| 1 | /***************************************************************************** |
| 2 | * Copyright (C) 2014 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 "common.h" |
| 25 | #include "frame.h" |
| 26 | #include "framedata.h" |
| 27 | #include "picyuv.h" |
| 28 | #include "mv.h" |
| 29 | #include "cudata.h" |
| 30 | |
| 31 | using namespace x265; |
| 32 | |
| 33 | namespace { |
| 34 | // file private namespace |
| 35 | |
| 36 | /* for all bcast* and copy* functions, dst and src are aligned to MIN(size, 32) */ |
| 37 | |
| 38 | void bcast1(uint8_t* dst, uint8_t val) { dst[0] = val; } |
| 39 | |
| 40 | void copy4(uint8_t* dst, uint8_t* src) { ((uint32_t*)dst)[0] = ((uint32_t*)src)[0]; } |
| 41 | void bcast4(uint8_t* dst, uint8_t val) { ((uint32_t*)dst)[0] = 0x01010101 * val; } |
| 42 | |
| 43 | void copy16(uint8_t* dst, uint8_t* src) { ((uint64_t*)dst)[0] = ((uint64_t*)src)[0]; ((uint64_t*)dst)[1] = ((uint64_t*)src)[1]; } |
| 44 | void bcast16(uint8_t* dst, uint8_t val) { uint64_t bval = 0x0101010101010101ULL * val; ((uint64_t*)dst)[0] = bval; ((uint64_t*)dst)[1] = bval; } |
| 45 | |
| 46 | void copy64(uint8_t* dst, uint8_t* src) { ((uint64_t*)dst)[0] = ((uint64_t*)src)[0]; ((uint64_t*)dst)[1] = ((uint64_t*)src)[1]; |
| 47 | ((uint64_t*)dst)[2] = ((uint64_t*)src)[2]; ((uint64_t*)dst)[3] = ((uint64_t*)src)[3]; |
| 48 | ((uint64_t*)dst)[4] = ((uint64_t*)src)[4]; ((uint64_t*)dst)[5] = ((uint64_t*)src)[5]; |
| 49 | ((uint64_t*)dst)[6] = ((uint64_t*)src)[6]; ((uint64_t*)dst)[7] = ((uint64_t*)src)[7]; } |
| 50 | void bcast64(uint8_t* dst, uint8_t val) { uint64_t bval = 0x0101010101010101ULL * val; |
| 51 | ((uint64_t*)dst)[0] = bval; ((uint64_t*)dst)[1] = bval; ((uint64_t*)dst)[2] = bval; ((uint64_t*)dst)[3] = bval; |
| 52 | ((uint64_t*)dst)[4] = bval; ((uint64_t*)dst)[5] = bval; ((uint64_t*)dst)[6] = bval; ((uint64_t*)dst)[7] = bval; } |
| 53 | |
| 54 | /* at 256 bytes, memset/memcpy will probably use SIMD more effectively than our uint64_t hack, |
| 55 | * but hand-written assembly would beat it. */ |
| 56 | void copy256(uint8_t* dst, uint8_t* src) { memcpy(dst, src, 256); } |
| 57 | void bcast256(uint8_t* dst, uint8_t val) { memset(dst, val, 256); } |
| 58 | |
| 59 | /* Check whether 2 addresses point to the same column */ |
| 60 | inline bool isEqualCol(int addrA, int addrB, int numUnitsPerRow) |
| 61 | { |
| 62 | // addrA % numUnitsPerRow == addrB % numUnitsPerRow |
| 63 | return ((addrA ^ addrB) & (numUnitsPerRow - 1)) == 0; |
| 64 | } |
| 65 | |
| 66 | /* Check whether 2 addresses point to the same row */ |
| 67 | inline bool isEqualRow(int addrA, int addrB, int numUnitsPerRow) |
| 68 | { |
| 69 | // addrA / numUnitsPerRow == addrB / numUnitsPerRow |
| 70 | return ((addrA ^ addrB) & ~(numUnitsPerRow - 1)) == 0; |
| 71 | } |
| 72 | |
| 73 | /* Check whether 2 addresses point to the same row or column */ |
| 74 | inline bool isEqualRowOrCol(int addrA, int addrB, int numUnitsPerRow) |
| 75 | { |
| 76 | return isEqualCol(addrA, addrB, numUnitsPerRow) | isEqualRow(addrA, addrB, numUnitsPerRow); |
| 77 | } |
| 78 | |
| 79 | /* Check whether one address points to the first column */ |
| 80 | inline bool isZeroCol(int addr, int numUnitsPerRow) |
| 81 | { |
| 82 | // addr % numUnitsPerRow == 0 |
| 83 | return (addr & (numUnitsPerRow - 1)) == 0; |
| 84 | } |
| 85 | |
| 86 | /* Check whether one address points to the first row */ |
| 87 | inline bool isZeroRow(int addr, int numUnitsPerRow) |
| 88 | { |
| 89 | // addr / numUnitsPerRow == 0 |
| 90 | return (addr & ~(numUnitsPerRow - 1)) == 0; |
| 91 | } |
| 92 | |
| 93 | /* Check whether one address points to a column whose index is smaller than a given value */ |
| 94 | inline bool lessThanCol(int addr, int val, int numUnitsPerRow) |
| 95 | { |
| 96 | // addr % numUnitsPerRow < val |
| 97 | return (addr & (numUnitsPerRow - 1)) < val; |
| 98 | } |
| 99 | |
| 100 | /* Check whether one address points to a row whose index is smaller than a given value */ |
| 101 | inline bool lessThanRow(int addr, int val, int numUnitsPerRow) |
| 102 | { |
| 103 | // addr / numUnitsPerRow < val |
| 104 | return addr < val * numUnitsPerRow; |
| 105 | } |
| 106 | |
| 107 | inline MV scaleMv(MV mv, int scale) |
| 108 | { |
| 109 | int mvx = Clip3(-32768, 32767, (scale * mv.x + 127 + (scale * mv.x < 0)) >> 8); |
| 110 | int mvy = Clip3(-32768, 32767, (scale * mv.y + 127 + (scale * mv.y < 0)) >> 8); |
| 111 | |
| 112 | return MV((int16_t)mvx, (int16_t)mvy); |
| 113 | } |
| 114 | |
| 115 | // Partition table. |
| 116 | // First index is partitioning mode. Second index is partition index. |
| 117 | // Third index is 0 for partition sizes, 1 for partition offsets. The |
| 118 | // sizes and offsets are encoded as two packed 4-bit values (X,Y). |
| 119 | // X and Y represent 1/4 fractions of the block size. |
| 120 | const uint32_t partTable[8][4][2] = |
| 121 | { |
| 122 | // XY |
| 123 | { { 0x44, 0x00 }, { 0x00, 0x00 }, { 0x00, 0x00 }, { 0x00, 0x00 } }, // SIZE_2Nx2N. |
| 124 | { { 0x42, 0x00 }, { 0x42, 0x02 }, { 0x00, 0x00 }, { 0x00, 0x00 } }, // SIZE_2NxN. |
| 125 | { { 0x24, 0x00 }, { 0x24, 0x20 }, { 0x00, 0x00 }, { 0x00, 0x00 } }, // SIZE_Nx2N. |
| 126 | { { 0x22, 0x00 }, { 0x22, 0x20 }, { 0x22, 0x02 }, { 0x22, 0x22 } }, // SIZE_NxN. |
| 127 | { { 0x41, 0x00 }, { 0x43, 0x01 }, { 0x00, 0x00 }, { 0x00, 0x00 } }, // SIZE_2NxnU. |
| 128 | { { 0x43, 0x00 }, { 0x41, 0x03 }, { 0x00, 0x00 }, { 0x00, 0x00 } }, // SIZE_2NxnD. |
| 129 | { { 0x14, 0x00 }, { 0x34, 0x10 }, { 0x00, 0x00 }, { 0x00, 0x00 } }, // SIZE_nLx2N. |
| 130 | { { 0x34, 0x00 }, { 0x14, 0x30 }, { 0x00, 0x00 }, { 0x00, 0x00 } } // SIZE_nRx2N. |
| 131 | }; |
| 132 | |
| 133 | // Partition Address table. |
| 134 | // First index is partitioning mode. Second index is partition address. |
| 135 | const uint32_t partAddrTable[8][4] = |
| 136 | { |
| 137 | { 0x00, 0x00, 0x00, 0x00 }, // SIZE_2Nx2N. |
| 138 | { 0x00, 0x08, 0x08, 0x08 }, // SIZE_2NxN. |
| 139 | { 0x00, 0x04, 0x04, 0x04 }, // SIZE_Nx2N. |
| 140 | { 0x00, 0x04, 0x08, 0x0C }, // SIZE_NxN. |
| 141 | { 0x00, 0x02, 0x02, 0x02 }, // SIZE_2NxnU. |
| 142 | { 0x00, 0x0A, 0x0A, 0x0A }, // SIZE_2NxnD. |
| 143 | { 0x00, 0x01, 0x01, 0x01 }, // SIZE_nLx2N. |
| 144 | { 0x00, 0x05, 0x05, 0x05 } // SIZE_nRx2N. |
| 145 | }; |
| 146 | |
| 147 | } |
| 148 | |
| 149 | cubcast_t CUData::s_partSet[NUM_FULL_DEPTH] = { NULL, NULL, NULL, NULL, NULL }; |
| 150 | uint32_t CUData::s_numPartInCUSize; |
| 151 | |
| 152 | CUData::CUData() |
| 153 | { |
| 154 | memset(this, 0, sizeof(*this)); |
| 155 | } |
| 156 | |
| 157 | void CUData::initialize(const CUDataMemPool& dataPool, uint32_t depth, int csp, int instance) |
| 158 | { |
| 159 | m_chromaFormat = csp; |
| 160 | m_hChromaShift = CHROMA_H_SHIFT(csp); |
| 161 | m_vChromaShift = CHROMA_V_SHIFT(csp); |
| 162 | m_numPartitions = NUM_CU_PARTITIONS >> (depth * 2); |
| 163 | |
| 164 | if (!s_partSet[0]) |
| 165 | { |
| 166 | s_numPartInCUSize = 1 << g_maxFullDepth; |
| 167 | switch (g_maxLog2CUSize) |
| 168 | { |
| 169 | case 6: |
| 170 | s_partSet[0] = bcast256; |
| 171 | s_partSet[1] = bcast64; |
| 172 | s_partSet[2] = bcast16; |
| 173 | s_partSet[3] = bcast4; |
| 174 | s_partSet[4] = bcast1; |
| 175 | break; |
| 176 | case 5: |
| 177 | s_partSet[0] = bcast64; |
| 178 | s_partSet[1] = bcast16; |
| 179 | s_partSet[2] = bcast4; |
| 180 | s_partSet[3] = bcast1; |
| 181 | s_partSet[4] = NULL; |
| 182 | break; |
| 183 | case 4: |
| 184 | s_partSet[0] = bcast16; |
| 185 | s_partSet[1] = bcast4; |
| 186 | s_partSet[2] = bcast1; |
| 187 | s_partSet[3] = NULL; |
| 188 | s_partSet[4] = NULL; |
| 189 | break; |
| 190 | default: |
| 191 | X265_CHECK(0, "unexpected CTU size\n"); |
| 192 | break; |
| 193 | } |
| 194 | } |
| 195 | |
| 196 | switch (m_numPartitions) |
| 197 | { |
| 198 | case 256: // 64x64 CU |
| 199 | m_partCopy = copy256; |
| 200 | m_partSet = bcast256; |
| 201 | m_subPartCopy = copy64; |
| 202 | m_subPartSet = bcast64; |
| 203 | break; |
| 204 | case 64: // 32x32 CU |
| 205 | m_partCopy = copy64; |
| 206 | m_partSet = bcast64; |
| 207 | m_subPartCopy = copy16; |
| 208 | m_subPartSet = bcast16; |
| 209 | break; |
| 210 | case 16: // 16x16 CU |
| 211 | m_partCopy = copy16; |
| 212 | m_partSet = bcast16; |
| 213 | m_subPartCopy = copy4; |
| 214 | m_subPartSet = bcast4; |
| 215 | break; |
| 216 | case 4: // 8x8 CU |
| 217 | m_partCopy = copy4; |
| 218 | m_partSet = bcast4; |
| 219 | m_subPartCopy = NULL; |
| 220 | m_subPartSet = NULL; |
| 221 | break; |
| 222 | default: |
| 223 | X265_CHECK(0, "unexpected CU partition count\n"); |
| 224 | break; |
| 225 | } |
| 226 | |
| 227 | /* Each CU's data is layed out sequentially within the charMemBlock */ |
| 228 | uint8_t *charBuf = dataPool.charMemBlock + (m_numPartitions * BytesPerPartition) * instance; |
| 229 | |
| 230 | m_qp = (char*)charBuf; charBuf += m_numPartitions; |
| 231 | m_log2CUSize = charBuf; charBuf += m_numPartitions; |
| 232 | m_partSize = charBuf; charBuf += m_numPartitions; |
| 233 | m_predMode = charBuf; charBuf += m_numPartitions; |
| 234 | m_lumaIntraDir = charBuf; charBuf += m_numPartitions; |
| 235 | m_tqBypass = charBuf; charBuf += m_numPartitions; |
| 236 | m_refIdx[0] = (char*)charBuf; charBuf += m_numPartitions; |
| 237 | m_refIdx[1] = (char*)charBuf; charBuf += m_numPartitions; |
| 238 | m_cuDepth = charBuf; charBuf += m_numPartitions; |
| 239 | m_skipFlag = charBuf; charBuf += m_numPartitions; /* the order up to here is important in initCTU() and initSubCU() */ |
| 240 | m_mergeFlag = charBuf; charBuf += m_numPartitions; |
| 241 | m_interDir = charBuf; charBuf += m_numPartitions; |
| 242 | m_mvpIdx[0] = charBuf; charBuf += m_numPartitions; |
| 243 | m_mvpIdx[1] = charBuf; charBuf += m_numPartitions; |
| 244 | m_tuDepth = charBuf; charBuf += m_numPartitions; |
| 245 | m_transformSkip[0] = charBuf; charBuf += m_numPartitions; |
| 246 | m_transformSkip[1] = charBuf; charBuf += m_numPartitions; |
| 247 | m_transformSkip[2] = charBuf; charBuf += m_numPartitions; |
| 248 | m_cbf[0] = charBuf; charBuf += m_numPartitions; |
| 249 | m_cbf[1] = charBuf; charBuf += m_numPartitions; |
| 250 | m_cbf[2] = charBuf; charBuf += m_numPartitions; |
| 251 | m_chromaIntraDir = charBuf; charBuf += m_numPartitions; |
| 252 | |
| 253 | X265_CHECK(charBuf == dataPool.charMemBlock + (m_numPartitions * BytesPerPartition) * (instance + 1), "CU data layout is broken\n"); |
| 254 | |
| 255 | m_mv[0] = dataPool.mvMemBlock + (instance * 4) * m_numPartitions; |
| 256 | m_mv[1] = m_mv[0] + m_numPartitions; |
| 257 | m_mvd[0] = m_mv[1] + m_numPartitions; |
| 258 | m_mvd[1] = m_mvd[0] + m_numPartitions; |
| 259 | |
| 260 | uint32_t cuSize = g_maxCUSize >> depth; |
| 261 | uint32_t sizeL = cuSize * cuSize; |
| 262 | uint32_t sizeC = sizeL >> (m_hChromaShift + m_vChromaShift); |
| 263 | m_trCoeff[0] = dataPool.trCoeffMemBlock + instance * (sizeL + sizeC * 2); |
| 264 | m_trCoeff[1] = m_trCoeff[0] + sizeL; |
| 265 | m_trCoeff[2] = m_trCoeff[0] + sizeL + sizeC; |
| 266 | } |
| 267 | |
