krd [rsp + 16 * 4]
+ %define zero [rsp + 16 * 5]
+
+ GET_FILTERS_4
+
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 3] ;load src
+
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm0
+ movdqa xmm7, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm5, xmm0
+ movdqa xmm4, xmm0
+
+ psrldq xmm1, 1
+ psrldq xmm6, 6
+ psrldq xmm7, 7
+ psrldq xmm2, 2
+ psrldq xmm3, 3
+ psrldq xmm5, 5
+ psrldq xmm4, 4
+
+ APPLY_FILTER_4 1
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 6
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_h8_avg_sse2) PRIVATE
+sym(vpx_filter_block1d8_h8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 10
+ %define k0 [rsp + 16 * 0]
+ %define k1 [rsp + 16 * 1]
+ %define k2 [rsp + 16 * 2]
+ %define k3 [rsp + 16 * 3]
+ %define k4 [rsp + 16 * 4]
+ %define k5 [rsp + 16 * 5]
+ %define k6 [rsp + 16 * 6]
+ %define k7 [rsp + 16 * 7]
+ %define krd [rsp + 16 * 8]
+ %define zero [rsp + 16 * 9]
+
+ GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 3] ;load src
+
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm0
+ movdqa xmm7, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm5, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm4, xmm0
+
+ psrldq xmm1, 1
+ psrldq xmm6, 6
+ psrldq xmm7, 7
+ psrldq xmm2, 2
+ psrldq xmm5, 5
+ psrldq xmm3, 3
+ psrldq xmm4, 4
+
+ APPLY_FILTER_8 1, 0
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 10
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_h8_avg_sse2) PRIVATE
+sym(vpx_filter_block1d16_h8_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 10
+ %define k0 [rsp + 16 * 0]
+ %define k1 [rsp + 16 * 1]
+ %define k2 [rsp + 16 * 2]
+ %define k3 [rsp + 16 * 3]
+ %define k4 [rsp + 16 * 4]
+ %define k5 [rsp + 16 * 5]
+ %define k6 [rsp + 16 * 6]
+ %define k7 [rsp + 16 * 7]
+ %define krd [rsp + 16 * 8]
+ %define zero [rsp + 16 * 9]
+
+ GET_FILTERS
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+
+.loop:
+ movdqu xmm0, [rsi - 3] ;load src
+
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm0
+ movdqa xmm7, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm5, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm4, xmm0
+
+ psrldq xmm1, 1
+ psrldq xmm6, 6
+ psrldq xmm7, 7
+ psrldq xmm2, 2
+ psrldq xmm5, 5
+ psrldq xmm3, 3
+ psrldq xmm4, 4
+
+ APPLY_FILTER_8 1, 0
+
+ movdqu xmm0, [rsi + 5] ;load src
+
+ movdqa xmm1, xmm0
+ movdqa xmm6, xmm0
+ movdqa xmm7, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm5, xmm0
+ movdqa xmm3, xmm0
+ movdqa xmm4, xmm0
+
+ psrldq xmm1, 1
+ psrldq xmm6, 6
+ psrldq xmm7, 7
+ psrldq xmm2, 2
+ psrldq xmm5, 5
+ psrldq xmm3, 3
+ psrldq xmm4, 4
+
+ APPLY_FILTER_8 1, 8
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+
+ add rsp, 16 * 10
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
--- /dev/null
+++ b/vpx_dsp/x86/vpx_subpixel_8t_ssse3.asm
@@ -1,0 +1,1071 @@
+;
+; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro VERTx4 1
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm4, [rdx] ;load filters
+ movq xmm5, rcx
+ packsswb xmm4, xmm4
+ pshuflw xmm0, xmm4, 0b ;k0_k1
+ pshuflw xmm1, xmm4, 01010101b ;k2_k3
+ pshuflw xmm2, xmm4, 10101010b ;k4_k5
+ pshuflw xmm3, xmm4, 11111111b ;k6_k7
+
+ punpcklqdq xmm0, xmm0
+ punpcklqdq xmm1, xmm1
+ punpcklqdq xmm2, xmm2
+ punpcklqdq xmm3, xmm3
+
+ movdqa k0k1, xmm0
+ movdqa k2k3, xmm1
+ pshufd xmm5, xmm5, 0
+ movdqa k4k5, xmm2
+ movdqa k6k7, xmm3
+ movdqa krd, xmm5
+
+ movsxd rdx, DWORD PTR arg(1) ;pixels_per_line
+
+%if ABI_IS_32BIT=0
+ movsxd r8, DWORD PTR arg(3) ;out_pitch
+%endif
+ mov rax, rsi
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+ add rax, rdx
+
+ lea rbx, [rdx + rdx*4]
