ref: 1794624c18a5642c62e0e5f586a87e5af06ecd1f
parent: 78629508f2f57596a76b13adf1b9ff2affd6a801
parent: 89dcc139397ca2a0982cc40fa7067ff1903a688d
author: James Zern <[email protected]>
date: Fri Aug 14 15:45:27 EDT 2015
Merge changes I2fe52bfb,I5e5084eb * changes: VPX: removed filter == 128 checks from mips convolve code VPX: removed step checks from mips convolve code
--- a/vpx_dsp/mips/convolve2_avg_dspr2.c
+++ b/vpx_dsp/mips/convolve2_avg_dspr2.c
@@ -233,47 +233,41 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {- if (16 == y_step_q4) {- uint32_t pos = 38;
+ uint32_t pos = 38;
- /* bit positon for extract from acc */
- __asm__ __volatile__ (
- "wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r" (pos)
- );
+ assert(y_step_q4 == 16);
- prefetch_store(dst);
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
- switch (w) {- case 4:
- case 8:
- case 16:
- case 32:
- convolve_bi_avg_vert_4_dspr2(src, src_stride,
- dst, dst_stride,
- filter_y, w, h);
- break;
- case 64:
- prefetch_store(dst + 32);
- convolve_bi_avg_vert_64_dspr2(src, src_stride,
- dst, dst_stride,
- filter_y, 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 {- vpx_convolve8_avg_vert_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
+ prefetch_store(dst);
+
+ switch (w) {+ case 4:
+ case 8:
+ case 16:
+ case 32:
+ convolve_bi_avg_vert_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, w, h);
+ break;
+ case 64:
+ prefetch_store(dst + 32);
+ convolve_bi_avg_vert_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, 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;
}
}
#endif
--- a/vpx_dsp/mips/convolve2_avg_horiz_dspr2.c
+++ b/vpx_dsp/mips/convolve2_avg_horiz_dspr2.c
@@ -768,64 +768,58 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {- if (16 == x_step_q4) {- uint32_t pos = 38;
+ uint32_t pos = 38;
- /* bit positon for extract from acc */
- __asm__ __volatile__ (
- "wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r" (pos)
- );
+ assert(x_step_q4 == 16);
- /* prefetch data to cache memory */
- prefetch_load(src);
- prefetch_load(src + 32);
- prefetch_store(dst);
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
- switch (w) {- case 4:
- convolve_bi_avg_horiz_4_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h);
- break;
- case 8:
- convolve_bi_avg_horiz_8_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h);
- break;
- case 16:
- convolve_bi_avg_horiz_16_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h, 1);
- break;
- case 32:
- convolve_bi_avg_horiz_16_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h, 2);
- break;
- case 64:
- prefetch_load(src + 64);
- prefetch_store(dst + 32);
+ /* prefetch data to cache memory */
+ prefetch_load(src);
+ prefetch_load(src + 32);
+ prefetch_store(dst);
- convolve_bi_avg_horiz_64_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, 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 {- vpx_convolve8_avg_horiz_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
+ switch (w) {+ case 4:
+ convolve_bi_avg_horiz_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ case 8:
+ convolve_bi_avg_horiz_8_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ case 16:
+ convolve_bi_avg_horiz_16_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h, 1);
+ break;
+ case 32:
+ convolve_bi_avg_horiz_16_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h, 2);
+ break;
+ case 64:
+ prefetch_load(src + 64);
+ prefetch_store(dst + 32);
+
+ convolve_bi_avg_horiz_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, 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;
}
}
#endif
--- a/vpx_dsp/mips/convolve2_horiz_dspr2.c
+++ b/vpx_dsp/mips/convolve2_horiz_dspr2.c
@@ -646,66 +646,60 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {- if (16 == x_step_q4) {- uint32_t pos = 38;
+ uint32_t pos = 38;
