1 /*
2 * Copyright (c) 2017 The WebM project authors. All Rights Reserved.
3 *
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.
9 */
10
11 #include <math.h>
12 #include <stdlib.h>
13 #include <string.h>
14 #include <tuple>
15
16 #include "third_party/googletest/src/include/gtest/gtest.h"
17
18 #include "./vp9_rtcd.h"
19 #include "./vpx_dsp_rtcd.h"
20 #include "test/acm_random.h"
21 #include "test/buffer.h"
22 #include "test/clear_system_state.h"
23 #include "test/register_state_check.h"
24 #include "test/util.h"
25 #include "vp9/common/vp9_entropy.h"
26 #include "vpx/vpx_codec.h"
27 #include "vpx/vpx_integer.h"
28 #include "vpx_ports/mem.h"
29
30 using libvpx_test::ACMRandom;
31 using libvpx_test::Buffer;
32 using std::make_tuple;
33 using std::tuple;
34
35 namespace {
36 typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
37 typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
38 typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
39 int tx_type);
40 typedef void (*FhtFuncRef)(const Buffer<int16_t> &in, Buffer<tran_low_t> *out,
41 int size, int tx_type);
42 typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
43 int tx_type);
44 typedef void (*IhtWithBdFunc)(const tran_low_t *in, uint8_t *out, int stride,
45 int tx_type, int bd);
46
47 template <FdctFunc fn>
fdct_wrapper(const int16_t * in,tran_low_t * out,int stride,int tx_type)48 void fdct_wrapper(const int16_t *in, tran_low_t *out, int stride, int tx_type) {
49 (void)tx_type;
50 fn(in, out, stride);
51 }
52
53 template <IdctFunc fn>
idct_wrapper(const tran_low_t * in,uint8_t * out,int stride,int tx_type,int bd)54 void idct_wrapper(const tran_low_t *in, uint8_t *out, int stride, int tx_type,
55 int bd) {
56 (void)tx_type;
57 (void)bd;
58 fn(in, out, stride);
59 }
60
61 template <IhtFunc fn>
iht_wrapper(const tran_low_t * in,uint8_t * out,int stride,int tx_type,int bd)62 void iht_wrapper(const tran_low_t *in, uint8_t *out, int stride, int tx_type,
63 int bd) {
64 (void)bd;
65 fn(in, out, stride, tx_type);
66 }
67
68 #if CONFIG_VP9_HIGHBITDEPTH
69 typedef void (*HighbdIdctFunc)(const tran_low_t *in, uint16_t *out, int stride,
70 int bd);
71
72 typedef void (*HighbdIhtFunc)(const tran_low_t *in, uint16_t *out, int stride,
73 int tx_type, int bd);
74
75 template <HighbdIdctFunc fn>
highbd_idct_wrapper(const tran_low_t * in,uint8_t * out,int stride,int tx_type,int bd)76 void highbd_idct_wrapper(const tran_low_t *in, uint8_t *out, int stride,
77 int tx_type, int bd) {
78 (void)tx_type;
79 fn(in, CAST_TO_SHORTPTR(out), stride, bd);
80 }
81
82 template <HighbdIhtFunc fn>
highbd_iht_wrapper(const tran_low_t * in,uint8_t * out,int stride,int tx_type,int bd)83 void highbd_iht_wrapper(const tran_low_t *in, uint8_t *out, int stride,
84 int tx_type, int bd) {
85 fn(in, CAST_TO_SHORTPTR(out), stride, tx_type, bd);
86 }
87 #endif // CONFIG_VP9_HIGHBITDEPTH
88
89 struct FuncInfo {
90 FhtFunc ft_func;
91 IhtWithBdFunc it_func;
92 int size;
93 int pixel_size;
94 };
95
96 /* forward transform, inverse transform, size, transform type, bit depth */
97 typedef tuple<int, const FuncInfo *, int, vpx_bit_depth_t> DctParam;
98
fdct_ref(const Buffer<int16_t> & in,Buffer<tran_low_t> * out,int size,int)99 void fdct_ref(const Buffer<int16_t> &in, Buffer<tran_low_t> *out, int size,
100 int /*tx_type*/) {
101 const int16_t *i = in.TopLeftPixel();
102 const int i_stride = in.stride();
103 tran_low_t *o = out->TopLeftPixel();
104 if (size == 4) {
105 vpx_fdct4x4_c(i, o, i_stride);
106 } else if (size == 8) {
107 vpx_fdct8x8_c(i, o, i_stride);
108 } else if (size == 16) {
109 vpx_fdct16x16_c(i, o, i_stride);
110 } else if (size == 32) {
111 vpx_fdct32x32_c(i, o, i_stride);
112 }
113 }
114
fht_ref(const Buffer<int16_t> & in,Buffer<tran_low_t> * out,int size,int tx_type)115 void fht_ref(const Buffer<int16_t> &in, Buffer<tran_low_t> *out, int size,
116 int tx_type) {
117 const int16_t *i = in.TopLeftPixel();
118 const int i_stride = in.stride();
119 tran_low_t *o = out->TopLeftPixel();
120 if (size == 4) {
121 vp9_fht4x4_c(i, o, i_stride, tx_type);
122 } else if (size == 8) {
123 vp9_fht8x8_c(i, o, i_stride, tx_type);
124 } else if (size == 16) {
125 vp9_fht16x16_c(i, o, i_stride, tx_type);
126 }
127 }
128
fwht_ref(const Buffer<int16_t> & in,Buffer<tran_low_t> * out,int size,int)129 void fwht_ref(const Buffer<int16_t> &in, Buffer<tran_low_t> *out, int size,
130 int /*tx_type*/) {
131 ASSERT_EQ(size, 4);
132 vp9_fwht4x4_c(in.TopLeftPixel(), out->TopLeftPixel(), in.stride());
133 }
134
135 class TransTestBase : public ::testing::TestWithParam<DctParam> {
136 public:
SetUp()137 virtual void SetUp() {
138 rnd_.Reset(ACMRandom::DeterministicSeed());
139 const int idx = GET_PARAM(0);
140 const FuncInfo *func_info = &(GET_PARAM(1)[idx]);
141 tx_type_ = GET_PARAM(2);
142 bit_depth_ = GET_PARAM(3);
143 fwd_txfm_ = func_info->ft_func;
144 inv_txfm_ = func_info->it_func;
145 size_ = func_info->size;
146 pixel_size_ = func_info->pixel_size;
147 max_pixel_value_ = (1 << bit_depth_) - 1;
148
149 // Randomize stride_ to a value less than or equal to 1024
150 stride_ = rnd_(1024) + 1;
151 if (stride_ < size_) {
152 stride_ = size_;
153 }
154 // Align stride_ to 16 if it's bigger than 16.
155 if (stride_ > 16) {
156 stride_ &= ~15;
157 }
158
159 block_size_ = size_ * stride_;
160
161 src_ = reinterpret_cast<uint8_t *>(
162 vpx_memalign(16, pixel_size_ * block_size_));
163 ASSERT_TRUE(src_ != NULL);
164 dst_ = reinterpret_cast<uint8_t *>(
165 vpx_memalign(16, pixel_size_ * block_size_));
166 ASSERT_TRUE(dst_ != NULL);
167 }
168
TearDown()169 virtual void TearDown() {
170 vpx_free(src_);
171 src_ = NULL;
172 vpx_free(dst_);
173 dst_ = NULL;
174 libvpx_test::ClearSystemState();
175 }
176
InitMem()177 void InitMem() {
178 if (pixel_size_ == 1 && bit_depth_ > VPX_BITS_8) return;
179 if (pixel_size_ == 1) {
180 for (int j = 0; j < block_size_; ++j) {
181 src_[j] = rnd_.Rand16() & max_pixel_value_;
182 }
183 for (int j = 0; j < block_size_; ++j) {
184 dst_[j] = rnd_.Rand16() & max_pixel_value_;
185 }
186 } else {
187 ASSERT_EQ(pixel_size_, 2);
188 uint16_t *const src = reinterpret_cast<uint16_t *>(src_);
189 uint16_t *const dst = reinterpret_cast<uint16_t *>(dst_);
190 for (int j = 0; j < block_size_; ++j) {
191 src[j] = rnd_.Rand16() & max_pixel_value_;
192 }
193 for (int j = 0; j < block_size_; ++j) {
194 dst[j] = rnd_.Rand16() & max_pixel_value_;
195 }
196 }
197 }
198
RunFwdTxfm(const Buffer<int16_t> & in,Buffer<tran_low_t> * out)199 void RunFwdTxfm(const Buffer<int16_t> &in, Buffer<tran_low_t> *out) {
200 fwd_txfm_(in.TopLeftPixel(), out->TopLeftPixel(), in.stride(), tx_type_);
201 }
202
RunInvTxfm(const Buffer<tran_low_t> & in,uint8_t * out)203 void RunInvTxfm(const Buffer<tran_low_t> &in, uint8_t *out) {
204 inv_txfm_(in.TopLeftPixel(), out, stride_, tx_type_, bit_depth_);
205 }
206
207 protected:
RunAccuracyCheck(int limit)208 void RunAccuracyCheck(int limit) {
209 if (pixel_size_ == 1 && bit_depth_ > VPX_BITS_8) return;
210 ACMRandom rnd(ACMRandom::DeterministicSeed());
211 Buffer<int16_t> test_input_block =
212 Buffer<int16_t>(size_, size_, 8, size_ == 4 ? 0 : 16);
213 ASSERT_TRUE(test_input_block.Init());
214 ASSERT_TRUE(test_input_block.TopLeftPixel() != NULL);
215 Buffer<tran_low_t> test_temp_block =
216 Buffer<tran_low_t>(size_, size_, 0, 16);
217 ASSERT_TRUE(test_temp_block.Init());
218 uint32_t max_error = 0;
219 int64_t total_error = 0;
220 const int count_test_block = 10000;
221 for (int i = 0; i < count_test_block; ++i) {
222 InitMem();
223 for (int h = 0; h < size_; ++h) {
224 for (int w = 0; w < size_; ++w) {
225 if (pixel_size_ == 1) {
226 test_input_block.TopLeftPixel()[h * test_input_block.stride() + w] =
227 src_[h * stride_ + w] - dst_[h * stride_ + w];
228 } else {
229 ASSERT_EQ(pixel_size_, 2);
230 const uint16_t *const src = reinterpret_cast<uint16_t *>(src_);
231 const uint16_t *const dst = reinterpret_cast<uint16_t *>(dst_);
232 test_input_block.TopLeftPixel()[h * test_input_block.stride() + w] =
233 src[h * stride_ + w] - dst[h * stride_ + w];
234 }
235 }
236 }
237
238 ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block, &test_temp_block));
239 ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst_));
240
241 for (int h = 0; h < size_; ++h) {
242 for (int w = 0; w < size_; ++w) {
243 int diff;
244 if (pixel_size_ == 1) {
245 diff = dst_[h * stride_ + w] - src_[h * stride_ + w];
246 } else {
247 ASSERT_EQ(pixel_size_, 2);
248 const uint16_t *const src = reinterpret_cast<uint16_t *>(src_);
249 const uint16_t *const dst = reinterpret_cast<uint16_t *>(dst_);
250 diff = dst[h * stride_ + w] - src[h * stride_ + w];
251 }
252 const uint32_t error = diff * diff;
253 if (max_error < error) max_error = error;
254 total_error += error;
255 }
256 }
257 }
258
259 EXPECT_GE(static_cast<uint32_t>(limit), max_error)
260 << "Error: " << size_ << "x" << size_
261 << " transform/inverse transform has an individual round trip error > "
262 << limit;
263
264 EXPECT_GE(count_test_block * limit, total_error)
265 << "Error: " << size_ << "x" << size_
266 << " transform/inverse transform has average round trip error > "
267 << limit << " per block";
268 }
269
RunCoeffCheck()270 void RunCoeffCheck() {
271 if (pixel_size_ == 1 && bit_depth_ > VPX_BITS_8) return;
272 ACMRandom rnd(ACMRandom::DeterministicSeed());
273 const int count_test_block = 5000;
274 Buffer<int16_t> input_block =
275 Buffer<int16_t>(size_, size_, 8, size_ == 4 ? 0 : 16);
276 ASSERT_TRUE(input_block.Init());
277 Buffer<tran_low_t> output_ref_block = Buffer<tran_low_t>(size_, size_, 0);
278 ASSERT_TRUE(output_ref_block.Init());
279 Buffer<tran_low_t> output_block = Buffer<tran_low_t>(size_, size_, 0, 16);
280 ASSERT_TRUE(output_block.Init());
281
282 for (int i = 0; i < count_test_block; ++i) {
283 // Initialize a test block with input range [-max_pixel_value_,
284 // max_pixel_value_].
285 input_block.Set(&rnd, -max_pixel_value_, max_pixel_value_);
286
287 fwd_txfm_ref(input_block, &output_ref_block, size_, tx_type_);
288 ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, &output_block));
289
290 // The minimum quant value is 4.
291 EXPECT_TRUE(output_block.CheckValues(output_ref_block));
292 if (::testing::Test::HasFailure()) {
293 printf("Size: %d Transform type: %d\n", size_, tx_type_);
294 output_block.PrintDifference(output_ref_block);
295 return;
296 }
297 }
298 }
299
RunMemCheck()300 void RunMemCheck() {
301 if (pixel_size_ == 1 && bit_depth_ > VPX_BITS_8) return;
302 ACMRandom rnd(ACMRandom::DeterministicSeed());
303 const int count_test_block = 5000;
304 Buffer<int16_t> input_extreme_block =
305 Buffer<int16_t>(size_, size_, 8, size_ == 4 ? 0 : 16);
306 ASSERT_TRUE(input_extreme_block.Init());
307 Buffer<tran_low_t> output_ref_block = Buffer<tran_low_t>(size_, size_, 0);
308 ASSERT_TRUE(output_ref_block.Init());
309 Buffer<tran_low_t> output_block = Buffer<tran_low_t>(size_, size_, 0, 16);
310 ASSERT_TRUE(output_block.Init());
311
312 for (int i = 0; i < count_test_block; ++i) {
313 // Initialize a test block with -max_pixel_value_ or max_pixel_value_.
314 if (i == 0) {
315 input_extreme_block.Set(max_pixel_value_);
316 } else if (i == 1) {
317 input_extreme_block.Set(-max_pixel_value_);
318 } else {
319 ASSERT_TRUE(input_extreme_block.TopLeftPixel() != NULL);
320 for (int h = 0; h < size_; ++h) {
321 for (int w = 0; w < size_; ++w) {
322 input_extreme_block
323 .TopLeftPixel()[h * input_extreme_block.stride() + w] =
324 rnd.Rand8() % 2 ? max_pixel_value_ : -max_pixel_value_;
325 }
326 }
327 }
328
329 fwd_txfm_ref(input_extreme_block, &output_ref_block, size_, tx_type_);
330 ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block, &output_block));
331
332 // The minimum quant value is 4.
333 EXPECT_TRUE(output_block.CheckValues(output_ref_block));
334 ASSERT_TRUE(output_block.TopLeftPixel() != NULL);
335 for (int h = 0; h < size_; ++h) {
336 for (int w = 0; w < size_; ++w) {
337 EXPECT_GE(
338 4 * DCT_MAX_VALUE << (bit_depth_ - 8),
339 abs(output_block.TopLeftPixel()[h * output_block.stride() + w]))
340 << "Error: " << size_ << "x" << size_
341 << " transform has coefficient larger than 4*DCT_MAX_VALUE"
342 << " at " << w << "," << h;
343 if (::testing::Test::HasFailure()) {
344 printf("Size: %d Transform type: %d\n", size_, tx_type_);
345 output_block.DumpBuffer();
346 return;
347 }
348 }
349 }
350 }
351 }
352
RunInvAccuracyCheck(int limit)353 void RunInvAccuracyCheck(int limit) {
354 if (pixel_size_ == 1 && bit_depth_ > VPX_BITS_8) return;
355 ACMRandom rnd(ACMRandom::DeterministicSeed());
356 const int count_test_block = 1000;
357 Buffer<int16_t> in = Buffer<int16_t>(size_, size_, 4);
358 ASSERT_TRUE(in.Init());
359 Buffer<tran_low_t> coeff = Buffer<tran_low_t>(size_, size_, 0, 16);
360 ASSERT_TRUE(coeff.Init());
361 Buffer<uint8_t> dst = Buffer<uint8_t>(size_, size_, 0, 16);
362 ASSERT_TRUE(dst.Init());
363 Buffer<uint8_t> src = Buffer<uint8_t>(size_, size_, 0);
364 ASSERT_TRUE(src.Init());
365 Buffer<uint16_t> dst16 = Buffer<uint16_t>(size_, size_, 0, 16);
366 ASSERT_TRUE(dst16.Init());
367 Buffer<uint16_t> src16 = Buffer<uint16_t>(size_, size_, 0);
368 ASSERT_TRUE(src16.Init());
369
370 for (int i = 0; i < count_test_block; ++i) {
371 InitMem();
372 ASSERT_TRUE(in.TopLeftPixel() != NULL);
373 // Initialize a test block with input range [-max_pixel_value_,
374 // max_pixel_value_].
375 for (int h = 0; h < size_; ++h) {
376 for (int w = 0; w < size_; ++w) {
377 if (pixel_size_ == 1) {
378 in.TopLeftPixel()[h * in.stride() + w] =
379 src_[h * stride_ + w] - dst_[h * stride_ + w];
380 } else {
381 ASSERT_EQ(pixel_size_, 2);
382 const uint16_t *const src = reinterpret_cast<uint16_t *>(src_);
383 const uint16_t *const dst = reinterpret_cast<uint16_t *>(dst_);
384 in.TopLeftPixel()[h * in.stride() + w] =
385 src[h * stride_ + w] - dst[h * stride_ + w];
386 }
387 }
388 }
389
390 fwd_txfm_ref(in, &coeff, size_, tx_type_);
391
392 ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst_));
393
394 for (int h = 0; h < size_; ++h) {
395 for (int w = 0; w < size_; ++w) {
396 int diff;
397 if (pixel_size_ == 1) {
398 diff = dst_[h * stride_ + w] - src_[h * stride_ + w];
399 } else {
400 ASSERT_EQ(pixel_size_, 2);
401 const uint16_t *const src = reinterpret_cast<uint16_t *>(src_);
402 const uint16_t *const dst = reinterpret_cast<uint16_t *>(dst_);
403 diff = dst[h * stride_ + w] - src[h * stride_ + w];
404 }
405 const uint32_t error = diff * diff;
406 EXPECT_GE(static_cast<uint32_t>(limit), error)
407 << "Error: " << size_ << "x" << size_
408 << " inverse transform has error " << error << " at " << w << ","
409 << h;
410 if (::testing::Test::HasFailure()) {
411 printf("Size: %d Transform type: %d\n", size_, tx_type_);
412 return;
413 }
414 }
415 }
416 }
417 }
418
419 FhtFunc fwd_txfm_;
420 FhtFuncRef fwd_txfm_ref;
421 IhtWithBdFunc inv_txfm_;
422 ACMRandom rnd_;
423 uint8_t *src_;
424 uint8_t *dst_;
425 vpx_bit_depth_t bit_depth_;
426 int tx_type_;
427 int max_pixel_value_;
428 int size_;
429 int stride_;
430 int pixel_size_;
431 int block_size_;
432 };
433
434 /* -------------------------------------------------------------------------- */
435
436 class TransDCT : public TransTestBase {
437 public:
TransDCT()438 TransDCT() { fwd_txfm_ref = fdct_ref; }
439 };
440
TEST_P(TransDCT,AccuracyCheck)441 TEST_P(TransDCT, AccuracyCheck) {
442 int t = 1;
443 if (size_ == 16 && bit_depth_ > 10 && pixel_size_ == 2) {
444 t = 2;
445 } else if (size_ == 32 && bit_depth_ > 10 && pixel_size_ == 2) {
446 t = 7;
447 }
448 RunAccuracyCheck(t);
449 }
450
TEST_P(TransDCT,CoeffCheck)451 TEST_P(TransDCT, CoeffCheck) { RunCoeffCheck(); }
452
TEST_P(TransDCT,MemCheck)453 TEST_P(TransDCT, MemCheck) { RunMemCheck(); }
454
TEST_P(TransDCT,InvAccuracyCheck)455 TEST_P(TransDCT, InvAccuracyCheck) { RunInvAccuracyCheck(1); }
456
457 static const FuncInfo dct_c_func_info[] = {
458 #if CONFIG_VP9_HIGHBITDEPTH
459 { &fdct_wrapper<vpx_highbd_fdct4x4_c>,
460 &highbd_idct_wrapper<vpx_highbd_idct4x4_16_add_c>, 4, 2 },
461 { &fdct_wrapper<vpx_highbd_fdct8x8_c>,
462 &highbd_idct_wrapper<vpx_highbd_idct8x8_64_add_c>, 8, 2 },
463 { &fdct_wrapper<vpx_highbd_fdct16x16_c>,
464 &highbd_idct_wrapper<vpx_highbd_idct16x16_256_add_c>, 16, 2 },
465 { &fdct_wrapper<vpx_highbd_fdct32x32_c>,
466 &highbd_idct_wrapper<vpx_highbd_idct32x32_1024_add_c>, 32, 2 },
467 #endif
468 { &fdct_wrapper<vpx_fdct4x4_c>, &idct_wrapper<vpx_idct4x4_16_add_c>, 4, 1 },
469 { &fdct_wrapper<vpx_fdct8x8_c>, &idct_wrapper<vpx_idct8x8_64_add_c>, 8, 1 },
470 { &fdct_wrapper<vpx_fdct16x16_c>, &idct_wrapper<vpx_idct16x16_256_add_c>, 16,
471 1 },
472 { &fdct_wrapper<vpx_fdct32x32_c>, &idct_wrapper<vpx_idct32x32_1024_add_c>, 32,
473 1 }
474 };
475
476 INSTANTIATE_TEST_CASE_P(
477 C, TransDCT,
478 ::testing::Combine(
479 ::testing::Range(0, static_cast<int>(sizeof(dct_c_func_info) /
480 sizeof(dct_c_func_info[0]))),
481 ::testing::Values(dct_c_func_info), ::testing::Values(0),
482 ::testing::Values(VPX_BITS_8, VPX_BITS_10, VPX_BITS_12)));
483
484 #if !CONFIG_EMULATE_HARDWARE
485
486 #if HAVE_SSE2
487 static const FuncInfo dct_sse2_func_info[] = {
488 #if CONFIG_VP9_HIGHBITDEPTH
489 { &fdct_wrapper<vpx_highbd_fdct4x4_sse2>,
490 &highbd_idct_wrapper<vpx_highbd_idct4x4_16_add_sse2>, 4, 2 },
491 { &fdct_wrapper<vpx_highbd_fdct8x8_sse2>,
492 &highbd_idct_wrapper<vpx_highbd_idct8x8_64_add_sse2>, 8, 2 },
493 { &fdct_wrapper<vpx_highbd_fdct16x16_sse2>,
494 &highbd_idct_wrapper<vpx_highbd_idct16x16_256_add_sse2>, 16, 2 },
495 { &fdct_wrapper<vpx_highbd_fdct32x32_sse2>,
496 &highbd_idct_wrapper<vpx_highbd_idct32x32_1024_add_sse2>, 32, 2 },
497 #endif
498 { &fdct_wrapper<vpx_fdct4x4_sse2>, &idct_wrapper<vpx_idct4x4_16_add_sse2>, 4,
499 1 },
500 { &fdct_wrapper<vpx_fdct8x8_sse2>, &idct_wrapper<vpx_idct8x8_64_add_sse2>, 8,
501 1 },
502 { &fdct_wrapper<vpx_fdct16x16_sse2>,
503 &idct_wrapper<vpx_idct16x16_256_add_sse2>, 16, 1 },
504 { &fdct_wrapper<vpx_fdct32x32_sse2>,
505 &idct_wrapper<vpx_idct32x32_1024_add_sse2>, 32, 1 }
506 };
507
508 INSTANTIATE_TEST_CASE_P(
509 SSE2, TransDCT,
510 ::testing::Combine(
511 ::testing::Range(0, static_cast<int>(sizeof(dct_sse2_func_info) /
512 sizeof(dct_sse2_func_info[0]))),
513 ::testing::Values(dct_sse2_func_info), ::testing::Values(0),
514 ::testing::Values(VPX_BITS_8, VPX_BITS_10, VPX_BITS_12)));
515 #endif // HAVE_SSE2
516
517 #if HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH && VPX_ARCH_X86_64
518 // vpx_fdct8x8_ssse3 is only available in 64 bit builds.
519 static const FuncInfo dct_ssse3_func_info = {
520 &fdct_wrapper<vpx_fdct8x8_ssse3>, &idct_wrapper<vpx_idct8x8_64_add_sse2>, 8, 1
521 };
522
523 // TODO(johannkoenig): high bit depth fdct8x8.
524 INSTANTIATE_TEST_CASE_P(SSSE3, TransDCT,
525 ::testing::Values(make_tuple(0, &dct_ssse3_func_info, 0,
526 VPX_BITS_8)));
527 #endif // HAVE_SSSE3 && !CONFIG_VP9_HIGHBITDEPTH && VPX_ARCH_X86_64
528
529 #if HAVE_AVX2 && !CONFIG_VP9_HIGHBITDEPTH
530 static const FuncInfo dct_avx2_func_info = {
531 &fdct_wrapper<vpx_fdct32x32_avx2>, &idct_wrapper<vpx_idct32x32_1024_add_sse2>,
532 32, 1
533 };
534
535 // TODO(johannkoenig): high bit depth fdct32x32.
536 INSTANTIATE_TEST_CASE_P(AVX2, TransDCT,
537 ::testing::Values(make_tuple(0, &dct_avx2_func_info, 0,
538 VPX_BITS_8)));
539 #endif // HAVE_AVX2 && !CONFIG_VP9_HIGHBITDEPTH
540
541 #if HAVE_NEON
542 static const FuncInfo dct_neon_func_info[4] = {
543 { &fdct_wrapper<vpx_fdct4x4_neon>, &idct_wrapper<vpx_idct4x4_16_add_neon>, 4,
544 1 },
545 { &fdct_wrapper<vpx_fdct8x8_neon>, &idct_wrapper<vpx_idct8x8_64_add_neon>, 8,
546 1 },
547 { &fdct_wrapper<vpx_fdct16x16_neon>,
548 &idct_wrapper<vpx_idct16x16_256_add_neon>, 16, 1 },
549 { &fdct_wrapper<vpx_fdct32x32_neon>,
550 &idct_wrapper<vpx_idct32x32_1024_add_neon>, 32, 1 }
551 };
552
553 INSTANTIATE_TEST_CASE_P(
554 NEON, TransDCT,
555 ::testing::Combine(::testing::Range(0, 4),
556 ::testing::Values(dct_neon_func_info),
557 ::testing::Values(0), ::testing::Values(VPX_BITS_8)));
558 #endif // HAVE_NEON
559
560 #if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH
561 static const FuncInfo dct_msa_func_info[4] = {
562 { &fdct_wrapper<vpx_fdct4x4_msa>, &idct_wrapper<vpx_idct4x4_16_add_msa>, 4,
563 1 },
564 { &fdct_wrapper<vpx_fdct8x8_msa>, &idct_wrapper<vpx_idct8x8_64_add_msa>, 8,
565 1 },
566 { &fdct_wrapper<vpx_fdct16x16_msa>, &idct_wrapper<vpx_idct16x16_256_add_msa>,
567 16, 1 },
568 { &fdct_wrapper<vpx_fdct32x32_msa>, &idct_wrapper<vpx_idct32x32_1024_add_msa>,
569 32, 1 }
570 };
571
572 INSTANTIATE_TEST_CASE_P(MSA, TransDCT,
573 ::testing::Combine(::testing::Range(0, 4),
574 ::testing::Values(dct_msa_func_info),
575 ::testing::Values(0),
576 ::testing::Values(VPX_BITS_8)));
577 #endif // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH
578
579 #if HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH
580 static const FuncInfo dct_vsx_func_info = {
581 &fdct_wrapper<vpx_fdct4x4_c>, &idct_wrapper<vpx_idct4x4_16_add_vsx>, 4, 1
582 };
583
584 INSTANTIATE_TEST_CASE_P(VSX, TransDCT,
585 ::testing::Values(make_tuple(0, &dct_vsx_func_info, 0,
586 VPX_BITS_8)));
587 #endif // HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH &&
588
589 #endif // !CONFIG_EMULATE_HARDWARE
590
591 /* -------------------------------------------------------------------------- */
592
593 class TransHT : public TransTestBase {
594 public:
TransHT()595 TransHT() { fwd_txfm_ref = fht_ref; }
596 };
597
TEST_P(TransHT,AccuracyCheck)598 TEST_P(TransHT, AccuracyCheck) {
599 RunAccuracyCheck(size_ == 16 && bit_depth_ > 10 && pixel_size_ == 2 ? 2 : 1);
600 }
601
TEST_P(TransHT,CoeffCheck)602 TEST_P(TransHT, CoeffCheck) { RunCoeffCheck(); }
603
TEST_P(TransHT,MemCheck)604 TEST_P(TransHT, MemCheck) { RunMemCheck(); }
605
TEST_P(TransHT,InvAccuracyCheck)606 TEST_P(TransHT, InvAccuracyCheck) { RunInvAccuracyCheck(1); }
607
608 static const FuncInfo ht_c_func_info[] = {
609 #if CONFIG_VP9_HIGHBITDEPTH
610 { &vp9_highbd_fht4x4_c, &highbd_iht_wrapper<vp9_highbd_iht4x4_16_add_c>, 4,
611 2 },
612 { &vp9_highbd_fht8x8_c, &highbd_iht_wrapper<vp9_highbd_iht8x8_64_add_c>, 8,
613 2 },
614 { &vp9_highbd_fht16x16_c, &highbd_iht_wrapper<vp9_highbd_iht16x16_256_add_c>,
615 16, 2 },
616 #endif
617 { &vp9_fht4x4_c, &iht_wrapper<vp9_iht4x4_16_add_c>, 4, 1 },
618 { &vp9_fht8x8_c, &iht_wrapper<vp9_iht8x8_64_add_c>, 8, 1 },
619 { &vp9_fht16x16_c, &iht_wrapper<vp9_iht16x16_256_add_c>, 16, 1 }
620 };
621
622 INSTANTIATE_TEST_CASE_P(
623 C, TransHT,
624 ::testing::Combine(
625 ::testing::Range(0, static_cast<int>(sizeof(ht_c_func_info) /
626 sizeof(ht_c_func_info[0]))),
627 ::testing::Values(ht_c_func_info), ::testing::Range(0, 4),
628 ::testing::Values(VPX_BITS_8, VPX_BITS_10, VPX_BITS_12)));
629
630 #if !CONFIG_EMULATE_HARDWARE
631
632 #if HAVE_NEON
633
634 static const FuncInfo ht_neon_func_info[] = {
635 #if CONFIG_VP9_HIGHBITDEPTH
636 { &vp9_highbd_fht4x4_c, &highbd_iht_wrapper<vp9_highbd_iht4x4_16_add_neon>, 4,
637 2 },
638 { &vp9_highbd_fht8x8_c, &highbd_iht_wrapper<vp9_highbd_iht8x8_64_add_neon>, 8,
639 2 },
640 { &vp9_highbd_fht16x16_c,
641 &highbd_iht_wrapper<vp9_highbd_iht16x16_256_add_neon>, 16, 2 },
642 #endif
643 { &vp9_fht4x4_c, &iht_wrapper<vp9_iht4x4_16_add_neon>, 4, 1 },
644 { &vp9_fht8x8_c, &iht_wrapper<vp9_iht8x8_64_add_neon>, 8, 1 },
645 { &vp9_fht16x16_c, &iht_wrapper<vp9_iht16x16_256_add_neon>, 16, 1 }
646 };
647
648 INSTANTIATE_TEST_CASE_P(
649 NEON, TransHT,
650 ::testing::Combine(
651 ::testing::Range(0, static_cast<int>(sizeof(ht_neon_func_info) /
652 sizeof(ht_neon_func_info[0]))),
653 ::testing::Values(ht_neon_func_info), ::testing::Range(0, 4),
654 ::testing::Values(VPX_BITS_8, VPX_BITS_10, VPX_BITS_12)));
655 #endif // HAVE_NEON
656
657 #if HAVE_SSE2
658
659 static const FuncInfo ht_sse2_func_info[3] = {
660 { &vp9_fht4x4_sse2, &iht_wrapper<vp9_iht4x4_16_add_sse2>, 4, 1 },
661 { &vp9_fht8x8_sse2, &iht_wrapper<vp9_iht8x8_64_add_sse2>, 8, 1 },
662 { &vp9_fht16x16_sse2, &iht_wrapper<vp9_iht16x16_256_add_sse2>, 16, 1 }
663 };
664
665 INSTANTIATE_TEST_CASE_P(SSE2, TransHT,
666 ::testing::Combine(::testing::Range(0, 3),
667 ::testing::Values(ht_sse2_func_info),
668 ::testing::Range(0, 4),
669 ::testing::Values(VPX_BITS_8)));
670 #endif // HAVE_SSE2
671
672 #if HAVE_SSE4_1 && CONFIG_VP9_HIGHBITDEPTH
673 static const FuncInfo ht_sse4_1_func_info[3] = {
674 { &vp9_highbd_fht4x4_c, &highbd_iht_wrapper<vp9_highbd_iht4x4_16_add_sse4_1>,
675 4, 2 },
676 { vp9_highbd_fht8x8_c, &highbd_iht_wrapper<vp9_highbd_iht8x8_64_add_sse4_1>,
677 8, 2 },
678 { &vp9_highbd_fht16x16_c,
679 &highbd_iht_wrapper<vp9_highbd_iht16x16_256_add_sse4_1>, 16, 2 }
680 };
681
682 INSTANTIATE_TEST_CASE_P(
683 SSE4_1, TransHT,
684 ::testing::Combine(::testing::Range(0, 3),
685 ::testing::Values(ht_sse4_1_func_info),
686 ::testing::Range(0, 4),
687 ::testing::Values(VPX_BITS_8, VPX_BITS_10,
688 VPX_BITS_12)));
689 #endif // HAVE_SSE4_1 && CONFIG_VP9_HIGHBITDEPTH
690
691 #if HAVE_VSX && !CONFIG_EMULATE_HARDWARE && !CONFIG_VP9_HIGHBITDEPTH
692 static const FuncInfo ht_vsx_func_info[3] = {
693 { &vp9_fht4x4_c, &iht_wrapper<vp9_iht4x4_16_add_vsx>, 4, 1 },
694 { &vp9_fht8x8_c, &iht_wrapper<vp9_iht8x8_64_add_vsx>, 8, 1 },
695 { &vp9_fht16x16_c, &iht_wrapper<vp9_iht16x16_256_add_vsx>, 16, 1 }
696 };
697
698 INSTANTIATE_TEST_CASE_P(VSX, TransHT,
699 ::testing::Combine(::testing::Range(0, 3),
700 ::testing::Values(ht_vsx_func_info),
701 ::testing::Range(0, 4),
702 ::testing::Values(VPX_BITS_8)));
703 #endif // HAVE_VSX
704 #endif // !CONFIG_EMULATE_HARDWARE
705
706 /* -------------------------------------------------------------------------- */
707
708 class TransWHT : public TransTestBase {
709 public:
TransWHT()710 TransWHT() { fwd_txfm_ref = fwht_ref; }
711 };
712
TEST_P(TransWHT,AccuracyCheck)713 TEST_P(TransWHT, AccuracyCheck) { RunAccuracyCheck(0); }
714
TEST_P(TransWHT,CoeffCheck)715 TEST_P(TransWHT, CoeffCheck) { RunCoeffCheck(); }
716
TEST_P(TransWHT,MemCheck)717 TEST_P(TransWHT, MemCheck) { RunMemCheck(); }
718
TEST_P(TransWHT,InvAccuracyCheck)719 TEST_P(TransWHT, InvAccuracyCheck) { RunInvAccuracyCheck(0); }
720
721 static const FuncInfo wht_c_func_info[] = {
722 #if CONFIG_VP9_HIGHBITDEPTH
723 { &fdct_wrapper<vp9_highbd_fwht4x4_c>,
724 &highbd_idct_wrapper<vpx_highbd_iwht4x4_16_add_c>, 4, 2 },
725 #endif
726 { &fdct_wrapper<vp9_fwht4x4_c>, &idct_wrapper<vpx_iwht4x4_16_add_c>, 4, 1 }
727 };
728
729 INSTANTIATE_TEST_CASE_P(
730 C, TransWHT,
731 ::testing::Combine(
732 ::testing::Range(0, static_cast<int>(sizeof(wht_c_func_info) /
733 sizeof(wht_c_func_info[0]))),
734 ::testing::Values(wht_c_func_info), ::testing::Values(0),
735 ::testing::Values(VPX_BITS_8, VPX_BITS_10, VPX_BITS_12)));
736
737 #if HAVE_SSE2 && !CONFIG_EMULATE_HARDWARE
738 static const FuncInfo wht_sse2_func_info = {
739 &fdct_wrapper<vp9_fwht4x4_sse2>, &idct_wrapper<vpx_iwht4x4_16_add_sse2>, 4, 1
740 };
741
742 INSTANTIATE_TEST_CASE_P(SSE2, TransWHT,
743 ::testing::Values(make_tuple(0, &wht_sse2_func_info, 0,
744 VPX_BITS_8)));
745 #endif // HAVE_SSE2 && !CONFIG_EMULATE_HARDWARE
746
747 #if HAVE_VSX && !CONFIG_EMULATE_HARDWARE && !CONFIG_VP9_HIGHBITDEPTH
748 static const FuncInfo wht_vsx_func_info = {
749 &fdct_wrapper<vp9_fwht4x4_c>, &idct_wrapper<vpx_iwht4x4_16_add_vsx>, 4, 1
750 };
751
752 INSTANTIATE_TEST_CASE_P(VSX, TransWHT,
753 ::testing::Values(make_tuple(0, &wht_vsx_func_info, 0,
754 VPX_BITS_8)));
755 #endif // HAVE_VSX && !CONFIG_EMULATE_HARDWARE
756 } // namespace
757