1// Copyright 2011 Google Inc. All Rights Reserved. 2// 3// This code is licensed under the same terms as WebM: 4// Software License Agreement: http://www.webmproject.org/license/software/ 5// Additional IP Rights Grant: http://www.webmproject.org/license/additional/ 6// ----------------------------------------------------------------------------- 7// 8// VP8Iterator: block iterator 9// 10// Author: Skal (pascal.massimino@gmail.com) 11 12#include <string.h> 13 14#include "./vp8enci.h" 15 16#if defined(__cplusplus) || defined(c_plusplus) 17extern "C" { 18#endif 19 20//------------------------------------------------------------------------------ 21// VP8Iterator 22//------------------------------------------------------------------------------ 23 24static void InitLeft(VP8EncIterator* const it) { 25 const VP8Encoder* const enc = it->enc_; 26 enc->y_left_[-1] = enc->u_left_[-1] = enc->v_left_[-1] = 27 (it->y_ > 0) ? 129 : 127; 28 memset(enc->y_left_, 129, 16); 29 memset(enc->u_left_, 129, 8); 30 memset(enc->v_left_, 129, 8); 31 it->left_nz_[8] = 0; 32} 33 34static void InitTop(VP8EncIterator* const it) { 35 const VP8Encoder* const enc = it->enc_; 36 const size_t top_size = enc->mb_w_ * 16; 37 memset(enc->y_top_, 127, 2 * top_size); 38 memset(enc->nz_, 0, enc->mb_w_ * sizeof(*enc->nz_)); 39} 40 41void VP8IteratorReset(VP8EncIterator* const it) { 42 VP8Encoder* const enc = it->enc_; 43 it->x_ = 0; 44 it->y_ = 0; 45 it->y_offset_ = 0; 46 it->uv_offset_ = 0; 47 it->mb_ = enc->mb_info_; 48 it->preds_ = enc->preds_; 49 it->nz_ = enc->nz_; 50 it->bw_ = &enc->parts_[0]; 51 it->done_ = enc->mb_w_* enc->mb_h_; 52 InitTop(it); 53 InitLeft(it); 54 memset(it->bit_count_, 0, sizeof(it->bit_count_)); 55 it->do_trellis_ = 0; 56} 57 58void VP8IteratorInit(VP8Encoder* const enc, VP8EncIterator* const it) { 59 it->enc_ = enc; 60 it->y_stride_ = enc->pic_->y_stride; 61 it->uv_stride_ = enc->pic_->uv_stride; 62 // TODO(later): for multithreading, these should be owned by 'it'. 63 it->yuv_in_ = enc->yuv_in_; 64 it->yuv_out_ = enc->yuv_out_; 65 it->yuv_out2_ = enc->yuv_out2_; 66 it->yuv_p_ = enc->yuv_p_; 67 it->lf_stats_ = enc->lf_stats_; 68 it->percent0_ = enc->percent_; 69 VP8IteratorReset(it); 70} 71 72int VP8IteratorProgress(const VP8EncIterator* const it, int delta) { 73 VP8Encoder* const enc = it->enc_; 74 if (delta && enc->pic_->progress_hook) { 75 const int percent = (enc->mb_h_ <= 1) 76 ? it->percent0_ 77 : it->percent0_ + delta * it->y_ / (enc->mb_h_ - 1); 78 return WebPReportProgress(enc->pic_, percent, &enc->percent_); 79 } 80 return 1; 81} 82 83//------------------------------------------------------------------------------ 84// Import the source samples into the cache. Takes care of replicating 85// boundary pixels if necessary. 86 87static void ImportBlock(const uint8_t* src, int src_stride, 88 uint8_t* dst, int w, int h, int size) { 89 int i; 90 for (i = 0; i < h; ++i) { 91 memcpy(dst, src, w); 92 if (w < size) { 93 memset(dst + w, dst[w - 1], size - w); 94 } 95 dst += BPS; 96 src += src_stride; 97 } 98 for (i = h; i < size; ++i) { 99 memcpy(dst, dst - BPS, size); 100 dst += BPS; 101 } 102} 103 104void VP8IteratorImport(const VP8EncIterator* const it) { 105 const VP8Encoder* const enc = it->enc_; 106 const int x = it->x_, y = it->y_; 107 