| 1 | /***************************************************************************** |
| 2 | * Copyright (C) 2013 x265 project |
| 3 | * |
| 4 | * Authors: Steve Borho <steve@borho.org> |
| 5 | * |
| 6 | * This program 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 | * This program 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 |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. |
| 19 | * |
| 20 | * This program is also available under a commercial proprietary license. |
| 21 | * For more information, contact us at license @ x265.com. |
| 22 | *****************************************************************************/ |
| 23 | |
| 24 | #include "pixelharness.h" |
| 25 | #include "primitives.h" |
| 26 | |
| 27 | using namespace x265; |
| 28 | |
| 29 | PixelHarness::PixelHarness() |
| 30 | { |
| 31 | /* [0] --- Random values |
| 32 | * [1] --- Minimum |
| 33 | * [2] --- Maximum */ |
| 34 | for (int i = 0; i < BUFFSIZE; i++) |
| 35 | { |
| 36 | pixel_test_buff[0][i] = rand() % PIXEL_MAX; |
| 37 | short_test_buff[0][i] = (rand() % (2 * SMAX + 1)) - SMAX - 1; // max(SHORT_MIN, min(rand(), SMAX)); |
| 38 | short_test_buff1[0][i] = rand() & PIXEL_MAX; // For block copy only |
| 39 | short_test_buff2[0][i] = rand() % 16383; // for addAvg |
| 40 | int_test_buff[0][i] = rand() % SHORT_MAX; |
| 41 | ushort_test_buff[0][i] = rand() % ((1 << 16) - 1); |
| 42 | uchar_test_buff[0][i] = rand() % ((1 << 8) - 1); |
| 43 | |
| 44 | pixel_test_buff[1][i] = PIXEL_MIN; |
| 45 | short_test_buff[1][i] = SMIN; |
| 46 | short_test_buff1[1][i] = PIXEL_MIN; |
| 47 | short_test_buff2[1][i] = -16384; |
| 48 | int_test_buff[1][i] = SHORT_MIN; |
| 49 | ushort_test_buff[1][i] = PIXEL_MIN; |
| 50 | uchar_test_buff[1][i] = PIXEL_MIN; |
| 51 | |
| 52 | pixel_test_buff[2][i] = PIXEL_MAX; |
| 53 | short_test_buff[2][i] = SMAX; |
| 54 | short_test_buff1[2][i] = PIXEL_MAX; |
| 55 | short_test_buff2[2][i] = 16383; |
| 56 | int_test_buff[2][i] = SHORT_MAX; |
| 57 | ushort_test_buff[2][i] = ((1 << 16) - 1); |
| 58 | uchar_test_buff[2][i] = 255; |
| 59 | |
| 60 | pbuf1[i] = rand() & PIXEL_MAX; |
| 61 | pbuf2[i] = rand() & PIXEL_MAX; |
| 62 | pbuf3[i] = rand() & PIXEL_MAX; |
| 63 | pbuf4[i] = rand() & PIXEL_MAX; |
| 64 | |
| 65 | sbuf1[i] = (rand() % (2 * SMAX + 1)) - SMAX - 1; //max(SHORT_MIN, min(rand(), SMAX)); |
| 66 | sbuf2[i] = (rand() % (2 * SMAX + 1)) - SMAX - 1; //max(SHORT_MIN, min(rand(), SMAX)); |
| 67 | ibuf1[i] = (rand() % (2 * SMAX + 1)) - SMAX - 1; |
| 68 | psbuf1[i] = (rand() % 65) - 32; // range is between -32 to 32 |
| 69 | sbuf3[i] = rand() % PIXEL_MAX; // for blockcopy only |
| 70 | } |
| 71 | } |
| 72 | |
| 73 | bool PixelHarness::check_pixelcmp(pixelcmp_t ref, pixelcmp_t opt) |
| 74 | { |
| 75 | int j = 0; |
| 76 | intptr_t stride = STRIDE; |
| 77 | |
| 78 | for (int i = 0; i < ITERS; i++) |
| 79 | { |
| 80 | int index1 = rand() % TEST_CASES; |
| 81 | int index2 = rand() % TEST_CASES; |
| 82 | int vres = (int)checked(opt, pixel_test_buff[index1], stride, pixel_test_buff[index2] + j, stride); |
| 83 | int cres = ref(pixel_test_buff[index1], stride, pixel_test_buff[index2] + j, stride); |
| 84 | if (vres != cres) |
| 85 | return false; |
| 86 | |
| 87 | reportfail(); |
| 88 | j += INCR; |
| 89 | } |
| 90 | |
| 91 | return true; |
| 92 | } |
| 93 | |
| 94 | bool PixelHarness::check_pixelcmp_sp(pixelcmp_sp_t ref, pixelcmp_sp_t opt) |
| 95 | { |
| 96 | int j = 0; |
| 97 | intptr_t stride = STRIDE; |
| 98 | |
| 99 | for (int i = 0; i < ITERS; i++) |
| 100 | { |
| 101 | int index1 = rand() % TEST_CASES; |
| 102 | int index2 = rand() % TEST_CASES; |
| 103 | int vres = (int)checked(opt, short_test_buff[index1], stride, pixel_test_buff[index2] + j, stride); |
| 104 | int cres = ref(short_test_buff[index1], stride, pixel_test_buff[index2] + j, stride); |
| 105 | if (vres != cres) |
| 106 | return false; |
| 107 | |
| 108 | reportfail(); |
| 109 | j += INCR; |
| 110 | } |
| 111 | |
| 112 | return true; |
| 113 | } |
| 114 | |
| 115 | bool PixelHarness::check_pixelcmp_ss(pixelcmp_ss_t ref, pixelcmp_ss_t opt) |
| 116 | { |
| 117 | int j = 0; |
| 118 | intptr_t stride = STRIDE; |
| 119 | |
| 120 | for (int i = 0; i < ITERS; i++) |
| 121 | { |
| 122 | int index1 = rand() % TEST_CASES; |
| 123 | int index2 = rand() % TEST_CASES; |
| 124 | int vres = (int)checked(opt, short_test_buff[index1], stride, short_test_buff[index2] + j, stride); |
| 125 | int cres = ref(short_test_buff[index1], stride, short_test_buff[index2] + j, stride); |
| 126 | if (vres != cres) |
| 127 | return false; |
| 128 | |
| 129 | reportfail(); |
| 130 | j += INCR; |
| 131 | } |
| 132 | |
| 133 | return true; |
| 134 | } |
| 135 | |
| 136 | bool PixelHarness::check_pixelcmp_x3(pixelcmp_x3_t ref, pixelcmp_x3_t opt) |
| 137 | { |
| 138 | ALIGN_VAR_16(int, cres[16]); |
| 139 | ALIGN_VAR_16(int, vres[16]); |
| 140 | int j = 0; |
| 141 | intptr_t stride = FENC_STRIDE - 5; |
| 142 | for (int i = 0; i < ITERS; i++) |
| 143 | { |
| 144 | int index1 = rand() % TEST_CASES; |
| 145 | int index2 = rand() % TEST_CASES; |
| 146 | checked(opt, pixel_test_buff[index1], |
| 147 | pixel_test_buff[index2] + j, |
| 148 | pixel_test_buff[index2] + j + 1, |
| 149 | pixel_test_buff[index2] + j + 2, stride, &vres[0]); |
| 150 | ref(pixel_test_buff[index1], |
| 151 | pixel_test_buff[index2] + j, |
| 152 | pixel_test_buff[index2] + j + 1, |
| 153 | pixel_test_buff[index2] + j + 2, stride, &cres[0]); |
| 154 | if ((vres[0] != cres[0]) || ((vres[1] != cres[1])) || ((vres[2] != cres[2]))) |
| 155 | return false; |
| 156 | |
| 157 | reportfail(); |
| 158 | j += INCR; |
| 159 | } |
| 160 | |
| 161 | return true; |
| 162 | } |
| 163 | |
| 164 | bool PixelHarness::check_pixelcmp_x4(pixelcmp_x4_t ref, pixelcmp_x4_t opt) |
| 165 | { |
| 166 | ALIGN_VAR_16(int, cres[16]); |
| 167 | ALIGN_VAR_16(int, vres[16]); |
| 168 | int j = 0; |
| 169 | intptr_t stride = FENC_STRIDE - 5; |
| 170 | for (int i = 0; i < ITERS; i++) |
| 171 | { |
| 172 | int index1 = rand() % TEST_CASES; |
| 173 | int index2 = rand() % TEST_CASES; |
| 174 | checked(opt, pixel_test_buff[index1], |
| 175 | pixel_test_buff[index2] + j, |
| 176 | pixel_test_buff[index2] + j + 1, |
| 177 | pixel_test_buff[index2] + j + 2, |
| 178 | pixel_test_buff[index2] + j + 3, stride, &vres[0]); |
| 179 | ref(pixel_test_buff[index1], |
| 180 | pixel_test_buff[index2] + j, |
| 181 | pixel_test_buff[index2] + j + 1, |
| 182 | pixel_test_buff[index2] + j + 2, |
| 183 | pixel_test_buff[index2] + j + 3, stride, &cres[0]); |
| 184 | |
| 185 | if ((vres[0] != cres[0]) || ((vres[1] != cres[1])) || ((vres[2] != cres[2])) || ((vres[3] != cres[3]))) |
| 186 | return false; |
| 187 | |
| 188 | reportfail(); |
| 189 | j += INCR; |
| 190 | } |
| 191 | |
| 192 | return true; |
| 193 | } |
| 194 | |
| 195 | bool PixelHarness::check_calresidual(calcresidual_t ref, calcresidual_t opt) |
| 196 | { |
| 197 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 198 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 199 | memset(ref_dest, 0, 64 * 64 * sizeof(int16_t)); |
| 200 | memset(opt_dest, 0, 64 * 64 * sizeof(int16_t)); |
| 201 | |
| 202 | int j = 0; |
| 203 | intptr_t stride = STRIDE; |
| 204 | for (int i = 0; i < ITERS; i++) |
| 205 | { |
| 206 | int index = i % TEST_CASES; |
| 207 | checked(opt, pbuf1 + j, pixel_test_buff[index] + j, opt_dest, stride); |
| 208 | ref(pbuf1 + j, pixel_test_buff[index] + j, ref_dest, stride); |
| 209 | |
| 210 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 211 | return false; |
| 212 | |
| 213 | reportfail(); |