| 268 | void CUData::initCTU(const Frame& frame, uint32_t cuAddr, int qp) |
| 269 | { |
| 270 | m_encData = frame.m_encData; |
| 271 | m_slice = m_encData->m_slice; |
| 272 | m_cuAddr = cuAddr; |
| 273 | m_cuPelX = (cuAddr % m_slice->m_sps->numCuInWidth) << g_maxLog2CUSize; |
| 274 | m_cuPelY = (cuAddr / m_slice->m_sps->numCuInWidth) << g_maxLog2CUSize; |
| 275 | m_absIdxInCTU = 0; |
| 276 | m_numPartitions = NUM_CU_PARTITIONS; |
| 277 | |
| 278 | /* sequential memsets */ |
| 279 | m_partSet((uint8_t*)m_qp, (uint8_t)qp); |
| 280 | m_partSet(m_log2CUSize, (uint8_t)g_maxLog2CUSize); |
| 281 | m_partSet(m_partSize, (uint8_t)SIZE_NONE); |
| 282 | m_partSet(m_predMode, (uint8_t)MODE_NONE); |
| 283 | m_partSet(m_lumaIntraDir, (uint8_t)DC_IDX); |
| 284 | m_partSet(m_tqBypass, (uint8_t)frame.m_encData->m_param->bLossless); |
| 285 | if (m_slice->m_sliceType != I_SLICE) |
| 286 | { |
| 287 | m_partSet((uint8_t*)m_refIdx[0], (uint8_t)REF_NOT_VALID); |
| 288 | m_partSet((uint8_t*)m_refIdx[1], (uint8_t)REF_NOT_VALID); |
| 289 | } |
| 290 | |
| 291 | X265_CHECK(!(frame.m_encData->m_param->bLossless && !m_slice->m_pps->bTransquantBypassEnabled), "lossless enabled without TQbypass in PPS\n"); |
| 292 | |
| 293 | /* initialize the remaining CU data in one memset */ |
| 294 | memset(m_cuDepth, 0, (BytesPerPartition - 8) * m_numPartitions); |
| 295 | |
| 296 | uint32_t widthInCU = m_slice->m_sps->numCuInWidth; |
| 297 | m_cuLeft = (m_cuAddr % widthInCU) ? m_encData->getPicCTU(m_cuAddr - 1) : NULL; |
| 298 | m_cuAbove = (m_cuAddr / widthInCU) ? m_encData->getPicCTU(m_cuAddr - widthInCU) : NULL; |
| 299 | m_cuAboveLeft = (m_cuLeft && m_cuAbove) ? m_encData->getPicCTU(m_cuAddr - widthInCU - 1) : NULL; |
| 300 | m_cuAboveRight = (m_cuAbove && ((m_cuAddr % widthInCU) < (widthInCU - 1))) ? m_encData->getPicCTU(m_cuAddr - widthInCU + 1) : NULL; |
| 301 | } |
| 302 | |
| 303 | // initialize Sub partition |
| 304 | void CUData::initSubCU(const CUData& ctu, const CUGeom& cuGeom) |
| 305 | { |
| 306 | m_absIdxInCTU = cuGeom.encodeIdx; |
| 307 | m_encData = ctu.m_encData; |
| 308 | m_slice = ctu.m_slice; |
| 309 | m_cuAddr = ctu.m_cuAddr; |
| 310 | m_cuPelX = ctu.m_cuPelX + g_zscanToPelX[cuGeom.encodeIdx]; |
| 311 | m_cuPelY = ctu.m_cuPelY + g_zscanToPelY[cuGeom.encodeIdx]; |
| 312 | m_cuLeft = ctu.m_cuLeft; |
| 313 | m_cuAbove = ctu.m_cuAbove; |
| 314 | m_cuAboveLeft = ctu.m_cuAboveLeft; |
| 315 | m_cuAboveRight = ctu.m_cuAboveRight; |
| 316 | X265_CHECK(m_numPartitions == cuGeom.numPartitions, "initSubCU() size mismatch\n"); |
| 317 | |
| 318 | /* sequential memsets */ |
| 319 | m_partSet((uint8_t*)m_qp, (uint8_t)ctu.m_qp[0]); |
| 320 | m_partSet(m_log2CUSize, (uint8_t)cuGeom.log2CUSize); |
| 321 | m_partSet(m_partSize, (uint8_t)SIZE_NONE); |
| 322 | m_partSet(m_predMode, (uint8_t)MODE_NONE); |
| 323 | m_partSet(m_lumaIntraDir, (uint8_t)DC_IDX); |
| 324 | m_partSet(m_tqBypass, (uint8_t)m_encData->m_param->bLossless); |
| 325 | m_partSet((uint8_t*)m_refIdx[0], (uint8_t)REF_NOT_VALID); |
| 326 | m_partSet((uint8_t*)m_refIdx[1], (uint8_t)REF_NOT_VALID); |
| 327 | m_partSet(m_cuDepth, (uint8_t)cuGeom.depth); |
| 328 | |
| 329 | /* initialize the remaining CU data in one memset */ |
| 330 | memset(m_skipFlag, 0, (BytesPerPartition - 9) * m_numPartitions); |
| 331 | } |
| 332 | |
| 333 | /* Copy the results of a sub-part (split) CU to the parent CU */ |
| 334 | void CUData::copyPartFrom(const CUData& subCU, const CUGeom& childGeom, uint32_t subPartIdx) |
| 335 | { |
| 336 | X265_CHECK(subPartIdx < 4, "part unit should be less than 4\n"); |
| 337 | |
| 338 | uint32_t offset = childGeom.numPartitions * subPartIdx; |
| 339 | |
| 340 | m_subPartCopy((uint8_t*)m_qp + offset, (uint8_t*)subCU.m_qp); |
| 341 | m_subPartCopy(m_log2CUSize + offset, subCU.m_log2CUSize); |
| 342 | m_subPartCopy(m_partSize + offset, subCU.m_partSize); |
| 343 | m_subPartCopy(m_predMode + offset, subCU.m_predMode); |
| 344 | m_subPartCopy(m_lumaIntraDir + offset, subCU.m_lumaIntraDir); |
| 345 | m_subPartCopy(m_tqBypass + offset, subCU.m_tqBypass); |
| 346 | m_subPartCopy((uint8_t*)m_refIdx[0] + offset, (uint8_t*)subCU.m_refIdx[0]); |
| 347 | m_subPartCopy((uint8_t*)m_refIdx[1] + offset, (uint8_t*)subCU.m_refIdx[1]); |
| 348 | m_subPartCopy(m_cuDepth + offset, subCU.m_cuDepth); |
| 349 | m_subPartCopy(m_skipFlag + offset, subCU.m_skipFlag); |
| 350 | m_subPartCopy(m_mergeFlag + offset, subCU.m_mergeFlag); |
| 351 | m_subPartCopy(m_interDir + offset, subCU.m_interDir); |
| 352 | m_subPartCopy(m_mvpIdx[0] + offset, subCU.m_mvpIdx[0]); |
| 353 | m_subPartCopy(m_mvpIdx[1] + offset, subCU.m_mvpIdx[1]); |
| 354 | m_subPartCopy(m_tuDepth + offset, subCU.m_tuDepth); |
| 355 | m_subPartCopy(m_transformSkip[0] + offset, subCU.m_transformSkip[0]); |
| 356 | m_subPartCopy(m_transformSkip[1] + offset, subCU.m_transformSkip[1]); |
| 357 | m_subPartCopy(m_transformSkip[2] + offset, subCU.m_transformSkip[2]); |
| 358 | m_subPartCopy(m_cbf[0] + offset, subCU.m_cbf[0]); |
| 359 | m_subPartCopy(m_cbf[1] + offset, subCU.m_cbf[1]); |
| 360 | m_subPartCopy(m_cbf[2] + offset, subCU.m_cbf[2]); |
| 361 | m_subPartCopy(m_chromaIntraDir + offset, subCU.m_chromaIntraDir); |
| 362 | |
| 363 | memcpy(m_mv[0] + offset, subCU.m_mv[0], childGeom.numPartitions * sizeof(MV)); |
| 364 | memcpy(m_mv[1] + offset, subCU.m_mv[1], childGeom.numPartitions * sizeof(MV)); |
| 365 | memcpy(m_mvd[0] + offset, subCU.m_mvd[0], childGeom.numPartitions * sizeof(MV)); |
| 366 | memcpy(m_mvd[1] + offset, subCU.m_mvd[1], childGeom.numPartitions * sizeof(MV)); |
| 367 | |
| 368 | uint32_t tmp = 1 << ((g_maxLog2CUSize - childGeom.depth) * 2); |
| 369 | uint32_t tmp2 = subPartIdx * tmp; |
| 370 | memcpy(m_trCoeff[0] + tmp2, subCU.m_trCoeff[0], sizeof(coeff_t) * tmp); |
| 371 | |
| 372 | uint32_t tmpC = tmp >> (m_hChromaShift + m_vChromaShift); |
| 373 | uint32_t tmpC2 = tmp2 >> (m_hChromaShift + m_vChromaShift); |
| 374 | memcpy(m_trCoeff[1] + tmpC2, subCU.m_trCoeff[1], sizeof(coeff_t) * tmpC); |
| 375 | memcpy(m_trCoeff[2] + tmpC2, subCU.m_trCoeff[2], sizeof(coeff_t) * tmpC); |
| 376 | } |
| 377 | |
| 378 | /* If a sub-CU part is not present (off the edge of the picture) its depth and |
| 379 | * log2size should still be configured */ |
| 380 | void CUData::setEmptyPart(const CUGeom& childGeom, uint32_t subPartIdx) |
| 381 | { |
| 382 | uint32_t offset = childGeom.numPartitions * subPartIdx; |
| 383 | m_subPartSet(m_cuDepth + offset, (uint8_t)childGeom.depth); |
| 384 | m_subPartSet(m_log2CUSize + offset, (uint8_t)childGeom.log2CUSize); |
| 385 | } |
| 386 | |
| 387 | /* Copy all CU data from one instance to the next, except set lossless flag |
| 388 | * This will only get used when --cu-lossless is enabled but --lossless is not. */ |
| 389 | void CUData::initLosslessCU(const CUData& cu, const CUGeom& cuGeom) |
| 390 | { |
| 391 | /* Start by making an exact copy */ |
| 392 | m_encData = cu.m_encData; |
| 393 | m_slice = cu.m_slice; |
| 394 | m_cuAddr = cu.m_cuAddr; |
| 395 | m_cuPelX = cu.m_cuPelX; |
| 396 | m_cuPelY = cu.m_cuPelY; |
| 397 | m_cuLeft = cu.m_cuLeft; |
| 398 | m_cuAbove = cu.m_cuAbove; |
| 399 | m_cuAboveLeft = cu.m_cuAboveLeft; |
| 400 | m_cuAboveRight = cu.m_cuAboveRight; |
| 401 | m_absIdxInCTU = cuGeom.encodeIdx; |
| 402 | m_numPartitions = cuGeom.numPartitions; |
| 403 | memcpy(m_qp, cu.m_qp, BytesPerPartition * m_numPartitions); |
| 404 | memcpy(m_mv[0], cu.m_mv[0], m_numPartitions * sizeof(MV)); |
| 405 | memcpy(m_mv[1], cu.m_mv[1], m_numPartitions * sizeof(MV)); |
| 406 | memcpy(m_mvd[0], cu.m_mvd[0], m_numPartitions * sizeof(MV)); |
| 407 | memcpy(m_mvd[1], cu.m_mvd[1], m_numPartitions * sizeof(MV)); |
| 408 | |
| 409 | /* force TQBypass to true */ |
| 410 | m_partSet(m_tqBypass, true); |
| 411 | |
| 412 | /* clear residual coding flags */ |
| 413 | m_partSet(m_skipFlag, 0); |
| 414 | m_partSet(m_tuDepth, 0); |
| 415 | m_partSet(m_transformSkip[0], 0); |
| 416 | m_partSet(m_transformSkip[1], 0); |
| 417 | m_partSet(m_transformSkip[2], 0); |
| 418 | m_partSet(m_cbf[0], 0); |
| 419 | m_partSet(m_cbf[1], 0); |
| 420 | m_partSet(m_cbf[2], 0); |
| 421 | } |
| 422 | |
| 423 | /* Copy completed predicted CU to CTU in picture */ |
| 424 | void CUData::copyToPic(uint32_t depth) const |
| 425 | { |
| 426 | CUData& ctu = *m_encData->getPicCTU(m_cuAddr); |
| 427 | |
| 428 | m_partCopy((uint8_t*)ctu.m_qp + m_absIdxInCTU, (uint8_t*)m_qp); |
| 429 | m_partCopy(ctu.m_log2CUSize + m_absIdxInCTU, m_log2CUSize); |
| 430 | m_partCopy(ctu.m_partSize + m_absIdxInCTU, m_partSize); |
| 431 | m_partCopy(ctu.m_predMode + m_absIdxInCTU, m_predMode); |
| 432 | m_partCopy(ctu.m_lumaIntraDir + m_absIdxInCTU, m_lumaIntraDir); |
| 433 | m_partCopy(ctu.m_tqBypass + m_absIdxInCTU, m_tqBypass); |
| 434 | m_partCopy((uint8_t*)ctu.m_refIdx[0] + m_absIdxInCTU, (uint8_t*)m_refIdx[0]); |
| 435 | m_partCopy((uint8_t*)ctu.m_refIdx[1] + m_absIdxInCTU, (uint8_t*)m_refIdx[1]); |
| 436 | m_partCopy(ctu.m_cuDepth + m_absIdxInCTU, m_cuDepth); |
| 437 | m_partCopy(ctu.m_skipFlag + m_absIdxInCTU, m_skipFlag); |
| 438 | m_partCopy(ctu.m_mergeFlag + m_absIdxInCTU, m_mergeFlag); |
| 439 | m_partCopy(ctu.m_interDir + m_absIdxInCTU, m_interDir); |
| 440 | m_partCopy(ctu.m_mvpIdx[0] + m_absIdxInCTU, m_mvpIdx[0]); |
| 441 | m_partCopy(ctu.m_mvpIdx[1] + m_absIdxInCTU, m_mvpIdx[1]); |
| 442 | m_partCopy(ctu.m_tuDepth + m_absIdxInCTU, m_tuDepth); |
| 443 | m_partCopy(ctu.m_transformSkip[0] + m_absIdxInCTU, m_transformSkip[0]); |
| 444 | m_partCopy(ctu.m_transformSkip[1] + m_absIdxInCTU, m_transformSkip[1]); |
| 445 | m_partCopy(ctu.m_transformSkip[2] + m_absIdxInCTU, m_transformSkip[2]); |
| 446 | m_partCopy(ctu.m_cbf[0] + m_absIdxInCTU, m_cbf[0]); |
| 447 | m_partCopy(ctu.m_cbf[1] + m_absIdxInCTU, m_cbf[1]); |
| 448 | m_partCopy(ctu.m_cbf[2] + m_absIdxInCTU, m_cbf[2]); |
| 449 | m_partCopy(ctu.m_chromaIntraDir + m_absIdxInCTU, m_chromaIntraDir); |
| 450 | |
| 451 | memcpy(ctu.m_mv[0] + m_absIdxInCTU, m_mv[0], m_numPartitions * sizeof(MV)); |
| 452 | memcpy(ctu.m_mv[1] + m_absIdxInCTU, m_mv[1], m_numPartitions * sizeof(MV)); |
| 453 | memcpy(ctu.m_mvd[0] + m_absIdxInCTU, m_mvd[0], m_numPartitions * sizeof(MV)); |
| 454 | memcpy(ctu.m_mvd[1] + m_absIdxInCTU, m_mvd[1], m_numPartitions * sizeof(MV)); |
| 455 | |
| 456 | uint32_t tmpY = 1 << ((g_maxLog2CUSize - depth) * 2); |
| 457 | uint32_t tmpY2 = m_absIdxInCTU << (LOG2_UNIT_SIZE * 2); |
| 458 | memcpy(ctu.m_trCoeff[0] + tmpY2, m_trCoeff[0], sizeof(coeff_t) * tmpY); |
| 459 | |
| 460 | uint32_t tmpC = tmpY >> (m_hChromaShift + m_vChromaShift); |
| 461 | uint32_t tmpC2 = tmpY2 >> (m_hChromaShift + m_vChromaShift); |
| 462 | memcpy(ctu.m_trCoeff[1] + tmpC2, m_trCoeff[1], sizeof(coeff_t) * tmpC); |
| 463 | memcpy(ctu.m_trCoeff[2] + tmpC2, m_trCoeff[2], sizeof(coeff_t) * tmpC); |
| 464 | } |
| 465 | |
| 466 | /* The reverse of copyToPic, called only by encodeResidue */ |
| 467 | void CUData::copyFromPic(const CUData& ctu, const CUGeom& cuGeom) |
| 468 | { |
| 469 | m_encData = ctu.m_encData; |
| 470 | m_slice = ctu.m_slice; |
