| OLD | NEW |
|---|
| (Empty) | |
| 1 /* |
| 2 * Copyright (c) 2014 The WebM project authors. All Rights Reserved. |
| 3 * |
| 4 * Use of this source code is governed by a BSD-style license |
| 5 * that can be found in the LICENSE file in the root of the source |
| 6 * tree. An additional intellectual property rights grant can be found |
| 7 * in the file PATENTS. All contributing project authors may |
| 8 * be found in the AUTHORS file in the root of the source tree. |
| 9 */ |
| 10 #include "./vpx_config.h" |
| 11 #include "vp9/common/vp9_common.h" |
| 12 |
| 13 #include "vp9/encoder/vp9_variance.h" |
| 14 #include "vpx_ports/mem.h" |
| 15 |
| 16 typedef uint32_t (*high_variance_fn_t) (const uint16_t *src, int src_stride, |
| 17 const uint16_t *ref, int ref_stride, |
| 18 uint32_t *sse, int *sum); |
| 19 |
| 20 uint32_t vp9_highbd_calc8x8var_sse2(const uint16_t *src, int src_stride, |
| 21 const uint16_t *ref, int ref_stride, |
| 22 uint32_t *sse, int *sum); |
| 23 |
| 24 uint32_t vp9_highbd_calc16x16var_sse2(const uint16_t *src, int src_stride, |
| 25 const uint16_t *ref, int ref_stride, |
| 26 uint32_t *sse, int *sum); |
| 27 |
| 28 static void highbd_variance_sse2(const uint16_t *src, int src_stride, |
| 29 const uint16_t *ref, int ref_stride, |
| 30 int w, int h, uint32_t *sse, int *sum, |
| 31 high_variance_fn_t var_fn, int block_size) { |
| 32 int i, j; |
| 33 |
| 34 *sse = 0; |
| 35 *sum = 0; |
| 36 |
| 37 for (i = 0; i < h; i += block_size) { |
| 38 for (j = 0; j < w; j += block_size) { |
| 39 unsigned int sse0; |
| 40 int sum0; |
| 41 var_fn(src + src_stride * i + j, src_stride, |
| 42 ref + ref_stride * i + j, ref_stride, &sse0, &sum0); |
| 43 *sse += sse0; |
| 44 *sum += sum0; |
| 45 } |
| 46 } |
| 47 } |
| 48 |
| 49 static void highbd_10_variance_sse2(const uint16_t *src, int src_stride, |
| 50 const uint16_t *ref, int ref_stride, |
| 51 int w, int h, uint32_t *sse, int *sum, |
| 52 high_variance_fn_t var_fn, int block_size) { |
| 53 int i, j; |
| 54 uint64_t sse_long = 0; |
| 55 int64_t sum_long = 0; |
| 56 |
| 57 for (i = 0; i < h; i += block_size) { |
| 58 for (j = 0; j < w; j += block_size) { |
| 59 unsigned int sse0; |
| 60 int sum0; |
| 61 var_fn(src + src_stride * i + j, src_stride, |
| 62 ref + ref_stride * i + j, ref_stride, &sse0, &sum0); |
| 63 sse_long += sse0; |
| 64 sum_long += sum0; |
| 65 } |
| 66 } |
| 67 *sum = ROUND_POWER_OF_TWO(sum_long, 2); |
| 68 *sse = ROUND_POWER_OF_TWO(sse_long, 4); |
| 69 } |
| 70 |
| 71 static void highbd_12_variance_sse2(const uint16_t *src, int src_stride, |
| 72 const uint16_t *ref, int ref_stride, |
| 73 int w, int h, uint32_t *sse, int *sum, |
| 74 high_variance_fn_t var_fn, int block_size) { |
| 75 int i, j; |
| 76 uint64_t sse_long = 0; |
| 77 int64_t sum_long = 0; |
| 78 |
| 79 for (i = 0; i < h; i += block_size) { |
| 80 for (j = 0; j < w; j += block_size) { |
| 81 unsigned int sse0; |
| 82 int sum0; |
| 83 var_fn(src + src_stride * i + j, src_stride, |
| 84 ref + ref_stride * i + j, ref_stride, &sse0, &sum0); |
| 85 sse_long += sse0; |
| 86 sum_long += sum0; |
| 87 } |
| 88 } |
| 89 *sum = ROUND_POWER_OF_TWO(sum_long, 4); |
| 90 *sse = ROUND_POWER_OF_TWO(sse_long, 8); |