| 214 | j += INCR; |
| 215 | } |
| 216 | |
| 217 | return true; |
| 218 | } |
| 219 | |
| 220 | bool PixelHarness::check_ssd_s(pixel_ssd_s_t ref, pixel_ssd_s_t opt) |
| 221 | { |
| 222 | int j = 0; |
| 223 | for (int i = 0; i < ITERS; i++) |
| 224 | { |
| 225 | // NOTE: stride must be multiple of 16, because minimum block is 4x4 |
| 226 | int stride = (STRIDE + (rand() % STRIDE)) & ~15; |
| 227 | int cres = ref(sbuf1 + j, stride); |
| 228 | int vres = (int)checked(opt, sbuf1 + j, (intptr_t)stride); |
| 229 | |
| 230 | if (cres != vres) |
| 231 | { |
| 232 | return false; |
| 233 | } |
| 234 | |
| 235 | reportfail(); |
| 236 | j += INCR; |
| 237 | } |
| 238 | |
| 239 | return true; |
| 240 | } |
| 241 | |
| 242 | bool PixelHarness::check_weightp(weightp_sp_t ref, weightp_sp_t opt) |
| 243 | { |
| 244 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 245 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 246 | |
| 247 | memset(ref_dest, 0, 64 * 64 * sizeof(pixel)); |
| 248 | memset(opt_dest, 0, 64 * 64 * sizeof(pixel)); |
| 249 | int j = 0; |
| 250 | int width = 2 * (rand() % 32 + 1); |
| 251 | int height = 8; |
| 252 | int w0 = rand() % 128; |
| 253 | int shift = rand() % 15; |
| 254 | int round = shift ? (1 << (shift - 1)) : 0; |
| 255 | int offset = (rand() % 256) - 128; |
| 256 | intptr_t stride = 64; |
| 257 | for (int i = 0; i < ITERS; i++) |
| 258 | { |
| 259 | int index = i % TEST_CASES; |
| 260 | checked(opt, short_test_buff[index] + j, opt_dest, stride, stride, width, height, w0, round, shift, offset); |
| 261 | ref(short_test_buff[index] + j, ref_dest, stride, stride, width, height, w0, round, shift, offset); |
| 262 | |
| 263 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 264 | return false; |
| 265 | |
| 266 | reportfail(); |
| 267 | j += INCR; |
| 268 | } |
| 269 | |
| 270 | return true; |
| 271 | } |
| 272 | |
| 273 | bool PixelHarness::check_weightp(weightp_pp_t ref, weightp_pp_t opt) |
| 274 | { |
| 275 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 276 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 277 | |
| 278 | memset(ref_dest, 0, 64 * 64 * sizeof(pixel)); |
| 279 | memset(opt_dest, 0, 64 * 64 * sizeof(pixel)); |
| 280 | int j = 0; |
| 281 | int width = 16 * (rand() % 4 + 1); |
| 282 | int height = 8; |
| 283 | int w0 = rand() % 128; |
| 284 | int shift = rand() % 15; |
| 285 | int round = shift ? (1 << (shift - 1)) : 0; |
| 286 | int offset = (rand() % 256) - 128; |
| 287 | intptr_t stride = 64; |
| 288 | for (int i = 0; i < ITERS; i++) |
| 289 | { |
| 290 | int index = i % TEST_CASES; |
| 291 | checked(opt, pixel_test_buff[index] + j, opt_dest, stride, width, height, w0, round, shift, offset); |
| 292 | ref(pixel_test_buff[index] + j, ref_dest, stride, width, height, w0, round, shift, offset); |
| 293 | |
| 294 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 295 | return false; |
| 296 | |
| 297 | reportfail(); |
| 298 | j += INCR; |
| 299 | } |
| 300 | |
| 301 | return true; |
| 302 | } |
| 303 | |
| 304 | bool PixelHarness::check_downscale_t(downscale_t ref, downscale_t opt) |
| 305 | { |
| 306 | ALIGN_VAR_16(pixel, ref_destf[32 * 32]); |
| 307 | ALIGN_VAR_16(pixel, opt_destf[32 * 32]); |
| 308 | |
| 309 | ALIGN_VAR_16(pixel, ref_desth[32 * 32]); |
| 310 | ALIGN_VAR_16(pixel, opt_desth[32 * 32]); |
| 311 | |
| 312 | ALIGN_VAR_16(pixel, ref_destv[32 * 32]); |
| 313 | ALIGN_VAR_16(pixel, opt_destv[32 * 32]); |
| 314 | |
| 315 | ALIGN_VAR_16(pixel, ref_destc[32 * 32]); |
| 316 | ALIGN_VAR_16(pixel, opt_destc[32 * 32]); |
| 317 | |
| 318 | intptr_t src_stride = 64; |
| 319 | intptr_t dst_stride = 32; |
| 320 | int bx = 32; |
| 321 | int by = 32; |
| 322 | int j = 0; |
| 323 | for (int i = 0; i < ITERS; i++) |
| 324 | { |
| 325 | int index = i % TEST_CASES; |
| 326 | ref(pixel_test_buff[index] + j, ref_destf, ref_desth, ref_destv, |
| 327 | ref_destc, src_stride, dst_stride, bx, by); |
| 328 | checked(opt, pixel_test_buff[index] + j, opt_destf, opt_desth, opt_destv, |
| 329 | opt_destc, src_stride, dst_stride, bx, by); |
| 330 | |
| 331 | if (memcmp(ref_destf, opt_destf, 32 * 32 * sizeof(pixel))) |
| 332 | return false; |
| 333 | if (memcmp(ref_desth, opt_desth, 32 * 32 * sizeof(pixel))) |
| 334 | return false; |
| 335 | if (memcmp(ref_destv, opt_destv, 32 * 32 * sizeof(pixel))) |
| 336 | return false; |
| 337 | if (memcmp(ref_destc, opt_destc, 32 * 32 * sizeof(pixel))) |
| 338 | return false; |
| 339 | |
| 340 | reportfail(); |
| 341 | j += INCR; |
| 342 | } |
| 343 | |
| 344 | return true; |
| 345 | } |
| 346 | |
| 347 | bool PixelHarness::check_cvt32to16_shr_t(cvt32to16_shr_t ref, cvt32to16_shr_t opt) |
| 348 | { |
| 349 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 350 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 351 | |
| 352 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 353 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 354 | |
| 355 | int j = 0; |
| 356 | intptr_t stride = STRIDE; |
| 357 | for (int i = 0; i < ITERS; i++) |
| 358 | { |
| 359 | int shift = (rand() % 7 + 1); |
| 360 | |
| 361 | int index = i % TEST_CASES; |
| 362 | checked(opt, opt_dest, int_test_buff[index] + j, stride, shift, (int)STRIDE); |
| 363 | ref(ref_dest, int_test_buff[index] + j, stride, shift, (int)STRIDE); |
| 364 | |
| 365 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 366 | return false; |
| 367 | |
| 368 | reportfail(); |
| 369 | j += INCR; |
| 370 | } |
| 371 | |
| 372 | return true; |
| 373 | } |
| 374 | |
| 375 | bool PixelHarness::check_cvt16to32_shl_t(cvt16to32_shl_t ref, cvt16to32_shl_t opt) |
| 376 | { |
| 377 | ALIGN_VAR_16(int32_t, ref_dest[64 * 64]); |
| 378 | ALIGN_VAR_16(int32_t, opt_dest[64 * 64]); |
| 379 | |
| 380 | int j = 0; |
| 381 | intptr_t stride = STRIDE; |
| 382 | for (int i = 0; i < ITERS; i++) |
| 383 | { |
| 384 | int shift = (rand() % 7 + 1); |
| 385 | |
| 386 | int index = i % TEST_CASES; |
| 387 | checked(opt, opt_dest, short_test_buff[index] + j, stride, shift, (int)stride); |
| 388 | ref(ref_dest, short_test_buff[index] + j, stride, shift, (int)stride); |
| 389 | |
| 390 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int32_t))) |
| 391 | return false; |
| 392 | |
| 393 | reportfail(); |
| 394 | j += INCR; |
| 395 | } |
| 396 | |
| 397 | return true; |
| 398 | } |
| 399 | |
| 400 | bool PixelHarness::check_cvt16to32_shr_t(cvt16to32_shr_t ref, cvt16to32_shr_t opt) |
| 401 | { |
| 402 | ALIGN_VAR_16(int32_t, ref_dest[64 * 64]); |
| 403 | ALIGN_VAR_16(int32_t, opt_dest[64 * 64]); |
| 404 | |
| 405 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 406 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 407 | |
| 408 | int j = 0; |
| 409 | intptr_t stride = STRIDE; |
| 410 | for (int i = 0; i < ITERS; i++) |
| 411 | { |
| 412 | int shift = (rand() % 7 + 1); |
| 413 | |
| 414 | int index = i % TEST_CASES; |
| 415 | checked(opt, opt_dest, short_test_buff[index] + j, stride, shift, (int)stride); |
| 416 | ref(ref_dest, short_test_buff[index] + j, stride, shift, (int)stride); |
| 417 | |
| 418 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int32_t))) |
| 419 | return false; |
| 420 | |
| 421 | reportfail(); |
| 422 | j += INCR; |
| 423 | } |
| 424 | |
| 425 | return true; |
| 426 | } |
| 427 | |
| 428 | bool PixelHarness::check_cvt32to16_shl_t(cvt32to16_shl_t ref, cvt32to16_shl_t opt) |
| 429 | { |
| 430 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 431 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 432 | |
| 433 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 434 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 435 | |