| 471 | m_cuAddr = ctu.m_cuAddr; |
| 472 | m_cuPelX = ctu.m_cuPelX + g_zscanToPelX[cuGeom.encodeIdx]; |
| 473 | m_cuPelY = ctu.m_cuPelY + g_zscanToPelY[cuGeom.encodeIdx]; |
| 474 | m_absIdxInCTU = cuGeom.encodeIdx; |
| 475 | m_numPartitions = cuGeom.numPartitions; |
| 476 | |
| 477 | /* copy out all prediction info for this part */ |
| 478 | m_partCopy((uint8_t*)m_qp, (uint8_t*)ctu.m_qp + m_absIdxInCTU); |
| 479 | m_partCopy(m_log2CUSize, ctu.m_log2CUSize + m_absIdxInCTU); |
| 480 | m_partCopy(m_partSize, ctu.m_partSize + m_absIdxInCTU); |
| 481 | m_partCopy(m_predMode, ctu.m_predMode + m_absIdxInCTU); |
| 482 | m_partCopy(m_lumaIntraDir, ctu.m_lumaIntraDir + m_absIdxInCTU); |
| 483 | m_partCopy(m_tqBypass, ctu.m_tqBypass + m_absIdxInCTU); |
| 484 | m_partCopy((uint8_t*)m_refIdx[0], (uint8_t*)ctu.m_refIdx[0] + m_absIdxInCTU); |
| 485 | m_partCopy((uint8_t*)m_refIdx[1], (uint8_t*)ctu.m_refIdx[1] + m_absIdxInCTU); |
| 486 | m_partCopy(m_cuDepth, ctu.m_cuDepth + m_absIdxInCTU); |
| 487 | m_partCopy(m_mergeFlag, ctu.m_mergeFlag + m_absIdxInCTU); |
| 488 | m_partCopy(m_interDir, ctu.m_interDir + m_absIdxInCTU); |
| 489 | m_partCopy(m_mvpIdx[0], ctu.m_mvpIdx[0] + m_absIdxInCTU); |
| 490 | m_partCopy(m_mvpIdx[1], ctu.m_mvpIdx[1] + m_absIdxInCTU); |
| 491 | m_partCopy(m_chromaIntraDir, ctu.m_chromaIntraDir + m_absIdxInCTU); |
| 492 | |
| 493 | memcpy(m_mv[0], ctu.m_mv[0] + m_absIdxInCTU, m_numPartitions * sizeof(MV)); |
| 494 | memcpy(m_mv[1], ctu.m_mv[1] + m_absIdxInCTU, m_numPartitions * sizeof(MV)); |
| 495 | memcpy(m_mvd[0], ctu.m_mvd[0] + m_absIdxInCTU, m_numPartitions * sizeof(MV)); |
| 496 | memcpy(m_mvd[1], ctu.m_mvd[1] + m_absIdxInCTU, m_numPartitions * sizeof(MV)); |
| 497 | |
| 498 | /* clear residual coding flags */ |
| 499 | m_partSet(m_skipFlag, 0); |
| 500 | m_partSet(m_tuDepth, 0); |
| 501 | m_partSet(m_transformSkip[0], 0); |
| 502 | m_partSet(m_transformSkip[1], 0); |
| 503 | m_partSet(m_transformSkip[2], 0); |
| 504 | m_partSet(m_cbf[0], 0); |
| 505 | m_partSet(m_cbf[1], 0); |
| 506 | m_partSet(m_cbf[2], 0); |
| 507 | } |
| 508 | |
| 509 | /* Only called by encodeResidue, these fields can be modified during inter/intra coding */ |
| 510 | void CUData::updatePic(uint32_t depth) const |
| 511 | { |
| 512 | CUData& ctu = *m_encData->getPicCTU(m_cuAddr); |
| 513 | |
| 514 | m_partCopy((uint8_t*)ctu.m_qp + m_absIdxInCTU, (uint8_t*)m_qp); |
| 515 | m_partCopy(ctu.m_transformSkip[0] + m_absIdxInCTU, m_transformSkip[0]); |
| 516 | m_partCopy(ctu.m_transformSkip[1] + m_absIdxInCTU, m_transformSkip[1]); |
| 517 | m_partCopy(ctu.m_transformSkip[2] + m_absIdxInCTU, m_transformSkip[2]); |
| 518 | m_partCopy(ctu.m_skipFlag + m_absIdxInCTU, m_skipFlag); |
| 519 | m_partCopy(ctu.m_tuDepth + m_absIdxInCTU, m_tuDepth); |
| 520 | m_partCopy(ctu.m_cbf[0] + m_absIdxInCTU, m_cbf[0]); |
| 521 | m_partCopy(ctu.m_cbf[1] + m_absIdxInCTU, m_cbf[1]); |
| 522 | m_partCopy(ctu.m_cbf[2] + m_absIdxInCTU, m_cbf[2]); |
| 523 | m_partCopy(ctu.m_chromaIntraDir + m_absIdxInCTU, m_chromaIntraDir); |
| 524 | |
| 525 | uint32_t tmpY = 1 << ((g_maxLog2CUSize - depth) * 2); |
| 526 | uint32_t tmpY2 = m_absIdxInCTU << (LOG2_UNIT_SIZE * 2); |
| 527 | memcpy(ctu.m_trCoeff[0] + tmpY2, m_trCoeff[0], sizeof(coeff_t) * tmpY); |
| 528 | tmpY >>= m_hChromaShift + m_vChromaShift; |
| 529 | tmpY2 >>= m_hChromaShift + m_vChromaShift; |
| 530 | memcpy(ctu.m_trCoeff[1] + tmpY2, m_trCoeff[1], sizeof(coeff_t) * tmpY); |
| 531 | memcpy(ctu.m_trCoeff[2] + tmpY2, m_trCoeff[2], sizeof(coeff_t) * tmpY); |
| 532 | } |
| 533 | |
| 534 | const CUData* CUData::getPULeft(uint32_t& lPartUnitIdx, uint32_t curPartUnitIdx) const |
| 535 | { |
| 536 | uint32_t absPartIdx = g_zscanToRaster[curPartUnitIdx]; |
| 537 | |
| 538 | if (!isZeroCol(absPartIdx, s_numPartInCUSize)) |
| 539 | { |
| 540 | uint32_t absZorderCUIdx = g_zscanToRaster[m_absIdxInCTU]; |
| 541 | lPartUnitIdx = g_rasterToZscan[absPartIdx - 1]; |
| 542 | if (isEqualCol(absPartIdx, absZorderCUIdx, s_numPartInCUSize)) |
| 543 | return m_encData->getPicCTU(m_cuAddr); |
| 544 | else |
| 545 | { |
| 546 | lPartUnitIdx -= m_absIdxInCTU; |
| 547 | return this; |
| 548 | } |
| 549 | } |
| 550 | |
| 551 | lPartUnitIdx = g_rasterToZscan[absPartIdx + s_numPartInCUSize - 1]; |
| 552 | return m_cuLeft; |
| 553 | } |
| 554 | |
| 555 | const CUData* CUData::getPUAbove(uint32_t& aPartUnitIdx, uint32_t curPartUnitIdx, bool planarAtCTUBoundary) const |
| 556 | { |
| 557 | uint32_t absPartIdx = g_zscanToRaster[curPartUnitIdx]; |
| 558 | |
| 559 | if (!isZeroRow(absPartIdx, s_numPartInCUSize)) |
| 560 | { |
| 561 | uint32_t absZorderCUIdx = g_zscanToRaster[m_absIdxInCTU]; |
| 562 | aPartUnitIdx = g_rasterToZscan[absPartIdx - s_numPartInCUSize]; |
| 563 | if (isEqualRow(absPartIdx, absZorderCUIdx, s_numPartInCUSize)) |
| 564 | return m_encData->getPicCTU(m_cuAddr); |
| 565 | else |
| 566 | { |
| 567 | aPartUnitIdx -= m_absIdxInCTU; |
| 568 | return this; |
| 569 | } |
| 570 | } |
| 571 | |
| 572 | if (planarAtCTUBoundary) |
| 573 | return NULL; |
| 574 | |
| 575 | aPartUnitIdx = g_rasterToZscan[absPartIdx + NUM_CU_PARTITIONS - s_numPartInCUSize]; |
| 576 | return m_cuAbove; |
| 577 | } |
| 578 | |
| 579 | const CUData* CUData::getPUAboveLeft(uint32_t& alPartUnitIdx, uint32_t curPartUnitIdx) const |
| 580 | { |
| 581 | uint32_t absPartIdx = g_zscanToRaster[curPartUnitIdx]; |
| 582 | |
| 583 | if (!isZeroCol(absPartIdx, s_numPartInCUSize)) |
| 584 | { |
| 585 | if (!isZeroRow(absPartIdx, s_numPartInCUSize)) |
| 586 | { |
| 587 | uint32_t absZorderCUIdx = g_zscanToRaster[m_absIdxInCTU]; |
| 588 | alPartUnitIdx = g_rasterToZscan[absPartIdx - s_numPartInCUSize - 1]; |
| 589 | if (isEqualRowOrCol(absPartIdx, absZorderCUIdx, s_numPartInCUSize)) |
| 590 | return m_encData->getPicCTU(m_cuAddr); |
| 591 | else |
| 592 | { |
| 593 | alPartUnitIdx -= m_absIdxInCTU; |
| 594 | return this; |
| 595 | } |
| 596 | } |
| 597 | alPartUnitIdx = g_rasterToZscan[absPartIdx + NUM_CU_PARTITIONS - s_numPartInCUSize - 1]; |
| 598 | return m_cuAbove; |
| 599 | } |
| 600 | |
| 601 | if (!isZeroRow(absPartIdx, s_numPartInCUSize)) |
| 602 | { |
| 603 | alPartUnitIdx = g_rasterToZscan[absPartIdx - 1]; |
| 604 | return m_cuLeft; |
| 605 | } |
| 606 | |
| 607 | alPartUnitIdx = g_rasterToZscan[NUM_CU_PARTITIONS - 1]; |
| 608 | return m_cuAboveLeft; |
| 609 | } |
| 610 | |
| 611 | const CUData* CUData::getPUAboveRight(uint32_t& arPartUnitIdx, uint32_t curPartUnitIdx) const |
| 612 | { |
| 613 | if ((m_encData->getPicCTU(m_cuAddr)->m_cuPelX + g_zscanToPelX[curPartUnitIdx] + UNIT_SIZE) >= m_slice->m_sps->picWidthInLumaSamples) |
| 614 | return NULL; |
| 615 | |
| 616 | uint32_t absPartIdxRT = g_zscanToRaster[curPartUnitIdx]; |
| 617 | |
| 618 | if (lessThanCol(absPartIdxRT, s_numPartInCUSize - 1, s_numPartInCUSize)) |
| 619 | { |
| 620 | if (!isZeroRow(absPartIdxRT, s_numPartInCUSize)) |
| 621 | { |
| 622 | if (curPartUnitIdx > g_rasterToZscan[absPartIdxRT - s_numPartInCUSize + 1]) |
| 623 | { |
| 624 | uint32_t absZorderCUIdx = g_zscanToRaster[m_absIdxInCTU] + (1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE)) - 1; |
| 625 | arPartUnitIdx = g_rasterToZscan[absPartIdxRT - s_numPartInCUSize + 1]; |
| 626 | if (isEqualRowOrCol(absPartIdxRT, absZorderCUIdx, s_numPartInCUSize)) |
| 627 | return m_encData->getPicCTU(m_cuAddr); |
| 628 | else |
| 629 | { |
| 630 | arPartUnitIdx -= m_absIdxInCTU; |
| 631 | return this; |
| 632 | } |
| 633 | } |
| 634 | return NULL; |
| 635 | } |
| 636 | arPartUnitIdx = g_rasterToZscan[absPartIdxRT + NUM_CU_PARTITIONS - s_numPartInCUSize + 1]; |
| 637 | return m_cuAbove; |
| 638 | } |
| 639 | |
| 640 | if (!isZeroRow(absPartIdxRT, s_numPartInCUSize)) |
| 641 | return NULL; |
| 642 | |
| 643 | arPartUnitIdx = g_rasterToZscan[NUM_CU_PARTITIONS - s_numPartInCUSize]; |
| 644 | return m_cuAboveRight; |
| 645 | } |
| 646 | |
| 647 | const CUData* CUData::getPUBelowLeft(uint32_t& blPartUnitIdx, uint32_t curPartUnitIdx) const |
| 648 | { |
| 649 | if ((m_encData->getPicCTU(m_cuAddr)->m_cuPelY + g_zscanToPelY[curPartUnitIdx] + UNIT_SIZE) >= m_slice->m_sps->picHeightInLumaSamples) |
| 650 | return NULL; |
| 651 | |
| 652 | uint32_t absPartIdxLB = g_zscanToRaster[curPartUnitIdx]; |
| 653 | |
| 654 | if (lessThanRow(absPartIdxLB, s_numPartInCUSize - 1, s_numPartInCUSize)) |
| 655 | { |
| 656 | if (!isZeroCol(absPartIdxLB, s_numPartInCUSize)) |
| 657 | { |
| 658 | if (curPartUnitIdx > g_rasterToZscan[absPartIdxLB + s_numPartInCUSize - 1]) |
| 659 | { |
| 660 | uint32_t absZorderCUIdxLB = g_zscanToRaster[m_absIdxInCTU] + ((1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE)) - 1) * s_numPartInCUSize; |
| 661 | blPartUnitIdx = g_rasterToZscan[absPartIdxLB + s_numPartInCUSize - 1]; |
| 662 | if (isEqualRowOrCol(absPartIdxLB, absZorderCUIdxLB, s_numPartInCUSize)) |
| 663 | return m_encData->getPicCTU(m_cuAddr); |
| 664 | else |
| 665 | { |
| 666 | blPartUnitIdx -= m_absIdxInCTU; |
| 667 | return this; |
| 668 | } |
| 669 | } |
| 670 | return NULL; |
| 671 | } |
| 672 | blPartUnitIdx = g_rasterToZscan[absPartIdxLB + s_numPartInCUSize * 2 - 1]; |
| 673 | return m_cuLeft; |
| 674 | } |
| 675 | |
| 676 | return NULL; |
| 677 | } |
| 678 | |
| 679 | const CUData* CUData::getPUBelowLeftAdi(uint32_t& blPartUnitIdx, uint32_t curPartUnitIdx, uint32_t partUnitOffset) const |
| 680 | { |
| 681 | if ((m_encData->getPicCTU(m_cuAddr)->m_cuPelY + g_zscanToPelY[curPartUnitIdx] + (partUnitOffset << LOG2_UNIT_SIZE)) >= m_slice->m_sps->picHeightInLumaSamples) |
| 682 | return NULL; |
| 683 | |
| 684 | uint32_t absPartIdxLB = g_zscanToRaster[curPartUnitIdx]; |
| 685 | |
| 686 | if (lessThanRow(absPartIdxLB, s_numPartInCUSize - partUnitOffset, s_numPartInCUSize)) |
| 687 | { |
| 688 | if (!isZeroCol(absPartIdxLB, s_numPartInCUSize)) |
| 689 | { |
| 690 | if (curPartUnitIdx > g_rasterToZscan[absPartIdxLB + partUnitOffset * s_numPartInCUSize - 1]) |
| 691 | { |
| 692 | uint32_t absZorderCUIdxLB = g_zscanToRaster[m_absIdxInCTU] + ((1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE)) - 1) * s_numPartInCUSize; |
| 693 | blPartUnitIdx = g_rasterToZscan[absPartIdxLB + partUnitOffset * s_numPartInCUSize - 1]; |
| 694 | if (isEqualRowOrCol(absPartIdxLB, absZorderCUIdxLB, s_numPartInCUSize)) |
| 695 | return m_encData->getPicCTU(m_cuAddr); |
| 696 | else |
| 697 | { |
| 698 | blPartUnitIdx -= m_absIdxInCTU; |
| 699 | return this; |
| 700 | } |
| 701 | } |
| 702 | return NULL; |
| 703 | } |
| 704 | blPartUnitIdx = g_rasterToZscan[absPartIdxLB + (1 + partUnitOffset) * s_numPartInCUSize - 1]; |
| 705 | if (!m_cuLeft || !m_cuLeft->m_slice) |
| 706 | return NULL; |
| 707 | return m_cuLeft; |
| 708 | } |
| 709 | |
| 710 | return NULL; |
| 711 | } |
| 712 | |
| 713 | const CUData* CUData::getPUAboveRightAdi(uint32_t& arPartUnitIdx, uint32_t curPartUnitIdx, uint32_t partUnitOffset) const |
| 714 | { |
| 715 | if ((m_encData->getPicCTU(m_cuAddr)->m_cuPelX + g_zscanToPelX[curPartUnitIdx] + (partUnitOffset << LOG2_UNIT_SIZE)) >= m_slice->m_sps->picWidthInLumaSamples) |
| 716 | return NULL; |
| 717 | |
| 718 | uint32_t absPartIdxRT = g_zscanToRaster[curPartUnitIdx]; |
| 719 | |
| 720 | if (lessThanCol(absPartIdxRT, s_numPartInCUSize - partUnitOffset, s_numPartInCUSize)) |
| 721 | { |
| 722 | if (!isZeroRow(absPartIdxRT, s_numPartInCUSize)) |
| 723 | { |
| 724 | if (curPartUnitIdx > g_rasterToZscan[absPartIdxRT - s_numPartInCUSize + partUnitOffset]) |
| 725 | { |
| 726 | uint32_t absZorderCUIdx = g_zscanToRaster[m_absIdxInCTU] + (1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE)) - 1; |
| 727 | arPartUnitIdx = g_rasterToZscan[absPartIdxRT - s_numPartInCUSize + partUnitOffset]; |
| 728 | if (isEqualRowOrCol(absPartIdxRT, absZorderCUIdx, s_numPartInCUSize)) |
| 729 | return m_encData->getPicCTU(m_cuAddr); |
| 730 | else |
| 731 | { |
| 732 | arPartUnitIdx -= m_absIdxInCTU; |
| 733 | return this; |
| 734 | } |
| 735 | } |
| 736 | return NULL; |
| 737 | } |
| 738 | arPartUnitIdx = g_rasterToZscan[absPartIdxRT + NUM_CU_PARTITIONS - s_numPartInCUSize + partUnitOffset]; |
| 739 | if (!m_cuAbove || !m_cuAbove->m_slice) |
| 740 | return NULL; |
| 741 | return m_cuAbove; |
| 742 | } |
| 743 | |
| 744 | if (!isZeroRow(absPartIdxRT, s_numPartInCUSize)) |
| 745 | return NULL; |
| 746 | |
| 747 | arPartUnitIdx = g_rasterToZscan[NUM_CU_PARTITIONS - s_numPartInCUSize + partUnitOffset - 1]; |
| 748 | if ((m_cuAboveRight == NULL || m_cuAboveRight->m_slice == NULL || (m_cuAboveRight->m_cuAddr) > m_cuAddr)) |
| 749 | return NULL; |
| 750 | return m_cuAboveRight; |
| 751 | } |
| 752 | |
| 753 | /* Get left QpMinCu */ |
| 754 | const CUData* CUData::getQpMinCuLeft(uint32_t& lPartUnitIdx, uint32_t curAbsIdxInCTU) const |
| 755 | { |
| 756 | uint32_t absZorderQpMinCUIdx = curAbsIdxInCTU & (0xFF << (g_maxFullDepth - m_slice->m_pps->maxCuDQPDepth) * 2); |
| 757 | uint32_t absRorderQpMinCUIdx = g_zscanToRaster[absZorderQpMinCUIdx]; |
| 758 | |
| 759 | // check for left CTU boundary |
| 760 | if (isZeroCol(absRorderQpMinCUIdx, s_numPartInCUSize)) |
| 761 | return NULL; |
| 762 | |
| 763 | // get index of left-CU relative to top-left corner of current quantization group |
| 764 | lPartUnitIdx = g_rasterToZscan[absRorderQpMinCUIdx - 1]; |
| 765 | |
| 766 | // return pointer to current CTU |
| 767 | return m_encData->getPicCTU(m_cuAddr); |
| 768 | } |
| 769 | |
| 770 | /* Get above QpMinCu */ |
| 771 | const CUData* CUData::getQpMinCuAbove(uint32_t& aPartUnitIdx, uint32_t curAbsIdxInCTU) const |
| 772 | { |
| 773 | uint32_t absZorderQpMinCUIdx = curAbsIdxInCTU & (0xFF << (g_maxFullDepth - m_slice->m_pps->maxCuDQPDepth) * 2); |
| 774 | uint32_t absRorderQpMinCUIdx = g_zscanToRaster[absZorderQpMinCUIdx]; |
| 775 | |
| 776 | // check for top CTU boundary |
| 777 | if (isZeroRow(absRorderQpMinCUIdx, s_numPartInCUSize)) |
| 778 | return NULL; |
| 779 | |
| 780 | // get index of top-CU relative to top-left corner of current quantization group |
| 781 | aPartUnitIdx = g_rasterToZscan[absRorderQpMinCUIdx - s_numPartInCUSize]; |
| 782 | |
| 783 | // return pointer to current CTU |
| 784 | return m_encData->getPicCTU(m_cuAddr); |
| 785 | } |
| 786 | |
| 787 | /* Get reference QP from left QpMinCu or latest coded QP */ |
| 788 | char CUData::getRefQP(uint32_t curAbsIdxInCTU) const |
| 789 | { |
| 790 | uint32_t lPartIdx = 0, aPartIdx = 0; |
| 791 | const CUData* cULeft = getQpMinCuLeft(lPartIdx, m_absIdxInCTU + curAbsIdxInCTU); |
| 792 | const CUData* cUAbove = getQpMinCuAbove(aPartIdx, m_absIdxInCTU + curAbsIdxInCTU); |
| 793 | |
| 794 | return ((cULeft ? cULeft->m_qp[lPartIdx] : getLastCodedQP(curAbsIdxInCTU)) + (cUAbove ? cUAbove->m_qp[aPartIdx] : getLastCodedQP(curAbsIdxInCTU)) + 1) >> 1; |
| 795 | } |
| 796 | |
| 797 | int CUData::getLastValidPartIdx(int absPartIdx) const |
| 798 | { |
| 799 | int lastValidPartIdx = absPartIdx - 1; |
| 800 | |
| 801 | while (lastValidPartIdx >= 0 && m_predMode[lastValidPartIdx] == MODE_NONE) |
| 802 | { |
| 803 | uint32_t depth = m_cuDepth[lastValidPartIdx]; |
| 804 | lastValidPartIdx -= m_numPartitions >> (depth << 1); |
| 805 | } |
| 806 | |
| 807 | return lastValidPartIdx; |
| 808 | } |
| 809 | |
| 810 | char CUData::getLastCodedQP(uint32_t absPartIdx) const |
| 811 | { |
| 812 | uint32_t quPartIdxMask = 0xFF << (g_maxFullDepth - m_slice->m_pps->maxCuDQPDepth) * 2; |
| 813 | int lastValidPartIdx = getLastValidPartIdx(absPartIdx & quPartIdxMask); |
| 814 | |
| 815 | if (lastValidPartIdx >= 0) |
| 816 | return m_qp[lastValidPartIdx]; |
| 817 | else |
| 818 | { |
| 819 | if (m_absIdxInCTU) |
| 820 | return m_encData->getPicCTU(m_cuAddr)->getLastCodedQP(m_absIdxInCTU); |
| 821 | else if (m_cuAddr > 0 && !(m_slice->m_pps->bEntropyCodingSyncEnabled && !(m_cuAddr % m_slice->m_sps->numCuInWidth))) |
| 822 | return m_encData->getPicCTU(m_cuAddr - 1)->getLastCodedQP(NUM_CU_PARTITIONS); |
| 823 | else |
| 824 | return (char)m_slice->m_sliceQp; |
| 825 | } |
| 826 | } |
| 827 | |
| 828 | /* Get allowed chroma intra modes */ |
| 829 | void CUData::getAllowedChromaDir(uint32_t absPartIdx, uint32_t* modeList) const |
| 830 | { |
| 831 | modeList[0] = PLANAR_IDX; |
| 832 | modeList[1] = VER_IDX; |
| 833 | modeList[2] = HOR_IDX; |
| 834 | modeList[3] = DC_IDX; |
| 835 | modeList[4] = DM_CHROMA_IDX; |
| 836 | |
| 837 | uint32_t lumaMode = m_lumaIntraDir[absPartIdx]; |
| 838 | |
| 839 | for (int i = 0; i < NUM_CHROMA_MODE - 1; i++) |
| 840 | { |
| 841 | if (lumaMode == modeList[i]) |
| 842 | { |
| 843 | modeList[i] = 34; // VER+8 mode |
| 844 | break; |
| 845 | } |
| 846 | } |
| 847 | } |
| 848 | |
| 849 | /* Get most probable intra modes */ |
| 850 | int CUData::getIntraDirLumaPredictor(uint32_t absPartIdx, uint32_t* intraDirPred) const |
| 851 | { |
| 852 | const CUData* tempCU; |
| 853 | uint32_t tempPartIdx; |
| 854 | uint32_t leftIntraDir, aboveIntraDir; |
| 855 | |
| 856 | // Get intra direction of left PU |
| 857 | tempCU = getPULeft(tempPartIdx, m_absIdxInCTU + absPartIdx); |
| 858 | |
| 859 | leftIntraDir = (tempCU && tempCU->isIntra(tempPartIdx)) ? tempCU->m_lumaIntraDir[tempPartIdx] : DC_IDX; |
| 860 | |
| 861 | // Get intra direction of above PU |
| 862 | tempCU = getPUAbove(tempPartIdx, m_absIdxInCTU + absPartIdx, true); |
| 863 | |
| 864 | aboveIntraDir = (tempCU && tempCU->isIntra(tempPartIdx)) ? tempCU->m_lumaIntraDir[tempPartIdx] : DC_IDX; |
| 865 | |
| 866 | if (leftIntraDir == aboveIntraDir) |
| 867 | { |
| 868 | if (leftIntraDir >= 2) // angular modes |
| 869 | { |
| 870 | intraDirPred[0] = leftIntraDir; |
| 871 | intraDirPred[1] = ((leftIntraDir - 2 + 31) & 31) + 2; |
| 872 | intraDirPred[2] = ((leftIntraDir - 2 + 1) & 31) + 2; |
| 873 | } |
| 874 | else //non-angular |
| 875 | { |
| 876 | intraDirPred[0] = PLANAR_IDX; |
| 877 | intraDirPred[1] = DC_IDX; |
| 878 | intraDirPred[2] = VER_IDX; |
| 879 | } |
| 880 | return 1; |
| 881 | } |
| 882 | else |
| 883 | { |
| 884 | intraDirPred[0] = leftIntraDir; |
| 885 | intraDirPred[1] = aboveIntraDir; |
| 886 | |
| 887 | if (leftIntraDir && aboveIntraDir) //both modes are non-planar |
| 888 | intraDirPred[2] = PLANAR_IDX; |
| 889 | else |
| 890 | intraDirPred[2] = (leftIntraDir + aboveIntraDir) < 2 ? VER_IDX : DC_IDX; |
| 891 | return 2; |
| 892 | } |
| 893 | } |
| 894 | |
| 895 | uint32_t CUData::getCtxSplitFlag(uint32_t absPartIdx, uint32_t depth) const |
| 896 | { |
| 897 | const CUData* tempCU; |
| 898 | uint32_t tempPartIdx; |
| 899 | uint32_t ctx; |
| 900 | |
| 901 | // Get left split flag |
| 902 | tempCU = getPULeft(tempPartIdx, m_absIdxInCTU + absPartIdx); |
| 903 | ctx = (tempCU) ? ((tempCU->m_cuDepth[tempPartIdx] > depth) ? 1 : 0) : 0; |
| 904 | |
| 905 | // Get above split flag |
| 906 | tempCU = getPUAbove(tempPartIdx, m_absIdxInCTU + absPartIdx); |
| 907 | ctx += (tempCU) ? ((tempCU->m_cuDepth[tempPartIdx] > depth) ? 1 : 0) : 0; |
| 908 | |
| 909 | return ctx; |
| 910 | } |
| 911 | |
| 912 | void CUData::getIntraTUQtDepthRange(uint32_t tuDepthRange[2], uint32_t absPartIdx) const |
| 913 | { |
| 914 | uint32_t log2CUSize = m_log2CUSize[absPartIdx]; |
| 915 | uint32_t splitFlag = m_partSize[absPartIdx] == SIZE_NxN; |
| 916 | |
| 917 | tuDepthRange[0] = m_slice->m_sps->quadtreeTULog2MinSize; |
| 918 | tuDepthRange[1] = m_slice->m_sps->quadtreeTULog2MaxSize; |
| 919 | |
| 920 | tuDepthRange[0] = X265_MAX(tuDepthRange[0], X265_MIN(log2CUSize - (m_slice->m_sps->quadtreeTUMaxDepthIntra - 1 + splitFlag), tuDepthRange[1])); |
| 921 | } |
| 922 | |
| 923 | void CUData::getInterTUQtDepthRange(uint32_t tuDepthRange[2], uint32_t absPartIdx) const |
| 924 | { |
| 925 | uint32_t log2CUSize = m_log2CUSize[absPartIdx]; |
| 926 | uint32_t quadtreeTUMaxDepth = m_slice->m_sps->quadtreeTUMaxDepthInter; |
| 927 | uint32_t splitFlag = quadtreeTUMaxDepth == 1 && m_partSize[absPartIdx] != SIZE_2Nx2N; |
| 928 | |
| 929 | tuDepthRange[0] = m_slice->m_sps->quadtreeTULog2MinSize; |
| 930 | tuDepthRange[1] = m_slice->m_sps->quadtreeTULog2MaxSize; |
| 931 | |
| 932 | tuDepthRange[0] = X265_MAX(tuDepthRange[0], X265_MIN(log2CUSize - (quadtreeTUMaxDepth - 1 + splitFlag), tuDepthRange[1])); |
| 933 | } |
| 934 | |
| 935 | uint32_t CUData::getCtxSkipFlag(uint32_t absPartIdx) const |
| 936 | { |
| 937 | const CUData* tempCU; |
| 938 | uint32_t tempPartIdx; |
| 939 | uint32_t ctx; |
| 940 | |
| 941 | // Get BCBP of left PU |
| 942 | tempCU = getPULeft(tempPartIdx, m_absIdxInCTU + absPartIdx); |
| 943 | ctx = tempCU ? tempCU->isSkipped(tempPartIdx) : 0; |
| 944 | |
| 945 | // Get BCBP of above PU |
| 946 | tempCU = getPUAbove(tempPartIdx, m_absIdxInCTU + absPartIdx); |
| 947 | ctx += tempCU ? tempCU->isSkipped(tempPartIdx) : 0; |
| 948 | |
| 949 | return ctx; |
| 950 | } |
| 951 | |
| 952 | bool CUData::setQPSubCUs(char qp, uint32_t absPartIdx, uint32_t depth) |
| 953 | { |
| 954 | uint32_t curPartNumb = NUM_CU_PARTITIONS >> (depth << 1); |
| 955 | uint32_t curPartNumQ = curPartNumb >> 2; |
| 956 | |
| 957 | if (m_cuDepth[absPartIdx] > depth) |
| 958 | { |
| 959 | for (uint32_t subPartIdx = 0; subPartIdx < 4; subPartIdx++) |
| 960 | if (setQPSubCUs(qp, absPartIdx + subPartIdx * curPartNumQ, depth + 1)) |
| 961 | return true; |
| 962 | } |
| 963 | else |
| 964 | { |
| 965 | if (getQtRootCbf(absPartIdx)) |
| 966 | return true; |
| 967 | else |
| 968 | setQPSubParts(qp, absPartIdx, depth); |
| 969 | } |
| 970 | |
| 971 | return false; |
| 972 | } |
| 973 | |
| 974 | void CUData::setPUInterDir(uint8_t dir, uint32_t absPartIdx, uint32_t puIdx) |
| 975 | { |
| 976 | uint32_t curPartNumQ = m_numPartitions >> 2; |
| 977 | X265_CHECK(puIdx < 2, "unexpected part unit index\n"); |
| 978 | |
| 979 | switch (m_partSize[absPartIdx]) |
| 980 | { |
| 981 | case SIZE_2Nx2N: |
| 982 | memset(m_interDir + absPartIdx, dir, 4 * curPartNumQ); |
| 983 | break; |
| 984 | case SIZE_2NxN: |
| 985 | memset(m_interDir + absPartIdx, dir, 2 * curPartNumQ); |
| 986 | break; |
| 987 | case SIZE_Nx2N: |
| 988 | memset(m_interDir + absPartIdx, dir, curPartNumQ); |
| 989 | memset(m_interDir + absPartIdx + 2 * curPartNumQ, dir, curPartNumQ); |
| 990 | break; |
| 991 | case SIZE_NxN: |
| 992 | memset(m_interDir + absPartIdx, dir, curPartNumQ); |
| 993 | break; |
| 994 | case SIZE_2NxnU: |
| 995 | if (!puIdx) |
| 996 | { |
| 997 | memset(m_interDir + absPartIdx, dir, (curPartNumQ >> 1)); |
| 998 | memset(m_interDir + absPartIdx + curPartNumQ, dir, (curPartNumQ >> 1)); |
| 999 | } |
| 1000 | else |
| 1001 | { |
| 1002 | memset(m_interDir + absPartIdx, dir, (curPartNumQ >> 1)); |
| 1003 | memset(m_interDir + absPartIdx + curPartNumQ, dir, ((curPartNumQ >> 1) + (curPartNumQ << 1))); |
| 1004 | } |
| 1005 | break; |
| 1006 | case SIZE_2NxnD: |
| 1007 | if (!puIdx) |
| 1008 | { |
| 1009 | memset(m_interDir + absPartIdx, dir, ((curPartNumQ << 1) + (curPartNumQ >> 1))); |
| 1010 | memset(m_interDir + absPartIdx + (curPartNumQ << 1) + curPartNumQ, dir, (curPartNumQ >> 1)); |
| 1011 | } |
| 1012 | else |
| 1013 | { |
| 1014 | memset(m_interDir + absPartIdx, dir, (curPartNumQ >> 1)); |
| 1015 | memset(m_interDir + absPartIdx + curPartNumQ, dir, (curPartNumQ >> 1)); |
| 1016 | } |
| 1017 | break; |
| 1018 | case SIZE_nLx2N: |
| 1019 | if (!puIdx) |
| 1020 | { |
| 1021 | memset(m_interDir + absPartIdx, dir, (curPartNumQ >> 2)); |
| 1022 | memset(m_interDir + absPartIdx + (curPartNumQ >> 1), dir, (curPartNumQ >> 2)); |
| 1023 | memset(m_interDir + absPartIdx + (curPartNumQ << 1), dir, (curPartNumQ >> 2)); |
| 1024 | memset(m_interDir + absPartIdx + (curPartNumQ << 1) + (curPartNumQ >> 1), dir, (curPartNumQ >> 2)); |
| 1025 | } |
| 1026 | else |
| 1027 | { |
| 1028 | memset(m_interDir + absPartIdx, dir, (curPartNumQ >> 2)); |
| 1029 | memset(m_interDir + absPartIdx + (curPartNumQ >> 1), dir, (curPartNumQ + (curPartNumQ >> 2))); |
| 1030 | memset(m_interDir + absPartIdx + (curPartNumQ << 1), dir, (curPartNumQ >> 2)); |
| 1031 | memset(m_interDir + absPartIdx + (curPartNumQ << 1) + (curPartNumQ >> 1), dir, (curPartNumQ + (curPartNumQ >> 2))); |
| 1032 | } |
| 1033 | break; |
| 1034 | case SIZE_nRx2N: |
| 1035 | if (!puIdx) |
| 1036 | { |
| 1037 | memset(m_interDir + absPartIdx, dir, (curPartNumQ + (curPartNumQ >> 2))); |
| 1038 | memset(m_interDir + absPartIdx + curPartNumQ + (curPartNumQ >> 1), dir, (curPartNumQ >> 2)); |
| 1039 | memset(m_interDir + absPartIdx + (curPartNumQ << 1), dir, (curPartNumQ + (curPartNumQ >> 2))); |
| 1040 | memset(m_interDir + absPartIdx + (curPartNumQ << 1) + curPartNumQ + (curPartNumQ >> 1), dir, (curPartNumQ >> 2)); |
| 1041 | } |
| 1042 | else |
| 1043 | { |
| 1044 | memset(m_interDir + absPartIdx, dir, (curPartNumQ >> 2)); |
| 1045 | memset(m_interDir + absPartIdx + (curPartNumQ >> 1), dir, (curPartNumQ >> 2)); |
| 1046 | memset(m_interDir + absPartIdx + (curPartNumQ << 1), dir, (curPartNumQ >> 2)); |
| 1047 | memset(m_interDir + absPartIdx + (curPartNumQ << 1) + (curPartNumQ >> 1), dir, (curPartNumQ >> 2)); |
| 1048 | } |
| 1049 | break; |
| 1050 | default: |
| 1051 | X265_CHECK(0, "unexpected part type\n"); |
| 1052 | break; |
| 1053 | } |
| 1054 | } |
| 1055 | |
| 1056 | template<typename T> |
| 1057 | void CUData::setAllPU(T* p, const T& val, int absPartIdx, int puIdx) |
| 1058 | { |
| 1059 | int i; |
| 1060 | |
| 1061 | p += absPartIdx; |
| 1062 | int numElements = m_numPartitions; |
| 1063 | |
| 1064 | switch (m_partSize[absPartIdx]) |
| 1065 | { |
| 1066 | case SIZE_2Nx2N: |
| 1067 | for (i = 0; i < numElements; i++) |
| 1068 | p[i] = val; |
| 1069 | break; |
| 1070 | |
| 1071 | case SIZE_2NxN: |
| 1072 | numElements >>= 1; |
| 1073 | for (i = 0; i < numElements; i++) |
| 1074 | p[i] = val; |
| 1075 | break; |
| 1076 | |
| 1077 | case SIZE_Nx2N: |
| 1078 | numElements >>= 2; |
| 1079 | for (i = 0; i < numElements; i++) |
| 1080 | { |
| 1081 | p[i] = val; |
| 1082 | p[i + 2 * numElements] = val; |
| 1083 | } |
| 1084 | break; |
| 1085 | |
| 1086 | case SIZE_2NxnU: |
| 1087 | { |
| 1088 | int curPartNumQ = numElements >> 2; |
| 1089 | if (!puIdx) |
| 1090 | { |
| 1091 | T *pT = p; |
| 1092 | T *pT2 = p + curPartNumQ; |
| 1093 | for (i = 0; i < (curPartNumQ >> 1); i++) |
| 1094 | { |
| 1095 | pT[i] = val; |
| 1096 | pT2[i] = val; |
| 1097 | } |
| 1098 | } |
| 1099 | else |
| 1100 | { |
| 1101 | T *pT = p; |
| 1102 | for (i = 0; i < (curPartNumQ >> 1); i++) |
| 1103 | pT[i] = val; |
| 1104 | |
| 1105 | pT = p + curPartNumQ; |
| 1106 | for (i = 0; i < ((curPartNumQ >> 1) + (curPartNumQ << 1)); i++) |
| 1107 | pT[i] = val; |
| 1108 | } |
| 1109 | break; |
| 1110 | } |
| 1111 | |
| 1112 | case SIZE_2NxnD: |
| 1113 | { |
| 1114 | int curPartNumQ = numElements >> 2; |
| 1115 | if (!puIdx) |
| 1116 | { |
| 1117 | T *pT = p; |
| 1118 | for (i = 0; i < ((curPartNumQ >> 1) + (curPartNumQ << 1)); i++) |
| 1119 | pT[i] = val; |
| 1120 | |
| 1121 | pT = p + (numElements - curPartNumQ); |
| 1122 | for (i = 0; i < (curPartNumQ >> 1); i++) |
| 1123 | pT[i] = val; |
| 1124 | } |
| 1125 | else |
| 1126 | { |
| 1127 | T *pT = p; |
| 1128 | T *pT2 = p + curPartNumQ; |
| 1129 | for (i = 0; i < (curPartNumQ >> 1); i++) |
| 1130 | { |
| 1131 | pT[i] = val; |
| 1132 | pT2[i] = val; |
| 1133 | } |
| 1134 | } |
| 1135 | break; |
| 1136 | } |
| 1137 | |
| 1138 | case SIZE_nLx2N: |
| 1139 | { |
| 1140 | int curPartNumQ = numElements >> 2; |
| 1141 | if (!puIdx) |
| 1142 | { |
| 1143 | T *pT = p; |
| 1144 | T *pT2 = p + (curPartNumQ << 1); |
| 1145 | T *pT3 = p + (curPartNumQ >> 1); |
| 1146 | T *pT4 = p + (curPartNumQ << 1) + (curPartNumQ >> 1); |
| 1147 | |
| 1148 | for (i = 0; i < (curPartNumQ >> 2); i++) |
| 1149 | { |
| 1150 | pT[i] = val; |
| 1151 | pT2[i] = val; |
| 1152 | pT3[i] = val; |
| 1153 | pT4[i] = val; |
| 1154 | } |
| 1155 | } |
| 1156 | else |
| 1157 | { |
| 1158 | T *pT = p; |
| 1159 | T *pT2 = p + (curPartNumQ << 1); |
| 1160 | for (i = 0; i < (curPartNumQ >> 2); i++) |
| 1161 | { |
| 1162 | pT[i] = val; |
| 1163 | pT2[i] = val; |
| 1164 | } |
| 1165 | |
| 1166 | pT = p + (curPartNumQ >> 1); |
| 1167 | pT2 = p + (curPartNumQ << 1) + (curPartNumQ >> 1); |
| 1168 | for (i = 0; i < ((curPartNumQ >> 2) + curPartNumQ); i++) |
| 1169 | { |
| 1170 | pT[i] = val; |
| 1171 | pT2[i] = val; |
| 1172 | } |
| 1173 | } |
| 1174 | break; |
| 1175 | } |
| 1176 | |
| 1177 | case SIZE_nRx2N: |
| 1178 | { |
| 1179 | int curPartNumQ = numElements >> 2; |
| 1180 | if (!puIdx) |
| 1181 | { |
| 1182 | T *pT = p; |
| 1183 | T *pT2 = p + (curPartNumQ << 1); |
| 1184 | for (i = 0; i < ((curPartNumQ >> 2) + curPartNumQ); i++) |
| 1185 | { |
| 1186 | pT[i] = val; |
| 1187 | pT2[i] = val; |
| 1188 | } |
| 1189 | |
| 1190 | pT = p + curPartNumQ + (curPartNumQ >> 1); |
| 1191 | pT2 = p + numElements - curPartNumQ + (curPartNumQ >> 1); |
| 1192 | for (i = 0; i < (curPartNumQ >> 2); i++) |
| 1193 | { |
| 1194 | pT[i] = val; |
| 1195 | pT2[i] = val; |
| 1196 | } |
| 1197 | } |
| 1198 | else |
| 1199 | { |
| 1200 | T *pT = p; |
| 1201 | T *pT2 = p + (curPartNumQ >> 1); |
| 1202 | T *pT3 = p + (curPartNumQ << 1); |
| 1203 | T *pT4 = p + (curPartNumQ << 1) + (curPartNumQ >> 1); |
| 1204 | for (i = 0; i < (curPartNumQ >> 2); i++) |
| 1205 | { |
| 1206 | pT[i] = val; |
| 1207 | pT2[i] = val; |
| 1208 | pT3[i] = val; |
| 1209 | pT4[i] = val; |
| 1210 | } |
| 1211 | } |
| 1212 | break; |
| 1213 | } |
| 1214 | |
| 1215 | case SIZE_NxN: |
| 1216 | default: |
| 1217 | X265_CHECK(0, "unknown partition type\n"); |
| 1218 | break; |
| 1219 | } |
| 1220 | } |
| 1221 | |
| 1222 | void CUData::setPUMv(int list, const MV& mv, int absPartIdx, int puIdx) |
| 1223 | { |
| 1224 | setAllPU(m_mv[list], mv, absPartIdx, puIdx); |
| 1225 | } |
| 1226 | |
| 1227 | void CUData::setPURefIdx(int list, char refIdx, int absPartIdx, int puIdx) |
| 1228 | { |
| 1229 | setAllPU(m_refIdx[list], refIdx, absPartIdx, puIdx); |
| 1230 | } |
| 1231 | |
| 1232 | void CUData::getPartIndexAndSize(uint32_t partIdx, uint32_t& outPartAddr, int& outWidth, int& outHeight) const |
| 1233 | { |
| 1234 | int cuSize = 1 << m_log2CUSize[0]; |
| 1235 | int partType = m_partSize[0]; |
| 1236 | |
| 1237 | int tmp = partTable[partType][partIdx][0]; |
| 1238 | outWidth = ((tmp >> 4) * cuSize) >> 2; |
| 1239 | outHeight = ((tmp & 0xF) * cuSize) >> 2; |
| 1240 | outPartAddr = (partAddrTable[partType][partIdx] * m_numPartitions) >> 4; |
| 1241 | } |
| 1242 | |
| 1243 | void CUData::getMvField(const CUData* cu, uint32_t absPartIdx, int picList, MVField& outMvField) const |
| 1244 | { |
| 1245 | if (cu) |
| 1246 | { |
| 1247 | outMvField.mv = cu->m_mv[picList][absPartIdx]; |
| 1248 | outMvField.refIdx = cu->m_refIdx[picList][absPartIdx]; |
| 1249 | } |
| 1250 | else |
| 1251 | { |
| 1252 | // OUT OF BOUNDARY |
| 1253 | outMvField.mv.word = 0; |
| 1254 | outMvField.refIdx = REF_NOT_VALID; |
| 1255 | } |
| 1256 | } |
| 1257 | |
| 1258 | void CUData::deriveLeftRightTopIdx(uint32_t partIdx, uint32_t& partIdxLT, uint32_t& partIdxRT) const |
| 1259 | { |
| 1260 | partIdxLT = m_absIdxInCTU; |
| 1261 | partIdxRT = g_rasterToZscan[g_zscanToRaster[partIdxLT] + (1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE)) - 1]; |
| 1262 | |
| 1263 | switch (m_partSize[0]) |
| 1264 | { |
| 1265 | case SIZE_2Nx2N: break; |
| 1266 | case SIZE_2NxN: |
| 1267 | partIdxLT += (partIdx == 0) ? 0 : m_numPartitions >> 1; |
| 1268 | partIdxRT += (partIdx == 0) ? 0 : m_numPartitions >> 1; |
| 1269 | break; |
| 1270 | case SIZE_Nx2N: |
| 1271 | partIdxLT += (partIdx == 0) ? 0 : m_numPartitions >> 2; |
| 1272 | partIdxRT -= (partIdx == 1) ? 0 : m_numPartitions >> 2; |
| 1273 | break; |
| 1274 | case SIZE_NxN: |
| 1275 | partIdxLT += (m_numPartitions >> 2) * partIdx; |
| 1276 | partIdxRT += (m_numPartitions >> 2) * (partIdx - 1); |
| 1277 | break; |
| 1278 | case SIZE_2NxnU: |
| 1279 | partIdxLT += (partIdx == 0) ? 0 : m_numPartitions >> 3; |
| 1280 | partIdxRT += (partIdx == 0) ? 0 : m_numPartitions >> 3; |
| 1281 | break; |
| 1282 | case SIZE_2NxnD: |
| 1283 | partIdxLT += (partIdx == 0) ? 0 : (m_numPartitions >> 1) + (m_numPartitions >> 3); |
| 1284 | partIdxRT += (partIdx == 0) ? 0 : (m_numPartitions >> 1) + (m_numPartitions >> 3); |
| 1285 | break; |
| 1286 | case SIZE_nLx2N: |
| 1287 | partIdxLT += (partIdx == 0) ? 0 : m_numPartitions >> 4; |
| 1288 | partIdxRT -= (partIdx == 1) ? 0 : (m_numPartitions >> 2) + (m_numPartitions >> 4); |
| 1289 | break; |
| 1290 | case SIZE_nRx2N: |
| 1291 | partIdxLT += (partIdx == 0) ? 0 : (m_numPartitions >> 2) + (m_numPartitions >> 4); |
| 1292 | partIdxRT -= (partIdx == 1) ? 0 : m_numPartitions >> 4; |
| 1293 | break; |
| 1294 | default: |
| 1295 | X265_CHECK(0, "unexpected part index\n"); |
| 1296 | break; |
| 1297 | } |
| 1298 | } |
| 1299 | |
| 1300 | uint32_t CUData::deriveLeftBottomIdx(uint32_t puIdx) const |
| 1301 | { |
| 1302 | uint32_t outPartIdxLB; |
| 1303 | outPartIdxLB = g_rasterToZscan[g_zscanToRaster[m_absIdxInCTU] + ((1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE - 1)) - 1) * s_numPartInCUSize]; |
| 1304 | |
| 1305 | switch (m_partSize[0]) |
| 1306 | { |
| 1307 | case SIZE_2Nx2N: |
| 1308 | outPartIdxLB += m_numPartitions >> 1; |
| 1309 | break; |
| 1310 | case SIZE_2NxN: |
| 1311 | outPartIdxLB += puIdx ? m_numPartitions >> 1 : 0; |
| 1312 | break; |
| 1313 | case SIZE_Nx2N: |
| 1314 | outPartIdxLB += puIdx ? (m_numPartitions >> 2) * 3 : m_numPartitions >> 1; |
| 1315 | break; |
| 1316 | case SIZE_NxN: |
| 1317 | outPartIdxLB += (m_numPartitions >> 2) * puIdx; |
| 1318 | break; |
| 1319 | case SIZE_2NxnU: |
| 1320 | outPartIdxLB += puIdx ? m_numPartitions >> 1 : -((int)m_numPartitions >> 3); |
| 1321 | break; |
| 1322 | case SIZE_2NxnD: |
| 1323 | outPartIdxLB += puIdx ? m_numPartitions >> 1 : (m_numPartitions >> 2) + (m_numPartitions >> 3); |
| 1324 | break; |
| 1325 | case SIZE_nLx2N: |
| 1326 | outPartIdxLB += puIdx ? (m_numPartitions >> 1) + (m_numPartitions >> 4) : m_numPartitions >> 1; |
| 1327 | break; |
| 1328 | case SIZE_nRx2N: |
| 1329 | outPartIdxLB += puIdx ? (m_numPartitions >> 1) + (m_numPartitions >> 2) + (m_numPartitions >> 4) : m_numPartitions >> 1; |
| 1330 | break; |
| 1331 | default: |
| 1332 | X265_CHECK(0, "unexpected part index\n"); |
| 1333 | break; |
| 1334 | } |
| 1335 | return outPartIdxLB; |
| 1336 | } |
| 1337 | |
| 1338 | /* Derives the partition index of neighboring bottom right block */ |
| 1339 | uint32_t CUData::deriveRightBottomIdx(uint32_t puIdx) const |
| 1340 | { |
| 1341 | uint32_t outPartIdxRB; |
| 1342 | outPartIdxRB = g_rasterToZscan[g_zscanToRaster[m_absIdxInCTU] + |
| 1343 | ((1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE - 1)) - 1) * s_numPartInCUSize + |
| 1344 | (1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE)) - 1]; |
| 1345 | |
| 1346 | switch (m_partSize[0]) |
| 1347 | { |
| 1348 | case SIZE_2Nx2N: |
| 1349 | outPartIdxRB += m_numPartitions >> 1; |
| 1350 | break; |
| 1351 | case SIZE_2NxN: |
| 1352 | outPartIdxRB += puIdx ? m_numPartitions >> 1 : 0; |
| 1353 | break; |
| 1354 | case SIZE_Nx2N: |
| 1355 | outPartIdxRB += puIdx ? m_numPartitions >> 1 : m_numPartitions >> 2; |
| 1356 | break; |
| 1357 | case SIZE_NxN: |
| 1358 | outPartIdxRB += (m_numPartitions >> 2) * (puIdx - 1); |
| 1359 | break; |
| 1360 | case SIZE_2NxnU: |
| 1361 | outPartIdxRB += puIdx ? m_numPartitions >> 1 : -((int)m_numPartitions >> 3); |
| 1362 | break; |
| 1363 | case SIZE_2NxnD: |
| 1364 | outPartIdxRB += puIdx ? m_numPartitions >> 1 : (m_numPartitions >> 2) + (m_numPartitions >> 3); |
| 1365 | break; |
| 1366 | case SIZE_nLx2N: |
| 1367 | outPartIdxRB += puIdx ? m_numPartitions >> 1 : (m_numPartitions >> 3) + (m_numPartitions >> 4); |
| 1368 | break; |
| 1369 | case SIZE_nRx2N: |
| 1370 | outPartIdxRB += puIdx ? m_numPartitions >> 1 : (m_numPartitions >> 2) + (m_numPartitions >> 3) + (m_numPartitions >> 4); |
| 1371 | break; |
| 1372 | default: |
| 1373 | X265_CHECK(0, "unexpected part index\n"); |
| 1374 | break; |
| 1375 | } |
| 1376 | return outPartIdxRB; |
| 1377 | } |
| 1378 | |
| 1379 | void CUData::deriveLeftRightTopIdxAdi(uint32_t& outPartIdxLT, uint32_t& outPartIdxRT, uint32_t partOffset, uint32_t partDepth) const |
| 1380 | { |
| 1381 | uint32_t numPartInWidth = 1 << (m_log2CUSize[0] - LOG2_UNIT_SIZE - partDepth); |
| 1382 | |
| 1383 | outPartIdxLT = m_absIdxInCTU + partOffset; |
| 1384 | outPartIdxRT = g_rasterToZscan[g_zscanToRaster[outPartIdxLT] + numPartInWidth - 1]; |
| 1385 | } |
| 1386 | |
| 1387 | bool CUData::hasEqualMotion(uint32_t absPartIdx, const CUData& candCU, uint32_t candAbsPartIdx) const |
| 1388 | { |
| 1389 | if (m_interDir[absPartIdx] != candCU.m_interDir[candAbsPartIdx]) |
| 1390 | return false; |
| 1391 | |
| 1392 | for (uint32_t refListIdx = 0; refListIdx < 2; refListIdx++) |
| 1393 | { |
| 1394 | if (m_interDir[absPartIdx] & (1 << refListIdx)) |
| 1395 | { |
| 1396 | if (m_mv[refListIdx][absPartIdx] != candCU.m_mv[refListIdx][candAbsPartIdx] || |
| 1397 | m_refIdx[refListIdx][absPartIdx] != candCU.m_refIdx[refListIdx][candAbsPartIdx]) |
| 1398 | return false; |
| 1399 | } |
| 1400 | } |
| 1401 | |
| 1402 | return true; |
| 1403 | } |
| 1404 | |
| 1405 | /* Construct list of merging candidates */ |
| 1406 | uint32_t CUData::getInterMergeCandidates(uint32_t absPartIdx, uint32_t puIdx, MVField(*mvFieldNeighbours)[2], uint8_t* interDirNeighbours) const |
| 1407 | { |
| 1408 | uint32_t absPartAddr = m_absIdxInCTU + absPartIdx; |
| 1409 | const bool isInterB = m_slice->isInterB(); |
| 1410 | |
| 1411 | const uint32_t maxNumMergeCand = m_slice->m_maxNumMergeCand; |
| 1412 | |
| 1413 | for (uint32_t i = 0; i < maxNumMergeCand; ++i) |
| 1414 | { |
| 1415 | mvFieldNeighbours[i][0].refIdx = REF_NOT_VALID; |
| 1416 | mvFieldNeighbours[i][1].refIdx = REF_NOT_VALID; |
| 1417 | } |
| 1418 | |
| 1419 | /* calculate the location of upper-left corner pixel and size of the current PU */ |
| 1420 | int xP, yP, nPSW, nPSH; |
| 1421 | |
| 1422 | int cuSize = 1 << m_log2CUSize[0]; |
| 1423 | int partMode = m_partSize[0]; |
| 1424 | |
| 1425 | int tmp = partTable[partMode][puIdx][0]; |
| 1426 | nPSW = ((tmp >> 4) * cuSize) >> 2; |
| 1427 | nPSH = ((tmp & 0xF) * cuSize) >> 2; |
| 1428 | |
| 1429 | tmp = partTable[partMode][puIdx][1]; |
| 1430 | xP = ((tmp >> 4) * cuSize) >> 2; |
| 1431 | yP = ((tmp & 0xF) * cuSize) >> 2; |
| 1432 | |
| 1433 | uint32_t count = 0; |
| 1434 | |
| 1435 | uint32_t partIdxLT, partIdxRT, partIdxLB = deriveLeftBottomIdx(puIdx); |
| 1436 | PartSize curPS = (PartSize)m_partSize[absPartIdx]; |
| 1437 | |
| 1438 | // left |
| 1439 | uint32_t leftPartIdx = 0; |
| 1440 | const CUData* cuLeft = getPULeft(leftPartIdx, partIdxLB); |
| 1441 | bool isAvailableA1 = cuLeft && |
| 1442 | cuLeft->isDiffMER(xP - 1, yP + nPSH - 1, xP, yP) && |
| 1443 | !(puIdx == 1 && (curPS == SIZE_Nx2N || curPS == SIZE_nLx2N || curPS == SIZE_nRx2N)) && |
| 1444 | !cuLeft->isIntra(leftPartIdx); |
| 1445 | if (isAvailableA1) |
| 1446 | { |
| 1447 | // get Inter Dir |
| 1448 | interDirNeighbours[count] = cuLeft->m_interDir[leftPartIdx]; |
| 1449 | // get Mv from Left |
| 1450 | cuLeft->getMvField(cuLeft, leftPartIdx, 0, mvFieldNeighbours[count][0]); |
| 1451 | if (isInterB) |
| 1452 | cuLeft->getMvField(cuLeft, leftPartIdx, 1, mvFieldNeighbours[count][1]); |
| 1453 | |
| 1454 | count++; |
| 1455 | |
| 1456 | if (count == maxNumMergeCand) |
| 1457 | return maxNumMergeCand; |
| 1458 | } |
| 1459 | |
| 1460 | deriveLeftRightTopIdx(puIdx, partIdxLT, partIdxRT); |
| 1461 | |
| 1462 | // above |
| 1463 | uint32_t abovePartIdx = 0; |
| 1464 | const CUData* cuAbove = getPUAbove(abovePartIdx, partIdxRT); |
| 1465 | bool isAvailableB1 = cuAbove && |
| 1466 | cuAbove->isDiffMER(xP + nPSW - 1, yP - 1, xP, yP) && |
| 1467 | !(puIdx == 1 && (curPS == SIZE_2NxN || curPS == SIZE_2NxnU || curPS == SIZE_2NxnD)) && |
| 1468 | !cuAbove->isIntra(abovePartIdx); |
| 1469 | if (isAvailableB1 && (!isAvailableA1 || !cuLeft->hasEqualMotion(leftPartIdx, *cuAbove, abovePartIdx))) |
| 1470 | { |
| 1471 | // get Inter Dir |
| 1472 | interDirNeighbours[count] = cuAbove->m_interDir[abovePartIdx]; |
| 1473 | // get Mv from Left |
| 1474 | cuAbove->getMvField(cuAbove, abovePartIdx, 0, mvFieldNeighbours[count][0]); |
| 1475 | if (isInterB) |
| 1476 | cuAbove->getMvField(cuAbove, abovePartIdx, 1, mvFieldNeighbours[count][1]); |
| 1477 | |
| 1478 | count++; |
| 1479 | |
| 1480 | if (count == maxNumMergeCand) |
| 1481 | return maxNumMergeCand; |
| 1482 | } |
| 1483 | |
| 1484 | // above right |
| 1485 | uint32_t aboveRightPartIdx = 0; |
| 1486 | const CUData* cuAboveRight = getPUAboveRight(aboveRightPartIdx, partIdxRT); |
| 1487 | bool isAvailableB0 = cuAboveRight && |
| 1488 | cuAboveRight->isDiffMER(xP + nPSW, yP - 1, xP, yP) && |
| 1489 | !cuAboveRight->isIntra(aboveRightPartIdx); |
| 1490 | if (isAvailableB0 && (!isAvailableB1 || !cuAbove->hasEqualMotion(abovePartIdx, *cuAboveRight, aboveRightPartIdx))) |
| 1491 | { |
| 1492 | // get Inter Dir |
| 1493 | interDirNeighbours[count] = cuAboveRight->m_interDir[aboveRightPartIdx]; |
| 1494 | // get Mv from Left |
| 1495 | cuAboveRight->getMvField(cuAboveRight, aboveRightPartIdx, 0, mvFieldNeighbours[count][0]); |
| 1496 | if (isInterB) |
| 1497 | cuAboveRight->getMvField(cuAboveRight, aboveRightPartIdx, 1, mvFieldNeighbours[count][1]); |
| 1498 | |
| 1499 | count++; |
| 1500 | |
| 1501 | if (count == maxNumMergeCand) |
| 1502 | return maxNumMergeCand; |
| 1503 | } |
| 1504 | |
| 1505 | // left bottom |
| 1506 | uint32_t leftBottomPartIdx = 0; |
| 1507 | const CUData* cuLeftBottom = this->getPUBelowLeft(leftBottomPartIdx, partIdxLB); |
| 1508 | bool isAvailableA0 = cuLeftBottom && |
| 1509 | cuLeftBottom->isDiffMER(xP - 1, yP + nPSH, xP, yP) && |
| 1510 | !cuLeftBottom->isIntra(leftBottomPartIdx); |
| 1511 | if (isAvailableA0 && (!isAvailableA1 || !cuLeft->hasEqualMotion(leftPartIdx, *cuLeftBottom, leftBottomPartIdx))) |
| 1512 | { |
| 1513 | // get Inter Dir |
| 1514 | interDirNeighbours[count] = cuLeftBottom->m_interDir[leftBottomPartIdx]; |
| 1515 | // get Mv from Left |
| 1516 | cuLeftBottom->getMvField(cuLeftBottom, leftBottomPartIdx, 0, mvFieldNeighbours[count][0]); |
| 1517 | if (isInterB) |
| 1518 | cuLeftBottom->getMvField(cuLeftBottom, leftBottomPartIdx, 1, mvFieldNeighbours[count][1]); |
| 1519 | |
| 1520 | count++; |
| 1521 | |
| 1522 | if (count == maxNumMergeCand) |
| 1523 | return maxNumMergeCand; |
| 1524 | } |
| 1525 | |
| 1526 | // above left |
| 1527 | if (count < 4) |
| 1528 | { |
| 1529 | uint32_t aboveLeftPartIdx = 0; |
| 1530 | const CUData* cuAboveLeft = getPUAboveLeft(aboveLeftPartIdx, absPartAddr); |
| 1531 | bool isAvailableB2 = cuAboveLeft && |
| 1532 | cuAboveLeft->isDiffMER(xP - 1, yP - 1, xP, yP) && |
| 1533 | !cuAboveLeft->isIntra(aboveLeftPartIdx); |
| 1534 | if (isAvailableB2 && (!isAvailableA1 || !cuLeft->hasEqualMotion(leftPartIdx, *cuAboveLeft, aboveLeftPartIdx)) |
| 1535 | && (!isAvailableB1 || !cuAbove->hasEqualMotion(abovePartIdx, *cuAboveLeft, aboveLeftPartIdx))) |
| 1536 | { |
| 1537 | // get Inter Dir |
| 1538 | interDirNeighbours[count] = cuAboveLeft->m_interDir[aboveLeftPartIdx]; |
| 1539 | // get Mv from Left |
| 1540 | cuAboveLeft->getMvField(cuAboveLeft, aboveLeftPartIdx, 0, mvFieldNeighbours[count][0]); |
| 1541 | if (isInterB) |
| 1542 | cuAboveLeft->getMvField(cuAboveLeft, aboveLeftPartIdx, 1, mvFieldNeighbours[count][1]); |
| 1543 | |
| 1544 | count++; |
| 1545 | |
| 1546 | if (count == maxNumMergeCand) |
| 1547 | return maxNumMergeCand; |
| 1548 | } |
| 1549 | } |
| 1550 | if (m_slice->m_sps->bTemporalMVPEnabled) |
| 1551 | { |
| 1552 | uint32_t partIdxRB = deriveRightBottomIdx(puIdx); |
| 1553 | MV colmv; |
| 1554 | int ctuIdx = -1; |
| 1555 | |
| 1556 | // image boundary check |
| 1557 | if (m_encData->getPicCTU(m_cuAddr)->m_cuPelX + g_zscanToPelX[partIdxRB] + UNIT_SIZE < m_slice->m_sps->picWidthInLumaSamples && |
| 1558 | m_encData->getPicCTU(m_cuAddr)->m_cuPelY + g_zscanToPelY[partIdxRB] + UNIT_SIZE < m_slice->m_sps->picHeightInLumaSamples) |
| 1559 | { |
| 1560 | uint32_t absPartIdxRB = g_zscanToRaster[partIdxRB]; |
| 1561 | uint32_t numPartInCUSize = s_numPartInCUSize; |
| 1562 | bool bNotLastCol = lessThanCol(absPartIdxRB, numPartInCUSize - 1, numPartInCUSize); // is not at the last column of CTU |
