2 * Copyright (c) 2012 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.
15 #include "third_party/googletest/src/include/gtest/gtest.h"
17 #include "./vp9_rtcd.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_entropy.h"
24 #include "vp9/common/vp9_scan.h"
25 #include "vpx/vpx_codec.h"
26 #include "vpx/vpx_integer.h"
27 #include "vpx_ports/mem.h"
28 #include "vpx_ports/msvc.h" // for round()
30 using libvpx_test::ACMRandom;
34 const int kNumCoeffs = 256;
35 const double C1 = 0.995184726672197;
36 const double C2 = 0.98078528040323;
37 const double C3 = 0.956940335732209;
38 const double C4 = 0.923879532511287;
39 const double C5 = 0.881921264348355;
40 const double C6 = 0.831469612302545;
41 const double C7 = 0.773010453362737;
42 const double C8 = 0.707106781186548;
43 const double C9 = 0.634393284163646;
44 const double C10 = 0.555570233019602;
45 const double C11 = 0.471396736825998;
46 const double C12 = 0.38268343236509;
47 const double C13 = 0.290284677254462;
48 const double C14 = 0.195090322016128;
49 const double C15 = 0.098017140329561;
51 void butterfly_16x16_dct_1d(double input[16], double output[16]) {
53 double intermediate[16];
57 step[0] = input[0] + input[15];
58 step[1] = input[1] + input[14];
59 step[2] = input[2] + input[13];
60 step[3] = input[3] + input[12];
61 step[4] = input[4] + input[11];
62 step[5] = input[5] + input[10];
63 step[6] = input[6] + input[9];
64 step[7] = input[7] + input[8];
65 step[8] = input[7] - input[8];
66 step[9] = input[6] - input[9];
67 step[10] = input[5] - input[10];
68 step[11] = input[4] - input[11];
69 step[12] = input[3] - input[12];
70 step[13] = input[2] - input[13];
71 step[14] = input[1] - input[14];
72 step[15] = input[0] - input[15];
75 output[0] = step[0] + step[7];
76 output[1] = step[1] + step[6];
77 output[2] = step[2] + step[5];
78 output[3] = step[3] + step[4];
79 output[4] = step[3] - step[4];
80 output[5] = step[2] - step[5];
81 output[6] = step[1] - step[6];
82 output[7] = step[0] - step[7];
85 temp2 = step[15] * C9;
86 output[8] = temp1 + temp2;
88 temp1 = step[9] * C11;
89 temp2 = step[14] * C5;
90 output[9] = temp1 - temp2;
92 temp1 = step[10] * C3;
93 temp2 = step[13] * C13;
94 output[10] = temp1 + temp2;
96 temp1 = step[11] * C15;
97 temp2 = step[12] * C1;
98 output[11] = temp1 - temp2;
100 temp1 = step[11] * C1;
101 temp2 = step[12] * C15;
102 output[12] = temp2 + temp1;
104 temp1 = step[10] * C13;
105 temp2 = step[13] * C3;
106 output[13] = temp2 - temp1;
108 temp1 = step[9] * C5;
109 temp2 = step[14] * C11;
110 output[14] = temp2 + temp1;
112 temp1 = step[8] * C9;
113 temp2 = step[15] * C7;
114 output[15] = temp2 - temp1;
117 step[0] = output[0] + output[3];
118 step[1] = output[1] + output[2];
119 step[2] = output[1] - output[2];
120 step[3] = output[0] - output[3];
122 temp1 = output[4] * C14;
123 temp2 = output[7] * C2;
124 step[4] = temp1 + temp2;
126 temp1 = output[5] * C10;
127 temp2 = output[6] * C6;
128 step[5] = temp1 + temp2;
130 temp1 = output[5] * C6;
131 temp2 = output[6] * C10;
132 step[6] = temp2 - temp1;
134 temp1 = output[4] * C2;
135 temp2 = output[7] * C14;
136 step[7] = temp2 - temp1;
138 step[8] = output[8] + output[11];
139 step[9] = output[9] + output[10];
140 step[10] = output[9] - output[10];
141 step[11] = output[8] - output[11];
143 step[12] = output[12] + output[15];
144 step[13] = output[13] + output[14];
145 step[14] = output[13] - output[14];
146 step[15] = output[12] - output[15];
149 output[0] = (step[0] + step[1]);
150 output[8] = (step[0] - step[1]);
152 temp1 = step[2] * C12;
153 temp2 = step[3] * C4;
154 temp1 = temp1 + temp2;
155 output[4] = 2 * (temp1 * C8);
