generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

VPX: removed step checks from mips convolve code

Browse Source 
The check is handled by the predictor table.

Change-Id: I5e5084ebb46be8087c8c9d80b5f76e919a1cd05b
This commit is contained in:
Scott LaVarnway
2015-08-13 11:27:04 -07:00
parent 1aa84e03fd
commit aeea00cc4f
15 changed files with 337 additions and 420 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
vpx_dsp/mips/convolve2_avg_dspr2.c
View File

@@ -233,47 +233,41 @@ void vpx_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
const int16_t *filter_x, int x_step_q4, const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) { int w, int h) {
if (16 == y_step_q4) { uint32_t pos = 38;
uint32_t pos = 38;
/* bit positon for extract from acc */ assert(y_step_q4 == 16);
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
prefetch_store(dst); /* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
switch (w) { prefetch_store(dst);
case 4:
case 8: switch (w) {
case 16: case 4:
case 32: case 8:
convolve_bi_avg_vert_4_dspr2(src, src_stride, case 16:
dst, dst_stride, case 32:
filter_y, w, h); convolve_bi_avg_vert_4_dspr2(src, src_stride,
break; dst, dst_stride,
case 64: filter_y, w, h);
prefetch_store(dst + 32); break;
convolve_bi_avg_vert_64_dspr2(src, src_stride, case 64:
dst, dst_stride, prefetch_store(dst + 32);
filter_y, h); convolve_bi_avg_vert_64_dspr2(src, src_stride,
break; dst, dst_stride,
default: filter_y, h);
vpx_convolve8_avg_vert_c(src, src_stride, break;
dst, dst_stride, default:
filter_x, x_step_q4, vpx_convolve8_avg_vert_c(src, src_stride,
filter_y, y_step_q4, dst, dst_stride,
w, h); filter_x, x_step_q4,
break; filter_y, y_step_q4,
} w, h);
} else { break;
vpx_convolve8_avg_vert_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
} }
} }
#endif #endif

102
vpx_dsp/mips/convolve2_avg_horiz_dspr2.c
View File

@@ -768,64 +768,58 @@ void vpx_convolve2_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
const int16_t *filter_x, int x_step_q4, const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) { int w, int h) {
if (16 == x_step_q4) { uint32_t pos = 38;
uint32_t pos = 38;
/* bit positon for extract from acc */ assert(x_step_q4 == 16);
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
/* prefetch data to cache memory */ /* bit positon for extract from acc */
prefetch_load(src); __asm__ __volatile__ (
prefetch_load(src + 32); "wrdsp %[pos], 1 \n\t"
prefetch_store(dst); :
: [pos] "r" (pos)
);
switch (w) { /* prefetch data to cache memory */
case 4: prefetch_load(src);
convolve_bi_avg_horiz_4_dspr2(src, src_stride, prefetch_load(src + 32);
dst, dst_stride, prefetch_store(dst);
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, switch (w) {
dst, dst_stride, case 4:
filter_x, h); convolve_bi_avg_horiz_4_dspr2(src, src_stride,
break; dst, dst_stride,
default: filter_x, h);
vpx_convolve8_avg_horiz_c(src, src_stride, break;
dst, dst_stride, case 8:
filter_x, x_step_q4, convolve_bi_avg_horiz_8_dspr2(src, src_stride,
filter_y, y_step_q4, dst, dst_stride,
w, h); filter_x, h);
break; break;
} case 16:
} else { convolve_bi_avg_horiz_16_dspr2(src, src_stride,
vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride,
dst, dst_stride, filter_x, h, 1);
filter_x, x_step_q4, break;
filter_y, y_step_q4, case 32:
w, h); 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 #endif

