vp8dsp.c (34933B)
1 /* 2 * Copyright (C) 2010 David Conrad 3 * Copyright (C) 2010 Ronald S. Bultje 4 * Copyright (C) 2014 Peter Ross 5 * 6 * This file is part of FFmpeg. 7 * 8 * FFmpeg is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU Lesser General Public 10 * License as published by the Free Software Foundation; either 11 * version 2.1 of the License, or (at your option) any later version. 12 * 13 * FFmpeg is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 16 * Lesser General Public License for more details. 17 * 18 * You should have received a copy of the GNU Lesser General Public 19 * License along with FFmpeg; if not, write to the Free Software 20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 21 */ 22 23 /** 24 * @file 25 * VP8 compatible video decoder 26 */ 27 28 #include "config_components.h" 29 30 #include "libavutil/common.h" 31 #include "libavutil/intreadwrite.h" 32 33 #include "mathops.h" 34 #include "vp8dsp.h" 35 36 #define MK_IDCT_DC_ADD4_C(name) \ 37 static void name ## _idct_dc_add4uv_c(uint8_t *dst, int16_t block[4][16], \ 38 ptrdiff_t stride) \ 39 { \ 40 name ## _idct_dc_add_c(dst + stride * 0 + 0, block[0], stride); \ 41 name ## _idct_dc_add_c(dst + stride * 0 + 4, block[1], stride); \ 42 name ## _idct_dc_add_c(dst + stride * 4 + 0, block[2], stride); \ 43 name ## _idct_dc_add_c(dst + stride * 4 + 4, block[3], stride); \ 44 } \ 45 \ 46 static void name ## _idct_dc_add4y_c(uint8_t *dst, int16_t block[4][16], \ 47 ptrdiff_t stride) \ 48 { \ 49 name ## _idct_dc_add_c(dst + 0, block[0], stride); \ 50 name ## _idct_dc_add_c(dst + 4, block[1], stride); \ 51 name ## _idct_dc_add_c(dst + 8, block[2], stride); \ 52 name ## _idct_dc_add_c(dst + 12, block[3], stride); \ 53 } 54 55 #if CONFIG_VP7_DECODER 56 static void vp7_luma_dc_wht_c(int16_t block[4][4][16], int16_t dc[16]) 57 { 58 int i; 59 unsigned a1, b1, c1, d1; 60 int16_t tmp[16]; 61 62 for (i = 0; i < 4; i++) { 63 a1 = (dc[i * 4 + 0] + dc[i * 4 + 2]) * 23170; 64 b1 = (dc[i * 4 + 0] - dc[i * 4 + 2]) * 23170; 65 c1 = dc[i * 4 + 1] * 12540 - dc[i * 4 + 3] * 30274; 66 d1 = dc[i * 4 + 1] * 30274 + dc[i * 4 + 3] * 12540; 67 tmp[i * 4 + 0] = (int)(a1 + d1) >> 14; 68 tmp[i * 4 + 3] = (int)(a1 - d1) >> 14; 69 tmp[i * 4 + 1] = (int)(b1 + c1) >> 14; 70 tmp[i * 4 + 2] = (int)(b1 - c1) >> 14; 71 } 72 73 for (i = 0; i < 4; i++) { 74 a1 = (tmp[i + 0] + tmp[i + 8]) * 23170; 75 b1 = (tmp[i + 0] - tmp[i + 8]) * 23170; 76 c1 = tmp[i + 4] * 12540 - tmp[i + 12] * 30274; 77 d1 = tmp[i + 4] * 30274 + tmp[i + 12] * 12540; 78 AV_ZERO64(dc + i * 4); 79 block[0][i][0] = (int)(a1 + d1 + 0x20000) >> 18; 80 block[3][i][0] = (int)(a1 - d1 + 0x20000) >> 18; 81 block[1][i][0] = (int)(b1 + c1 + 0x20000) >> 18; 82 block[2][i][0] = (int)(b1 - c1 + 0x20000) >> 18; 83 } 84 } 85 86 static void vp7_luma_dc_wht_dc_c(int16_t block[4][4][16], int16_t dc[16]) 87 { 88 int i, val = (23170 * (23170 * dc[0] >> 14) + 0x20000) >> 18; 89 dc[0] = 