+ add rbx, rdx ;pitch * 6
+
+.loop:
+ movd xmm0, [rsi] ;A
+ movd xmm1, [rsi + rdx] ;B
+ movd xmm2, [rsi + rdx * 2] ;C
+ movd xmm3, [rax + rdx * 2] ;D
+ movd xmm4, [rsi + rdx * 4] ;E
+ movd xmm5, [rax + rdx * 4] ;F
+
+ punpcklbw xmm0, xmm1 ;A B
+ punpcklbw xmm2, xmm3 ;C D
+ punpcklbw xmm4, xmm5 ;E F
+
+ movd xmm6, [rsi + rbx] ;G
+ movd xmm7, [rax + rbx] ;H
+
+ pmaddubsw xmm0, k0k1
+ pmaddubsw xmm2, k2k3
+ punpcklbw xmm6, xmm7 ;G H
+ pmaddubsw xmm4, k4k5
+ pmaddubsw xmm6, k6k7
+
+ movdqa xmm1, xmm2
+ paddsw xmm0, xmm6
+ pmaxsw xmm2, xmm4
+ pminsw xmm4, xmm1
+ paddsw xmm0, xmm4
+ paddsw xmm0, xmm2
+
+ paddsw xmm0, krd
+ psraw xmm0, 7
+ packuswb xmm0, xmm0
+
+ add rsi, rdx
+ add rax, rdx
+%if %1
+ movd xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movd [rdi], xmm0
+
+%if ABI_IS_32BIT
+ add rdi, DWORD PTR arg(3) ;out_pitch
+%else
+ add rdi, r8
+%endif
+ dec rcx
+ jnz .loop
+%endm
+
+%macro VERTx8 1
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm4, [rdx] ;load filters
+ movq xmm5, rcx
+ packsswb xmm4, xmm4
+ pshuflw xmm0, xmm4, 0b ;k0_k1
+ pshuflw xmm1, xmm4, 01010101b ;k2_k3
+ pshuflw xmm2, xmm4, 10101010b ;k4_k5
+ pshuflw xmm3, xmm4, 11111111b ;k6_k7
+
+ punpcklqdq xmm0, xmm0
+ punpcklqdq xmm1, xmm1
+ punpcklqdq xmm2, xmm2
+ punpcklqdq xmm3, xmm3
+
+ movdqa k0k1, xmm0
+ movdqa k2k3, xmm1
+ pshufd xmm5, xmm5, 0
+ movdqa k4k5, xmm2
+ movdqa k6k7, xmm3
+ movdqa krd, xmm5
+
+ movsxd rdx, DWORD PTR arg(1) ;pixels_per_line
+
+%if ABI_IS_32BIT=0
+ movsxd r8, DWORD PTR arg(3) ;out_pitch
+%endif
+ mov rax, rsi
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+ add rax, rdx
+
+ lea rbx, [rdx + rdx*4]
+ add rbx, rdx ;pitch * 6
+
+.loop:
+ movq xmm0, [rsi] ;A
+ movq xmm1, [rsi + rdx] ;B
+ movq xmm2, [rsi + rdx * 2] ;C
+ movq xmm3, [rax + rdx * 2] ;D
+ movq xmm4, [rsi + rdx * 4] ;E
+ movq xmm5, [rax + rdx * 4] ;F
+
+ punpcklbw xmm0, xmm1 ;A B
+ punpcklbw xmm2, xmm3 ;C D
+ punpcklbw xmm4, xmm5 ;E F
+
+ movq xmm6, [rsi + rbx] ;G
+ movq xmm7, [rax + rbx] ;H
+
+ pmaddubsw xmm0, k0k1
+ pmaddubsw xmm2, k2k3
+ punpcklbw xmm6, xmm7 ;G H
+ pmaddubsw xmm4, k4k5
+ pmaddubsw xmm6, k6k7
+
+ paddsw xmm0, xmm6
+ movdqa xmm1, xmm2
+ pmaxsw xmm2, xmm4
+ pminsw xmm4, xmm1
+ paddsw xmm0, xmm4
+ paddsw xmm0, xmm2
+
+ paddsw xmm0, krd
+ psraw xmm0, 7
+ packuswb xmm0, xmm0
+
+ add rsi, rdx
+ add rax, rdx
+%if %1
+ movq xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movq [rdi], xmm0
+
+%if ABI_IS_32BIT
+ add rdi, DWORD PTR arg(3) ;out_pitch
+%else
+ add rdi, r8
+%endif
+ dec rcx
+ jnz .loop
+%endm
+
+
+%macro VERTx16 1
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm4, [rdx] ;load filters
+ movq xmm5, rcx
+ packsswb xmm4, xmm4
+ pshuflw xmm0, xmm4, 0b ;k0_k1
+ pshuflw xmm1, xmm4, 01010101b ;k2_k3
+ pshuflw xmm2, xmm4, 10101010b ;k4_k5
+ pshuflw xmm3, xmm4, 11111111b ;k6_k7
+
+ punpcklqdq xmm0, xmm0
+ punpcklqdq xmm1, xmm1
+ punpcklqdq xmm2, xmm2
+ punpcklqdq xmm3, xmm3
+
+ movdqa k0k1, xmm0
+ movdqa k2k3, xmm1
+ pshufd xmm5, xmm5, 0
+ movdqa k4k5, xmm2
+ movdqa k6k7, xmm3
+ movdqa krd, xmm5
+
+ movsxd rdx, DWORD PTR arg(1) ;pixels_per_line
+
+%if ABI_IS_32BIT=0
+ movsxd r8, DWORD PTR arg(3) ;out_pitch
+%endif
+ mov rax, rsi
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+ add rax, rdx
+
+ lea rbx, [rdx + rdx*4]
+ add rbx, rdx ;pitch * 6
+
+.loop:
+ movq xmm0, [rsi] ;A
+ movq xmm1, [rsi + rdx] ;B
+ movq xmm2, [rsi + rdx * 2] ;C
+ movq xmm3, [rax + rdx * 2] ;D
+ movq xmm4, [rsi + rdx * 4] ;E
+ movq xmm5, [rax + rdx * 4] ;F
+
+ punpcklbw xmm0, xmm1 ;A B
+ punpcklbw xmm2, xmm3 ;C D
+ punpcklbw xmm4, xmm5 ;E F
+
+ movq xmm6, [rsi + rbx] ;G
+ movq xmm7, [rax + rbx] ;H
