| 1 | /***************************************************************************** |
| 2 | * Copyright (C) 2013 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 "framedata.h" |
| 26 | #include "scalinglist.h" |
| 27 | #include "quant.h" |
| 28 | #include "contexts.h" |
| 29 | #include "picyuv.h" |
| 30 | |
| 31 | #include "sao.h" |
| 32 | #include "entropy.h" |
| 33 | |
| 34 | #define CU_DQP_TU_CMAX 5 // max number bins for truncated unary |
| 35 | #define CU_DQP_EG_k 0 // exp-golomb order |
| 36 | #define START_VALUE 8 // start value for dpcm mode |
| 37 | |
| 38 | static const uint32_t g_puOffset[8] = { 0, 8, 4, 4, 2, 10, 1, 5 }; |
| 39 | |
| 40 | namespace x265 { |
| 41 | |
| 42 | Entropy::Entropy() |
| 43 | { |
| 44 | markValid(); |
| 45 | m_fracBits = 0; |
| 46 | X265_CHECK(sizeof(m_contextState) >= sizeof(m_contextState[0]) * MAX_OFF_CTX_MOD, "context state table is too small\n"); |
| 47 | } |
| 48 | |
| 49 | void Entropy::codeVPS(const VPS& vps) |
| 50 | { |
| 51 | WRITE_CODE(0, 4, "vps_video_parameter_set_id"); |
| 52 | WRITE_CODE(3, 2, "vps_reserved_three_2bits"); |
| 53 | WRITE_CODE(0, 6, "vps_reserved_zero_6bits"); |
| 54 | WRITE_CODE(0, 3, "vps_max_sub_layers_minus1"); |
| 55 | WRITE_FLAG(1, "vps_temporal_id_nesting_flag"); |
| 56 | WRITE_CODE(0xffff, 16, "vps_reserved_ffff_16bits"); |
| 57 | |
| 58 | codeProfileTier(vps.ptl); |
| 59 | |
| 60 | WRITE_FLAG(true, "vps_sub_layer_ordering_info_present_flag"); |
| 61 | WRITE_UVLC(vps.maxDecPicBuffering - 1, "vps_max_dec_pic_buffering_minus1[i]"); |
| 62 | WRITE_UVLC(vps.numReorderPics, "vps_num_reorder_pics[i]"); |
| 63 | |
| 64 | WRITE_UVLC(0, "vps_max_latency_increase_plus1[i]"); |
| 65 | WRITE_CODE(0, 6, "vps_max_nuh_reserved_zero_layer_id"); |
| 66 | WRITE_UVLC(0, "vps_max_op_sets_minus1"); |
| 67 | WRITE_FLAG(0, "vps_timing_info_present_flag"); /* we signal timing info in SPS-VUI */ |
| 68 | WRITE_FLAG(0, "vps_extension_flag"); |
| 69 | } |
| 70 | |
| 71 | void Entropy::codeSPS(const SPS& sps, const ScalingList& scalingList, const ProfileTierLevel& ptl) |
| 72 | { |
| 73 | WRITE_CODE(0, 4, "sps_video_parameter_set_id"); |
| 74 | WRITE_CODE(0, 3, "sps_max_sub_layers_minus1"); |
| 75 | WRITE_FLAG(1, "sps_temporal_id_nesting_flag"); |
| 76 | |
| 77 | codeProfileTier(ptl); |
| 78 | |
| 79 | WRITE_UVLC(0, "sps_seq_parameter_set_id"); |
| 80 | WRITE_UVLC(sps.chromaFormatIdc, "chroma_format_idc"); |
| 81 | |
| 82 | if (sps.chromaFormatIdc == X265_CSP_I444) |
| 83 | WRITE_FLAG(0, "separate_colour_plane_flag"); |
| 84 | |
| 85 | WRITE_UVLC(sps.picWidthInLumaSamples, "pic_width_in_luma_samples"); |
| 86 | WRITE_UVLC(sps.picHeightInLumaSamples, "pic_height_in_luma_samples"); |
| 87 | |
| 88 | const Window& conf = sps.conformanceWindow; |
| 89 | WRITE_FLAG(conf.bEnabled, "conformance_window_flag"); |
| 90 | if (conf.bEnabled) |
| 91 | { |
| 92 | int hShift = CHROMA_H_SHIFT(sps.chromaFormatIdc), vShift = CHROMA_V_SHIFT(sps.chromaFormatIdc); |
| 93 | WRITE_UVLC(conf.leftOffset >> hShift, "conf_win_left_offset"); |
| 94 | WRITE_UVLC(conf.rightOffset >> hShift, "conf_win_right_offset"); |
| 95 | WRITE_UVLC(conf.topOffset >> vShift, "conf_win_top_offset"); |
| 96 | WRITE_UVLC(conf.bottomOffset >> vShift, "conf_win_bottom_offset"); |
| 97 | } |
| 98 | |
| 99 | WRITE_UVLC(X265_DEPTH - 8, "bit_depth_luma_minus8"); |
| 100 | WRITE_UVLC(X265_DEPTH - 8, "bit_depth_chroma_minus8"); |
| 101 | WRITE_UVLC(BITS_FOR_POC - 4, "log2_max_pic_order_cnt_lsb_minus4"); |
| 102 | WRITE_FLAG(true, "sps_sub_layer_ordering_info_present_flag"); |
| 103 | |
| 104 | WRITE_UVLC(sps.maxDecPicBuffering - 1, "sps_max_dec_pic_buffering_minus1[i]"); |
| 105 | WRITE_UVLC(sps.numReorderPics, "sps_num_reorder_pics[i]"); |
| 106 | WRITE_UVLC(0, "sps_max_latency_increase_plus1[i]"); |
| 107 | |
| 108 | WRITE_UVLC(sps.log2MinCodingBlockSize - 3, "log2_min_coding_block_size_minus3"); |
| 109 | WRITE_UVLC(sps.log2DiffMaxMinCodingBlockSize, "log2_diff_max_min_coding_block_size"); |
| 110 | WRITE_UVLC(sps.quadtreeTULog2MinSize - 2, "log2_min_transform_block_size_minus2"); |
| 111 | WRITE_UVLC(sps.quadtreeTULog2MaxSize - sps.quadtreeTULog2MinSize, "log2_diff_max_min_transform_block_size"); |
| 112 | WRITE_UVLC(sps.quadtreeTUMaxDepthInter - 1, "max_transform_hierarchy_depth_inter"); |
| 113 | WRITE_UVLC(sps.quadtreeTUMaxDepthIntra - 1, "max_transform_hierarchy_depth_intra"); |
| 114 | WRITE_FLAG(scalingList.m_bEnabled, "scaling_list_enabled_flag"); |
| 115 | if (scalingList.m_bEnabled) |
| 116 | { |
| 117 | WRITE_FLAG(scalingList.m_bDataPresent, "sps_scaling_list_data_present_flag"); |
| 118 | if (scalingList.m_bDataPresent) |
| 119 | codeScalingList(scalingList); |
| 120 | } |
| 121 | WRITE_FLAG(sps.bUseAMP, "amp_enabled_flag"); |
| 122 | WRITE_FLAG(sps.bUseSAO, "sample_adaptive_offset_enabled_flag"); |
| 123 | |
| 124 | WRITE_FLAG(0, "pcm_enabled_flag"); |
| 125 | WRITE_UVLC(0, "num_short_term_ref_pic_sets"); |
| 126 | WRITE_FLAG(0, "long_term_ref_pics_present_flag"); |
| 127 | |
| 128 | WRITE_FLAG(sps.bTemporalMVPEnabled, "sps_temporal_mvp_enable_flag"); |
| 129 | WRITE_FLAG(sps.bUseStrongIntraSmoothing, "sps_strong_intra_smoothing_enable_flag"); |
| 130 | |
| 131 | WRITE_FLAG(1, "vui_parameters_present_flag"); |
| 132 | codeVUI(sps.vuiParameters); |
| 133 | |
| 134 | WRITE_FLAG(0, "sps_extension_flag"); |
| 135 | } |
| 136 | |
| 137 | void Entropy::codePPS(const PPS& pps) |
| 138 | { |
| 139 | WRITE_UVLC(0, "pps_pic_parameter_set_id"); |
| 140 | WRITE_UVLC(0, "pps_seq_parameter_set_id"); |
| 141 | WRITE_FLAG(0, "dependent_slice_segments_enabled_flag"); |
| 142 | WRITE_FLAG(0, "output_flag_present_flag"); |
| 143 | WRITE_CODE(0, 3, "num_extra_slice_header_bits"); |
| 144 | WRITE_FLAG(pps.bSignHideEnabled, "sign_data_hiding_flag"); |