const WebPPicture* const pic = enc->pic_; 108 const uint8_t* const ysrc = pic->y + (y * pic->y_stride + x) * 16; 109 const uint8_t* const usrc = pic->u + (y * pic->uv_stride + x) * 8; 110 const uint8_t* const vsrc = pic->v + (y * pic->uv_stride + x) * 8; 111 uint8_t* const ydst = it->yuv_in_ + Y_OFF; 112 uint8_t* const udst = it->yuv_in_ + U_OFF; 113 uint8_t* const vdst = it->yuv_in_ + V_OFF; 114 int w = (pic->width - x * 16); 115 int h = (pic->height - y * 16); 116 117 if (w > 16) w = 16; 118 if (h > 16) h = 16; 119 120 // Luma plane 121 ImportBlock(ysrc, pic->y_stride, ydst, w, h, 16); 122 123 { // U/V planes 124 const int uv_w = (w + 1) >> 1; 125 const int uv_h = (h + 1) >> 1; 126 ImportBlock(usrc, pic->uv_stride, udst, uv_w, uv_h, 8); 127 ImportBlock(vsrc, pic->uv_stride, vdst, uv_w, uv_h, 8); 128 } 129} 130 131//------------------------------------------------------------------------------ 132// Copy back the compressed samples into user space if requested. 133 134static void ExportBlock(const uint8_t* src, uint8_t* dst, int dst_stride, 135 int w, int h) { 136 while (h-- > 0) { 137 memcpy(dst, src, w); 138 dst += dst_stride; 139 src += BPS; 140 } 141} 142 143void VP8IteratorExport(const VP8EncIterator* const it) { 144 const VP8Encoder* const enc = it->enc_; 145 if (enc->config_->show_compressed) { 146 const int x = it->x_, y = it->y_; 147 const uint8_t* const ysrc = it->yuv_out_ + Y_OFF; 148 const uint8_t* const usrc = it->yuv_out_ + U_OFF; 149 const uint8_t* const vsrc = it->yuv_out_ + V_OFF; 150 const WebPPicture* const pic = enc->pic_; 151 uint8_t* const ydst = pic->y + (y * pic->y_stride + x) * 16; 152 uint8_t* const udst = pic->u + (y * pic->uv_stride + x) * 8; 153 uint8_t* const vdst = pic->v + (y * pic->uv_stride + x) * 8; 154 int w = (pic->width - x * 16); 155 int h = (pic->height - y * 16); 156 157 if (w > 16) w = 16; 158 if (h > 16) h = 16; 159 160 // Luma plane 161 ExportBlock(ysrc, ydst, pic->y_stride, w, h); 162 163 { // U/V planes 164 const int uv_w = (w + 1) >> 1; 165 const int uv_h = (h + 1) >> 1; 166 ExportBlock(usrc, udst, pic->uv_stride, uv_w, uv_h); 167 ExportBlock(vsrc, vdst, pic->uv_stride, uv_w, uv_h); 168 } 169 } 170} 171 172//------------------------------------------------------------------------------ 173// Non-zero contexts setup/teardown 174 175// Nz bits: 176// 0 1 2 3 Y 177// 4 5 6 7 178// 8 9 10 11 179// 12 13 14 15 180// 16 17 U 181// 18 19 182// 20 21 V 183// 22 23 184// 24 DC-intra16 185 186// Convert packed context to byte array 187#define BIT(nz, n) (!!((nz) & (1 << (n)))) 188 189void VP8IteratorNzToBytes(VP8EncIterator* const it) { 190 const int tnz = it->nz_[0], lnz = it->nz_[-1]; 191 int* const top_nz = it->top_nz_; 192 int* const left_nz = it->left_nz_; 193 194 // Top-Y 195 top_nz[0] = BIT(tnz, 12); 196 top_nz[1] = BIT(tnz, 13); 197 top_nz[2] = BIT(tnz, 14); 198 top_nz[3] = BIT(tnz, 15); 199 // Top-U 200 top_nz[4] = BIT(tnz, 18); 201 top_nz[5] = BIT(tnz, 19); 202 // Top-V 203 top_nz[6] = BIT(tnz, 22); 204 top_nz[7] = BIT(tnz, 23); 205 // DC 206 top_nz[8] = BIT(tnz, 24); 207 208 // left-Y 209 left_nz[0] = BIT(lnz, 3); 210 left_nz[1] = BIT(lnz, 7); 211 left_nz[2] = BIT(lnz, 11); 212 left_nz[3] = BIT(lnz, 15); 213 // left-U 214 left_nz[4] = BIT(lnz, 17); 215 left_nz[5] = BIT(lnz, 19); 216 // left-V 217 left_nz[6] = BIT(lnz, 21); 218 left_nz[7] = BIT(lnz, 23); 219 // left-DC is special, iterated separately 220} 221 222void VP8IteratorBytesToNz(VP8EncIterator* const it) { 223 uint32_t nz = 0; 224 const int* const top_nz = it->top_nz_; 225 const int* const left_nz = it->left_nz_; 226 // top 227 nz |= (top_nz[0] << 12) | (top_nz[1] << 13); 228 nz |= (top_nz[2] << 14) | (top_nz[3] << 15); 229 nz |= (top_nz[4] << 18) | (top_nz[5] << 19); 230 nz |= (top_nz[6] << 22) | (top_nz[7] << 23); 231 nz |= (top_nz[8] << 24); // we propagate the _top_ bit, esp. for intra4 232 // left 233 nz |= (left_nz[0] << 3) | (left_nz[1] << 7); 234 nz |= (left_nz[2] << 11); 235 nz |= (left_nz[4] << 17) | (left_nz[6] << 21); 236 237 *it->nz_ = nz; 238} 239 240#undef BIT 241 242//------------------------------------------------------------------------------ 243// Advance to the next position, doing the bookeeping. 244 245int VP8IteratorNext(VP8EncIterator* const it, 246 const uint8_t* const block_to_save) { 247 VP8Encoder* const enc = it->enc_; 248 if (block_to_save) { 249 const int x = it->x_, y = it->y_; 250 const uint8_t* const ysrc = block_to_save + Y_OFF; 251 const uint8_t* const usrc = block_to_save + U_OFF; 252 if (x < enc->mb_w_ - 1) { // left 253 int i; 254 for (i = 0; i < 16; ++i) { 255 enc->y_left_[i] = ysrc[15 + i * BPS]; 256 } 257 for (i = 0; i < 8; ++i) { 258 enc->u_left_[i] = usrc[7 + i * BPS]; 259 enc->v_left_[i] = usrc[15 + i * BPS]; 260 } 261 // top-left (before 'top'!) 262 enc->y_left_[-1] = enc->y_top_[x * 16 + 15]; 263 enc->u_left_[-1] = enc->uv_top_[x * 16 + 0 + 7]; 264 enc->v_left_[-1] = enc->uv_top_[x * 16 + 8 + 7]; 265 } 266 if (y < enc->mb_h_ - 1) { // top 267 memcpy(enc->y_top_ + x * 16, ysrc + 15 * BPS, 16); 268 memcpy(enc->uv_top_ + x * 16, usrc + 7 * BPS, 8 + 8); 269 } 270 } 271 272 it->mb_++; 273 it->preds_ += 4; 274 it->nz_++; 275 it->x_++; 276 if (it->x_ == enc->mb_w_) { 277 it->x_ = 0; 278 it->y_++; 279 it->bw_ = &enc->parts_[it->y_ & (enc->num_parts_ - 1)]; 280 it->preds_ = enc->preds_ + it->y_ * 4 * enc->preds_w_; 281 it->nz_ = enc->nz_; 282 InitLeft(it); 283 } 284 return (0 < --it->done_); 285} 286 287//------------------------------------------------------------------------------ 288// Helper function to set mode properties 289 290void VP8SetIntra16Mode(const VP8EncIterator* const it, int mode) { 291 uint8_t* preds = it->preds_; 292 int y; 293 for (y = 0; y < 4; ++y) { 294 memset(preds, mode, 4); 295 preds += it->enc_->preds_w_; 296 } 297 it->mb_->type_ = 1; 298} 299 300void VP8SetIntra4Mode(const VP8EncIterator* const it, const uint8_t* modes) { 301 uint8_t* preds = it->preds_; 302 int y; 303 for (y = 4; y > 0; --y) { 304 memcpy(preds, modes, 4 * sizeof(*modes)); 305 preds += it->enc_->preds_w_; 306 modes += 4; 307 } 308 it->mb_->type_ = 0; 309} 310 311void VP8SetIntraUVMode(const VP8EncIterator* const it, int mode) { 312 it->mb_->uv_mode_ = mode; 313} 314 315void VP8SetSkip(const VP8EncIterator* const it, int skip) { 316 it->mb_->skip_ = skip; 317} 318 319void VP8SetSegment(const VP8EncIterator* const it, int segment) { 320 it->mb_->segment_ = segment; 321} 322 323//------------------------------------------------------------------------------ 324// Intra4x4 sub-blocks iteration 325// 326// We store