+
+ pmaddubsw xmm0, k0k1
+ pmaddubsw xmm2, k2k3
+ punpcklbw xmm6, xmm7 ;G H
+ pmaddubsw xmm4, k4k5
+ pmaddubsw xmm6, k6k7
+
+ paddsw xmm0, xmm6
+ movdqa xmm1, xmm2
+ pmaxsw xmm2, xmm4
+ pminsw xmm4, xmm1
+ paddsw xmm0, xmm4
+ paddsw xmm0, xmm2
+
+ paddsw xmm0, krd
+ psraw xmm0, 7
+ packuswb xmm0, xmm0
+%if %1
+ movq xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movq [rdi], xmm0
+
+ movq xmm0, [rsi + 8] ;A
+ movq xmm1, [rsi + rdx + 8] ;B
+ movq xmm2, [rsi + rdx * 2 + 8] ;C
+ movq xmm3, [rax + rdx * 2 + 8] ;D
+ movq xmm4, [rsi + rdx * 4 + 8] ;E
+ movq xmm5, [rax + rdx * 4 + 8] ;F
+
+ punpcklbw xmm0, xmm1 ;A B
+ punpcklbw xmm2, xmm3 ;C D
+ punpcklbw xmm4, xmm5 ;E F
+
+ movq xmm6, [rsi + rbx + 8] ;G
+ movq xmm7, [rax + rbx + 8] ;H
+ punpcklbw xmm6, xmm7 ;G H
+
+ pmaddubsw xmm0, k0k1
+ pmaddubsw xmm2, k2k3
+ pmaddubsw xmm4, k4k5
+ pmaddubsw xmm6, k6k7
+
+ paddsw xmm0, xmm6
+ movdqa xmm1, xmm2
+ pmaxsw xmm2, xmm4
+ pminsw xmm4, xmm1
+ paddsw xmm0, xmm4
+ paddsw xmm0, xmm2
+
+ paddsw xmm0, krd
+ psraw xmm0, 7
+ packuswb xmm0, xmm0
+
+ add rsi, rdx
+ add rax, rdx
+%if %1
+ movq xmm1, [rdi+8]
+ pavgb xmm0, xmm1
+%endif
+
+ movq [rdi+8], xmm0
+
+%if ABI_IS_32BIT
+ add rdi, DWORD PTR arg(3) ;out_pitch
+%else
+ add rdi, r8
+%endif
+ dec rcx
+ jnz .loop
+%endm
+
+;void vpx_filter_block1d8_v8_ssse3
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d4_v8_ssse3) PRIVATE
+sym(vpx_filter_block1d4_v8_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ VERTx4 0
+
+ add rsp, 16*5
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d8_v8_ssse3
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d8_v8_ssse3) PRIVATE
+sym(vpx_filter_block1d8_v8_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ VERTx8 0
+
+ add rsp, 16*5
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d16_v8_ssse3
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pitch,
+; unsigned char *output_ptr,
+; unsigned int out_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d16_v8_ssse3) PRIVATE
+sym(vpx_filter_block1d16_v8_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ VERTx16 0
+
+ add rsp, 16*5
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+
+global sym(vpx_filter_block1d4_v8_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d4_v8_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ VERTx4 1
+
+ add rsp, 16*5
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_v8_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d8_v8_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ VERTx8 1
+
+ add rsp, 16*5
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_v8_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d16_v8_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ push rbx
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ VERTx16 1
+
+ add rsp, 16*5
+ pop rsp
+ pop rbx
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+%macro HORIZx4_ROW 2
+ movdqa %2, %1
+ pshufb %1, [GLOBAL(shuf_t0t1)]
+ pshufb %2, [GLOBAL(shuf_t2t3)]
+ pmaddubsw %1, k0k1k4k5
+ pmaddubsw %2, k2k3k6k7
+
+ movdqa xmm4, %1
+ movdqa xmm5, %2
+ psrldq %1, 8
+ psrldq %2, 8
+ movdqa xmm6, xmm5
+
+ paddsw xmm4, %2
+ pmaxsw xmm5, %1
+ pminsw %1, xmm6
+ paddsw %1, xmm4
+ paddsw %1, xmm5
+
+ paddsw %1, krd
+ psraw %1, 7
+ packuswb %1, %1
+%endm
+
+%macro HORIZx4 1
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm4, [rdx] ;load filters
+ movq xmm5, rcx
+ packsswb xmm4, xmm4
+ pshuflw xmm6, xmm4, 0b ;k0_k1
+ pshufhw xmm6, xmm6, 10101010b ;k0_k1_k4_k5
+ pshuflw xmm7, xmm4, 01010101b ;k2_k3
+ pshufhw xmm7, xmm7, 11111111b ;k2_k3_k6_k7
+ pshufd xmm5, xmm5, 0 ;rounding
+
+ movdqa k0k1k4k5, xmm6