157 temp1 = step[2] * C4;
158 temp2 = step[3] * C12;
159 temp1 = temp2 - temp1;
160 output[12] = 2 * (temp1 * C8);
162 output[2] = 2 * ((step[4] + step[5]) * C8);
163 output[14] = 2 * ((step[7] - step[6]) * C8);
165 temp1 = step[4] - step[5];
166 temp2 = step[6] + step[7];
167 output[6] = (temp1 + temp2);
168 output[10] = (temp1 - temp2);
170 intermediate[8] = step[8] + step[14];
171 intermediate[9] = step[9] + step[15];
173 temp1 = intermediate[8] * C12;
174 temp2 = intermediate[9] * C4;
175 temp1 = temp1 - temp2;
176 output[3] = 2 * (temp1 * C8);
178 temp1 = intermediate[8] * C4;
179 temp2 = intermediate[9] * C12;
180 temp1 = temp2 + temp1;
181 output[13] = 2 * (temp1 * C8);
183 output[9] = 2 * ((step[10] + step[11]) * C8);
185 intermediate[11] = step[10] - step[11];
186 intermediate[12] = step[12] + step[13];
187 intermediate[13] = step[12] - step[13];
188 intermediate[14] = step[8] - step[14];
189 intermediate[15] = step[9] - step[15];
191 output[15] = (intermediate[11] + intermediate[12]);
192 output[1] = -(intermediate[11] - intermediate[12]);
194 output[7] = 2 * (intermediate[13] * C8);
196 temp1 = intermediate[14] * C12;
197 temp2 = intermediate[15] * C4;
198 temp1 = temp1 - temp2;
199 output[11] = -2 * (temp1 * C8);
201 temp1 = intermediate[14] * C4;
202 temp2 = intermediate[15] * C12;
203 temp1 = temp2 + temp1;
204 output[5] = 2 * (temp1 * C8);
207 void reference_16x16_dct_2d(int16_t input[256], double output[256]) {
208 // First transform columns
209 for (int i = 0; i < 16; ++i) {
210 double temp_in[16], temp_out[16];
211 for (int j = 0; j < 16; ++j) temp_in[j] = input[j * 16 + i];
212 butterfly_16x16_dct_1d(temp_in, temp_out);
213 for (int j = 0; j < 16; ++j) output[j * 16 + i] = temp_out[j];
215 // Then transform rows
216 for (int i = 0; i < 16; ++i) {
217 double temp_in[16], temp_out[16];
218 for (int j = 0; j < 16; ++j) temp_in[j] = output[j + i * 16];
219 butterfly_16x16_dct_1d(temp_in, temp_out);
220 // Scale by some magic number
221 for (int j = 0; j < 16; ++j) output[j + i * 16] = temp_out[j] / 2;
225 typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
226 typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
227 typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
229 typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
232 typedef ::testing::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t>
234 typedef ::testing::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht16x16Param;
235 typedef ::testing::tuple<IdctFunc, IdctFunc, int, vpx_bit_depth_t>
238 void fdct16x16_ref(const int16_t *in, tran_low_t *out, int stride,
240 vpx_fdct16x16_c(in, out, stride);
243 void idct16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
245 vpx_idct16x16_256_add_c(in, dest, stride);
248 void fht16x16_ref(const int16_t *in, tran_low_t *out, int stride, int tx_type) {
249 vp9_fht16x16_c(in, out, stride, tx_type);
252 void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
254 vp9_iht16x16_256_add_c(in, dest, stride, tx_type);
257 #if CONFIG_VP9_HIGHBITDEPTH
258 void idct16x16_10(const tran_low_t *in, uint8_t *out, int stride) {
259 vpx_highbd_idct16x16_256_add_c(in, CAST_TO_SHORTPTR(out), stride, 10);
262 void idct16x16_12(const tran_low_t *in, uint8_t *out, int stride) {
263 vpx_highbd_idct16x16_256_add_c(in, CAST_TO_SHORTPTR(out), stride, 12);
266 void idct16x16_10_ref(const tran_low_t *in, uint8_t *out, int stride,
268 idct16x16_10(in, out, stride);
271 void idct16x16_12_ref(const tran_low_t *in, uint8_t *out, int stride,
273 idct16x16_12(in, out, stride);
276 void iht16x16_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
277 vp9_highbd_iht16x16_256_add_c(in, CAST_TO_SHORTPTR(out), stride, tx_type, 10);
280 void iht16x16_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
281 vp9_highbd_iht16x16_256_add_c(in, CAST_TO_SHORTPTR(out), stride, tx_type, 12);
285 void idct16x16_10_add_10_c(const tran_low_t *in, uint8_t *out, int stride) {
286 vpx_highbd_idct16x16_10_add_c(in, CAST_TO_SHORTPTR(out), stride, 10);
289 void idct16x16_10_add_12_c(const tran_low_t *in, uint8_t *out, int stride) {
290 vpx_highbd_idct16x16_10_add_c(in, CAST_TO_SHORTPTR(out), stride, 12);
293 void idct16x16_256_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
294 vpx_highbd_idct16x16_256_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 10);
297 void idct16x16_256_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
298 vpx_highbd_idct16x16_256_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 12);
301 void idct16x16_10_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
302 vpx_highbd_idct16x16_10_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 10);
305 void idct16x16_10_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
306 vpx_highbd_idct16x16_10_add_sse2(in, CAST_TO_SHORTPTR(out), stride, 12);
309 #endif // CONFIG_VP9_HIGHBITDEPTH
311 class Trans16x16TestBase {
313 virtual ~Trans16x16TestBase() {}
316 virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
318 virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
320 void RunAccuracyCheck() {
321 ACMRandom rnd(ACMRandom::DeterministicSeed());
322 uint32_t max_error = 0;
323 int64_t total_error = 0;
324 const int count_test_block = 10000;
325 for (int i = 0; i < count_test_block; ++i) {
326 DECLARE_ALIGNED(16, int16_t, test_input_block[kNumCoeffs]);
327 DECLARE_ALIGNED(16, tran_low_t, test_temp_block[kNumCoeffs]);
328 DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
329 DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
330 #if CONFIG_VP9_HIGHBITDEPTH
331 DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
332 DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
335 // Initialize a test block with input range [-mask_, mask_].
336 for (int j = 0; j < kNumCoeffs; ++j) {
337 if (bit_depth_ == VPX_BITS_8) {
338 src[j] = rnd.Rand8();
339 dst[j] = rnd.Rand8();
340 test_input_block[j] = src[j] - dst[j];
341 #if CONFIG_VP9_HIGHBITDEPTH
343 src16[j] = rnd.Rand16() & mask_;
344 dst16[j] = rnd.Rand16() & mask_;
345 test_input_block[j] = src16[j] - dst16[j];
350 ASM_REGISTER_STATE_CHECK(
351 RunFwdTxfm(test_input_block, test_temp_block, pitch_));
352 if (bit_depth_ == VPX_BITS_8) {
353 ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
354 #if CONFIG_VP9_HIGHBITDEPTH
356 ASM_REGISTER_STATE_CHECK(
357 RunInvTxfm(test_temp_block, CAST_TO_BYTEPTR(dst16), pitch_));
361 for (int j = 0; j < kNumCoeffs; ++j) {
362 #if CONFIG_VP9_HIGHBITDEPTH
364 bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
366 const int32_t diff = dst[j] - src[j];
368 const uint32_t error = diff * diff;
369 if (max_error < error) max_error = error;
370 total_error += error;
374 EXPECT_GE(1u << 2 * (bit_depth_ - 8), max_error)
375 << "Error: 16x16 FHT/IHT has an individual round trip error > 1";
377 EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
378 << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
381 void RunCoeffCheck() {
382 ACMRandom rnd(ACMRandom::DeterministicSeed());
383 const int count_test_block = 1000;
384 DECLARE_ALIGNED(16, int16_t, input_block[kNumCoeffs]);
385 DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
386 DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
388 for (int i = 0; i < count_test_block; ++i) {
389 // Initialize a test block with input range [-mask_, mask_].
390 for (int j = 0; j < kNumCoeffs; ++j) {
391 input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
394 fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
395 ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
397 // The minimum quant value is 4.
398 for (int j = 0; j < kNumCoeffs; ++j)
399 EXPECT_EQ(output_block[j], output_ref_block[j]);
404 ACMRandom rnd(ACMRandom::DeterministicSeed());
405 const int count_test_block = 1000;
406 DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
407 DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
408 DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
410 for (int i = 0; i < count_test_block; ++i) {
411 // Initialize a test block with input range [-mask_, mask_].
412 for (int j = 0; j < kNumCoeffs; ++j) {
413 input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
416 for (int j = 0; j < kNumCoeffs; ++j) input_extreme_block[j] = mask_;
418 for (int j = 0; j < kNumCoeffs; ++j) input_extreme_block[j] = -mask_;
421 fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
422 ASM_REGISTER_STATE_CHECK(
423 RunFwdTxfm(input_extreme_block, output_block, pitch_));
425 // The minimum quant value is 4.
426 for (int j = 0; j < kNumCoeffs; ++j) {
427 EXPECT_EQ(output_block[j], output_ref_block[j]);
428 EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
429 << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
434 void RunQuantCheck(int dc_thred, int ac_thred) {
435 ACMRandom rnd(ACMRandom::DeterministicSeed());
436 const int count_test_block = 100000;
437 DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
438 DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
440 DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
441 DECLARE_ALIGNED(16, uint8_t, ref[kNumCoeffs]);
442 #if CONFIG_VP9_HIGHBITDEPTH
443 DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
444 DECLARE_ALIGNED(16, uint16_t, ref16[kNumCoeffs]);
447 for (int i = 0; i < count_test_block; ++i) {
448 // Initialize a test block with input range [-mask_, mask_].
449 for (int j = 0; j < kNumCoeffs; ++j) {
450 input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
453 for (int j = 0; j < kNumCoeffs; ++j) input_extreme_block[j] = mask_;
456 for (int j = 0; j < kNumCoeffs; ++j) input_extreme_block[j] = -mask_;
459 fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
461 // clear reconstructed pixel buffers
462 memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
463 memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
464 #if CONFIG_VP9_HIGHBITDEPTH
465 memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
466 memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
469 // quantization with maximum allowed step sizes
470 output_ref_block[0] = (output_ref_block[0] / dc_thred) * dc_thred;
471 for (int j = 1; j < kNumCoeffs; ++j) {
472 output_ref_block[j] = (output_ref_block[j] / ac_thred) * ac_thred;
474 if (bit_depth_ == VPX_BITS_8) {
475 inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
476 ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
477 #if CONFIG_VP9_HIGHBITDEPTH
479 inv_txfm_ref(output_ref_block, CAST_TO_BYTEPTR(ref16), pitch_,
481 ASM_REGISTER_STATE_CHECK(
482 RunInvTxfm(output_ref_block, CAST_TO_BYTEPTR(dst16), pitch_));
485 if (bit_depth_ == VPX_BITS_8) {
486 for (int j = 0; j < kNumCoeffs; ++j) EXPECT_EQ(ref[j], dst[j]);
487 #if CONFIG_VP9_HIGHBITDEPTH
489 for (int j = 0; j < kNumCoeffs; ++j) EXPECT_EQ(ref16[j], dst16[j]);
495 void RunInvAccuracyCheck() {
496 ACMRandom rnd(ACMRandom::DeterministicSeed());
497 const int count_test_block = 1000;
498 DECLARE_ALIGNED(16, int16_t, in[kNumCoeffs]);
499 DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
500 DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
501 DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
502 #if CONFIG_VP9_HIGHBITDEPTH
503 DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
504 DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
505 #endif // CONFIG_VP9_HIGHBITDEPTH
507 for (int i = 0; i < count_test_block; ++i) {
508 double out_r[kNumCoeffs];
510 // Initialize a test block with input range [-255, 255].
511 for (int j = 0; j < kNumCoeffs; ++j) {
512 if (bit_depth_ == VPX_BITS_8) {
513 src[j] = rnd.Rand8();
514 dst[j] = rnd.Rand8();
515 in[j] = src[j] - dst[j];
516 #if CONFIG_VP9_HIGHBITDEPTH
518 src16[j] = rnd.Rand16() & mask_;
519 dst16[j] = rnd.Rand16() & mask_;
520 in[j] = src16[j] - dst16[j];
521 #endif // CONFIG_VP9_HIGHBITDEPTH
525 reference_16x16_dct_2d(in, out_r);
526 for (int j = 0; j < kNumCoeffs; ++j) {
527 coeff[j] = static_cast<tran_low_t>(round(out_r[j]));
530 if (bit_depth_ == VPX_BITS_8) {
531 ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
532 #if CONFIG_VP9_HIGHBITDEPTH
534 ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CAST_TO_BYTEPTR(dst16), 16));
535 #endif // CONFIG_VP9_HIGHBITDEPTH
538 for (int j = 0; j < kNumCoeffs; ++j) {
539 #if CONFIG_VP9_HIGHBITDEPTH
540 const uint32_t diff =
541 bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
543 const uint32_t diff = dst[j] - src[j];
544 #endif // CONFIG_VP9_HIGHBITDEPTH
545 const uint32_t error = diff * diff;
547 << "Error: 16x16 IDCT has error " << error << " at index " << j;
552 void CompareInvReference(IdctFunc ref_txfm, int thresh) {
553 ACMRandom rnd(ACMRandom::DeterministicSeed());
554 const int count_test_block = 10000;
556 const int16_t *scan = vp9_default_scan_orders[TX_16X16].scan;
557 DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
558 DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
559 DECLARE_ALIGNED(16, uint8_t, ref[kNumCoeffs]);
560 #if CONFIG_VP9_HIGHBITDEPTH
561 DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
562 DECLARE_ALIGNED(16, uint16_t, ref16[kNumCoeffs]);
563 #endif // CONFIG_VP9_HIGHBITDEPTH
565 for (int i = 0; i < count_test_block; ++i) {
566 for (int j = 0; j < kNumCoeffs; ++j) {
568 // Random values less than the threshold, either positive or negative
569 coeff[scan[j]] = rnd(thresh) * (1 - 2 * (i % 2));
573 if (bit_depth_ == VPX_BITS_8) {
576 #if CONFIG_VP9_HIGHBITDEPTH
580 #endif // CONFIG_VP9_HIGHBITDEPTH
583 if (bit_depth_ == VPX_BITS_8) {
584 ref_txfm(coeff, ref, pitch_);
585 ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
587 #if CONFIG_VP9_HIGHBITDEPTH
588 ref_txfm(coeff, CAST_TO_BYTEPTR(ref16), pitch_);
589 ASM_REGISTER_STATE_CHECK(
590 RunInvTxfm(coeff, CAST_TO_BYTEPTR(dst16), pitch_));
591 #endif // CONFIG_VP9_HIGHBITDEPTH
594 for (int j = 0; j < kNumCoeffs; ++j) {
595 #if CONFIG_VP9_HIGHBITDEPTH
596 const uint32_t diff =
597 bit_depth_ == VPX_BITS_8 ? dst[j] - ref[j] : dst16[j] - ref16[j];
599 const uint32_t diff = dst[j] - ref[j];
600 #endif // CONFIG_VP9_HIGHBITDEPTH
601 const uint32_t error = diff * diff;
602 EXPECT_EQ(0u, error) << "Error: 16x16 IDCT Comparison has error "
603 << error << " at index " << j;
610 vpx_bit_depth_t bit_depth_;
612 FhtFunc fwd_txfm_ref;
613 IhtFunc inv_txfm_ref;
616 class Trans16x16DCT : public Trans16x16TestBase,
617 public ::testing::TestWithParam<Dct16x16Param> {
619 virtual ~Trans16x16DCT() {}
621 virtual void SetUp() {
622 fwd_txfm_ = GET_PARAM(0);
623 inv_txfm_ = GET_PARAM(1);
624 tx_type_ = GET_PARAM(2);
625 bit_depth_ = GET_PARAM(3);
627 fwd_txfm_ref = fdct16x16_ref;
628 inv_txfm_ref = idct16x16_ref;
629 mask_ = (1 << bit_depth_) - 1;
630 #if CONFIG_VP9_HIGHBITDEPTH
631 switch (bit_depth_) {
632 case VPX_BITS_10: inv_txfm_ref = idct16x16_10_ref; break;
633 case VPX_BITS_12: inv_txfm_ref = idct16x16_12_ref; break;
634 default: inv_txfm_ref = idct16x16_ref; break;
637 inv_txfm_ref = idct16x16_ref;
640 virtual void TearDown() { libvpx_test::ClearSystemState(); }
643 void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
644 fwd_txfm_(in, out, stride);
646 void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
647 inv_txfm_(out, dst, stride);
654 TEST_P(Trans16x16DCT, AccuracyCheck) { RunAccuracyCheck(); }
656 TEST_P(Trans16x16DCT, CoeffCheck) { RunCoeffCheck(); }
658 TEST_P(Trans16x16DCT, MemCheck) { RunMemCheck(); }
660 TEST_P(Trans16x16DCT, QuantCheck) {
661 // Use maximally allowed quantization step sizes for DC and AC
662 // coefficients respectively.
663 RunQuantCheck(1336, 1828);
666 TEST_P(Trans16x16DCT, InvAccuracyCheck) { RunInvAccuracyCheck(); }
668 class Trans16x16HT : public Trans16x16TestBase,
669 public ::testing::TestWithParam<Ht16x16Param> {
671 virtual ~Trans16x16HT() {}
673 virtual void SetUp() {
674 fwd_txfm_ = GET_PARAM(0);
675 inv_txfm_ = GET_PARAM(1);
676 tx_type_ = GET_PARAM(2);
677 bit_depth_ = GET_PARAM(3);
679 fwd_txfm_ref = fht16x16_ref;
680 inv_txfm_ref = iht16x16_ref;
681 mask_ = (1 << bit_depth_) - 1;
682 #if CONFIG_VP9_HIGHBITDEPTH
683 switch (bit_depth_) {
684 case VPX_BITS_10: inv_txfm_ref = iht16x16_10; break;
685 case VPX_BITS_12: inv_txfm_ref = iht16x16_12; break;
686 default: inv_txfm_ref = iht16x16_ref; break;
689 inv_txfm_ref = iht16x16_ref;
692 virtual void TearDown() { libvpx_test::ClearSystemState(); }
695 void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
696 fwd_txfm_(in, out, stride, tx_type_);
698 void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
699 inv_txfm_(out, dst, stride, tx_type_);
706 TEST_P(Trans16x16HT, AccuracyCheck) { RunAccuracyCheck(); }
708 TEST_P(Trans16x16HT, CoeffCheck) { RunCoeffCheck(); }
710 TEST_P(Trans16x16HT, MemCheck) { RunMemCheck(); }
712 TEST_P(Trans16x16HT, QuantCheck) {
713 // The encoder skips any non-DC intra prediction modes,
714 // when the quantization step size goes beyond 988.
715 RunQuantCheck(429, 729);
718 class InvTrans16x16DCT : public Trans16x16TestBase,
719 public ::testing::TestWithParam<Idct16x16Param> {
721 virtual ~InvTrans16x16DCT() {}
723 virtual void SetUp() {
724 ref_txfm_ = GET_PARAM(0);
725 inv_txfm_ = GET_PARAM(1);
726 thresh_ = GET_PARAM(2);
727 bit_depth_ = GET_PARAM(3);
729 mask_ = (1 << bit_depth_) - 1;
731 virtual void TearDown() { libvpx_test::ClearSystemState(); }
734 void RunFwdTxfm(int16_t * /*in*/, tran_low_t * /*out*/, int /*stride*/) {}
735 void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
736 inv_txfm_(out, dst, stride);
744 TEST_P(InvTrans16x16DCT, CompareReference) {
745 CompareInvReference(ref_txfm_, thresh_);
748 using ::testing::make_tuple;
750 #if CONFIG_VP9_HIGHBITDEPTH
751 INSTANTIATE_TEST_CASE_P(
754 make_tuple(&vpx_highbd_fdct16x16_c, &idct16x16_10, 0, VPX_BITS_10),
755 make_tuple(&vpx_highbd_fdct16x16_c, &idct16x16_12, 0, VPX_BITS_12),
756 make_tuple(&vpx_fdct16x16_c, &vpx_idct16x16_256_add_c, 0, VPX_BITS_8)));
758 INSTANTIATE_TEST_CASE_P(C, Trans16x16DCT,
759 ::testing::Values(make_tuple(&vpx_fdct16x16_c,
760 &vpx_idct16x16_256_add_c,
762 #endif // CONFIG_VP9_HIGHBITDEPTH
764 #if CONFIG_VP9_HIGHBITDEPTH
765 INSTANTIATE_TEST_CASE_P(
768 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 0, VPX_BITS_10),
769 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 1, VPX_BITS_10),
770 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 2, VPX_BITS_10),
771 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 3, VPX_BITS_10),
772 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 0, VPX_BITS_12),
773 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 1, VPX_BITS_12),
774 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 2, VPX_BITS_12),
775 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 3, VPX_BITS_12),
776 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
777 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
778 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
