ref: c91dd55eda7717d7a5f9726977138fb430cbbcfb
parent: f075fdc474898329eaa37ccc47433d6ad45792de
author: Jian Zhou <[email protected]>
date: Fri Dec 18 10:25:27 EST 2015
Code clean of highbd_v_predictor_4x4 MMX replaced with SSE2, same performance. Change-Id: I2ab8f30a71e5fadbbc172fb385093dec1e11a696
--- a/test/vp9_intrapred_test.cc
+++ b/test/vp9_intrapred_test.cc
@@ -147,7 +147,7 @@
&vpx_highbd_dc_predictor_8x8_c, 8, 8),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16, 8),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 8),
@@ -168,7 +168,7 @@
&vpx_highbd_dc_predictor_8x8_c, 8, 8),
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16, 8),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 8),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 8),
@@ -201,7 +201,7 @@
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
10),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 10),
@@ -225,7 +225,7 @@
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
10),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 10),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 10),
@@ -258,7 +258,7 @@
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
12),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 12),
@@ -282,7 +282,7 @@
make_tuple(&vpx_highbd_dc_predictor_16x16_sse2,
&vpx_highbd_dc_predictor_16x16_c, 16,
12),
- make_tuple(&vpx_highbd_v_predictor_4x4_sse,
+ make_tuple(&vpx_highbd_v_predictor_4x4_sse2,
&vpx_highbd_v_predictor_4x4_c, 4, 12),
make_tuple(&vpx_highbd_v_predictor_8x8_sse2,
&vpx_highbd_v_predictor_8x8_c, 8, 12),
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -288,7 +288,7 @@
specialize qw/vpx_highbd_d153_predictor_4x4/;
add_proto qw/void vpx_highbd_v_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_v_predictor_4x4/, "$sse_x86inc";
+ specialize qw/vpx_highbd_v_predictor_4x4/, "$sse2_x86inc";
add_proto qw/void vpx_highbd_tm_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
specialize qw/vpx_highbd_tm_predictor_4x4/, "$sse_x86inc";
--- a/vpx_dsp/x86/highbd_intrapred_sse2.asm
+++ b/vpx_dsp/x86/highbd_intrapred_sse2.asm
@@ -183,7 +183,7 @@
REP_RET
%endif
-INIT_MMX sse
+INIT_XMM sse2
cglobal highbd_v_predictor_4x4, 3, 3, 1, dst, stride, above
movq m0, [aboveq]
movq [dstq ], m0