+ movdqa k2k3k6k7, xmm7
+ movdqa krd, xmm5
+
+ movsxd rax, dword ptr arg(1) ;src_pixels_per_line
+ movsxd rdx, dword ptr arg(3) ;output_pitch
+ movsxd rcx, dword ptr arg(4) ;output_height
+ shr rcx, 1
+.loop:
+ ;Do two rows once
+ movq xmm0, [rsi - 3] ;load src
+ movq xmm1, [rsi + 5]
+ movq xmm2, [rsi + rax - 3]
+ movq xmm3, [rsi + rax + 5]
+ punpcklqdq xmm0, xmm1
+ punpcklqdq xmm2, xmm3
+
+ HORIZx4_ROW xmm0, xmm1
+ HORIZx4_ROW xmm2, xmm3
+%if %1
+ movd xmm1, [rdi]
+ pavgb xmm0, xmm1
+ movd xmm3, [rdi + rdx]
+ pavgb xmm2, xmm3
+%endif
+ movd [rdi], xmm0
+ movd [rdi +rdx], xmm2
+
+ lea rsi, [rsi + rax]
+ prefetcht0 [rsi + 4 * rax - 3]
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + 2 * rdx]
+ prefetcht0 [rsi + 2 * rax - 3]
+
+ dec rcx
+ jnz .loop
+
+ ; Do last row if output_height is odd
+ movsxd rcx, dword ptr arg(4) ;output_height
+ and rcx, 1
+ je .done
+
+ movq xmm0, [rsi - 3] ; load src
+ movq xmm1, [rsi + 5]
+ punpcklqdq xmm0, xmm1
+
+ HORIZx4_ROW xmm0, xmm1
+%if %1
+ movd xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movd [rdi], xmm0
+.done
+%endm
+
+%macro HORIZx8_ROW 4
+ movdqa %2, %1
+ movdqa %3, %1
+ movdqa %4, %1
+
+ pshufb %1, [GLOBAL(shuf_t0t1)]
+ pshufb %2, [GLOBAL(shuf_t2t3)]
+ pshufb %3, [GLOBAL(shuf_t4t5)]
+ pshufb %4, [GLOBAL(shuf_t6t7)]
+
+ pmaddubsw %1, k0k1
+ pmaddubsw %2, k2k3
+ pmaddubsw %3, k4k5
+ pmaddubsw %4, k6k7
+
+ paddsw %1, %4
+ movdqa %4, %2
+ pmaxsw %2, %3
+ pminsw %3, %4
+ paddsw %1, %3
+ paddsw %1, %2
+
+ paddsw %1, krd
+ psraw %1, 7
+ packuswb %1, %1
+%endm
+
+%macro HORIZx8 1
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm4, [rdx] ;load filters
+ movq xmm5, rcx
+ packsswb xmm4, xmm4
+ pshuflw xmm0, xmm4, 0b ;k0_k1
+ pshuflw xmm1, xmm4, 01010101b ;k2_k3
+ pshuflw xmm2, xmm4, 10101010b ;k4_k5
+ pshuflw xmm3, xmm4, 11111111b ;k6_k7
+
+ punpcklqdq xmm0, xmm0
+ punpcklqdq xmm1, xmm1
+ punpcklqdq xmm2, xmm2
+ punpcklqdq xmm3, xmm3
+
+ movdqa k0k1, xmm0
+ movdqa k2k3, xmm1
+ pshufd xmm5, xmm5, 0
+ movdqa k4k5, xmm2
+ movdqa k6k7, xmm3
+ movdqa krd, xmm5
+
+ movsxd rax, dword ptr arg(1) ;src_pixels_per_line
+ movsxd rdx, dword ptr arg(3) ;output_pitch
+ movsxd rcx, dword ptr arg(4) ;output_height
+ shr rcx, 1
+
+.loop:
+ movq xmm0, [rsi - 3] ;load src
+ movq xmm3, [rsi + 5]
+ movq xmm4, [rsi + rax - 3]
+ movq xmm7, [rsi + rax + 5]
+ punpcklqdq xmm0, xmm3
+ punpcklqdq xmm4, xmm7
+
+ HORIZx8_ROW xmm0, xmm1, xmm2, xmm3
+ HORIZx8_ROW xmm4, xmm5, xmm6, xmm7
+%if %1
+ movq xmm1, [rdi]
+ movq xmm2, [rdi + rdx]
+ pavgb xmm0, xmm1
+ pavgb xmm4, xmm2
+%endif
+ movq [rdi], xmm0
+ movq [rdi + rdx], xmm4
+
+ lea rsi, [rsi + rax]
+ prefetcht0 [rsi + 4 * rax - 3]
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + 2 * rdx]
+ prefetcht0 [rsi + 2 * rax - 3]
+ dec rcx
+ jnz .loop
+
+ ;Do last row if output_height is odd
+ movsxd rcx, dword ptr arg(4) ;output_height
+ and rcx, 1
+ je .done
+
+ movq xmm0, [rsi - 3]
+ movq xmm3, [rsi + 5]
+ punpcklqdq xmm0, xmm3
+
+ HORIZx8_ROW xmm0, xmm1, xmm2, xmm3
+%if %1
+ movq xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movq [rdi], xmm0
+.done
+%endm
+
+%macro HORIZx16 1
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm4, [rdx] ;load filters
+ movq xmm5, rcx
+ packsswb xmm4, xmm4
+ pshuflw xmm0, xmm4, 0b ;k0_k1
+ pshuflw xmm1, xmm4, 01010101b ;k2_k3
+ pshuflw xmm2, xmm4, 10101010b ;k4_k5
+ pshuflw xmm3, xmm4, 11111111b ;k6_k7
+
+ punpcklqdq xmm0, xmm0
+ punpcklqdq xmm1, xmm1
+ punpcklqdq xmm2, xmm2
+ punpcklqdq xmm3, xmm3
+
+ movdqa k0k1, xmm0
+ movdqa k2k3, xmm1
+ pshufd xmm5, xmm5, 0
+ movdqa k4k5, xmm2
+ movdqa k6k7, xmm3
+ movdqa krd, xmm5
+