| 91 } |
| 92 |
| 93 |
| 94 #define HIGH_GET_VAR(S) \ |
| 95 void vp9_highbd_get##S##x##S##var_sse2(const uint8_t *src8, int src_stride, \ |
| 96 const uint8_t *ref8, int ref_stride, \ |
| 97 uint32_t *sse, int *sum) { \ |
| 98 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 99 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \ |
| 100 vp9_highbd_calc##S##x##S##var_sse2(src, src_stride, ref, ref_stride, \ |
| 101 sse, sum); \ |
| 102 } \ |
| 103 \ |
| 104 void vp9_highbd_10_get##S##x##S##var_sse2(const uint8_t *src8, int src_stride, \ |
| 105 const uint8_t *ref8, int ref_stride, \ |
| 106 uint32_t *sse, int *sum) { \ |
| 107 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 108 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \ |
| 109 vp9_highbd_calc##S##x##S##var_sse2(src, src_stride, ref, ref_stride, \ |
| 110 sse, sum); \ |
| 111 *sum = ROUND_POWER_OF_TWO(*sum, 2); \ |
| 112 *sse = ROUND_POWER_OF_TWO(*sse, 4); \ |
| 113 } \ |
| 114 \ |
| 115 void vp9_highbd_12_get##S##x##S##var_sse2(const uint8_t *src8, int src_stride, \ |
| 116 const uint8_t *ref8, int ref_stride, \ |
| 117 uint32_t *sse, int *sum) { \ |
| 118 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 119 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \ |
| 120 vp9_highbd_calc##S##x##S##var_sse2(src, src_stride, ref, ref_stride, \ |
| 121 sse, sum); \ |
| 122 *sum = ROUND_POWER_OF_TWO(*sum, 4); \ |
| 123 *sse = ROUND_POWER_OF_TWO(*sse, 8); \ |
| 124 } |
| 125 |
| 126 HIGH_GET_VAR(16); |
| 127 HIGH_GET_VAR(8); |
| 128 |
| 129 #undef HIGH_GET_VAR |
| 130 |
| 131 #define VAR_FN(w, h, block_size, shift) \ |
| 132 uint32_t vp9_highbd_variance##w##x##h##_sse2( \ |
| 133 const uint8_t *src8, int src_stride, \ |
| 134 const uint8_t *ref8, int ref_stride, uint32_t *sse) { \ |
| 135 int sum; \ |
| 136 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 137 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \ |
| 138 highbd_variance_sse2(src, src_stride, ref, ref_stride, w, h, sse, &sum, \ |
| 139 vp9_highbd_calc##block_size##x##block_size##var_sse2, \ |
| 140 block_size); \ |
| 141 return *sse - (((int64_t)sum * sum) >> shift); \ |
| 142 } \ |
| 143 \ |
| 144 uint32_t vp9_highbd_10_variance##w##x##h##_sse2( \ |
| 145 const uint8_t *src8, int src_stride, \ |
| 146 const uint8_t *ref8, int ref_stride, uint32_t *sse) { \ |
| 147 int sum; \ |
| 148 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 149 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \ |
| 150 highbd_10_variance_sse2( \ |
| 151 src, src_stride, ref, ref_stride, w, h, sse, &sum, \ |
| 152 vp9_highbd_calc##block_size##x##block_size##var_sse2, block_size); \ |
| 153 return *sse - (((int64_t)sum * sum) >> shift); \ |
| 154 } \ |
| 155 \ |
| 156 uint32_t vp9_highbd_12_variance##w##x##h##_sse2( \ |
| 157 const uint8_t *src8, int src_stride, \ |
| 158 const uint8_t *ref8, int ref_stride, uint32_t *sse) { \ |
| 159 int sum; \ |
| 160 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 161 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \ |
| 162 highbd_12_variance_sse2( \ |
| 163 src, src_stride, ref, ref_stride, w, h, sse, &sum, \ |
| 164 vp9_highbd_calc##block_size##x##block_size##var_sse2, block_size); \ |
| 165 return *sse - (((int64_t)sum * sum) >> shift); \ |
| 166 } |
| 167 |
| 168 VAR_FN(64, 64, 16, 12); |
| 169 VAR_FN(64, 32, 16, 11); |
| 170 VAR_FN(32, 64, 16, 11); |
| 171 VAR_FN(32, 32, 16, 10); |
| 172 VAR_FN(32, 16, 16, 9); |
| 173 VAR_FN(16, 32, 16, 9); |
| 174 VAR_FN(16, 16, 16, 8); |
| 175 VAR_FN(16, 8, 8, 7); |
| 176 VAR_FN(8, 16, 8, 7); |
| 177 VAR_FN(8, 8, 8, 6); |
| 178 |
| 179 #undef VAR_FN |
| 180 |
| 181 unsigned int vp9_highbd_mse16x16_sse2(const uint8_t *src8, int src_stride, |
| 182 const uint8_t *ref8, int ref_stride, |
| 183 unsigned int *sse) { |
| 184 int sum; |
| 185 uint16_t *src = CONVERT_TO_SHORTPTR(src8); |
| 186 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); |
| 187 highbd_variance_sse2(src, src_stride, ref, ref_stride, 16, 16, |
| 188 sse, &sum, vp9_highbd_calc16x16var_sse2, 16); |
| 189 return *sse; |
| 190 } |
| 191 |
| 192 unsigned int vp9_highbd_10_mse16x16_sse2(const uint8_t *src8, int src_stride, |
| 193 const uint8_t *ref8, int ref_stride, |
| 194 unsigned int *sse) { |
| 195 int sum; |
| 196 uint16_t *src = CONVERT_TO_SHORTPTR(src8); |
| 197 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); |
| 198 highbd_10_variance_sse2(src, src_stride, ref, ref_stride, 16, 16, |
| 199 sse, &sum, vp9_highbd_calc16x16var_sse2, 16); |
| 200 return *sse; |
| 201 } |
| 202 |
| 203 unsigned int vp9_highbd_12_mse16x16_sse2(const uint8_t *src8, int src_stride, |
| 204 const uint8_t *ref8, int ref_stride, |
| 205 unsigned int *sse) { |
| 206 int sum; |
| 207 uint16_t *src = CONVERT_TO_SHORTPTR(src8); |
| 208 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); |
| 209 highbd_12_variance_sse2(src, src_stride, ref, ref_stride, 16, 16, |
| 210 sse, &sum, vp9_highbd_calc16x16var_sse2, 16); |
| 211 return *sse; |
| 212 } |
| 213 |
| 214 unsigned int vp9_highbd_mse8x8_sse2(const uint8_t *src8, int src_stride, |
| 215 const uint8_t *ref8, int ref_stride, |
| 216 unsigned int *sse) { |
| 217 int sum; |
| 218 uint16_t *src = CONVERT_TO_SHORTPTR(src8); |
| 219 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); |
| 220 highbd_variance_sse2(src, src_stride, ref, ref_stride, 8, 8, |
| 221 sse, &sum, vp9_highbd_calc8x8var_sse2, 8); |
| 222 return *sse; |
| 223 } |
| 224 |
| 225 unsigned int vp9_highbd_10_mse8x8_sse2(const uint8_t *src8, int src_stride, |
| 226 const uint8_t *ref8, int ref_stride, |
| 227 unsigned int *sse) { |
| 228 int sum; |
| 229 uint16_t *src = CONVERT_TO_SHORTPTR(src8); |
| 230 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); |
| 231 highbd_10_variance_sse2(src, src_stride, ref, ref_stride, 8, 8, |
| 232 sse, &sum, vp9_highbd_calc8x8var_sse2, 8); |
| 233 return *sse; |
| 234 } |
| 235 |
| 236 unsigned int vp9_highbd_12_mse8x8_sse2(const uint8_t *src8, int src_stride, |
| 237 const uint8_t *ref8, int ref_stride, |
| 238 unsigned int *sse) { |
| 239 int sum; |
| 240 uint16_t *src = CONVERT_TO_SHORTPTR(src8); |
| 241 uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); |
| 242 highbd_12_variance_sse2(src, src_stride, ref, ref_stride, 8, 8, |
| 243 sse, &sum, vp9_highbd_calc8x8var_sse2, 8); |
| 244 return *sse; |
| 245 } |
| 246 |