| 145 | WRITE_FLAG(0, "cabac_init_present_flag"); |
| 146 | WRITE_UVLC(0, "num_ref_idx_l0_default_active_minus1"); |
| 147 | WRITE_UVLC(0, "num_ref_idx_l1_default_active_minus1"); |
| 148 | |
| 149 | WRITE_SVLC(0, "init_qp_minus26"); |
| 150 | WRITE_FLAG(pps.bConstrainedIntraPred, "constrained_intra_pred_flag"); |
| 151 | WRITE_FLAG(pps.bTransformSkipEnabled, "transform_skip_enabled_flag"); |
| 152 | |
| 153 | WRITE_FLAG(pps.bUseDQP, "cu_qp_delta_enabled_flag"); |
| 154 | if (pps.bUseDQP) |
| 155 | WRITE_UVLC(pps.maxCuDQPDepth, "diff_cu_qp_delta_depth"); |
| 156 | |
| 157 | WRITE_SVLC(pps.chromaCbQpOffset, "pps_cb_qp_offset"); |
| 158 | WRITE_SVLC(pps.chromaCrQpOffset, "pps_cr_qp_offset"); |
| 159 | WRITE_FLAG(0, "pps_slice_chroma_qp_offsets_present_flag"); |
| 160 | |
| 161 | WRITE_FLAG(pps.bUseWeightPred, "weighted_pred_flag"); |
| 162 | WRITE_FLAG(pps.bUseWeightedBiPred, "weighted_bipred_flag"); |
| 163 | WRITE_FLAG(pps.bTransquantBypassEnabled, "transquant_bypass_enable_flag"); |
| 164 | WRITE_FLAG(0, "tiles_enabled_flag"); |
| 165 | WRITE_FLAG(pps.bEntropyCodingSyncEnabled, "entropy_coding_sync_enabled_flag"); |
| 166 | WRITE_FLAG(1, "loop_filter_across_slices_enabled_flag"); |
| 167 | |
| 168 | WRITE_FLAG(pps.bDeblockingFilterControlPresent, "deblocking_filter_control_present_flag"); |
| 169 | if (pps.bDeblockingFilterControlPresent) |
| 170 | { |
| 171 | WRITE_FLAG(0, "deblocking_filter_override_enabled_flag"); |
| 172 | WRITE_FLAG(pps.bPicDisableDeblockingFilter, "pps_disable_deblocking_filter_flag"); |
| 173 | if (!pps.bPicDisableDeblockingFilter) |
| 174 | { |
| 175 | WRITE_SVLC(pps.deblockingFilterBetaOffsetDiv2, "pps_beta_offset_div2"); |
| 176 | WRITE_SVLC(pps.deblockingFilterTcOffsetDiv2, "pps_tc_offset_div2"); |
| 177 | } |
| 178 | } |
| 179 | |
| 180 | WRITE_FLAG(0, "pps_scaling_list_data_present_flag"); |
| 181 | WRITE_FLAG(0, "lists_modification_present_flag"); |
| 182 | WRITE_UVLC(0, "log2_parallel_merge_level_minus2"); |
| 183 | WRITE_FLAG(0, "slice_segment_header_extension_present_flag"); |
| 184 | WRITE_FLAG(0, "pps_extension_flag"); |
| 185 | } |
| 186 | |
| 187 | void Entropy::codeProfileTier(const ProfileTierLevel& ptl) |
| 188 | { |
| 189 | WRITE_CODE(0, 2, "XXX_profile_space[]"); |
| 190 | WRITE_FLAG(ptl.tierFlag, "XXX_tier_flag[]"); |
| 191 | WRITE_CODE(ptl.profileIdc, 5, "XXX_profile_idc[]"); |
| 192 | for (int j = 0; j < 32; j++) |
| 193 | WRITE_FLAG(ptl.profileCompatibilityFlag[j], "XXX_profile_compatibility_flag[][j]"); |
| 194 | |
| 195 | WRITE_FLAG(ptl.progressiveSourceFlag, "general_progressive_source_flag"); |
| 196 | WRITE_FLAG(ptl.interlacedSourceFlag, "general_interlaced_source_flag"); |
| 197 | WRITE_FLAG(ptl.nonPackedConstraintFlag, "general_non_packed_constraint_flag"); |
| 198 | WRITE_FLAG(ptl.frameOnlyConstraintFlag, "general_frame_only_constraint_flag"); |
| 199 | |
| 200 | if (ptl.profileIdc == Profile::MAINREXT || ptl.profileIdc == Profile::HIGHTHROUGHPUTREXT) |
| 201 | { |
| 202 | uint32_t bitDepthConstraint = ptl.bitDepthConstraint; |
| 203 | int csp = ptl.chromaFormatConstraint; |
| 204 | WRITE_FLAG(bitDepthConstraint<=12, "general_max_12bit_constraint_flag"); |
| 205 | WRITE_FLAG(bitDepthConstraint<=10, "general_max_10bit_constraint_flag"); |
| 206 | WRITE_FLAG(bitDepthConstraint<= 8 && csp != X265_CSP_I422 , "general_max_8bit_constraint_flag"); |
| 207 | WRITE_FLAG(csp == X265_CSP_I422 || csp == X265_CSP_I420 || csp == X265_CSP_I400, "general_max_422chroma_constraint_flag"); |
| 208 | WRITE_FLAG(csp == X265_CSP_I420 || csp == X265_CSP_I400, "general_max_420chroma_constraint_flag"); |
| 209 | WRITE_FLAG(csp == X265_CSP_I400, "general_max_monochrome_constraint_flag"); |
| 210 | WRITE_FLAG(ptl.intraConstraintFlag, "general_intra_constraint_flag"); |
| 211 | WRITE_FLAG(0, "general_one_picture_only_constraint_flag"); |
| 212 | WRITE_FLAG(ptl.lowerBitRateConstraintFlag, "general_lower_bit_rate_constraint_flag"); |
| 213 | WRITE_CODE(0 , 16, "XXX_reserved_zero_35bits[0..15]"); |
| 214 | WRITE_CODE(0 , 16, "XXX_reserved_zero_35bits[16..31]"); |
| 215 | WRITE_CODE(0 , 3, "XXX_reserved_zero_35bits[32..34]"); |
| 216 | } |
| 217 | else |
| 218 | { |
| 219 | WRITE_CODE(0, 16, "XXX_reserved_zero_44bits[0..15]"); |
| 220 | WRITE_CODE(0, 16, "XXX_reserved_zero_44bits[16..31]"); |
| 221 | WRITE_CODE(0, 12, "XXX_reserved_zero_44bits[32..43]"); |
| 222 | } |
| 223 | |
| 224 | WRITE_CODE(ptl.levelIdc, 8, "general_level_idc"); |
| 225 | } |
| 226 | |
| 227 | void Entropy::codeVUI(const VUI& vui) |
| 228 | { |
| 229 | WRITE_FLAG(vui.aspectRatioInfoPresentFlag, "aspect_ratio_info_present_flag"); |
| 230 | if (vui.aspectRatioInfoPresentFlag) |
| 231 | { |
| 232 | WRITE_CODE(vui.aspectRatioIdc, 8, "aspect_ratio_idc"); |
| 233 | if (vui.aspectRatioIdc == 255) |
| 234 | { |
| 235 | WRITE_CODE(vui.sarWidth, 16, "sar_width"); |
| 236 | WRITE_CODE(vui.sarHeight, 16, "sar_height"); |
| 237 | } |
| 238 | } |
| 239 | |
| 240 | WRITE_FLAG(vui.overscanInfoPresentFlag, "overscan_info_present_flag"); |
| 241 | if (vui.overscanInfoPresentFlag) |
| 242 | WRITE_FLAG(vui.overscanAppropriateFlag, "overscan_appropriate_flag"); |
| 243 | |
| 244 | WRITE_FLAG(vui.videoSignalTypePresentFlag, "video_signal_type_present_flag"); |
| 245 | if (vui.videoSignalTypePresentFlag) |
| 246 | { |
| 247 | WRITE_CODE(vui.videoFormat, 3, "video_format"); |
| 248 | WRITE_FLAG(vui.videoFullRangeFlag, "video_full_range_flag"); |
| 249 | WRITE_FLAG(vui.colourDescriptionPresentFlag, "colour_description_present_flag"); |
| 250 | if (vui.colourDescriptionPresentFlag) |
| 251 | { |
| 252 | WRITE_CODE(vui.colourPrimaries, 8, "colour_primaries"); |
| 253 | WRITE_CODE(vui.transferCharacteristics, 8, "transfer_characteristics"); |
| 254 | WRITE_CODE(vui.matrixCoefficients, 8, "matrix_coefficients"); |
