2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
14 #include "third_party/googletest/src/include/gtest/gtest.h"
16 #include "./vp9_rtcd.h"
17 #include "./vpx_config.h"
18 #include "./vpx_dsp_rtcd.h"
19 #include "test/acm_random.h"
20 #include "test/clear_system_state.h"
21 #include "test/register_state_check.h"
22 #include "test/util.h"
23 #include "vp9/common/vp9_common.h"
24 #include "vp9/common/vp9_filter.h"
25 #include "vpx_dsp/vpx_dsp_common.h"
26 #include "vpx_dsp/vpx_filter.h"
27 #include "vpx_mem/vpx_mem.h"
28 #include "vpx_ports/mem.h"
29 #include "vpx_ports/vpx_timer.h"
33 static const unsigned int kMaxDimension = 64;
35 typedef void (*ConvolveFunc)(const uint8_t *src, ptrdiff_t src_stride,
36 uint8_t *dst, ptrdiff_t dst_stride,
37 const InterpKernel *filter, int x0_q4,
38 int x_step_q4, int y0_q4, int y_step_q4, int w,
41 typedef void (*WrapperFilterBlock2d8Func)(
42 const uint8_t *src_ptr, const unsigned int src_stride,
43 const int16_t *hfilter, const int16_t *vfilter, uint8_t *dst_ptr,
44 unsigned int dst_stride, unsigned int output_width,
45 unsigned int output_height, int use_highbd);
47 struct ConvolveFunctions {
48 ConvolveFunctions(ConvolveFunc copy, ConvolveFunc avg, ConvolveFunc h8,
49 ConvolveFunc h8_avg, ConvolveFunc v8, ConvolveFunc v8_avg,
50 ConvolveFunc hv8, ConvolveFunc hv8_avg, ConvolveFunc sh8,
51 ConvolveFunc sh8_avg, ConvolveFunc sv8,
52 ConvolveFunc sv8_avg, ConvolveFunc shv8,
53 ConvolveFunc shv8_avg, int bd)
71 ConvolveFunc copy_[2];
75 ConvolveFunc sh8_[2]; // scaled horiz
76 ConvolveFunc sv8_[2]; // scaled vert
77 ConvolveFunc shv8_[2]; // scaled horiz/vert
78 int use_highbd_; // 0 if high bitdepth not used, else the actual bit depth.
81 typedef std::tuple<int, int, const ConvolveFunctions *> ConvolveParam;
83 #define ALL_SIZES(convolve_fn) \
84 make_tuple(4, 4, &convolve_fn), make_tuple(8, 4, &convolve_fn), \
85 make_tuple(4, 8, &convolve_fn), make_tuple(8, 8, &convolve_fn), \
86 make_tuple(16, 8, &convolve_fn), make_tuple(8, 16, &convolve_fn), \
87 make_tuple(16, 16, &convolve_fn), make_tuple(32, 16, &convolve_fn), \
88 make_tuple(16, 32, &convolve_fn), make_tuple(32, 32, &convolve_fn), \
89 make_tuple(64, 32, &convolve_fn), make_tuple(32, 64, &convolve_fn), \
90 make_tuple(64, 64, &convolve_fn)
92 // Reference 8-tap subpixel filter, slightly modified to fit into this test.
93 #define VP9_FILTER_WEIGHT 128
94 #define VP9_FILTER_SHIFT 7
95 uint8_t clip_pixel(int x) { return x < 0 ? 0 : x > 255 ? 255 : x; }
97 void filter_block2d_8_c(const uint8_t *src_ptr, const unsigned int src_stride,
98 const int16_t *hfilter, const int16_t *vfilter,
99 uint8_t *dst_ptr, unsigned int dst_stride,
100 unsigned int output_width, unsigned int output_height) {
101 // Between passes, we use an intermediate buffer whose height is extended to
102 // have enough horizontally filtered values as input for the vertical pass.
103 // This buffer is allocated to be big enough for the largest block type we
105 const int kInterp_Extend = 4;
106 const unsigned int intermediate_height =
107 (kInterp_Extend - 1) + output_height + kInterp_Extend;
110 // Size of intermediate_buffer is max_intermediate_height * filter_max_width,
111 // where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
115 // and filter_max_width = 16
117 uint8_t intermediate_buffer[71 * kMaxDimension];
118 vp9_zero(intermediate_buffer);
119 const int intermediate_next_stride =
120 1 - static_cast<int>(intermediate_height * output_width);
122 // Horizontal pass (src -> transposed intermediate).
123 uint8_t *output_ptr = intermediate_buffer;
124 const int src_next_row_stride = src_stride - output_width;
125 src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
126 for (i = 0; i < intermediate_height; ++i) {
127 for (j = 0; j < output_width; ++j) {
129 const int temp = (src_ptr[0] * hfilter[0]) + (src_ptr[1] * hfilter[1]) +
130 (src_ptr[2] * hfilter[2]) + (src_ptr[3] * hfilter[3]) +
131 (src_ptr[4] * hfilter[4]) + (src_ptr[5] * hfilter[5]) +
132 (src_ptr[6] * hfilter[6]) + (src_ptr[7] * hfilter[7]) +
133 (VP9_FILTER_WEIGHT >> 1); // Rounding
135 // Normalize back to 0-255...
136 *output_ptr = clip_pixel(temp >> VP9_FILTER_SHIFT);
138 output_ptr += intermediate_height;
140 src_ptr += src_next_row_stride;
141 output_ptr += intermediate_next_stride;
144 // Vertical pass (transposed intermediate -> dst).
145 src_ptr = intermediate_buffer;
146 const int dst_next_row_stride = dst_stride - output_width;
147 for (i = 0; i < output_height; ++i) {
148 for (j = 0; j < output_width; ++j) {
150 const int temp = (src_ptr[0] * vfilter[0]) + (src_ptr[1] * vfilter[1]) +
151 (src_ptr[2] * vfilter[2]) + (src_ptr[3] * vfilter[3]) +
152 (src_ptr[4] * vfilter[4]) + (src_ptr[5] * vfilter[5]) +
153 (src_ptr[6] * vfilter[6]) + (src_ptr[7] * vfilter[7]) +
154 (VP9_FILTER_WEIGHT >> 1); // Rounding
156 // Normalize back to 0-255...
157 *dst_ptr++ = clip_pixel(temp >> VP9_FILTER_SHIFT);
158 src_ptr += intermediate_height;
160 src_ptr += intermediate_next_stride;
161 dst_ptr += dst_next_row_stride;
165 void block2d_average_c(uint8_t *src, unsigned int src_stride,
166 uint8_t *output_ptr, unsigned int output_stride,
167 unsigned int output_width, unsigned int output_height) {
169 for (i = 0; i < output_height; ++i) {
170 for (j = 0; j < output_width; ++j) {
171 output_ptr[j] = (output_ptr[j] + src[i * src_stride + j] + 1) >> 1;
173 output_ptr += output_stride;
177 void filter_average_block2d_8_c(const uint8_t *src_ptr,
178 const unsigned int src_stride,
179 const int16_t *hfilter, const int16_t *vfilter,
180 uint8_t *dst_ptr, unsigned int dst_stride,
181 unsigned int output_width,
182 unsigned int output_height) {
183 uint8_t tmp[kMaxDimension * kMaxDimension];
185 assert(output_width <= kMaxDimension);
186 assert(output_height <= kMaxDimension);
187 filter_block2d_8_c(src_ptr, src_stride, hfilter, vfilter, tmp, 64,
188 output_width, output_height);
189 block2d_average_c(tmp, 64, dst_ptr, dst_stride, output_width, output_height);
192 #if CONFIG_VP9_HIGHBITDEPTH
193 void highbd_filter_block2d_8_c(const uint16_t *src_ptr,
194 const unsigned int src_stride,
195 const int16_t *hfilter, const int16_t *vfilter,
196 uint16_t *dst_ptr, unsigned int dst_stride,
197 unsigned int output_width,
198 unsigned int output_height, int bd) {
199 // Between passes, we use an intermediate buffer whose height is extended to
200 // have enough horizontally filtered values as input for the vertical pass.
201 // This buffer is allocated to be big enough for the largest block type we
203 const int kInterp_Extend = 4;
204 const unsigned int intermediate_height =
205 (kInterp_Extend - 1) + output_height + kInterp_Extend;
207 /* Size of intermediate_buffer is max_intermediate_height * filter_max_width,
208 * where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
212 * and filter_max_width = 16
214 uint16_t intermediate_buffer[71 * kMaxDimension];
215 const int intermediate_next_stride =
216 1 - static_cast<int>(intermediate_height * output_width);
218 vp9_zero(intermediate_buffer);
220 // Horizontal pass (src -> transposed intermediate).
222 uint16_t *output_ptr = intermediate_buffer;
223 const int src_next_row_stride = src_stride - output_width;
225 src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
226 for (i = 0; i < intermediate_height; ++i) {
227 for (j = 0; j < output_width; ++j) {
229 const int temp = (src_ptr[0] * hfilter[0]) + (src_ptr[1] * hfilter[1]) +
230 (src_ptr[2] * hfilter[2]) + (src_ptr[3] * hfilter[3]) +
231 (src_ptr[4] * hfilter[4]) + (src_ptr[5] * hfilter[5]) +
232 (src_ptr[6] * hfilter[6]) + (src_ptr[7] * hfilter[7]) +
233 (VP9_FILTER_WEIGHT >> 1); // Rounding
235 // Normalize back to 0-255...
236 *output_ptr = clip_pixel_highbd(temp >> VP9_FILTER_SHIFT, bd);
238 output_ptr += intermediate_height;
240 src_ptr += src_next_row_stride;
241 output_ptr += intermediate_next_stride;
245 // Vertical pass (transposed intermediate -> dst).
247 uint16_t *src_ptr = intermediate_buffer;
248 const int dst_next_row_stride = dst_stride - output_width;
250 for (i = 0; i < output_height; ++i) {
251 for (j = 0; j < output_width; ++j) {
253 const int temp = (src_ptr[0] * vfilter[0]) + (src_ptr[1] * vfilter[1]) +
254 (src_ptr[2] * vfilter[2]) + (src_ptr[3] * vfilter[3]) +
255 (src_ptr[4] * vfilter[4]) + (src_ptr[5] * vfilter[5]) +
256 (src_ptr[6] * vfilter[6]) + (src_ptr[7] * vfilter[7]) +
257 (VP9_FILTER_WEIGHT >> 1); // Rounding
259 // Normalize back to 0-255...
260 *dst_ptr++ = clip_pixel_highbd(temp >> VP9_FILTER_SHIFT, bd);
261 src_ptr += intermediate_height;
263 src_ptr += intermediate_next_stride;
264 dst_ptr += dst_next_row_stride;
269 void highbd_block2d_average_c(uint16_t *src, unsigned int src_stride,
270 uint16_t *output_ptr, unsigned int output_stride,
271 unsigned int output_width,
272 unsigned int output_height) {
274 for (i = 0; i < output_height; ++i) {
275 for (j = 0; j < output_width; ++j) {
276 output_ptr[j] = (output_ptr[j] + src[i * src_stride + j] + 1) >> 1;
278 output_ptr += output_stride;
282 void highbd_filter_average_block2d_8_c(
283 const uint16_t *src_ptr, const unsigned int src_stride,
284 const int16_t *hfilter, const int16_t *vfilter, uint16_t *dst_ptr,
285 unsigned int dst_stride, unsigned int output_width,
286 unsigned int output_height, int bd) {
287 uint16_t tmp[kMaxDimension * kMaxDimension];
289 assert(output_width <= kMaxDimension);
290 assert(output_height <= kMaxDimension);
291 highbd_filter_block2d_8_c(src_ptr, src_stride, hfilter, vfilter, tmp, 64,
292 output_width, output_height, bd);
293 highbd_block2d_average_c(tmp, 64, dst_ptr, dst_stride, output_width,
296 #endif // CONFIG_VP9_HIGHBITDEPTH
298 void wrapper_filter_average_block2d_8_c(
299 const uint8_t *src_ptr, const unsigned int src_stride,
300 const int16_t *hfilter, const int16_t *vfilter, uint8_t *dst_ptr,
301 unsigned int dst_stride, unsigned int output_width,
302 unsigned int output_height, int use_highbd) {
303 #if CONFIG_VP9_HIGHBITDEPTH
304 if (use_highbd == 0) {
305 filter_average_block2d_8_c(src_ptr, src_stride, hfilter, vfilter, dst_ptr,
306 dst_stride, output_width, output_height);
308 highbd_filter_average_block2d_8_c(CAST_TO_SHORTPTR(src_ptr), src_stride,
310 CAST_TO_SHORTPTR(dst_ptr), dst_stride,
311 output_width, output_height, use_highbd);
314 ASSERT_EQ(0, use_highbd);
315 filter_average_block2d_8_c(src_ptr, src_stride, hfilter, vfilter, dst_ptr,
316 dst_stride, output_width, output_height);
320 void wrapper_filter_block2d_8_c(const uint8_t *src_ptr,
321 const unsigned int src_stride,
322 const int16_t *hfilter, const int16_t *vfilter,
323 uint8_t *dst_ptr, unsigned int dst_stride,
324 unsigned int output_width,
325 unsigned int output_height, int use_highbd) {
326 #if CONFIG_VP9_HIGHBITDEPTH
327 if (use_highbd == 0) {
328 filter_block2d_8_c(src_ptr, src_stride, hfilter, vfilter, dst_ptr,
329 dst_stride, output_width, output_height);
331 highbd_filter_block2d_8_c(CAST_TO_SHORTPTR(src_ptr), src_stride, hfilter,
332 vfilter, CAST_TO_SHORTPTR(dst_ptr), dst_stride,
333 output_width, output_height, use_highbd);
336 ASSERT_EQ(0, use_highbd);
337 filter_block2d_8_c(src_ptr, src_stride, hfilter, vfilter, dst_ptr, dst_stride,
338 output_width, output_height);
342 class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
344 static void SetUpTestSuite() {
345 // Force input_ to be unaligned, output to be 16 byte aligned.
