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];
84 temp1 = step[ 8] * C7;
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)
212 temp_in[j] = input[j * 16 + i];
213 butterfly_16x16_dct_1d(temp_in, temp_out);
214 for (int j = 0; j < 16; ++j)
215 output[j * 16 + i] = temp_out[j];
217 // Then transform rows
218 for (int i = 0; i < 16; ++i) {
219 double temp_in[16], temp_out[16];
220 for (int j = 0; j < 16; ++j)
221 temp_in[j] = output[j + i * 16];
222 butterfly_16x16_dct_1d(temp_in, temp_out);
223 // Scale by some magic number
224 for (int j = 0; j < 16; ++j)
225 output[j + i * 16] = temp_out[j]/2;
229 typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
230 typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
231 typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
233 typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
236 typedef std::tr1::tuple<FdctFunc, IdctFunc, int, vpx_bit_depth_t> Dct16x16Param;
237 typedef std::tr1::tuple<FhtFunc, IhtFunc, int, vpx_bit_depth_t> Ht16x16Param;
238 typedef std::tr1::tuple<IdctFunc, IdctFunc, int, vpx_bit_depth_t>
241 void fdct16x16_ref(const int16_t *in, tran_low_t *out, int stride,
243 vpx_fdct16x16_c(in, out, stride);
246 void idct16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
248 vpx_idct16x16_256_add_c(in, dest, stride);
251 void fht16x16_ref(const int16_t *in, tran_low_t *out, int stride,
253 vp9_fht16x16_c(in, out, stride, tx_type);
256 void iht16x16_ref(const tran_low_t *in, uint8_t *dest, int stride,
258 vp9_iht16x16_256_add_c(in, dest, stride, tx_type);
261 #if CONFIG_VP9_HIGHBITDEPTH
262 void idct16x16_10(const tran_low_t *in, uint8_t *out, int stride) {
263 vpx_highbd_idct16x16_256_add_c(in, out, stride, 10);
266 void idct16x16_12(const tran_low_t *in, uint8_t *out, int stride) {
267 vpx_highbd_idct16x16_256_add_c(in, out, stride, 12);
270 void idct16x16_10_ref(const tran_low_t *in, uint8_t *out, int stride,
272 idct16x16_10(in, out, stride);
275 void idct16x16_12_ref(const tran_low_t *in, uint8_t *out, int stride,
277 idct16x16_12(in, out, stride);
280 void iht16x16_10(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
281 vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 10);
284 void iht16x16_12(const tran_low_t *in, uint8_t *out, int stride, int tx_type) {
285 vp9_highbd_iht16x16_256_add_c(in, out, stride, tx_type, 12);
289 void idct16x16_10_add_10_c(const tran_low_t *in, uint8_t *out, int stride) {
290 vpx_highbd_idct16x16_10_add_c(in, out, stride, 10);
293 void idct16x16_10_add_12_c(const tran_low_t *in, uint8_t *out, int stride) {
294 vpx_highbd_idct16x16_10_add_c(in, out, stride, 12);
297 void idct16x16_256_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
298 vpx_highbd_idct16x16_256_add_sse2(in, out, stride, 10);
301 void idct16x16_256_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
302 vpx_highbd_idct16x16_256_add_sse2(in, out, stride, 12);
305 void idct16x16_10_add_10_sse2(const tran_low_t *in, uint8_t *out, int stride) {
306 vpx_highbd_idct16x16_10_add_sse2(in, out, stride, 10);
309 void idct16x16_10_add_12_sse2(const tran_low_t *in, uint8_t *out, int stride) {
310 vpx_highbd_idct16x16_10_add_sse2(in, out, stride, 12);
313 #endif // CONFIG_VP9_HIGHBITDEPTH
315 class Trans16x16TestBase {
317 virtual ~Trans16x16TestBase() {}
320 virtual void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) = 0;
322 virtual void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) = 0;
324 void RunAccuracyCheck() {
325 ACMRandom rnd(ACMRandom::DeterministicSeed());
326 uint32_t max_error = 0;
327 int64_t total_error = 0;
328 const int count_test_block = 10000;
329 for (int i = 0; i < count_test_block; ++i) {
330 DECLARE_ALIGNED(16, int16_t, test_input_block[kNumCoeffs]);
331 DECLARE_ALIGNED(16, tran_low_t, test_temp_block[kNumCoeffs]);
332 DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
333 DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
334 #if CONFIG_VP9_HIGHBITDEPTH
335 DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
336 DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
339 // Initialize a test block with input range [-mask_, mask_].
340 for (int j = 0; j < kNumCoeffs; ++j) {
341 if (bit_depth_ == VPX_BITS_8) {
342 src[j] = rnd.Rand8();
343 dst[j] = rnd.Rand8();
344 test_input_block[j] = src[j] - dst[j];
345 #if CONFIG_VP9_HIGHBITDEPTH
347 src16[j] = rnd.Rand16() & mask_;
348 dst16[j] = rnd.Rand16() & mask_;
349 test_input_block[j] = src16[j] - dst16[j];
354 ASM_REGISTER_STATE_CHECK(RunFwdTxfm(test_input_block,
355 test_temp_block, pitch_));
356 if (bit_depth_ == VPX_BITS_8) {
357 ASM_REGISTER_STATE_CHECK(
358 RunInvTxfm(test_temp_block, dst, pitch_));
359 #if CONFIG_VP9_HIGHBITDEPTH
361 ASM_REGISTER_STATE_CHECK(
362 RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
366 for (int j = 0; j < kNumCoeffs; ++j) {
367 #if CONFIG_VP9_HIGHBITDEPTH
369 bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
371 const int32_t diff = dst[j] - src[j];
373 const uint32_t error = diff * diff;
374 if (max_error < error)
376 total_error += error;
380 EXPECT_GE(1u << 2 * (bit_depth_ - 8), max_error)
381 << "Error: 16x16 FHT/IHT has an individual round trip error > 1";
383 EXPECT_GE(count_test_block << 2 * (bit_depth_ - 8), total_error)
384 << "Error: 16x16 FHT/IHT has average round trip error > 1 per block";
387 void RunCoeffCheck() {
388 ACMRandom rnd(ACMRandom::DeterministicSeed());
389 const int count_test_block = 1000;
390 DECLARE_ALIGNED(16, int16_t, input_block[kNumCoeffs]);
391 DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
392 DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
394 for (int i = 0; i < count_test_block; ++i) {
395 // Initialize a test block with input range [-mask_, mask_].
396 for (int j = 0; j < kNumCoeffs; ++j)
397 input_block[j] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
399 fwd_txfm_ref(input_block, output_ref_block, pitch_, tx_type_);
400 ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, pitch_));
402 // The minimum quant value is 4.
403 for (int j = 0; j < kNumCoeffs; ++j)
404 EXPECT_EQ(output_block[j], output_ref_block[j]);
409 ACMRandom rnd(ACMRandom::DeterministicSeed());
410 const int count_test_block = 1000;
411 DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
412 DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
413 DECLARE_ALIGNED(16, tran_low_t, output_block[kNumCoeffs]);
415 for (int i = 0; i < count_test_block; ++i) {
416 // Initialize a test block with input range [-mask_, mask_].
417 for (int j = 0; j < kNumCoeffs; ++j) {
418 input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
421 for (int j = 0; j < kNumCoeffs; ++j)
422 input_extreme_block[j] = mask_;
424 for (int j = 0; j < kNumCoeffs; ++j)
425 input_extreme_block[j] = -mask_;
428 fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
429 ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_extreme_block,
430 output_block, pitch_));
432 // The minimum quant value is 4.
433 for (int j = 0; j < kNumCoeffs; ++j) {
434 EXPECT_EQ(output_block[j], output_ref_block[j]);
435 EXPECT_GE(4 * DCT_MAX_VALUE << (bit_depth_ - 8), abs(output_block[j]))
436 << "Error: 16x16 FDCT has coefficient larger than 4*DCT_MAX_VALUE";
441 void RunQuantCheck(int dc_thred, int ac_thred) {
442 ACMRandom rnd(ACMRandom::DeterministicSeed());
443 const int count_test_block = 100000;
444 DECLARE_ALIGNED(16, int16_t, input_extreme_block[kNumCoeffs]);
445 DECLARE_ALIGNED(16, tran_low_t, output_ref_block[kNumCoeffs]);
447 DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
448 DECLARE_ALIGNED(16, uint8_t, ref[kNumCoeffs]);
449 #if CONFIG_VP9_HIGHBITDEPTH
450 DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
451 DECLARE_ALIGNED(16, uint16_t, ref16[kNumCoeffs]);
454 for (int i = 0; i < count_test_block; ++i) {
455 // Initialize a test block with input range [-mask_, mask_].
456 for (int j = 0; j < kNumCoeffs; ++j) {
457 input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
460 for (int j = 0; j < kNumCoeffs; ++j)
461 input_extreme_block[j] = mask_;
463 for (int j = 0; j < kNumCoeffs; ++j)
464 input_extreme_block[j] = -mask_;
466 fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, tx_type_);
468 // clear reconstructed pixel buffers
469 memset(dst, 0, kNumCoeffs * sizeof(uint8_t));
470 memset(ref, 0, kNumCoeffs * sizeof(uint8_t));
471 #if CONFIG_VP9_HIGHBITDEPTH
472 memset(dst16, 0, kNumCoeffs * sizeof(uint16_t));
473 memset(ref16, 0, kNumCoeffs * sizeof(uint16_t));
476 // quantization with maximum allowed step sizes
477 output_ref_block[0] = (output_ref_block[0] / dc_thred) * dc_thred;
478 for (int j = 1; j < kNumCoeffs; ++j)
479 output_ref_block[j] = (output_ref_block[j] / ac_thred) * ac_thred;
480 if (bit_depth_ == VPX_BITS_8) {
481 inv_txfm_ref(output_ref_block, ref, pitch_, tx_type_);
482 ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block, dst, pitch_));
483 #if CONFIG_VP9_HIGHBITDEPTH
485 inv_txfm_ref(output_ref_block, CONVERT_TO_BYTEPTR(ref16), pitch_,
487 ASM_REGISTER_STATE_CHECK(RunInvTxfm(output_ref_block,
488 CONVERT_TO_BYTEPTR(dst16), pitch_));
491 if (bit_depth_ == VPX_BITS_8) {
492 for (int j = 0; j < kNumCoeffs; ++j)
493 EXPECT_EQ(ref[j], dst[j]);
494 #if CONFIG_VP9_HIGHBITDEPTH
496 for (int j = 0; j < kNumCoeffs; ++j)
497 EXPECT_EQ(ref16[j], dst16[j]);
503 void RunInvAccuracyCheck() {
504 ACMRandom rnd(ACMRandom::DeterministicSeed());
505 const int count_test_block = 1000;
506 DECLARE_ALIGNED(16, int16_t, in[kNumCoeffs]);
507 DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
508 DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
509 DECLARE_ALIGNED(16, uint8_t, src[kNumCoeffs]);
510 #if CONFIG_VP9_HIGHBITDEPTH
511 DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
512 DECLARE_ALIGNED(16, uint16_t, src16[kNumCoeffs]);
513 #endif // CONFIG_VP9_HIGHBITDEPTH
515 for (int i = 0; i < count_test_block; ++i) {
516 double out_r[kNumCoeffs];
518 // Initialize a test block with input range [-255, 255].
519 for (int j = 0; j < kNumCoeffs; ++j) {
520 if (bit_depth_ == VPX_BITS_8) {
521 src[j] = rnd.Rand8();
522 dst[j] = rnd.Rand8();
523 in[j] = src[j] - dst[j];
524 #if CONFIG_VP9_HIGHBITDEPTH
526 src16[j] = rnd.Rand16() & mask_;
527 dst16[j] = rnd.Rand16() & mask_;
528 in[j] = src16[j] - dst16[j];
529 #endif // CONFIG_VP9_HIGHBITDEPTH
533 reference_16x16_dct_2d(in, out_r);
534 for (int j = 0; j < kNumCoeffs; ++j)
535 coeff[j] = static_cast<tran_low_t>(round(out_r[j]));
537 if (bit_depth_ == VPX_BITS_8) {
538 ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, 16));
539 #if CONFIG_VP9_HIGHBITDEPTH
541 ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
543 #endif // CONFIG_VP9_HIGHBITDEPTH
546 for (int j = 0; j < kNumCoeffs; ++j) {
547 #if CONFIG_VP9_HIGHBITDEPTH
548 const uint32_t diff =
549 bit_depth_ == VPX_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
551 const uint32_t diff = dst[j] - src[j];
552 #endif // CONFIG_VP9_HIGHBITDEPTH
553 const uint32_t error = diff * diff;
555 << "Error: 16x16 IDCT has error " << error
556 << " at index " << j;
561 void CompareInvReference(IdctFunc ref_txfm, int thresh) {
562 ACMRandom rnd(ACMRandom::DeterministicSeed());
563 const int count_test_block = 10000;
565 const int16_t *scan = vp9_default_scan_orders[TX_16X16].scan;
566 DECLARE_ALIGNED(16, tran_low_t, coeff[kNumCoeffs]);
567 DECLARE_ALIGNED(16, uint8_t, dst[kNumCoeffs]);
568 DECLARE_ALIGNED(16, uint8_t, ref[kNumCoeffs]);
569 #if CONFIG_VP9_HIGHBITDEPTH
570 DECLARE_ALIGNED(16, uint16_t, dst16[kNumCoeffs]);
571 DECLARE_ALIGNED(16, uint16_t, ref16[kNumCoeffs]);
572 #endif // CONFIG_VP9_HIGHBITDEPTH
574 for (int i = 0; i < count_test_block; ++i) {
575 for (int j = 0; j < kNumCoeffs; ++j) {
577 // Random values less than the threshold, either positive or negative
578 coeff[scan[j]] = rnd(thresh) * (1 - 2 * (i % 2));
582 if (bit_depth_ == VPX_BITS_8) {
585 #if CONFIG_VP9_HIGHBITDEPTH
589 #endif // CONFIG_VP9_HIGHBITDEPTH
592 if (bit_depth_ == VPX_BITS_8) {
593 ref_txfm(coeff, ref, pitch_);
594 ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
596 #if CONFIG_VP9_HIGHBITDEPTH
597 ref_txfm(coeff, CONVERT_TO_BYTEPTR(ref16), pitch_);
598 ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16),
600 #endif // CONFIG_VP9_HIGHBITDEPTH
603 for (int j = 0; j < kNumCoeffs; ++j) {
604 #if CONFIG_VP9_HIGHBITDEPTH
605 const uint32_t diff =
606 bit_depth_ == VPX_BITS_8 ? dst[j] - ref[j] : dst16[j] - ref16[j];
608 const uint32_t diff = dst[j] - ref[j];
609 #endif // CONFIG_VP9_HIGHBITDEPTH
610 const uint32_t error = diff * diff;
612 << "Error: 16x16 IDCT Comparison has error " << error
613 << " at index " << j;
620 vpx_bit_depth_t bit_depth_;
622 FhtFunc fwd_txfm_ref;
623 IhtFunc inv_txfm_ref;
627 : public Trans16x16TestBase,
628 public ::testing::TestWithParam<Dct16x16Param> {
630 virtual ~Trans16x16DCT() {}
632 virtual void SetUp() {
633 fwd_txfm_ = GET_PARAM(0);
634 inv_txfm_ = GET_PARAM(1);
635 tx_type_ = GET_PARAM(2);
636 bit_depth_ = GET_PARAM(3);
638 fwd_txfm_ref = fdct16x16_ref;
639 inv_txfm_ref = idct16x16_ref;
640 mask_ = (1 << bit_depth_) - 1;
641 #if CONFIG_VP9_HIGHBITDEPTH
642 switch (bit_depth_) {
644 inv_txfm_ref = idct16x16_10_ref;
647 inv_txfm_ref = idct16x16_12_ref;
650 inv_txfm_ref = idct16x16_ref;
654 inv_txfm_ref = idct16x16_ref;
657 virtual void TearDown() { libvpx_test::ClearSystemState(); }
660 void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
661 fwd_txfm_(in, out, stride);
663 void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
664 inv_txfm_(out, dst, stride);
671 TEST_P(Trans16x16DCT, AccuracyCheck) {
675 TEST_P(Trans16x16DCT, CoeffCheck) {
679 TEST_P(Trans16x16DCT, MemCheck) {
683 TEST_P(Trans16x16DCT, QuantCheck) {
684 // Use maximally allowed quantization step sizes for DC and AC
685 // coefficients respectively.
686 RunQuantCheck(1336, 1828);
689 TEST_P(Trans16x16DCT, InvAccuracyCheck) {
690 RunInvAccuracyCheck();
694 : public Trans16x16TestBase,
695 public ::testing::TestWithParam<Ht16x16Param> {
697 virtual ~Trans16x16HT() {}
699 virtual void SetUp() {
700 fwd_txfm_ = GET_PARAM(0);
701 inv_txfm_ = GET_PARAM(1);
702 tx_type_ = GET_PARAM(2);
703 bit_depth_ = GET_PARAM(3);
705 fwd_txfm_ref = fht16x16_ref;
706 inv_txfm_ref = iht16x16_ref;
707 mask_ = (1 << bit_depth_) - 1;
708 #if CONFIG_VP9_HIGHBITDEPTH
709 switch (bit_depth_) {
711 inv_txfm_ref = iht16x16_10;
714 inv_txfm_ref = iht16x16_12;
717 inv_txfm_ref = iht16x16_ref;
721 inv_txfm_ref = iht16x16_ref;
724 virtual void TearDown() { libvpx_test::ClearSystemState(); }
727 void RunFwdTxfm(int16_t *in, tran_low_t *out, int stride) {
728 fwd_txfm_(in, out, stride, tx_type_);
730 void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
731 inv_txfm_(out, dst, stride, tx_type_);
738 TEST_P(Trans16x16HT, AccuracyCheck) {
742 TEST_P(Trans16x16HT, CoeffCheck) {
746 TEST_P(Trans16x16HT, MemCheck) {
750 TEST_P(Trans16x16HT, QuantCheck) {
751 // The encoder skips any non-DC intra prediction modes,
752 // when the quantization step size goes beyond 988.
753 RunQuantCheck(429, 729);
756 class InvTrans16x16DCT
757 : public Trans16x16TestBase,
758 public ::testing::TestWithParam<Idct16x16Param> {
760 virtual ~InvTrans16x16DCT() {}
762 virtual void SetUp() {
763 ref_txfm_ = GET_PARAM(0);
764 inv_txfm_ = GET_PARAM(1);
765 thresh_ = GET_PARAM(2);
766 bit_depth_ = GET_PARAM(3);
768 mask_ = (1 << bit_depth_) - 1;
770 virtual void TearDown() { libvpx_test::ClearSystemState(); }
773 void RunFwdTxfm(int16_t * /*in*/, tran_low_t * /*out*/, int /*stride*/) {}
774 void RunInvTxfm(tran_low_t *out, uint8_t *dst, int stride) {
775 inv_txfm_(out, dst, stride);
783 TEST_P(InvTrans16x16DCT, CompareReference) {
784 CompareInvReference(ref_txfm_, thresh_);
787 class PartialTrans16x16Test
788 : public ::testing::TestWithParam<
789 std::tr1::tuple<FdctFunc, vpx_bit_depth_t> > {
791 virtual ~PartialTrans16x16Test() {}
792 virtual void SetUp() {
793 fwd_txfm_ = GET_PARAM(0);
794 bit_depth_ = GET_PARAM(1);
797 virtual void TearDown() { libvpx_test::ClearSystemState(); }
800 vpx_bit_depth_t bit_depth_;
804 TEST_P(PartialTrans16x16Test, Extremes) {
805 #if CONFIG_VP9_HIGHBITDEPTH
806 const int16_t maxval =
807 static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
809 const int16_t maxval = 255;
811 const int minval = -maxval;
812 DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
813 DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
815 for (int i = 0; i < kNumCoeffs; ++i) input[i] = maxval;
817 ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
818 EXPECT_EQ((maxval * kNumCoeffs) >> 1, output[0]);
820 for (int i = 0; i < kNumCoeffs; ++i) input[i] = minval;
822 ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
823 EXPECT_EQ((minval * kNumCoeffs) >> 1, output[0]);
826 TEST_P(PartialTrans16x16Test, Random) {
827 #if CONFIG_VP9_HIGHBITDEPTH
828 const int16_t maxval =
829 static_cast<int16_t>(clip_pixel_highbd(1 << 30, bit_depth_));
831 const int16_t maxval = 255;
833 DECLARE_ALIGNED(16, int16_t, input[kNumCoeffs]);
834 DECLARE_ALIGNED(16, tran_low_t, output[kNumCoeffs]);
835 ACMRandom rnd(ACMRandom::DeterministicSeed());
838 for (int i = 0; i < kNumCoeffs; ++i) {
839 const int val = (i & 1) ? -rnd(maxval + 1) : rnd(maxval + 1);
844 ASM_REGISTER_STATE_CHECK(fwd_txfm_(input, output, 16));
845 EXPECT_EQ(sum >> 1, output[0]);
848 using std::tr1::make_tuple;
850 #if CONFIG_VP9_HIGHBITDEPTH
851 INSTANTIATE_TEST_CASE_P(
854 make_tuple(&vpx_highbd_fdct16x16_c, &idct16x16_10, 0, VPX_BITS_10),
855 make_tuple(&vpx_highbd_fdct16x16_c, &idct16x16_12, 0, VPX_BITS_12),
856 make_tuple(&vpx_fdct16x16_c, &vpx_idct16x16_256_add_c, 0, VPX_BITS_8)));
858 INSTANTIATE_TEST_CASE_P(
861 make_tuple(&vpx_fdct16x16_c, &vpx_idct16x16_256_add_c, 0, VPX_BITS_8)));
862 #endif // CONFIG_VP9_HIGHBITDEPTH
864 #if CONFIG_VP9_HIGHBITDEPTH
865 INSTANTIATE_TEST_CASE_P(
868 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 0, VPX_BITS_10),
869 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 1, VPX_BITS_10),
870 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 2, VPX_BITS_10),
871 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_10, 3, VPX_BITS_10),
872 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 0, VPX_BITS_12),
873 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 1, VPX_BITS_12),
874 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 2, VPX_BITS_12),
875 make_tuple(&vp9_highbd_fht16x16_c, &iht16x16_12, 3, VPX_BITS_12),
876 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
877 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
878 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
879 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
880 INSTANTIATE_TEST_CASE_P(
881 C, PartialTrans16x16Test,
882 ::testing::Values(make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_8),
883 make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_10),
884 make_tuple(&vpx_highbd_fdct16x16_1_c, VPX_BITS_12)));
886 INSTANTIATE_TEST_CASE_P(
889 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
890 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
891 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
892 make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_c, 3, VPX_BITS_8)));
893 INSTANTIATE_TEST_CASE_P(C, PartialTrans16x16Test,
894 ::testing::Values(make_tuple(&vpx_fdct16x16_1_c,
896 #endif // CONFIG_VP9_HIGHBITDEPTH
898 #if HAVE_NEON_ASM && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
899 INSTANTIATE_TEST_CASE_P(
902 make_tuple(&vpx_fdct16x16_c,
903 &vpx_idct16x16_256_add_neon, 0, VPX_BITS_8)));
906 #if HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
907 INSTANTIATE_TEST_CASE_P(
910 make_tuple(&vpx_fdct16x16_sse2,
911 &vpx_idct16x16_256_add_sse2, 0, VPX_BITS_8)));
912 INSTANTIATE_TEST_CASE_P(
915 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 0,
917 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 1,
919 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 2,
921 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_sse2, 3,
923 INSTANTIATE_TEST_CASE_P(SSE2, PartialTrans16x16Test,
924 ::testing::Values(make_tuple(&vpx_fdct16x16_1_sse2,
926 #endif // HAVE_SSE2 && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
928 #if HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
929 INSTANTIATE_TEST_CASE_P(
932 make_tuple(&vpx_highbd_fdct16x16_sse2,
933 &idct16x16_10, 0, VPX_BITS_10),
934 make_tuple(&vpx_highbd_fdct16x16_c,
935 &idct16x16_256_add_10_sse2, 0, VPX_BITS_10),
936 make_tuple(&vpx_highbd_fdct16x16_sse2,
937 &idct16x16_12, 0, VPX_BITS_12),
938 make_tuple(&vpx_highbd_fdct16x16_c,
939 &idct16x16_256_add_12_sse2, 0, VPX_BITS_12),
940 make_tuple(&vpx_fdct16x16_sse2,
941 &vpx_idct16x16_256_add_c, 0, VPX_BITS_8)));
942 INSTANTIATE_TEST_CASE_P(
945 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 0, VPX_BITS_8),
946 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 1, VPX_BITS_8),
947 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 2, VPX_BITS_8),
948 make_tuple(&vp9_fht16x16_sse2, &vp9_iht16x16_256_add_c, 3,
950 // Optimizations take effect at a threshold of 3155, so we use a value close to
951 // that to test both branches.
952 INSTANTIATE_TEST_CASE_P(
953 SSE2, InvTrans16x16DCT,
955 make_tuple(&idct16x16_10_add_10_c,
956 &idct16x16_10_add_10_sse2, 3167, VPX_BITS_10),
957 make_tuple(&idct16x16_10,
958 &idct16x16_256_add_10_sse2, 3167, VPX_BITS_10),
959 make_tuple(&idct16x16_10_add_12_c,
960 &idct16x16_10_add_12_sse2, 3167, VPX_BITS_12),
961 make_tuple(&idct16x16_12,
962 &idct16x16_256_add_12_sse2, 3167, VPX_BITS_12)));
963 INSTANTIATE_TEST_CASE_P(SSE2, PartialTrans16x16Test,
964 ::testing::Values(make_tuple(&vpx_fdct16x16_1_sse2,
966 #endif // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
968 #if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
969 INSTANTIATE_TEST_CASE_P(
972 make_tuple(&vpx_fdct16x16_msa,
973 &vpx_idct16x16_256_add_msa, 0, VPX_BITS_8)));
974 INSTANTIATE_TEST_CASE_P(
977 make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 0, VPX_BITS_8),
978 make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 1, VPX_BITS_8),
979 make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 2, VPX_BITS_8),
980 make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 3,
982 INSTANTIATE_TEST_CASE_P(MSA, PartialTrans16x16Test,
983 ::testing::Values(make_tuple(&vpx_fdct16x16_1_msa,
985 #endif // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE