1/* 2 * Copyright (c) 2010 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 11#include "./vpx_config.h" 12 13#include "vpx_mem/vpx_mem.h" 14#include "vpx_ports/vpx_once.h" 15 16#include "./vp9_rtcd.h" 17 18#include "vp9/common/vp9_reconintra.h" 19#include "vp9/common/vp9_onyxc_int.h" 20 21const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES] = { 22 DCT_DCT, // DC 23 ADST_DCT, // V 24 DCT_ADST, // H 25 DCT_DCT, // D45 26 ADST_ADST, // D135 27 ADST_DCT, // D117 28 DCT_ADST, // D153 29 DCT_ADST, // D207 30 ADST_DCT, // D63 31 ADST_ADST, // TM 32}; 33 34#define intra_pred_sized(type, size) \ 35 void vp9_##type##_predictor_##size##x##size##_c(uint8_t *dst, \ 36 ptrdiff_t stride, \ 37 const uint8_t *above, \ 38 const uint8_t *left) { \ 39 type##_predictor(dst, stride, size, above, left); \ 40 } 41 42#define intra_pred_allsizes(type) \ 43 intra_pred_sized(type, 4) \ 44 intra_pred_sized(type, 8) \ 45 intra_pred_sized(type, 16) \ 46 intra_pred_sized(type, 32) 47 48static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 49 const uint8_t *above, const uint8_t *left) { 50 int r, c; 51 52 // first column 53 for (r = 0; r < bs - 1; ++r) 54 dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1); 55 dst[(bs - 1) * stride] = left[bs - 1]; 56 dst++; 57 58 // second column 59 for (r = 0; r < bs - 2; ++r) 60 dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1] * 2 + 61 left[r + 2], 2); 62 dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] + 63 left[bs - 1] * 3, 2); 64 dst[(bs - 1) * stride] = left[bs - 1]; 65 dst++; 66 67 // rest of last row 68 for (c = 0; c < bs - 2; ++c) 69 dst[(bs - 1) * stride + c] = left[bs - 1]; 70 71 for (r = bs - 2; r >= 0; --r) 72 for (c = 0; c < bs - 2; ++c) 73 dst[r * stride + c] = dst[(r + 1) * stride + c - 2]; 74} 75intra_pred_allsizes(d207) 76 77static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 78 const uint8_t *above, const uint8_t *left) { 79 int r, c; 80 for (r = 0; r < bs; ++r) { 81 for (c = 0; c < bs; ++c) 82 dst[c] = r & 1 ? ROUND_POWER_OF_TWO(above[r/2 + c] + 83 above[r/2 + c + 1] * 2 + 84 above[r/2 + c + 2], 2) 85 : ROUND_POWER_OF_TWO(above[r/2 + c] + 86 above[r/2 + c + 1], 1); 87 dst += stride; 88 } 89} 90intra_pred_allsizes(d63) 91 92static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 93 const uint8_t *above, const uint8_t *left) { 94 int r, c; 95 for (r = 0; r < bs; ++r) { 96 for (c = 0; c < bs; ++c) 97 dst[c] = r + c + 2 < bs * 2 ? ROUND_POWER_OF_TWO(above[r + c] + 98 above[r + c + 1] * 2 + 99 above[r + c + 2], 2) 100 : above[bs * 2 - 1]; 101 dst += stride; 102 } 103} 104intra_pred_allsizes(d45) 105 106static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 107 const uint8_t *above, const uint8_t *left) { 108 int r, c; 109 110 // first row 111 for (c = 0; c < bs; c++) 112 dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1); 113 dst += stride; 114 115 // second row 116 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2); 117 for (c = 1; c < bs; c++) 118 dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2); 119 dst += stride; 120 121 // the rest of first col 122 dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2); 123 for (r = 3; r < bs; ++r) 124 dst[(r - 2) * stride] = ROUND_POWER_OF_TWO(left[r - 3] + left[r - 2] * 2 + 125 left[r - 1], 2); 126 127 // the rest of the block 128 for (r = 2; r < bs; ++r) { 129 for (c = 1; c < bs; c++) 130 dst[c] = dst[-2 * stride + c - 1]; 131 dst += stride; 132 } 133} 134intra_pred_allsizes(d117) 135 136static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 137 const uint8_t *above, const uint8_t *left) { 138 int r, c; 139 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2); 140 for (c = 1; c < bs; c++) 141 dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2); 142 143 dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2); 144 for (r = 2; r < bs; ++r) 145 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 + 146 left[r], 2); 147 148 dst += stride; 149 for (r = 1; r < bs; ++r) { 150 for (c = 1; c < bs; c++) 151 dst[c] = dst[-stride + c - 1]; 152 dst += stride; 153 } 154} 155intra_pred_allsizes(d135) 156 157static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 158 const uint8_t *above, const uint8_t *left) { 159 int r, c; 160 dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1); 161 for (r = 1; r < bs; r++) 162 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1); 163 dst++; 164 165 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2); 166 dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2); 167 for (r = 2; r < bs; r++) 168 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 + 169 left[r], 2); 170 dst++; 171 172 for (c = 0; c < bs - 2; c++) 173 dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c] * 2 + above[c + 1], 2); 174 dst += stride; 175 176 for (r = 1; r < bs; ++r) { 177 for (c = 0; c < bs - 2; c++) 178 dst[c] = dst[-stride + c - 2]; 179 dst += stride; 180 } 181} 182intra_pred_allsizes(d153) 183 184static INLINE void v_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 185 const uint8_t *above, const uint8_t *left) { 186 int r; 187 188 for (r = 0; r < bs; r++) { 189 vpx_memcpy(dst, above, bs); 190 dst += stride; 191 } 192} 193intra_pred_allsizes(v) 194 195static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 196 const uint8_t *above, const uint8_t *left) { 197 int r; 198 199 for (r = 0; r < bs; r++) { 200 vpx_memset(dst, left[r], bs); 201 dst += stride; 202 } 203} 204intra_pred_allsizes(h) 205 206static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 207 const uint8_t *above, const uint8_t *left) { 208 int r, c; 209 int ytop_left = above[-1]; 210 211 for (r = 0; r < bs; r++) { 212 for (c = 0; c < bs; c++) 213 dst[c] = clip_pixel(left[r] + above[c] - ytop_left); 214 dst += stride; 215 } 216} 217intra_pred_allsizes(tm) 218 219static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 220 const uint8_t *above, const uint8_t *left) { 221 int r; 222 223 for (r = 0; r < bs; r++) { 224 vpx_memset(dst, 128, bs); 225 dst += stride; 226 } 227} 228intra_pred_allsizes(dc_128) 229 230static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 231 const uint8_t *above, 232 const uint8_t *left) { 233 int i, r, expected_dc, sum = 0; 234 235 for (i = 0; i < bs; i++) 236 sum += left[i]; 237 expected_dc = (sum + (bs >> 1)) / bs; 238 239 for (r = 0; r < bs; r++) { 240 vpx_memset(dst, expected_dc, bs); 241 dst += stride; 242 } 243} 244intra_pred_allsizes(dc_left) 245 246static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 247 const uint8_t *above, const uint8_t *left) { 248 int i, r, expected_dc, sum = 0; 249 250 for (i = 0; i < bs; i++) 251 sum += above[i]; 252 expected_dc = (sum + (bs >> 1)) / bs; 253 254 for (r = 0; r < bs; r++) { 255 vpx_memset(dst, expected_dc, bs); 256 dst += stride; 257 } 258} 259intra_pred_allsizes(dc_top) 260 261static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bs, 262 const uint8_t *above, const uint8_t *left) { 263 int i, r, expected_dc, sum = 0; 264 const int count = 2 * bs; 265 266 for (i = 0; i < bs; i++) { 267 sum += above[i]; 268 sum += left[i]; 269 } 270 271 expected_dc = (sum + (count >> 1)) / count; 272 273 for (r = 0; r < bs; r++) { 274 vpx_memset(dst, expected_dc, bs); 275 dst += stride; 276 } 277} 278intra_pred_allsizes(dc) 279#undef intra_pred_allsizes 280 281typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride, 282 const uint8_t *above, const uint8_t *left); 283 284static intra_pred_fn pred[INTRA_MODES][4]; 285static intra_pred_fn dc_pred[2][2][4]; 286 287static void init_intra_pred_fn_ptrs(void) { 288#define intra_pred_allsizes(l, type) \ 289 l[0] = vp9_##type##_predictor_4x4; \ 290 l[1] = vp9_##type##_predictor_8x8; \ 291 l[2] = vp9_##type##_predictor_16x16; \ 292 l[3] = vp9_##type##_predictor_32x32 293 294 intra_pred_allsizes(pred[V_PRED], v); 295 intra_pred_allsizes(pred[H_PRED], h); 296 intra_pred_allsizes(pred[D207_PRED], d207); 297 intra_pred_allsizes(pred[D45_PRED], d45); 298 intra_pred_allsizes(pred[D63_PRED], d63); 299 intra_pred_allsizes(pred[D117_PRED], d117); 300 intra_pred_allsizes(pred[D135_PRED], d135); 301 intra_pred_allsizes(pred[D153_PRED], d153); 302 intra_pred_allsizes(pred[TM_PRED], tm); 303 304 intra_pred_allsizes(dc_pred[0][0], dc_128); 305 intra_pred_allsizes(dc_pred[0][1], dc_top); 306 intra_pred_allsizes(dc_pred[1][0], dc_left); 307 intra_pred_allsizes(dc_pred[1][1], dc); 308 309#undef intra_pred_allsizes 310} 311 312static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref, 313 int ref_stride, uint8_t *dst, int dst_stride, 314 MB_PREDICTION_MODE mode, TX_SIZE tx_size, 315 int up_available, int left_available, 316 int right_available, int x, int y, 317 int plane) { 318 int i; 319 DECLARE_ALIGNED_ARRAY(16, uint8_t, left_col, 64); 320 DECLARE_ALIGNED_ARRAY(16, uint8_t, above_data, 128 + 16); 321 uint8_t *above_row = above_data + 16; 322 const uint8_t *const_above_row = above_row; 323 const int bs = 4 << tx_size; 324 int frame_width, frame_height; 325 int x0, y0; 326 const struct macroblockd_plane *const pd = &xd->plane[plane]; 327 328 // 127 127 127 .. 