| 436 | int j = 0; |
| 437 | intptr_t stride = STRIDE; |
| 438 | for (int i = 0; i < ITERS; i++) |
| 439 | { |
| 440 | int shift = (rand() % 7 + 1); |
| 441 | |
| 442 | int index = i % TEST_CASES; |
| 443 | checked(opt, opt_dest, int_test_buff[index] + j, stride, shift); |
| 444 | ref(ref_dest, int_test_buff[index] + j, stride, shift); |
| 445 | |
| 446 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 447 | return false; |
| 448 | |
| 449 | reportfail(); |
| 450 | j += INCR; |
| 451 | } |
| 452 | |
| 453 | return true; |
| 454 | } |
| 455 | |
| 456 | bool PixelHarness::check_copy_cnt_t(copy_cnt_t ref, copy_cnt_t opt) |
| 457 | { |
| 458 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 459 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 460 | |
| 461 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 462 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 463 | |
| 464 | int j = 0; |
| 465 | intptr_t stride = STRIDE; |
| 466 | for (int i = 0; i < ITERS; i++) |
| 467 | { |
| 468 | int index = i % TEST_CASES; |
| 469 | int opt_cnt = (int)checked(opt, opt_dest, short_test_buff1[index] + j, stride); |
| 470 | int ref_cnt = ref(ref_dest, short_test_buff1[index] + j, stride); |
| 471 | |
| 472 | if ((ref_cnt != opt_cnt) || memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 473 | return false; |
| 474 | |
| 475 | reportfail(); |
| 476 | j += INCR; |
| 477 | } |
| 478 | |
| 479 | return true; |
| 480 | } |
| 481 | |
| 482 | bool PixelHarness::check_copy_shr_t(copy_shr_t ref, copy_shr_t opt) |
| 483 | { |
| 484 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 485 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 486 | |
| 487 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 488 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 489 | |
| 490 | int j = 0; |
| 491 | intptr_t stride = STRIDE; |
| 492 | for (int i = 0; i < ITERS; i++) |
| 493 | { |
| 494 | int shift = (rand() % 7 + 1); |
| 495 | |
| 496 | int index = i % TEST_CASES; |
| 497 | checked(opt, opt_dest, short_test_buff[index] + j, stride, shift, (int)STRIDE); |
| 498 | ref(ref_dest, short_test_buff[index] + j, stride, shift, (int)STRIDE); |
| 499 | |
| 500 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 501 | return false; |
| 502 | |
| 503 | reportfail(); |
| 504 | j += INCR; |
| 505 | } |
| 506 | |
| 507 | return true; |
| 508 | } |
| 509 | |
| 510 | bool PixelHarness::check_copy_shl_t(copy_shl_t ref, copy_shl_t opt) |
| 511 | { |
| 512 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 513 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 514 | |
| 515 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 516 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 517 | |
| 518 | int j = 0; |
| 519 | intptr_t stride = STRIDE; |
| 520 | for (int i = 0; i < ITERS; i++) |
| 521 | { |
| 522 | int shift = (rand() % 7 + 1); |
| 523 | |
| 524 | int index = i % TEST_CASES; |
| 525 | checked(opt, opt_dest, short_test_buff[index] + j, stride, shift); |
| 526 | ref(ref_dest, short_test_buff[index] + j, stride, shift); |
| 527 | |
| 528 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 529 | return false; |
| 530 | |
| 531 | reportfail(); |
| 532 | j += INCR; |
| 533 | } |
| 534 | |
| 535 | return true; |
| 536 | } |
| 537 | |
| 538 | bool PixelHarness::check_pixelavg_pp(pixelavg_pp_t ref, pixelavg_pp_t opt) |
| 539 | { |
| 540 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 541 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 542 | |
| 543 | int j = 0; |
| 544 | |
| 545 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 546 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 547 | |
| 548 | intptr_t stride = STRIDE; |
| 549 | for (int i = 0; i < ITERS; i++) |
| 550 | { |
| 551 | int index1 = rand() % TEST_CASES; |
| 552 | int index2 = rand() % TEST_CASES; |
| 553 | checked(ref, ref_dest, stride, pixel_test_buff[index1] + j, |
| 554 | stride, pixel_test_buff[index2] + j, stride, 32); |
| 555 | opt(opt_dest, stride, pixel_test_buff[index1] + j, |
| 556 | stride, pixel_test_buff[index2] + j, stride, 32); |
| 557 | |
| 558 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 559 | return false; |
| 560 | |
| 561 | reportfail(); |
| 562 | j += INCR; |
| 563 | } |
| 564 | |
| 565 | return true; |
| 566 | } |
| 567 | |
| 568 | bool PixelHarness::check_copy_pp(copy_pp_t ref, copy_pp_t opt) |
| 569 | { |
| 570 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 571 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 572 | |
| 573 | // we don't know the partition size so we are checking the entire output buffer so |
| 574 | // we must initialize the buffers |
| 575 | memset(ref_dest, 0, sizeof(ref_dest)); |
| 576 | memset(opt_dest, 0, sizeof(opt_dest)); |
| 577 | |
| 578 | int j = 0; |
| 579 | intptr_t stride = STRIDE; |
| 580 | for (int i = 0; i < ITERS; i++) |
| 581 | { |
| 582 | int index = i % TEST_CASES; |
| 583 | checked(opt, opt_dest, stride, pixel_test_buff[index] + j, stride); |
| 584 | ref(ref_dest, stride, pixel_test_buff[index] + j, stride); |
| 585 | |
| 586 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 587 | return false; |
| 588 | |
| 589 | reportfail(); |
| 590 | j += INCR; |
| 591 | } |
| 592 | |
| 593 | return true; |
| 594 | } |
| 595 | |
| 596 | bool PixelHarness::check_copy_sp(copy_sp_t ref, copy_sp_t opt) |
| 597 | { |
| 598 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 599 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 600 | |
| 601 | // we don't know the partition size so we are checking the entire output buffer so |
| 602 | // we must initialize the buffers |
| 603 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 604 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 605 | |
| 606 | int j = 0; |
| 607 | intptr_t stride1 = 64, stride2 = STRIDE; |
| 608 | for (int i = 0; i < ITERS; i++) |
| 609 | { |
| 610 | int index = i % TEST_CASES; |
| 611 | checked(opt, opt_dest, stride1, short_test_buff1[index] + j, stride2); |
| 612 | ref(ref_dest, stride1, short_test_buff1[index] + j, stride2); |
| 613 | |
| 614 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 615 | return false; |
| 616 | |
| 617 | reportfail(); |
| 618 | j += INCR; |
| 619 | } |
| 620 | |
| 621 | return true; |
| 622 | } |
| 623 | |
| 624 | bool PixelHarness::check_copy_ps(copy_ps_t ref, copy_ps_t opt) |
| 625 | { |
| 626 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 627 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 628 | |
| 629 | // we don't know the partition size so we are checking the entire output buffer so |
| 630 | // we must initialize the buffers |
| 631 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 632 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 633 | |
| 634 | int j = 0; |
| 635 | intptr_t stride = STRIDE; |
| 636 | for (int i = 0; i < ITERS; i++) |
| 637 | { |
| 638 | int index = i % TEST_CASES; |
| 639 | checked(opt, opt_dest, stride, pixel_test_buff[index] + j, stride); |
| 640 | ref(ref_dest, stride, pixel_test_buff[index] + j, stride); |
| 641 | |
| 642 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 643 | return false; |
| 644 | |
| 645 | reportfail(); |
| 646 | j += INCR; |
| 647 | } |
| 648 | |
| 649 | return true; |
| 650 | } |
| 651 | |
| 652 | bool PixelHarness::check_copy_ss(copy_ss_t ref, copy_ss_t opt) |
| 653 | { |
| 654 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 655 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 656 | |