+ movsxd rax, dword ptr arg(1) ;src_pixels_per_line
+ movsxd rdx, dword ptr arg(3) ;output_pitch
+ movsxd rcx, dword ptr arg(4) ;output_height
+
+.loop:
+ prefetcht0 [rsi + 2 * rax -3]
+
+ movq xmm0, [rsi - 3] ;load src data
+ movq xmm4, [rsi + 5]
+ movq xmm6, [rsi + 13]
+ punpcklqdq xmm0, xmm4
+ punpcklqdq xmm4, xmm6
+
+ movdqa xmm7, xmm0
+
+ punpcklbw xmm7, xmm7
+ punpckhbw xmm0, xmm0
+ movdqa xmm1, xmm0
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm0
+
+ palignr xmm0, xmm7, 1
+ palignr xmm1, xmm7, 5
+ pmaddubsw xmm0, k0k1
+ palignr xmm2, xmm7, 9
+ pmaddubsw xmm1, k2k3
+ palignr xmm3, xmm7, 13
+
+ pmaddubsw xmm2, k4k5
+ pmaddubsw xmm3, k6k7
+ paddsw xmm0, xmm3
+
+ movdqa xmm3, xmm4
+ punpcklbw xmm3, xmm3
+ punpckhbw xmm4, xmm4
+
+ movdqa xmm5, xmm4
+ movdqa xmm6, xmm4
+ movdqa xmm7, xmm4
+
+ palignr xmm4, xmm3, 1
+ palignr xmm5, xmm3, 5
+ palignr xmm6, xmm3, 9
+ palignr xmm7, xmm3, 13
+
+ movdqa xmm3, xmm1
+ pmaddubsw xmm4, k0k1
+ pmaxsw xmm1, xmm2
+ pmaddubsw xmm5, k2k3
+ pminsw xmm2, xmm3
+ pmaddubsw xmm6, k4k5
+ paddsw xmm0, xmm2
+ pmaddubsw xmm7, k6k7
+ paddsw xmm0, xmm1
+
+ paddsw xmm4, xmm7
+ movdqa xmm7, xmm5
+ pmaxsw xmm5, xmm6
+ pminsw xmm6, xmm7
+ paddsw xmm4, xmm6
+ paddsw xmm4, xmm5
+
+ paddsw xmm0, krd
+ paddsw xmm4, krd
+ psraw xmm0, 7
+ psraw xmm4, 7
+ packuswb xmm0, xmm0
+ packuswb xmm4, xmm4
+ punpcklqdq xmm0, xmm4
+%if %1
+ movdqa xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+
+ lea rsi, [rsi + rax]
+ movdqa [rdi], xmm0
+
+ lea rdi, [rdi + rdx]
+ dec rcx
+ jnz .loop
+%endm
+
+;void vpx_filter_block1d4_h8_ssse3
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d4_h8_ssse3) PRIVATE
+sym(vpx_filter_block1d4_h8_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ GET_GOT rbx
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 3
+ %define k0k1k4k5 [rsp + 16 * 0]
+ %define k2k3k6k7 [rsp + 16 * 1]
+ %define krd [rsp + 16 * 2]
+
+ HORIZx4 0
+
+ add rsp, 16 * 3
+ pop rsp
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_GOT
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d8_h8_ssse3
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d8_h8_ssse3) PRIVATE
+sym(vpx_filter_block1d8_h8_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ GET_GOT rbx
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ HORIZx8 0
+
+ add rsp, 16*5
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_GOT
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vpx_filter_block1d16_h8_ssse3
+;(
+; unsigned char *src_ptr,
+; unsigned int src_pixels_per_line,
+; unsigned char *output_ptr,
+; unsigned int output_pitch,
+; unsigned int output_height,
+; short *filter
+;)
+global sym(vpx_filter_block1d16_h8_ssse3) PRIVATE
+sym(vpx_filter_block1d16_h8_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ GET_GOT rbx
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ HORIZx16 0
+
+ add rsp, 16*5
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_GOT
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_h8_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d4_h8_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ GET_GOT rbx
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16 * 3
+ %define k0k1k4k5 [rsp + 16 * 0]
+ %define k2k3k6k7 [rsp + 16 * 1]
+ %define krd [rsp + 16 * 2]
+
+ HORIZx4 1
+
+ add rsp, 16 * 3
+ pop rsp
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_GOT
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_h8_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d8_h8_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ GET_GOT rbx
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ HORIZx8 1
+
+ add rsp, 16*5