- prefetch_load((const uint8_t *)filter_x);
+ assert(x_step_q4 == 16);
- /* bit positon for extract from acc */
- __asm__ __volatile__ (
- "wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r" (pos)
- );
+ prefetch_load((const uint8_t *)filter_x);
- /* prefetch data to cache memory */
- prefetch_load(src);
- prefetch_load(src + 32);
- prefetch_store(dst);
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
- switch (w) {- case 4:
- convolve_bi_horiz_4_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h);
- break;
- case 8:
- convolve_bi_horiz_8_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h);
- break;
- case 16:
- convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h, 1);
- break;
- case 32:
- convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h, 2);
- break;
- case 64:
- prefetch_load(src + 64);
- prefetch_store(dst + 32);
+ /* prefetch data to cache memory */
+ prefetch_load(src);
+ prefetch_load(src + 32);
+ prefetch_store(dst);
- convolve_bi_horiz_64_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)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 {- vpx_convolve8_horiz_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
+ switch (w) {+ case 4:
+ convolve_bi_horiz_4_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ case 8:
+ convolve_bi_horiz_8_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ case 16:
+ convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h, 1);
+ break;
+ case 32:
+ convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h, 2);
+ break;
+ case 64:
+ prefetch_load(src + 64);
+ prefetch_store(dst + 32);
+
+ convolve_bi_horiz_64_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)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;
}
}
#endif
--- a/vpx_dsp/mips/convolve2_vert_dspr2.c
+++ b/vpx_dsp/mips/convolve2_vert_dspr2.c
@@ -218,47 +218,41 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {- if (16 == y_step_q4) {- uint32_t pos = 38;
+ uint32_t pos = 38;
- /* bit positon for extract from acc */
- __asm__ __volatile__ (
- "wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r" (pos)
- );
+ assert(y_step_q4 == 16);
- prefetch_store(dst);
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
- switch (w) {- case 4 :
- case 8 :
- case 16 :
- case 32 :
- convolve_bi_vert_4_dspr2(src, src_stride,
- dst, dst_stride,
- filter_y, w, h);
- break;
- case 64 :
- prefetch_store(dst + 32);
- convolve_bi_vert_64_dspr2(src, src_stride,
- dst, dst_stride,
- filter_y, 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 {- vpx_convolve8_vert_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
+ prefetch_store(dst);
+
+ switch (w) {+ case 4 :
+ case 8 :
+ case 16 :
+ case 32 :
+ convolve_bi_vert_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, w, h);
+ break;
+ case 64 :
+ prefetch_store(dst + 32);
+ convolve_bi_vert_64_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, 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;
}
}
#endif
--- a/vpx_dsp/mips/convolve8_avg_dspr2.c
+++ b/vpx_dsp/mips/convolve8_avg_dspr2.c
@@ -347,13 +347,10 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {- 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);
- } else if (((const int32_t *)filter_y)[0] == 0) {+ assert(y_step_q4 == 16);
+ assert(((const int32_t *)filter_y)[1] != 0x800000);
+
+ if (((const int32_t *)filter_y)[0] == 0) {vpx_convolve2_avg_vert_dspr2(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
@@ -360,47 +357,39 @@
filter_y, y_step_q4,
w, h);
} else {- if (16 == y_step_q4) {- uint32_t pos = 38;
+ uint32_t pos = 38;
- /* bit positon for extract from acc */
- __asm__ __volatile__ (
- "wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r" (pos)
- );
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
- prefetch_store(dst);
+ prefetch_store(dst);
- switch (w) {- case 4:
- case 8:
- case 16:
- case 32:
- convolve_avg_vert_4_dspr2(src, src_stride,
- dst, dst_stride,
- filter_y, w, h);
- break;
- case 64:
- prefetch_store(dst + 32);
- convolve_avg_vert_64_dspr2(src, src_stride,
- dst, dst_stride,
- filter_y, h);
- break;
- default:
- vpx_convolve8_avg_vert_c(src, src_stride,
+ switch (w) {+ case 4:
+ case 8:
+ case 16:
+ case 32:
+ convolve_avg_vert_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, w, h);
+ break;
+ case 64:
+ prefetch_store(dst + 32);
+ convolve_avg_vert_64_dspr2(src, src_stride,
dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
- break;
- }
- } else {- vpx_convolve8_avg_vert_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
+ filter_y, 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;
}
}
}
@@ -416,16 +405,11 @@
assert(w <= 64);
assert(h <= 64);
+ assert(x_step_q4 == 16);
+ assert(y_step_q4 == 16);
if (intermediate_height < h)
intermediate_height = h;
-
- if (x_step_q4 != 16 || y_step_q4 != 16)
- return vpx_convolve8_avg_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
vpx_convolve8_horiz(src - (src_stride * 3), src_stride,
temp, 64,
--- a/vpx_dsp/mips/convolve8_avg_horiz_dspr2.c
+++ b/vpx_dsp/mips/convolve8_avg_horiz_dspr2.c
@@ -957,13 +957,10 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {- 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);
- } else if (((const int32_t *)filter_x)[0] == 0) {+ assert(x_step_q4 == 16);
+ assert(((const int32_t *)filter_x)[1] != 0x800000);
+
+ if (((const int32_t *)filter_x)[0] == 0) {vpx_convolve2_avg_horiz_dspr2(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
@@ -970,66 +967,58 @@
filter_y, y_step_q4,
w, h);
} else {- if (16 == x_step_q4) {- uint32_t pos = 38;
+ uint32_t pos = 38;
- src -= 3;
+ src -= 3;
- /* bit positon for extract from acc */
- __asm__ __volatile__ (
- "wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r" (pos)
- );
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
- /* prefetch data to cache memory */
- prefetch_load(src);
- prefetch_load(src + 32);
- prefetch_store(dst);
+ /* prefetch data to cache memory */
+ prefetch_load(src);
+ prefetch_load(src + 32);
+ prefetch_store(dst);
- switch (w) {- case 4:
- convolve_avg_horiz_4_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h);
- break;
- case 8:
- convolve_avg_horiz_8_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h);
- break;
- case 16:
- convolve_avg_horiz_16_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h, 1);
- break;
- case 32:
- convolve_avg_horiz_16_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h, 2);
- break;
- case 64:
- prefetch_load(src + 64);
- prefetch_store(dst + 32);
+ switch (w) {+ case 4:
+ convolve_avg_horiz_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ case 8:
+ convolve_avg_horiz_8_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h);
+ break;
+ case 16:
+ convolve_avg_horiz_16_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h, 1);
+ break;
+ case 32:
+ convolve_avg_horiz_16_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_x, h, 2);
+ break;
+ case 64:
+ prefetch_load(src + 64);
+ prefetch_store(dst + 32);
- convolve_avg_horiz_64_dspr2(src, src_stride,
- dst, dst_stride,
- filter_x, h);
- break;
- default:
- vpx_convolve8_avg_horiz_c(src + 3, src_stride,
+ convolve_avg_horiz_64_dspr2(src, src_stride,
dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
- break;
- }
- } else {- vpx_convolve8_avg_horiz_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
+ filter_x, h);
+ break;
+ default:
+ vpx_convolve8_avg_horiz_c(src + 3, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
}
}
}
--- a/vpx_dsp/mips/convolve8_dspr2.c
+++ b/vpx_dsp/mips/convolve8_dspr2.c
@@ -936,6 +936,12 @@
int32_t intermediate_height = ((h * y_step_q4) >> 4) + 7;
uint32_t pos = 38;
+ assert(x_step_q4 == 16);
+ assert(y_step_q4 == 16);