| 247 #define DECL(w, opt) \ |
| 248 int vp9_highbd_sub_pixel_variance##w##xh_##opt(const uint16_t *src, \ |
| 249 ptrdiff_t src_stride, \ |
| 250 int x_offset, int y_offset, \ |
| 251 const uint16_t *dst, \ |
| 252 ptrdiff_t dst_stride, \ |
| 253 int height, unsigned int *sse); |
| 254 #define DECLS(opt1, opt2) \ |
| 255 DECL(8, opt1); \ |
| 256 DECL(16, opt1) |
| 257 |
| 258 DECLS(sse2, sse); |
| 259 // DECLS(ssse3, ssse3); |
| 260 #undef DECLS |
| 261 #undef DECL |
| 262 |
| 263 #define FN(w, h, wf, wlog2, hlog2, opt, cast) \ |
| 264 uint32_t vp9_highbd_sub_pixel_variance##w##x##h##_##opt(const uint8_t *src8, \ |
| 265 int src_stride, \ |
| 266 int x_offset, \ |
| 267 int y_offset, \ |
| 268 const uint8_t *dst8, \ |
| 269 int dst_stride, \ |
| 270 uint32_t *sse_ptr) { \ |
| 271 uint32_t sse; \ |
| 272 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 273 uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \ |
| 274 int se = vp9_highbd_sub_pixel_variance##wf##xh_##opt(src, src_stride, \ |
| 275 x_offset, y_offset, \ |
| 276 dst, dst_stride, h, \ |
| 277 &sse); \ |
| 278 if (w > wf) { \ |
| 279 unsigned int sse2; \ |
| 280 int se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt(src + 16, \ |
| 281 src_stride, \ |
| 282 x_offset, y_offset, \ |
| 283 dst + 16, \ |
| 284 dst_stride, \ |
| 285 h, &sse2); \ |
| 286 se += se2; \ |
| 287 sse += sse2; \ |
| 288 if (w > wf * 2) { \ |
| 289 se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt(src + 32, src_stride, \ |
| 290 x_offset, y_offset, \ |
| 291 dst + 32, dst_stride, \ |
| 292 h, &sse2); \ |
| 293 se += se2; \ |
| 294 sse += sse2; \ |
| 295 se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt( \ |
| 296 src + 48, src_stride, x_offset, y_offset, \ |
| 297 dst + 48, dst_stride, h, &sse2); \ |
| 298 se += se2; \ |
| 299 sse += sse2; \ |
| 300 } \ |
| 301 } \ |
| 302 *sse_ptr = sse; \ |
| 303 return sse - ((cast se * se) >> (wlog2 + hlog2)); \ |
| 304 } \ |
| 305 \ |
| 306 uint32_t vp9_highbd_10_sub_pixel_variance##w##x##h##_##opt( \ |
| 307 const uint8_t *src8, int src_stride, int x_offset, int y_offset, \ |
| 308 const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr) { \ |
| 309 uint32_t sse; \ |
| 310 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 311 uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \ |
| 312 int se = vp9_highbd_sub_pixel_variance##wf##xh_##opt(src, src_stride, \ |
| 313 x_offset, y_offset, \ |
| 314 dst, dst_stride, \ |
| 315 h, &sse); \ |
| 316 if (w > wf) { \ |
| 317 uint32_t sse2; \ |
| 318 int se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt(src + 16, \ |
| 319 src_stride, \ |
| 320 x_offset, y_offset, \ |
| 321 dst + 16, \ |
| 322 dst_stride, \ |
| 323 h, &sse2); \ |
| 324 se += se2; \ |
| 325 sse += sse2; \ |
| 326 if (w > wf * 2) { \ |
| 327 se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt(src + 32, src_stride, \ |
| 328 x_offset, y_offset, \ |
| 329 dst + 32, dst_stride, \ |
| 330 h, &sse2); \ |
| 331 se += se2; \ |
| 332 sse += sse2; \ |
| 333 se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt(src + 48, src_stride, \ |
| 334 x_offset, y_offset, \ |
| 335 dst + 48, dst_stride, \ |
| 336 h, &sse2); \ |
| 337 se += se2; \ |
| 338 sse += sse2; \ |
| 339 } \ |
| 340 } \ |
| 341 se = ROUND_POWER_OF_TWO(se, 2); \ |
| 342 sse = ROUND_POWER_OF_TWO(sse, 4); \ |
| 343 *sse_ptr = sse; \ |
| 344 return sse - ((cast se * se) >> (wlog2 + hlog2)); \ |
| 345 } \ |
| 346 \ |
| 347 uint32_t vp9_highbd_12_sub_pixel_variance##w##x##h##_##opt( \ |
| 348 const uint8_t *src8, int src_stride, int x_offset, int y_offset, \ |
| 349 const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr) { \ |
| 350 int start_row; \ |
| 351 uint32_t sse; \ |
| 352 int se = 0; \ |
| 353 uint64_t long_sse = 0; \ |
| 354 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 355 uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \ |
| 356 for (start_row = 0; start_row < h; start_row +=16) { \ |
| 357 uint32_t sse2; \ |
| 358 int height = h - start_row < 16 ? h - start_row : 16; \ |
| 359 int se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt( \ |
| 360 src + (start_row * src_stride), src_stride, \ |
| 361 x_offset, y_offset, dst + (start_row * dst_stride), \ |
| 362 dst_stride, height, &sse2); \ |
| 363 se += se2; \ |
| 364 long_sse += sse2; \ |
| 365 if (w > wf) { \ |
| 366 se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt( \ |
| 367 src + 16 + (start_row * src_stride), src_stride, \ |
| 368 x_offset, y_offset, dst + 16 + (start_row * dst_stride), \ |
| 369 dst_stride, height, &sse2); \ |
| 370 se += se2; \ |
| 371 long_sse += sse2; \ |
| 372 if (w > wf * 2) { \ |
| 373 se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt( \ |
| 374 src + 32 + (start_row * src_stride), src_stride, \ |
| 375 x_offset, y_offset, dst + 32 + (start_row * dst_stride), \ |
| 376 dst_stride, height, &sse2); \ |
| 377 se += se2; \ |
| 378 long_sse += sse2; \ |
| 379 se2 = vp9_highbd_sub_pixel_variance##wf##xh_##opt( \ |
| 380 src + 48 + (start_row * src_stride), src_stride, \ |
| 381 x_offset, y_offset, dst + 48 + (start_row * dst_stride), \ |
| 382 dst_stride, height, &sse2); \ |
| 383 se += se2; \ |
| 384 long_sse += sse2; \ |
| 385 }\ |
| 386 } \ |
| 387 } \ |
| 388 se = ROUND_POWER_OF_TWO(se, 4); \ |
| 389 sse = ROUND_POWER_OF_TWO(long_sse, 8); \ |
| 390 *sse_ptr = sse; \ |
| 391 return sse - ((cast se * se) >> (wlog2 + hlog2)); \ |
| 392 } |
| 393 |
| 394 #define FNS(opt1, opt2) \ |
| 395 FN(64, 64, 16, 6, 6, opt1, (int64_t)); \ |
| 396 FN(64, 32, 16, 6, 5, opt1, (int64_t)); \ |
| 397 FN(32, 64, 16, 5, 6, opt1, (int64_t)); \ |
| 398 FN(32, 32, 16, 5, 5, opt1, (int64_t)); \ |
| 399 FN(32, 16, 16, 5, 4, opt1, (int64_t)); \ |
| 400 FN(16, 32, 16, 4, 5, opt1, (int64_t)); \ |
| 401 FN(16, 16, 16, 4, 4, opt1, (int64_t)); \ |
| 402 FN(16, 8, 16, 4, 3, opt1, (int64_t)); \ |
| 403 FN(8, 16, 8, 3, 4, opt1, (int64_t)); \ |
| 404 FN(8, 8, 8, 3, 3, opt1, (int64_t)); \ |
| 405 FN(8, 4, 8, 3, 2, opt1, (int64_t)); |
| 406 |
| 407 |
| 408 FNS(sse2, sse); |
| 409 |
| 410 #undef FNS |
| 411 #undef FN |
| 412 |
| 413 #define DECL(w, opt) \ |
| 414 int vp9_highbd_sub_pixel_avg_variance##w##xh_##opt(const uint16_t *src, \ |
| 415 ptrdiff_t src_stride, \ |
| 416 int x_offset, int y_offset, \ |
| 417 const uint16_t *dst, \ |
| 418 ptrdiff_t dst_stride, \ |
| 419 const uint16_t *sec, \ |