0; 90 91 for (i = 0; i < 4; i++) { 92 block[i][0][0] = val; 93 block[i][1][0] = val; 94 block[i][2][0] = val; 95 block[i][3][0] = val; 96 } 97 } 98 99 static void vp7_idct_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride) 100 { 101 int i; 102 unsigned a1, b1, c1, d1; 103 int16_t tmp[16]; 104 105 for (i = 0; i < 4; i++) { 106 a1 = (block[i * 4 + 0] + block[i * 4 + 2]) * 23170; 107 b1 = (block[i * 4 + 0] - block[i * 4 + 2]) * 23170; 108 c1 = block[i * 4 + 1] * 12540 - block[i * 4 + 3] * 30274; 109 d1 = block[i * 4 + 1] * 30274 + block[i * 4 + 3] * 12540; 110 AV_ZERO64(block + i * 4); 111 tmp[i * 4 + 0] = (int)(a1 + d1) >> 14; 112 tmp[i * 4 + 3] = (int)(a1 - d1) >> 14; 113 tmp[i * 4 + 1] = (int)(b1 + c1) >> 14; 114 tmp[i * 4 + 2] = (int)(b1 - c1) >> 14; 115 } 116 117 for (i = 0; i < 4; i++) { 118 a1 = (tmp[i + 0] + tmp[i + 8]) * 23170; 119 b1 = (tmp[i + 0] - tmp[i + 8]) * 23170; 120 c1 = tmp[i + 4] * 12540 - tmp[i + 12] * 30274; 121 d1 = tmp[i + 4] * 30274 + tmp[i + 12] * 12540; 122 dst[0 * stride + i] = av_clip_uint8(dst[0 * stride + i] + 123 ((int)(a1 + d1 + 0x20000) >> 18)); 124 dst[3 * stride + i] = av_clip_uint8(dst[3 * stride + i] + 125 ((int)(a1 - d1 + 0x20000) >> 18)); 126 dst[1 * stride + i] = av_clip_uint8(dst[1 * stride + i] + 127 ((int)(b1 + c1 + 0x20000) >> 18)); 128 dst[2 * stride + i] = av_clip_uint8(dst[2 * stride + i] + 129 ((int)(b1 - c1 + 0x20000) >> 18)); 130 } 131 } 132 133 static void vp7_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride) 134 { 135 int i, dc = (23170 * (23170 * block[0] >> 14) + 0x20000) >> 18; 136 block[0] = 0; 137 138 for (i = 0; i < 4; i++) { 139 dst[0] = av_clip_uint8(dst[0] + dc); 140 dst[1] = av_clip_uint8(dst[1] + dc); 141 dst[2] = av_clip_uint8(dst[2] + dc); 142 dst[3] = av_clip_uint8(dst[3] + dc); 143 dst += stride; 144 } 145 } 146 147 MK_IDCT_DC_ADD4_C(vp7) 148 #endif /* CONFIG_VP7_DECODER */ 149 150 // TODO: Maybe add dequant 151 #if CONFIG_VP8_DECODER 152 static void vp8_luma_dc_wht_c(int16_t block[4][4][16], int16_t dc[16]) 153 { 154 int i, t0, t1, t2, t3; 155 156 for (i = 0; i < 4; i++) { 157 t0 = dc[0 * 4 + i] + dc[3 * 4 + i]; 158 t1 = dc[1 * 4 + i] + dc[2 * 4 + i]; 159 t2 = dc[1 * 4 + i] - dc[2 * 4 + i]; 160 t3 = dc[0 * 4 + i] - dc[3 * 4 + i]; 161 162 dc[0 * 4 + i] = t0 + t1; 163 dc[1 * 4 + i] = t3 + t2; 164 dc[2 * 4 + i] = t0 - t1; 165 dc[3 * 4 + i] = t3 - t2; 166 } 167 168 for (i = 0; i < 4; i++) { 169 t0 = dc[i * 4 + 0] + dc[i * 4 + 3] + 3; // rounding 170 t1 = dc[i * 4 + 1] + dc[i * 4 + 2]; 171 t2 = dc[i * 4 + 1] - dc[i * 4 + 2]; 172 t3 = dc[i * 4 + 0] - dc[i * 4 + 3] + 3; // rounding 173 AV_ZERO64(dc + i * 4); 174 175 block[i][0][0] = (t0 + t1) >> 3; 176 block[i][1][0] = (t3 + t2) >> 3; 177 block[i][2][0] = (t0 - t1) >> 3; 178 block[i][3][0] = (t3 - t2) >> 3; 179 } 180 } 181 182 static void vp8_luma_dc_wht_dc_c(int16_t block[4][4][16], int16_t dc[16]) 183 { 184 int i, val = (dc[0] + 3) >> 3; 185 dc[0] = 