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_GOT
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_h8_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d16_h8_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ GET_GOT rbx
+ push rsi
+ push rdi
+ ; end prolog
+
+ ALIGN_STACK 16, rax
+ sub rsp, 16*5
+ %define k0k1 [rsp + 16*0]
+ %define k2k3 [rsp + 16*1]
+ %define k4k5 [rsp + 16*2]
+ %define k6k7 [rsp + 16*3]
+ %define krd [rsp + 16*4]
+
+ HORIZx16 1
+
+ add rsp, 16*5
+ pop rsp
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_GOT
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+SECTION_RODATA
+align 16
+shuf_t0t1:
+ db 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
+align 16
+shuf_t2t3:
+ db 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
+align 16
+shuf_t4t5:
+ db 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
+align 16
+shuf_t6t7:
+ db 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
--- /dev/null
+++ b/vpx_dsp/x86/vpx_subpixel_bilinear_sse2.asm
@@ -1,0 +1,448 @@
+;
+; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro GET_PARAM_4 0
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm3, [rdx] ;load filters
+ pshuflw xmm4, xmm3, 11111111b ;k3
+ psrldq xmm3, 8
+ pshuflw xmm3, xmm3, 0b ;k4
+ punpcklqdq xmm4, xmm3 ;k3k4
+
+ movq xmm3, rcx ;rounding
+ pshufd xmm3, xmm3, 0
+
+ pxor xmm2, xmm2
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+%endm
+
+%macro APPLY_FILTER_4 1
+
+ punpckldq xmm0, xmm1 ;two row in one register
+ punpcklbw xmm0, xmm2 ;unpack to word
+ pmullw xmm0, xmm4 ;multiply the filter factors
+
+ movdqa xmm1, xmm0
+ psrldq xmm1, 8
+ paddsw xmm0, xmm1
+
+ paddsw xmm0, xmm3 ;rounding
+ psraw xmm0, 7 ;shift
+ packuswb xmm0, xmm0 ;pack to byte
+
+%if %1
+ movd xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+
+ movd [rdi], xmm0
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+%endm
+
+%macro GET_PARAM 0
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm7, [rdx] ;load filters
+
+ pshuflw xmm6, xmm7, 11111111b ;k3
+ pshufhw xmm7, xmm7, 0b ;k4
+ punpcklwd xmm6, xmm6
+ punpckhwd xmm7, xmm7
+
+ movq xmm4, rcx ;rounding
+ pshufd xmm4, xmm4, 0
+
+ pxor xmm5, xmm5
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+%endm
+
+%macro APPLY_FILTER_8 1
+ punpcklbw xmm0, xmm5
+ punpcklbw xmm1, xmm5
+
+ pmullw xmm0, xmm6
+ pmullw xmm1, xmm7
+ paddsw xmm0, xmm1
+ paddsw xmm0, xmm4 ;rounding
+ psraw xmm0, 7 ;shift
+ packuswb xmm0, xmm0 ;pack back to byte
+%if %1
+ movq xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movq [rdi], xmm0 ;store the result
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+%endm
+
+%macro APPLY_FILTER_16 1
+ punpcklbw xmm0, xmm5
+ punpcklbw xmm1, xmm5
+ punpckhbw xmm2, xmm5
+ punpckhbw xmm3, xmm5
+
+ pmullw xmm0, xmm6
+ pmullw xmm1, xmm7
+ pmullw xmm2, xmm6
+ pmullw xmm3, xmm7
+
+ paddsw xmm0, xmm1
+ paddsw xmm2, xmm3
+
+ paddsw xmm0, xmm4 ;rounding
+ paddsw xmm2, xmm4
+ psraw xmm0, 7 ;shift
+ psraw xmm2, 7
+ packuswb xmm0, xmm2 ;pack back to byte
+%if %1
+ movdqu xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movdqu [rdi], xmm0 ;store the result
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+%endm
+
+global sym(vpx_filter_block1d4_v2_sse2) PRIVATE
+sym(vpx_filter_block1d4_v2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM_4
+.loop:
+ movd xmm0, [rsi] ;load src
+ movd xmm1, [rsi + rax]
+
+ APPLY_FILTER_4 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_v2_sse2) PRIVATE
+sym(vpx_filter_block1d8_v2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movq xmm0, [rsi] ;0
+ movq xmm1, [rsi + rax] ;1
+
+ APPLY_FILTER_8 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_v2_sse2) PRIVATE