| 420 ptrdiff_t sec_stride, \ |
| 421 int height, \ |
| 422 unsigned int *sse); |
| 423 #define DECLS(opt1) \ |
| 424 DECL(16, opt1) \ |
| 425 DECL(8, opt1) |
| 426 |
| 427 DECLS(sse2); |
| 428 #undef DECL |
| 429 #undef DECLS |
| 430 |
| 431 #define FN(w, h, wf, wlog2, hlog2, opt, cast) \ |
| 432 uint32_t vp9_highbd_sub_pixel_avg_variance##w##x##h##_##opt( \ |
| 433 const uint8_t *src8, int src_stride, int x_offset, int y_offset, \ |
| 434 const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr, \ |
| 435 const uint8_t *sec8) { \ |
| 436 uint32_t sse; \ |
| 437 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 438 uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \ |
| 439 uint16_t *sec = CONVERT_TO_SHORTPTR(sec8); \ |
| 440 int se = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 441 src, src_stride, x_offset, \ |
| 442 y_offset, dst, dst_stride, sec, w, h, &sse); \ |
| 443 if (w > wf) { \ |
| 444 uint32_t sse2; \ |
| 445 int se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 446 src + 16, src_stride, x_offset, y_offset, \ |
| 447 dst + 16, dst_stride, sec + 16, w, h, &sse2); \ |
| 448 se += se2; \ |
| 449 sse += sse2; \ |
| 450 if (w > wf * 2) { \ |
| 451 se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 452 src + 32, src_stride, x_offset, y_offset, \ |
| 453 dst + 32, dst_stride, sec + 32, w, h, &sse2); \ |
| 454 se += se2; \ |
| 455 sse += sse2; \ |
| 456 se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 457 src + 48, src_stride, x_offset, y_offset, \ |
| 458 dst + 48, dst_stride, sec + 48, w, h, &sse2); \ |
| 459 se += se2; \ |
| 460 sse += sse2; \ |
| 461 } \ |
| 462 } \ |
| 463 *sse_ptr = sse; \ |
| 464 return sse - ((cast se * se) >> (wlog2 + hlog2)); \ |
| 465 } \ |
| 466 \ |
| 467 uint32_t vp9_highbd_10_sub_pixel_avg_variance##w##x##h##_##opt( \ |
| 468 const uint8_t *src8, int src_stride, int x_offset, int y_offset, \ |
| 469 const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr, \ |
| 470 const uint8_t *sec8) { \ |
| 471 uint32_t sse; \ |
| 472 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 473 uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \ |
| 474 uint16_t *sec = CONVERT_TO_SHORTPTR(sec8); \ |
| 475 int se = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 476 src, src_stride, x_offset, \ |
| 477 y_offset, dst, dst_stride, \ |
| 478 sec, w, h, &sse); \ |
| 479 if (w > wf) { \ |
| 480 uint32_t sse2; \ |
| 481 int se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 482 src + 16, src_stride, \ |
| 483 x_offset, y_offset, \ |
| 484 dst + 16, dst_stride, \ |
| 485 sec + 16, w, h, &sse2); \ |
| 486 se += se2; \ |
| 487 sse += sse2; \ |
| 488 if (w > wf * 2) { \ |
| 489 se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 490 src + 32, src_stride, \ |
| 491 x_offset, y_offset, \ |
| 492 dst + 32, dst_stride, \ |
| 493 sec + 32, w, h, &sse2); \ |
| 494 se += se2; \ |
| 495 sse += sse2; \ |
| 496 se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 497 src + 48, src_stride, \ |
| 498 x_offset, y_offset, \ |
| 499 dst + 48, dst_stride, \ |
| 500 sec + 48, w, h, &sse2); \ |
| 501 se += se2; \ |
| 502 sse += sse2; \ |
| 503 } \ |
| 504 } \ |