127 127 127 127 127 127 329 // 129 A B .. Y Z 330 // 129 C D .. W X 331 // 129 E F .. U V 332 // 129 G H .. S T T T T T 333 // .. 334 335 once(init_intra_pred_fn_ptrs); 336 337 // Get current frame pointer, width and height. 338 if (plane == 0) { 339 frame_width = xd->cur_buf->y_width; 340 frame_height = xd->cur_buf->y_height; 341 } else { 342 frame_width = xd->cur_buf->uv_width; 343 frame_height = xd->cur_buf->uv_height; 344 } 345 346 // Get block position in current frame. 347 x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x; 348 y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y; 349 350 vpx_memset(left_col, 129, 64); 351 352 // left 353 if (left_available) { 354 if (xd->mb_to_bottom_edge < 0) { 355 /* slower path if the block needs border extension */ 356 if (y0 + bs <= frame_height) { 357 for (i = 0; i < bs; ++i) 358 left_col[i] = ref[i * ref_stride - 1]; 359 } else { 360 const int extend_bottom = frame_height - y0; 361 for (i = 0; i < extend_bottom; ++i) 362 left_col[i] = ref[i * ref_stride - 1]; 363 for (; i < bs; ++i) 364 left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1]; 365 } 366 } else { 367 /* faster path if the block does not need extension */ 368 for (i = 0; i < bs; ++i) 369 left_col[i] = ref[i * ref_stride - 1]; 370 } 371 } 372 373 // TODO(hkuang) do not extend 2*bs pixels for all modes. 374 // above 375 if (up_available) { 376 const uint8_t *above_ref = ref - ref_stride; 377 if (xd->mb_to_right_edge < 0) { 378 /* slower path if the block needs border extension */ 379 if (x0 + 2 * bs <= frame_width) { 380 if (right_available && bs == 4) { 381 vpx_memcpy(above_row, above_ref, 2 * bs); 382 } else { 383 vpx_memcpy(above_row, above_ref, bs); 384 vpx_memset(above_row + bs, above_row[bs - 1], bs); 385 } 386 } else if (x0 + bs <= frame_width) { 387 const int r = frame_width - x0; 388 if (right_available && bs == 4) { 389 vpx_memcpy(above_row, above_ref, r); 390 vpx_memset(above_row + r, above_row[r - 1], 391 x0 + 2 * bs - frame_width); 392 } else { 393 vpx_memcpy(above_row, above_ref, bs); 394 vpx_memset(above_row + bs, above_row[bs - 1], bs); 395 } 396 } else if (x0 <= frame_width) { 397 const int r = frame_width - x0; 398 if (right_available && bs == 4) { 399 vpx_memcpy(above_row, above_ref, r); 400 vpx_memset(above_row + r, above_row[r - 1], 401 x0 + 2 * bs - frame_width); 402 } else { 403 vpx_memcpy(above_row, above_ref, r); 404 vpx_memset(above_row + r, above_row[r - 1], 405 x0 + 2 * bs - frame_width); 406 } 407 } 408 above_row[-1] = left_available ? above_ref[-1] : 129; 409 } else { 410 /* faster path if the block does not need extension */ 411 if (bs == 4 && right_available && left_available) { 412 const_above_row = above_ref; 413 } else { 414 vpx_memcpy(above_row, above_ref, bs); 415 if (bs == 4 && right_available) 416 vpx_memcpy(above_row + bs, above_ref + bs, bs); 417 else 418 vpx_memset(above_row + bs, above_row[bs - 1], bs); 419 above_row[-1] = left_available ? above_ref[-1] : 129; 420 } 421 } 422 } else { 423 vpx_memset(above_row, 127, bs * 2); 424 above_row[-1] = 127; 425 } 426 427 // predict 428 if (mode == DC_PRED) { 429 dc_pred[left_available][up_available][tx_size](dst, dst_stride, 430 const_above_row, left_col); 431 } else { 432 pred[mode][tx_size](dst, dst_stride, const_above_row, left_col); 433 } 434} 435 436void vp9_predict_intra_block(const MACROBLOCKD *xd, int block_idx, int bwl_in, 437 TX_SIZE tx_size, MB_PREDICTION_MODE mode, 438 const uint8_t *ref, int ref_stride, 439 uint8_t *dst, int dst_stride, 440 int aoff, int loff, int plane) { 441 const int bwl = bwl_in - tx_size; 442 const int wmask = (1 << bwl) - 1; 443 const int have_top = (block_idx >> bwl) || xd->up_available; 444 const int have_left = (block_idx & wmask) || xd->left_available; 445 const int have_right = ((block_idx & wmask) != wmask); 446 const int x = aoff * 4; 447 const int y = loff * 4; 448 449 assert(bwl >= 0); 450 build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size, 451 have_top, have_left, have_right, x, y, plane); 452} 453