0; 186 187 for (i = 0; i < 4; i++) { 188 block[i][0][0] = val; 189 block[i][1][0] = val; 190 block[i][2][0] = val; 191 block[i][3][0] = val; 192 } 193 } 194 195 #define MUL_20091(a) ((((a) * 20091) >> 16) + (a)) 196 #define MUL_35468(a) (((a) * 35468) >> 16) 197 198 static void vp8_idct_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride) 199 { 200 int i, t0, t1, t2, t3; 201 int16_t tmp[16]; 202 203 for (i = 0; i < 4; i++) { 204 t0 = block[0 * 4 + i] + block[2 * 4 + i]; 205 t1 = block[0 * 4 + i] - block[2 * 4 + i]; 206 t2 = MUL_35468(block[1 * 4 + i]) - MUL_20091(block[3 * 4 + i]); 207 t3 = MUL_20091(block[1 * 4 + i]) + MUL_35468(block[3 * 4 + i]); 208 block[0 * 4 + i] = 0; 209 block[1 * 4 + i] = 0; 210 block[2 * 4 + i] = 0; 211 block[3 * 4 + i] = 0; 212 213 tmp[i * 4 + 0] = t0 + t3; 214 tmp[i * 4 + 1] = t1 + t2; 215 tmp[i * 4 + 2] = t1 - t2; 216 tmp[i * 4 + 3] = t0 - t3; 217 } 218 219 for (i = 0; i < 4; i++) { 220 t0 = tmp[0 * 4 + i] + tmp[2 * 4 + i]; 221 t1 = tmp[0 * 4 + i] - tmp[2 * 4 + i]; 222 t2 = MUL_35468(tmp[1 * 4 + i]) - MUL_20091(tmp[3 * 4 + i]); 223 t3 = MUL_20091(tmp[1 * 4 + i]) + MUL_35468(tmp[3 * 4 + i]); 224 225 dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3)); 226 dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3)); 227 dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3)); 228 dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3)); 229 dst += stride; 230 } 231 } 232 233 static void vp8_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride) 234 { 235 int i, dc = (block[0] + 4) >> 3; 236 block[0] = 0; 237 238 for (i = 0; i < 4; i++) { 239 dst[0] = av_clip_uint8(dst[0] + dc); 240 dst[1] = av_clip_uint8(dst[1] + dc); 241 dst[2] = av_clip_uint8(dst[2] + dc); 242 dst[3] = av_clip_uint8(dst[3] + dc); 243 dst += stride; 244 } 245 } 246 247 MK_IDCT_DC_ADD4_C(vp8) 248 #endif /* CONFIG_VP8_DECODER */ 249 250 // because I like only having two parameters to pass functions... 251 #define LOAD_PIXELS \ 252 int av_unused p3 = p[-4 * stride]; \ 253 int av_unused p2 = p[-3 * stride]; \ 254 int av_unused p1 = p[-2 * stride]; \ 255 int av_unused p0 = p[-1 * stride]; \ 256 int av_unused q0 = p[ 0 * stride]; \ 257 int av_unused q1 = p[ 1 * stride]; \ 258 int av_unused q2 = p[ 2 * stride]; \ 259 int av_unused q3 = p[ 3 * stride]; 260 261 #define clip_int8(n) (cm[(n) + 0x80] - 0x80) 262 263 static av_always_inline void filter_common(uint8_t *p, ptrdiff_t stride, 264 int is4tap, int is_vp7) 265 { 266 LOAD_PIXELS 267 int a, f1, f2; 268 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP; 269 270 a = 3 * (q0 - p0); 271 272 if (is4tap) 273 a += clip_int8(p1 - q1); 274 275 a = clip_int8(a); 276 277 // We deviate from the spec here with c(a+3) >> 3 278 // since that's what libvpx does. 279 f1 = FFMIN(a + 4, 127) >> 3; 280 281 if (is_vp7) 282 f2 = f1 - ((a & 7) == 4); 283 else 284 f2 = FFMIN(a + 3, 127) >> 3; 285 286 // Despite what the spec says, we do need to clamp here to 287 // be bitexact with libvpx. 