+sym(vpx_filter_block1d16_v2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;0
+ movdqu xmm1, [rsi + rax] ;1
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm1
+
+ APPLY_FILTER_16 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_v2_avg_sse2) PRIVATE
+sym(vpx_filter_block1d4_v2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM_4
+.loop:
+ movd xmm0, [rsi] ;load src
+ movd xmm1, [rsi + rax]
+
+ APPLY_FILTER_4 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_v2_avg_sse2) PRIVATE
+sym(vpx_filter_block1d8_v2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movq xmm0, [rsi] ;0
+ movq xmm1, [rsi + rax] ;1
+
+ APPLY_FILTER_8 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_v2_avg_sse2) PRIVATE
+sym(vpx_filter_block1d16_v2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;0
+ movdqu xmm1, [rsi + rax] ;1
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm1
+
+ APPLY_FILTER_16 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_h2_sse2) PRIVATE
+sym(vpx_filter_block1d4_h2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM_4
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+
+ APPLY_FILTER_4 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_h2_sse2) PRIVATE
+sym(vpx_filter_block1d8_h2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+
+ APPLY_FILTER_8 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_h2_sse2) PRIVATE
+sym(vpx_filter_block1d16_h2_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqu xmm1, [rsi + 1]
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm1
+
+ APPLY_FILTER_16 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_h2_avg_sse2) PRIVATE
+sym(vpx_filter_block1d4_h2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM_4
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+
+ APPLY_FILTER_4 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_h2_avg_sse2) PRIVATE
+sym(vpx_filter_block1d8_h2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+
+ APPLY_FILTER_8 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_h2_avg_sse2) PRIVATE
+sym(vpx_filter_block1d16_h2_avg_sse2):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqu xmm1, [rsi + 1]
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm1
+
+ APPLY_FILTER_16 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
--- /dev/null
+++ b/vpx_dsp/x86/vpx_subpixel_bilinear_ssse3.asm
@@ -1,0 +1,422 @@
+;
+; Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+%include "vpx_ports/x86_abi_support.asm"
+
+%macro GET_PARAM_4 0
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm3, [rdx] ;load filters
+ psrldq xmm3, 6
+ packsswb xmm3, xmm3
+ pshuflw xmm3, xmm3, 0b ;k3_k4
+
+ movq xmm2, rcx ;rounding
+ pshufd xmm2, xmm2, 0
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+%endm
+
+%macro APPLY_FILTER_4 1
+ punpcklbw xmm0, xmm1
+ pmaddubsw xmm0, xmm3
+
+ paddsw xmm0, xmm2 ;rounding
+ psraw xmm0, 7 ;shift
+ packuswb xmm0, xmm0 ;pack to byte
+
+%if %1
+ movd xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movd [rdi], xmm0
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+%endm
+
+%macro GET_PARAM 0
+ mov rdx, arg(5) ;filter ptr
+ mov rsi, arg(0) ;src_ptr
+ mov rdi, arg(2) ;output_ptr
+ mov rcx, 0x0400040
+
+ movdqa xmm7, [rdx] ;load filters
+ psrldq xmm7, 6
+ packsswb xmm7, xmm7
+ pshuflw xmm7, xmm7, 0b ;k3_k4
+ punpcklwd xmm7, xmm7
+
+ movq xmm6, rcx ;rounding
+ pshufd xmm6, xmm6, 0
+
+ movsxd rax, DWORD PTR arg(1) ;pixels_per_line
+ movsxd rdx, DWORD PTR arg(3) ;out_pitch
+ movsxd rcx, DWORD PTR arg(4) ;output_height
+%endm
+
+%macro APPLY_FILTER_8 1
+ punpcklbw xmm0, xmm1
+ pmaddubsw xmm0, xmm7
+
+ paddsw xmm0, xmm6 ;rounding
+ psraw xmm0, 7 ;shift
+ packuswb xmm0, xmm0 ;pack back to byte
+
+%if %1
+ movq xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movq [rdi], xmm0 ;store the result
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+%endm
+
+%macro APPLY_FILTER_16 1
+ punpcklbw xmm0, xmm1
+ punpckhbw xmm2, xmm1
+ pmaddubsw xmm0, xmm7
+ pmaddubsw xmm2, xmm7
+
+ paddsw xmm0, xmm6 ;rounding
+ paddsw xmm2, xmm6
+ psraw xmm0, 7 ;shift
+ psraw xmm2, 7
+ packuswb xmm0, xmm2 ;pack back to byte
+
+%if %1
+ movdqu xmm1, [rdi]
+ pavgb xmm0, xmm1
+%endif
+ movdqu [rdi], xmm0 ;store the result
+
+ lea rsi, [rsi + rax]
+ lea rdi, [rdi + rdx]
+ dec rcx
+%endm
+
+global sym(vpx_filter_block1d4_v2_ssse3) PRIVATE
+sym(vpx_filter_block1d4_v2_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM_4
+.loop:
+ movd xmm0, [rsi] ;load src
+ movd xmm1, [rsi + rax]
+
+ APPLY_FILTER_4 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_v2_ssse3) PRIVATE
+sym(vpx_filter_block1d8_v2_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movq xmm0, [rsi] ;0
+ movq xmm1, [rsi + rax] ;1
+
+ APPLY_FILTER_8 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_v2_ssse3) PRIVATE
+sym(vpx_filter_block1d16_v2_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;0
+ movdqu xmm1, [rsi + rax] ;1
+ movdqa xmm2, xmm0
+
+ APPLY_FILTER_16 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_v2_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d4_v2_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM_4
+.loop:
+ movd xmm0, [rsi] ;load src
+ movd xmm1, [rsi + rax]
+
+ APPLY_FILTER_4 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_v2_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d8_v2_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movq xmm0, [rsi] ;0
+ movq xmm1, [rsi + rax] ;1
+
+ APPLY_FILTER_8 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_v2_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d16_v2_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;0
+ movdqu xmm1, [rsi + rax] ;1
+ movdqa xmm2, xmm0
+
+ APPLY_FILTER_16 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_h2_ssse3) PRIVATE
+sym(vpx_filter_block1d4_h2_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM_4
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+
+ APPLY_FILTER_4 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_h2_ssse3) PRIVATE
+sym(vpx_filter_block1d8_h2_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+
+ APPLY_FILTER_8 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_h2_ssse3) PRIVATE
+sym(vpx_filter_block1d16_h2_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqu xmm1, [rsi + 1]
+ movdqa xmm2, xmm0
+
+ APPLY_FILTER_16 0
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d4_h2_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d4_h2_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM_4
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+
+ APPLY_FILTER_4 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d8_h2_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d8_h2_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+
+ APPLY_FILTER_8 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+global sym(vpx_filter_block1d16_h2_avg_ssse3) PRIVATE
+sym(vpx_filter_block1d16_h2_avg_ssse3):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 6
+ SAVE_XMM 7
+ push rsi
+ push rdi
+ ; end prolog
+
+ GET_PARAM
+.loop:
+ movdqu xmm0, [rsi] ;load src
+ movdqu xmm1, [rsi + 1]
+ movdqa xmm2, xmm0
+
+ APPLY_FILTER_16 1
+ jnz .loop
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ RESTORE_XMM
+ UNSHADOW_ARGS
+ pop rbp
+ ret