| 657 | // we don't know the partition size so we are checking the entire output buffer so |
| 658 | // we must initialize the buffers |
| 659 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 660 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 661 | |
| 662 | int j = 0; |
| 663 | intptr_t stride = STRIDE; |
| 664 | for (int i = 0; i < ITERS; i++) |
| 665 | { |
| 666 | int index = i % TEST_CASES; |
| 667 | checked(opt, opt_dest, stride, short_test_buff1[index] + j, stride); |
| 668 | ref(ref_dest, stride, short_test_buff1[index] + j, stride); |
| 669 | |
| 670 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 671 | return false; |
| 672 | |
| 673 | reportfail(); |
| 674 | j += INCR; |
| 675 | } |
| 676 | |
| 677 | return true; |
| 678 | } |
| 679 | |
| 680 | bool PixelHarness::check_blockfill_s(blockfill_s_t ref, blockfill_s_t opt) |
| 681 | { |
| 682 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 683 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 684 | |
| 685 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 686 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 687 | |
| 688 | intptr_t stride = 64; |
| 689 | for (int i = 0; i < ITERS; i++) |
| 690 | { |
| 691 | int16_t value = (rand() % SHORT_MAX) + 1; |
| 692 | |
| 693 | checked(opt, opt_dest, stride, value); |
| 694 | ref(ref_dest, stride, value); |
| 695 | |
| 696 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 697 | return false; |
| 698 | |
| 699 | reportfail(); |
| 700 | } |
| 701 | |
| 702 | return true; |
| 703 | } |
| 704 | |
| 705 | bool PixelHarness::check_pixel_sub_ps(pixel_sub_ps_t ref, pixel_sub_ps_t opt) |
| 706 | { |
| 707 | ALIGN_VAR_16(int16_t, ref_dest[64 * 64]); |
| 708 | ALIGN_VAR_16(int16_t, opt_dest[64 * 64]); |
| 709 | |
| 710 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 711 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 712 | |
| 713 | int j = 0; |
| 714 | intptr_t stride2 = 64, stride = STRIDE; |
| 715 | for (int i = 0; i < 1; i++) |
| 716 | { |
| 717 | int index1 = rand() % TEST_CASES; |
| 718 | int index2 = rand() % TEST_CASES; |
| 719 | checked(opt, opt_dest, stride2, pixel_test_buff[index1] + j, |
| 720 | pixel_test_buff[index2] + j, stride, stride); |
| 721 | ref(ref_dest, stride2, pixel_test_buff[index1] + j, |
| 722 | pixel_test_buff[index2] + j, stride, stride); |
| 723 | |
| 724 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(int16_t))) |
| 725 | return false; |
| 726 | |
| 727 | reportfail(); |
| 728 | j += INCR; |
| 729 | } |
| 730 | |
| 731 | return true; |
| 732 | } |
| 733 | |
| 734 | bool PixelHarness::check_scale_pp(scale_t ref, scale_t opt) |
| 735 | { |
| 736 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 737 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 738 | |
| 739 | memset(ref_dest, 0, sizeof(ref_dest)); |
| 740 | memset(opt_dest, 0, sizeof(opt_dest)); |
| 741 | |
| 742 | int j = 0; |
| 743 | intptr_t stride = STRIDE; |
| 744 | for (int i = 0; i < ITERS; i++) |
| 745 | { |
| 746 | int index = i % TEST_CASES; |
| 747 | checked(opt, opt_dest, pixel_test_buff[index] + j, stride); |
| 748 | ref(ref_dest, pixel_test_buff[index] + j, stride); |
| 749 | |
| 750 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 751 | return false; |
| 752 | |
| 753 | reportfail(); |
| 754 | j += INCR; |
| 755 | } |
| 756 | |
| 757 | return true; |
| 758 | } |
| 759 | |
| 760 | bool PixelHarness::check_transpose(transpose_t ref, transpose_t opt) |
| 761 | { |
| 762 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 763 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 764 | |
| 765 | memset(ref_dest, 0, sizeof(ref_dest)); |
| 766 | memset(opt_dest, 0, sizeof(opt_dest)); |
| 767 | |
| 768 | int j = 0; |
| 769 | intptr_t stride = STRIDE; |
| 770 | for (int i = 0; i < ITERS; i++) |
| 771 | { |
| 772 | int index = i % TEST_CASES; |
| 773 | checked(opt, opt_dest, pixel_test_buff[index] + j, stride); |
| 774 | ref(ref_dest, pixel_test_buff[index] + j, stride); |
| 775 | |
| 776 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 777 | return false; |
| 778 | |
| 779 | reportfail(); |
| 780 | j += INCR; |
| 781 | } |
| 782 | |
| 783 | return true; |
| 784 | } |
| 785 | |
| 786 | bool PixelHarness::check_pixel_add_ps(pixel_add_ps_t ref, pixel_add_ps_t opt) |
| 787 | { |
| 788 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 789 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 790 | |
| 791 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 792 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 793 | |
| 794 | int j = 0; |
| 795 | intptr_t stride2 = 64, stride = STRIDE; |
| 796 | for (int i = 0; i < ITERS; i++) |
| 797 | { |
| 798 | int index1 = rand() % TEST_CASES; |
| 799 | int index2 = rand() % TEST_CASES; |
| 800 | checked(opt, opt_dest, stride2, pixel_test_buff[index1] + j, short_test_buff[index2] + j, stride, stride); |
| 801 | ref(ref_dest, stride2, pixel_test_buff[index1] + j, short_test_buff[index2] + j, stride, stride); |
| 802 | |
| 803 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 804 | return false; |
| 805 | |
| 806 | reportfail(); |
| 807 | j += INCR; |
| 808 | } |
| 809 | |
| 810 | return true; |
| 811 | } |
| 812 | |
| 813 | bool PixelHarness::check_pixel_var(var_t ref, var_t opt) |
| 814 | { |
| 815 | int j = 0; |
| 816 | |
| 817 | intptr_t stride = STRIDE; |
| 818 | |
| 819 | for (int i = 0; i < ITERS; i++) |
| 820 | { |
| 821 | int index = i % TEST_CASES; |
| 822 | uint64_t vres = checked(opt, pixel_test_buff[index], stride); |
| 823 | uint64_t cres = ref(pixel_test_buff[index], stride); |
| 824 | if (vres != cres) |
| 825 | return false; |
| 826 | |
| 827 | reportfail(); |
| 828 | j += INCR; |
| 829 | } |
| 830 | |
| 831 | return true; |
| 832 | } |
| 833 | |
| 834 | bool PixelHarness::check_ssim_4x4x2_core(ssim_4x4x2_core_t ref, ssim_4x4x2_core_t opt) |
| 835 | { |
| 836 | ALIGN_VAR_32(int, sum0[2][4]); |
| 837 | ALIGN_VAR_32(int, sum1[2][4]); |
| 838 | |
| 839 | for (int i = 0; i < ITERS; i++) |
| 840 | { |
| 841 | intptr_t stride = rand() % 64; |
| 842 | int index1 = rand() % TEST_CASES; |
| 843 | int index2 = rand() % TEST_CASES; |
| 844 | ref(pixel_test_buff[index1] + i, stride, pixel_test_buff[index2] + i, stride, sum0); |
| 845 | checked(opt, pixel_test_buff[index1] + i, stride, pixel_test_buff[index2] + i, stride, sum1); |
| 846 | |
| 847 | if (memcmp(sum0, sum1, sizeof(sum0))) |
| 848 | return false; |
| 849 | |
| 850 | reportfail(); |
| 851 | } |
| 852 | |
| 853 | return true; |
| 854 | } |
| 855 | |
| 856 | /* TODO: This function causes crashes when checked. Is this a real bug? */ |
| 857 | bool PixelHarness::check_ssim_end(ssim_end4_t ref, ssim_end4_t opt) |
| 858 | { |
| 859 | ALIGN_VAR_32(int, sum0[5][4]); |
| 860 | ALIGN_VAR_32(int, sum1[5][4]); |
| 861 | |
| 862 | for (int i = 0; i < ITERS; i++) |
| 863 | { |
| 864 | for (int j = 0; j < 5; j++) |
| 865 | { |
| 866 | for (int k = 0; k < 4; k++) |
| 867 | { |
| 868 | sum0[j][k] = rand() % (1 << 12); |
| 869 | sum1[j][k] = rand() % (1 << 12); |
| 870 | } |
| 871 | } |
| 872 | |
| 873 | int width = (rand() % 4) + 1; // range[1-4] |
| 874 | float cres = ref(sum0, sum1, width); |
| 875 | float vres = checked_float(opt, sum0, sum1, width); |
| 876 | if (fabs(vres - cres) > 0.00001) |
| 877 | return false; |
| 878 | |
| 879 | reportfail(); |
| 880 | } |
| 881 | |
| 882 | return true; |
| 883 | } |
| 884 | |
| 885 | bool PixelHarness::check_addAvg(addAvg_t ref, addAvg_t opt) |
| 886 | { |
| 887 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 888 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 889 | |
