Remove get_filter_base() and get_filter_offset() in convolve

so that the convolve functions are independent of table alignment.

Change-Id: Ieab132a30d72c6e75bbe9473544fbe2cf51541ee
This commit is contained in:
Linfeng Zhang 2017-08-28 10:35:43 -07:00
parent d49a1a5329
commit d331e7a1c0
48 changed files with 760 additions and 816 deletions

View File

@ -33,9 +33,9 @@ static const unsigned int kMaxDimension = 64;
typedef void (*ConvolveFunc)(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride,
int w, int h);
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h);
typedef void (*WrapperFilterBlock2d8Func)(
const uint8_t *src_ptr, const unsigned int src_stride,
@ -550,7 +550,7 @@ TEST_P(ConvolveTest, DISABLED_Copy_Speed) {
vpx_usec_timer_start(&timer);
for (int n = 0; n < kNumTests; ++n) {
UUT_->copy_[0](in, kInputStride, out, kOutputStride, NULL, 0, NULL, 0,
UUT_->copy_[0](in, kInputStride, out, kOutputStride, NULL, 0, 0, 0, 0,
width, height);
}
vpx_usec_timer_mark(&timer);
@ -570,7 +570,7 @@ TEST_P(ConvolveTest, DISABLED_Avg_Speed) {
vpx_usec_timer_start(&timer);
for (int n = 0; n < kNumTests; ++n) {
UUT_->copy_[1](in, kInputStride, out, kOutputStride, NULL, 0, NULL, 0,
UUT_->copy_[1](in, kInputStride, out, kOutputStride, NULL, 0, 0, 0, 0,
width, height);
}
vpx_usec_timer_mark(&timer);
@ -585,7 +585,7 @@ TEST_P(ConvolveTest, Copy) {
uint8_t *const out = output();
ASM_REGISTER_STATE_CHECK(UUT_->copy_[0](in, kInputStride, out, kOutputStride,
NULL, 0, NULL, 0, Width(), Height()));
NULL, 0, 0, 0, 0, Width(), Height()));
CheckGuardBlocks();
@ -604,7 +604,7 @@ TEST_P(ConvolveTest, Avg) {
CopyOutputToRef();
ASM_REGISTER_STATE_CHECK(UUT_->copy_[1](in, kInputStride, out, kOutputStride,
NULL, 0, NULL, 0, Width(), Height()));
NULL, 0, 0, 0, 0, Width(), Height()));
CheckGuardBlocks();
@ -621,12 +621,10 @@ TEST_P(ConvolveTest, Avg) {
TEST_P(ConvolveTest, CopyHoriz) {
uint8_t *const in = input();
uint8_t *const out = output();
DECLARE_ALIGNED(256, const int16_t,
filter8[8]) = { 0, 0, 0, 128, 0, 0, 0, 0 };
ASM_REGISTER_STATE_CHECK(UUT_->sh8_[0](in, kInputStride, out, kOutputStride,
filter8, 16, filter8, 16, Width(),
Height()));
vp9_filter_kernels[0], 0, 16, 0, 16,
Width(), Height()));
CheckGuardBlocks();
@ -641,12 +639,10 @@ TEST_P(ConvolveTest, CopyHoriz) {
TEST_P(ConvolveTest, CopyVert) {
uint8_t *const in = input();
uint8_t *const out = output();
DECLARE_ALIGNED(256, const int16_t,
filter8[8]) = { 0, 0, 0, 128, 0, 0, 0, 0 };
ASM_REGISTER_STATE_CHECK(UUT_->sv8_[0](in, kInputStride, out, kOutputStride,
filter8, 16, filter8, 16, Width(),
Height()));
vp9_filter_kernels[0], 0, 16, 0, 16,
Width(), Height()));
CheckGuardBlocks();
@ -661,12 +657,10 @@ TEST_P(ConvolveTest, CopyVert) {
TEST_P(ConvolveTest, Copy2D) {
uint8_t *const in = input();
uint8_t *const out = output();
DECLARE_ALIGNED(256, const int16_t,
filter8[8]) = { 0, 0, 0, 128, 0, 0, 0, 0 };
ASM_REGISTER_STATE_CHECK(UUT_->shv8_[0](in, kInputStride, out, kOutputStride,
filter8, 16, filter8, 16, Width(),
Height()));
vp9_filter_kernels[0], 0, 16, 0, 16,
Width(), Height()));
CheckGuardBlocks();
@ -702,7 +696,6 @@ TEST(ConvolveTest, FiltersWontSaturateWhenAddedPairwise) {
}
}
const int16_t kInvalidFilter[8] = { 0 };
const WrapperFilterBlock2d8Func wrapper_filter_block2d_8[2] = {
wrapper_filter_block2d_8_c, wrapper_filter_average_block2d_8_c
};
@ -755,21 +748,21 @@ TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
Width(), Height(), UUT_->use_highbd_);
if (filter_x && filter_y)
ASM_REGISTER_STATE_CHECK(UUT_->hv8_[i](
in, kInputStride, out, kOutputStride, filters[filter_x], 16,
filters[filter_y], 16, Width(), Height()));
ASM_REGISTER_STATE_CHECK(
UUT_->hv8_[i](in, kInputStride, out, kOutputStride, filters,
filter_x, 16, filter_y, 16, Width(), Height()));
else if (filter_y)
ASM_REGISTER_STATE_CHECK(UUT_->v8_[i](
in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
filters[filter_y], 16, Width(), Height()));
ASM_REGISTER_STATE_CHECK(
UUT_->v8_[i](in, kInputStride, out, kOutputStride, filters, 0,
16, filter_y, 16, Width(), Height()));
else if (filter_x)
ASM_REGISTER_STATE_CHECK(UUT_->h8_[i](
in, kInputStride, out, kOutputStride, filters[filter_x], 16,
kInvalidFilter, 16, Width(), Height()));
ASM_REGISTER_STATE_CHECK(
UUT_->h8_[i](in, kInputStride, out, kOutputStride, filters,
filter_x, 16, 0, 16, Width(), Height()));
else
ASM_REGISTER_STATE_CHECK(UUT_->copy_[i](
in, kInputStride, out, kOutputStride, kInvalidFilter, 0,
kInvalidFilter, 0, Width(), Height()));
ASM_REGISTER_STATE_CHECK(UUT_->copy_[i](in, kInputStride, out,
kOutputStride, NULL, 0, 0,
0, 0, Width(), Height()));
CheckGuardBlocks();
@ -853,21 +846,21 @@ TEST_P(ConvolveTest, FilterExtremes) {
filters[filter_y], ref, kOutputStride,
Width(), Height(), UUT_->use_highbd_);
if (filter_x && filter_y)
ASM_REGISTER_STATE_CHECK(UUT_->hv8_[0](
in, kInputStride, out, kOutputStride, filters[filter_x], 16,
filters[filter_y], 16, Width(), Height()));
ASM_REGISTER_STATE_CHECK(
UUT_->hv8_[0](in, kInputStride, out, kOutputStride, filters,
filter_x, 16, filter_y, 16, Width(), Height()));
else if (filter_y)
ASM_REGISTER_STATE_CHECK(UUT_->v8_[0](
in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
filters[filter_y], 16, Width(), Height()));
ASM_REGISTER_STATE_CHECK(
UUT_->v8_[0](in, kInputStride, out, kOutputStride, filters, 0,
16, filter_y, 16, Width(), Height()));
else if (filter_x)
ASM_REGISTER_STATE_CHECK(UUT_->h8_[0](
in, kInputStride, out, kOutputStride, filters[filter_x], 16,
kInvalidFilter, 16, Width(), Height()));
ASM_REGISTER_STATE_CHECK(
UUT_->h8_[0](in, kInputStride, out, kOutputStride, filters,
filter_x, 16, 0, 16, Width(), Height()));
else
ASM_REGISTER_STATE_CHECK(UUT_->copy_[0](
in, kInputStride, out, kOutputStride, kInvalidFilter, 0,
kInvalidFilter, 0, Width(), Height()));
ASM_REGISTER_STATE_CHECK(UUT_->copy_[0](in, kInputStride, out,
kOutputStride, NULL, 0, 0,
0, 0, Width(), Height()));
for (int y = 0; y < Height(); ++y) {
for (int x = 0; x < Width(); ++x)
@ -897,8 +890,8 @@ TEST_P(ConvolveTest, CheckScalingFiltering) {
for (int step = 1; step <= 32; ++step) {
/* Test the horizontal and vertical filters in combination. */
ASM_REGISTER_STATE_CHECK(
UUT_->shv8_[0](in, kInputStride, out, kOutputStride, eighttap[frac],
step, eighttap[frac], step, Width(), Height()));
UUT_->shv8_[0](in, kInputStride, out, kOutputStride, eighttap, frac,
step, frac, step, Width(), Height()));
CheckGuardBlocks();
@ -917,14 +910,14 @@ TEST_P(ConvolveTest, CheckScalingFiltering) {
using std::tr1::make_tuple;
#if CONFIG_VP9_HIGHBITDEPTH
#define WRAP(func, bd) \
void wrap_##func##_##bd( \
const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \
ptrdiff_t dst_stride, const int16_t *filter_x, int filter_x_stride, \
const int16_t *filter_y, int filter_y_stride, int w, int h) { \
vpx_highbd_##func(reinterpret_cast<const uint16_t *>(src), src_stride, \
reinterpret_cast<uint16_t *>(dst), dst_stride, filter_x, \
filter_x_stride, filter_y, filter_y_stride, w, h, bd); \
#define WRAP(func, bd) \
void wrap_##func##_##bd( \
const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \
ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, \
int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \
vpx_highbd_##func(reinterpret_cast<const uint16_t *>(src), src_stride, \
reinterpret_cast<uint16_t *>(dst), dst_stride, filter, \
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, bd); \
}
#if HAVE_SSE2 && ARCH_X86_64

View File

@ -26,9 +26,9 @@ static INLINE void inter_predictor(const uint8_t *src, int src_stride,
const struct scale_factors *sf, int w, int h,
int ref, const InterpKernel *kernel, int xs,
int ys) {
sf->predict[subpel_x != 0][subpel_y != 0][ref](
src, src_stride, dst, dst_stride, kernel[subpel_x], xs, kernel[subpel_y],
ys, w, h);
sf->predict[subpel_x != 0][subpel_y != 0][ref](src, src_stride, dst,
dst_stride, kernel, subpel_x,
xs, subpel_y, ys, w, h);
}
#if CONFIG_VP9_HIGHBITDEPTH
@ -37,8 +37,8 @@ static INLINE void highbd_inter_predictor(
const int subpel_x, const int subpel_y, const struct scale_factors *sf,
int w, int h, int ref, const InterpKernel *kernel, int xs, int ys, int bd) {
sf->highbd_predict[subpel_x != 0][subpel_y != 0][ref](
src, src_stride, dst, dst_stride, kernel[subpel_x], xs, kernel[subpel_y],
ys, w, h, bd);
src, src_stride, dst, dst_stride, kernel, subpel_x, xs, subpel_y, ys, w,
h, bd);
}
#endif // CONFIG_VP9_HIGHBITDEPTH

View File

@ -390,12 +390,12 @@ void vp9_denoiser_denoise(VP9_COMP *cpi, MACROBLOCK *mb, int mi_row, int mi_col,
}
if (decision == FILTER_BLOCK) {
vpx_convolve_copy(avg_start, avg.y_stride, src.buf, src.stride, NULL, 0,
NULL, 0, num_4x4_blocks_wide_lookup[bs] << 2,
vpx_convolve_copy(avg_start, avg.y_stride, src.buf, src.stride, NULL, 0, 0,
0, 0, num_4x4_blocks_wide_lookup[bs] << 2,
num_4x4_blocks_high_lookup[bs] << 2);
} else { // COPY_BLOCK
vpx_convolve_copy(src.buf, src.stride, avg_start, avg.y_stride, NULL, 0,
NULL, 0, num_4x4_blocks_wide_lookup[bs] << 2,
vpx_convolve_copy(src.buf, src.stride, avg_start, avg.y_stride, NULL, 0, 0,
0, 0, num_4x4_blocks_wide_lookup[bs] << 2,
num_4x4_blocks_high_lookup[bs] << 2);
}
*denoiser_decision = decision;

View File

@ -2645,15 +2645,14 @@ static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
vpx_highbd_convolve8(CONVERT_TO_SHORTPTR(src_ptr), src_stride,
CONVERT_TO_SHORTPTR(dst_ptr), dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h,
16 / factor, 16 / factor, bd);
CONVERT_TO_SHORTPTR(dst_ptr), dst_stride, kernel,
x_q4 & 0xf, 16 * src_w / dst_w, y_q4 & 0xf,
16 * src_h / dst_h, 16 / factor, 16 / factor,
bd);
} else {
vpx_scaled_2d(src_ptr, src_stride, dst_ptr, dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h, 16 / factor,
16 / factor);
vpx_scaled_2d(src_ptr, src_stride, dst_ptr, dst_stride, kernel,
x_q4 & 0xf, 16 * src_w / dst_w, y_q4 & 0xf,
16 * src_h / dst_h, 16 / factor, 16 / factor);
}
}
}

View File

@ -43,10 +43,9 @@ void vp9_scale_and_extend_frame_c(const YV12_BUFFER_CONFIG *src,
(x / factor) * src_w / dst_w;
uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
vpx_scaled_2d(src_ptr, src_stride, dst_ptr, dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h, 16 / factor,
16 / factor);
vpx_scaled_2d(src_ptr, src_stride, dst_ptr, dst_stride, kernel,
x_q4 & 0xf, 16 * src_w / dst_w, y_q4 & 0xf,
16 * src_h / dst_h, 16 / factor, 16 / factor);
}
}
}

View File

@ -2162,15 +2162,15 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
vpx_highbd_convolve_copy(
CONVERT_TO_SHORTPTR(best_pred->data), best_pred->stride,
CONVERT_TO_SHORTPTR(this_mode_pred->data), this_mode_pred->stride,
NULL, 0, NULL, 0, bw, bh, xd->bd);
NULL, 0, 0, 0, 0, bw, bh, xd->bd);
else
vpx_convolve_copy(best_pred->data, best_pred->stride,
this_mode_pred->data, this_mode_pred->stride, NULL,
0, NULL, 0, bw, bh);
0, 0, 0, 0, bw, bh);
#else
vpx_convolve_copy(best_pred->data, best_pred->stride,
this_mode_pred->data, this_mode_pred->stride, NULL, 0,
NULL, 0, bw, bh);
0, 0, 0, bw, bh);
#endif // CONFIG_VP9_HIGHBITDEPTH
best_pred = this_mode_pred;
}
@ -2264,14 +2264,14 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
if (cm->use_highbitdepth)
vpx_highbd_convolve_copy(
CONVERT_TO_SHORTPTR(best_pred->data), best_pred->stride,
CONVERT_TO_SHORTPTR(pd->dst.buf), pd->dst.stride, NULL, 0, NULL, 0,
CONVERT_TO_SHORTPTR(pd->dst.buf), pd->dst.stride, NULL, 0, 0, 0, 0,
bw, bh, xd->bd);
else
vpx_convolve_copy(best_pred->data, best_pred->stride, pd->dst.buf,
pd->dst.stride, NULL, 0, NULL, 0, bw, bh);
pd->dst.stride, NULL, 0, 0, 0, 0, bw, bh);
#else
vpx_convolve_copy(best_pred->data, best_pred->stride, pd->dst.buf,
pd->dst.stride, NULL, 0, NULL, 0, bw, bh);
pd->dst.stride, NULL, 0, 0, 0, 0, bw, bh);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}

View File

@ -600,7 +600,7 @@ static void dist_block(const VP9_COMP *cpi, MACROBLOCK *x, int plane,
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
vpx_highbd_convolve_copy(CONVERT_TO_SHORTPTR(dst), dst_stride, recon16,
32, NULL, 0, NULL, 0, bs, bs, xd->bd);
32, NULL, 0, 0, 0, 0, bs, bs, xd->bd);
if (xd->lossless) {
vp9_highbd_iwht4x4_add(dqcoeff, recon16, 32, *eob, xd->bd);
} else {
@ -623,7 +623,7 @@ static void dist_block(const VP9_COMP *cpi, MACROBLOCK *x, int plane,
recon = CONVERT_TO_BYTEPTR(recon16);
} else {
#endif // CONFIG_VP9_HIGHBITDEPTH
vpx_convolve_copy(dst, dst_stride, recon, 32, NULL, 0, NULL, 0, bs, bs);
vpx_convolve_copy(dst, dst_stride, recon, 32, NULL, 0, 0, 0, 0, bs, bs);
switch (tx_size) {
case TX_32X32: vp9_idct32x32_add(dqcoeff, recon, 32, *eob); break;
case TX_16X16: vp9_idct16x16_add(dqcoeff, recon, 32, *eob); break;

View File

@ -137,15 +137,14 @@ static INLINE uint16x8_t convolve8_8(const int16x8_t s0, const int16x8_t s1,
void vpx_highbd_convolve8_horiz_neon(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, // unused
int y_step_q4, // unused
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
if (x_step_q4 != 16) {
vpx_highbd_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h, bd);
vpx_highbd_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, bd);
} else {
const int16x8_t filters = vld1q_s16(filter_x);
const int16x8_t filters = vld1q_s16(filter[x0_q4]);
const uint16x8_t max = vdupq_n_u16((1 << bd) - 1);
uint16x8_t t0, t1, t2, t3;
@ -337,15 +336,15 @@ void vpx_highbd_convolve8_horiz_neon(const uint16_t *src, ptrdiff_t src_stride,
void vpx_highbd_convolve8_avg_horiz_neon(const uint16_t *src,
ptrdiff_t src_stride, uint16_t *dst,
ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, // unused
int y_step_q4, // unused
int w, int h, int bd) {
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4,
int y_step_q4, int w, int h, int bd) {
if (x_step_q4 != 16) {
vpx_highbd_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h, bd);
vpx_highbd_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h,
bd);
} else {
const int16x8_t filters = vld1q_s16(filter_x);
const int16x8_t filters = vld1q_s16(filter[x0_q4]);
const uint16x8_t max = vdupq_n_u16((1 << bd) - 1);
uint16x8_t t0, t1, t2, t3;
@ -566,15 +565,14 @@ void vpx_highbd_convolve8_avg_horiz_neon(const uint16_t *src,
void vpx_highbd_convolve8_vert_neon(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, // unused
int x_step_q4, // unused
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
if (y_step_q4 != 16) {
vpx_highbd_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h, bd);
vpx_highbd_convolve8_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h, bd);
} else {
const int16x8_t filters = vld1q_s16(filter_y);
const int16x8_t filters = vld1q_s16(filter[y0_q4]);
const uint16x8_t max = vdupq_n_u16((1 << bd) - 1);
assert(!((intptr_t)dst & 3));
@ -732,15 +730,15 @@ void vpx_highbd_convolve8_vert_neon(const uint16_t *src, ptrdiff_t src_stride,
void vpx_highbd_convolve8_avg_vert_neon(const uint16_t *src,
ptrdiff_t src_stride, uint16_t *dst,
ptrdiff_t dst_stride,
const int16_t *filter_x, // unused
int x_step_q4, // unused
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
if (y_step_q4 != 16) {
vpx_highbd_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h, bd);
vpx_highbd_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h,
bd);
} else {
const int16x8_t filters = vld1q_s16(filter_y);
const int16x8_t filters = vld1q_s16(filter[y0_q4]);
const uint16x8_t max = vdupq_n_u16((1 << bd) - 1);
assert(!((intptr_t)dst & 3));

View File

@ -15,13 +15,14 @@
void vpx_highbd_convolve_avg_neon(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
(void)bd;
if (w < 8) { // avg4

View File

@ -15,13 +15,14 @@
void vpx_highbd_convolve_copy_neon(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
(void)bd;
if (w < 8) { // copy4

View File

@ -15,10 +15,9 @@
void vpx_highbd_convolve8_neon(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h, int bd) {
const int y0_q4 = get_filter_offset(filter_y, get_filter_base(filter_y));
// + 1 to make it divisible by 4
uint16_t temp[64 * 136];
const int intermediate_height =
@ -29,20 +28,19 @@ void vpx_highbd_convolve8_neon(const uint16_t *src, ptrdiff_t src_stride,
* buffer which has lots of extra room and is subsequently discarded this is
* safe if somewhat less than ideal. */
vpx_highbd_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w,
filter_x, x_step_q4, filter_y, y_step_q4, w,
filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w,
intermediate_height, bd);
/* Step into the temp buffer 3 lines to get the actual frame data */
vpx_highbd_convolve8_vert_neon(temp + w * 3, w, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h, bd);
vpx_highbd_convolve8_vert_neon(temp + w * 3, w, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_avg_neon(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
const int y0_q4 = get_filter_offset(filter_y, get_filter_base(filter_y));
// + 1 to make it divisible by 4
uint16_t temp[64 * 136];
const int intermediate_height =
@ -52,8 +50,9 @@ void vpx_highbd_convolve8_avg_neon(const uint16_t *src, ptrdiff_t src_stride,
* to average the values after both passes.
*/
vpx_highbd_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w,
filter_x, x_step_q4, filter_y, y_step_q4, w,
filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w,
intermediate_height, bd);
vpx_highbd_convolve8_avg_vert_neon(temp + w * 3, w, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h, bd);
vpx_highbd_convolve8_avg_vert_neon(temp + w * 3, w, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h,
bd);
}

View File

@ -42,10 +42,11 @@
; r1 int src_stride
; r2 uint8_t *dst
; r3 int dst_stride
; sp[]const int16_t *filter_x
; sp[]int x_step_q4
; sp[]const int16_t *filter_y ; unused
; sp[]int y_step_q4 ; unused
; sp[]const int16_t *filter
; sp[]int x0_q4
; sp[]int x_step_q4 ; unused
; sp[]int y0_q4
; sp[]int y_step_q4 ; unused
; sp[]int w
; sp[]int h
@ -54,11 +55,11 @@
sub r0, r0, #3 ; adjust for taps
ldr r5, [sp, #32] ; filter_x
ldr r6, [sp, #48] ; w
ldr r7, [sp, #52] ; h
ldrd r4, r5, [sp, #32] ; filter, x0_q4
add r4, r5, lsl #4
ldrd r6, r7, [sp, #52] ; w, h
vld1.s16 {q0}, [r5] ; filter_x
vld1.s16 {q0}, [r4] ; filter
sub r8, r1, r1, lsl #2 ; -src_stride * 3
add r8, r8, #4 ; -src_stride * 3 + 4
@ -127,7 +128,7 @@ vpx_convolve8_avg_loop_horiz
sub r2, r2, r3, lsl #2 ; reset for store
; src[] * filter_x
; src[] * filter
MULTIPLY_BY_Q0 q1, d16, d17, d20, d22, d18, d19, d23, d24
MULTIPLY_BY_Q0 q2, d17, d20, d22, d18, d19, d23, d24, d26
MULTIPLY_BY_Q0 q14, d20, d22, d18, d19, d23, d24, d26, d27
@ -184,11 +185,13 @@ vpx_convolve8_avg_loop_horiz
sub r0, r0, r1
sub r0, r0, r1, lsl #1
ldr r4, [sp, #32] ; filter_y
ldr r6, [sp, #40] ; w
ldr lr, [sp, #44] ; h
ldr r4, [sp, #24] ; filter
ldr r5, [sp, #36] ; y0_q4
add r4, r5, lsl #4
ldr r6, [sp, #44] ; w
ldr lr, [sp, #48] ; h
vld1.s16 {q0}, [r4] ; filter_y
vld1.s16 {q0}, [r4] ; filter
lsl r1, r1, #1
lsl r3, r3, #1
@ -232,7 +235,7 @@ vpx_convolve8_avg_loop_vert
pld [r7]
pld [r4]
; src[] * filter_y
; src[] * filter
MULTIPLY_BY_Q0 q1, d16, d17, d18, d19, d20, d21, d22, d24
pld [r7, r1]

View File

@ -125,11 +125,10 @@ static INLINE int16x8_t convolve8_8(int16x8_t s0, int16x8_t s1, int16x8_t s2,
void vpx_convolve8_horiz_neon(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, // unused
int y_step_q4, // unused
int w, int h) {
const int16x8_t filters = vld1q_s16(filter_x);
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16x8_t filters = vld1q_s16(filter[x0_q4]);
uint8x8_t t0, t1, t2, t3;
assert(!((intptr_t)dst & 3));
@ -137,8 +136,8 @@ void vpx_convolve8_horiz_neon(const uint8_t *src, ptrdiff_t src_stride,
assert(x_step_q4 == 16);
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
(void)filter_y;
src -= 3;
@ -390,11 +389,10 @@ void vpx_convolve8_horiz_neon(const uint8_t *src, ptrdiff_t src_stride,
void vpx_convolve8_avg_horiz_neon(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, // unused
int y_step_q4, // unused
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const int16x8_t filters = vld1q_s16(filter_x);
const int16x8_t filters = vld1q_s16(filter[x0_q4]);
uint8x8_t t0, t1, t2, t3;
assert(!((intptr_t)dst & 3));
@ -402,8 +400,8 @@ void vpx_convolve8_avg_horiz_neon(const uint8_t *src, ptrdiff_t src_stride,
assert(x_step_q4 == 16);
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
(void)filter_y;
src -= 3;
@ -692,19 +690,18 @@ void vpx_convolve8_avg_horiz_neon(const uint8_t *src, ptrdiff_t src_stride,
void vpx_convolve8_vert_neon(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, // unused
int x_step_q4, // unused
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16x8_t filters = vld1q_s16(filter_y);
const int16x8_t filters = vld1q_s16(filter[y0_q4]);
assert(!((intptr_t)dst & 3));
assert(!(dst_stride & 3));
assert(y_step_q4 == 16);
(void)x0_q4;
(void)x_step_q4;
(void)y_step_q4;
(void)filter_x;
src -= 3 * src_stride;
@ -864,19 +861,18 @@ void vpx_convolve8_vert_neon(const uint8_t *src, ptrdiff_t src_stride,
void vpx_convolve8_avg_vert_neon(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, // unused
int x_step_q4, // unused
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16x8_t filters = vld1q_s16(filter_y);
const int16x8_t filters = vld1q_s16(filter[y0_q4]);
assert(!((intptr_t)dst & 3));
assert(!(dst_stride & 3));
assert(y_step_q4 == 16);
(void)x0_q4;
(void)x_step_q4;
(void)y_step_q4;
(void)filter_x;
src -= 3 * src_stride;

View File

@ -42,10 +42,11 @@
; r1 int src_stride
; r2 uint8_t *dst
; r3 int dst_stride
; sp[]const int16_t *filter_x
; sp[]int x_step_q4
; sp[]const int16_t *filter_y ; unused
; sp[]int y_step_q4 ; unused
; sp[]const int16_t *filter
; sp[]int x0_q4
; sp[]int x_step_q4 ; unused
; sp[]int y0_q4
; sp[]int y_step_q4 ; unused
; sp[]int w
; sp[]int h
@ -54,11 +55,11 @@
sub r0, r0, #3 ; adjust for taps
ldr r5, [sp, #32] ; filter_x
ldr r6, [sp, #48] ; w
ldr r7, [sp, #52] ; h
ldrd r4, r5, [sp, #32] ; filter, x0_q4
add r4, r5, lsl #4
ldrd r6, r7, [sp, #52] ; w, h
vld1.s16 {q0}, [r5] ; filter_x
vld1.s16 {q0}, [r4] ; filter
sub r8, r1, r1, lsl #2 ; -src_stride * 3
add r8, r8, #4 ; -src_stride * 3 + 4
@ -119,7 +120,7 @@ vpx_convolve8_loop_horiz
pld [r5, r1, lsl #1]
; src[] * filter_x
; src[] * filter
MULTIPLY_BY_Q0 q1, d16, d17, d20, d22, d18, d19, d23, d24
MULTIPLY_BY_Q0 q2, d17, d20, d22, d18, d19, d23, d24, d26
MULTIPLY_BY_Q0 q14, d20, d22, d18, d19, d23, d24, d26, d27
@ -173,11 +174,13 @@ vpx_convolve8_loop_horiz
sub r0, r0, r1
sub r0, r0, r1, lsl #1
ldr r4, [sp, #32] ; filter_y
ldr r6, [sp, #40] ; w
ldr lr, [sp, #44] ; h
ldr r4, [sp, #24] ; filter
ldr r5, [sp, #36] ; y0_q4
add r4, r5, lsl #4
ldr r6, [sp, #44] ; w
ldr lr, [sp, #48] ; h
vld1.s16 {q0}, [r4] ; filter_y
vld1.s16 {q0}, [r4] ; filter
lsl r1, r1, #1
lsl r3, r3, #1
@ -216,7 +219,7 @@ vpx_convolve8_loop_vert
pld [r5]
pld [r8]
; src[] * filter_y
; src[] * filter
MULTIPLY_BY_Q0 q1, d16, d17, d18, d19, d20, d21, d22, d24
pld [r5, r3]

View File

@ -15,13 +15,13 @@
void vpx_convolve_avg_neon(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride, int w,
int h) {
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
const InterpKernel *filter, int x0_q4, int x_step_q4,
int y0_q4, int y_step_q4, int w, int h) {
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
if (w < 8) { // avg4
uint8x8_t s0, s1;

View File

@ -17,7 +17,7 @@
|vpx_convolve_avg_neon| PROC
push {r4-r6, lr}
ldrd r4, r5, [sp, #32]
ldrd r4, r5, [sp, #36]
mov r6, r2
cmp r4, #32

View File

@ -15,13 +15,14 @@
void vpx_convolve_copy_neon(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
if (w < 8) { // copy4
do {

View File

@ -17,7 +17,7 @@
|vpx_convolve_copy_neon| PROC
push {r4-r5, lr}
ldrd r4, r5, [sp, #28]
ldrd r4, r5, [sp, #32]
cmp r4, #32
bgt copy64

View File

@ -15,8 +15,8 @@
#include "vpx_ports/mem.h"
void vpx_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
/* Given our constraints: w <= 64, h <= 64, taps == 8 we can reduce the
* maximum buffer size to 64 * 64 + 7 (+ 1 to make it divisible by 4).
@ -33,19 +33,19 @@ void vpx_convolve8_neon(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
* height and filter a multiple of 4 lines. Since this goes in to the temp
* buffer which has lots of extra room and is subsequently discarded this is
* safe if somewhat less than ideal. */
vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w, filter_x,
x_step_q4, filter_y, y_step_q4, w,
vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w,
intermediate_height);
/* Step into the temp buffer 3 lines to get the actual frame data */
vpx_convolve8_vert_neon(temp + w * 3, w, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
vpx_convolve8_vert_neon(temp + w * 3, w, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
}
void vpx_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
uint8_t temp[64 * 72];
const int intermediate_height = h + 7;
@ -56,9 +56,9 @@ void vpx_convolve8_avg_neon(const uint8_t *src, ptrdiff_t src_stride,
/* This implementation has the same issues as above. In addition, we only want
* to average the values after both passes.
*/
vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w, filter_x,
x_step_q4, filter_y, y_step_q4, w,
vpx_convolve8_horiz_neon(src - src_stride * 3, src_stride, temp, w, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w,
intermediate_height);
vpx_convolve8_avg_vert_neon(temp + w * 3, w, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h);
vpx_convolve8_avg_vert_neon(temp + w * 3, w, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
}

View File

@ -219,9 +219,10 @@ static void convolve_bi_avg_vert_64_dspr2(const uint8_t *src,
void vpx_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const int16_t *const filter_y = filter[y0_q4];
uint32_t pos = 38;
assert(y_step_q4 == 16);
@ -247,8 +248,8 @@ void vpx_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
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);
vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -751,9 +751,10 @@ static void convolve_bi_avg_horiz_64_dspr2(const uint8_t *src_ptr,
void vpx_convolve2_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const int16_t *const filter_x = filter[x0_q4];
uint32_t pos = 38;
assert(x_step_q4 == 16);
@ -793,8 +794,8 @@ void vpx_convolve2_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
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);
vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -628,9 +628,10 @@ static void convolve_bi_horiz_64_dspr2(const uint8_t *src_ptr,
void vpx_convolve2_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const int16_t *const filter_x = filter[x0_q4];
uint32_t pos = 38;
assert(x_step_q4 == 16);
@ -672,8 +673,8 @@ void vpx_convolve2_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
(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);
vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -201,9 +201,10 @@ static void convolve_bi_vert_64_dspr2(const uint8_t *src, int32_t src_stride,
void vpx_convolve2_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const int16_t *const filter_y = filter[y0_q4];
uint32_t pos = 38;
assert(y_step_q4 == 16);
@ -228,8 +229,8 @@ void vpx_convolve2_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
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);
vpx_convolve8_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -334,15 +334,16 @@ static void convolve_avg_vert_64_dspr2(const uint8_t *src, int32_t src_stride,
void vpx_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const int16_t *const filter_y = filter[y0_q4];
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, filter_y, y_step_q4, w, h);
vpx_convolve2_avg_vert_dspr2(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
} else {
uint32_t pos = 38;
@ -367,8 +368,8 @@ void vpx_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
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);
vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}
@ -376,8 +377,8 @@ void vpx_convolve8_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
void vpx_convolve8_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
/* Fixed size intermediate buffer places limits on parameters. */
DECLARE_ALIGNED(32, uint8_t, temp[64 * 135]);
@ -390,24 +391,26 @@ void vpx_convolve8_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
if (intermediate_height < h) intermediate_height = h;
vpx_convolve8_horiz(src - (src_stride * 3), src_stride, temp, 64, filter_x,
x_step_q4, filter_y, y_step_q4, w, intermediate_height);
vpx_convolve8_horiz(src - (src_stride * 3), src_stride, temp, 64, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w,
intermediate_height);
vpx_convolve8_avg_vert(temp + 64 * 3, 64, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h);
vpx_convolve8_avg_vert(temp + 64 * 3, 64, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
}
void vpx_convolve_avg_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride, int w,
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
int x, y;
uint32_t tp1, tp2, tn1, tp3, tp4, tn2;
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
/* prefetch data to cache memory */
prefetch_load(src);

View File

@ -938,15 +938,16 @@ static void convolve_avg_horiz_64_dspr2(const uint8_t *src_ptr,
void vpx_convolve8_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const int16_t *const filter_x = filter[x0_q4];
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, filter_y, y_step_q4, w, h);
vpx_convolve2_avg_horiz_dspr2(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
} else {
uint32_t pos = 38;
@ -987,9 +988,8 @@ void vpx_convolve8_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
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);
vpx_convolve8_avg_horiz_c(src + 3, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -1296,9 +1296,11 @@ void copy_horiz_transposed(const uint8_t *src, ptrdiff_t src_stride,
}
void vpx_convolve8_dspr2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const int16_t *const filter_x = filter[x0_q4];
const int16_t *const filter_y = filter[y0_q4];
DECLARE_ALIGNED(32, uint8_t, temp[64 * 135]);
int32_t intermediate_height = ((h * y_step_q4) >> 4) + 7;
uint32_t pos = 38;
@ -1395,14 +1397,15 @@ void vpx_convolve8_dspr2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
void vpx_convolve_copy_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride,
int w, int h) {
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
int x, y;
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
/* prefetch data to cache memory */
prefetch_load(src);

View File

@ -818,15 +818,16 @@ static void convolve_horiz_64_dspr2(const uint8_t *src_ptr, int32_t src_stride,
void vpx_convolve8_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16_t *const filter_x = filter[x0_q4];
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, filter_y, y_step_q4, w, h);
vpx_convolve2_horiz_dspr2(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
} else {
uint32_t pos = 38;
@ -868,8 +869,8 @@ void vpx_convolve8_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
(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);
vpx_convolve8_horiz_c(src + 3, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -318,15 +318,16 @@ static void convolve_vert_64_dspr2(const uint8_t *src, int32_t src_stride,
void vpx_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16_t *const filter_y = filter[y0_q4];
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, filter_y, y_step_q4, w, h);
vpx_convolve2_vert_dspr2(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
} else {
uint32_t pos = 38;
@ -349,8 +350,8 @@ void vpx_convolve8_vert_dspr2(const uint8_t *src, ptrdiff_t 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, filter_x,
x_step_q4, filter_y, y_step_q4, w, h);
vpx_convolve8_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -24,21 +24,21 @@ extern "C" {
#if HAVE_DSPR2
void vpx_convolve2_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h);
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h);
void vpx_convolve2_avg_horiz_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h);
void vpx_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h);
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h);
void vpx_convolve2_dspr2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter, int w,
@ -46,9 +46,9 @@ void vpx_convolve2_dspr2(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
void vpx_convolve2_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h);
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4,
int w, int h);
#endif // #if HAVE_DSPR2
#ifdef __cplusplus

View File

@ -633,9 +633,10 @@ static void common_hz_2t_and_aver_dst_64w_msa(const uint8_t *src,
void vpx_convolve8_avg_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16_t *const filter_x = filter[x0_q4];
int8_t cnt, filt_hor[8];
assert(x_step_q4 == 16);
@ -668,8 +669,8 @@ void vpx_convolve8_avg_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
(int32_t)dst_stride, &filt_hor[3], 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);
vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
} else {
@ -695,8 +696,8 @@ void vpx_convolve8_avg_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
(int32_t)dst_stride, filt_hor, 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);
vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -516,9 +516,10 @@ static void common_hv_2ht_2vt_and_aver_dst_64w_msa(
void vpx_convolve8_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
const InterpKernel *filter, int x0_q4, int x_step_q4,
int y0_q4, int y_step_q4, int w, int h) {
const int16_t *const filter_x = filter[x0_q4];
const int16_t *const filter_y = filter[y0_q4];
int8_t cnt, filt_hor[8], filt_ver[8];
assert(x_step_q4 == 16);
@ -560,14 +561,14 @@ void vpx_convolve8_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
&filt_hor[3], &filt_ver[3], h);
break;
default:
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h);
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
} else if (((const int32_t *)filter_x)[0] == 0 ||
((const int32_t *)filter_y)[0] == 0) {
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
} else {
switch (w) {
case 4:
@ -596,8 +597,8 @@ void vpx_convolve8_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
filt_ver, h);
break;
default:
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h);
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -605,9 +605,10 @@ static void common_vt_2t_and_aver_dst_64w_msa(const uint8_t *src,
void vpx_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16_t *const filter_y = filter[y0_q4];
int8_t cnt, filt_ver[8];
assert(y_step_q4 == 16);
@ -640,8 +641,8 @@ void vpx_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
(int32_t)dst_stride, &filt_ver[3], 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);
vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
} else {
@ -668,8 +669,8 @@ void vpx_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
(int32_t)dst_stride, filt_ver, 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);
vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter,
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -621,9 +621,10 @@ static void common_hz_2t_64w_msa(const uint8_t *src, int32_t src_stride,
void vpx_convolve8_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16_t *const filter_x = filter[x0_q4];
int8_t cnt, filt_hor[8];
assert(x_step_q4 == 16);
@ -656,8 +657,8 @@ void vpx_convolve8_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
&filt_hor[3], 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);
vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
} else {
@ -683,8 +684,8 @@ void vpx_convolve8_horiz_msa(const uint8_t *src, ptrdiff_t src_stride,
filt_hor, 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);
vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -541,9 +541,11 @@ static void common_hv_2ht_2vt_64w_msa(const uint8_t *src, int32_t src_stride,
}
void vpx_convolve8_msa(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int32_t x_step_q4, const int16_t *filter_y,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int32_t x_step_q4, int y0_q4,
int32_t y_step_q4, int32_t w, int32_t h) {
const int16_t *const filter_x = filter[x0_q4];
const int16_t *const filter_y = filter[y0_q4];
int8_t cnt, filt_hor[8], filt_ver[8];
assert(x_step_q4 == 16);
@ -585,14 +587,14 @@ void vpx_convolve8_msa(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
&filt_ver[3], (int32_t)h);
break;
default:
vpx_convolve8_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
vpx_convolve8_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
} else if (((const int32_t *)filter_x)[0] == 0 ||
((const int32_t *)filter_y)[0] == 0) {
vpx_convolve8_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
vpx_convolve8_c(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4,
y0_q4, y_step_q4, w, h);
} else {
switch (w) {
case 4:
@ -621,8 +623,8 @@ void vpx_convolve8_msa(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
(int32_t)h);
break;
default:
vpx_convolve8_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
vpx_convolve8_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -628,9 +628,10 @@ static void common_vt_2t_64w_msa(const uint8_t *src, int32_t src_stride,
void vpx_convolve8_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const int16_t *const filter_y = filter[y0_q4];
int8_t cnt, filt_ver[8];
assert(y_step_q4 == 16);
@ -663,8 +664,8 @@ void vpx_convolve8_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
&filt_ver[3], 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);
vpx_convolve8_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
} else {
@ -690,8 +691,8 @@ void vpx_convolve8_vert_msa(const uint8_t *src, ptrdiff_t src_stride,
filt_ver, 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);
vpx_convolve8_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}

View File

@ -189,13 +189,14 @@ static void avg_width64_msa(const uint8_t *src, int32_t src_stride,
void vpx_convolve_avg_msa(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int32_t filter_x_stride,
const int16_t *filter_y, int32_t filter_y_stride,
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int32_t y_step_q4,
int32_t w, int32_t h) {
(void)filter_x;
(void)filter_y;
(void)filter_x_stride;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
switch (w) {
case 4: {

View File

@ -199,13 +199,14 @@ static void copy_width64_msa(const uint8_t *src, int32_t src_stride,
void vpx_convolve_copy_msa(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int32_t filter_x_stride,
const int16_t *filter_y, int32_t filter_y_stride,
const InterpKernel *filter, int x0_q4,
int32_t x_step_q4, int y0_q4, int32_t y_step_q4,
int32_t w, int32_t h) {
(void)filter_x;
(void)filter_y;
(void)filter_x_stride;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
switch (w) {
case 4: {

View File

@ -53,13 +53,13 @@ static inline void copy_w64(const uint8_t *src, ptrdiff_t src_stride,
void vpx_convolve_copy_vsx(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int32_t filter_x_stride,
const int16_t *filter_y, int32_t filter_y_stride,
int32_t w, int32_t h) {
(void)filter_x;
(void)filter_y;
(void)filter_x_stride;
(void)filter_y_stride;
const InterpKernel *filter, int x0_q4, int x_step_q4,
int y0_q4, int32_t y_step_q4, int32_t w, int32_t h) {
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
switch (w) {
case 16: {
@ -132,14 +132,8 @@ static inline void avg_w64(const uint8_t *src, ptrdiff_t src_stride,
void vpx_convolve_avg_vsx(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int32_t filter_x_stride,
const int16_t *filter_y, int32_t filter_y_stride,
int32_t w, int32_t h) {
(void)filter_x;
(void)filter_y;
(void)filter_x_stride;
(void)filter_y_stride;
const InterpKernel *filter, int x0_q4, int x_step_q4,
int y0_q4, int32_t y_step_q4, int32_t w, int32_t h) {
switch (w) {
case 16: {
avg_w16(src, src_stride, dst, dst_stride, h);
@ -154,8 +148,8 @@ void vpx_convolve_avg_vsx(const uint8_t *src, ptrdiff_t src_stride,
break;
}
default: {
vpx_convolve_avg_c(src, src_stride, dst, dst_stride, filter_x,
filter_x_stride, filter_y, filter_y_stride, w, h);
vpx_convolve_avg_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
break;
}
}
@ -299,9 +293,9 @@ static inline void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
static inline void convolve(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const InterpKernel *const x_filters, int x0_q4,
int x_step_q4, const InterpKernel *const y_filters,
int y0_q4, int y_step_q4, int w, int h) {
const InterpKernel *const filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
// Note: Fixed size intermediate buffer, temp, places limits on parameters.
// 2d filtering proceeds in 2 steps:
// (1) Interpolate horizontally into an intermediate buffer, temp.
@ -324,95 +318,77 @@ static inline void convolve(const uint8_t *src, ptrdiff_t src_stride,
assert(x_step_q4 <= 32);
convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp, 64,
x_filters, x0_q4, x_step_q4, w, intermediate_height);
convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride,
y_filters, y0_q4, y_step_q4, w, h);
filter, x0_q4, x_step_q4, w, intermediate_height);
convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride, filter,
y0_q4, y_step_q4, w, h);
}
void vpx_convolve8_horiz_vsx(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
(void)filter_y;
(void)y0_q4;
(void)y_step_q4;
convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
w, h);
convolve_horiz(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4, w,
h);
}
void vpx_convolve8_avg_horiz_vsx(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
(void)filter_y;
(void)y0_q4;
(void)y_step_q4;
convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
x_step_q4, w, h);
convolve_avg_horiz(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4,
w, h);
}
void vpx_convolve8_vert_vsx(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
(void)filter_x;
(void)x0_q4;
(void)x_step_q4;
convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4, y_step_q4,
w, h);
convolve_vert(src, src_stride, dst, dst_stride, filter, y0_q4, y_step_q4, w,
h);
}
void vpx_convolve8_avg_vert_vsx(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
(void)filter_x;
(void)x0_q4;
(void)x_step_q4;
convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
y_step_q4, w, h);
convolve_avg_vert(src, src_stride, dst, dst_stride, filter, y0_q4, y_step_q4,
w, h);
}
void vpx_convolve8_vsx(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
convolve(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
filters_y, y0_q4, y_step_q4, w, h);
convolve(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4, y0_q4,
y_step_q4, w, h);
}
void vpx_convolve8_avg_vsx(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
const InterpKernel *filter, int x0_q4, int x_step_q4,
int y0_q4, int y_step_q4, int w, int h) {
// Fixed size intermediate buffer places limits on parameters.
DECLARE_ALIGNED(16, uint8_t, temp[64 * 64]);
assert(w <= 64);
assert(h <= 64);
vpx_convolve8_vsx(src, src_stride, temp, 64, filter_x, x_step_q4, filter_y,
vpx_convolve8_vsx(src, src_stride, temp, 64, filter, x0_q4, x_step_q4, y0_q4,
y_step_q4, w, h);
vpx_convolve_avg_vsx(temp, 64, dst, dst_stride, NULL, 0, NULL, 0, w, h);
vpx_convolve_avg_vsx(temp, 64, dst, dst_stride, NULL, 0, 0, 0, 0, w, h);
}

View File

@ -114,10 +114,9 @@ static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
}
static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const InterpKernel *const x_filters,
int x0_q4, int x_step_q4,
const InterpKernel *const y_filters, int y0_q4,
int y_step_q4, int w, int h) {
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
// Note: Fixed size intermediate buffer, temp, places limits on parameters.
// 2d filtering proceeds in 2 steps:
// (1) Interpolate horizontally into an intermediate buffer, temp.
@ -140,108 +139,86 @@ static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
assert(x_step_q4 <= 32);
convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride, temp, 64,
x_filters, x0_q4, x_step_q4, w, intermediate_height);
convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride,
y_filters, y0_q4, y_step_q4, w, h);
filter, x0_q4, x_step_q4, w, intermediate_height);
convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride, filter,
y0_q4, y_step_q4, w, h);
}
void vpx_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
(void)filter_y;
const InterpKernel *filter, int x0_q4, int x_step_q4,
int y0_q4, int y_step_q4, int w, int h) {
(void)y0_q4;
(void)y_step_q4;
convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
w, h);
convolve_horiz(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4, w,
h);
}
void vpx_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
(void)filter_y;
(void)y0_q4;
(void)y_step_q4;
convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
x_step_q4, w, h);
convolve_avg_horiz(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4,
w, h);
}
void vpx_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
(void)filter_x;
const InterpKernel *filter, int x0_q4, int x_step_q4,
int y0_q4, int y_step_q4, int w, int h) {
(void)x0_q4;
(void)x_step_q4;
convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4, y_step_q4,
w, h);
convolve_vert(src, src_stride, dst, dst_stride, filter, y0_q4, y_step_q4, w,
h);
}
void vpx_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
(void)filter_x;
(void)x0_q4;
(void)x_step_q4;
convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
y_step_q4, w, h);
convolve_avg_vert(src, src_stride, dst, dst_stride, filter, y0_q4, y_step_q4,
w, h);
}
void vpx_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
convolve(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
filters_y, y0_q4, y_step_q4, w, h);
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
convolve(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4, y0_q4,
y_step_q4, w, h);
}
void vpx_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
// Fixed size intermediate buffer places limits on parameters.
DECLARE_ALIGNED(16, uint8_t, temp[64 * 64]);
assert(w <= 64);
assert(h <= 64);
vpx_convolve8_c(src, src_stride, temp, 64, filter_x, x_step_q4, filter_y,
vpx_convolve8_c(src, src_stride, temp, 64, filter, x0_q4, x_step_q4, y0_q4,
y_step_q4, w, h);
vpx_convolve_avg_c(temp, 64, dst, dst_stride, NULL, 0, NULL, 0, w, h);
vpx_convolve_avg_c(temp, 64, dst, dst_stride, NULL, 0, 0, 0, 0, w, h);
}
void vpx_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int filter_x_stride, const int16_t *filter_y,
int filter_y_stride, int w, int h) {
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
int r;
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
for (r = h; r > 0; --r) {
memcpy(dst, src, w);
@ -251,15 +228,16 @@ void vpx_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
}
void vpx_convolve_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int filter_x_stride, const int16_t *filter_y,
int filter_y_stride, int w, int h) {
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
int x, y;
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
for (y = 0; y < h; ++y) {
for (x = 0; x < w; ++x) dst[x] = ROUND_POWER_OF_TWO(dst[x] + src[x], 1);
@ -269,53 +247,52 @@ void vpx_convolve_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
}
void vpx_scaled_horiz_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
vpx_convolve8_horiz_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
}
void vpx_scaled_vert_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
vpx_convolve8_vert_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
vpx_convolve8_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
}
void vpx_scaled_2d_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
int w, int h) {
vpx_convolve8_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
vpx_convolve8_c(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4,
y0_q4, y_step_q4, w, h);
}
void vpx_scaled_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h);
vpx_convolve8_avg_horiz_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
}
void vpx_scaled_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h) {
vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter_x,
x_step_q4, filter_y, y_step_q4, w, h);
const InterpKernel *filter, int x0_q4, int x_step_q4,
int y0_q4, int y_step_q4, int w, int h) {
vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
}
void vpx_scaled_avg_2d_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter_x, x_step_q4,
filter_y, y_step_q4, w, h);
vpx_convolve8_avg_c(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, y0_q4, y_step_q4, w, h);
}
#if CONFIG_VP9_HIGHBITDEPTH
@ -417,9 +394,9 @@ static void highbd_convolve_avg_vert(const uint16_t *src, ptrdiff_t src_stride,
static void highbd_convolve(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const InterpKernel *const x_filters, int x0_q4,
int x_step_q4, const InterpKernel *const y_filters,
int y0_q4, int y_step_q4, int w, int h, int bd) {
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h, int bd) {
// Note: Fixed size intermediate buffer, temp, places limits on parameters.
// 2d filtering proceeds in 2 steps:
// (1) Interpolate horizontally into an intermediate buffer, temp.
@ -442,113 +419,97 @@ static void highbd_convolve(const uint16_t *src, ptrdiff_t src_stride,
assert(x_step_q4 <= 32);
highbd_convolve_horiz(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride,
temp, 64, x_filters, x0_q4, x_step_q4, w,
temp, 64, filter, x0_q4, x_step_q4, w,
intermediate_height, bd);
highbd_convolve_vert(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst, dst_stride,
y_filters, y0_q4, y_step_q4, w, h, bd);
filter, y0_q4, y_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_horiz_c(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
int h, int bd) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
(void)filter_y;
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
(void)y0_q4;
(void)y_step_q4;
highbd_convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
highbd_convolve_horiz(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_avg_horiz_c(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
(void)filter_y;
(void)y0_q4;
(void)y_step_q4;
highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
highbd_convolve_avg_horiz(src, src_stride, dst, dst_stride, filter, x0_q4,
x_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_vert_c(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h, int bd) {
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
(void)filter_x;
(void)x0_q4;
(void)x_step_q4;
highbd_convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
highbd_convolve_vert(src, src_stride, dst, dst_stride, filter, y0_q4,
y_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_avg_vert_c(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd) {
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
(void)filter_x;
(void)x0_q4;
(void)x_step_q4;
highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
highbd_convolve_avg_vert(src, src_stride, dst, dst_stride, filter, y0_q4,
y_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_c(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h, int bd) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
highbd_convolve(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
filters_y, y0_q4, y_step_q4, w, h, bd);
highbd_convolve(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4,
y0_q4, y_step_q4, w, h, bd);
}
void vpx_highbd_convolve8_avg_c(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h, int bd) {
// Fixed size intermediate buffer places limits on parameters.
DECLARE_ALIGNED(16, uint16_t, temp[64 * 64]);
assert(w <= 64);
assert(h <= 64);
vpx_highbd_convolve8_c(src, src_stride, temp, 64, filter_x, x_step_q4,
filter_y, y_step_q4, w, h, bd);
vpx_highbd_convolve_avg_c(temp, 64, dst, dst_stride, NULL, 0, NULL, 0, w, h,
vpx_highbd_convolve8_c(src, src_stride, temp, 64, filter, x0_q4, x_step_q4,
y0_q4, y_step_q4, w, h, bd);
vpx_highbd_convolve_avg_c(temp, 64, dst, dst_stride, NULL, 0, 0, 0, 0, w, h,
bd);
}
void vpx_highbd_convolve_copy_c(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride,
int w, int h, int bd) {
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h, int bd) {
int r;
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
(void)bd;
for (r = h; r > 0; --r) {
@ -560,15 +521,16 @@ void vpx_highbd_convolve_copy_c(const uint16_t *src, ptrdiff_t src_stride,
void vpx_highbd_convolve_avg_c(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride,
int w, int h, int bd) {
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h, int bd) {
int x, y;
(void)filter_x;
(void)filter_x_stride;
(void)filter_y;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
(void)bd;
for (y = 0; y < h; ++y) {

View File

@ -19,15 +19,15 @@ extern "C" {
typedef void (*convolve_fn_t)(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4, int w,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h);
#if CONFIG_VP9_HIGHBITDEPTH
typedef void (*highbd_convolve_fn_t)(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int w, int h, int bd);
#endif

View File

@ -6,6 +6,7 @@ print <<EOF
#include "vpx/vpx_integer.h"
#include "vpx_dsp/vpx_dsp_common.h"
#include "vpx_dsp/vpx_filter.h"
EOF
}
@ -331,69 +332,69 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
#
# Sub Pixel Filters
#
add_proto qw/void vpx_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_convolve_copy neon dspr2 msa sse2 vsx/;
add_proto qw/void vpx_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_convolve_avg neon dspr2 msa sse2 vsx/;
add_proto qw/void vpx_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_convolve8 sse2 ssse3 avx2 neon dspr2 msa vsx/;
add_proto qw/void vpx_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_convolve8_horiz sse2 ssse3 avx2 neon dspr2 msa vsx/;
add_proto qw/void vpx_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_convolve8_vert sse2 ssse3 avx2 neon dspr2 msa vsx/;
add_proto qw/void vpx_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_convolve8_avg sse2 ssse3 neon dspr2 msa vsx/;
add_proto qw/void vpx_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_convolve8_avg_horiz sse2 ssse3 neon dspr2 msa vsx/;
add_proto qw/void vpx_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_convolve8_avg_vert sse2 ssse3 neon dspr2 msa vsx/;
add_proto qw/void vpx_scaled_2d/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_2d/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
specialize qw/vpx_scaled_2d ssse3/;
add_proto qw/void vpx_scaled_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_avg_2d/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_avg_2d/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
add_proto qw/void vpx_scaled_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h";
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
#
# Sub Pixel Filters
#
add_proto qw/void vpx_highbd_convolve_copy/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void vpx_highbd_convolve_copy/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bps";
specialize qw/vpx_highbd_convolve_copy sse2 avx2 neon/;
add_proto qw/void vpx_highbd_convolve_avg/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void vpx_highbd_convolve_avg/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bps";
specialize qw/vpx_highbd_convolve_avg sse2 avx2 neon/;
add_proto qw/void vpx_highbd_convolve8/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void vpx_highbd_convolve8/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bps";
specialize qw/vpx_highbd_convolve8 avx2 neon/, "$sse2_x86_64";
add_proto qw/void vpx_highbd_convolve8_horiz/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void vpx_highbd_convolve8_horiz/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bps";
specialize qw/vpx_highbd_convolve8_horiz avx2 neon/, "$sse2_x86_64";
add_proto qw/void vpx_highbd_convolve8_vert/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void vpx_highbd_convolve8_vert/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bps";
specialize qw/vpx_highbd_convolve8_vert avx2 neon/, "$sse2_x86_64";
add_proto qw/void vpx_highbd_convolve8_avg/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void vpx_highbd_convolve8_avg/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bps";
specialize qw/vpx_highbd_convolve8_avg avx2 neon/, "$sse2_x86_64";
add_proto qw/void vpx_highbd_convolve8_avg_horiz/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void vpx_highbd_convolve8_avg_horiz/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bps";
specialize qw/vpx_highbd_convolve8_avg_horiz avx2 neon/, "$sse2_x86_64";
add_proto qw/void vpx_highbd_convolve8_avg_vert/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
add_proto qw/void vpx_highbd_convolve8_avg_vert/, "const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bps";
specialize qw/vpx_highbd_convolve8_avg_vert avx2 neon/, "$sse2_x86_64";
} # CONFIG_VP9_HIGHBITDEPTH

View File

@ -26,17 +26,6 @@ extern "C" {
typedef int16_t InterpKernel[SUBPEL_TAPS];
static INLINE const InterpKernel *get_filter_base(const int16_t *filter) {
// NOTE: This assumes that the filter table is 256-byte aligned.
// TODO(agrange) Modify to make independent of table alignment.
return (const InterpKernel *)(((intptr_t)filter) & ~((intptr_t)0xFF));
}
static INLINE int get_filter_offset(const int16_t *f,
const InterpKernel *base) {
return (int)((const InterpKernel *)(intptr_t)f - base);
}
#ifdef __cplusplus
} // extern "C"
#endif

View File

@ -20,14 +20,15 @@ typedef void filter8_1dfunction(const uint8_t *src_ptr, ptrdiff_t src_pitch,
uint8_t *output_ptr, ptrdiff_t out_pitch,
uint32_t output_height, const int16_t *filter);
#define FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
#define FUN_CONV_1D(name, offset, step_q4, dir, src_start, avg, opt) \
void vpx_convolve8_##name##_##opt( \
const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \
ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, \
const int16_t *filter_y, int y_step_q4, int w, int h) { \
(void)filter_x; \
ptrdiff_t dst_stride, const InterpKernel *filter_kernel, int x0_q4, \
int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \
const int16_t *filter = filter_kernel[offset]; \
(void)x0_q4; \
(void)x_step_q4; \
(void)filter_y; \
(void)y0_q4; \
(void)y_step_q4; \
assert(filter[3] != 128); \
assert(step_q4 == 16); \
@ -64,32 +65,36 @@ typedef void filter8_1dfunction(const uint8_t *src_ptr, ptrdiff_t src_pitch,
} \
}
#define FUN_CONV_2D(avg, opt) \
void vpx_convolve8_##avg##opt( \
const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \
ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, \
const int16_t *filter_y, int y_step_q4, int w, int h) { \
assert(filter_x[3] != 128); \
assert(filter_y[3] != 128); \
assert(w <= 64); \
assert(h <= 64); \
assert(x_step_q4 == 16); \
assert(y_step_q4 == 16); \
if (filter_x[0] | filter_x[1] | filter_x[2]) { \
DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71]); \
vpx_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \
filter_x, x_step_q4, filter_y, y_step_q4, w, \
h + 7); \
vpx_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \
filter_x, x_step_q4, filter_y, \
y_step_q4, w, h); \
} else { \
DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 65]); \
vpx_convolve8_horiz_##opt(src, src_stride, fdata2, 64, filter_x, \
x_step_q4, filter_y, y_step_q4, w, h + 1); \
vpx_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, filter_x, \
x_step_q4, filter_y, y_step_q4, w, h); \
} \
#define FUN_CONV_2D(avg, opt) \
void vpx_convolve8_##avg##opt( \
const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \
ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, \
int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \
const int16_t *filter_x = filter[x0_q4]; \
const int16_t *filter_y = filter[y0_q4]; \
(void)filter_y; \
assert(filter_x[3] != 128); \
assert(filter_y[3] != 128); \
assert(w <= 64); \
assert(h <= 64); \
assert(x_step_q4 == 16); \
assert(y_step_q4 == 16); \
if (filter_x[0] | filter_x[1] | filter_x[2]) { \
DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71]); \
vpx_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \
filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w, \
h + 7); \
vpx_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \
filter, x0_q4, x_step_q4, y0_q4, \
y_step_q4, w, h); \
} else { \
DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 65]); \
vpx_convolve8_horiz_##opt(src, src_stride, fdata2, 64, filter, x0_q4, \
x_step_q4, y0_q4, y_step_q4, w, h + 1); \
vpx_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, filter, \
x0_q4, x_step_q4, y0_q4, y_step_q4, w, \
h); \
} \
}
#if CONFIG_VP9_HIGHBITDEPTH
@ -101,95 +106,97 @@ typedef void highbd_filter8_1dfunction(const uint16_t *src_ptr,
unsigned int output_height,
const int16_t *filter, int bd);
#define HIGH_FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
void vpx_highbd_convolve8_##name##_##opt( \
const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, \
ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, \
const int16_t *filter_y, int y_step_q4, int w, int h, int bd) { \
if (step_q4 == 16 && filter[3] != 128) { \
if (filter[0] | filter[1] | filter[2]) { \
while (w >= 16) { \
vpx_highbd_filter_block1d16_##dir##8_##avg##opt( \
src_start, src_stride, dst, dst_stride, h, filter, bd); \
src += 16; \
dst += 16; \
w -= 16; \
} \
while (w >= 8) { \
vpx_highbd_filter_block1d8_##dir##8_##avg##opt( \
src_start, src_stride, dst, dst_stride, h, filter, bd); \
src += 8; \
dst += 8; \
w -= 8; \
} \
while (w >= 4) { \
vpx_highbd_filter_block1d4_##dir##8_##avg##opt( \
src_start, src_stride, dst, dst_stride, h, filter, bd); \
src += 4; \
dst += 4; \
w -= 4; \
} \
} else { \
while (w >= 16) { \
vpx_highbd_filter_block1d16_##dir##2_##avg##opt( \
src, src_stride, dst, dst_stride, h, filter, bd); \
src += 16; \
dst += 16; \
w -= 16; \
} \
while (w >= 8) { \
vpx_highbd_filter_block1d8_##dir##2_##avg##opt( \
src, src_stride, dst, dst_stride, h, filter, bd); \
src += 8; \
dst += 8; \
w -= 8; \
} \
while (w >= 4) { \
vpx_highbd_filter_block1d4_##dir##2_##avg##opt( \
src, src_stride, dst, dst_stride, h, filter, bd); \
src += 4; \
dst += 4; \
w -= 4; \
} \
} \
} \
if (w) { \
vpx_highbd_convolve8_##name##_c(src, src_stride, dst, dst_stride, \
filter_x, x_step_q4, filter_y, \
y_step_q4, w, h, bd); \
} \
#define HIGH_FUN_CONV_1D(name, offset, step_q4, dir, src_start, avg, opt) \
void vpx_highbd_convolve8_##name##_##opt( \
const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, \
ptrdiff_t dst_stride, const InterpKernel *filter_kernel, int x0_q4, \
int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bd) { \
const int16_t *filter = filter_kernel[offset]; \
if (step_q4 == 16 && filter[3] != 128) { \
if (filter[0] | filter[1] | filter[2]) { \
while (w >= 16) { \
vpx_highbd_filter_block1d16_##dir##8_##avg##opt( \
src_start, src_stride, dst, dst_stride, h, filter, bd); \
src += 16; \
dst += 16; \
w -= 16; \
} \
while (w >= 8) { \
vpx_highbd_filter_block1d8_##dir##8_##avg##opt( \
src_start, src_stride, dst, dst_stride, h, filter, bd); \
src += 8; \
dst += 8; \
w -= 8; \
} \
while (w >= 4) { \
vpx_highbd_filter_block1d4_##dir##8_##avg##opt( \
src_start, src_stride, dst, dst_stride, h, filter, bd); \
src += 4; \
dst += 4; \
w -= 4; \
} \
} else { \
while (w >= 16) { \
vpx_highbd_filter_block1d16_##dir##2_##avg##opt( \
src, src_stride, dst, dst_stride, h, filter, bd); \
src += 16; \
dst += 16; \
w -= 16; \
} \
while (w >= 8) { \
vpx_highbd_filter_block1d8_##dir##2_##avg##opt( \
src, src_stride, dst, dst_stride, h, filter, bd); \
src += 8; \
dst += 8; \
w -= 8; \
} \
while (w >= 4) { \
vpx_highbd_filter_block1d4_##dir##2_##avg##opt( \
src, src_stride, dst, dst_stride, h, filter, bd); \
src += 4; \
dst += 4; \
w -= 4; \
} \
} \
} \
if (w) { \
vpx_highbd_convolve8_##name##_c(src, src_stride, dst, dst_stride, \
filter_kernel, x0_q4, x_step_q4, y0_q4, \
y_step_q4, w, h, bd); \
} \
}
#define HIGH_FUN_CONV_2D(avg, opt) \
void vpx_highbd_convolve8_##avg##opt( \
const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, \
ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, \
const int16_t *filter_y, int y_step_q4, int w, int h, int bd) { \
assert(w <= 64); \
assert(h <= 64); \
if (x_step_q4 == 16 && y_step_q4 == 16) { \
if ((filter_x[0] | filter_x[1] | filter_x[2]) || filter_x[3] == 128) { \
DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 71]); \
vpx_highbd_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \
fdata2, 64, filter_x, x_step_q4, \
filter_y, y_step_q4, w, h + 7, bd); \
vpx_highbd_convolve8_##avg##vert_##opt( \
fdata2 + 192, 64, dst, dst_stride, filter_x, x_step_q4, filter_y, \
y_step_q4, w, h, bd); \
} else { \
DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 65]); \
vpx_highbd_convolve8_horiz_##opt(src, src_stride, fdata2, 64, \
filter_x, x_step_q4, filter_y, \
y_step_q4, w, h + 1, bd); \
vpx_highbd_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, \
filter_x, x_step_q4, filter_y, \
y_step_q4, w, h, bd); \
} \
} else { \
vpx_highbd_convolve8_##avg##c(src, src_stride, dst, dst_stride, \
filter_x, x_step_q4, filter_y, y_step_q4, \
w, h, bd); \
} \
#define HIGH_FUN_CONV_2D(avg, opt) \
void vpx_highbd_convolve8_##avg##opt( \
const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, \
ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, \
int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bd) { \
const int16_t *filter_x = filter[x0_q4]; \
assert(w <= 64); \
assert(h <= 64); \
if (x_step_q4 == 16 && y_step_q4 == 16) { \
if ((filter_x[0] | filter_x[1] | filter_x[2]) || filter_x[3] == 128) { \
DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 71]); \
vpx_highbd_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \
fdata2, 64, filter, x0_q4, x_step_q4, \
y0_q4, y_step_q4, w, h + 7, bd); \
vpx_highbd_convolve8_##avg##vert_##opt( \
fdata2 + 192, 64, dst, dst_stride, filter, x0_q4, x_step_q4, \
y0_q4, y_step_q4, w, h, bd); \
} else { \
DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 65]); \
vpx_highbd_convolve8_horiz_##opt(src, src_stride, fdata2, 64, filter, \
x0_q4, x_step_q4, y0_q4, y_step_q4, \
w, h + 1, bd); \
vpx_highbd_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, \
filter, x0_q4, x_step_q4, \
y0_q4, y_step_q4, w, h, bd); \
} \
} else { \
vpx_highbd_convolve8_##avg##c(src, src_stride, dst, dst_stride, filter, \
x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, \
bd); \
} \
}
#endif // CONFIG_VP9_HIGHBITDEPTH

View File

@ -18,13 +18,14 @@
void vpx_highbd_convolve_copy_avx2(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int width, int h, int bd) {
(void)filter_x;
(void)filter_y;
(void)filter_x_stride;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
(void)bd;
assert(width % 4 == 0);
@ -99,13 +100,14 @@ void vpx_highbd_convolve_copy_avx2(const uint16_t *src, ptrdiff_t src_stride,
void vpx_highbd_convolve_avg_avx2(const uint16_t *src, ptrdiff_t src_stride,
uint16_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int filter_x_stride,
const int16_t *filter_y, int filter_y_stride,
const InterpKernel *filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4,
int width, int h, int bd) {
(void)filter_x;
(void)filter_y;
(void)filter_x_stride;
(void)filter_y_stride;
(void)filter;
(void)x0_q4;
(void)x_step_q4;
(void)y0_q4;
(void)y_step_q4;
(void)bd;
assert(width % 4 == 0);
@ -1073,8 +1075,8 @@ void vpx_highbd_filter_block1d4_v2_sse2(const uint16_t *, ptrdiff_t, uint16_t *,
#define vpx_highbd_filter_block1d4_v8_avx2 vpx_highbd_filter_block1d4_v8_sse2
#define vpx_highbd_filter_block1d4_v2_avx2 vpx_highbd_filter_block1d4_v2_sse2
HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , avx2);
HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , avx2);
HIGH_FUN_CONV_1D(horiz, x0_q4, x_step_q4, h, src, , avx2);
HIGH_FUN_CONV_1D(vert, y0_q4, y_step_q4, v, src - src_stride * 3, , avx2);
HIGH_FUN_CONV_2D(, avx2);
void vpx_highbd_filter_block1d4_h8_avg_sse2(const uint16_t *, ptrdiff_t,
@ -1098,8 +1100,8 @@ void vpx_highbd_filter_block1d4_v2_avg_sse2(const uint16_t *, ptrdiff_t,
#define vpx_highbd_filter_block1d4_v2_avg_avx2 \
vpx_highbd_filter_block1d4_v2_avg_sse2
HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, avx2);
HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
HIGH_FUN_CONV_1D(avg_horiz, x0_q4, x_step_q4, h, src, avg_, avx2);
HIGH_FUN_CONV_1D(avg_vert, y0_q4, y_step_q4, v, src - src_stride * 3, avg_,
avx2);
HIGH_FUN_CONV_2D(avg_, avx2);

View File

@ -41,38 +41,38 @@ filter8_1dfunction vpx_filter_block1d4_h2_avg_sse2;
// void vpx_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
// void vpx_convolve8_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
// void vpx_convolve8_avg_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4,
// int y_step_q4, int w, int h);
// void vpx_convolve8_avg_vert_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, sse2);
FUN_CONV_1D(horiz, x0_q4, x_step_q4, h, src, , sse2);
FUN_CONV_1D(vert, y0_q4, y_step_q4, v, src - src_stride * 3, , sse2);
FUN_CONV_1D(avg_horiz, x0_q4, x_step_q4, h, src, avg_, sse2);
FUN_CONV_1D(avg_vert, y0_q4, y_step_q4, v, src - src_stride * 3, avg_, sse2);
// void vpx_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
// void vpx_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
FUN_CONV_2D(, sse2);
FUN_CONV_2D(avg_, sse2);
@ -140,22 +140,22 @@ highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h2_avg_sse2;
// const int16_t *filter_y,
// int y_step_q4,
// int w, int h, int bd);
HIGH_FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
HIGH_FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
HIGH_FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
HIGH_FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
HIGH_FUN_CONV_1D(horiz, x0_q4, x_step_q4, h, src, , sse2);
HIGH_FUN_CONV_1D(vert, y0_q4, y_step_q4, v, src - src_stride * 3, , sse2);
HIGH_FUN_CONV_1D(avg_horiz, x0_q4, x_step_q4, h, src, avg_, sse2);
HIGH_FUN_CONV_1D(avg_vert, y0_q4, y_step_q4, v, src - src_stride * 3, avg_,
sse2);
// void vpx_highbd_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h, int bd);
// void vpx_highbd_convolve8_avg_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h, int bd);
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4,
// int y_step_q4, int w, int h, int bd);
HIGH_FUN_CONV_2D(, sse2);
HIGH_FUN_CONV_2D(avg_, sse2);
#endif // CONFIG_VP9_HIGHBITDEPTH && ARCH_X86_64

View File

@ -20,14 +20,14 @@ SECTION .text
%endif
%ifidn %2, highbd
%define pavg pavgw
cglobal %2_convolve_%1, 4, 7, 4+AUX_XMM_REGS, src, src_stride, \
cglobal %2_convolve_%1, 4, 8, 4+AUX_XMM_REGS, src, src_stride, \
dst, dst_stride, \
fx, fxs, fy, fys, w, h, bd
f, fxo, fxs, fyo, fys, w, h, bd
%else
%define pavg pavgb
cglobal convolve_%1, 4, 7, 4+AUX_XMM_REGS, src, src_stride, \
cglobal convolve_%1, 4, 8, 4+AUX_XMM_REGS, src, src_stride, \
dst, dst_stride, \
fx, fxs, fy, fys, w, h
f, fxo, fxs, fyo, fys, w, h
%endif
mov r4d, dword wm
%ifidn %2, highbd

View File

@ -554,21 +554,21 @@ filter8_1dfunction vpx_filter_block1d4_h2_ssse3;
#define vpx_filter_block1d4_h2_avx2 vpx_filter_block1d4_h2_ssse3
// void vpx_convolve8_horiz_avx2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
// void vpx_convolve8_vert_avx2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , avx2);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , avx2);
FUN_CONV_1D(horiz, x0_q4, x_step_q4, h, src, , avx2);
FUN_CONV_1D(vert, y0_q4, y_step_q4, v, src - src_stride * 3, , avx2);
// void vpx_convolve8_avx2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
FUN_CONV_2D(, avx2);
#endif // HAVE_AX2 && HAVE_SSSE3

View File

@ -306,29 +306,28 @@ filter8_1dfunction vpx_filter_block1d4_h2_avg_ssse3;
// void vpx_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
// void vpx_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
// void vpx_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4,
// int y_step_q4, int w, int h);
// void vpx_convolve8_avg_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , ssse3);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , ssse3);
FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, ssse3);
FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
ssse3);
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4,
// int y_step_q4, int w, int h);
FUN_CONV_1D(horiz, x0_q4, x_step_q4, h, src, , ssse3);
FUN_CONV_1D(vert, y0_q4, y_step_q4, v, src - src_stride * 3, , ssse3);
FUN_CONV_1D(avg_horiz, x0_q4, x_step_q4, h, src, avg_, ssse3);
FUN_CONV_1D(avg_vert, y0_q4, y_step_q4, v, src - src_stride * 3, avg_, ssse3);
#define TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
out2, out3, out4, out5, out6, out7) \
@ -813,9 +812,9 @@ static void scaledconvolve_vert_w16(const uint8_t *src, ptrdiff_t src_stride,
static void scaledconvolve2d(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const InterpKernel *const x_filters, int x0_q4,
int x_step_q4, const InterpKernel *const y_filters,
int y0_q4, int y_step_q4, int w, int h) {
const InterpKernel *const filter, int x0_q4,
int x_step_q4, int y0_q4, int y_step_q4, int w,
int h) {
// Note: Fixed size intermediate buffer, temp, places limits on parameters.
// 2d filtering proceeds in 2 steps:
// (1) Interpolate horizontally into an intermediate buffer, temp.
@ -840,49 +839,43 @@ static void scaledconvolve2d(const uint8_t *src, ptrdiff_t src_stride,
if (w >= 8) {
scaledconvolve_horiz_w8(src - src_stride * (SUBPEL_TAPS / 2 - 1),
src_stride, temp, 64, x_filters, x0_q4, x_step_q4,
w, intermediate_height);
src_stride, temp, 64, filter, x0_q4, x_step_q4, w,
intermediate_height);
} else {
scaledconvolve_horiz_w4(src - src_stride * (SUBPEL_TAPS / 2 - 1),
src_stride, temp, 64, x_filters, x0_q4, x_step_q4,
w, intermediate_height);
src_stride, temp, 64, filter, x0_q4, x_step_q4, w,
intermediate_height);
}
if (w >= 16) {
scaledconvolve_vert_w16(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst,
dst_stride, y_filters, y0_q4, y_step_q4, w, h);
dst_stride, filter, y0_q4, y_step_q4, w, h);
} else if (w == 8) {
scaledconvolve_vert_w8(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst,
dst_stride, y_filters, y0_q4, y_step_q4, w, h);
dst_stride, filter, y0_q4, y_step_q4, w, h);
} else {
scaledconvolve_vert_w4(temp + 64 * (SUBPEL_TAPS / 2 - 1), 64, dst,
dst_stride, y_filters, y0_q4, y_step_q4, w, h);
dst_stride, filter, y0_q4, y_step_q4, w, h);
}
}
void vpx_scaled_2d_ssse3(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
ptrdiff_t dst_stride, const int16_t *filter_x,
int x_step_q4, const int16_t *filter_y, int y_step_q4,
ptrdiff_t dst_stride, const InterpKernel *filter,
int x0_q4, int x_step_q4, int y0_q4, int y_step_q4,
int w, int h) {
const InterpKernel *const filters_x = get_filter_base(filter_x);
const int x0_q4 = get_filter_offset(filter_x, filters_x);
const InterpKernel *const filters_y = get_filter_base(filter_y);
const int y0_q4 = get_filter_offset(filter_y, filters_y);
scaledconvolve2d(src, src_stride, dst, dst_stride, filters_x, x0_q4,
x_step_q4, filters_y, y0_q4, y_step_q4, w, h);
scaledconvolve2d(src, src_stride, dst, dst_stride, filter, x0_q4, x_step_q4,
y0_q4, y_step_q4, w, h);
}
// void vp9_convolve8_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
// void vpx_convolve8_avg_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// const InterpKernel *filter, int x0_q4,
// int32_t x_step_q4, int y0_q4, int y_step_q4,
// int w, int h);
FUN_CONV_2D(, ssse3);
FUN_CONV_2D(avg_, ssse3);