| 1563 | bool bNotLastRow = lessThanRow(absPartIdxRB, numPartInCUSize - 1, numPartInCUSize); // is not at the last row of CTU |
| 1564 | |
| 1565 | if (bNotLastCol && bNotLastRow) |
| 1566 | { |
| 1567 | absPartAddr = g_rasterToZscan[absPartIdxRB + numPartInCUSize + 1]; |
| 1568 | ctuIdx = m_cuAddr; |
| 1569 | } |
| 1570 | else if (bNotLastCol) |
| 1571 | absPartAddr = g_rasterToZscan[(absPartIdxRB + numPartInCUSize + 1) & (numPartInCUSize - 1)]; |
| 1572 | else if (bNotLastRow) |
| 1573 | { |
| 1574 | absPartAddr = g_rasterToZscan[absPartIdxRB + 1]; |
| 1575 | ctuIdx = m_cuAddr + 1; |
| 1576 | } |
| 1577 | else // is the right bottom corner of CTU |
| 1578 | absPartAddr = 0; |
| 1579 | } |
| 1580 | |
| 1581 | int refIdx = 0; |
| 1582 | uint32_t partIdxCenter = deriveCenterIdx(puIdx); |
| 1583 | uint32_t curCTUIdx = m_cuAddr; |
| 1584 | int dir = 0; |
| 1585 | bool bExistMV = ctuIdx >= 0 && getColMVP(colmv, refIdx, 0, ctuIdx, absPartAddr); |
| 1586 | if (!bExistMV) |
| 1587 | bExistMV = getColMVP(colmv, refIdx, 0, curCTUIdx, partIdxCenter); |
| 1588 | if (bExistMV) |
| 1589 | { |
| 1590 | dir |= 1; |
| 1591 | mvFieldNeighbours[count][0].mv = colmv; |
| 1592 | mvFieldNeighbours[count][0].refIdx = refIdx; |
| 1593 | } |
| 1594 | |
| 1595 | if (isInterB) |
| 1596 | { |
| 1597 | bExistMV = ctuIdx >= 0 && getColMVP(colmv, refIdx, 1, ctuIdx, absPartAddr); |
| 1598 | if (!bExistMV) |
| 1599 | bExistMV = getColMVP(colmv, refIdx, 1, curCTUIdx, partIdxCenter); |
| 1600 | |
| 1601 | if (bExistMV) |
| 1602 | { |
| 1603 | dir |= 2; |
| 1604 | mvFieldNeighbours[count][1].mv = colmv; |
| 1605 | mvFieldNeighbours[count][1].refIdx = refIdx; |
| 1606 | } |
| 1607 | } |
| 1608 | |
| 1609 | if (dir != 0) |
| 1610 | { |
| 1611 | interDirNeighbours[count] = (uint8_t)dir; |
| 1612 | |
| 1613 | count++; |
| 1614 | |
| 1615 | if (count == maxNumMergeCand) |
| 1616 | return maxNumMergeCand; |
| 1617 | } |
| 1618 | } |
| 1619 | |
| 1620 | if (isInterB) |
| 1621 | { |
| 1622 | const uint32_t cutoff = count * (count - 1); |
| 1623 | uint32_t priorityList0 = 0xEDC984; // { 0, 1, 0, 2, 1, 2, 0, 3, 1, 3, 2, 3 } |
| 1624 | uint32_t priorityList1 = 0xB73621; // { 1, 0, 2, 0, 2, 1, 3, 0, 3, 1, 3, 2 } |
| 1625 | |
| 1626 | for (uint32_t idx = 0; idx < cutoff; idx++) |
| 1627 | { |
| 1628 | int i = priorityList0 & 3; |
| 1629 | int j = priorityList1 & 3; |
| 1630 | priorityList0 >>= 2; |
| 1631 | priorityList1 >>= 2; |
| 1632 | |
| 1633 | if ((interDirNeighbours[i] & 0x1) && (interDirNeighbours[j] & 0x2)) |
| 1634 | { |
| 1635 | // get Mv from cand[i] and cand[j] |
| 1636 | int refIdxL0 = mvFieldNeighbours[i][0].refIdx; |
| 1637 | int refIdxL1 = mvFieldNeighbours[j][1].refIdx; |
| 1638 | int refPOCL0 = m_slice->m_refPOCList[0][refIdxL0]; |
| 1639 | int refPOCL1 = m_slice->m_refPOCList[1][refIdxL1]; |
| 1640 | if (!(refPOCL0 == refPOCL1 && mvFieldNeighbours[i][0].mv == mvFieldNeighbours[j][1].mv)) |
| 1641 | { |
| 1642 | mvFieldNeighbours[count][0].mv = mvFieldNeighbours[i][0].mv; |
| 1643 | mvFieldNeighbours[count][0].refIdx = refIdxL0; |
| 1644 | mvFieldNeighbours[count][1].mv = mvFieldNeighbours[j][1].mv; |
| 1645 | mvFieldNeighbours[count][1].refIdx = refIdxL1; |
| 1646 | interDirNeighbours[count] = 3; |
| 1647 | |
| 1648 | count++; |
| 1649 | |
| 1650 | if (count == maxNumMergeCand) |
| 1651 | return maxNumMergeCand; |
| 1652 | } |
| 1653 | } |
| 1654 | } |
| 1655 | } |
| 1656 | int numRefIdx = (isInterB) ? X265_MIN(m_slice->m_numRefIdx[0], m_slice->m_numRefIdx[1]) : m_slice->m_numRefIdx[0]; |
| 1657 | int r = 0; |
| 1658 | int refcnt = 0; |
| 1659 | while (count < maxNumMergeCand) |
| 1660 | { |
| 1661 | interDirNeighbours[count] = 1; |
| 1662 | mvFieldNeighbours[count][0].mv.word = 0; |
| 1663 | mvFieldNeighbours[count][0].refIdx = r; |
| 1664 | |
| 1665 | if (isInterB) |
| 1666 | { |
| 1667 | interDirNeighbours[count] = 3; |
| 1668 | mvFieldNeighbours[count][1].mv.word = 0; |
| 1669 | mvFieldNeighbours[count][1].refIdx = r; |
| 1670 | } |
| 1671 | |
| 1672 | count++; |
| 1673 | |
| 1674 | if (refcnt == numRefIdx - 1) |
| 1675 | r = 0; |
| 1676 | else |
| 1677 | { |
| 1678 | ++r; |
| 1679 | ++refcnt; |
| 1680 | } |
| 1681 | } |
| 1682 | |
| 1683 | return count; |
| 1684 | } |
| 1685 | |
| 1686 | /* Check whether the current PU and a spatial neighboring PU are in a same ME region */ |
| 1687 | bool CUData::isDiffMER(int xN, int yN, int xP, int yP) const |
| 1688 | { |
| 1689 | uint32_t plevel = 2; |
| 1690 | |
| 1691 | if ((xN >> plevel) != (xP >> plevel)) |
| 1692 | return true; |
| 1693 | if ((yN >> plevel) != (yP >> plevel)) |
| 1694 | return true; |
| 1695 | return false; |
| 1696 | } |
| 1697 | |
| 1698 | /* Constructs a list of candidates for AMVP, and a larger list of motion candidates */ |
| 1699 | int CUData::fillMvpCand(uint32_t puIdx, uint32_t absPartIdx, int picList, int refIdx, MV* amvpCand, MV* mvc) const |
| 1700 | { |
| 1701 | int num = 0; |
| 1702 | |
| 1703 | // spatial MV |
| 1704 | uint32_t partIdxLT, partIdxRT, partIdxLB = deriveLeftBottomIdx(puIdx); |
| 1705 | |
| 1706 | deriveLeftRightTopIdx(puIdx, partIdxLT, partIdxRT); |
| 1707 | |
| 1708 | MV mv[MD_ABOVE_LEFT + 1]; |
| 1709 | MV mvOrder[MD_ABOVE_LEFT + 1]; |
| 1710 | bool valid[MD_ABOVE_LEFT + 1]; |
| 1711 | bool validOrder[MD_ABOVE_LEFT + 1]; |
| 1712 | |
| 1713 | valid[MD_BELOW_LEFT] = addMVPCand(mv[MD_BELOW_LEFT], picList, refIdx, partIdxLB, MD_BELOW_LEFT); |
| 1714 | valid[MD_LEFT] = addMVPCand(mv[MD_LEFT], picList, refIdx, partIdxLB, MD_LEFT); |
| 1715 | valid[MD_ABOVE_RIGHT] = addMVPCand(mv[MD_ABOVE_RIGHT], picList, refIdx, partIdxRT, MD_ABOVE_RIGHT); |
| 1716 | valid[MD_ABOVE] = addMVPCand(mv[MD_ABOVE], picList, refIdx, partIdxRT, MD_ABOVE); |
| 1717 | valid[MD_ABOVE_LEFT] = addMVPCand(mv[MD_ABOVE_LEFT], picList, refIdx, partIdxLT, MD_ABOVE_LEFT); |
| 1718 | |
| 1719 | validOrder[MD_BELOW_LEFT] = addMVPCandOrder(mvOrder[MD_BELOW_LEFT], picList, refIdx, partIdxLB, MD_BELOW_LEFT); |
| 1720 | validOrder[MD_LEFT] = addMVPCandOrder(mvOrder[MD_LEFT], picList, refIdx, partIdxLB, MD_LEFT); |
| 1721 | validOrder[MD_ABOVE_RIGHT] = addMVPCandOrder(mvOrder[MD_ABOVE_RIGHT], picList, refIdx, partIdxRT, MD_ABOVE_RIGHT); |
| 1722 | validOrder[MD_ABOVE] = addMVPCandOrder(mvOrder[MD_ABOVE], picList, refIdx, partIdxRT, MD_ABOVE); |
| 1723 | validOrder[MD_ABOVE_LEFT] = addMVPCandOrder(mvOrder[MD_ABOVE_LEFT], picList, refIdx, partIdxLT, MD_ABOVE_LEFT); |
| 1724 | |
| 1725 | // Left predictor search |
| 1726 | if (valid[MD_BELOW_LEFT]) |
| 1727 | amvpCand[num++] = mv[MD_BELOW_LEFT]; |
| 1728 | else if (valid[MD_LEFT]) |
| 1729 | amvpCand[num++] = mv[MD_LEFT]; |
| 1730 | else if (validOrder[MD_BELOW_LEFT]) |
| 1731 | amvpCand[num++] = mvOrder[MD_BELOW_LEFT]; |
| 1732 | else if (validOrder[MD_LEFT]) |
| 1733 | amvpCand[num++] = mvOrder[MD_LEFT]; |
| 1734 | |
| 1735 | bool bAddedSmvp = num > 0; |
| 1736 | |
| 1737 | // Above predictor search |
| 1738 | if (valid[MD_ABOVE_RIGHT]) |
| 1739 | amvpCand[num++] = mv[MD_ABOVE_RIGHT]; |
| 1740 | else if (valid[MD_ABOVE]) |
| 1741 | amvpCand[num++] = mv[MD_ABOVE]; |
| 1742 | else if (valid[MD_ABOVE_LEFT]) |
| 1743 | amvpCand[num++] = mv[MD_ABOVE_LEFT]; |
| 1744 | |
| 1745 | if (!bAddedSmvp) |
| 1746 | { |
| 1747 | if (validOrder[MD_ABOVE_RIGHT]) |
| 1748 | amvpCand[num++] = mvOrder[MD_ABOVE_RIGHT]; |
| 1749 | else if (validOrder[MD_ABOVE]) |
| 1750 | amvpCand[num++] = mvOrder[MD_ABOVE]; |
| 1751 | else if (validOrder[MD_ABOVE_LEFT]) |
| 1752 | amvpCand[num++] = mvOrder[MD_ABOVE_LEFT]; |
| 1753 | } |
| 1754 | |
| 1755 | int numMvc = 0; |
| 1756 | for (int dir = MD_LEFT; dir <= MD_ABOVE_LEFT; dir++) |
| 1757 | { |
| 1758 | if (valid[dir] && mv[dir].notZero()) |
| 1759 | mvc[numMvc++] = mv[dir]; |
| 1760 | |
| 1761 | if (validOrder[dir] && mvOrder[dir].notZero()) |
| 1762 | mvc[numMvc++] = mvOrder[dir]; |
| 1763 | } |
| 1764 | |
| 1765 | if (num == 2) |
| 1766 | { |
| 1767 | if (amvpCand[0] == amvpCand[1]) |
| 1768 | num = 1; |
| 1769 | else |
| 1770 | /* AMVP_NUM_CANDS = 2 */ |
| 1771 | return numMvc; |
| 1772 | } |
| 1773 | |
| 1774 | if (m_slice->m_sps->bTemporalMVPEnabled) |
| 1775 | { |
| 1776 | uint32_t absPartAddr = m_absIdxInCTU + absPartIdx; |
| 1777 | uint32_t partIdxRB = deriveRightBottomIdx(puIdx); |
| 1778 | MV colmv; |
| 1779 | |
| 1780 | // co-located RightBottom temporal predictor (H) |
| 1781 | int ctuIdx = -1; |
| 1782 | |
| 1783 | // image boundary check |
| 1784 | if (m_encData->getPicCTU(m_cuAddr)->m_cuPelX + g_zscanToPelX[partIdxRB] + UNIT_SIZE < m_slice->m_sps->picWidthInLumaSamples && |
| 1785 | m_encData->getPicCTU(m_cuAddr)->m_cuPelY + g_zscanToPelY[partIdxRB] + UNIT_SIZE < m_slice->m_sps->picHeightInLumaSamples) |
| 1786 | { |
| 1787 | uint32_t absPartIdxRB = g_zscanToRaster[partIdxRB]; |
| 1788 | uint32_t numPartInCUSize = s_numPartInCUSize; |
| 1789 | bool bNotLastCol = lessThanCol(absPartIdxRB, numPartInCUSize - 1, numPartInCUSize); // is not at the last column of CTU |
| 1790 | bool bNotLastRow = lessThanRow(absPartIdxRB, numPartInCUSize - 1, numPartInCUSize); // is not at the last row of CTU |
| 1791 | |
| 1792 | if (bNotLastCol && bNotLastRow) |
| 1793 | { |
| 1794 | absPartAddr = g_rasterToZscan[absPartIdxRB + numPartInCUSize + 1]; |
| 1795 | ctuIdx = m_cuAddr; |
| 1796 | } |
| 1797 | else if (bNotLastCol) |
| 1798 | absPartAddr = g_rasterToZscan[(absPartIdxRB + numPartInCUSize + 1) & (numPartInCUSize - 1)]; |
| 1799 | else if (bNotLastRow) |
| 1800 | { |
| 1801 | absPartAddr = g_rasterToZscan[absPartIdxRB + 1]; |
| 1802 | ctuIdx = m_cuAddr + 1; |
| 1803 | } |
| 1804 | else // is the right bottom corner of CTU |
| 1805 | absPartAddr = 0; |
| 1806 | } |
| 1807 | if (ctuIdx >= 0 && getColMVP(colmv, refIdx, picList, ctuIdx, absPartAddr)) |
| 1808 | { |
| 1809 | amvpCand[num++] = colmv; |
| 1810 | mvc[numMvc++] = colmv; |
| 1811 | } |
| 1812 | else |
| 1813 | { |
| 1814 | uint32_t partIdxCenter = deriveCenterIdx(puIdx); |
| 1815 | uint32_t curCTUIdx = m_cuAddr; |
| 1816 | if (getColMVP(colmv, refIdx, picList, curCTUIdx, partIdxCenter)) |
| 1817 | { |
| 1818 | amvpCand[num++] = colmv; |
| 1819 | mvc[numMvc++] = colmv; |
| 1820 | } |
| 1821 | } |
| 1822 | } |
| 1823 | |
| 1824 | while (num < AMVP_NUM_CANDS) |
| 1825 | amvpCand[num++] = 0; |
| 1826 | |
| 1827 | return numMvc; |
| 1828 | } |
| 1829 | |
| 1830 | void CUData::clipMv(MV& outMV) const |
| 1831 | { |
| 1832 | int mvshift = 2; |
| 1833 | int offset = 8; |
| 1834 | int xmax = (m_slice->m_sps->picWidthInLumaSamples + offset - m_cuPelX - 1) << mvshift; |
| 1835 | int xmin = (-(int)g_maxCUSize - offset - (int)m_cuPelX + 1) << mvshift; |
| 1836 | |
| 1837 | int ymax = (m_slice->m_sps->picHeightInLumaSamples + offset - m_cuPelY - 1) << mvshift; |
| 1838 | int ymin = (-(int)g_maxCUSize - offset - (int)m_cuPelY + 1) << mvshift; |
| 1839 | |
| 1840 | outMV.x = (int16_t)X265_MIN(xmax, X265_MAX(xmin, (int)outMV.x)); |
| 1841 | outMV.y = (int16_t)X265_MIN(ymax, X265_MAX(ymin, (int)outMV.y)); |
| 1842 | } |
| 1843 | |
| 1844 | bool CUData::addMVPCand(MV& mvp, int picList, int refIdx, uint32_t partUnitIdx, MVP_DIR dir) const |
| 1845 | { |
| 1846 | const CUData* tmpCU = NULL; |
| 1847 | uint32_t idx = 0; |
| 1848 | |
| 1849 | switch (dir) |
| 1850 | { |
| 1851 | case MD_LEFT: |
| 1852 | tmpCU = getPULeft(idx, partUnitIdx); |
| 1853 | break; |
| 1854 | case MD_ABOVE: |
| 1855 | tmpCU = getPUAbove(idx, partUnitIdx); |
| 1856 | break; |
| 1857 | case MD_ABOVE_RIGHT: |
| 1858 | tmpCU = getPUAboveRight(idx, partUnitIdx); |
| 1859 | break; |
| 1860 | case MD_BELOW_LEFT: |
| 1861 | tmpCU = getPUBelowLeft(idx, partUnitIdx); |
| 1862 | break; |
| 1863 | case MD_ABOVE_LEFT: |
| 1864 | tmpCU = getPUAboveLeft(idx, partUnitIdx); |
| 1865 | break; |
| 1866 | default: |
| 1867 | return false; |
| 1868 | } |
| 1869 | |
| 1870 | if (!tmpCU) |
| 1871 | return false; |
| 1872 | |
| 1873 | int refPOC = m_slice->m_refPOCList[picList][refIdx]; |
| 1874 | int partRefIdx = tmpCU->m_refIdx[picList][idx]; |
| 1875 | if (partRefIdx >= 0 && refPOC == tmpCU->m_slice->m_refPOCList[picList][partRefIdx]) |
| 1876 | { |
| 1877 | mvp = tmpCU->m_mv[picList][idx]; |
| 1878 | return true; |
| 1879 | } |
| 1880 | |
| 1881 | int refPicList2nd = 0; |
| 1882 | if (picList == 0) |
| 1883 | refPicList2nd = 1; |
| 1884 | else if (picList == 1) |
| 1885 | refPicList2nd = 0; |
| 1886 | |
| 1887 | int curRefPOC = m_slice->m_refPOCList[picList][refIdx]; |
| 1888 | int neibRefPOC; |
| 1889 | |
| 1890 | partRefIdx = tmpCU->m_refIdx[refPicList2nd][idx]; |
| 1891 | if (partRefIdx >= 0) |
| 1892 | { |
| 1893 | neibRefPOC = tmpCU->m_slice->m_refPOCList[refPicList2nd][partRefIdx]; |
| 1894 | if (neibRefPOC == curRefPOC) |
| 1895 | { |
| 1896 | // Same reference frame but different list |
| 1897 | mvp = tmpCU->m_mv[refPicList2nd][idx]; |
| 1898 | return true; |
| 1899 | } |
| 1900 | } |
| 1901 | return false; |
| 1902 | } |
| 1903 | |
| 1904 | bool CUData::addMVPCandOrder(MV& outMV, int picList, int refIdx, uint32_t partUnitIdx, MVP_DIR dir) const |
| 1905 | { |
| 1906 | const CUData* tmpCU = NULL; |
| 1907 | uint32_t idx = 0; |
| 1908 | |
| 1909 | switch (dir) |
| 1910 | { |
| 1911 | case MD_LEFT: |
| 1912 | tmpCU = getPULeft(idx, partUnitIdx); |
| 1913 | break; |
| 1914 | case MD_ABOVE: |
| 1915 | tmpCU = getPUAbove(idx, partUnitIdx); |
| 1916 | break; |
| 1917 | case MD_ABOVE_RIGHT: |
| 1918 | tmpCU = getPUAboveRight(idx, partUnitIdx); |
| 1919 | break; |
| 1920 | case MD_BELOW_LEFT: |
| 1921 | tmpCU = getPUBelowLeft(idx, partUnitIdx); |
| 1922 | break; |
| 1923 | case MD_ABOVE_LEFT: |
| 1924 | tmpCU = getPUAboveLeft(idx, partUnitIdx); |
| 1925 | break; |
| 1926 | default: |
| 1927 | return false; |
| 1928 | } |
| 1929 | |
| 1930 | if (!tmpCU) |
| 1931 | return false; |
| 1932 | |
| 1933 | int refPicList2nd = 0; |
| 1934 | if (picList == 0) |
| 1935 | refPicList2nd = 1; |
| 1936 | else if (picList == 1) |
| 1937 | refPicList2nd = 0; |
| 1938 | |
| 1939 | int curPOC = m_slice->m_poc; |
| 1940 | int curRefPOC = m_slice->m_refPOCList[picList][refIdx]; |
| 1941 | int neibPOC = curPOC; |
| 1942 | int neibRefPOC; |
| 1943 | |
| 1944 | int partRefIdx = tmpCU->m_refIdx[picList][idx]; |
| 1945 | if (partRefIdx >= 0) |
| 1946 | { |
| 1947 | neibRefPOC = tmpCU->m_slice->m_refPOCList[picList][partRefIdx]; |
| 1948 | MV mvp = tmpCU->m_mv[picList][idx]; |
| 1949 | |
| 1950 | scaleMvByPOCDist(outMV, mvp, curPOC, curRefPOC, neibPOC, neibRefPOC); |
| 1951 | return true; |
| 1952 | } |
| 1953 | |
| 1954 | partRefIdx = tmpCU->m_refIdx[refPicList2nd][idx]; |
| 1955 | if (partRefIdx >= 0) |
| 1956 | { |
| 1957 | neibRefPOC = tmpCU->m_slice->m_refPOCList[refPicList2nd][partRefIdx]; |
| 1958 | MV mvp = tmpCU->m_mv[refPicList2nd][idx]; |
| 1959 | |
| 1960 | scaleMvByPOCDist(outMV, mvp, curPOC, curRefPOC, neibPOC, neibRefPOC); |
| 1961 | return true; |
| 1962 | } |
| 1963 | |
| 1964 | return false; |
| 1965 | } |
| 1966 | |
| 1967 | bool CUData::getColMVP(MV& outMV, int& outRefIdx, int picList, int cuAddr, int partUnitIdx) const |
| 1968 | { |
| 1969 | uint32_t absPartAddr = partUnitIdx & TMVP_UNIT_MASK; |
| 1970 | |
| 1971 | int colRefPicList; |
| 1972 | int colPOC, colRefPOC, curPOC, curRefPOC; |
| 1973 | MV colmv; |
| 1974 | |
| 1975 | // use coldir. |
| 1976 | Frame *colPic = m_slice->m_refPicList[m_slice->isInterB() ? 1 - m_slice->m_colFromL0Flag : 0][m_slice->m_colRefIdx]; |
| 1977 | CUData *colCU = colPic->m_encData->getPicCTU(cuAddr); |
| 1978 | |
| 1979 | if (colCU->m_partSize[partUnitIdx] == SIZE_NONE) |
| 1980 | return false; |
| 1981 | |
| 1982 | curPOC = m_slice->m_poc; |
| 1983 | colPOC = colCU->m_slice->m_poc; |
| 1984 | |
| 1985 | if (colCU->isIntra(absPartAddr)) |
| 1986 | return false; |
| 1987 | |
| 1988 | colRefPicList = m_slice->m_bCheckLDC ? picList : m_slice->m_colFromL0Flag; |
| 1989 | |
| 1990 | int colRefIdx = colCU->m_refIdx[colRefPicList][absPartAddr]; |
| 1991 | |
| 1992 | if (colRefIdx < 0) |
| 1993 | { |
| 1994 | colRefPicList = 1 - colRefPicList; |
| 1995 | colRefIdx = colCU->m_refIdx[colRefPicList][absPartAddr]; |
| 1996 | |
| 1997 | if (colRefIdx < 0) |
| 1998 | return false; |
| 1999 | } |
| 2000 | |
| 2001 | // Scale the vector |
| 2002 | colRefPOC = colCU->m_slice->m_refPOCList[colRefPicList][colRefIdx]; |
| 2003 | colmv = colCU->m_mv[colRefPicList][absPartAddr]; |
| 2004 | curRefPOC = m_slice->m_refPOCList[picList][outRefIdx]; |
| 2005 | |
| 2006 | scaleMvByPOCDist(outMV, colmv, curPOC, curRefPOC, colPOC, colRefPOC); |
| 2007 | return true; |
| 2008 | } |
| 2009 | |
| 2010 | void CUData::scaleMvByPOCDist(MV& outMV, const MV& inMV, int curPOC, int curRefPOC, int colPOC, int colRefPOC) const |
| 2011 | { |
| 2012 | int diffPocD = colPOC - colRefPOC; |
| 2013 | int diffPocB = curPOC - curRefPOC; |
| 2014 | |
| 2015 | if (diffPocD == diffPocB) |
| 2016 | outMV = inMV; |
| 2017 | else |
| 2018 | { |
| 2019 | int tdb = Clip3(-128, 127, diffPocB); |
| 2020 | int tdd = Clip3(-128, 127, diffPocD); |
| 2021 | int x = (0x4000 + abs(tdd / 2)) / tdd; |
| 2022 | int scale = Clip3(-4096, 4095, (tdb * x + 32) >> 6); |
| 2023 | outMV = scaleMv(inMV, scale); |
| 2024 | } |
| 2025 | } |
| 2026 | |
| 2027 | uint32_t CUData::deriveCenterIdx(uint32_t puIdx) const |
| 2028 | { |
| 2029 | uint32_t absPartIdx; |
| 2030 | int puWidth, puHeight; |
| 2031 | |
| 2032 | getPartIndexAndSize(puIdx, absPartIdx, puWidth, puHeight); |
| 2033 | |
| 2034 | return g_rasterToZscan[g_zscanToRaster[m_absIdxInCTU + absPartIdx] |
| 2035 | + (puHeight >> (LOG2_UNIT_SIZE + 1)) * s_numPartInCUSize |
| 2036 | + (puWidth >> (LOG2_UNIT_SIZE + 1))]; |
| 2037 | } |
| 2038 | |
| 2039 | ScanType CUData::getCoefScanIdx(uint32_t absPartIdx, uint32_t log2TrSize, bool bIsLuma, bool bIsIntra) const |
| 2040 | { |
| 2041 | uint32_t dirMode; |
| 2042 | |
| 2043 | if (!bIsIntra) |
| 2044 | return SCAN_DIAG; |
| 2045 | |
| 2046 | // check that MDCS can be used for this TU |
| 2047 | if (bIsLuma) |
| 2048 | { |
| 2049 | if (log2TrSize > MDCS_LOG2_MAX_SIZE) |
| 2050 | return SCAN_DIAG; |
| 2051 | |
| 2052 | dirMode = m_lumaIntraDir[absPartIdx]; |
| 2053 | } |
| 2054 | else |
| 2055 | { |
| 2056 | if (log2TrSize > (uint32_t)(MDCS_LOG2_MAX_SIZE - m_hChromaShift)) |
| 2057 | return SCAN_DIAG; |
| 2058 | |
| 2059 | dirMode = m_chromaIntraDir[absPartIdx]; |
| 2060 | if (dirMode == DM_CHROMA_IDX) |
| 2061 | { |
| 2062 | dirMode = m_lumaIntraDir[(m_chromaFormat == X265_CSP_I444) ? absPartIdx : absPartIdx & 0xFC]; |
| 2063 | dirMode = (m_chromaFormat == X265_CSP_I422) ? g_chroma422IntraAngleMappingTable[dirMode] : dirMode; |
| 2064 | } |
| 2065 | } |
| 2066 | |
| 2067 | if (abs((int)dirMode - VER_IDX) <= MDCS_ANGLE_LIMIT) |
| 2068 | return SCAN_HOR; |
| 2069 | else if (abs((int)dirMode - HOR_IDX) <= MDCS_ANGLE_LIMIT) |
| 2070 | return SCAN_VER; |
| 2071 | else |
| 2072 | return SCAN_DIAG; |
| 2073 | } |
| 2074 | |
| 2075 | void CUData::getTUEntropyCodingParameters(TUEntropyCodingParameters &result, uint32_t absPartIdx, uint32_t log2TrSize, bool bIsLuma) const |
| 2076 | { |
| 2077 | // set the group layout |
| 2078 | result.log2TrSizeCG = log2TrSize - 2; |
| 2079 | |
| 2080 | // set the scan orders |
| 2081 | result.scanType = getCoefScanIdx(absPartIdx, log2TrSize, bIsLuma, isIntra(absPartIdx)); |
| 2082 | result.scan = g_scanOrder[result.scanType][log2TrSize - 2]; |
| 2083 | result.scanCG = g_scanOrderCG[result.scanType][result.log2TrSizeCG]; |
| 2084 | |
| 2085 | if (log2TrSize == 2) |
| 2086 | result.firstSignificanceMapContext = 0; |
| 2087 | else if (log2TrSize == 3) |
| 2088 | { |
| 2089 | result.firstSignificanceMapContext = 9; |
| 2090 | if (result.scanType != SCAN_DIAG && bIsLuma) |
| 2091 | result.firstSignificanceMapContext += 6; |
| 2092 | } |
| 2093 | else |
| 2094 | result.firstSignificanceMapContext = bIsLuma ? 21 : 12; |
| 2095 | } |
| 2096 | |
| 2097 | #define CU_SET_FLAG(bitfield, flag, value) (bitfield) = ((bitfield) & (~(flag))) | ((~((value) - 1)) & (flag)) |
| 2098 | |
| 2099 | void CUData::calcCTUGeoms(uint32_t picWidth, uint32_t picHeight, uint32_t maxCUSize, CUGeom cuDataArray[CUGeom::MAX_GEOMS]) const |
| 2100 | { |
| 2101 | // Initialize the coding blocks inside the CTB |
| 2102 | for (uint32_t log2CUSize = g_log2Size[maxCUSize], rangeCUIdx = 0; log2CUSize >= MIN_LOG2_CU_SIZE; log2CUSize--) |
| 2103 | { |
| 2104 | uint32_t blockSize = 1 << log2CUSize; |
| 2105 | uint32_t sbWidth = 1 << (g_log2Size[maxCUSize] - log2CUSize); |
| 2106 | int32_t lastLevelFlag = log2CUSize == MIN_LOG2_CU_SIZE; |
| 2107 | for (uint32_t sbY = 0; sbY < sbWidth; sbY++) |
| 2108 | { |
| 2109 | for (uint32_t sbX = 0; sbX < sbWidth; sbX++) |
| 2110 | { |
| 2111 | uint32_t depthIdx = g_depthScanIdx[sbY][sbX]; |
| 2112 | uint32_t cuIdx = rangeCUIdx + depthIdx; |
| 2113 | uint32_t childIdx = rangeCUIdx + sbWidth * sbWidth + (depthIdx << 2); |
| 2114 | uint32_t px = m_cuPelX + sbX * blockSize; |
| 2115 | uint32_t py = m_cuPelY + sbY * blockSize; |
| 2116 | int32_t presentFlag = px < picWidth && py < picHeight; |
| 2117 | int32_t splitMandatoryFlag = presentFlag && !lastLevelFlag && (px + blockSize > picWidth || py + blockSize > picHeight); |
| 2118 | |
| 2119 | /* Offset of the luma CU in the X, Y direction in terms of pixels from the CTU origin */ |
| 2120 | uint32_t xOffset = (sbX * blockSize) >> 3; |
| 2121 | uint32_t yOffset = (sbY * blockSize) >> 3; |
| 2122 | X265_CHECK(cuIdx < CUGeom::MAX_GEOMS, "CU geom index bug\n"); |
| 2123 | |
| 2124 | CUGeom *cu = cuDataArray + cuIdx; |
| 2125 | cu->log2CUSize = log2CUSize; |
| 2126 | cu->childOffset = childIdx - cuIdx; |
| 2127 | cu->encodeIdx = g_depthScanIdx[yOffset][xOffset] * 4; |
| 2128 | cu->numPartitions = (NUM_CU_PARTITIONS >> ((g_maxLog2CUSize - cu->log2CUSize) * 2)); |
| 2129 | cu->depth = g_log2Size[maxCUSize] - log2CUSize; |
| 2130 | |
| 2131 | cu->flags = 0; |
| 2132 | CU_SET_FLAG(cu->flags, CUGeom::PRESENT, presentFlag); |
| 2133 | CU_SET_FLAG(cu->flags, CUGeom::SPLIT_MANDATORY | CUGeom::SPLIT, splitMandatoryFlag); |
| 2134 | CU_SET_FLAG(cu->flags, CUGeom::LEAF, lastLevelFlag); |
| 2135 | } |
| 2136 | } |
| 2137 | rangeCUIdx += sbWidth * sbWidth; |
| 2138 | } |
| 2139 | } |