288 p[-1 * stride] = cm[p0 + f2]; 289 p[ 0 * stride] = cm[q0 - f1]; 290 291 // only used for _inner on blocks without high edge variance 292 if (!is4tap) { 293 a = (f1 + 1) >> 1; 294 p[-2 * stride] = cm[p1 + a]; 295 p[ 1 * stride] = cm[q1 - a]; 296 } 297 } 298 299 static av_always_inline void vp7_filter_common(uint8_t *p, ptrdiff_t stride, 300 int is4tap) 301 { 302 filter_common(p, stride, is4tap, IS_VP7); 303 } 304 305 static av_always_inline void vp8_filter_common(uint8_t *p, ptrdiff_t stride, 306 int is4tap) 307 { 308 filter_common(p, stride, is4tap, IS_VP8); 309 } 310 311 static av_always_inline int vp7_simple_limit(uint8_t *p, ptrdiff_t stride, 312 int flim) 313 { 314 LOAD_PIXELS 315 return FFABS(p0 - q0) <= flim; 316 } 317 318 static av_always_inline int vp8_simple_limit(uint8_t *p, ptrdiff_t stride, 319 int flim) 320 { 321 LOAD_PIXELS 322 return 2 * FFABS(p0 - q0) + (FFABS(p1 - q1) >> 1) <= flim; 323 } 324 325 /** 326 * E - limit at the macroblock edge 327 * I - limit for interior difference 328 */ 329 #define NORMAL_LIMIT(vpn) \ 330 static av_always_inline int vp ## vpn ## _normal_limit(uint8_t *p, \ 331 ptrdiff_t stride, \ 332 int E, int I) \ 333 { \ 334 LOAD_PIXELS \ 335 return vp ## vpn ## _simple_limit(p, stride, E) && \ 336 FFABS(p3 - p2) <= I && FFABS(p2 - p1) <= I && \ 337 FFABS(p1 - p0) <= I && FFABS(q3 - q2) <= I && \ 338 FFABS(q2 - q1) <= I && FFABS(q1 - q0) <= I; \ 339 } 340 341 NORMAL_LIMIT(7) 342 NORMAL_LIMIT(8) 343 344 // high edge variance 345 static av_always_inline int hev(uint8_t *p, ptrdiff_t stride, int thresh) 346 { 347 LOAD_PIXELS 348 return FFABS(p1 - p0) > thresh || FFABS(q1 - q0) > thresh; 349 } 350 351 static av_always_inline void filter_mbedge(uint8_t *p, ptrdiff_t stride) 352 { 353 int a0, a1, a2, w; 354 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP; 355 356 LOAD_PIXELS 357 358 w = clip_int8(p1 - q1); 359 w = clip_int8(w + 3 * (q0 - p0)); 360 361 a0 = (27 * w + 63) >> 7; 362 a1 = (18 * w + 63) >> 7; 363 a2 = (9 * w + 63) >> 7; 364 365 p[-3 * stride] = cm[p2 + a2]; 366 p[-2 * stride] = cm[p1 + a1]; 367 p[-1 * stride] = cm[p0 + a0]; 368 p[ 0 * stride] = cm[q0 - a0]; 369 p[ 1 * stride] = cm[q1 - a1]; 370 p[ 2 * stride] = cm[q2 - a2]; 371 } 372 373 #define LOOP_FILTER(vpn, dir, size, stridea, strideb, maybe_inline) \ 374 static maybe_inline \ 375 void vpn ## _ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, \ 376 ptrdiff_t stride, \ 377 int flim_E, int flim_I, \ 378 int hev_thresh) \ 379 { \ 380 int i; \ 381 for (i = 0; i < size; i++) \ 382 if (vpn ## _normal_limit(dst + i * stridea, strideb, \ 383 flim_E, flim_I)) { \ 384 if (hev(dst + i * stridea, strideb, hev_thresh)) \ 385 vpn ## _filter_common(dst + i * stridea, strideb, 1); \ 386 else \ 387 filter_mbedge(dst + i * stridea, strideb); \ 388 } \ 389 } \ 390 \ 391 static maybe_inline \ 392 void vpn ## _ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, \ 393 ptrdiff_t stride, \ 394 int flim_E, \ 395 int flim_I, \ 396 int hev_thresh) \ 397 { \ 398 int i; \ 399 for (i = 0; i < size; i++) \ 400 if (vpn ## _normal_limit(dst + i * stridea, strideb, \ 401 flim_E, flim_I)) { \ 402 int hv = hev(dst + i * stridea, strideb, hev_thresh); \ 403 if (hv) \ 404 vpn ## _filter_common(dst + i * stridea, strideb, 1); \ 405 else \ 406 vpn ## _filter_common(dst + i * stridea, strideb, 0); \ 407 } \ 408 } 409 410 #define UV_LOOP_FILTER(vpn, dir, stridea, strideb) \ 411 LOOP_FILTER(vpn, dir, 8, stridea, strideb, av_always_inline) \ 412 static void vpn ## _ ## dir ## _loop_filter8uv_c(uint8_t *dstU, \ 413 uint8_t *dstV, \ 414 ptrdiff_t stride, int fE, \ 415 int fI, int hev_thresh) \ 416 { \ 417 vpn ## _ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh); \ 418 vpn ## _ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh); \ 419 } \ 420 \ 421 static void vpn ## _ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU, \ 422 uint8_t *dstV, \ 423 ptrdiff_t stride, \ 424 int fE, int fI, \ 425 int hev_thresh) \ 426 { \ 427 vpn ## _ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI, \ 428 hev_thresh); \ 429 vpn ## _ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI, \ 430 hev_thresh); \ 431 } 432 433 #define LOOP_FILTER_SIMPLE(vpn) \ 434 static void vpn ## _v_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride, \ 435 int flim) \ 436 { \ 437 int i; \ 438 for (i = 0; i < 16; i++) \ 439 if (vpn ## _simple_limit(dst + i, stride, flim)) \ 440 vpn ## _filter_common(dst + i, stride, 1); \ 441 } \ 442 \ 443 static void vpn ## _h_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride, \ 444 int flim) \ 445 { \ 446 int i; \ 447 for (i = 0; i < 16; i++) \ 448 if (vpn ## _simple_limit(dst + i * stride, 1, flim)) \ 449 vpn ## _filter_common(dst + i * stride, 1, 1); \ 450 } 451 452 #define LOOP_FILTERS(vpn) \ 453 LOOP_FILTER(vpn, v, 16, 1, stride, ) \ 454 LOOP_FILTER(vpn, h, 16, stride, 1, ) \ 455 UV_LOOP_FILTER(vpn, v, 1, stride) \ 456 UV_LOOP_FILTER(vpn, h, stride, 1) \ 457 LOOP_FILTER_SIMPLE(vpn) \ 458 459 static const uint8_t subpel_filters[7][6] = { 460 { 0, 6, 123, 12, 1, 0 }, 461 { 2, 11, 108, 36, 8, 1 }, 462 { 0, 9, 93, 50, 6, 0 }, 463 { 3, 16, 77, 77, 16, 3 }, 464 { 0, 6, 50, 93, 9, 0 }, 465 { 1, 8, 36, 108, 11, 2 }, 466 { 0, 1, 12, 123, 6, 0 }, 467 }; 468 469 #define PUT_PIXELS(WIDTH) \ 470 static void put_vp8_pixels ## WIDTH ## _c(uint8_t *dst, ptrdiff_t dststride, \ 471 const uint8_t *src, ptrdiff_t srcstride, \ 472 int h, int x, int y) \ 473 { \ 474 int i; \ 475 for (i = 0; i < h; i++, dst += dststride, src += srcstride) \ 476 memcpy(dst, src, WIDTH); \ 477 } 478 479 PUT_PIXELS(16) 480 PUT_PIXELS(8) 481 PUT_PIXELS(4) 482 483 #define FILTER_6TAP(src, F, stride) \ 484 cm[(F[2] * src[x + 0 * stride] - F[1] * src[x - 1 * stride] + \ 485 F[0] * src[x - 2 * stride] + F[3] * src[x + 1 * stride] - \ 486 F[4] * src[x + 2 * stride] + F[5] * src[x + 3 * stride] + 64) >> 7] 487 488 #define FILTER_4TAP(src, F, stride) \ 489 cm[(F[2] * src[x + 0 * stride] - F[1] * src[x - 1 * stride] + \ 490 F[3] * src[x + 1 * stride] - F[4] * src[x + 2 * stride] + 64) >> 7] 491 492 #define VP8_EPEL_H(SIZE, TAPS) \ 493 static void put_vp8_epel ## SIZE ## _h ## TAPS ## _c(uint8_t *dst, \ 494 ptrdiff_t dststride, \ 495 const uint8_t *src, \ 496 ptrdiff_t srcstride, \ 497 int h, int mx, int my) \ 498 { \ 499 const uint8_t *filter = subpel_filters[mx - 1]; \ 500 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP; \ 501 int x, y; \ 502 for (y = 0; y < h; y++) { \ 503 for (x = 0; x < SIZE; x++) \ 504 dst[x] = FILTER_ ## TAPS ## TAP(src, filter, 1); \ 505 dst += dststride; \ 506 src += srcstride; \ 507 } \ 508 } 509 510 #define VP8_EPEL_V(SIZE, TAPS) \ 511 static void put_vp8_epel ## SIZE ## _v ## TAPS ## _c(uint8_t *dst, \ 512 ptrdiff_t dststride, \ 513 const uint8_t *src, \ 514 ptrdiff_t srcstride, \ 515 int h, int mx, int my) \ 516 { \ 517 const uint8_t *filter = subpel_filters[my - 1]; \ 518 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP; \ 519 int x, y; \ 520 for (y = 0; y < h; y++) { \ 521 for (x = 0; x < SIZE; x++) \ 522 dst[x] = FILTER_ ## TAPS ## TAP(src, filter, srcstride); \ 523 dst += dststride; \ 524 src += srcstride; \ 525 } \ 526 } 527 528 #define VP8_EPEL_HV(SIZE, HTAPS, VTAPS) \ 529 static void \ 530 put_vp8_epel ## SIZE ## _h ## HTAPS ## v ## VTAPS ## _c(uint8_t *dst, \ 531 ptrdiff_t dststride, \ 532 const uint8_t *src, \ 533 ptrdiff_t srcstride, \ 534 int h, int mx, \ 535 int my) \ 536 { \ 537 const uint8_t *filter = subpel_filters[mx - 1]; \ 538 const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP; \ 539 int x, y; \ 540 uint8_t tmp_array[(2 * SIZE + VTAPS - 1) * SIZE]; \ 541 uint8_t *tmp = tmp_array; \ 542 src -= (2 - (VTAPS == 4)) * srcstride; \ 543 \ 544 for (y = 0; y < h + VTAPS - 1; y++) { \ 545 for (x = 0; x < SIZE; x++) \ 546 tmp[x] = FILTER_ ## HTAPS ## TAP(src, filter, 1); \ 547 tmp += SIZE; \ 548 src += srcstride; \ 549 } \ 550 tmp = tmp_array + (2 - (VTAPS == 4)) * SIZE; \ 551 filter = subpel_filters[my - 1]; \ 552 \ 553 for (y = 0; y < h; y++) { \ 554 for (x = 0; x < SIZE; x++) \ 555 dst[x] = FILTER_ ## VTAPS ## TAP(tmp, filter, SIZE); \ 556 dst += dststride; \ 557 tmp += SIZE; \ 558 } \ 559 } 560 561 VP8_EPEL_H(16, 4) 562 VP8_EPEL_H(8, 4) 563 VP8_EPEL_H(4, 4) 564 VP8_EPEL_H(16, 6) 565 VP8_EPEL_H(8, 6) 566 VP8_EPEL_H(4, 6) 567 VP8_EPEL_V(16, 4) 568 VP8_EPEL_V(8, 4) 569 VP8_EPEL_V(4, 4) 570 VP8_EPEL_V(16, 6) 571 VP8_EPEL_V(8, 6) 572 VP8_EPEL_V(4, 6) 573 574 VP8_EPEL_HV(16, 4, 4) 575 VP8_EPEL_HV(8, 4, 4) 576 VP8_EPEL_HV(4, 4, 4) 577 VP8_EPEL_HV(16, 4, 6) 578 VP8_EPEL_HV(8, 4, 6) 579 VP8_EPEL_HV(4, 4, 6) 580 VP8_EPEL_HV(16, 6, 4) 581 VP8_EPEL_HV(8, 6, 4) 582 VP8_EPEL_HV(4, 6, 4) 583 VP8_EPEL_HV(16, 6, 6) 584 VP8_EPEL_HV(8, 6, 6) 585 VP8_EPEL_HV(4, 6, 6) 586 587 #define