and update the boundary samples into an array of 37 pixels. They 327// are updated as we iterate and reconstructs each intra4x4 blocks in turn. 328// The position of the samples has the following snake pattern: 329// 330// 16|17 18 19 20|21 22 23 24|25 26 27 28|29 30 31 32|33 34 35 36 <- Top-right 331// --+-----------+-----------+-----------+-----------+ 332// 15| 19| 23| 27| 31| 333// 14| 18| 22| 26| 30| 334// 13| 17| 21| 25| 29| 335// 12|13 14 15 16|17 18 19 20|21 22 23 24|25 26 27 28| 336// --+-----------+-----------+-----------+-----------+ 337// 11| 15| 19| 23| 27| 338// 10| 14| 18| 22| 26| 339// 9| 13| 17| 21| 25| 340// 8| 9 10 11 12|13 14 15 16|17 18 19 20|21 22 23 24| 341// --+-----------+-----------+-----------+-----------+ 342// 7| 11| 15| 19| 23| 343// 6| 10| 14| 18| 22| 344// 5| 9| 13| 17| 21| 345// 4| 5 6 7 8| 9 10 11 12|13 14 15 16|17 18 19 20| 346// --+-----------+-----------+-----------+-----------+ 347// 3| 7| 11| 15| 19| 348// 2| 6| 10| 14| 18| 349// 1| 5| 9| 13| 17| 350// 0| 1 2 3 4| 5 6 7 8| 9 10 11 12|13 14 15 16| 351// --+-----------+-----------+-----------+-----------+ 352 353// Array to record the position of the top sample to pass to the prediction 354// functions in dsp.c. 355static const uint8_t VP8TopLeftI4[16] = { 356 17, 21, 25, 29, 357 13, 17, 21, 25, 358 9, 13, 17, 21, 359 5, 9, 13, 17 360}; 361 362void VP8IteratorStartI4(VP8EncIterator* const it) { 363 const VP8Encoder* const enc = it->enc_; 364 int i; 365 366 it->i4_ = 0; // first 4x4 sub-block 367 it->i4_top_ = it->i4_boundary_ + VP8TopLeftI4[0]; 368 369 // Import the boundary samples 370 for (i = 0; i < 17; ++i) { // left 371 it->i4_boundary_[i] = enc->y_left_[15 - i]; 372 } 373 for (i = 0; i < 16; ++i) { // top 374 it->i4_boundary_[17 + i] = enc->y_top_[it->x_ * 16 + i]; 375 } 376 // top-right samples have a special case on the far right of the picture 377 if (it->x_ < enc->mb_w_ - 1) { 378 for (i = 16; i < 16 + 4; ++i) { 379 it->i4_boundary_[17 + i] = enc->y_top_[it->x_ * 16 + i]; 380 } 381 } else { // else, replicate the last valid pixel four times 382 for (i = 16; i < 16 + 4; ++i) { 383 it->i4_boundary_[17 + i] = it->i4_boundary_[17 + 15]; 384 } 385 } 386 VP8IteratorNzToBytes(it); // import the non-zero context 387} 388 389int VP8IteratorRotateI4(VP8EncIterator* const it, 390 const uint8_t* const yuv_out) { 391 const uint8_t* const blk = yuv_out + VP8Scan[it->i4_]; 392 uint8_t* const top = it->i4_top_; 393 int i; 394 395 // Update the cache with 7 fresh samples 396 for (i = 0; i <= 3; ++i) { 397 top[-4 + i] = blk[i + 3 * BPS]; // store future top samples 398 } 399 if ((it->i4_ & 3) != 3) { // if not on the right sub-blocks #3, #7, #11, #15 400 for (i = 0; i <= 2; ++i) { // store future left samples 401 top[i] = blk[3 + (2 - i) * BPS]; 402 } 403 } else { // else replicate top-right samples, as says the specs. 404 for (i = 0; i <= 3; ++i) { 405 top[i] = top[i + 4]; 406 } 407 } 408 // move pointers to next sub-block 409 ++it->i4_; 410 if (it->i4_ == 16) { // we're done 411 return 0; 412 } 413 414 it->i4_top_ = it->i4_boundary_ + VP8TopLeftI4[it->i4_]; 415 return 1; 416} 417 418//------------------------------------------------------------------------------ 419 420#if defined(__cplusplus) || defined(c_plusplus) 421} // extern "C" 422#endif 423