+ assert(((const int32_t *)filter_x)[1] != 0x800000);
+ assert(((const int32_t *)filter_y)[1] != 0x800000);
+
+
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
@@ -945,21 +951,6 @@
if (intermediate_height < h)
intermediate_height = h;
-
- if (x_step_q4 != 16 || y_step_q4 != 16)
- return vpx_convolve8_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
-
- if ((((const int32_t *)filter_x)[1] == 0x800000)
- && (((const int32_t *)filter_y)[1] == 0x800000))
- return vpx_convolve_copy(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
/* copy the src to dst */
if (filter_x[3] == 0x80) {--- a/vpx_dsp/mips/convolve8_horiz_dspr2.c
+++ b/vpx_dsp/mips/convolve8_horiz_dspr2.c
@@ -841,13 +841,10 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {- 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);
- } else if (((const int32_t *)filter_x)[0] == 0) {+ assert(x_step_q4 == 16);
+ assert(((const int32_t *)filter_x)[1] != 0x800000);
+
+ if (((const int32_t *)filter_x)[0] == 0) {vpx_convolve2_horiz_dspr2(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
@@ -854,67 +851,59 @@
filter_y, y_step_q4,
w, h);
} else {- if (16 == x_step_q4) {- uint32_t pos = 38;
+ uint32_t pos = 38;
- prefetch_load((const uint8_t *)filter_x);
- src -= 3;
+ prefetch_load((const uint8_t *)filter_x);
+ src -= 3;
- /* bit positon for extract from acc */
- __asm__ __volatile__ (
- "wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r" (pos)
- );
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
- /* prefetch data to cache memory */
- prefetch_load(src);
- prefetch_load(src + 32);
- prefetch_store(dst);
+ /* prefetch data to cache memory */
+ prefetch_load(src);
+ prefetch_load(src + 32);
+ prefetch_store(dst);
- switch (w) {- case 4:
- convolve_horiz_4_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h);
- break;
- case 8:
- convolve_horiz_8_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h);
- break;
- case 16:
- convolve_horiz_16_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h, 1);
- break;
- case 32:
- convolve_horiz_16_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h, 2);
- break;
- case 64:
- prefetch_load(src + 64);
- prefetch_store(dst + 32);
+ switch (w) {+ case 4:
+ convolve_horiz_4_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ case 8:
+ convolve_horiz_8_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ case 16:
+ convolve_horiz_16_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h, 1);
+ break;
+ case 32:
+ convolve_horiz_16_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h, 2);
+ break;
+ case 64:
+ prefetch_load(src + 64);
+ prefetch_store(dst + 32);
- convolve_horiz_64_dspr2(src, (int32_t)src_stride,
- dst, (int32_t)dst_stride,
- filter_x, (int32_t)h);
- break;
- default:
- vpx_convolve8_horiz_c(src + 3, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
- break;
- }
- } else {- vpx_convolve8_horiz_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
+ convolve_horiz_64_dspr2(src, (int32_t)src_stride,
+ dst, (int32_t)dst_stride,
+ filter_x, (int32_t)h);
+ break;
+ default:
+ vpx_convolve8_horiz_c(src + 3, src_stride,
+ dst, dst_stride,
+ filter_x, x_step_q4,
+ filter_y, y_step_q4,
+ w, h);
+ break;
}
}
}
--- a/vpx_dsp/mips/convolve8_vert_dspr2.c
+++ b/vpx_dsp/mips/convolve8_vert_dspr2.c
@@ -333,13 +333,10 @@
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {- 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);
- } else if (((const int32_t *)filter_y)[0] == 0) {+ assert(y_step_q4 == 16);
+ assert(((const int32_t *)filter_y)[1] != 0x800000);
+
+ if (((const int32_t *)filter_y)[0] == 0) {vpx_convolve2_vert_dspr2(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
@@ -346,47 +343,39 @@
filter_y, y_step_q4,
w, h);
} else {- if (16 == y_step_q4) {- uint32_t pos = 38;
+ uint32_t pos = 38;