104
vpx_dsp/mips/convolve2_horiz_dspr2.c
View File

@@ -646,66 +646,60 @@ void vpx_convolve2_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
const int16_t *filter_x, int x_step_q4, const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) { 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 */ prefetch_load((const uint8_t *)filter_x);
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
/* prefetch data to cache memory */ /* bit positon for extract from acc */
prefetch_load(src); __asm__ __volatile__ (
prefetch_load(src + 32); "wrdsp %[pos], 1 \n\t"
prefetch_store(dst); :
: [pos] "r" (pos)
);
switch (w) { /* prefetch data to cache memory */
case 4: prefetch_load(src);
convolve_bi_horiz_4_dspr2(src, (int32_t)src_stride, prefetch_load(src + 32);
dst, (int32_t)dst_stride, prefetch_store(dst);
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, switch (w) {
dst, (int32_t)dst_stride, case 4:
filter_x, (int32_t)h); convolve_bi_horiz_4_dspr2(src, (int32_t)src_stride,
break; dst, (int32_t)dst_stride,
default: filter_x, (int32_t)h);
vpx_convolve8_horiz_c(src, src_stride, break;
dst, dst_stride, case 8:
filter_x, x_step_q4, convolve_bi_horiz_8_dspr2(src, (int32_t)src_stride,
filter_y, y_step_q4, dst, (int32_t)dst_stride,
w, h); filter_x, (int32_t)h);
break; break;
} case 16:
} else { convolve_bi_horiz_16_dspr2(src, (int32_t)src_stride,
vpx_convolve8_horiz_c(src, src_stride, dst, (int32_t)dst_stride,
dst, dst_stride, filter_x, (int32_t)h, 1);
filter_x, x_step_q4, break;
filter_y, y_step_q4, case 32:
w, h); 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 #endif

70
vpx_dsp/mips/convolve2_vert_dspr2.c
View File

@@ -218,47 +218,41 @@ void vpx_convolve2_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
const int16_t *filter_x, int x_step_q4, const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) { int w, int h) {
if (16 == y_step_q4) { uint32_t pos = 38;
uint32_t pos = 38;
/* bit positon for extract from acc */ assert(y_step_q4 == 16);
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
prefetch_store(dst); /* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
switch (w) { prefetch_store(dst);
case 4 :
case 8 : switch (w) {
case 16 : case 4 :
case 32 : case 8 :
convolve_bi_vert_4_dspr2(src, src_stride, case 16 :
dst, dst_stride, case 32 :
filter_y, w, h); convolve_bi_vert_4_dspr2(src, src_stride,
break; dst, dst_stride,
case 64 : filter_y, w, h);
prefetch_store(dst + 32); break;
convolve_bi_vert_64_dspr2(src, src_stride, case 64 :
dst, dst_stride, prefetch_store(dst + 32);
filter_y, h); convolve_bi_vert_64_dspr2(src, src_stride,
break; dst, dst_stride,
default: filter_y, h);
vpx_convolve8_vert_c(src, src_stride, break;
dst, dst_stride, default:
filter_x, x_step_q4, vpx_convolve8_vert_c(src, src_stride,
filter_y, y_step_q4, dst, dst_stride,
w, h); filter_x, x_step_q4,
break; filter_y, y_step_q4,
} w, h);
} else { break;
vpx_convolve8_vert_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
} }
} }
#endif #endif

76
vpx_dsp/mips/convolve8_avg_dspr2.c
View File

@@ -347,6 +347,7 @@ void vpx_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
const int16_t *filter_x, int x_step_q4, const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) { int w, int h) {
assert(y_step_q4 == 16);
if (((const int32_t *)filter_y)[1] == 0x800000) { if (((const int32_t *)filter_y)[1] == 0x800000) {
vpx_convolve_avg(src, src_stride, vpx_convolve_avg(src, src_stride,
dst, dst_stride, dst, dst_stride,
@@ -360,47 +361,39 @@ void vpx_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
filter_y, y_step_q4, filter_y, y_step_q4,
w, h); w, h);
} else { } else {
if (16 == y_step_q4) { uint32_t pos = 38;
uint32_t pos = 38;
/* bit positon for extract from acc */ /* bit positon for extract from acc */
__asm__ __volatile__ ( __asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t" "wrdsp %[pos], 1 \n\t"
: :
: [pos] "r" (pos) : [pos] "r" (pos)
); );
prefetch_store(dst); prefetch_store(dst);
switch (w) { switch (w) {
case 4: case 4:
case 8: case 8:
case 16: case 16:
case 32: case 32:
convolve_avg_vert_4_dspr2(src, src_stride, convolve_avg_vert_4_dspr2(src, src_stride,
dst, dst_stride, dst, dst_stride,
filter_y, w, h); filter_y, w, h);
break; break;
case 64: case 64:
prefetch_store(dst + 32); prefetch_store(dst + 32);
convolve_avg_vert_64_dspr2(src, src_stride, convolve_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, dst, dst_stride,
filter_x, x_step_q4, filter_y, h);
filter_y, y_step_q4, break;
w, h); default:
break; vpx_convolve8_avg_vert_c(src, src_stride,
} dst, dst_stride,
} else { filter_x, x_step_q4,
vpx_convolve8_avg_vert_c(src, src_stride, filter_y, y_step_q4,
dst, dst_stride, w, h);
filter_x, x_step_q4, break;
filter_y, y_step_q4,
w, h);
} }
} }
} }
@@ -416,17 +409,12 @@ void vpx_convolve8_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
assert(w <= 64); assert(w <= 64);
assert(h <= 64); assert(h <= 64);
assert(x_step_q4 == 16);
assert(y_step_q4 == 16);
if (intermediate_height < h) if (intermediate_height < h)
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, vpx_convolve8_horiz(src - (src_stride * 3), src_stride,
temp, 64, temp, 64,
filter_x, x_step_q4, filter_x, x_step_q4,

103
vpx_dsp/mips/convolve8_avg_horiz_dspr2.c
View File

@@ -957,6 +957,7 @@ void vpx_convolve8_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
const int16_t *filter_x, int x_step_q4, const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) { int w, int h) {
assert(x_step_q4 == 16);
if (((const int32_t *)filter_x)[1] == 0x800000) { if (((const int32_t *)filter_x)[1] == 0x800000) {
vpx_convolve_avg(src, src_stride, vpx_convolve_avg(src, src_stride,
dst, dst_stride, dst, dst_stride,
@@ -970,66 +971,58 @@ void vpx_convolve8_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
filter_y, y_step_q4, filter_y, y_step_q4,
w, h); w, h);
} else { } else {
if (16 == x_step_q4) { uint32_t pos = 38;
uint32_t pos = 38;
src -= 3; src -= 3;
/* bit positon for extract from acc */ /* bit positon for extract from acc */
__asm__ __volatile__ ( __asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t" "wrdsp %[pos], 1 \n\t"
: :
: [pos] "r" (pos) : [pos] "r" (pos)
); );
/* prefetch data to cache memory */ /* prefetch data to cache memory */
prefetch_load(src); prefetch_load(src);
prefetch_load(src + 32); prefetch_load(src + 32);
prefetch_store(dst); prefetch_store(dst);
switch (w) { switch (w) {
case 4: case 4:
convolve_avg_horiz_4_dspr2(src, src_stride, convolve_avg_horiz_4_dspr2(src, src_stride,
dst, dst_stride, dst, dst_stride,
filter_x, h); filter_x, h);
break; break;
case 8: case 8:
convolve_avg_horiz_8_dspr2(src, src_stride, convolve_avg_horiz_8_dspr2(src, src_stride,
dst, dst_stride, dst, dst_stride,
filter_x, h); filter_x, h);
break; break;
case 16: case 16:
convolve_avg_horiz_16_dspr2(src, src_stride, 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,
dst, dst_stride, dst, dst_stride,
filter_x, x_step_q4, filter_x, h, 1);
filter_y, y_step_q4, break;
w, h); case 32:
break; convolve_avg_horiz_16_dspr2(src, src_stride,
} dst, dst_stride,
} else { filter_x, h, 2);
vpx_convolve8_avg_horiz_c(src, src_stride, break;
dst, dst_stride, case 64:
filter_x, x_step_q4, prefetch_load(src + 64);
filter_y, y_step_q4, prefetch_store(dst + 32);
w, h);
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,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
break;
} }
} }
} }

10
vpx_dsp/mips/convolve8_dspr2.c
View File

@@ -936,6 +936,9 @@ void vpx_convolve8_dspr2(const uint8_t *src, ptrdiff_t src_stride,
int32_t intermediate_height = ((h * y_step_q4) >> 4) + 7; int32_t intermediate_height = ((h * y_step_q4) >> 4) + 7;
uint32_t pos = 38; uint32_t pos = 38;
assert(x_step_q4 == 16);
assert(y_step_q4 == 16);
/* bit positon for extract from acc */ /* bit positon for extract from acc */
__asm__ __volatile__ ( __asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t" "wrdsp %[pos], 1 \n\t"
@@ -946,13 +949,6 @@ void vpx_convolve8_dspr2(const uint8_t *src, ptrdiff_t src_stride,
if (intermediate_height < h) if (intermediate_height < h)
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) if ((((const int32_t *)filter_x)[1] == 0x800000)
&& (((const int32_t *)filter_y)[1] == 0x800000)) && (((const int32_t *)filter_y)[1] == 0x800000))
return vpx_convolve_copy(src, src_stride, return vpx_convolve_copy(src, src_stride,

105
vpx_dsp/mips/convolve8_horiz_dspr2.c
View File

@@ -841,6 +841,7 @@ void vpx_convolve8_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
const int16_t *filter_x, int x_step_q4, const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) { int w, int h) {
assert(x_step_q4 == 16);
if (((const int32_t *)filter_x)[1] == 0x800000) { if (((const int32_t *)filter_x)[1] == 0x800000) {
vpx_convolve_copy(src, src_stride, vpx_convolve_copy(src, src_stride,
dst, dst_stride, dst, dst_stride,
@@ -854,67 +855,59 @@ void vpx_convolve8_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
filter_y, y_step_q4, filter_y, y_step_q4,
w, h); w, h);
} else { } else {
if (16 == x_step_q4) { uint32_t pos = 38;
uint32_t pos = 38;
prefetch_load((const uint8_t *)filter_x); prefetch_load((const uint8_t *)filter_x);
src -= 3; src -= 3;
/* bit positon for extract from acc */ /* bit positon for extract from acc */
__asm__ __volatile__ ( __asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t" "wrdsp %[pos], 1 \n\t"
: :
: [pos] "r" (pos) : [pos] "r" (pos)
); );
/* prefetch data to cache memory */ /* prefetch data to cache memory */
prefetch_load(src); prefetch_load(src);
prefetch_load(src + 32); prefetch_load(src + 32);
prefetch_store(dst); prefetch_store(dst);
switch (w) { switch (w) {
case 4: case 4:
convolve_horiz_4_dspr2(src, (int32_t)src_stride, convolve_horiz_4_dspr2(src, (int32_t)src_stride,
dst, (int32_t)dst_stride, dst, (int32_t)dst_stride,
filter_x, (int32_t)h); filter_x, (int32_t)h);
break; break;
case 8: case 8:
convolve_horiz_8_dspr2(src, (int32_t)src_stride, convolve_horiz_8_dspr2(src, (int32_t)src_stride,
dst, (int32_t)dst_stride, dst, (int32_t)dst_stride,
filter_x, (int32_t)h); filter_x, (int32_t)h);
break; break;
case 16: case 16:
convolve_horiz_16_dspr2(src, (int32_t)src_stride, convolve_horiz_16_dspr2(src, (int32_t)src_stride,
dst, (int32_t)dst_stride, dst, (int32_t)dst_stride,
filter_x, (int32_t)h, 1); filter_x, (int32_t)h, 1);
break; break;
case 32: case 32:
convolve_horiz_16_dspr2(src, (int32_t)src_stride, convolve_horiz_16_dspr2(src, (int32_t)src_stride,
dst, (int32_t)dst_stride, dst, (int32_t)dst_stride,
filter_x, (int32_t)h, 2); filter_x, (int32_t)h, 2);
break; break;
case 64: case 64:
prefetch_load(src + 64); prefetch_load(src + 64);
prefetch_store(dst + 32); prefetch_store(dst + 32);
convolve_horiz_64_dspr2(src, (int32_t)src_stride, convolve_horiz_64_dspr2(src, (int32_t)src_stride,
dst, (int32_t)dst_stride, dst, (int32_t)dst_stride,
filter_x, (int32_t)h); filter_x, (int32_t)h);
break; break;
default: default:
vpx_convolve8_horiz_c(src + 3, src_stride, vpx_convolve8_horiz_c(src + 3, src_stride,
dst, dst_stride, dst, dst_stride,
filter_x, x_step_q4, filter_x, x_step_q4,
filter_y, y_step_q4, filter_y, y_step_q4,
w, h); w, h);
break; break;
}
} else {
vpx_convolve8_horiz_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
} }
} }
} }