| 890 | int j = 0; |
| 891 | |
| 892 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 893 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 894 | intptr_t stride = STRIDE; |
| 895 | |
| 896 | for (int i = 0; i < ITERS; i++) |
| 897 | { |
| 898 | int index1 = rand() % TEST_CASES; |
| 899 | int index2 = rand() % TEST_CASES; |
| 900 | ref(short_test_buff2[index1] + j, short_test_buff2[index2] + j, ref_dest, stride, stride, stride); |
| 901 | checked(opt, short_test_buff2[index1] + j, short_test_buff2[index2] + j, opt_dest, stride, stride, stride); |
| 902 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 903 | return false; |
| 904 | |
| 905 | reportfail(); |
| 906 | j += INCR; |
| 907 | } |
| 908 | |
| 909 | return true; |
| 910 | } |
| 911 | |
| 912 | bool PixelHarness::check_saoCuOrgE0_t(saoCuOrgE0_t ref, saoCuOrgE0_t opt) |
| 913 | { |
| 914 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 915 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 916 | |
| 917 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 918 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 919 | |
| 920 | int j = 0; |
| 921 | |
| 922 | for (int i = 0; i < ITERS; i++) |
| 923 | { |
| 924 | int width = 16 * (rand() % 4 + 1); |
| 925 | int8_t sign = rand() % 3; |
| 926 | if (sign == 2) |
| 927 | { |
| 928 | sign = -1; |
| 929 | } |
| 930 | |
| 931 | ref(ref_dest, psbuf1 + j, width, sign); |
| 932 | checked(opt, opt_dest, psbuf1 + j, width, sign); |
| 933 | |
| 934 | if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel))) |
| 935 | return false; |
| 936 | |
| 937 | reportfail(); |
| 938 | j += INCR; |
| 939 | } |
| 940 | |
| 941 | return true; |
| 942 | } |
| 943 | |
| 944 | bool PixelHarness::check_planecopy_sp(planecopy_sp_t ref, planecopy_sp_t opt) |
| 945 | { |
| 946 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 947 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 948 | |
| 949 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 950 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 951 | |
| 952 | int width = 16 + rand() % 48; |
| 953 | int height = 16 + rand() % 48; |
| 954 | intptr_t srcStride = 64; |
| 955 | intptr_t dstStride = width; |
| 956 | int j = 0; |
| 957 | |
| 958 | for (int i = 0; i < ITERS; i++) |
| 959 | { |
| 960 | int index = i % TEST_CASES; |
| 961 | checked(opt, ushort_test_buff[index] + j, srcStride, opt_dest, dstStride, width, height, (int)8, (uint16_t)255); |
| 962 | ref(ushort_test_buff[index] + j, srcStride, ref_dest, dstStride, width, height, (int)8, (uint16_t)255); |
| 963 | |
| 964 | if (memcmp(ref_dest, opt_dest, width * height * sizeof(pixel))) |
| 965 | return false; |
| 966 | |
| 967 | reportfail(); |
| 968 | j += INCR; |
| 969 | } |
| 970 | |
| 971 | return true; |
| 972 | } |
| 973 | |
| 974 | bool PixelHarness::check_planecopy_cp(planecopy_cp_t ref, planecopy_cp_t opt) |
| 975 | { |
| 976 | ALIGN_VAR_16(pixel, ref_dest[64 * 64]); |
| 977 | ALIGN_VAR_16(pixel, opt_dest[64 * 64]); |
| 978 | |
| 979 | memset(ref_dest, 0xCD, sizeof(ref_dest)); |
| 980 | memset(opt_dest, 0xCD, sizeof(opt_dest)); |
| 981 | |
| 982 | int width = 16 + rand() % 48; |
| 983 | int height = 16 + rand() % 48; |
| 984 | intptr_t srcStride = 64; |
| 985 | intptr_t dstStride = width; |
| 986 | int j = 0; |
| 987 | |
| 988 | for (int i = 0; i < ITERS; i++) |
| 989 | { |
| 990 | int index = i % TEST_CASES; |
| 991 | checked(opt, uchar_test_buff[index] + j, srcStride, opt_dest, dstStride, width, height, (int)2); |
| 992 | ref(uchar_test_buff[index] + j, srcStride, ref_dest, dstStride, width, height, (int)2); |
| 993 | |
| 994 | if (memcmp(ref_dest, opt_dest, width * height * sizeof(pixel))) |
| 995 | return false; |
| 996 | |
| 997 | reportfail(); |
| 998 | j += INCR; |
| 999 | } |
| 1000 | |
| 1001 | return true; |
| 1002 | } |
| 1003 | |
| 1004 | bool PixelHarness::testPartition(int part, const EncoderPrimitives& ref, const EncoderPrimitives& opt) |
| 1005 | { |
| 1006 | if (opt.satd[part]) |
| 1007 | { |
| 1008 | if (!check_pixelcmp(ref.satd[part], opt.satd[part])) |
| 1009 | { |
| 1010 | printf("satd[%s]: failed!\n", lumaPartStr[part]); |
| 1011 | return false; |
| 1012 | } |
| 1013 | } |
| 1014 | |
| 1015 | if (opt.sa8d_inter[part]) |
| 1016 | { |
| 1017 | if (!check_pixelcmp(ref.sa8d_inter[part], opt.sa8d_inter[part])) |
| 1018 | { |
| 1019 | printf("sa8d_inter[%s]: failed!\n", lumaPartStr[part]); |
| 1020 | return false; |
| 1021 | } |
| 1022 | } |
| 1023 | |
| 1024 | if (opt.sad[part]) |
| 1025 | { |
| 1026 | if (!check_pixelcmp(ref.sad[part], opt.sad[part])) |
| 1027 | { |
| 1028 | printf("sad[%s]: failed!\n", lumaPartStr[part]); |
| 1029 | return false; |
| 1030 | } |
| 1031 | } |
| 1032 | |
| 1033 | if (opt.sse_pp[part]) |
| 1034 | { |
| 1035 | if (!check_pixelcmp(ref.sse_pp[part], opt.sse_pp[part])) |
| 1036 | { |
| 1037 | printf("sse_pp[%s]: failed!\n", lumaPartStr[part]); |
| 1038 | return false; |
| 1039 | } |
| 1040 | } |
| 1041 | |
| 1042 | if (opt.sse_sp[part]) |
| 1043 | { |
| 1044 | if (!check_pixelcmp_sp(ref.sse_sp[part], opt.sse_sp[part])) |
| 1045 | { |
| 1046 | printf("sse_sp[%s]: failed!\n", lumaPartStr[part]); |
| 1047 | return false; |
| 1048 | } |
| 1049 | } |
| 1050 | |
| 1051 | if (opt.sse_ss[part]) |
| 1052 | { |
| 1053 | if (!check_pixelcmp_ss(ref.sse_ss[part], opt.sse_ss[part])) |
| 1054 | { |
| 1055 | printf("sse_ss[%s]: failed!\n", lumaPartStr[part]); |
| 1056 | return false; |
| 1057 | } |
| 1058 | } |
| 1059 | |
| 1060 | if (opt.sad_x3[part]) |
| 1061 | { |
| 1062 | if (!check_pixelcmp_x3(ref.sad_x3[part], opt.sad_x3[part])) |
| 1063 | { |
| 1064 | printf("sad_x3[%s]: failed!\n", lumaPartStr[part]); |
| 1065 | return false; |
| 1066 | } |
| 1067 | } |
| 1068 | |
| 1069 | if (opt.sad_x4[part]) |
| 1070 | { |
| 1071 | if (!check_pixelcmp_x4(ref.sad_x4[part], opt.sad_x4[part])) |
| 1072 | { |
| 1073 | printf("sad_x4[%s]: failed!\n", lumaPartStr[part]); |
| 1074 | return false; |
| 1075 | } |
| 1076 | } |
| 1077 | |
| 1078 | if (opt.pixelavg_pp[part]) |
| 1079 | { |
| 1080 | if (!check_pixelavg_pp(ref.pixelavg_pp[part], opt.pixelavg_pp[part])) |
| 1081 | { |
| 1082 | printf("pixelavg_pp[%s]: failed!\n", lumaPartStr[part]); |
| 1083 | return false; |
| 1084 | } |
| 1085 | } |
| 1086 | |
| 1087 | if (opt.luma_copy_pp[part]) |
| 1088 | { |
| 1089 | if (!check_copy_pp(ref.luma_copy_pp[part], opt.luma_copy_pp[part])) |
| 1090 | { |
| 1091 | printf("luma_copy_pp[%s] failed\n", lumaPartStr[part]); |
| 1092 | return false; |
| 1093 | } |
| 1094 | } |
| 1095 | |
| 1096 | if (opt.luma_copy_sp[part]) |
| 1097 | { |
| 1098 | if (!check_copy_sp(ref.luma_copy_sp[part], opt.luma_copy_sp[part])) |
| 1099 | { |
| 1100 | printf("luma_copy_sp[%s] failed\n", lumaPartStr[part]); |
| 1101 | return false; |
| 1102 | } |
| 1103 | } |
| 1104 | |
| 1105 | if (opt.luma_copy_ps[part]) |
| 1106 | { |
| 1107 | if (!check_copy_ps(ref.luma_copy_ps[part], opt.luma_copy_ps[part])) |
| 1108 | { |
| 1109 | printf("luma_copy_ps[%s] failed\n", lumaPartStr[part]); |
| 1110 | return false; |
| 1111 | } |
| 1112 | } |
| 1113 | |
| 1114 | if (opt.luma_copy_ss[part]) |
| 1115 | { |
| 1116 | if (!check_copy_ss(ref.luma_copy_ss[part], opt.luma_copy_ss[part])) |
| 1117 | { |
| 1118 | printf("luma_copy_ss[%s] failed\n", lumaPartStr[part]); |
| 1119 | return false; |
| 1120 | } |
| 1121 | } |
| 1122 | |
| 1123 | if (opt.luma_addAvg[part]) |
| 1124 | { |
| 1125 | if (!check_addAvg(ref.luma_addAvg[part], opt.luma_addAvg[part])) |
| 1126 | { |
| 1127 | printf("luma_addAvg[%s] failed\n", lumaPartStr[part]); |
| 1128 | return false; |
| 1129 | } |
| 1130 | } |
| 1131 | |
| 1132 | if (part < NUM_SQUARE_BLOCKS) |
| 1133 | { |
| 1134 | if (opt.luma_sub_ps[part]) |
| 1135 | { |
| 1136 | if (!check_pixel_sub_ps(ref.luma_sub_ps[part], opt.luma_sub_ps[part])) |
| 1137 | { |
| 1138 | printf("luma_sub_ps[%s] failed\n", lumaPartStr[part]); |
| 1139 | return false; |
| 1140 | } |
| 1141 | } |
| 1142 | |
| 1143 | if (opt.luma_add_ps[part]) |
| 1144 | { |
| 1145 | if (!check_pixel_add_ps(ref.luma_add_ps[part], opt.luma_add_ps[part])) |
| 1146 | { |
| 1147 | printf("luma_add_ps[%s] failed\n", lumaPartStr[part]); |
| 1148 | return false; |
| 1149 | } |
| 1150 | } |
| 1151 | } |
| 1152 | |
| 1153 | for (int i = 0; i < X265_CSP_COUNT; i++) |
| 1154 | { |
| 1155 | if (opt.chroma[i].copy_pp[part]) |
| 1156 | { |
| 1157 | if (!check_copy_pp(ref.chroma[i].copy_pp[part], opt.chroma[i].copy_pp[part])) |
| 1158 | { |
| 1159 | printf("chroma_copy_pp[%s][%s] failed\n", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1160 | return false; |
| 1161 | } |
| 1162 | } |
| 1163 | if (opt.chroma[i].copy_sp[part]) |
| 1164 | { |
| 1165 | if (!check_copy_sp(ref.chroma[i].copy_sp[part], opt.chroma[i].copy_sp[part])) |
| 1166 | { |
| 1167 | printf("chroma_copy_sp[%s][%s] failed\n", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1168 | return false; |
| 1169 | } |
| 1170 | } |
| 1171 | if (opt.chroma[i].copy_ps[part]) |
| 1172 | { |
| 1173 | if (!check_copy_ps(ref.chroma[i].copy_ps[part], opt.chroma[i].copy_ps[part])) |
| 1174 | { |
| 1175 | printf("chroma_copy_ps[%s][%s] failed\n", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1176 | return false; |
| 1177 | } |
| 1178 | } |
| 1179 | if (opt.chroma[i].copy_ss[part]) |
| 1180 | { |
| 1181 | if (!check_copy_ss(ref.chroma[i].copy_ss[part], opt.chroma[i].copy_ss[part])) |
| 1182 | { |
| 1183 | printf("chroma_copy_ss[%s][%s] failed\n", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1184 | return false; |
| 1185 | } |
| 1186 | } |
| 1187 | if (opt.chroma[i].addAvg[part]) |
| 1188 | { |
| 1189 | if (!check_addAvg(ref.chroma[i].addAvg[part], opt.chroma[i].addAvg[part])) |
| 1190 | { |
| 1191 | printf("chroma_addAvg[%s][%s] failed\n", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1192 | return false; |
| 1193 | } |
| 1194 | } |
| 1195 | if (part < NUM_SQUARE_BLOCKS) |
| 1196 | { |
| 1197 | if (opt.chroma[i].sub_ps[part]) |
| 1198 | { |
| 1199 | if (!check_pixel_sub_ps(ref.chroma[i].sub_ps[part], opt.chroma[i].sub_ps[part])) |
| 1200 | { |
| 1201 | printf("chroma_sub_ps[%s][%s] failed\n", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1202 | return false; |
| 1203 | } |
| 1204 | } |
| 1205 | if (opt.chroma[i].add_ps[part]) |
| 1206 | { |
| 1207 | if (!check_pixel_add_ps(ref.chroma[i].add_ps[part], opt.chroma[i].add_ps[part])) |
| 1208 | { |
| 1209 | printf("chroma_add_ps[%s][%s] failed\n", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1210 | return false; |
| 1211 | } |
| 1212 | } |
| 1213 | } |
| 1214 | } |
| 1215 | |
| 1216 | return true; |
| 1217 | } |
| 1218 | |
| 1219 | bool PixelHarness::testCorrectness(const EncoderPrimitives& ref, const EncoderPrimitives& opt) |
| 1220 | { |
| 1221 | for (int size = 4; size <= 64; size *= 2) |
| 1222 | { |
| 1223 | int part = partitionFromSizes(size, size); // 2Nx2N |
| 1224 | if (!testPartition(part, ref, opt)) return false; |
| 1225 | |
| 1226 | if (size > 4) |
| 1227 | { |
| 1228 | part = partitionFromSizes(size, size >> 1); // 2NxN |
| 1229 | if (!testPartition(part, ref, opt)) return false; |
| 1230 | part = partitionFromSizes(size >> 1, size); // Nx2N |
| 1231 | if (!testPartition(part, ref, opt)) return false; |
| 1232 | } |
| 1233 | if (size > 8) |
| 1234 | { |
| 1235 | // 4 AMP modes |
| 1236 | part = partitionFromSizes(size, size >> 2); |
| 1237 | if (!testPartition(part, ref, opt)) return false; |
| 1238 | part = partitionFromSizes(size, 3 * (size >> 2)); |
| 1239 | if (!testPartition(part, ref, opt)) return false; |
| 1240 | |
| 1241 | part = partitionFromSizes(size >> 2, size); |
| 1242 | if (!testPartition(part, ref, opt)) return false; |
| 1243 | part = partitionFromSizes(3 * (size >> 2), size); |
| 1244 | if (!testPartition(part, ref, opt)) return false; |
| 1245 | } |
| 1246 | } |
| 1247 | |
| 1248 | for (int i = 0; i < NUM_SQUARE_BLOCKS; i++) |
| 1249 | { |
| 1250 | if (opt.calcresidual[i]) |
| 1251 | { |
| 1252 | if (!check_calresidual(ref.calcresidual[i], opt.calcresidual[i])) |
| 1253 | { |
| 1254 | printf("calcresidual width: %d failed!\n", 4 << i); |
| 1255 | return false; |
| 1256 | } |
| 1257 | } |
| 1258 | if (opt.sa8d[i]) |
| 1259 | { |
| 1260 | if (!check_pixelcmp(ref.sa8d[i], opt.sa8d[i])) |
| 1261 | { |
| 1262 | printf("sa8d[%dx%d]: failed!\n", 4 << i, 4 << i); |
| 1263 | return false; |
| 1264 | } |
| 1265 | } |
| 1266 | |
| 1267 | if ((i <= BLOCK_32x32) && opt.ssd_s[i]) |
| 1268 | { |
| 1269 | if (!check_ssd_s(ref.ssd_s[i], opt.ssd_s[i])) |
| 1270 | { |
| 1271 | printf("ssd_s[%dx%d]: failed!\n", 4 << i, 4 << i); |
| 1272 | return false; |
| 1273 | } |
| 1274 | } |
| 1275 | |
| 1276 | if (opt.blockfill_s[i]) |
| 1277 | { |
| 1278 | if (!check_blockfill_s(ref.blockfill_s[i], opt.blockfill_s[i])) |
| 1279 | { |
| 1280 | printf("blockfill_s[%dx%d]: failed!\n", 4 << i, 4 << i); |
| 1281 | return false; |
| 1282 | } |
| 1283 | } |
| 1284 | if (opt.transpose[i]) |
| 1285 | { |
| 1286 | if (!check_transpose(ref.transpose[i], opt.transpose[i])) |
| 1287 | { |
| 1288 | printf("transpose[%dx%d] failed\n", 4 << i, 4 << i); |
| 1289 | return false; |
| 1290 | } |
| 1291 | } |
| 1292 | |
| 1293 | if (opt.var[i]) |
| 1294 | { |
| 1295 | if (!check_pixel_var(ref.var[i], opt.var[i])) |
| 1296 | { |
| 1297 | printf("var[%dx%d] failed\n", 4 << i, 4 << i); |
| 1298 | return false; |
| 1299 | } |
| 1300 | } |
| 1301 | |
| 1302 | if ((i < BLOCK_64x64) && opt.copy_cnt[i]) |
| 1303 | { |
| 1304 | if (!check_copy_cnt_t(ref.copy_cnt[i], opt.copy_cnt[i])) |
| 1305 | { |
| 1306 | printf("copy_cnt[%dx%d] failed!\n", 4 << i, 4 << i); |
| 1307 | return false; |
| 1308 | } |
| 1309 | } |
| 1310 | |
| 1311 | if ((i < BLOCK_64x64) && opt.cvt16to32_shr[i]) |
| 1312 | { |
| 1313 | if (!check_cvt16to32_shr_t(ref.cvt16to32_shr[i], opt.cvt16to32_shr[i])) |
| 1314 | { |
| 1315 | printf("cvt16to32_shr failed!\n"); |
| 1316 | return false; |
| 1317 | } |
| 1318 | } |
| 1319 | |
| 1320 | if ((i < BLOCK_64x64) && opt.cvt32to16_shl[i]) |
| 1321 | { |
| 1322 | if (!check_cvt32to16_shl_t(ref.cvt32to16_shl[i], opt.cvt32to16_shl[i])) |
| 1323 | { |
| 1324 | printf("cvt32to16_shl failed!\n"); |
| 1325 | return false; |
| 1326 | } |
| 1327 | } |
| 1328 | |
| 1329 | if ((i < BLOCK_64x64) && opt.copy_shl[i]) |
| 1330 | { |
| 1331 | if (!check_copy_shl_t(ref.copy_shl[i], opt.copy_shl[i])) |
| 1332 | { |
| 1333 | printf("copy_shl[%dx%d] failed!\n", 4 << i, 4 << i); |
| 1334 | return false; |
| 1335 | } |
| 1336 | } |
| 1337 | |
| 1338 | } |
| 1339 | |
| 1340 | if (opt.cvt32to16_shr) |
| 1341 | { |
| 1342 | if (!check_cvt32to16_shr_t(ref.cvt32to16_shr, opt.cvt32to16_shr)) |
| 1343 | { |
| 1344 | printf("cvt32to16 failed!\n"); |
| 1345 | return false; |
| 1346 | } |
| 1347 | } |
| 1348 | |
| 1349 | if (opt.cvt16to32_shl) |
| 1350 | { |
| 1351 | if (!check_cvt16to32_shl_t(ref.cvt16to32_shl, opt.cvt16to32_shl)) |
| 1352 | { |
| 1353 | printf("cvt16to32_shl failed!\n"); |
| 1354 | return false; |
| 1355 | } |
| 1356 | } |
| 1357 | |
| 1358 | if (opt.weight_pp) |
| 1359 | { |
