Commit aeea00cc authored by Scott LaVarnway's avatar Scott LaVarnway
Browse files

VPX: removed step checks from mips convolve code

The check is handled by the predictor table.

Change-Id: I5e5084ebb46be8087c8c9d80b5f76e919a1cd05b
parent 1aa84e03
......@@ -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_y, int y_step_q4,
int w, int h) {
if (16 == y_step_q4) {
uint32_t pos = 38;
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
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;
}
} else {
vpx_convolve8_avg_vert_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
uint32_t pos = 38;
assert(y_step_q4 == 16);
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
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
......@@ -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_y, int y_step_q4,
int w, int h) {
if (16 == x_step_q4) {
uint32_t pos = 38;
/* 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);
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;
}
} else {
vpx_convolve8_avg_horiz_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
uint32_t pos = 38;
assert(x_step_q4 == 16);
/* 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);
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
......@@ -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_y, int y_step_q4,
int w, int h) {
if (16 == x_step_q4) {
uint32_t pos = 38;
prefetch_load((const uint8_t *)filter_x);
/* 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);
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;
}
} else {
vpx_convolve8_horiz_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
uint32_t pos = 38;
assert(x_step_q4 == 16);
prefetch_load((const uint8_t *)filter_x);
/* 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);
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
......@@ -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_y, int y_step_q4,
int w, int h) {
if (16 == y_step_q4) {
uint32_t pos = 38;
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
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;
}
} else {
vpx_convolve8_vert_c(src, src_stride,
dst, dst_stride,
filter_x, x_step_q4,
filter_y, y_step_q4,
w, h);
uint32_t pos = 38;
assert(y_step_q4 == 16);
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
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
......@@ -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_y, int y_step_q4,
int w, int h) {
assert(y_step_q4 == 16);
if (((const int32_t *)filter_y)[1] == 0x800000) {
vpx_convolve_avg(src, src_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,
w, h);
} else {
if (16 == y_step_q4) {
uint32_t pos = 38;
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
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,
uint32_t pos = 38;
/* bit positon for extract from acc */
__asm__ __volatile__ (
"wrdsp %[pos], 1 \n\t"
:
: [pos] "r" (pos)
);
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_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;
}
}
}
......@@ -416,17 +409,12 @@ void vpx_convolve8_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
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,
filter_x, x_step_q4,
......
......@@ -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_y, int y_step_q4,
int w, int h) {
assert(x_step_q4 == 16);
if (((const int32_t *)filter_x)[1] == 0x800000) {
vpx_convolve_avg(src, src_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,
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);
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);
/* 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);
convolve_avg_horiz_64_dspr2(src, src_stride,