VP8_BILINEAR(SIZE) \ 588 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, ptrdiff_t dstride, \ 589 const uint8_t *src, ptrdiff_t sstride, \ 590 int h, int mx, int my) \ 591 { \ 592 int a = 8 - mx, b = mx; \ 593 int x, y; \ 594 for (y = 0; y < h; y++) { \ 595 for (x = 0; x < SIZE; x++) \ 596 dst[x] = (a * src[x] + b * src[x + 1] + 4) >> 3; \ 597 dst += dstride; \ 598 src += sstride; \ 599 } \ 600 } \ 601 \ 602 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, ptrdiff_t dstride, \ 603 const uint8_t *src, ptrdiff_t sstride, \ 604 int h, int mx, int my) \ 605 { \ 606 int c = 8 - my, d = my; \ 607 int x, y; \ 608 for (y = 0; y < h; y++) { \ 609 for (x = 0; x < SIZE; x++) \ 610 dst[x] = (c * src[x] + d * src[x + sstride] + 4) >> 3; \ 611 dst += dstride; \ 612 src += sstride; \ 613 } \ 614 } \ 615 \ 616 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst, \ 617 ptrdiff_t dstride, \ 618 const uint8_t *src, \ 619 ptrdiff_t sstride, \ 620 int h, int mx, int my) \ 621 { \ 622 int a = 8 - mx, b = mx; \ 623 int c = 8 - my, d = my; \ 624 int x, y; \ 625 uint8_t tmp_array[(2 * SIZE + 1) * SIZE]; \ 626 uint8_t *tmp = tmp_array; \ 627 for (y = 0; y < h + 1; y++) { \ 628 for (x = 0; x < SIZE; x++) \ 629 tmp[x] = (a * src[x] + b * src[x + 1] + 4) >> 3; \ 630 tmp += SIZE; \ 631 src += sstride; \ 632 } \ 633 tmp = tmp_array; \ 634 for (y = 0; y < h; y++) { \ 635 for (x = 0; x < SIZE; x++) \ 636 dst[x] = (c * tmp[x] + d * tmp[x + SIZE] + 4) >> 3; \ 637 dst += dstride; \ 638 tmp += SIZE; \ 639 } \ 640 } 641 642 VP8_BILINEAR(16) 643 VP8_BILINEAR(8) 644 VP8_BILINEAR(4) 645 646 #define VP78_MC_FUNC(IDX, SIZE) \ 647 dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \ 648 dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c; \ 649 dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c; \ 650 dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c; \ 651 dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \ 652 dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \ 653 dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c; \ 654 dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \ 655 dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c 656 657 #define VP78_BILINEAR_MC_FUNC(IDX, SIZE) \ 658 dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c; \ 659 dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \ 660 dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \ 661 dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \ 662 dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \ 663 dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \ 664 dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \ 665 dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \ 666 dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c 667 668 av_cold void ff_vp78dsp_init(VP8DSPContext *dsp) 669 { 670 VP78_MC_FUNC(0, 16); 671 VP78_MC_FUNC(1, 8); 672 VP78_MC_FUNC(2, 4); 673 674 VP78_BILINEAR_MC_FUNC(0, 