| 1360 | if (!check_weightp(ref.weight_pp, opt.weight_pp)) |
| 1361 | { |
| 1362 | printf("Weighted Prediction (pixel) failed!\n"); |
| 1363 | return false; |
| 1364 | } |
| 1365 | } |
| 1366 | |
| 1367 | if (opt.weight_sp) |
| 1368 | { |
| 1369 | if (!check_weightp(ref.weight_sp, opt.weight_sp)) |
| 1370 | { |
| 1371 | printf("Weighted Prediction (short) failed!\n"); |
| 1372 | return false; |
| 1373 | } |
| 1374 | } |
| 1375 | |
| 1376 | if (opt.frame_init_lowres_core) |
| 1377 | { |
| 1378 | if (!check_downscale_t(ref.frame_init_lowres_core, opt.frame_init_lowres_core)) |
| 1379 | { |
| 1380 | printf("downscale failed!\n"); |
| 1381 | return false; |
| 1382 | } |
| 1383 | } |
| 1384 | |
| 1385 | if (opt.scale1D_128to64) |
| 1386 | { |
| 1387 | if (!check_scale_pp(ref.scale1D_128to64, opt.scale1D_128to64)) |
| 1388 | { |
| 1389 | printf("scale1D_128to64 failed!\n"); |
| 1390 | return false; |
| 1391 | } |
| 1392 | } |
| 1393 | |
| 1394 | if (opt.scale2D_64to32) |
| 1395 | { |
| 1396 | if (!check_scale_pp(ref.scale2D_64to32, opt.scale2D_64to32)) |
| 1397 | { |
| 1398 | printf("scale2D_64to32 failed!\n"); |
| 1399 | return false; |
| 1400 | } |
| 1401 | } |
| 1402 | |
| 1403 | if (opt.ssim_4x4x2_core) |
| 1404 | { |
| 1405 | if (!check_ssim_4x4x2_core(ref.ssim_4x4x2_core, opt.ssim_4x4x2_core)) |
| 1406 | { |
| 1407 | printf("ssim_end_4 failed!\n"); |
| 1408 | return false; |
| 1409 | } |
| 1410 | } |
| 1411 | |
| 1412 | if (opt.ssim_end_4) |
| 1413 | { |
| 1414 | if (!check_ssim_end(ref.ssim_end_4, opt.ssim_end_4)) |
| 1415 | { |
| 1416 | printf("ssim_end_4 failed!\n"); |
| 1417 | return false; |
| 1418 | } |
| 1419 | } |
| 1420 | |
| 1421 | if (opt.saoCuOrgE0) |
| 1422 | { |
| 1423 | if (!check_saoCuOrgE0_t(ref.saoCuOrgE0, opt.saoCuOrgE0)) |
| 1424 | { |
| 1425 | printf("SAO_EO_0 failed\n"); |
| 1426 | return false; |
| 1427 | } |
| 1428 | } |
| 1429 | |
| 1430 | if (opt.planecopy_sp) |
| 1431 | { |
| 1432 | if (!check_planecopy_sp(ref.planecopy_sp, opt.planecopy_sp)) |
| 1433 | { |
| 1434 | printf("planecopy_sp failed\n"); |
| 1435 | return false; |
| 1436 | } |
| 1437 | } |
| 1438 | |
| 1439 | if (opt.planecopy_cp) |
| 1440 | { |
| 1441 | if (!check_planecopy_cp(ref.planecopy_cp, opt.planecopy_cp)) |
| 1442 | { |
| 1443 | printf("planecopy_cp failed\n"); |
| 1444 | return false; |
| 1445 | } |
| 1446 | } |
| 1447 | |
| 1448 | if (opt.copy_shr) |
| 1449 | { |
| 1450 | if (!check_copy_shr_t(ref.copy_shr, opt.copy_shr)) |
| 1451 | { |
| 1452 | printf("copy_shr failed!\n"); |
| 1453 | return false; |
| 1454 | } |
| 1455 | } |
| 1456 | |
| 1457 | return true; |
| 1458 | } |
| 1459 | |
| 1460 | void PixelHarness::measurePartition(int part, const EncoderPrimitives& ref, const EncoderPrimitives& opt) |
| 1461 | { |
| 1462 | ALIGN_VAR_16(int, cres[16]); |
| 1463 | pixel *fref = pbuf2 + 2 * INCR; |
| 1464 | char header[128]; |
| 1465 | #define HEADER(str, ...) sprintf(header, str, __VA_ARGS__); printf("%22s", header); |
| 1466 | |
| 1467 | if (opt.satd[part]) |
| 1468 | { |
| 1469 | HEADER("satd[%s]", lumaPartStr[part]); |
| 1470 | REPORT_SPEEDUP(opt.satd[part], ref.satd[part], pbuf1, STRIDE, fref, STRIDE); |
| 1471 | } |
| 1472 | |
| 1473 | if (opt.pixelavg_pp[part]) |
| 1474 | { |
| 1475 | HEADER("avg_pp[%s]", lumaPartStr[part]); |
| 1476 | REPORT_SPEEDUP(opt.pixelavg_pp[part], ref.pixelavg_pp[part], pbuf1, STRIDE, pbuf2, STRIDE, pbuf3, STRIDE, 32); |
| 1477 | } |
| 1478 | |
| 1479 | if (opt.sa8d_inter[part]) |
| 1480 | { |
| 1481 | HEADER("sa8d[%s]", lumaPartStr[part]); |
| 1482 | REPORT_SPEEDUP(opt.sa8d_inter[part], ref.sa8d_inter[part], pbuf1, STRIDE, fref, STRIDE); |
| 1483 | } |
| 1484 | |
| 1485 | if (opt.sad[part]) |
| 1486 | { |
| 1487 | HEADER("sad[%s]", lumaPartStr[part]); |
| 1488 | REPORT_SPEEDUP(opt.sad[part], ref.sad[part], pbuf1, STRIDE, fref, STRIDE); |
| 1489 | } |
| 1490 | |
| 1491 | if (opt.sad_x3[part]) |
| 1492 | { |
| 1493 | HEADER("sad_x3[%s]", lumaPartStr[part]); |
| 1494 | REPORT_SPEEDUP(opt.sad_x3[part], ref.sad_x3[part], pbuf1, fref, fref + 1, fref - 1, FENC_STRIDE + 5, &cres[0]); |
| 1495 | } |
| 1496 | |
| 1497 | if (opt.sad_x4[part]) |
| 1498 | { |
| 1499 | HEADER("sad_x4[%s]", lumaPartStr[part]); |
| 1500 | REPORT_SPEEDUP(opt.sad_x4[part], ref.sad_x4[part], pbuf1, fref, fref + 1, fref - 1, fref - INCR, FENC_STRIDE + 5, &cres[0]); |
| 1501 | } |
| 1502 | |
| 1503 | if (opt.sse_pp[part]) |
| 1504 | { |
| 1505 | HEADER("sse_pp[%s]", lumaPartStr[part]); |
| 1506 | REPORT_SPEEDUP(opt.sse_pp[part], ref.sse_pp[part], pbuf1, STRIDE, fref, STRIDE); |
| 1507 | } |
| 1508 | |
| 1509 | if (opt.sse_sp[part]) |
| 1510 | { |
| 1511 | HEADER("sse_sp[%s]", lumaPartStr[part]); |
| 1512 | REPORT_SPEEDUP(opt.sse_sp[part], ref.sse_sp[part], (int16_t*)pbuf1, STRIDE, fref, STRIDE); |
| 1513 | } |
| 1514 | |
| 1515 | if (opt.sse_ss[part]) |
| 1516 | { |
| 1517 | HEADER("sse_ss[%s]", lumaPartStr[part]); |
| 1518 | REPORT_SPEEDUP(opt.sse_ss[part], ref.sse_ss[part], (int16_t*)pbuf1, STRIDE, (int16_t*)fref, STRIDE); |
| 1519 | } |
| 1520 | |
| 1521 | if (opt.luma_copy_pp[part]) |
| 1522 | { |
| 1523 | HEADER("luma_copy_pp[%s]", lumaPartStr[part]); |
| 1524 | REPORT_SPEEDUP(opt.luma_copy_pp[part], ref.luma_copy_pp[part], pbuf1, 64, pbuf2, 128); |
| 1525 | } |
| 1526 | |
| 1527 | if (opt.luma_copy_sp[part]) |
| 1528 | { |
| 1529 | HEADER("luma_copy_sp[%s]", lumaPartStr[part]); |
| 1530 | REPORT_SPEEDUP(opt.luma_copy_sp[part], ref.luma_copy_sp[part], pbuf1, 64, sbuf3, 128); |
| 1531 | } |
| 1532 | |
| 1533 | if (opt.luma_copy_ps[part]) |
| 1534 | { |
| 1535 | HEADER("luma_copy_ps[%s]", lumaPartStr[part]); |
| 1536 | REPORT_SPEEDUP(opt.luma_copy_ps[part], ref.luma_copy_ps[part], sbuf1, 64, pbuf1, 128); |
| 1537 | } |
| 1538 | if (opt.luma_copy_ss[part]) |
| 1539 | { |
| 1540 | HEADER("luma_copy_ss[%s]", lumaPartStr[part]); |
| 1541 | REPORT_SPEEDUP(opt.luma_copy_ss[part], ref.luma_copy_ss[part], sbuf1, 64, sbuf2, 128); |
| 1542 | } |
| 1543 | if (opt.luma_addAvg[part]) |
| 1544 | { |
| 1545 | HEADER("luma_addAvg[%s]", lumaPartStr[part]); |
| 1546 | REPORT_SPEEDUP(opt.luma_addAvg[part], ref.luma_addAvg[part], sbuf1, sbuf2, pbuf1, STRIDE, STRIDE, STRIDE); |
| 1547 | } |
| 1548 | if (part < NUM_SQUARE_BLOCKS) |
| 1549 | { |
| 1550 | if (opt.luma_sub_ps[part]) |
| 1551 | { |
| 1552 | HEADER("luma_sub_ps[%s]", lumaPartStr[part]); |
| 1553 | REPORT_SPEEDUP(opt.luma_sub_ps[part], ref.luma_sub_ps[part], (int16_t*)pbuf1, FENC_STRIDE, pbuf2, pbuf1, STRIDE, STRIDE); |
| 1554 | } |
| 1555 | if (opt.luma_add_ps[part]) |
| 1556 | { |
| 1557 | HEADER("luma_add_ps[%s]", lumaPartStr[part]); |
| 1558 | REPORT_SPEEDUP(opt.luma_add_ps[part], ref.luma_add_ps[part], pbuf1, FENC_STRIDE, pbuf2, sbuf1, STRIDE, STRIDE); |
| 1559 | } |
| 1560 | } |
| 1561 | |
| 1562 | for (int i = 0; i < X265_CSP_COUNT; i++) |
| 1563 | { |
| 1564 | if (opt.chroma[i].copy_pp[part]) |
| 1565 | { |
| 1566 | HEADER("[%s] copy_pp[%s]", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1567 | REPORT_SPEEDUP(opt.chroma[i].copy_pp[part], ref.chroma[i].copy_pp[part], pbuf1, 64, pbuf2, 128); |
| 1568 | } |
| 1569 | if (opt.chroma[i].copy_sp[part]) |
| 1570 | { |
| 1571 | HEADER("[%s] copy_sp[%s]", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1572 | REPORT_SPEEDUP(opt.chroma[i].copy_sp[part], ref.chroma[i].copy_sp[part], pbuf1, 64, sbuf3, 128); |
| 1573 | } |
| 1574 | if (opt.chroma[i].copy_ps[part]) |
| 1575 | { |
| 1576 | HEADER("[%s] copy_ps[%s]", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1577 | REPORT_SPEEDUP(opt.chroma[i].copy_ps[part], ref.chroma[i].copy_ps[part], sbuf1, 64, pbuf1, 128); |