779 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
781 INSTANTIATE_TEST_CASE_P(
784 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
785 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
786 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
787 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
788 #endif // CONFIG_VP9_HIGHBITDEPTH
790 #if HAVE_NEON && !CONFIG_EMULATE_HARDWARE
791 INSTANTIATE_TEST_CASE_P(
793 ::testing::Values(make_tuple(&vpx_fdct16x16_neon,
794 &vpx_idct16x16_256_add_neon, 0, VPX_BITS_8)));
795 #endif // HAVE_NEON && !CONFIG_EMULATE_HARDWARE
797 #if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
798 INSTANTIATE_TEST_CASE_P(
800 ::testing::Values(make_tuple(&vpx_fdct16x16_sse2,
801 &vpx_idct16x16_256_add_sse2, 0, VPX_BITS_8)));
802 INSTANTIATE_TEST_CASE_P(
804 ::testing::Values(make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2,
806 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2,
808 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2,
810 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2,
812 #endif // HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
814 #if HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
815 INSTANTIATE_TEST_CASE_P(
818 make_tuple(&vpx_highbd_fdct16x16_sse2, &idct16x16_10, 0, VPX_BITS_10),
819 make_tuple(&vpx_highbd_fdct16x16_c, &idct16x16_256_add_10_sse2, 0,
821 make_tuple(&vpx_highbd_fdct16x16_sse2, &idct16x16_12, 0, VPX_BITS_12),
822 make_tuple(&vpx_highbd_fdct16x16_c, &idct16x16_256_add_12_sse2, 0,
824 make_tuple(&vpx_fdct16x16_sse2, &vpx_idct16x16_256_add_c, 0,
826 INSTANTIATE_TEST_CASE_P(
829 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
830 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
831 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
832 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 3,
834 // Optimizations take effect at a threshold of 3155, so we use a value close to
835 // that to test both branches.
836 INSTANTIATE_TEST_CASE_P(
837 SSE2, InvTrans16x16DCT,
838 ::testing::Values(make_tuple(&idct16x16_10_add_10_c,
839 &idct16x16_10_add_10_sse2, 3167, VPX_BITS_10),
840 make_tuple(&idct16x16_10, &idct16x16_256_add_10_sse2,
842 make_tuple(&idct16x16_10_add_12_c,
843 &idct16x16_10_add_12_sse2, 3167, VPX_BITS_12),
844 make_tuple(&idct16x16_12, &idct16x16_256_add_12_sse2,
845 3167, VPX_BITS_12)));
846 #endif // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
848 #if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
849 INSTANTIATE_TEST_CASE_P(MSA, Trans16x16DCT,
850 ::testing::Values(make_tuple(&vpx_fdct16x16_msa,
851 &vpx_idct16x16_256_add_msa,
853 INSTANTIATE_TEST_CASE_P(
856 make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 0, VPX_BITS_8),
857 make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 1, VPX_BITS_8),
858 make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 2, VPX_BITS_8),
859 make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 3,
861 #endif // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
863 #if HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
864 INSTANTIATE_TEST_CASE_P(VSX, Trans16x16DCT,
865 ::testing::Values(make_tuple(&vpx_fdct16x16_c,
866 &vpx_idct16x16_256_add_vsx,
868 #endif // HAVE_VSX && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE