FFmpeg: tests/checkasm/vp9dsp.c Source File

FFmpeg

[フレーム]

vp9dsp.c

Go to the documentation of this file.

1 /*

3 *

4 * This file is part of FFmpeg.

5 *

6 * FFmpeg is free software; you can redistribute it and/or modify

7 * it under the terms of the GNU General Public License as published by

8 * the Free Software Foundation; either version 2 of the License, or

9 * (at your option) any later version.

10 *

11 * FFmpeg is distributed in the hope that it will be useful,

12 * but WITHOUT ANY WARRANTY; without even the implied warranty of

13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the

14 * GNU General Public License for more details.

15 *

16 * You should have received a copy of the GNU General Public License along

17 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,

18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.

19 */

21 #include <math.h>

22 #include <string.h>

23 #include "checkasm.h"

24 #include "libavcodec/vp9data.h"

25 #include "libavcodec/vp9dsp.h"

26 #include "libavutil/common.h"

27 #include "libavutil/internal.h"

28 #include "libavutil/intreadwrite.h"

29 #include "libavutil/mathematics.h"

31 static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff };

32 #define SIZEOF_PIXEL ((bit_depth + 7) / 8)

34 #define randomize_buffers() \

35 do { \

36 uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \

37 int k; \

38 for (k = -4; k < SIZEOF_PIXEL * FFMAX(8, size); k += 4) { \

39 uint32_t r = rnd() & mask; \

40 AV_WN32A(a + k, r); \

41 } \

42 for (k = 0; k < size * SIZEOF_PIXEL; k += 4) { \

43 uint32_t r = rnd() & mask; \

44 AV_WN32A(l + k, r); \

45 } \

46 } while (0)

48 static void check_ipred(void)

49 {

50 LOCAL_ALIGNED_32(uint8_t, a_buf, [64 * 2]);

51 uint8_t *a = &a_buf[32 * 2];

52 LOCAL_ALIGNED_32(uint8_t, l, [32 * 2]);

53 LOCAL_ALIGNED_32(uint8_t, dst0, [32 * 32 * 2]);

54 LOCAL_ALIGNED_32(uint8_t, dst1, [32 * 32 * 2]);

55 VP9DSPContext dsp;

56 int tx, mode, bit_depth;

57 declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, void, uint8_t *dst, ptrdiff_t stride,

58 const uint8_t *left, const uint8_t *top);

59 static const char *const mode_names[N_INTRA_PRED_MODES] = {

60 [VERT_PRED] = "vert",

61 [HOR_PRED] = "hor",

62 [DC_PRED] = "dc",

63 [DIAG_DOWN_LEFT_PRED] = "diag_downleft",

64 [DIAG_DOWN_RIGHT_PRED] = "diag_downright",

65 [VERT_RIGHT_PRED] = "vert_right",

66 [HOR_DOWN_PRED] = "hor_down",

67 [VERT_LEFT_PRED] = "vert_left",

68 [HOR_UP_PRED] = "hor_up",

69 [TM_VP8_PRED] = "tm",

70 [LEFT_DC_PRED] = "dc_left",

71 [TOP_DC_PRED] = "dc_top",

72 [DC_128_PRED] = "dc_128",

73 [DC_127_PRED] = "dc_127",

74 [DC_129_PRED] = "dc_129",

75 };

77 for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {

78 ff_vp9dsp_init(&dsp, bit_depth, 0);

79 for (tx = 0; tx < 4; tx++) {

80 int size = 4 << tx;

82 for (mode = 0; mode < N_INTRA_PRED_MODES; mode++) {

83 if (check_func(dsp.intra_pred[tx][mode], "vp9_%s_%dx%d_%dbpp",

84 mode_names[mode], size, size, bit_depth)) {

85 randomize_buffers();

86 call_ref(dst0, size * SIZEOF_PIXEL, l, a);

87 call_new(dst1, size * SIZEOF_PIXEL, l, a);

88 if (memcmp(dst0, dst1, size * size * SIZEOF_PIXEL))

89 fail();

90 bench_new(dst1, size * SIZEOF_PIXEL,l, a);

91 }

92 }

93 }

94 }

95 report("ipred");

96 }

98 #undef randomize_buffers

100 #define randomize_buffers() \

101 do { \

102 uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \

103 for (y = 0; y < sz; y++) { \

104 for (x = 0; x < sz * SIZEOF_PIXEL; x += 4) { \

105 uint32_t r = rnd() & mask; \

106 AV_WN32A(dst + y * sz * SIZEOF_PIXEL + x, r); \

107 AV_WN32A(src + y * sz * SIZEOF_PIXEL + x, rnd() & mask); \

108 } \

109 for (x = 0; x < sz; x++) { \

110 if (bit_depth == 8) { \

111 coef[y * sz + x] = src[y * sz + x] - dst[y * sz + x]; \

112 } else { \

113 ((int32_t *) coef)[y * sz + x] = \

114 ((uint16_t *) src)[y * sz + x] - \

115 ((uint16_t *) dst)[y * sz + x]; \

116 } \

117 } \

118 } \

119 } while(0)

120

121 // wht function copied from libvpx

122 static void fwht_1d(double *out, const double *in, int sz)

123 {

124 double t0 = in[0] + in[1];

125 double t3 = in[3] - in[2];

126 double t4 = trunc((t0 - t3) * 0.5);

127 double t1 = t4 - in[1];

128 double t2 = t4 - in[2];

129

130 out[0] = t0 - t2;

131 out[1] = t2;

132 out[2] = t3 + t1;

133 out[3] = t1;

134 }

135

136 // standard DCT-II

137 static void fdct_1d(double *out, const double *in, int sz)

138 {

139 int k, n;

140

141 for (k = 0; k < sz; k++) {

142 out[k] = 0.0;

143 for (n = 0; n < sz; n++)

144 out[k] += in[n] * cos(M_PI * (2 * n + 1) * k / (sz * 2.0));

145 }

146 out[0] *= M_SQRT1_2;

147 }

148

149 // see "Towards jointly optimal spatial prediction and adaptive transform in

150 // video/image coding", by J. Han, A. Saxena, and K. Rose

151 // IEEE Proc. ICASSP, pp. 726-729, Mar. 2010.

152 static void fadst4_1d(double *out, const double *in, int sz)

153 {

154 int k, n;

155

156 for (k = 0; k < sz; k++) {

157 out[k] = 0.0;

158 for (n = 0; n < sz; n++)

159 out[k] += in[n] * sin(M_PI * (n + 1) * (2 * k + 1) / (sz * 2.0 + 1.0));

160 }

161 }

162

163 // see "A Butterfly Structured Design of The Hybrid Transform Coding Scheme",

164 // by Jingning Han, Yaowu Xu, and Debargha Mukherjee

165 // http://static.googleusercontent.com/media/research.google.com/en//pubs/archive/41418.pdf

166 static void fadst_1d(double *out, const double *in, int sz)

167 {

168 int k, n;

169

170 for (k = 0; k < sz; k++) {

171 out[k] = 0.0;

172 for (n = 0; n < sz; n++)

173 out[k] += in[n] * sin(M_PI * (2 * n + 1) * (2 * k + 1) / (sz * 4.0));

174 }

175 }

176

177 typedef void (*ftx1d_fn)(double *out, const double *in, int sz);

178 static void ftx_2d(double *out, const double *in, enum TxfmMode tx,

179 enum TxfmType txtp, int sz)

180 {

181 static const double scaling_factors[5][4] = {

182 { 4.0, 16.0 * M_SQRT1_2 / 3.0, 16.0 * M_SQRT1_2 / 3.0, 32.0 / 9.0 },

183 { 2.0, 2.0, 2.0, 2.0 },

184 { 1.0, 1.0, 1.0, 1.0 },

185 { 0.25 },

186 { 4.0 }

187 };

188 static const ftx1d_fn ftx1d_tbl[5][4][2] = {

189 {

190 { fdct_1d, fdct_1d },

191 { fadst4_1d, fdct_1d },

192 { fdct_1d, fadst4_1d },

193 { fadst4_1d, fadst4_1d },

194 }, {

195 { fdct_1d, fdct_1d },

196 { fadst_1d, fdct_1d },

197 { fdct_1d, fadst_1d },

198 { fadst_1d, fadst_1d },

199 }, {

200 { fdct_1d, fdct_1d },

201 { fadst_1d, fdct_1d },

202 { fdct_1d, fadst_1d },

203 { fadst_1d, fadst_1d },

204 }, {

205 { fdct_1d, fdct_1d },

206 }, {

207 { fwht_1d, fwht_1d },

208 },

209 };

210 double temp[1024];

211 double scaling_factor = scaling_factors[tx][txtp];

212 int i, j;

213

214 // cols

215 for (i = 0; i < sz; ++i) {

216 double temp_out[32];

217

218 ftx1d_tbl[tx][txtp][0](temp_out, &in[i * sz], sz);

219 // scale and transpose

220 for (j = 0; j < sz; ++j)

221 temp[j * sz + i] = temp_out[j] * scaling_factor;

222 }

223

224 // rows

225 for (i = 0; i < sz; i++)

226 ftx1d_tbl[tx][txtp][1](&out[i * sz], &temp[i * sz], sz);

227 }

228

229 static void ftx(int16_t *buf, enum TxfmMode tx,

230 enum TxfmType txtp, int sz, int bit_depth)

231 {

232 double ind[1024], outd[1024];

233 int n;

234

235 emms_c();

236 for (n = 0; n < sz * sz; n++) {

237 if (bit_depth == 8)

238 ind[n] = buf[n];

239 else

240 ind[n] = ((int32_t *) buf)[n];

241 }

242 ftx_2d(outd, ind, tx, txtp, sz);

243 for (n = 0; n < sz * sz; n++) {

244 if (bit_depth == 8)

245 buf[n] = lrint(outd[n]);

246 else

247 ((int32_t *) buf)[n] = lrint(outd[n]);

248 }

249 }

250

251 static int copy_subcoefs(int16_t *out, const int16_t *in, enum TxfmMode tx,

252 enum TxfmType txtp, int sz, int sub, int bit_depth)

253 {

254 // copy the topleft coefficients such that the return value (being the

255 // coefficient scantable index for the eob token) guarantees that only

256 // the topleft $sub out of $sz (where $sz >= $sub) coefficients in both

257 // dimensions are non-zero. This leads to braching to specific optimized

258 // simd versions (e.g. dc-only) so that we get full asm coverage in this

259 // test

260

261 int n;

262 const int16_t *scan = vp9_scans[tx][txtp];

263 int eob;

264

265 for (n = 0; n < sz * sz; n++) {

266 int rc = scan[n], rcx = rc % sz, rcy = rc / sz;

267

268 // find eob for this sub-idct

269 if (rcx >= sub || rcy >= sub)

270 break;

271

272 // copy coef

273 if (bit_depth == 8) {

274 out[rc] = in[rc];

275 } else {

276 AV_COPY32(&out[rc * 2], &in[rc * 2]);

277 }

278 }

279

280 eob = n;

281

282 for (; n < sz * sz; n++) {

283 int rc = scan[n];

284

285 // zero

286 if (bit_depth == 8) {

287 out[rc] = 0;

288 } else {

289 AV_ZERO32(&out[rc * 2]);

290 }

291 }

292

293 return eob;

294 }

295

296 static int iszero(const int16_t *c, int sz)

297 {

298 int n;

299

300 for (n = 0; n < sz / sizeof(int16_t); n += 2)

301 if (AV_RN32A(&c[n]))

302 return 0;

303

304 return 1;

305 }

306

307 #define SIZEOF_COEF (2 * ((bit_depth + 7) / 8))

308

309 static void check_itxfm(void)

310 {

311 LOCAL_ALIGNED_32(uint8_t, src, [32 * 32 * 2]);

312 LOCAL_ALIGNED_32(uint8_t, dst, [32 * 32 * 2]);

313 LOCAL_ALIGNED_32(uint8_t, dst0, [32 * 32 * 2]);

314 LOCAL_ALIGNED_32(uint8_t, dst1, [32 * 32 * 2]);

315 LOCAL_ALIGNED_32(int16_t, coef, [32 * 32 * 2]);

316 LOCAL_ALIGNED_32(int16_t, subcoef0, [32 * 32 * 2]);

317 LOCAL_ALIGNED_32(int16_t, subcoef1, [32 * 32 * 2]);

318 declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, void, uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob);

319 VP9DSPContext dsp;

320 int y, x, tx, txtp, bit_depth, sub;

321 static const char *const txtp_types[N_TXFM_TYPES] = {

322 [DCT_DCT] = "dct_dct", [DCT_ADST] = "adst_dct",

323 [ADST_DCT] = "dct_adst", [ADST_ADST] = "adst_adst"

324 };

325

326 for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {

327 ff_vp9dsp_init(&dsp, bit_depth, 0);

328

329 for (tx = TX_4X4; tx <= N_TXFM_SIZES /* 4 = lossless */; tx++) {

330 int sz = 4 << (tx & 3);

331 int n_txtps = tx < TX_32X32 ? N_TXFM_TYPES : 1;

332

333 for (txtp = 0; txtp < n_txtps; txtp++) {

334 if (check_func(dsp.itxfm_add[tx][txtp], "vp9_inv_%s_%dx%d_add_%d",

335 tx == 4 ? "wht_wht" : txtp_types[txtp], sz, sz,

336 bit_depth)) {

337 randomize_buffers();

338 ftx(coef, tx, txtp, sz, bit_depth);

339

340 for (sub = (txtp == 0) ? 1 : 2; sub <= sz; sub <<= 1) {

341 int eob;

342

343 if (sub < sz) {

344 eob = copy_subcoefs(subcoef0, coef, tx, txtp,

345 sz, sub, bit_depth);

346 } else {

347 eob = sz * sz;

348 memcpy(subcoef0, coef, sz * sz * SIZEOF_COEF);

349 }

350

351 memcpy(dst0, dst, sz * sz * SIZEOF_PIXEL);

352 memcpy(dst1, dst, sz * sz * SIZEOF_PIXEL);

353 memcpy(subcoef1, subcoef0, sz * sz * SIZEOF_COEF);

354 call_ref(dst0, sz * SIZEOF_PIXEL, subcoef0, eob);

355 call_new(dst1, sz * SIZEOF_PIXEL, subcoef1, eob);

356 if (memcmp(dst0, dst1, sz * sz * SIZEOF_PIXEL) ||

357 !iszero(subcoef0, sz * sz * SIZEOF_COEF) ||

358 !iszero(subcoef1, sz * sz * SIZEOF_COEF))

359 fail();

360 }

361 bench_new(dst, sz * SIZEOF_PIXEL, coef, sz * sz);

362 }

363 }

364 }

365 }

366 report("itxfm");

367 }

368

369 #undef randomize_buffers

370

371 #define setpx(a,b,c) \

372 do { \

373 if (SIZEOF_PIXEL == 1) { \

374 buf0[(a) + (b) * jstride] = av_clip_uint8(c); \

375 } else { \

376 ((uint16_t *)buf0)[(a) + (b) * jstride] = av_clip_uintp2(c, bit_depth); \

377 } \

378 } while (0)

379

380 // c can be an assignment and must not be put under ()

381 #define setdx(a,b,c,d) setpx(a,b,c-(d)+(rnd()%((d)*2+1)))

382 #define setsx(a,b,c,d) setdx(a,b,c,(d) << (bit_depth - 8))

383 static void randomize_loopfilter_buffers(int bidx, int lineoff, int str,

384 int bit_depth, int dir, const int *E,

385 const int *F, const int *H, const int *I,

386 uint8_t *buf0, uint8_t *buf1)

387 {

388 uint32_t mask = (1 << bit_depth) - 1;

389 int off = dir ? lineoff : lineoff * 16;

390 int istride = dir ? 1 : 16;

391 int jstride = dir ? str : 1;

392 int i, j;

393 for (i = 0; i < 2; i++) /* flat16 */ {

394 int idx = off + i * istride, p0, q0;

395 setpx(idx, 0, q0 = rnd() & mask);

396 setsx(idx, -1, p0 = q0, E[bidx] >> 2);

397 for (j = 1; j < 8; j++) {

398 setsx(idx, -1 - j, p0, F[bidx]);

399 setsx(idx, j, q0, F[bidx]);

400 }

401 }

402 for (i = 2; i < 4; i++) /* flat8 */ {

403 int idx = off + i * istride, p0, q0;

404 setpx(idx, 0, q0 = rnd() & mask);

405 setsx(idx, -1, p0 = q0, E[bidx] >> 2);

406 for (j = 1; j < 4; j++) {

407 setsx(idx, -1 - j, p0, F[bidx]);

408 setsx(idx, j, q0, F[bidx]);

409 }

410 for (j = 4; j < 8; j++) {

411 setpx(idx, -1 - j, rnd() & mask);

412 setpx(idx, j, rnd() & mask);

413 }

414 }

415 for (i = 4; i < 6; i++) /* regular */ {

416 int idx = off + i * istride, p2, p1, p0, q0, q1, q2;

417 setpx(idx, 0, q0 = rnd() & mask);

418 setsx(idx, 1, q1 = q0, I[bidx]);

419 setsx(idx, 2, q2 = q1, I[bidx]);

420 setsx(idx, 3, q2, I[bidx]);

421 setsx(idx, -1, p0 = q0, E[bidx] >> 2);

422 setsx(idx, -2, p1 = p0, I[bidx]);

423 setsx(idx, -3, p2 = p1, I[bidx]);

424 setsx(idx, -4, p2, I[bidx]);

425 for (j = 4; j < 8; j++) {

426 setpx(idx, -1 - j, rnd() & mask);

427 setpx(idx, j, rnd() & mask);

428 }

429 }

430 for (i = 6; i < 8; i++) /* off */ {

431 int idx = off + i * istride;

432 for (j = 0; j < 8; j++) {

433 setpx(idx, -1 - j, rnd() & mask);

434 setpx(idx, j, rnd() & mask);

435 }

436 }

437 }

438 #define randomize_buffers(bidx, lineoff, str) \

439 randomize_loopfilter_buffers(bidx, lineoff, str, bit_depth, dir, \

440 E, F, H, I, buf0, buf1)

441

442 static void check_loopfilter(void)

443 {

444 LOCAL_ALIGNED_32(uint8_t, base0, [32 + 16 * 16 * 2]);

445 LOCAL_ALIGNED_32(uint8_t, base1, [32 + 16 * 16 * 2]);

446 VP9DSPContext dsp;

447 int dir, wd, wd2, bit_depth;

448 static const char *const dir_name[2] = { "h", "v" };

449 static const int E[2] = { 20, 28 }, I[2] = { 10, 16 };

450 static const int H[2] = { 7, 11 }, F[2] = { 1, 1 };

451 declare_func(void, uint8_t *dst, ptrdiff_t stride, int E, int I, int H);

452

453 for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {

454 ff_vp9dsp_init(&dsp, bit_depth, 0);

455

456 for (dir = 0; dir < 2; dir++) {

457 int midoff = (dir ? 8 * 8 : 8) * SIZEOF_PIXEL;

458 int midoff_aligned = (dir ? 8 * 8 : 16) * SIZEOF_PIXEL;

459 uint8_t *buf0 = base0 + midoff_aligned;

460 uint8_t *buf1 = base1 + midoff_aligned;

461

462 for (wd = 0; wd < 3; wd++) {

463 // 4/8/16wd_8px

464 if (check_func(dsp.loop_filter_8[wd][dir],

465 "vp9_loop_filter_%s_%d_8_%dbpp",

466 dir_name[dir], 4 << wd, bit_depth)) {

467 randomize_buffers(0, 0, 8);

468 memcpy(buf1 - midoff, buf0 - midoff,

469 16 * 8 * SIZEOF_PIXEL);

470 call_ref(buf0, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]);

471 call_new(buf1, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]);

472 if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 8 * SIZEOF_PIXEL))

473 fail();

474 bench_new(buf1, 16 * SIZEOF_PIXEL >> dir, E[0], I[0], H[0]);

475 }

476 }

477

478 midoff = (dir ? 16 * 8 : 8) * SIZEOF_PIXEL;

479 midoff_aligned = (dir ? 16 * 8 : 16) * SIZEOF_PIXEL;

480

481 buf0 = base0 + midoff_aligned;

482 buf1 = base1 + midoff_aligned;

483

484 // 16wd_16px loopfilter

485 if (check_func(dsp.loop_filter_16[dir],

486 "vp9_loop_filter_%s_16_16_%dbpp",

487 dir_name[dir], bit_depth)) {

488 randomize_buffers(0, 0, 16);

489 randomize_buffers(0, 8, 16);

490 memcpy(buf1 - midoff, buf0 - midoff, 16 * 16 * SIZEOF_PIXEL);

491 call_ref(buf0, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]);

492 call_new(buf1, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]);

493 if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 16 * SIZEOF_PIXEL))

494 fail();

495 bench_new(buf1, 16 * SIZEOF_PIXEL, E[0], I[0], H[0]);

496 }

497

498 for (wd = 0; wd < 2; wd++) {

499 for (wd2 = 0; wd2 < 2; wd2++) {

500 // mix2 loopfilter

501 if (check_func(dsp.loop_filter_mix2[wd][wd2][dir],

502 "vp9_loop_filter_mix2_%s_%d%d_16_%dbpp",

503 dir_name[dir], 4 << wd, 4 << wd2, bit_depth)) {

504 randomize_buffers(0, 0, 16);

505 randomize_buffers(1, 8, 16);

506 memcpy(buf1 - midoff, buf0 - midoff, 16 * 16 * SIZEOF_PIXEL);

507 #define M(a) (((a)[1] << 8) | (a)[0])

508 call_ref(buf0, 16 * SIZEOF_PIXEL, M(E), M(I), M(H));

509 call_new(buf1, 16 * SIZEOF_PIXEL, M(E), M(I), M(H));

510 if (memcmp(buf0 - midoff, buf1 - midoff, 16 * 16 * SIZEOF_PIXEL))

511 fail();

512 bench_new(buf1, 16 * SIZEOF_PIXEL, M(E), M(I), M(H));

513 #undef M

514 }

515 }

516 }

517 }

518 }

519 report("loopfilter");

520 }

521

522 #undef setsx

523 #undef setpx

524 #undef setdx

525 #undef randomize_buffers

526

527 #define DST_BUF_SIZE (size * size * SIZEOF_PIXEL)

528 #define SRC_BUF_STRIDE 72

529 #define SRC_BUF_SIZE ((size + 7) * SRC_BUF_STRIDE * SIZEOF_PIXEL)

530 #define src (buf + 3 * SIZEOF_PIXEL * (SRC_BUF_STRIDE + 1))

531

532 #define randomize_buffers() \

533 do { \

534 uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \

535 int k; \

536 for (k = 0; k < SRC_BUF_SIZE; k += 4) { \

537 uint32_t r = rnd() & mask; \

538 AV_WN32A(buf + k, r); \

539 } \

540 if (op == 1) { \

541 for (k = 0; k < DST_BUF_SIZE; k += 4) { \

542 uint32_t r = rnd() & mask; \

543 AV_WN32A(dst0 + k, r); \

544 AV_WN32A(dst1 + k, r); \

545 } \

546 } \

547 } while (0)

548

549 static void check_mc(void)

550 {

551 LOCAL_ALIGNED_32(uint8_t, buf, [72 * 72 * 2]);

552 LOCAL_ALIGNED_32(uint8_t, dst0, [64 * 64 * 2]);

553 LOCAL_ALIGNED_32(uint8_t, dst1, [64 * 64 * 2]);

554 VP9DSPContext dsp;

555 int op, hsize, bit_depth, filter, dx, dy;

556 declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, void, uint8_t *dst, ptrdiff_t dst_stride,

557 const uint8_t *ref, ptrdiff_t ref_stride,

558 int h, int mx, int my);

559 static const char *const filter_names[4] = {

560 "8tap_smooth", "8tap_regular", "8tap_sharp", "bilin"

561 };

562 static const char *const subpel_names[2][2] = { { "", "h" }, { "v", "hv" } };

563 static const char *const op_names[2] = { "put", "avg" };

564 char str[256];

565

566 for (op = 0; op < 2; op++) {

567 for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {

568 ff_vp9dsp_init(&dsp, bit_depth, 0);

569 for (hsize = 0; hsize < 5; hsize++) {

570 int size = 64 >> hsize;

571

572 for (filter = 0; filter < 4; filter++) {

573 for (dx = 0; dx < 2; dx++) {

574 for (dy = 0; dy < 2; dy++) {

575 if (dx || dy) {

576 snprintf(str, sizeof(str),

577 "%s_%s_%d%s", op_names[op],

578 filter_names[filter], size,

579 subpel_names[dy][dx]);

580 } else {

581 snprintf(str, sizeof(str),

582 "%s%d", op_names[op], size);

583 }

584 if (check_func(dsp.mc[hsize][filter][op][dx][dy],

585 "vp9_%s_%dbpp", str, bit_depth)) {

586 int mx = dx ? 1 + (rnd() % 14) : 0;

587 int my = dy ? 1 + (rnd() % 14) : 0;

588 randomize_buffers();

589 call_ref(dst0, size * SIZEOF_PIXEL,

590 src, SRC_BUF_STRIDE * SIZEOF_PIXEL,

591 size, mx, my);

592 call_new(dst1, size * SIZEOF_PIXEL,

593 src, SRC_BUF_STRIDE * SIZEOF_PIXEL,

594 size, mx, my);

595 if (memcmp(dst0, dst1, DST_BUF_SIZE))

596 fail();

597

598 // simd implementations for each filter of subpel

599 // functions are identical

600 if (filter >= 1 && filter <= 2) continue;

601 // 10/12 bpp for bilin are identical

602 if (bit_depth == 12 && filter == 3) continue;

603

604 bench_new(dst1, size * SIZEOF_PIXEL,

605 src, SRC_BUF_STRIDE * SIZEOF_PIXEL,

606 size, mx, my);

607 }

608 }

609 }

610 }

611 }

612 }

613 }

614 report("mc");

615 }

616

617 void checkasm_check_vp9dsp(void)

618 {

619 check_ipred();

620 check_itxfm();

621 check_loopfilter();

622 check_mc();

623 }

HOR_PRED

Definition: vp9.h:83

TOP_DC_PRED

Definition: vp9.h:93

Definition: vp9dsp_template.c:2093

SRC_BUF_STRIDE

#define SRC_BUF_STRIDE

Definition: vp9dsp.c:528

check_loopfilter

static void check_loopfilter(void)

Definition: vp9dsp.c:442

N_TXFM_TYPES

Definition: vp9.h:78

temp

else temp

Definition: vf_mcdeint.c:259

M_SQRT1_2

#define M_SQRT1_2

Definition: mathematics.h:52

HOR_DOWN_PRED

Definition: vp9.h:88

VERT_LEFT_PRED

Definition: vp9.h:89

copy_subcoefs

static int copy_subcoefs(int16_t *out, const int16_t *in, enum TxfmMode tx, enum TxfmType txtp, int sz, int sub, int bit_depth)

Definition: vp9dsp.c:251

VP9DSPContext::intra_pred

void(* intra_pred[N_TXFM_SIZES][N_INTRA_PRED_MODES])(uint8_t *dst, ptrdiff_t stride, const uint8_t *left, const uint8_t *top)

Definition: vp9dsp.h:51

static const uint8_t q1[256]

Definition: twofish.c:96

setpx

#define setpx(a, b, c)

Definition: vp9dsp.c:371

AV_COPY32

#define AV_COPY32(d, s)

Definition: intreadwrite.h:586

setsx

#define setsx(a, b, c, d)

Definition: vp9dsp.c:382

report

#define report

Definition: checkasm.h:83

SIZEOF_COEF

#define SIZEOF_COEF

Definition: vp9dsp.c:307

mathematics.h

AV_RN32A

#define AV_RN32A(p)

Definition: intreadwrite.h:526

VP9DSPContext::mc

vp9_mc_func mc[5][4][2][2][2]

Definition: vp9dsp.h:114

DC_127_PRED

Definition: vp9.h:95

uint8_t

Definition: audio_convert.c:194

mode

Definition: f_perms.c:27

#define H

Definition: swscale-test.c:353

randomize_loopfilter_buffers

static void randomize_loopfilter_buffers(int bidx, int lineoff, int str, int bit_depth, int dir, const int *E, const int *F, const int *H, const int *I, uint8_t *buf0, uint8_t *buf1)

Definition: vp9dsp.c:383

TxfmType

Definition: vp9.h:73

fadst4_1d

static void fadst4_1d(double *out, const double *in, int sz)

Definition: vp9dsp.c:152

VERT_PRED

Definition: vp9.h:82

AV_CPU_FLAG_MMXEXT

#define AV_CPU_FLAG_MMXEXT

SSE integer functions or AMD MMX ext.

Definition: cpu.h:30

filter

static void filter(int16_t *output, ptrdiff_t out_stride, int16_t *low, ptrdiff_t low_stride, int16_t *high, ptrdiff_t high_stride, int len, uint8_t clip)

Definition: cfhd.c:82

N_TXFM_SIZES

Definition: vp9.h:68

#define t0

Definition: regdef.h:28

ftx

static void ftx(int16_t *buf, enum TxfmMode tx, enum TxfmType txtp, int sz, int bit_depth)

Definition: vp9dsp.c:229

DST_BUF_SIZE

#define DST_BUF_SIZE

Definition: vp9dsp.c:527

size

ptrdiff_t size

Definition: opengl_enc.c:101

DCT_DCT

Definition: vp9.h:74

fwht_1d

static void fwht_1d(double *out, const double *in, int sz)

Definition: vp9dsp.c:122

ff_vp9dsp_init

av_cold void ff_vp9dsp_init(VP9DSPContext *dsp, int bpp, int bitexact)

Definition: vp9dsp.c:28

TX_4X4

Definition: vp9.h:64

iszero

static int iszero(const int16_t *c, int sz)

Definition: vp9dsp.c:296

DC_129_PRED

Definition: vp9.h:96

checkasm.h

mask

static const uint16_t mask[17]

Definition: lzw.c:38

#define F

Definition: af_aformat.c:50

VP9DSPContext

Definition: vp9dsp.h:39

DCT_ADST

Definition: vp9.h:75

DC_128_PRED

Definition: vp9.h:94

declare_func

#define declare_func(ret,...)

Definition: checkasm.h:76

ADST_ADST

Definition: vp9.h:77

#define t1

Definition: regdef.h:29

fadst_1d

static void fadst_1d(double *out, const double *in, int sz)

Definition: vp9dsp.c:166

VERT_RIGHT_PRED

Definition: vp9.h:87

TxfmMode

Definition: vp9.h:63

TM_VP8_PRED

Definition: vp9.h:91

#define t3

Definition: regdef.h:31

fail

#define fail()

Definition: checkasm.h:80

static const uint8_t q0[256]

Definition: twofish.c:77

internal.h

common internal API header

vp9dsp.h

#define E

Definition: avdct.c:32

HOR_UP_PRED

Definition: vp9.h:90

intreadwrite.h

void

typedef void(APIENTRY *FF_PFNGLACTIVETEXTUREPROC)(GLenum texture)

trunc

static av_always_inline av_const double trunc(double x)

Definition: libm.h:458

VP9DSPContext::loop_filter_16

void(* loop_filter_16[2])(uint8_t *dst, ptrdiff_t stride, int mb_lim, int lim, int hev_thr)

Definition: vp9dsp.h:88

int32_t

Definition: audio_convert.c:194

int n

Definition: avisynth_c.h:547

SIZEOF_PIXEL

#define SIZEOF_PIXEL

Definition: vp9dsp.c:32

ftx1d_fn

void(* ftx1d_fn)(double *out, const double *in, int sz)

Definition: vp9dsp.c:177

src

#define src

Definition: vp9dsp.c:530

declare_func_emms

#define declare_func_emms(cpu_flags, ret,...)

Definition: checkasm.h:77

out

FILE * out

Definition: movenc-test.c:54

N_INTRA_PRED_MODES

Definition: vp9.h:97

call_ref

#define call_ref(...)

Definition: checkasm.h:86

check_mc

static void check_mc(void)

Definition: vp9dsp.c:549

VP9DSPContext::loop_filter_mix2

void(* loop_filter_mix2[2][2][2])(uint8_t *dst, ptrdiff_t stride, int mb_lim, int lim, int hev_thr)

Definition: vp9dsp.h:102

VP9DSPContext::loop_filter_8

void(* loop_filter_8[3][2])(uint8_t *dst, ptrdiff_t stride, int mb_lim, int lim, int hev_thr)

Definition: vp9dsp.h:80

DIAG_DOWN_RIGHT_PRED

Definition: vp9.h:86

AV_CPU_FLAG_MMX

#define AV_CPU_FLAG_MMX

standard MMX

Definition: cpu.h:29

uint8_t pi<< 24) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8, uint8_t,(*(constuint8_t *) pi-0x80)*(1.0f/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8, uint8_t,(*(constuint8_t *) pi-0x80)*(1.0/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16, int16_t,(*(constint16_t *) pi >>8)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, int16_t,*(constint16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, int16_t,*(constint16_t *) pi *(1.0/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32, int32_t,(*(constint32_t *) pi >>24)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, int32_t,*(constint32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, int32_t,*(constint32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, float, av_clip_uint8(lrintf(*(constfloat *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, float, av_clip_int16(lrintf(*(constfloat *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, float, av_clipl_int32(llrintf(*(constfloat *) pi *(1U<< 31)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, double, av_clip_uint8(lrint(*(constdouble *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, double, av_clip_int16(lrint(*(constdouble *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, double, av_clipl_int32(llrint(*(constdouble *) pi *(1U<< 31))))#defineSET_CONV_FUNC_GROUP(ofmt, ifmt) staticvoidset_generic_function(AudioConvert *ac){}voidff_audio_convert_free(AudioConvert **ac){if(!*ac) return;ff_dither_free(&(*ac) ->dc);av_freep(ac);}AudioConvert *ff_audio_convert_alloc(AVAudioResampleContext *avr, enumAVSampleFormatout_fmt, enumAVSampleFormatin_fmt, intchannels, intsample_rate, intapply_map){AudioConvert *ac;intin_planar, out_planar;ac=av_mallocz(sizeof(*ac));if(!ac) returnNULL;ac->avr=avr;ac->out_fmt=out_fmt;ac->in_fmt=in_fmt;ac->channels=channels;ac->apply_map=apply_map;if(avr->dither_method!=AV_RESAMPLE_DITHER_NONE &&av_get_packed_sample_fmt(out_fmt)==AV_SAMPLE_FMT_S16 &&av_get_bytes_per_sample(in_fmt)>2){ac->dc=ff_dither_alloc(avr, out_fmt, in_fmt, channels, sample_rate, apply_map);if(!ac->dc){av_free(ac);returnNULL;}returnac;}in_planar=ff_sample_fmt_is_planar(in_fmt, channels);out_planar=ff_sample_fmt_is_planar(out_fmt, channels);if(in_planar==out_planar){ac->func_type=CONV_FUNC_TYPE_FLAT;ac->planes=in_planar?ac->channels:1;}elseif(in_planar) ac->func_type=CONV_FUNC_TYPE_INTERLEAVE;elseac->func_type=CONV_FUNC_TYPE_DEINTERLEAVE;set_generic_function(ac);if(ARCH_AARCH64) ff_audio_convert_init_aarch64(ac);if(ARCH_ARM) ff_audio_convert_init_arm(ac);if(ARCH_X86) ff_audio_convert_init_x86(ac);returnac;}intff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in){intuse_generic=1;intlen=in->nb_samples;intp;if(ac->dc){av_log(ac->avr, AV_LOG_TRACE,"%dsamples-audio_convert:%sto%s(dithered)\n", len, av_get_sample_fmt_name(ac->in_fmt), av_get_sample_fmt_name(ac->out_fmt));returnff_convert_dither(ac-> in

Definition: audio_convert.c:194

buf

void * buf

Definition: avisynth_c.h:553

pixel_mask

static const uint32_t pixel_mask[3]

Definition: vp9dsp.c:31

check_func

#define check_func(func,...)

Definition: checkasm.h:72

vp9_scans

static const int16_t *const vp9_scans[5][4]