67
vpx_dsp/mips/convolve8_vert_dspr2.c
View File

@@ -333,6 +333,7 @@ void vpx_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
const int16_t *filter_x, int x_step_q4, const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) { int w, int h) {
assert(y_step_q4 == 16);
if (((const int32_t *)filter_y)[1] == 0x800000) { if (((const int32_t *)filter_y)[1] == 0x800000) {
vpx_convolve_copy(src, src_stride, vpx_convolve_copy(src, src_stride,
dst, dst_stride, dst, dst_stride,
@@ -346,47 +347,39 @@ void vpx_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
filter_y, y_step_q4, filter_y, y_step_q4,
w, h); w, h);
} else { } else {
if (16 == y_step_q4) { uint32_t pos = 38;
uint32_t pos = 38;
/* bit positon for extract from acc */ /* bit positon for extract from acc */
__asm__ __volatile__ ( __asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t" "wrdsp %[pos], 1 \n\t"
: :
: [pos] "r" (pos) : [pos] "r" (pos)
); );
prefetch_store(dst); prefetch_store(dst);
switch (w) { switch (w) {
case 4 : case 4 :
case 8 : case 8 :
case 16 : case 16 :
case 32 : case 32 :
convolve_vert_4_dspr2(src, src_stride, convolve_vert_4_dspr2(src, src_stride,
dst, dst_stride, dst, dst_stride,
filter_y, w, h); filter_y, w, h);
break; break;
case 64 : case 64 :
prefetch_store(dst + 32); prefetch_store(dst + 32);
convolve_vert_64_dspr2(src, src_stride, convolve_vert_64_dspr2(src, src_stride,
dst, dst_stride,
filter_y, h);
break;
default:
vpx_convolve8_vert_c(src, src_stride,
dst, dst_stride, dst, dst_stride,
filter_x, x_step_q4, filter_y, h);
filter_y, y_step_q4, break;
w, h); default:
break; vpx_convolve8_vert_c(src, src_stride,
} dst, dst_stride,
} else { filter_x, x_step_q4,
vpx_convolve8_vert_c(src, src_stride, filter_y, y_step_q4,
dst, dst_stride, w, h);
filter_x, x_step_q4, break;
filter_y, y_step_q4,
w, h);
} }
} }
} }

8
vpx_dsp/mips/vpx_convolve8_avg_horiz_msa.c
View File

@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree. * be found in the AUTHORS file in the root of the source tree.
*/ */
#include <assert.h>
#include "./vpx_dsp_rtcd.h" #include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h" #include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -665,12 +666,7 @@ void vpx_convolve8_avg_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
int w, int h) { int w, int h) {
int8_t cnt, filt_hor[8]; int8_t cnt, filt_hor[8];
if (16 != x_step_q4) { assert(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;
}
if (((const int32_t *)filter_x)[1] == 0x800000) { if (((const int32_t *)filter_x)[1] == 0x800000) {
vpx_convolve_avg(src, src_stride, dst, dst_stride, vpx_convolve_avg(src, src_stride, dst, dst_stride,