| 1578 | } |
| 1579 | if (opt.chroma[i].copy_ss[part]) |
| 1580 | { |
| 1581 | HEADER("[%s] copy_ss[%s]", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1582 | REPORT_SPEEDUP(opt.chroma[i].copy_ss[part], ref.chroma[i].copy_ss[part], sbuf1, 64, sbuf2, 128); |
| 1583 | } |
| 1584 | if (opt.chroma[i].addAvg[part]) |
| 1585 | { |
| 1586 | HEADER("[%s] addAvg[%s]", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1587 | REPORT_SPEEDUP(opt.chroma[i].addAvg[part], ref.chroma[i].addAvg[part], sbuf1, sbuf2, pbuf1, STRIDE, STRIDE, STRIDE); |
| 1588 | } |
| 1589 | if (part < NUM_SQUARE_BLOCKS) |
| 1590 | { |
| 1591 | if (opt.chroma[i].sub_ps[part]) |
| 1592 | { |
| 1593 | HEADER("[%s] sub_ps[%s]", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1594 | REPORT_SPEEDUP(opt.chroma[i].sub_ps[part], ref.chroma[i].sub_ps[part], (int16_t*)pbuf1, FENC_STRIDE, pbuf2, pbuf1, STRIDE, STRIDE); |
| 1595 | } |
| 1596 | if (opt.chroma[i].add_ps[part]) |
| 1597 | { |
| 1598 | HEADER("[%s] add_ps[%s]", x265_source_csp_names[i], chromaPartStr[i][part]); |
| 1599 | REPORT_SPEEDUP(opt.chroma[i].add_ps[part], ref.chroma[i].add_ps[part], pbuf1, FENC_STRIDE, pbuf2, sbuf1, STRIDE, STRIDE); |
| 1600 | } |
| 1601 | } |
| 1602 | } |
| 1603 | |
| 1604 | #undef HEADER |
| 1605 | } |
| 1606 | |
| 1607 | void PixelHarness::measureSpeed(const EncoderPrimitives& ref, const EncoderPrimitives& opt) |
| 1608 | { |
| 1609 | char header[128]; |
| 1610 | |
| 1611 | #define HEADER(str, ...) sprintf(header, str, __VA_ARGS__); printf("%22s", header); |
| 1612 | #define HEADER0(str) printf("%22s", str); |
| 1613 | |
| 1614 | for (int size = 4; size <= 64; size *= 2) |
| 1615 | { |
| 1616 | int part = partitionFromSizes(size, size); // 2Nx2N |
| 1617 | measurePartition(part, ref, opt); |
| 1618 | |
| 1619 | if (size > 4) |
| 1620 | { |
| 1621 | part = partitionFromSizes(size, size >> 1); // 2NxN |
| 1622 | measurePartition(part, ref, opt); |
| 1623 | part = partitionFromSizes(size >> 1, size); // Nx2N |
| 1624 | measurePartition(part, ref, opt); |
| 1625 | } |
| 1626 | if (size > 8) |
| 1627 | { |
| 1628 | // 4 AMP modes |
| 1629 | part = partitionFromSizes(size, size >> 2); |
| 1630 | measurePartition(part, ref, opt); |
| 1631 | part = partitionFromSizes(size, 3 * (size >> 2)); |
| 1632 | measurePartition(part, ref, opt); |
| 1633 | |
| 1634 | part = partitionFromSizes(size >> 2, size); |
| 1635 | measurePartition(part, ref, opt); |
| 1636 | part = partitionFromSizes(3 * (size >> 2), size); |
| 1637 | measurePartition(part, ref, opt); |
| 1638 | } |
| 1639 | } |
| 1640 | |
| 1641 | for (int i = 0; i < NUM_SQUARE_BLOCKS; i++) |
| 1642 | { |
| 1643 | if ((i <= BLOCK_32x32) && opt.ssd_s[i]) |
| 1644 | { |
| 1645 | HEADER("ssd_s[%dx%d]", 4 << i, 4 << i); |
| 1646 | REPORT_SPEEDUP(opt.ssd_s[i], ref.ssd_s[i], sbuf1, STRIDE); |
| 1647 | } |
| 1648 | if (opt.sa8d[i]) |
| 1649 | { |
| 1650 | HEADER("sa8d[%dx%d]", 4 << i, 4 << i); |
| 1651 | REPORT_SPEEDUP(opt.sa8d[i], ref.sa8d[i], pbuf1, STRIDE, pbuf2, STRIDE); |
| 1652 | } |
| 1653 | if (opt.calcresidual[i]) |
| 1654 | { |
| 1655 | HEADER("residual[%dx%d]", 4 << i, 4 << i); |
| 1656 | REPORT_SPEEDUP(opt.calcresidual[i], ref.calcresidual[i], pbuf1, pbuf2, sbuf1, 64); |
| 1657 | } |
| 1658 | |
| 1659 | if (opt.blockfill_s[i]) |
| 1660 | { |
| 1661 | HEADER("blkfill[%dx%d]", 4 << i, 4 << i); |
| 1662 | REPORT_SPEEDUP(opt.blockfill_s[i], ref.blockfill_s[i], sbuf1, 64, SHORT_MAX); |
| 1663 | } |
| 1664 | |
| 1665 | if (opt.transpose[i]) |
| 1666 | { |
| 1667 | HEADER("transpose[%dx%d]", 4 << i, 4 << i); |
| 1668 | REPORT_SPEEDUP(opt.transpose[i], ref.transpose[i], pbuf1, pbuf2, STRIDE); |
| 1669 | } |
| 1670 | |
| 1671 | if (opt.var[i]) |
| 1672 | { |
| 1673 | HEADER("var[%dx%d]", 4 << i, 4 << i); |
| 1674 | REPORT_SPEEDUP(opt.var[i], ref.var[i], pbuf1, STRIDE); |
| 1675 | } |
| 1676 | |
| 1677 | if ((i < BLOCK_64x64) && opt.cvt16to32_shr[i]) |
| 1678 | { |
| 1679 | HEADER("cvt16to32_shr[%dx%d]", 4 << i, 4 << i); |
| 1680 | REPORT_SPEEDUP(opt.cvt16to32_shr[i], ref.cvt16to32_shr[i], ibuf1, sbuf2, STRIDE, 3, 4); |
| 1681 | } |
| 1682 | |
| 1683 | if ((i < BLOCK_64x64) && opt.cvt32to16_shl[i]) |
| 1684 | { |
| 1685 | HEADER("cvt32to16_shl[%dx%d]", 4 << i, 4 << i); |
| 1686 | REPORT_SPEEDUP(opt.cvt32to16_shl[i], ref.cvt32to16_shl[i], sbuf2, ibuf1, STRIDE, 3); |
| 1687 | } |
| 1688 | |
| 1689 | if ((i < BLOCK_64x64) && opt.copy_cnt[i]) |
| 1690 | { |
| 1691 | HEADER("copy_cnt[%dx%d]", 4 << i, 4 << i); |
| 1692 | REPORT_SPEEDUP(opt.copy_cnt[i], ref.copy_cnt[i], sbuf1, sbuf2, STRIDE); |
| 1693 | } |
| 1694 | |
| 1695 | if ((i < BLOCK_64x64) && opt.copy_shl[i]) |
| 1696 | { |
| 1697 | HEADER("copy_shl[%dx%d]", 4 << i, 4 << i); |
| 1698 | REPORT_SPEEDUP(opt.copy_shl[i], ref.copy_shl[i], sbuf1, sbuf2, STRIDE, 64); |
| 1699 | } |
| 1700 | |
| 1701 | } |
| 1702 | |
| 1703 | if (opt.cvt32to16_shr) |
| 1704 | { |
| 1705 | HEADER0("cvt32to16_shr"); |
| 1706 | REPORT_SPEEDUP(opt.cvt32to16_shr, ref.cvt32to16_shr, sbuf1, ibuf1, 64, 5, 64); |
| 1707 | } |
| 1708 | |
| 1709 | if (opt.cvt16to32_shl) |
| 1710 | { |
| 1711 | HEADER0("cvt16to32_shl"); |
| 1712 | REPORT_SPEEDUP(opt.cvt16to32_shl, ref.cvt16to32_shl, ibuf1, sbuf1, 64, 5, 64); |
| 1713 | } |
| 1714 | |
| 1715 | if (opt.weight_pp) |
| 1716 | { |
| 1717 | HEADER0("weight_pp"); |
| 1718 | REPORT_SPEEDUP(opt.weight_pp, ref.weight_pp, pbuf1, pbuf2, 64, 32, 32, 128, 1 << 9, 10, 100); |
| 1719 | } |
| 1720 | |
| 1721 | if (opt.weight_sp) |
| 1722 | { |
| 1723 | HEADER0("weight_sp"); |
| 1724 | REPORT_SPEEDUP(opt.weight_sp, ref.weight_sp, (int16_t*)sbuf1, pbuf1, 64, 64, 32, 32, 128, 1 << 9, 10, 100); |
| 1725 | } |
| 1726 | |
| 1727 | if (opt.frame_init_lowres_core) |
| 1728 | { |
| 1729 | HEADER0("downscale"); |
| 1730 | REPORT_SPEEDUP(opt.frame_init_lowres_core, ref.frame_init_lowres_core, pbuf2, pbuf1, pbuf2, pbuf3, pbuf4, 64, 64, 64, 64); |
| 1731 | } |
| 1732 | |
| 1733 | if (opt.scale1D_128to64) |
| 1734 | { |
| 1735 | HEADER0("scale1D_128to64"); |
| 1736 | REPORT_SPEEDUP(opt.scale1D_128to64, ref.scale1D_128to64, pbuf2, pbuf1, 64); |
| 1737 | } |
| 1738 | |
| 1739 | if (opt.scale2D_64to32) |
| 1740 | { |
| 1741 | HEADER0("scale2D_64to32"); |
| 1742 | REPORT_SPEEDUP(opt.scale2D_64to32, ref.scale2D_64to32, pbuf2, pbuf1, 64); |
| 1743 | } |
| 1744 | |
| 1745 | if (opt.ssim_4x4x2_core) |
| 1746 | { |
| 1747 | HEADER0("ssim_4x4x2_core"); |
| 1748 | REPORT_SPEEDUP(opt.ssim_4x4x2_core, ref.ssim_4x4x2_core, pbuf1, 64, pbuf2, 64, (int(*)[4])sbuf1); |
| 1749 | } |
| 1750 | |
| 1751 | if (opt.ssim_end_4) |
| 1752 | { |
| 1753 | HEADER0("ssim_end_4"); |
| 1754 | REPORT_SPEEDUP(opt.ssim_end_4, ref.ssim_end_4, (int(*)[4])pbuf2, (int(*)[4])pbuf1, 4); |
| 1755 | } |
| 1756 | |
| 1757 | if (opt.saoCuOrgE0) |
| 1758 | { |
| 1759 | HEADER0("SAO_EO_0"); |
| 1760 | REPORT_SPEEDUP(opt.saoCuOrgE0, ref.saoCuOrgE0, pbuf1, psbuf1, 64, 1); |
| 1761 | } |
| 1762 | |
| 1763 | if (opt.planecopy_sp) |
| 1764 | { |
| 1765 | HEADER0("planecopy_sp"); |
| 1766 | REPORT_SPEEDUP(opt.planecopy_sp, ref.planecopy_sp, ushort_test_buff[0], 64, pbuf1, 64, 64, 64, 8, 255); |
| 1767 | } |
| 1768 | |
| 1769 | if (opt.planecopy_cp) |
| 1770 | { |
| 1771 | HEADER0("planecopy_cp"); |
| 1772 | REPORT_SPEEDUP(opt.planecopy_cp, ref.planecopy_cp, uchar_test_buff[0], 64, pbuf1, 64, 64, 64, 2); |
| 1773 | } |
| 1774 | |
| 1775 | if (opt.copy_shr) |
| 1776 | { |
| 1777 | HEADER0("copy_shr"); |
| 1778 | REPORT_SPEEDUP(opt.copy_shr, ref.copy_shr, sbuf1, sbuf2, 64, 5, 64); |
| 1779 | } |
| 1780 | |
| 1781 | } |