346 input_ = reinterpret_cast<uint8_t *>(
347 vpx_memalign(kDataAlignment, kInputBufferSize + 1)) +
349 output_ = reinterpret_cast<uint8_t *>(
350 vpx_memalign(kDataAlignment, kOutputBufferSize));
351 output_ref_ = reinterpret_cast<uint8_t *>(
352 vpx_memalign(kDataAlignment, kOutputBufferSize));
353 #if CONFIG_VP9_HIGHBITDEPTH
354 input16_ = reinterpret_cast<uint16_t *>(vpx_memalign(
355 kDataAlignment, (kInputBufferSize + 1) * sizeof(uint16_t))) +
357 output16_ = reinterpret_cast<uint16_t *>(
358 vpx_memalign(kDataAlignment, (kOutputBufferSize) * sizeof(uint16_t)));
359 output16_ref_ = reinterpret_cast<uint16_t *>(
360 vpx_memalign(kDataAlignment, (kOutputBufferSize) * sizeof(uint16_t)));
364 virtual void TearDown() { libvpx_test::ClearSystemState(); }
366 static void TearDownTestSuite() {
367 vpx_free(input_ - 1);
371 vpx_free(output_ref_);
372 output_ref_ = nullptr;
373 #if CONFIG_VP9_HIGHBITDEPTH
374 vpx_free(input16_ - 1);
378 vpx_free(output16_ref_);
379 output16_ref_ = nullptr;
384 static const int kDataAlignment = 16;
385 static const int kOuterBlockSize = 256;
386 static const int kInputStride = kOuterBlockSize;
387 static const int kOutputStride = kOuterBlockSize;
388 static const int kInputBufferSize = kOuterBlockSize * kOuterBlockSize;
389 static const int kOutputBufferSize = kOuterBlockSize * kOuterBlockSize;
391 int Width() const { return GET_PARAM(0); }
392 int Height() const { return GET_PARAM(1); }
393 int BorderLeft() const {
394 const int center = (kOuterBlockSize - Width()) / 2;
395 return (center + (kDataAlignment - 1)) & ~(kDataAlignment - 1);
397 int BorderTop() const { return (kOuterBlockSize - Height()) / 2; }
399 bool IsIndexInBorder(int i) {
400 return (i < BorderTop() * kOuterBlockSize ||
401 i >= (BorderTop() + Height()) * kOuterBlockSize ||
402 i % kOuterBlockSize < BorderLeft() ||
403 i % kOuterBlockSize >= (BorderLeft() + Width()));
406 virtual void SetUp() {
408 #if CONFIG_VP9_HIGHBITDEPTH
409 if (UUT_->use_highbd_ != 0) {
410 mask_ = (1 << UUT_->use_highbd_) - 1;
415 /* Set up guard blocks for an inner block centered in the outer block */
416 for (int i = 0; i < kOutputBufferSize; ++i) {
417 if (IsIndexInBorder(i)) {
419 #if CONFIG_VP9_HIGHBITDEPTH
420 output16_[i] = mask_;
424 #if CONFIG_VP9_HIGHBITDEPTH
430 ::libvpx_test::ACMRandom prng;
431 for (int i = 0; i < kInputBufferSize; ++i) {
434 #if CONFIG_VP9_HIGHBITDEPTH
438 input_[i] = prng.Rand8Extremes();
439 #if CONFIG_VP9_HIGHBITDEPTH
440 input16_[i] = prng.Rand16() & mask_;
446 void SetConstantInput(int value) {
447 memset(input_, value, kInputBufferSize);
448 #if CONFIG_VP9_HIGHBITDEPTH
449 vpx_memset16(input16_, value, kInputBufferSize);
453 void CopyOutputToRef() {
454 memcpy(output_ref_, output_, kOutputBufferSize);
455 #if CONFIG_VP9_HIGHBITDEPTH
456 memcpy(output16_ref_, output16_,
457 kOutputBufferSize * sizeof(output16_ref_[0]));
461 void CheckGuardBlocks() {
462 for (int i = 0; i < kOutputBufferSize; ++i) {
463 if (IsIndexInBorder(i)) {
464 EXPECT_EQ(255, output_[i]);
469 uint8_t *input() const {
470 const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
471 #if CONFIG_VP9_HIGHBITDEPTH
472 if (UUT_->use_highbd_ == 0) {
473 return input_ + offset;
475 return CAST_TO_BYTEPTR(input16_ + offset);
478 return input_ + offset;
482 uint8_t *output() const {
483 const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
484 #if CONFIG_VP9_HIGHBITDEPTH
485 if (UUT_->use_highbd_ == 0) {
486 return output_ + offset;
488 return CAST_TO_BYTEPTR(output16_ + offset);
491 return output_ + offset;
495 uint8_t *output_ref() const {
496 const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
497 #if CONFIG_VP9_HIGHBITDEPTH
498 if (UUT_->use_highbd_ == 0) {
499 return output_ref_ + offset;
501 return CAST_TO_BYTEPTR(output16_ref_ + offset);
504 return output_ref_ + offset;
508 uint16_t lookup(uint8_t *list, int index) const {
509 #if CONFIG_VP9_HIGHBITDEPTH
510 if (UUT_->use_highbd_ == 0) {
513 return CAST_TO_SHORTPTR(list)[index];
520 void assign_val(uint8_t *list, int index, uint16_t val) const {
521 #if CONFIG_VP9_HIGHBITDEPTH
522 if (UUT_->use_highbd_ == 0) {
523 list[index] = (uint8_t)val;
525 CAST_TO_SHORTPTR(list)[index] = val;
528 list[index] = (uint8_t)val;
532 const ConvolveFunctions *UUT_;
533 static uint8_t *input_;
534 static uint8_t *output_;
535 static uint8_t *output_ref_;
536 #if CONFIG_VP9_HIGHBITDEPTH
537 static uint16_t *input16_;
538 static uint16_t *output16_;
539 static uint16_t *output16_ref_;
544 uint8_t *ConvolveTest::input_ = nullptr;
545 uint8_t *ConvolveTest::output_ = nullptr;
546 uint8_t *ConvolveTest::output_ref_ = nullptr;
547 #if CONFIG_VP9_HIGHBITDEPTH
548 uint16_t *ConvolveTest::input16_ = nullptr;
549 uint16_t *ConvolveTest::output16_ = nullptr;
550 uint16_t *ConvolveTest::output16_ref_ = nullptr;
553 TEST_P(ConvolveTest, GuardBlocks) { CheckGuardBlocks(); }
555 TEST_P(ConvolveTest, DISABLED_Copy_Speed) {
556 const uint8_t *const in = input();
557 uint8_t *const out = output();
558 const int kNumTests = 5000000;
559 const int width = Width();
560 const int height = Height();
561 vpx_usec_timer timer;
563 vpx_usec_timer_start(&timer);
564 for (int n = 0; n < kNumTests; ++n) {
565 UUT_->copy_[0](in, kInputStride, out, kOutputStride, nullptr, 0, 0, 0, 0,
568 vpx_usec_timer_mark(&timer);
570 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
571 printf("convolve_copy_%dx%d_%d: %d us\n", width, height,
572 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
575 TEST_P(ConvolveTest, DISABLED_Avg_Speed) {
576 const uint8_t *const in = input();
577 uint8_t *const out = output();
578 const int kNumTests = 5000000;
579 const int width = Width();
580 const int height = Height();
581 vpx_usec_timer timer;
583 vpx_usec_timer_start(&timer);
584 for (int n = 0; n < kNumTests; ++n) {
585 UUT_->copy_[1](in, kInputStride, out, kOutputStride, nullptr, 0, 0, 0, 0,
588 vpx_usec_timer_mark(&timer);
590 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
591 printf("convolve_avg_%dx%d_%d: %d us\n", width, height,
592 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
595 TEST_P(ConvolveTest, DISABLED_Scale_Speed) {
596 const uint8_t *const in = input();
597 uint8_t *const out = output();
598 const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP];
599 const int kNumTests = 5000000;
600 const int width = Width();
601 const int height = Height();
602 vpx_usec_timer timer;
604 SetConstantInput(127);
606 vpx_usec_timer_start(&timer);
607 for (int n = 0; n < kNumTests; ++n) {
608 UUT_->shv8_[0](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
611 vpx_usec_timer_mark(&timer);
613 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
614 printf("convolve_scale_%dx%d_%d: %d us\n", width, height,
615 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
618 TEST_P(ConvolveTest, DISABLED_8Tap_Speed) {
619 const uint8_t *const in = input();
620 uint8_t *const out = output();
621 const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP_SHARP];
622 const int kNumTests = 5000000;
623 const int width = Width();
624 const int height = Height();
625 vpx_usec_timer timer;
627 SetConstantInput(127);
629 vpx_usec_timer_start(&timer);
630 for (int n = 0; n < kNumTests; ++n) {
631 UUT_->hv8_[0](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
634 vpx_usec_timer_mark(&timer);
636 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
637 printf("convolve8_%dx%d_%d: %d us\n", width, height,
638 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
641 TEST_P(ConvolveTest, DISABLED_8Tap_Horiz_Speed) {
642 const uint8_t *const in = input();
643 uint8_t *const out = output();
644 const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP_SHARP];
645 const int kNumTests = 5000000;
646 const int width = Width();
647 const int height = Height();
648 vpx_usec_timer timer;
650 SetConstantInput(127);
652 vpx_usec_timer_start(&timer);
653 for (int n = 0; n < kNumTests; ++n) {
654 UUT_->h8_[0](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
657 vpx_usec_timer_mark(&timer);
659 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
660 printf("convolve8_horiz_%dx%d_%d: %d us\n", width, height,
661 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
664 TEST_P(ConvolveTest, DISABLED_8Tap_Vert_Speed) {
665 const uint8_t *const in = input();
666 uint8_t *const out = output();
667 const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP_SHARP];
668 const int kNumTests = 5000000;
669 const int width = Width();
670 const int height = Height();
671 vpx_usec_timer timer;
673 SetConstantInput(127);
675 vpx_usec_timer_start(&timer);
676 for (int n = 0; n < kNumTests; ++n) {
677 UUT_->v8_[0](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
680 vpx_usec_timer_mark(&timer);
682 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
683 printf("convolve8_vert_%dx%d_%d: %d us\n", width, height,
684 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
687 TEST_P(ConvolveTest, DISABLED_4Tap_Speed) {
688 const uint8_t *const in = input();
689 uint8_t *const out = output();
690 const InterpKernel *const fourtap = vp9_filter_kernels[FOURTAP];
691 const int kNumTests = 5000000;
692 const int width = Width();
693 const int height = Height();
694 vpx_usec_timer timer;
696 SetConstantInput(127);
698 vpx_usec_timer_start(&timer);
699 for (int n = 0; n < kNumTests; ++n) {
700 UUT_->hv8_[0](in, kInputStride, out, kOutputStride, fourtap, 8, 16, 8, 16,
703 vpx_usec_timer_mark(&timer);
705 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
706 printf("convolve4_%dx%d_%d: %d us\n", width, height,
707 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
710 TEST_P(ConvolveTest, DISABLED_4Tap_Horiz_Speed) {
711 const uint8_t *const in = input();
712 uint8_t *const out = output();
713 const InterpKernel *const fourtap = vp9_filter_kernels[FOURTAP];
714 const int kNumTests = 5000000;
715 const int width = Width();
716 const int height = Height();
717 vpx_usec_timer timer;
719 SetConstantInput(127);
721 vpx_usec_timer_start(&timer);
722 for (int n = 0; n < kNumTests; ++n) {
723 UUT_->h8_[0](in, kInputStride, out, kOutputStride, fourtap, 8, 16, 8, 16,
726 vpx_usec_timer_mark(&timer);
728 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
729 printf("convolve4_horiz_%dx%d_%d: %d us\n", width, height,
730 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
733 TEST_P(ConvolveTest, DISABLED_4Tap_Vert_Speed) {
734 const uint8_t *const in = input();
735 uint8_t *const out = output();
736 const InterpKernel *const fourtap = vp9_filter_kernels[FOURTAP];
737 const int kNumTests = 5000000;
738 const int width = Width();
739 const int height = Height();
740 vpx_usec_timer timer;
742 SetConstantInput(127);
744 vpx_usec_timer_start(&timer);
745 for (int n = 0; n < kNumTests; ++n) {
746 UUT_->v8_[0](in, kInputStride, out, kOutputStride, fourtap, 8, 16, 8, 16,
749 vpx_usec_timer_mark(&timer);
751 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
752 printf("convolve4_vert_%dx%d_%d: %d us\n", width, height,
753 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
755 TEST_P(ConvolveTest, DISABLED_8Tap_Avg_Speed) {
756 const uint8_t *const in = input();
757 uint8_t *const out = output();
758 const InterpKernel *const eighttap = vp9_filter_kernels[EIGHTTAP_SHARP];
759 const int kNumTests = 5000000;
760 const int width = Width();
761 const int height = Height();
762 vpx_usec_timer timer;
764 SetConstantInput(127);
766 vpx_usec_timer_start(&timer);
767 for (int n = 0; n < kNumTests; ++n) {
768 UUT_->hv8_[1](in, kInputStride, out, kOutputStride, eighttap, 8, 16, 8, 16,
771 vpx_usec_timer_mark(&timer);
773 const int elapsed_time = static_cast<int>(vpx_usec_timer_elapsed(&timer));
774 printf("convolve8_avg_%dx%d_%d: %d us\n", width, height,
775 UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
778 TEST_P(ConvolveTest, Copy) {
779 uint8_t *const in = input();
780 uint8_t *const out = output();
782 ASM_REGISTER_STATE_CHECK(UUT_->copy_[0](in, kInputStride, out, kOutputStride,
783 nullptr, 0, 0, 0, 0, Width(),
788 for (int y = 0; y < Height(); ++y) {
789 for (int x = 0; x < Width(); ++x)
790 ASSERT_EQ(lookup(out, y * kOutputStride + x),
791 lookup(in, y * kInputStride + x))
792 << "(" << x << "," << y << ")";
796 TEST_P(ConvolveTest, Avg) {
797 uint8_t *const in = input();
798 uint8_t *const out = output();
799 uint8_t *const out_ref = output_ref();
802 ASM_REGISTER_STATE_CHECK(UUT_->copy_[1](in, kInputStride, out, kOutputStride,
803 nullptr, 0, 0, 0, 0, Width(),
808 for (int y = 0; y < Height(); ++y) {
809 for (int x = 0; x < Width(); ++x)
810 ASSERT_EQ(lookup(out, y * kOutputStride + x),
811 ROUND_POWER_OF_TWO(lookup(in, y * kInputStride + x) +
812 lookup(out_ref, y * kOutputStride + x),
814 << "(" << x << "," << y << ")";
818 TEST_P(ConvolveTest, CopyHoriz) {
819 uint8_t *const in = input();
820 uint8_t *const out = output();
822 ASM_REGISTER_STATE_CHECK(UUT_->sh8_[0](in, kInputStride, out, kOutputStride,
823 vp9_filter_kernels[0], 0, 16, 0, 16,
828 for (int y = 0; y < Height(); ++y) {
829 for (int x = 0; x < Width(); ++x)
830 ASSERT_EQ(lookup(out, y * kOutputStride + x),
831 lookup(in, y * kInputStride + x))
832 << "(" << x << "," << y << ")";
836 TEST_P(ConvolveTest, CopyVert) {
837 uint8_t *const in = input();
838 uint8_t *const out = output();
840 ASM_REGISTER_STATE_CHECK(UUT_->sv8_[0](in, kInputStride, out, kOutputStride,
841 vp9_filter_kernels[0], 0, 16, 0, 16,
846 for (int y = 0; y < Height(); ++y) {
847 for (int x = 0; x < Width(); ++x)
848 ASSERT_EQ(lookup(out, y * kOutputStride + x),
849 lookup(in, y * kInputStride + x))
850 << "(" << x << "," << y << ")";
854 TEST_P(ConvolveTest, Copy2D) {
855 uint8_t *const in = input();
856 uint8_t *const out = output();
858 ASM_REGISTER_STATE_CHECK(UUT_->shv8_[0](in, kInputStride, out, kOutputStride,
859 vp9_filter_kernels[0], 0, 16, 0, 16,
864 for (int y = 0; y < Height(); ++y) {
865 for (int x = 0; x < Width(); ++x)
866 ASSERT_EQ(lookup(out, y * kOutputStride + x),
867 lookup(in, y * kInputStride + x))
868 << "(" << x << "," << y << ")";
872 const int kNumFilterBanks = 5;
873 const int kNumFilters = 16;
875 TEST(ConvolveTest, FiltersWontSaturateWhenAddedPairwise) {
876 for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
877 const InterpKernel *filters =
878 vp9_filter_kernels[static_cast<INTERP_FILTER>(filter_bank)];
879 for (int i = 0; i < kNumFilters; i++) {
880 const int p0 = filters[i][0] + filters[i][1];
881 const int p1 = filters[i][2] + filters[i][3];
882 const int p2 = filters[i][4] + filters[i][5];
883 const int p3 = filters[i][6] + filters[i][7];
888 EXPECT_LE(p0 + p3, 128);
889 EXPECT_LE(p0 + p3 + p1, 128);
890 EXPECT_LE(p0 + p3 + p1 + p2, 128);
891 EXPECT_EQ(p0 + p1 + p2 + p3, 128);
896 const WrapperFilterBlock2d8Func wrapper_filter_block2d_8[2] = {
897 wrapper_filter_block2d_8_c, wrapper_filter_average_block2d_8_c
900 TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
901 for (int i = 0; i < 2; ++i) {
902 uint8_t *const in = input();
903 uint8_t *const out = output();
904 #if CONFIG_VP9_HIGHBITDEPTH
905 uint8_t ref8[kOutputStride * kMaxDimension];
906 uint16_t ref16[kOutputStride * kMaxDimension];
908 if (UUT_->use_highbd_ == 0) {
911 ref = CAST_TO_BYTEPTR(ref16);
914 uint8_t ref[kOutputStride * kMaxDimension];
917 // Populate ref and out with some random data
918 ::libvpx_test::ACMRandom prng;
919 for (int y = 0; y < Height(); ++y) {
920 for (int x = 0; x < Width(); ++x) {
922 #if CONFIG_VP9_HIGHBITDEPTH
923 if (UUT_->use_highbd_ == 0 || UUT_->use_highbd_ == 8) {
924 r = prng.Rand8Extremes();
926 r = prng.Rand16() & mask_;
929 r = prng.Rand8Extremes();
932 assign_val(out, y * kOutputStride + x, r);
933 assign_val(ref, y * kOutputStride + x, r);
937 for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
938 const InterpKernel *filters =
939 vp9_filter_kernels[static_cast<INTERP_FILTER>(filter_bank)];
941 for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
942 for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
943 wrapper_filter_block2d_8[i](in, kInputStride, filters[filter_x],
944 filters[filter_y], ref, kOutputStride,
945 Width(), Height(), UUT_->use_highbd_);
947 if (filter_x && filter_y)
948 ASM_REGISTER_STATE_CHECK(
949 UUT_->hv8_[i](in, kInputStride, out, kOutputStride, filters,
950 filter_x, 16, filter_y, 16, Width(), Height()));
952 ASM_REGISTER_STATE_CHECK(
953 UUT_->v8_[i](in, kInputStride, out, kOutputStride, filters, 0,
954 16, filter_y, 16, Width(), Height()));
956 ASM_REGISTER_STATE_CHECK(
957 UUT_->h8_[i](in, kInputStride, out, kOutputStride, filters,
958 filter_x, 16, 0, 16, Width(), Height()));
960 ASM_REGISTER_STATE_CHECK(
961 UUT_->copy_[i](in, kInputStride, out, kOutputStride, nullptr, 0,
962 0, 0, 0, Width(), Height()));
966 for (int y = 0; y < Height(); ++y) {
967 for (int x = 0; x < Width(); ++x)
968 ASSERT_EQ(lookup(ref, y * kOutputStride + x),
969 lookup(out, y * kOutputStride + x))
970 << "mismatch at (" << x << "," << y << "), "
971 << "filters (" << filter_bank << "," << filter_x << ","
980 TEST_P(ConvolveTest, FilterExtremes) {
981 uint8_t *const in = input();
982 uint8_t *const out = output();
983 #if CONFIG_VP9_HIGHBITDEPTH
984 uint8_t ref8[kOutputStride * kMaxDimension];
985 uint16_t ref16[kOutputStride * kMaxDimension];
987 if (UUT_->use_highbd_ == 0) {
990 ref = CAST_TO_BYTEPTR(ref16);
993 uint8_t ref[kOutputStride * kMaxDimension];
996 // Populate ref and out with some random data
997 ::libvpx_test::ACMRandom prng;
998 for (int y = 0; y < Height(); ++y) {
999 for (int x = 0; x < Width(); ++x) {
1001 #if CONFIG_VP9_HIGHBITDEPTH
1002 if (UUT_->use_highbd_ == 0 || UUT_->use_highbd_ == 8) {
1003 r = prng.Rand8Extremes();
1005 r = prng.Rand16() & mask_;
1008 r = prng.Rand8Extremes();
1010 assign_val(out, y * kOutputStride + x, r);
1011 assign_val(ref, y * kOutputStride + x, r);
1015 for (int axis = 0; axis < 2; axis++) {
1017 while (seed_val < 256) {
1018 for (int y = 0; y < 8; ++y) {
1019 for (int x = 0; x < 8; ++x) {
1020 #if CONFIG_VP9_HIGHBITDEPTH
1021 assign_val(in, y * kOutputStride + x - SUBPEL_TAPS / 2 + 1,
1022 ((seed_val >> (axis ? y : x)) & 1) * mask_);
1024 assign_val(in, y * kOutputStride + x - SUBPEL_TAPS / 2 + 1,
1025 ((seed_val >> (axis ? y : x)) & 1) * 255);
1027 if (axis) seed_val++;
1035 if (axis) seed_val += 8;
1037 for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
1038 const InterpKernel *filters =
1039 vp9_filter_kernels[static_cast<INTERP_FILTER>(filter_bank)];
1040 for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
1041 for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
1042 wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
1043 filters[filter_y], ref, kOutputStride,
1044 Width(), Height(), UUT_->use_highbd_);
1045 if (filter_x && filter_y)
1046 ASM_REGISTER_STATE_CHECK(
1047 UUT_->hv8_[0](in, kInputStride, out, kOutputStride, filters,
1048 filter_x, 16, filter_y, 16, Width(), Height()));
1050 ASM_REGISTER_STATE_CHECK(
1051 UUT_->v8_[0](in, kInputStride, out, kOutputStride, filters, 0,
1052 16, filter_y, 16, Width(), Height()));
1054 ASM_REGISTER_STATE_CHECK(
1055 UUT_->h8_[0](in, kInputStride, out, kOutputStride, filters,
1056 filter_x, 16, 0, 16, Width(), Height()));
1058 ASM_REGISTER_STATE_CHECK(
1059 UUT_->copy_[0](in, kInputStride, out, kOutputStride, nullptr,
1060 0, 0, 0, 0, Width(), Height()));
1062 for (int y = 0; y < Height(); ++y) {
1063 for (int x = 0; x < Width(); ++x)
1064 ASSERT_EQ(lookup(ref, y * kOutputStride + x),
1065 lookup(out, y * kOutputStride + x))
1066 << "mismatch at (" << x << "," << y << "), "
1067 << "filters (" << filter_bank << "," << filter_x << ","
1077 /* This test exercises that enough rows and columns are filtered with every
1078 possible initial fractional positions and scaling steps. */
1079 #if !CONFIG_VP9_HIGHBITDEPTH
1080 static const ConvolveFunc scaled_2d_c_funcs[2] = { vpx_scaled_2d_c,
1081 vpx_scaled_avg_2d_c };
1083 TEST_P(ConvolveTest, CheckScalingFiltering) {
1084 uint8_t *const in = input();
1085 uint8_t *const out = output();
1086 uint8_t ref[kOutputStride * kMaxDimension];
1088 ::libvpx_test::ACMRandom prng;
1089 for (int y = 0; y < Height(); ++y) {
1090 for (int x = 0; x < Width(); ++x) {
1091 const uint16_t r = prng.Rand8Extremes();
1092 assign_val(in, y * kInputStride + x, r);
1096 for (int i = 0; i < 2; ++i) {
1097 for (INTERP_FILTER filter_type = 0; filter_type < 4; ++filter_type) {
1098 const InterpKernel *const eighttap = vp9_filter_kernels[filter_type];
1099 for (int frac = 0; frac < 16; ++frac) {
1100 for (int step = 1; step <= 32; ++step) {
1101 /* Test the horizontal and vertical filters in combination. */
1102 scaled_2d_c_funcs[i](in, kInputStride, ref, kOutputStride, eighttap,
1103 frac, step, frac, step, Width(), Height());
1104 ASM_REGISTER_STATE_CHECK(
1105 UUT_->shv8_[i](in, kInputStride, out, kOutputStride, eighttap,
1106 frac, step, frac, step, Width(), Height()));
1110 for (int y = 0; y < Height(); ++y) {
1111 for (int x = 0; x < Width(); ++x) {
1112 ASSERT_EQ(lookup(ref, y * kOutputStride + x),
1113 lookup(out, y * kOutputStride + x))
1114 << "x == " << x << ", y == " << y << ", frac == " << frac
1115 << ", step == " << step;
1125 using std::make_tuple;
1127 #if CONFIG_VP9_HIGHBITDEPTH
1128 #define WRAP(func, bd) \
1129 void wrap_##func##_##bd( \
1130 const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \
1131 ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, \
1132 int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \
1133 vpx_highbd_##func(reinterpret_cast<const uint16_t *>(src), src_stride, \
1134 reinterpret_cast<uint16_t *>(dst), dst_stride, filter, \
1135 x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, bd); \
1138 #if HAVE_SSE2 && VPX_ARCH_X86_64
1139 WRAP(convolve_copy_sse2, 8)
1140 WRAP(convolve_avg_sse2, 8)
1141 WRAP(convolve_copy_sse2, 10)
1142 WRAP(convolve_avg_sse2, 10)
1143 WRAP(convolve_copy_sse2, 12)
1144 WRAP(convolve_avg_sse2, 12)
1145 WRAP(convolve8_horiz_sse2, 8)
1146 WRAP(convolve8_avg_horiz_sse2, 8)
1147 WRAP(convolve8_vert_sse2, 8)
1148 WRAP(convolve8_avg_vert_sse2, 8)
1149 WRAP(convolve8_sse2, 8)
1150 WRAP(convolve8_avg_sse2, 8)
1151 WRAP(convolve8_horiz_sse2, 10)
1152 WRAP(convolve8_avg_horiz_sse2, 10)
1153 WRAP(convolve8_vert_sse2, 10)
1154 WRAP(convolve8_avg_vert_sse2, 10)
1155 WRAP(convolve8_sse2, 10)
1156 WRAP(convolve8_avg_sse2, 10)
1157 WRAP(convolve8_horiz_sse2, 12)
1158 WRAP(convolve8_avg_horiz_sse2, 12)
1159 WRAP(convolve8_vert_sse2, 12)
1160 WRAP(convolve8_avg_vert_sse2, 12)
1161 WRAP(convolve8_sse2, 12)
1162 WRAP(convolve8_avg_sse2, 12)
1163 #endif // HAVE_SSE2 && VPX_ARCH_X86_64
1166 WRAP(convolve_copy_avx2, 8)
1167 WRAP(convolve_avg_avx2, 8)
1168 WRAP(convolve8_horiz_avx2, 8)
1169 WRAP(convolve8_avg_horiz_avx2, 8)
1170 WRAP(convolve8_vert_avx2, 8)
1171 WRAP(convolve8_avg_vert_avx2, 8)
1172 WRAP(convolve8_avx2, 8)
1173 WRAP(convolve8_avg_avx2, 8)
1175 WRAP(convolve_copy_avx2, 10)
1176 WRAP(convolve_avg_avx2, 10)
1177 WRAP(convolve8_avx2, 10)
1178 WRAP(convolve8_horiz_avx2, 10)
1179 WRAP(convolve8_vert_avx2, 10)
1180 WRAP(convolve8_avg_avx2, 10)
1181 WRAP(convolve8_avg_horiz_avx2, 10)
1182 WRAP(convolve8_avg_vert_avx2, 10)
1184 WRAP(convolve_copy_avx2, 12)
1185 WRAP(convolve_avg_avx2, 12)
1186 WRAP(convolve8_avx2, 12)
1187 WRAP(convolve8_horiz_avx2, 12)
1188 WRAP(convolve8_vert_avx2, 12)
1189 WRAP(convolve8_avg_avx2, 12)
1190 WRAP(convolve8_avg_horiz_avx2, 12)
1191 WRAP(convolve8_avg_vert_avx2, 12)
1195 WRAP(convolve_copy_neon, 8)
1196 WRAP(convolve_avg_neon, 8)
1197 WRAP(convolve_copy_neon, 10)
1198 WRAP(convolve_avg_neon, 10)
1199 WRAP(convolve_copy_neon, 12)
1200 WRAP(convolve_avg_neon, 12)
1201 WRAP(convolve8_horiz_neon, 8)
1202 WRAP(convolve8_avg_horiz_neon, 8)
1203 WRAP(convolve8_vert_neon, 8)
1204 WRAP(convolve8_avg_vert_neon, 8)
1205 WRAP(convolve8_neon, 8)
1206 WRAP(convolve8_avg_neon, 8)
1207 WRAP(convolve8_horiz_neon, 10)
1208 WRAP(convolve8_avg_horiz_neon, 10)
1209 WRAP(convolve8_vert_neon, 10)
1210 WRAP(convolve8_avg_vert_neon, 10)
1211 WRAP(convolve8_neon, 10)
1212 WRAP(convolve8_avg_neon, 10)
1213 WRAP(convolve8_horiz_neon, 12)
1214 WRAP(convolve8_avg_horiz_neon, 12)
1215 WRAP(convolve8_vert_neon, 12)
1216 WRAP(convolve8_avg_vert_neon, 12)
1217 WRAP(convolve8_neon, 12)
1218 WRAP(convolve8_avg_neon, 12)
1221 WRAP(convolve_copy_c, 8)
1222 WRAP(convolve_avg_c, 8)
1223 WRAP(convolve8_horiz_c, 8)
1224 WRAP(convolve8_avg_horiz_c, 8)
1225 WRAP(convolve8_vert_c, 8)
1226 WRAP(convolve8_avg_vert_c, 8)
1227 WRAP(convolve8_c, 8)
1228 WRAP(convolve8_avg_c, 8)
1229 WRAP(convolve_copy_c, 10)
1230 WRAP(convolve_avg_c, 10)
1231 WRAP(convolve8_horiz_c, 10)
1232 WRAP(convolve8_avg_horiz_c, 10)
1233 WRAP(convolve8_vert_c, 10)
1234 WRAP(convolve8_avg_vert_c, 10)
1235 WRAP(convolve8_c, 10)
1236 WRAP(convolve8_avg_c, 10)
1237 WRAP(convolve_copy_c, 12)
1238 WRAP(convolve_avg_c, 12)
1239 WRAP(convolve8_horiz_c, 12)
1240 WRAP(convolve8_avg_horiz_c, 12)
1241 WRAP(convolve8_vert_c, 12)
1242 WRAP(convolve8_avg_vert_c, 12)
1243 WRAP(convolve8_c, 12)
1244 WRAP(convolve8_avg_c, 12)
1247 const ConvolveFunctions convolve8_c(
1248 wrap_convolve_copy_c_8, wrap_convolve_avg_c_8, wrap_convolve8_horiz_c_8,
1249 wrap_convolve8_avg_horiz_c_8, wrap_convolve8_vert_c_8,
1250 wrap_convolve8_avg_vert_c_8, wrap_convolve8_c_8, wrap_convolve8_avg_c_8,
1251 wrap_convolve8_horiz_c_8, wrap_convolve8_avg_horiz_c_8,
1252 wrap_convolve8_vert_c_8, wrap_convolve8_avg_vert_c_8, wrap_convolve8_c_8,
1253 wrap_convolve8_avg_c_8, 8);
1254 const ConvolveFunctions convolve10_c(
1255 wrap_convolve_copy_c_10, wrap_convolve_avg_c_10, wrap_convolve8_horiz_c_10,
1256 wrap_convolve8_avg_horiz_c_10, wrap_convolve8_vert_c_10,
1257 wrap_convolve8_avg_vert_c_10, wrap_convolve8_c_10, wrap_convolve8_avg_c_10,
1258 wrap_convolve8_horiz_c_10, wrap_convolve8_avg_horiz_c_10,
1259 wrap_convolve8_vert_c_10, wrap_convolve8_avg_vert_c_10, wrap_convolve8_c_10,
1260 wrap_convolve8_avg_c_10, 10);
1261 const ConvolveFunctions convolve12_c(
1262 wrap_convolve_copy_c_12, wrap_convolve_avg_c_12, wrap_convolve8_horiz_c_12,
1263 wrap_convolve8_avg_horiz_c_12, wrap_convolve8_vert_c_12,
1264 wrap_convolve8_avg_vert_c_12, wrap_convolve8_c_12, wrap_convolve8_avg_c_12,
1265 wrap_convolve8_horiz_c_12, wrap_convolve8_avg_horiz_c_12,
1266 wrap_convolve8_vert_c_12, wrap_convolve8_avg_vert_c_12, wrap_convolve8_c_12,
1267 wrap_convolve8_avg_c_12, 12);
1268 const ConvolveParam kArrayConvolve_c[] = { ALL_SIZES(convolve8_c),
1269 ALL_SIZES(convolve10_c),
1270 ALL_SIZES(convolve12_c) };
1273 const ConvolveFunctions convolve8_c(
1274 vpx_convolve_copy_c, vpx_convolve_avg_c, vpx_convolve8_horiz_c,
1275 vpx_convolve8_avg_horiz_c, vpx_convolve8_vert_c, vpx_convolve8_avg_vert_c,
1276 vpx_convolve8_c, vpx_convolve8_avg_c, vpx_scaled_horiz_c,
1277 vpx_scaled_avg_horiz_c, vpx_scaled_vert_c, vpx_scaled_avg_vert_c,
1278 vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
1279 const ConvolveParam kArrayConvolve_c[] = { ALL_SIZES(convolve8_c) };
1281 INSTANTIATE_TEST_SUITE_P(C, ConvolveTest,
1282 ::testing::ValuesIn(kArrayConvolve_c));
1284 #if HAVE_SSE2 && VPX_ARCH_X86_64
1285 #if CONFIG_VP9_HIGHBITDEPTH
1286 const ConvolveFunctions convolve8_sse2(
1287 wrap_convolve_copy_sse2_8, wrap_convolve_avg_sse2_8,
1288 wrap_convolve8_horiz_sse2_8, wrap_convolve8_avg_horiz_sse2_8,
1289 wrap_convolve8_vert_sse2_8, wrap_convolve8_avg_vert_sse2_8,
1290 wrap_convolve8_sse2_8, wrap_convolve8_avg_sse2_8,
1291 wrap_convolve8_horiz_sse2_8, wrap_convolve8_avg_horiz_sse2_8,
1292 wrap_convolve8_vert_sse2_8, wrap_convolve8_avg_vert_sse2_8,
1293 wrap_convolve8_sse2_8, wrap_convolve8_avg_sse2_8, 8);
1294 const ConvolveFunctions convolve10_sse2(
1295 wrap_convolve_copy_sse2_10, wrap_convolve_avg_sse2_10,
1296 wrap_convolve8_horiz_sse2_10, wrap_convolve8_avg_horiz_sse2_10,
1297 wrap_convolve8_vert_sse2_10, wrap_convolve8_avg_vert_sse2_10,
1298 wrap_convolve8_sse2_10, wrap_convolve8_avg_sse2_10,
1299 wrap_convolve8_horiz_sse2_10, wrap_convolve8_avg_horiz_sse2_10,
1300 wrap_convolve8_vert_sse2_10, wrap_convolve8_avg_vert_sse2_10,
1301 wrap_convolve8_sse2_10, wrap_convolve8_avg_sse2_10, 10);
1302 const ConvolveFunctions convolve12_sse2(
1303 wrap_convolve_copy_sse2_12, wrap_convolve_avg_sse2_12,
1304 wrap_convolve8_horiz_sse2_12, wrap_convolve8_avg_horiz_sse2_12,
1305 wrap_convolve8_vert_sse2_12, wrap_convolve8_avg_vert_sse2_12,
1306 wrap_convolve8_sse2_12, wrap_convolve8_avg_sse2_12,
1307 wrap_convolve8_horiz_sse2_12, wrap_convolve8_avg_horiz_sse2_12,
1308 wrap_convolve8_vert_sse2_12, wrap_convolve8_avg_vert_sse2_12,
1309 wrap_convolve8_sse2_12, wrap_convolve8_avg_sse2_12, 12);
1310 const ConvolveParam kArrayConvolve_sse2[] = { ALL_SIZES(convolve8_sse2),
1311 ALL_SIZES(convolve10_sse2),
1312 ALL_SIZES(convolve12_sse2) };
1314 const ConvolveFunctions convolve8_sse2(
1315 vpx_convolve_copy_sse2, vpx_convolve_avg_sse2, vpx_convolve8_horiz_sse2,
1316 vpx_convolve8_avg_horiz_sse2, vpx_convolve8_vert_sse2,
1317 vpx_convolve8_avg_vert_sse2, vpx_convolve8_sse2, vpx_convolve8_avg_sse2,
1318 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1319 vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
1321 const ConvolveParam kArrayConvolve_sse2[] = { ALL_SIZES(convolve8_sse2) };
1322 #endif // CONFIG_VP9_HIGHBITDEPTH
1323 INSTANTIATE_TEST_SUITE_P(SSE2, ConvolveTest,
1324 ::testing::ValuesIn(kArrayConvolve_sse2));
1328 const ConvolveFunctions convolve8_ssse3(
1329 vpx_convolve_copy_c, vpx_convolve_avg_c, vpx_convolve8_horiz_ssse3,
1330 vpx_convolve8_avg_horiz_ssse3, vpx_convolve8_vert_ssse3,
1331 vpx_convolve8_avg_vert_ssse3, vpx_convolve8_ssse3, vpx_convolve8_avg_ssse3,
1332 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1333 vpx_scaled_avg_vert_c, vpx_scaled_2d_ssse3, vpx_scaled_avg_2d_c, 0);
1335 const ConvolveParam kArrayConvolve8_ssse3[] = { ALL_SIZES(convolve8_ssse3) };
1336 INSTANTIATE_TEST_SUITE_P(SSSE3, ConvolveTest,
1337 ::testing::ValuesIn(kArrayConvolve8_ssse3));
1341 #if CONFIG_VP9_HIGHBITDEPTH
1342 const ConvolveFunctions convolve8_avx2(
1343 wrap_convolve_copy_avx2_8, wrap_convolve_avg_avx2_8,
1344 wrap_convolve8_horiz_avx2_8, wrap_convolve8_avg_horiz_avx2_8,
1345 wrap_convolve8_vert_avx2_8, wrap_convolve8_avg_vert_avx2_8,
1346 wrap_convolve8_avx2_8, wrap_convolve8_avg_avx2_8, wrap_convolve8_horiz_c_8,
1347 wrap_convolve8_avg_horiz_c_8, wrap_convolve8_vert_c_8,
1348 wrap_convolve8_avg_vert_c_8, wrap_convolve8_c_8, wrap_convolve8_avg_c_8, 8);
1349 const ConvolveFunctions convolve10_avx2(
1350 wrap_convolve_copy_avx2_10, wrap_convolve_avg_avx2_10,
1351 wrap_convolve8_horiz_avx2_10, wrap_convolve8_avg_horiz_avx2_10,
1352 wrap_convolve8_vert_avx2_10, wrap_convolve8_avg_vert_avx2_10,
1353 wrap_convolve8_avx2_10, wrap_convolve8_avg_avx2_10,
1354 wrap_convolve8_horiz_c_10, wrap_convolve8_avg_horiz_c_10,
1355 wrap_convolve8_vert_c_10, wrap_convolve8_avg_vert_c_10, wrap_convolve8_c_10,
1356 wrap_convolve8_avg_c_10, 10);
1357 const ConvolveFunctions convolve12_avx2(
1358 wrap_convolve_copy_avx2_12, wrap_convolve_avg_avx2_12,
1359 wrap_convolve8_horiz_avx2_12, wrap_convolve8_avg_horiz_avx2_12,
1360 wrap_convolve8_vert_avx2_12, wrap_convolve8_avg_vert_avx2_12,
1361 wrap_convolve8_avx2_12, wrap_convolve8_avg_avx2_12,
1362 wrap_convolve8_horiz_c_12, wrap_convolve8_avg_horiz_c_12,
1363 wrap_convolve8_vert_c_12, wrap_convolve8_avg_vert_c_12, wrap_convolve8_c_12,
1364 wrap_convolve8_avg_c_12, 12);
1365 const ConvolveParam kArrayConvolve8_avx2[] = { ALL_SIZES(convolve8_avx2),
1366 ALL_SIZES(convolve10_avx2),
1367 ALL_SIZES(convolve12_avx2) };
1368 INSTANTIATE_TEST_SUITE_P(AVX2, ConvolveTest,
1369 ::testing::ValuesIn(kArrayConvolve8_avx2));
1370 #else // !CONFIG_VP9_HIGHBITDEPTH
1371 const ConvolveFunctions convolve8_avx2(
1372 vpx_convolve_copy_c, vpx_convolve_avg_c, vpx_convolve8_horiz_avx2,
1373 vpx_convolve8_avg_horiz_avx2, vpx_convolve8_vert_avx2,
1374 vpx_convolve8_avg_vert_avx2, vpx_convolve8_avx2, vpx_convolve8_avg_avx2,
1375 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1376 vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
1377 const ConvolveParam kArrayConvolve8_avx2[] = { ALL_SIZES(convolve8_avx2) };
1378 INSTANTIATE_TEST_SUITE_P(AVX2, ConvolveTest,
1379 ::testing::ValuesIn(kArrayConvolve8_avx2));
1380 #endif // CONFIG_VP9_HIGHBITDEPTH
1384 #if CONFIG_VP9_HIGHBITDEPTH
1385 const ConvolveFunctions convolve8_neon(
1386 wrap_convolve_copy_neon_8, wrap_convolve_avg_neon_8,
1387 wrap_convolve8_horiz_neon_8, wrap_convolve8_avg_horiz_neon_8,
1388 wrap_convolve8_vert_neon_8, wrap_convolve8_avg_vert_neon_8,
1389 wrap_convolve8_neon_8, wrap_convolve8_avg_neon_8,
1390 wrap_convolve8_horiz_neon_8, wrap_convolve8_avg_horiz_neon_8,
1391 wrap_convolve8_vert_neon_8, wrap_convolve8_avg_vert_neon_8,
1392 wrap_convolve8_neon_8, wrap_convolve8_avg_neon_8, 8);
1393 const ConvolveFunctions convolve10_neon(
1394 wrap_convolve_copy_neon_10, wrap_convolve_avg_neon_10,
1395 wrap_convolve8_horiz_neon_10, wrap_convolve8_avg_horiz_neon_10,
1396 wrap_convolve8_vert_neon_10, wrap_convolve8_avg_vert_neon_10,
1397 wrap_convolve8_neon_10, wrap_convolve8_avg_neon_10,
1398 wrap_convolve8_horiz_neon_10, wrap_convolve8_avg_horiz_neon_10,
1399 wrap_convolve8_vert_neon_10, wrap_convolve8_avg_vert_neon_10,
1400 wrap_convolve8_neon_10, wrap_convolve8_avg_neon_10, 10);
1401 const ConvolveFunctions convolve12_neon(
1402 wrap_convolve_copy_neon_12, wrap_convolve_avg_neon_12,
1403 wrap_convolve8_horiz_neon_12, wrap_convolve8_avg_horiz_neon_12,
1404 wrap_convolve8_vert_neon_12, wrap_convolve8_avg_vert_neon_12,
1405 wrap_convolve8_neon_12, wrap_convolve8_avg_neon_12,
1406 wrap_convolve8_horiz_neon_12, wrap_convolve8_avg_horiz_neon_12,
1407 wrap_convolve8_vert_neon_12, wrap_convolve8_avg_vert_neon_12,
1408 wrap_convolve8_neon_12, wrap_convolve8_avg_neon_12, 12);
1409 const ConvolveParam kArrayConvolve_neon[] = { ALL_SIZES(convolve8_neon),
1410 ALL_SIZES(convolve10_neon),
1411 ALL_SIZES(convolve12_neon) };
1413 const ConvolveFunctions convolve8_neon(
1414 vpx_convolve_copy_neon, vpx_convolve_avg_neon, vpx_convolve8_horiz_neon,
1415 vpx_convolve8_avg_horiz_neon, vpx_convolve8_vert_neon,
1416 vpx_convolve8_avg_vert_neon, vpx_convolve8_neon, vpx_convolve8_avg_neon,
1417 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1418 vpx_scaled_avg_vert_c, vpx_scaled_2d_neon, vpx_scaled_avg_2d_c, 0);
1420 const ConvolveParam kArrayConvolve_neon[] = { ALL_SIZES(convolve8_neon) };
1421 #endif // CONFIG_VP9_HIGHBITDEPTH
1422 INSTANTIATE_TEST_SUITE_P(NEON, ConvolveTest,
1423 ::testing::ValuesIn(kArrayConvolve_neon));
1427 const ConvolveFunctions convolve8_dspr2(
1428 vpx_convolve_copy_dspr2, vpx_convolve_avg_dspr2, vpx_convolve8_horiz_dspr2,
1429 vpx_convolve8_avg_horiz_dspr2, vpx_convolve8_vert_dspr2,
1430 vpx_convolve8_avg_vert_dspr2, vpx_convolve8_dspr2, vpx_convolve8_avg_dspr2,
1431 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1432 vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
1434 const ConvolveParam kArrayConvolve8_dspr2[] = { ALL_SIZES(convolve8_dspr2) };
1435 INSTANTIATE_TEST_SUITE_P(DSPR2, ConvolveTest,
1436 ::testing::ValuesIn(kArrayConvolve8_dspr2));
1437 #endif // HAVE_DSPR2
1440 const ConvolveFunctions convolve8_msa(
1441 vpx_convolve_copy_msa, vpx_convolve_avg_msa, vpx_convolve8_horiz_msa,
1442 vpx_convolve8_avg_horiz_msa, vpx_convolve8_vert_msa,
1443 vpx_convolve8_avg_vert_msa, vpx_convolve8_msa, vpx_convolve8_avg_msa,
1444 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1445 vpx_scaled_avg_vert_c, vpx_scaled_2d_msa, vpx_scaled_avg_2d_c, 0);
1447 const ConvolveParam kArrayConvolve8_msa[] = { ALL_SIZES(convolve8_msa) };
1448 INSTANTIATE_TEST_SUITE_P(MSA, ConvolveTest,
1449 ::testing::ValuesIn(kArrayConvolve8_msa));
1453 const ConvolveFunctions convolve8_lsx(
1454 vpx_convolve_copy_lsx, vpx_convolve_avg_lsx, vpx_convolve8_horiz_lsx,
1455 vpx_convolve8_avg_horiz_lsx, vpx_convolve8_vert_lsx,
1456 vpx_convolve8_avg_vert_lsx, vpx_convolve8_lsx, vpx_convolve8_avg_lsx,
1457 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1458 vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
1460 const ConvolveParam kArrayConvolve8_lsx[] = { ALL_SIZES(convolve8_lsx) };
1461 INSTANTIATE_TEST_SUITE_P(LSX, ConvolveTest,
1462 ::testing::ValuesIn(kArrayConvolve8_lsx));
1466 const ConvolveFunctions convolve8_vsx(
1467 vpx_convolve_copy_vsx, vpx_convolve_avg_vsx, vpx_convolve8_horiz_vsx,
1468 vpx_convolve8_avg_horiz_vsx, vpx_convolve8_vert_vsx,
1469 vpx_convolve8_avg_vert_vsx, vpx_convolve8_vsx, vpx_convolve8_avg_vsx,
1470 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1471 vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
1472 const ConvolveParam kArrayConvolve_vsx[] = { ALL_SIZES(convolve8_vsx) };
1473 INSTANTIATE_TEST_SUITE_P(VSX, ConvolveTest,
1474 ::testing::ValuesIn(kArrayConvolve_vsx));
1478 const ConvolveFunctions convolve8_mmi(
1479 vpx_convolve_copy_c, vpx_convolve_avg_mmi, vpx_convolve8_horiz_mmi,
1480 vpx_convolve8_avg_horiz_mmi, vpx_convolve8_vert_mmi,
1481 vpx_convolve8_avg_vert_mmi, vpx_convolve8_mmi, vpx_convolve8_avg_mmi,
1482 vpx_scaled_horiz_c, vpx_scaled_avg_horiz_c, vpx_scaled_vert_c,
1483 vpx_scaled_avg_vert_c, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
1484 const ConvolveParam kArrayConvolve_mmi[] = { ALL_SIZES(convolve8_mmi) };
1485 INSTANTIATE_TEST_SUITE_P(MMI, ConvolveTest,
1486 ::testing::ValuesIn(kArrayConvolve_mmi));