9
vpx_dsp/mips/vpx_convolve8_avg_msa.c
View File

@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree. * be found in the AUTHORS file in the root of the source tree.
*/ */
#include <assert.h>
#include "./vpx_dsp_rtcd.h" #include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h" #include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -574,12 +575,8 @@ void vpx_convolve8_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
int w, int h) { int w, int h) {
int8_t cnt, filt_hor[8], filt_ver[8]; int8_t cnt, filt_hor[8], filt_ver[8];
if (16 != x_step_q4 || 16 != y_step_q4) { assert(x_step_q4 == 16);
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, assert(y_step_q4 == 16);
filter_x, x_step_q4, filter_y, y_step_q4,
w, h);
return;
}
if (((const int32_t *)filter_x)[1] == 0x800000 && if (((const int32_t *)filter_x)[1] == 0x800000 &&
((const int32_t *)filter_y)[1] == 0x800000) { ((const int32_t *)filter_y)[1] == 0x800000) {

8
vpx_dsp/mips/vpx_convolve8_avg_vert_msa.c
View File

@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree. * be found in the AUTHORS file in the root of the source tree.
*/ */
#include <assert.h>
#include "./vpx_dsp_rtcd.h" #include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h" #include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -639,12 +640,7 @@ void vpx_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
int w, int h) { int w, int h) {
int8_t cnt, filt_ver[8]; int8_t cnt, filt_ver[8];
if (16 != y_step_q4) { assert(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;
}
if (((const int32_t *)filter_y)[1] == 0x800000) { if (((const int32_t *)filter_y)[1] == 0x800000) {
vpx_convolve_avg(src, src_stride, dst, dst_stride, vpx_convolve_avg(src, src_stride, dst, dst_stride,

8
vpx_dsp/mips/vpx_convolve8_horiz_msa.c
View File

@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree. * be found in the AUTHORS file in the root of the source tree.
*/ */
#include <assert.h>
#include "./vpx_dsp_rtcd.h" #include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h" #include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -625,12 +626,7 @@ void vpx_convolve8_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
int w, int h) { int w, int h) {
int8_t cnt, filt_hor[8]; int8_t cnt, filt_hor[8];
if (16 != x_step_q4) { assert(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;
}
if (((const int32_t *)filter_x)[1] == 0x800000) { if (((const int32_t *)filter_x)[1] == 0x800000) {
vpx_convolve_copy(src, src_stride, dst, dst_stride, vpx_convolve_copy(src, src_stride, dst, dst_stride,

9
vpx_dsp/mips/vpx_convolve8_msa.c
View File

@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree. * be found in the AUTHORS file in the root of the source tree.
*/ */
#include <assert.h>
#include "./vpx_dsp_rtcd.h" #include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h" #include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -548,12 +549,8 @@ void vpx_convolve8_msa(const uint8_t *src, ptrdiff_t src_stride,
int32_t w, int32_t h) { int32_t w, int32_t h) {
int8_t cnt, filt_hor[8], filt_ver[8]; int8_t cnt, filt_hor[8], filt_ver[8];
if (16 != x_step_q4 || 16 != y_step_q4) { assert(x_step_q4 == 16);
vpx_convolve8_c(src, src_stride, dst, dst_stride, assert(y_step_q4 == 16);
filter_x, x_step_q4, filter_y, y_step_q4,
w, h);
return;
}
if (((const int32_t *)filter_x)[1] == 0x800000 && if (((const int32_t *)filter_x)[1] == 0x800000 &&
((const int32_t *)filter_y)[1] == 0x800000) { ((const int32_t *)filter_y)[1] == 0x800000) {

8
vpx_dsp/mips/vpx_convolve8_vert_msa.c
View File

@@ -8,6 +8,7 @@
* be found in the AUTHORS file in the root of the source tree. * be found in the AUTHORS file in the root of the source tree.
*/ */
#include <assert.h>
#include "./vpx_dsp_rtcd.h" #include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/mips/vpx_convolve_msa.h" #include "vpx_dsp/mips/vpx_convolve_msa.h"
@@ -632,12 +633,7 @@ void vpx_convolve8_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
int w, int h) { int w, int h) {
int8_t cnt, filt_ver[8]; int8_t cnt, filt_ver[8];
if (16 != y_step_q4) { assert(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;
}
if (((const int32_t *)filter_y)[1] == 0x800000) { if (((const int32_t *)filter_y)[1] == 0x800000) {
vpx_convolve_copy(src, src_stride, dst, dst_stride, vpx_convolve_copy(src, src_stride, dst, dst_stride,