VPX: removed step checks from mips convolve code

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

View File

@@ -233,9 +233,10 @@ 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;
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"
@@ -268,12 +269,5 @@ void vpx_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
w, h); w, h);
break; 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);
}
} }
#endif #endif

View File

@@ -768,9 +768,10 @@ 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;
assert(x_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"
@@ -820,12 +821,5 @@ void vpx_convolve2_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
w, h); w, h);
break; 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);
}
} }
#endif #endif

View File

@@ -646,9 +646,10 @@ 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;
assert(x_step_q4 == 16);
prefetch_load((const uint8_t *)filter_x); prefetch_load((const uint8_t *)filter_x);
/* bit positon for extract from acc */ /* bit positon for extract from acc */
@@ -700,12 +701,5 @@ void vpx_convolve2_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
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);
}
} }
#endif #endif

View File

@@ -218,9 +218,10 @@ 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;
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"
@@ -253,12 +254,5 @@ void vpx_convolve2_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
w, h); w, h);
break; break;
} }
} else {
vpx_convolve8_vert_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
}
} }
#endif #endif

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,
@@ -360,7 +361,6 @@ 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 */
@@ -395,13 +395,6 @@ void vpx_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
w, h); w, h);
break; 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);
}
} }
} }
@@ -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,

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,
@@ -970,7 +971,6 @@ 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;
@@ -1024,13 +1024,6 @@ void vpx_convolve8_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
w, h); w, h);
break; 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);
}
} }
} }
#endif #endif

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,

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,
@@ -854,7 +855,6 @@ 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);
@@ -909,13 +909,6 @@ void vpx_convolve8_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
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);
}
} }
} }
#endif #endif

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,
@@ -346,7 +347,6 @@ 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 */
@@ -381,13 +381,6 @@ void vpx_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
w, h); w, h);
break; break;
} }
} else {
vpx_convolve8_vert_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
}
} }
} }

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,

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) {

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,

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,

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) {

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,