| 505 se = ROUND_POWER_OF_TWO(se, 2); \ |
| 506 sse = ROUND_POWER_OF_TWO(sse, 4); \ |
| 507 *sse_ptr = sse; \ |
| 508 return sse - ((cast se * se) >> (wlog2 + hlog2)); \ |
| 509 } \ |
| 510 \ |
| 511 uint32_t vp9_highbd_12_sub_pixel_avg_variance##w##x##h##_##opt( \ |
| 512 const uint8_t *src8, int src_stride, int x_offset, int y_offset, \ |
| 513 const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr, \ |
| 514 const uint8_t *sec8) { \ |
| 515 int start_row; \ |
| 516 uint32_t sse; \ |
| 517 int se = 0; \ |
| 518 uint64_t long_sse = 0; \ |
| 519 uint16_t *src = CONVERT_TO_SHORTPTR(src8); \ |
| 520 uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \ |
| 521 uint16_t *sec = CONVERT_TO_SHORTPTR(sec8); \ |
| 522 for (start_row = 0; start_row < h; start_row +=16) { \ |
| 523 uint32_t sse2; \ |
| 524 int height = h - start_row < 16 ? h - start_row : 16; \ |
| 525 int se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 526 src + (start_row * src_stride), src_stride, x_offset, \ |
| 527 y_offset, dst + (start_row * dst_stride), dst_stride, \ |
| 528 sec + (start_row * w), w, height, &sse2); \ |
| 529 se += se2; \ |
| 530 long_sse += sse2; \ |
| 531 if (w > wf) { \ |
| 532 se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 533 src + 16 + (start_row * src_stride), src_stride, \ |
| 534 x_offset, y_offset, \ |
| 535 dst + 16 + (start_row * dst_stride), dst_stride, \ |
| 536 sec + 16 + (start_row * w), w, height, &sse2); \ |
| 537 se += se2; \ |
| 538 long_sse += sse2; \ |
| 539 if (w > wf * 2) { \ |
| 540 se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 541 src + 32 + (start_row * src_stride), src_stride, \ |
| 542 x_offset, y_offset, \ |
| 543 dst + 32 + (start_row * dst_stride), dst_stride, \ |
| 544 sec + 32 + (start_row * w), w, height, &sse2); \ |
| 545 se += se2; \ |
| 546 long_sse += sse2; \ |
| 547 se2 = vp9_highbd_sub_pixel_avg_variance##wf##xh_##opt( \ |
| 548 src + 48 + (start_row * src_stride), src_stride, \ |
| 549 x_offset, y_offset, \ |
| 550 dst + 48 + (start_row * dst_stride), dst_stride, \ |
| 551 sec + 48 + (start_row * w), w, height, &sse2); \ |
| 552 se += se2; \ |
| 553 long_sse += sse2; \ |
| 554 } \ |
| 555 } \ |
| 556 } \ |
| 557 se = ROUND_POWER_OF_TWO(se, 4); \ |
| 558 sse = ROUND_POWER_OF_TWO(long_sse, 8); \ |
| 559 *sse_ptr = sse; \ |
| 560 return sse - ((cast se * se) >> (wlog2 + hlog2)); \ |
| 561 } |
| 562 |
| 563 |
| 564 #define FNS(opt1) \ |
| 565 FN(64, 64, 16, 6, 6, opt1, (int64_t)); \ |
| 566 FN(64, 32, 16, 6, 5, opt1, (int64_t)); \ |
| 567 FN(32, 64, 16, 5, 6, opt1, (int64_t)); \ |
| 568 FN(32, 32, 16, 5, 5, opt1, (int64_t)); \ |
| 569 FN(32, 16, 16, 5, 4, opt1, (int64_t)); \ |
| 570 FN(16, 32, 16, 4, 5, opt1, (int64_t)); \ |
| 571 FN(16, 16, 16, 4, 4, opt1, (int64_t)); \ |
| 572 FN(16, 8, 16, 4, 3, opt1, (int64_t)); \ |
| 573 FN(8, 16, 8, 4, 3, opt1, (int64_t)); \ |
| 574 FN(8, 8, 8, 3, 3, opt1, (int64_t)); \ |
| 575 FN(8, 4, 8, 3, 2, opt1, (int64_t)); |
| 576 |
| 577 FNS(sse2); |
| 578 |
| 579 #undef FNS |
| 580 #undef FN |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 756673003:
libvpx: Pull from upstream (Closed)
Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/libvpx/