- /* bit positon for extract from acc */
- __asm__ __volatile__ (
- "wrdsp %[pos], 1 \n\t"
- :
- : [pos] "r" (pos)
- );
+ /* bit positon for extract from acc */
+ __asm__ __volatile__ (
+ "wrdsp %[pos], 1 \n\t"
+ :
+ : [pos] "r" (pos)
+ );
- prefetch_store(dst);
+ prefetch_store(dst);
- switch (w) {- case 4 :
- case 8 :
- case 16 :
- case 32 :
- convolve_vert_4_dspr2(src, src_stride,
- dst, dst_stride,
- filter_y, w, h);
- break;
- case 64 :
- prefetch_store(dst + 32);
- convolve_vert_64_dspr2(src, src_stride,
- dst, dst_stride,
- filter_y, h);
- break;
- default:
- vpx_convolve8_vert_c(src, src_stride,
+ switch (w) {+ case 4 :
+ case 8 :
+ case 16 :
+ case 32 :
+ convolve_vert_4_dspr2(src, src_stride,
+ dst, dst_stride,
+ filter_y, w, h);
+ break;
+ case 64 :
+ prefetch_store(dst + 32);
+ convolve_vert_64_dspr2(src, src_stride,
dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
- break;
- }
- } else {- vpx_convolve8_vert_c(src, src_stride,
- dst, dst_stride,
- filter_x, x_step_q4,
- filter_y, y_step_q4,
- w, h);
+ filter_y, 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;
}
}
}
--- a/vpx_dsp/mips/vpx_convolve8_avg_horiz_msa.c
+++ b/vpx_dsp/mips/vpx_convolve8_avg_horiz_msa.c
@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <assert.h>
#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -665,19 +666,8 @@
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;
- }
+ assert(x_step_q4 == 16);
+ assert(((const int32_t *)filter_x)[1] != 0x800000);
for (cnt = 0; cnt < 8; ++cnt) {filt_hor[cnt] = filter_x[cnt];
--- a/vpx_dsp/mips/vpx_convolve8_avg_msa.c
+++ b/vpx_dsp/mips/vpx_convolve8_avg_msa.c
@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <assert.h>
#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -574,20 +575,10 @@
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;
- }
+ assert(x_step_q4 == 16);
+ assert(y_step_q4 == 16);
+ assert(((const int32_t *)filter_x)[1] != 0x800000);
+ assert(((const int32_t *)filter_y)[1] != 0x800000);
for (cnt = 0; cnt < 8; ++cnt) {filt_hor[cnt] = filter_x[cnt];
--- a/vpx_dsp/mips/vpx_convolve8_avg_vert_msa.c
+++ b/vpx_dsp/mips/vpx_convolve8_avg_vert_msa.c
@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <assert.h>
#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -639,19 +640,8 @@
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;
- }
+ assert(y_step_q4 == 16);
+ assert(((const int32_t *)filter_y)[1] != 0x800000);
for (cnt = 0; cnt < 8; ++cnt) {filt_ver[cnt] = filter_y[cnt];
--- a/vpx_dsp/mips/vpx_convolve8_horiz_msa.c
+++ b/vpx_dsp/mips/vpx_convolve8_horiz_msa.c
@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <assert.h>
#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -625,19 +626,8 @@
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;
- }
+ assert(x_step_q4 == 16);
+ assert(((const int32_t *)filter_x)[1] != 0x800000);
for (cnt = 0; cnt < 8; ++cnt) {filt_hor[cnt] = filter_x[cnt];
--- a/vpx_dsp/mips/vpx_convolve8_msa.c
+++ b/vpx_dsp/mips/vpx_convolve8_msa.c
@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <assert.h>
#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -548,20 +549,10 @@
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;
- }
+ assert(x_step_q4 == 16);
+ assert(y_step_q4 == 16);
+ assert(((const int32_t *)filter_x)[1] != 0x800000);
+ assert(((const int32_t *)filter_y)[1] != 0x800000);
for (cnt = 0; cnt < 8; ++cnt) {filt_hor[cnt] = filter_x[cnt];
--- a/vpx_dsp/mips/vpx_convolve8_vert_msa.c
+++ b/vpx_dsp/mips/vpx_convolve8_vert_msa.c
@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
+#include <assert.h>
#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -632,19 +633,8 @@
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;
- }
+ assert(y_step_q4 == 16);
+ assert(((const int32_t *)filter_y)[1] != 0x800000);
for (cnt = 8; cnt--;) {filt_ver[cnt] = filter_y[cnt];