16); 675 VP78_BILINEAR_MC_FUNC(1, 8); 676 VP78_BILINEAR_MC_FUNC(2, 4); 677 678 #if ARCH_AARCH64 679 ff_vp78dsp_init_aarch64(dsp); 680 #elif ARCH_ARM 681 ff_vp78dsp_init_arm(dsp); 682 #elif ARCH_PPC 683 ff_vp78dsp_init_ppc(dsp); 684 #elif ARCH_RISCV 685 ff_vp78dsp_init_riscv(dsp); 686 #elif ARCH_X86 687 ff_vp78dsp_init_x86(dsp); 688 #endif 689 } 690 691 #if CONFIG_VP7_DECODER 692 LOOP_FILTERS(vp7) 693 694 av_cold void ff_vp7dsp_init(VP8DSPContext *dsp) 695 { 696 dsp->vp8_luma_dc_wht = vp7_luma_dc_wht_c; 697 dsp->vp8_luma_dc_wht_dc = vp7_luma_dc_wht_dc_c; 698 dsp->vp8_idct_add = vp7_idct_add_c; 699 dsp->vp8_idct_dc_add = vp7_idct_dc_add_c; 700 dsp->vp8_idct_dc_add4y = vp7_idct_dc_add4y_c; 701 dsp->vp8_idct_dc_add4uv = vp7_idct_dc_add4uv_c; 702 703 dsp->vp8_v_loop_filter16y = vp7_v_loop_filter16_c; 704 dsp->vp8_h_loop_filter16y = vp7_h_loop_filter16_c; 705 dsp->vp8_v_loop_filter8uv = vp7_v_loop_filter8uv_c; 706 dsp->vp8_h_loop_filter8uv = vp7_h_loop_filter8uv_c; 707 708 dsp->vp8_v_loop_filter16y_inner = vp7_v_loop_filter16_inner_c; 709 dsp->vp8_h_loop_filter16y_inner = vp7_h_loop_filter16_inner_c; 710 dsp->vp8_v_loop_filter8uv_inner = vp7_v_loop_filter8uv_inner_c; 711 dsp->vp8_h_loop_filter8uv_inner = vp7_h_loop_filter8uv_inner_c; 712 713 dsp->vp8_v_loop_filter_simple = vp7_v_loop_filter_simple_c; 714 dsp->vp8_h_loop_filter_simple = vp7_h_loop_filter_simple_c; 715 716 #if ARCH_RISCV 717 ff_vp7dsp_init_riscv(dsp); 718 #endif 719 } 720 #endif /* CONFIG_VP7_DECODER */ 721 722 #if CONFIG_VP8_DECODER 723 LOOP_FILTERS(vp8) 724 725 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp) 726 { 727 dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c; 728 dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c; 729 dsp->vp8_idct_add = vp8_idct_add_c; 730 dsp->vp8_idct_dc_add = vp8_idct_dc_add_c; 731 dsp->vp8_idct_dc_add4y = vp8_idct_dc_add4y_c; 732 dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c; 733 734 dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c; 735 dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c; 736 dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c; 737 dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c; 738 739 dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c; 740 dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c; 741 dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c; 742 dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c; 743 744 dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c; 745 dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c; 746 747 #if ARCH_AARCH64 748 ff_vp8dsp_init_aarch64(dsp); 749 #elif ARCH_ARM 750 ff_vp8dsp_init_arm(dsp); 751 #elif ARCH_RISCV 752 ff_vp8dsp_init_riscv(dsp); 753 #elif ARCH_X86 754 ff_vp8dsp_init_x86(dsp); 755 #elif ARCH_MIPS 756 ff_vp8dsp_init_mips(dsp); 757 #elif ARCH_LOONGARCH 758 ff_vp8dsp_init_loongarch(dsp); 759 #endif 760 } 761 #endif /* CONFIG_VP8_DECODER */