| 255 | } |
| 256 | } |
| 257 | |
| 258 | WRITE_FLAG(vui.chromaLocInfoPresentFlag, "chroma_loc_info_present_flag"); |
| 259 | if (vui.chromaLocInfoPresentFlag) |
| 260 | { |
| 261 | WRITE_UVLC(vui.chromaSampleLocTypeTopField, "chroma_sample_loc_type_top_field"); |
| 262 | WRITE_UVLC(vui.chromaSampleLocTypeBottomField, "chroma_sample_loc_type_bottom_field"); |
| 263 | } |
| 264 | |
| 265 | WRITE_FLAG(0, "neutral_chroma_indication_flag"); |
| 266 | WRITE_FLAG(vui.fieldSeqFlag, "field_seq_flag"); |
| 267 | WRITE_FLAG(vui.frameFieldInfoPresentFlag, "frame_field_info_present_flag"); |
| 268 | |
| 269 | WRITE_FLAG(vui.defaultDisplayWindow.bEnabled, "default_display_window_flag"); |
| 270 | if (vui.defaultDisplayWindow.bEnabled) |
| 271 | { |
| 272 | WRITE_UVLC(vui.defaultDisplayWindow.leftOffset, "def_disp_win_left_offset"); |
| 273 | WRITE_UVLC(vui.defaultDisplayWindow.rightOffset, "def_disp_win_right_offset"); |
| 274 | WRITE_UVLC(vui.defaultDisplayWindow.topOffset, "def_disp_win_top_offset"); |
| 275 | WRITE_UVLC(vui.defaultDisplayWindow.bottomOffset, "def_disp_win_bottom_offset"); |
| 276 | } |
| 277 | |
| 278 | WRITE_FLAG(1, "vui_timing_info_present_flag"); |
| 279 | WRITE_CODE(vui.timingInfo.numUnitsInTick, 32, "vui_num_units_in_tick"); |
| 280 | WRITE_CODE(vui.timingInfo.timeScale, 32, "vui_time_scale"); |
| 281 | WRITE_FLAG(0, "vui_poc_proportional_to_timing_flag"); |
| 282 | |
| 283 | WRITE_FLAG(vui.hrdParametersPresentFlag, "vui_hrd_parameters_present_flag"); |
| 284 | if (vui.hrdParametersPresentFlag) |
| 285 | codeHrdParameters(vui.hrdParameters); |
| 286 | |
| 287 | WRITE_FLAG(0, "bitstream_restriction_flag"); |
| 288 | } |
| 289 | |
| 290 | void Entropy::codeScalingList(const ScalingList& scalingList) |
| 291 | { |
| 292 | for (int sizeId = 0; sizeId < ScalingList::NUM_SIZES; sizeId++) |
| 293 | { |
| 294 | for (int listId = 0; listId < ScalingList::NUM_LISTS; listId++) |
| 295 | { |
| 296 | int predList = scalingList.checkPredMode(sizeId, listId); |
| 297 | WRITE_FLAG(predList < 0, "scaling_list_pred_mode_flag"); |
| 298 | if (predList >= 0) |
| 299 | WRITE_UVLC(listId - predList, "scaling_list_pred_matrix_id_delta"); |
| 300 | else // DPCM Mode |
| 301 | codeScalingList(scalingList, sizeId, listId); |
| 302 | } |
| 303 | } |
| 304 | } |
| 305 | |
| 306 | void Entropy::codeScalingList(const ScalingList& scalingList, uint32_t sizeId, uint32_t listId) |
| 307 | { |
| 308 | int coefNum = X265_MIN(ScalingList::MAX_MATRIX_COEF_NUM, (int)ScalingList::s_numCoefPerSize[sizeId]); |
| 309 | const uint16_t* scan = (sizeId == 0 ? g_scan4x4[SCAN_DIAG] : g_scan8x8diag); |
| 310 | int nextCoef = START_VALUE; |
| 311 | int32_t *src = scalingList.m_scalingListCoef[sizeId][listId]; |
| 312 | int data; |
| 313 | |
| 314 | if (sizeId > BLOCK_8x8) |
| 315 | { |
| 316 | WRITE_SVLC(scalingList.m_scalingListDC[sizeId][listId] - 8, "scaling_list_dc_coef_minus8"); |
| 317 | nextCoef = scalingList.m_scalingListDC[sizeId][listId]; |
| 318 | } |
| 319 | for (int i = 0; i < coefNum; i++) |
| 320 | { |
| 321 | data = src[scan[i]] - nextCoef; |
| 322 | nextCoef = src[scan[i]]; |
| 323 | if (data > 127) |
| 324 | data = data - 256; |
| 325 | if (data < -128) |
| 326 | data = data + 256; |
| 327 | |
| 328 | WRITE_SVLC(data, "scaling_list_delta_coef"); |
| 329 | } |
| 330 | } |
| 331 | |
| 332 | void Entropy::codeHrdParameters(const HRDInfo& hrd) |
| 333 | { |
| 334 | WRITE_FLAG(1, "nal_hrd_parameters_present_flag"); |
| 335 | WRITE_FLAG(0, "vcl_hrd_parameters_present_flag"); |
| 336 | WRITE_FLAG(0, "sub_pic_hrd_params_present_flag"); |
| 337 | |
| 338 | WRITE_CODE(hrd.bitRateScale, 4, "bit_rate_scale"); |
| 339 | WRITE_CODE(hrd.cpbSizeScale, 4, "cpb_size_scale"); |
| 340 | |
| 341 | WRITE_CODE(hrd.initialCpbRemovalDelayLength - 1, 5, "initial_cpb_removal_delay_length_minus1"); |
| 342 | WRITE_CODE(hrd.cpbRemovalDelayLength - 1, 5, "au_cpb_removal_delay_length_minus1"); |
| 343 | WRITE_CODE(hrd.dpbOutputDelayLength - 1, 5, "dpb_output_delay_length_minus1"); |
| 344 | |
| 345 | WRITE_FLAG(1, "fixed_pic_rate_general_flag"); |
| 346 | WRITE_UVLC(0, "elemental_duration_in_tc_minus1"); |
| 347 | WRITE_UVLC(0, "cpb_cnt_minus1"); |
| 348 | |
| 349 | WRITE_UVLC(hrd.bitRateValue - 1, "bit_rate_value_minus1"); |
| 350 | WRITE_UVLC(hrd.cpbSizeValue - 1, "cpb_size_value_minus1"); |
| 351 | WRITE_FLAG(hrd.cbrFlag, "cbr_flag"); |
| 352 | } |
| 353 | |
| 354 | void Entropy::codeAUD(const Slice& slice) |
| 355 | { |
| 356 | int picType; |
| 357 | |
| 358 | switch (slice.m_sliceType) |
| 359 | { |
| 360 | case I_SLICE: |
| 361 | picType = 0; |
| 362 | break; |
| 363 | case P_SLICE: |
| 364 | picType = 1; |
| 365 | break; |
| 366 | case B_SLICE: |
| 367 | picType = 2; |
| 368 | break; |
| 369 | default: |
| 370 | picType = 7; |
| 371 | break; |
| 372 | } |
| 373 | |
| 374 | WRITE_CODE(picType, 3, "pic_type"); |
| 375 | } |
| 376 | |
| 377 | void Entropy::codeSliceHeader(const Slice& slice, FrameData& encData) |
| 378 | { |
| 379 | WRITE_FLAG(1, "first_slice_segment_in_pic_flag"); |
| 380 | if (slice.getRapPicFlag()) |
| 381 | WRITE_FLAG(0, "no_output_of_prior_pics_flag"); |
| 382 | |
| 383 | WRITE_UVLC(0, "slice_pic_parameter_set_id"); |
| 384 | |
| 385 | /* x265 does not use dependent slices, so always write all this data */ |
| 386 | |
| 387 | WRITE_UVLC(slice.m_sliceType, "slice_type"); |
| 388 | |
| 389 | if (!slice.getIdrPicFlag()) |
| 390 | { |
| 391 | int picOrderCntLSB = (slice.m_poc - slice.m_lastIDR + (1 << BITS_FOR_POC)) % (1 << BITS_FOR_POC); |
| 392 | WRITE_CODE(picOrderCntLSB, BITS_FOR_POC, "pic_order_cnt_lsb"); |
| 393 | |
| 394 | #if _DEBUG || CHECKED_BUILD |
| 395 | // check for bitstream restriction stating that: |
| 396 | // If the current picture is a BLA or CRA picture, the value of NumPocTotalCurr shall be equal to 0. |
| 397 | // Ideally this process should not be repeated for each slice in a picture |
| 398 | if (slice.isIRAP()) |
| 399 | for (int picIdx = 0; picIdx < slice.m_rps.numberOfPictures; picIdx++) |
| 400 | X265_CHECK(!slice.m_rps.bUsed[picIdx], "pic unused failure\n"); |
| 401 | #endif |
| 402 | |
| 403 | WRITE_FLAG(0, "short_term_ref_pic_set_sps_flag"); |
| 404 | codeShortTermRefPicSet(slice.m_rps); |
| 405 | |
| 406 | if (slice.m_sps->bTemporalMVPEnabled) |
| 407 | WRITE_FLAG(1, "slice_temporal_mvp_enable_flag"); |
| 408 | } |
| 409 | const SAOParam *saoParam = encData.m_saoParam; |
| 410 | if (slice.m_sps->bUseSAO) |
| 411 | { |
| 412 | WRITE_FLAG(saoParam->bSaoFlag[0], "slice_sao_luma_flag"); |
| 413 | WRITE_FLAG(saoParam->bSaoFlag[1], "slice_sao_chroma_flag"); |
| 414 | } |
| 415 | |
| 416 | // check if numRefIdx match the defaults (1, hard-coded in PPS). If not, override |
| 417 | // TODO: this might be a place to optimize a few bits per slice, by using param->refs for L0 default |
| 418 | |
| 419 | if (!slice.isIntra()) |
| 420 | { |
| 421 | bool overrideFlag = (slice.m_numRefIdx[0] != 1 || (slice.isInterB() && slice.m_numRefIdx[1] != 1)); |
| 422 | WRITE_FLAG(overrideFlag, "num_ref_idx_active_override_flag"); |
| 423 | if (overrideFlag) |
| 424 | { |
| 425 | WRITE_UVLC(slice.m_numRefIdx[0] - 1, "num_ref_idx_l0_active_minus1"); |
| 426 | if (slice.isInterB()) |
| 427 | WRITE_UVLC(slice.m_numRefIdx[1] - 1, "num_ref_idx_l1_active_minus1"); |
| 428 | else |
| 429 | { |
| 430 | X265_CHECK(slice.m_numRefIdx[1] == 0, "expected no L1 references for P slice\n"); |
| 431 | } |
| 432 | } |
| 433 | } |
| 434 | else |
| 435 | { |
| 436 | X265_CHECK(!slice.m_numRefIdx[0] && !slice.m_numRefIdx[1], "expected no references for I slice\n"); |
| 437 | } |
| 438 | |
| 439 | if (slice.isInterB()) |
| 440 | WRITE_FLAG(0, "mvd_l1_zero_flag"); |
| 441 | |
| 442 | if (slice.m_sps->bTemporalMVPEnabled) |
| 443 | { |
| 444 | if (slice.m_sliceType == B_SLICE) |
| 445 | WRITE_FLAG(slice.m_colFromL0Flag, "collocated_from_l0_flag"); |
| 446 | |
| 447 | if (slice.m_sliceType != I_SLICE && |
| 448 | ((slice.m_colFromL0Flag && slice.m_numRefIdx[0] > 1) || |
| 449 | (!slice.m_colFromL0Flag && slice.m_numRefIdx[1] > 1))) |
| 450 | { |
| 451 | WRITE_UVLC(slice.m_colRefIdx, "collocated_ref_idx"); |
| 452 | } |
| 453 | } |
| 454 | if ((slice.m_pps->bUseWeightPred && slice.m_sliceType == P_SLICE) || (slice.m_pps->bUseWeightedBiPred && slice.m_sliceType == B_SLICE)) |
| 455 | codePredWeightTable(slice); |
| 456 | |
| 457 | X265_CHECK(slice.m_maxNumMergeCand <= MRG_MAX_NUM_CANDS, "too many merge candidates\n"); |
| 458 | if (!slice.isIntra()) |
| 459 | WRITE_UVLC(MRG_MAX_NUM_CANDS - slice.m_maxNumMergeCand, "five_minus_max_num_merge_cand"); |
| 460 | |
| 461 | int code = slice.m_sliceQp - 26; |
| 462 | WRITE_SVLC(code, "slice_qp_delta"); |
| 463 | |
| 464 | bool isSAOEnabled = slice.m_sps->bUseSAO ? saoParam->bSaoFlag[0] || saoParam->bSaoFlag[1] : false; |
| 465 | bool isDBFEnabled = !slice.m_pps->bPicDisableDeblockingFilter; |
| 466 | |
| 467 | if (isSAOEnabled || isDBFEnabled) |
| 468 | WRITE_FLAG(slice.m_sLFaseFlag, "slice_loop_filter_across_slices_enabled_flag"); |
| 469 | } |
| 470 | |
| 471 | /** write wavefront substreams sizes for the slice header */ |
| 472 | void Entropy::codeSliceHeaderWPPEntryPoints(const Slice& slice, const uint32_t *substreamSizes, uint32_t maxOffset) |
| 473 | { |
| 474 | uint32_t offsetLen = 1; |
| 475 | while (maxOffset >= (1U << offsetLen)) |
| 476 | { |
| 477 | offsetLen++; |
| 478 | X265_CHECK(offsetLen < 32, "offsetLen is too large\n"); |
| 479 | } |
| 480 | |
| 481 | uint32_t numRows = slice.m_sps->numCuInHeight - 1; |
| 482 | WRITE_UVLC(numRows, "num_entry_point_offsets"); |
| 483 | if (numRows > 0) |
| 484 | WRITE_UVLC(offsetLen - 1, "offset_len_minus1"); |
| 485 | |
| 486 | for (uint32_t i = 0; i < numRows; i++) |
| 487 | WRITE_CODE(substreamSizes[i] - 1, offsetLen, "entry_point_offset_minus1"); |
| 488 | } |
| 489 | |
| 490 | void Entropy::codeShortTermRefPicSet(const RPS& rps) |
| 491 | { |
| 492 | WRITE_UVLC(rps.numberOfNegativePictures, "num_negative_pics"); |
| 493 | WRITE_UVLC(rps.numberOfPositivePictures, "num_positive_pics"); |
| 494 | int prev = 0; |
| 495 | for (int j = 0; j < rps.numberOfNegativePictures; j++) |
| 496 | { |
| 497 | WRITE_UVLC(prev - rps.deltaPOC[j] - 1, "delta_poc_s0_minus1"); |
| 498 | prev = rps.deltaPOC[j]; |
| 499 | WRITE_FLAG(rps.bUsed[j], "used_by_curr_pic_s0_flag"); |
| 500 | } |
| 501 | |
| 502 | prev = 0; |
| 503 | for (int j = rps.numberOfNegativePictures; j < rps.numberOfNegativePictures + rps.numberOfPositivePictures; j++) |
| 504 | { |
| 505 | WRITE_UVLC(rps.deltaPOC[j] - prev - 1, "delta_poc_s1_minus1"); |
| 506 | prev = rps.deltaPOC[j]; |
| 507 | WRITE_FLAG(rps.bUsed[j], "used_by_curr_pic_s1_flag"); |
| 508 | } |
| 509 | } |
| 510 | |
| 511 | void Entropy::encodeCTU(const CUData& ctu, const CUGeom& cuGeom) |
| 512 | { |
| 513 | bool bEncodeDQP = ctu.m_slice->m_pps->bUseDQP; |
| 514 | encodeCU(ctu, cuGeom, 0, 0, bEncodeDQP); |
| 515 | } |
| 516 | |
| 517 | /* encode a CU block recursively */ |
| 518 | void Entropy::encodeCU(const CUData& cu, const CUGeom& cuGeom, uint32_t absPartIdx, uint32_t depth, bool& bEncodeDQP) |
| 519 | { |
| 520 | const Slice* slice = cu.m_slice; |
| 521 | |
| 522 | if (depth <= slice->m_pps->maxCuDQPDepth && slice->m_pps->bUseDQP) |
| 523 | bEncodeDQP = true; |
| 524 | |
| 525 | int cuSplitFlag = !(cuGeom.flags & CUGeom::LEAF); |
| 526 | int cuUnsplitFlag = !(cuGeom.flags & CUGeom::SPLIT_MANDATORY); |
| 527 | |
| 528 | if (!cuUnsplitFlag) |
| 529 | { |
| 530 | uint32_t qNumParts = (NUM_CU_PARTITIONS >> (depth << 1)) >> 2; |
| 531 | for (uint32_t subPartIdx = 0; subPartIdx < 4; subPartIdx++, absPartIdx += qNumParts) |
| 532 | { |
| 533 | const CUGeom& childCuData = *(&cuGeom + cuGeom.childOffset + subPartIdx); |
| 534 | if (childCuData.flags & CUGeom::PRESENT) |
| 535 | encodeCU(cu, childCuData, absPartIdx, depth + 1, bEncodeDQP); |
| 536 | } |
| 537 | return; |
| 538 | } |
| 539 | |
| 540 | // We need to split, so don't try these modes. |
| 541 | if (cuSplitFlag) |
| 542 | codeSplitFlag(cu, absPartIdx, depth); |
| 543 | |
| 544 | if (depth < cu.m_cuDepth[absPartIdx] && depth < g_maxCUDepth) |
| 545 | { |
| 546 | uint32_t qNumParts = (NUM_CU_PARTITIONS >> (depth << 1)) >> 2; |
| 547 | |
| 548 | for (uint32_t subPartIdx = 0; subPartIdx < 4; subPartIdx++, absPartIdx += qNumParts) |
| 549 | { |
| 550 | const CUGeom& childCuData = *(&cuGeom + cuGeom.childOffset + subPartIdx); |
| 551 | encodeCU(cu, childCuData, absPartIdx, depth + 1, bEncodeDQP); |
| 552 | } |
| 553 | return; |
| 554 | } |
| 555 | |
| 556 | if (slice->m_pps->bTransquantBypassEnabled) |
| 557 | codeCUTransquantBypassFlag(cu.m_tqBypass[absPartIdx]); |
| 558 | |
| 559 | if (!slice->isIntra()) |
| 560 | codeSkipFlag(cu, absPartIdx); |
| 561 | |
| 562 | if (cu.isSkipped(absPartIdx)) |
| 563 | { |
| 564 | codeMergeIndex(cu, absPartIdx); |
| 565 | finishCU(cu, absPartIdx, depth); |
| 566 | return; |
| 567 | } |
| 568 | |
| 569 | if (!slice->isIntra()) |
| 570 | codePredMode(cu.m_predMode[absPartIdx]); |
| 571 | |
| 572 | codePartSize(cu, absPartIdx, depth); |
| 573 | |
| 574 | // prediction Info ( Intra : direction mode, Inter : Mv, reference idx ) |
| 575 | codePredInfo(cu, absPartIdx); |
| 576 | |
| 577 | uint32_t tuDepthRange[2]; |
| 578 | if (cu.isIntra(absPartIdx)) |
| 579 | cu.getIntraTUQtDepthRange(tuDepthRange, absPartIdx); |
| 580 | else |
| 581 | cu.getInterTUQtDepthRange(tuDepthRange, absPartIdx); |
| 582 | |
| 583 | // Encode Coefficients, allow codeCoeff() to modify bEncodeDQP |
| 584 | codeCoeff(cu, absPartIdx, depth, bEncodeDQP, tuDepthRange); |
| 585 | |
| 586 | // --- write terminating bit --- |
| 587 | finishCU(cu, absPartIdx, depth); |
| 588 | } |
| 589 | |
| 590 | /* finish encoding a cu and handle end-of-slice conditions */ |
| 591 | void Entropy::finishCU(const CUData& cu, uint32_t absPartIdx, uint32_t depth) |
| 592 | { |
| 593 | const Slice* slice = cu.m_slice; |
| 594 | X265_CHECK(cu.m_slice->m_endCUAddr == cu.m_slice->realEndAddress(slice->m_endCUAddr), "real end address expected\n"); |
| 595 | uint32_t realEndAddress = slice->m_endCUAddr; |
| 596 | uint32_t cuAddr = cu.getSCUAddr() + absPartIdx; |
| 597 | |
| 598 | uint32_t granularityMask = g_maxCUSize - 1; |
| 599 | uint32_t cuSize = 1 << cu.m_log2CUSize[absPartIdx]; |
| 600 | uint32_t rpelx = cu.m_cuPelX + g_zscanToPelX[absPartIdx] + cuSize; |
| 601 | uint32_t bpely = cu.m_cuPelY + g_zscanToPelY[absPartIdx] + cuSize; |
| 602 | bool granularityBoundary = (((rpelx & granularityMask) == 0 || (rpelx == slice->m_sps->picWidthInLumaSamples )) && |
| 603 | ((bpely & granularityMask) == 0 || (bpely == slice->m_sps->picHeightInLumaSamples))); |
| 604 | |
| 605 | if (granularityBoundary) |
| 606 | { |
| 607 | // Encode slice finish |
| 608 | bool bTerminateSlice = false; |
| 609 | if (cuAddr + (NUM_CU_PARTITIONS >> (depth << 1)) == realEndAddress) |
| 610 | bTerminateSlice = true; |
| 611 | |
| 612 | // The 1-terminating bit is added to all streams, so don't add it here when it's 1. |
| 613 | if (!bTerminateSlice) |
| 614 | encodeBinTrm(0); |
| 615 | |
| 616 | if (!m_bitIf) |
| 617 | resetBits(); // TODO: most likely unnecessary |
| 618 | } |
| 619 | } |
| 620 | |
| 621 | void Entropy::encodeTransform(const CUData& cu, CoeffCodeState& state, uint32_t offsetLuma, uint32_t offsetChroma, uint32_t absPartIdx, |
| 622 | uint32_t absPartIdxStep, uint32_t depth, uint32_t log2TrSize, uint32_t trIdx, bool& bCodeDQP, uint32_t depthRange[2]) |
| 623 | { |
| 624 | const bool subdiv = cu.m_tuDepth[absPartIdx] + cu.m_cuDepth[absPartIdx] > (uint8_t)depth; |
| 625 | uint32_t hChromaShift = cu.m_hChromaShift; |
| 626 | uint32_t vChromaShift = cu.m_vChromaShift; |
| 627 | uint32_t cbfY = cu.getCbf(absPartIdx, TEXT_LUMA, trIdx); |
| 628 | uint32_t cbfU = cu.getCbf(absPartIdx, TEXT_CHROMA_U, trIdx); |
| 629 | uint32_t cbfV = cu.getCbf(absPartIdx, TEXT_CHROMA_V, trIdx); |
| 630 | |
| 631 | if (!trIdx) |
| 632 | state.bakAbsPartIdxCU = absPartIdx; |
| 633 | |
| 634 | if (log2TrSize == 2 && cu.m_chromaFormat != X265_CSP_I444) |
| 635 | { |
| 636 | uint32_t partNum = NUM_CU_PARTITIONS >> ((depth - 1) << 1); |
| 637 | if (!(absPartIdx & (partNum - 1))) |
| 638 | { |
| 639 | state.bakAbsPartIdx = absPartIdx; |
| 640 | state.bakChromaOffset = offsetChroma; |
| 641 | } |
| 642 | else if ((absPartIdx & (partNum - 1)) == (partNum - 1)) |
| 643 | { |
| 644 | cbfU = cu.getCbf(state.bakAbsPartIdx, TEXT_CHROMA_U, trIdx); |
| 645 | cbfV = cu.getCbf(state.bakAbsPartIdx, TEXT_CHROMA_V, trIdx); |
| 646 | } |
| 647 | } |
| 648 | |
| 649 | /* in each of these conditions, the subdiv flag is implied and not signaled, |
| 650 | * so we have checks to make sure the implied value matches our intentions */ |
| 651 | if (cu.m_predMode[absPartIdx] == MODE_INTRA && cu.m_partSize[absPartIdx] == SIZE_NxN && depth == cu.m_cuDepth[absPartIdx]) |
| 652 | { |
| 653 | X265_CHECK(subdiv, "intra NxN requires TU depth below CU depth\n"); |
| 654 | } |
| 655 | else if (cu.m_predMode[absPartIdx] == MODE_INTER && (cu.m_partSize[absPartIdx] != SIZE_2Nx2N) && depth == cu.m_cuDepth[absPartIdx] && |
| 656 | cu.m_slice->m_sps->quadtreeTUMaxDepthInter == 1) |
| 657 | { |
| 658 | X265_CHECK(subdiv, "inter TU must be smaller than CU when not 2Nx2N part size: log2TrSize %d, depthRange[0] %d\n", log2TrSize, depthRange[0]); |
| 659 | } |
| 660 | else if (log2TrSize > depthRange[1]) |
| 661 | { |
| 662 | X265_CHECK(subdiv, "TU is larger than the max allowed, it should have been split\n"); |
| 663 | } |
| 664 | else if (log2TrSize == cu.m_slice->m_sps->quadtreeTULog2MinSize || log2TrSize == depthRange[0]) |
| 665 | { |
| 666 | X265_CHECK(!subdiv, "min sized TU cannot be subdivided\n"); |
| 667 | } |
| 668 | else |
| 669 | { |
| 670 | X265_CHECK(log2TrSize > depthRange[0], "transform size failure\n"); |
| 671 | codeTransformSubdivFlag(subdiv, 5 - log2TrSize); |
| 672 | } |
| 673 | |
| 674 | const uint32_t trDepthCurr = depth - cu.m_cuDepth[absPartIdx]; |
| 675 | const bool bFirstCbfOfCU = trDepthCurr == 0; |
| 676 | |
| 677 | bool mCodeAll = true; |
| 678 | const uint32_t numPels = 1 << (log2TrSize * 2 - hChromaShift - vChromaShift); |
| 679 | if (numPels < (MIN_TU_SIZE * MIN_TU_SIZE)) |
| 680 | mCodeAll = false; |
| 681 | |
| 682 | if (bFirstCbfOfCU || mCodeAll) |
| 683 | { |
| 684 | uint32_t tuSize = 1 << log2TrSize; |
| 685 | if (bFirstCbfOfCU || cu.getCbf(absPartIdx, TEXT_CHROMA_U, trDepthCurr - 1)) |
| 686 | codeQtCbf(cu, absPartIdx, absPartIdxStep, (tuSize >> hChromaShift), (tuSize >> vChromaShift), TEXT_CHROMA_U, trDepthCurr, (subdiv == 0)); |
| 687 | if (bFirstCbfOfCU || cu.getCbf(absPartIdx, TEXT_CHROMA_V, trDepthCurr - 1)) |
| 688 | codeQtCbf(cu, absPartIdx, absPartIdxStep, (tuSize >> hChromaShift), (tuSize >> vChromaShift), TEXT_CHROMA_V, trDepthCurr, (subdiv == 0)); |
| 689 | } |
| 690 | else |
| 691 | { |
| 692 | X265_CHECK(cu.getCbf(absPartIdx, TEXT_CHROMA_U, trDepthCurr) == cu.getCbf(absPartIdx, TEXT_CHROMA_U, trDepthCurr - 1), "chroma xform size match failure\n"); |
| 693 | X265_CHECK(cu.getCbf(absPartIdx, TEXT_CHROMA_V, trDepthCurr) == cu.getCbf(absPartIdx, TEXT_CHROMA_V, trDepthCurr - 1), "chroma xform size match failure\n"); |
| 694 | } |
| 695 | |
| 696 | if (subdiv) |
| 697 | { |
| 698 | log2TrSize--; |
| 699 | uint32_t numCoeff = 1 << (log2TrSize * 2); |
| 700 | uint32_t numCoeffC = (numCoeff >> (hChromaShift + vChromaShift)); |
| 701 | trIdx++; |
| 702 | ++depth; |
| 703 | absPartIdxStep >>= 2; |
| 704 | const uint32_t partNum = NUM_CU_PARTITIONS >> (depth << 1); |
| 705 | |
| 706 | encodeTransform(cu, state, offsetLuma, offsetChroma, absPartIdx, absPartIdxStep, depth, log2TrSize, trIdx, bCodeDQP, depthRange); |
| 707 | |
| 708 | absPartIdx += partNum; |
| 709 | offsetLuma += numCoeff; |
| 710 | offsetChroma += numCoeffC; |
| 711 | encodeTransform(cu, state, offsetLuma, offsetChroma, absPartIdx, absPartIdxStep, depth, log2TrSize, trIdx, bCodeDQP, depthRange); |
| 712 | |
| 713 | absPartIdx += partNum; |
| 714 | offsetLuma += numCoeff; |
| 715 | offsetChroma += numCoeffC; |
| 716 | encodeTransform(cu, state, offsetLuma, offsetChroma, absPartIdx, absPartIdxStep, depth, log2TrSize, trIdx, bCodeDQP, depthRange); |
| 717 | |
| 718 | absPartIdx += partNum; |
| 719 | offsetLuma += numCoeff; |
| 720 | offsetChroma += numCoeffC; |
| 721 | encodeTransform(cu, state, offsetLuma, offsetChroma, absPartIdx, absPartIdxStep, depth, log2TrSize, trIdx, bCodeDQP, depthRange); |
| 722 | } |
| 723 | else |
| 724 | { |
| 725 | if (cu.m_predMode[absPartIdx] != MODE_INTRA && depth == cu.m_cuDepth[absPartIdx] && !cu.getCbf(absPartIdx, TEXT_CHROMA_U, 0) && !cu.getCbf(absPartIdx, TEXT_CHROMA_V, 0)) |
| 726 | { |
| 727 | X265_CHECK(cu.getCbf(absPartIdx, TEXT_LUMA, 0), "CBF should have been set\n"); |
| 728 | } |
| 729 | else |
| 730 | codeQtCbf(cu, absPartIdx, TEXT_LUMA, cu.m_tuDepth[absPartIdx]); |
| 731 | |
| 732 | if (cbfY || cbfU || cbfV) |
| 733 | { |
| 734 | // dQP: only for CTU once |
| 735 | if (cu.m_slice->m_pps->bUseDQP) |
| 736 | { |
| 737 | if (bCodeDQP) |
| 738 | { |
| 739 | codeDeltaQP(cu, state.bakAbsPartIdxCU); |
| 740 | bCodeDQP = false; |
| 741 | } |
| 742 | } |
| 743 | } |
| 744 | if (cbfY) |
| 745 | codeCoeffNxN(cu, cu.m_trCoeff[0] + offsetLuma, absPartIdx, log2TrSize, TEXT_LUMA); |
| 746 | |
| 747 | int chFmt = cu.m_chromaFormat; |
| 748 | if (log2TrSize == 2 && chFmt != X265_CSP_I444) |
| 749 | { |
| 750 | uint32_t partNum = NUM_CU_PARTITIONS >> ((depth - 1) << 1); |
| 751 | if ((absPartIdx & (partNum - 1)) == (partNum - 1)) |
| 752 | { |
| 753 | const uint32_t log2TrSizeC = 2; |
| 754 | const bool splitIntoSubTUs = (chFmt == X265_CSP_I422); |
| 755 | |
| 756 | uint32_t curPartNum = NUM_CU_PARTITIONS >> ((depth - 1) << 1); |
| 757 | |
| 758 | for (uint32_t chromaId = TEXT_CHROMA_U; chromaId <= TEXT_CHROMA_V; chromaId++) |
| 759 | { |
| 760 | TURecurse tuIterator(splitIntoSubTUs ? VERTICAL_SPLIT : DONT_SPLIT, curPartNum, state.bakAbsPartIdx); |
| 761 | const coeff_t* coeffChroma = cu.m_trCoeff[chromaId]; |
| 762 | do |
| 763 | { |
| 764 | uint32_t cbf = cu.getCbf(tuIterator.absPartIdxTURelCU, (TextType)chromaId, trIdx + splitIntoSubTUs); |
| 765 | if (cbf) |
| 766 | { |
| 767 | uint32_t subTUOffset = tuIterator.section << (log2TrSizeC * 2); |
| 768 | codeCoeffNxN(cu, coeffChroma + state.bakChromaOffset + subTUOffset, tuIterator.absPartIdxTURelCU, log2TrSizeC, (TextType)chromaId); |
| 769 | } |
| 770 | } |
| 771 | while (tuIterator.isNextSection()); |
| 772 | } |
| 773 | } |
| 774 | } |
| 775 | else |
| 776 | { |
| 777 | uint32_t log2TrSizeC = log2TrSize - hChromaShift; |
| 778 | const bool splitIntoSubTUs = (chFmt == X265_CSP_I422); |
| 779 | uint32_t curPartNum = NUM_CU_PARTITIONS >> (depth << 1); |
| 780 | for (uint32_t chromaId = TEXT_CHROMA_U; chromaId <= TEXT_CHROMA_V; chromaId++) |
| 781 | { |
| 782 | TURecurse tuIterator(splitIntoSubTUs ? VERTICAL_SPLIT : DONT_SPLIT, curPartNum, absPartIdx); |
| 783 | const coeff_t* coeffChroma = cu.m_trCoeff[chromaId]; |
| 784 | do |
| 785 | { |
| 786 | uint32_t cbf = cu.getCbf(tuIterator.absPartIdxTURelCU, (TextType)chromaId, trIdx + splitIntoSubTUs); |
| 787 | if (cbf) |
| 788 | { |
| 789 | uint32_t subTUOffset = tuIterator.section << (log2TrSizeC * 2); |
| 790 | codeCoeffNxN(cu, coeffChroma + offsetChroma + subTUOffset, tuIterator.absPartIdxTURelCU, log2TrSizeC, (TextType)chromaId); |
| 791 | } |
| 792 | } |
| 793 | while (tuIterator.isNextSection()); |
| 794 | } |
| 795 | } |
| 796 | } |
| 797 | } |
| 798 | |
| 799 | void Entropy::codePredInfo(const CUData& cu, uint32_t absPartIdx) |
| 800 | { |
| 801 | if (cu.isIntra(absPartIdx)) // If it is intra mode, encode intra prediction mode. |
| 802 | { |
| 803 | codeIntraDirLumaAng(cu, absPartIdx, true); |
| 804 | if (cu.m_chromaFormat != X265_CSP_I400) |
| 805 | { |
| 806 | uint32_t chromaDirMode[NUM_CHROMA_MODE]; |
| 807 | cu.getAllowedChromaDir(absPartIdx, chromaDirMode); |
| 808 | |
| 809 | codeIntraDirChroma(cu, absPartIdx, chromaDirMode); |
| 810 | |
| 811 | if ((cu.m_chromaFormat == X265_CSP_I444) && (cu.m_partSize[absPartIdx] == SIZE_NxN)) |
| 812 | { |
| 813 | uint32_t partOffset = (NUM_CU_PARTITIONS >> (cu.m_cuDepth[absPartIdx] << 1)) >> 2; |
| 814 | for (uint32_t i = 1; i <= 3; i++) |
| 815 | { |
| 816 | uint32_t offset = absPartIdx + i * partOffset; |
| 817 | cu.getAllowedChromaDir(offset, chromaDirMode); |
| 818 | codeIntraDirChroma(cu, offset, chromaDirMode); |
| 819 | } |
| 820 | } |
| 821 | } |
| 822 | } |
| 823 | else // if it is inter mode, encode motion vector and reference index |
| 824 | codePUWise(cu, absPartIdx); |
| 825 | } |
| 826 | |
| 827 | /** encode motion information for every PU block */ |
| 828 | void Entropy::codePUWise(const CUData& cu, uint32_t absPartIdx) |
| 829 | { |
| 830 | PartSize partSize = (PartSize)cu.m_partSize[absPartIdx]; |
| 831 | uint32_t numPU = (partSize == SIZE_2Nx2N ? 1 : (partSize == SIZE_NxN ? 4 : 2)); |
| 832 | uint32_t depth = cu.m_cuDepth[absPartIdx]; |
| 833 | uint32_t puOffset = (g_puOffset[uint32_t(partSize)] << (g_maxFullDepth - depth) * 2) >> 4; |
| 834 | |
| 835 | for (uint32_t puIdx = 0, subPartIdx = absPartIdx; puIdx < numPU; puIdx++, subPartIdx += puOffset) |
| 836 | { |
| 837 | codeMergeFlag(cu, subPartIdx); |
| 838 | if (cu.m_mergeFlag[subPartIdx]) |
| 839 | codeMergeIndex(cu, subPartIdx); |
| 840 | else |
| 841 | { |
| 842 | if (cu.m_slice->isInterB()) |
| 843 | codeInterDir(cu, subPartIdx); |
| 844 | |
| 845 | uint32_t interDir = cu.m_interDir[subPartIdx]; |
| 846 | for (uint32_t list = 0; list < 2; list++) |
| 847 | { |
| 848 | if (interDir & (1 << list)) |
| 849 | { |
| 850 | X265_CHECK(cu.m_slice->m_numRefIdx[list] > 0, "numRefs should have been > 0\n"); |
| 851 | |
| 852 | codeRefFrmIdxPU(cu, subPartIdx, list); |
| 853 | codeMvd(cu, subPartIdx, list); |
| 854 | codeMVPIdx(cu.m_mvpIdx[list][subPartIdx]); |
| 855 | } |
| 856 | } |
| 857 | } |
| 858 | } |
| 859 | } |
| 860 | |
| 861 | /** encode reference frame index for a PU block */ |
| 862 | void Entropy::codeRefFrmIdxPU(const CUData& cu, uint32_t absPartIdx, int list) |
| 863 | { |
| 864 | X265_CHECK(!cu.isIntra(absPartIdx), "intra block not expected\n"); |
| 865 | |
| 866 | if (cu.m_slice->m_numRefIdx[list] > 1) |
| 867 | codeRefFrmIdx(cu, absPartIdx, list); |
| 868 | } |
| 869 | |
| 870 | void Entropy::codeCoeff(const CUData& cu, uint32_t absPartIdx, uint32_t depth, bool& bCodeDQP, uint32_t depthRange[2]) |
| 871 | { |
| 872 | if (!cu.isIntra(absPartIdx)) |
| 873 | { |
| 874 | if (!(cu.m_mergeFlag[absPartIdx] && cu.m_partSize[absPartIdx] == SIZE_2Nx2N)) |
| 875 | codeQtRootCbf(cu.getQtRootCbf(absPartIdx)); |
| 876 | if (!cu.getQtRootCbf(absPartIdx)) |
| 877 | return; |
| 878 | } |
| 879 | |
| 880 | uint32_t log2CUSize = cu.m_log2CUSize[absPartIdx]; |
| 881 | uint32_t lumaOffset = absPartIdx << (LOG2_UNIT_SIZE * 2); |
| 882 | uint32_t chromaOffset = lumaOffset >> (cu.m_hChromaShift + cu.m_vChromaShift); |
| 883 | uint32_t absPartIdxStep = NUM_CU_PARTITIONS >> (depth << 1); |
| 884 | CoeffCodeState state; |
| 885 | encodeTransform(cu, state, lumaOffset, chromaOffset, absPartIdx, absPartIdxStep, depth, log2CUSize, 0, bCodeDQP, depthRange); |
| 886 | } |
| 887 | |
| 888 | void Entropy::codeSaoOffset(const SaoCtuParam& ctuParam, int plane) |
| 889 | { |
| 890 | int typeIdx = ctuParam.typeIdx; |
| 891 | |
| 892 | if (plane != 2) |
| 893 | { |
| 894 | encodeBin(typeIdx >= 0, m_contextState[OFF_SAO_TYPE_IDX_CTX]); |
| 895 | if (typeIdx >= 0) |
| 896 | encodeBinEP(typeIdx < SAO_BO ? 1 : 0); |
| 897 | } |
| 898 | |
| 899 | if (typeIdx >= 0) |
| 900 | { |
| 901 | enum { OFFSET_THRESH = 1 << X265_MIN(X265_DEPTH - 5, 5) }; |
| 902 | if (typeIdx == SAO_BO) |
| 903 | { |
| 904 | for (int i = 0; i < SAO_BO_LEN; i++) |
| 905 | codeSaoMaxUvlc(abs(ctuParam.offset[i]), OFFSET_THRESH - 1); |
| 906 | |
| 907 | for (int i = 0; i < SAO_BO_LEN; i++) |
| 908 | if (ctuParam.offset[i] != 0) |
| 909 | encodeBinEP(ctuParam.offset[i] < 0); |
| 910 | |
| 911 | encodeBinsEP(ctuParam.bandPos, 5); |
| 912 | } |
| 913 | else // if (typeIdx < SAO_BO) |
| 914 | { |
| 915 | codeSaoMaxUvlc(ctuParam.offset[0], OFFSET_THRESH - 1); |
| 916 | codeSaoMaxUvlc(ctuParam.offset[1], OFFSET_THRESH - 1); |
| 917 | codeSaoMaxUvlc(-ctuParam.offset[2], OFFSET_THRESH - 1); |
| 918 | codeSaoMaxUvlc(-ctuParam.offset[3], OFFSET_THRESH - 1); |
| 919 | if (plane != 2) |
| 920 | encodeBinsEP((uint32_t)(typeIdx), 2); |
| 921 | } |
| 922 | } |
| 923 | } |
| 924 | |
| 925 | /** initialize context model with respect to QP and initialization value */ |
| 926 | uint8_t sbacInit(int qp, int initValue) |
| 927 | { |
| 928 | qp = Clip3(0, 51, qp); |
| 929 | |
| 930 | int slope = (initValue >> 4) * 5 - 45; |
| 931 | int offset = ((initValue & 15) << 3) - 16; |
| 932 | int initState = X265_MIN(X265_MAX(1, (((slope * qp) >> 4) + offset)), 126); |
| 933 | uint32_t mpState = (initState >= 64); |
| 934 | uint32_t state = ((mpState ? (initState - 64) : (63 - initState)) << 1) + mpState; |
| 935 | |
| 936 | return (uint8_t)state; |
| 937 | } |
| 938 | |
| 939 | static void initBuffer(uint8_t* contextModel, SliceType sliceType, int qp, uint8_t* ctxModel, int size) |
| 940 | { |
| 941 | ctxModel += sliceType * size; |
| 942 | |
| 943 | for (int n = 0; n < size; n++) |
| 944 | contextModel[n] = sbacInit(qp, ctxModel[n]); |
| 945 | } |
| 946 | |
| 947 | void Entropy::resetEntropy(const Slice& slice) |
| 948 | { |
| 949 | int qp = slice.m_sliceQp; |
| 950 | SliceType sliceType = slice.m_sliceType; |
| 951 | |
| 952 | initBuffer(&m_contextState[OFF_SPLIT_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_SPLIT_FLAG, NUM_SPLIT_FLAG_CTX); |
| 953 | initBuffer(&m_contextState[OFF_SKIP_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_SKIP_FLAG, NUM_SKIP_FLAG_CTX); |
| 954 | initBuffer(&m_contextState[OFF_MERGE_FLAG_EXT_CTX], sliceType, qp, (uint8_t*)INIT_MERGE_FLAG_EXT, NUM_MERGE_FLAG_EXT_CTX); |
| 955 | initBuffer(&m_contextState[OFF_MERGE_IDX_EXT_CTX], sliceType, qp, (uint8_t*)INIT_MERGE_IDX_EXT, NUM_MERGE_IDX_EXT_CTX); |
| 956 | initBuffer(&m_contextState[OFF_PART_SIZE_CTX], sliceType, qp, (uint8_t*)INIT_PART_SIZE, NUM_PART_SIZE_CTX); |
| 957 | initBuffer(&m_contextState[OFF_PRED_MODE_CTX], sliceType, qp, (uint8_t*)INIT_PRED_MODE, NUM_PRED_MODE_CTX); |
| 958 | initBuffer(&m_contextState[OFF_ADI_CTX], sliceType, qp, (uint8_t*)INIT_INTRA_PRED_MODE, NUM_ADI_CTX); |
| 959 | initBuffer(&m_contextState[OFF_CHROMA_PRED_CTX], sliceType, qp, (uint8_t*)INIT_CHROMA_PRED_MODE, NUM_CHROMA_PRED_CTX); |
| 960 | initBuffer(&m_contextState[OFF_DELTA_QP_CTX], sliceType, qp, (uint8_t*)INIT_DQP, NUM_DELTA_QP_CTX); |
| 961 | initBuffer(&m_contextState[OFF_INTER_DIR_CTX], sliceType, qp, (uint8_t*)INIT_INTER_DIR, NUM_INTER_DIR_CTX); |
| 962 | initBuffer(&m_contextState[OFF_REF_NO_CTX], sliceType, qp, (uint8_t*)INIT_REF_PIC, NUM_REF_NO_CTX); |
| 963 | initBuffer(&m_contextState[OFF_MV_RES_CTX], sliceType, qp, (uint8_t*)INIT_MVD, NUM_MV_RES_CTX); |
| 964 | initBuffer(&m_contextState[OFF_QT_CBF_CTX], sliceType, qp, (uint8_t*)INIT_QT_CBF, NUM_QT_CBF_CTX); |
| 965 | initBuffer(&m_contextState[OFF_TRANS_SUBDIV_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_TRANS_SUBDIV_FLAG, NUM_TRANS_SUBDIV_FLAG_CTX); |
| 966 | initBuffer(&m_contextState[OFF_QT_ROOT_CBF_CTX], sliceType, qp, (uint8_t*)INIT_QT_ROOT_CBF, NUM_QT_ROOT_CBF_CTX); |
| 967 | initBuffer(&m_contextState[OFF_SIG_CG_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_SIG_CG_FLAG, 2 * NUM_SIG_CG_FLAG_CTX); |
| 968 | initBuffer(&m_contextState[OFF_SIG_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_SIG_FLAG, NUM_SIG_FLAG_CTX); |
| 969 | initBuffer(&m_contextState[OFF_CTX_LAST_FLAG_X], sliceType, qp, (uint8_t*)INIT_LAST, NUM_CTX_LAST_FLAG_XY); |
| 970 | initBuffer(&m_contextState[OFF_CTX_LAST_FLAG_Y], sliceType, qp, (uint8_t*)INIT_LAST, NUM_CTX_LAST_FLAG_XY); |
| 971 | initBuffer(&m_contextState[OFF_ONE_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_ONE_FLAG, NUM_ONE_FLAG_CTX); |
| 972 | initBuffer(&m_contextState[OFF_ABS_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_ABS_FLAG, NUM_ABS_FLAG_CTX); |
| 973 | initBuffer(&m_contextState[OFF_MVP_IDX_CTX], sliceType, qp, (uint8_t*)INIT_MVP_IDX, NUM_MVP_IDX_CTX); |
| 974 | initBuffer(&m_contextState[OFF_SAO_MERGE_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_SAO_MERGE_FLAG, NUM_SAO_MERGE_FLAG_CTX); |
| 975 | initBuffer(&m_contextState[OFF_SAO_TYPE_IDX_CTX], sliceType, qp, (uint8_t*)INIT_SAO_TYPE_IDX, NUM_SAO_TYPE_IDX_CTX); |
| 976 | initBuffer(&m_contextState[OFF_TRANSFORMSKIP_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_TRANSFORMSKIP_FLAG, 2 * NUM_TRANSFORMSKIP_FLAG_CTX); |
| 977 | initBuffer(&m_contextState[OFF_TQUANT_BYPASS_FLAG_CTX], sliceType, qp, (uint8_t*)INIT_CU_TRANSQUANT_BYPASS_FLAG, NUM_TQUANT_BYPASS_FLAG_CTX); |
| 978 | // new structure |
| 979 | |
| 980 | start(); |
| 981 | } |
| 982 | |
| 983 | /* code explicit wp tables */ |
| 984 | void Entropy::codePredWeightTable(const Slice& slice) |
| 985 | { |
| 986 | const WeightParam *wp; |
| 987 | bool bChroma = true; // 4:0:0 not yet supported |
| 988 | bool bDenomCoded = false; |
| 989 | int numRefDirs = slice.m_sliceType == B_SLICE ? 2 : 1; |
| 990 | uint32_t totalSignalledWeightFlags = 0; |
| 991 | |
| 992 | if ((slice.m_sliceType == P_SLICE && slice.m_pps->bUseWeightPred) || |
| 993 | (slice.m_sliceType == B_SLICE && slice.m_pps->bUseWeightedBiPred)) |
| 994 | { |
| 995 | for (int list = 0; list < numRefDirs; list++) |
| 996 | { |
| 997 | for (int ref = 0; ref < slice.m_numRefIdx[list]; ref++) |
| 998 | { |
| 999 | wp = slice.m_weightPredTable[list][ref]; |
| 1000 | if (!bDenomCoded) |
| 1001 | { |
| 1002 | WRITE_UVLC(wp[0].log2WeightDenom, "luma_log2_weight_denom"); |
| 1003 | |
| 1004 | if (bChroma) |
| 1005 | { |
| 1006 | int deltaDenom = wp[1].log2WeightDenom - wp[0].log2WeightDenom; |
| 1007 | WRITE_SVLC(deltaDenom, "delta_chroma_log2_weight_denom"); |
| 1008 | } |
| 1009 | bDenomCoded = true; |
| 1010 | } |
| 1011 | WRITE_FLAG(wp[0].bPresentFlag, "luma_weight_lX_flag"); |
| 1012 | totalSignalledWeightFlags += wp[0].bPresentFlag; |
| 1013 | } |
| 1014 | |
| 1015 | if (bChroma) |
| 1016 | { |
| 1017 | for (int ref = 0; ref < slice.m_numRefIdx[list]; ref++) |
| 1018 | { |
| 1019 | wp = slice.m_weightPredTable[list][ref]; |
| 1020 | WRITE_FLAG(wp[1].bPresentFlag, "chroma_weight_lX_flag"); |
| 1021 | totalSignalledWeightFlags += 2 * wp[1].bPresentFlag; |
| 1022 | } |
| 1023 | } |
| 1024 | |
| 1025 | for (int ref = 0; ref < slice.m_numRefIdx[list]; ref++) |
| 1026 | { |
| 1027 | wp = slice.m_weightPredTable[list][ref]; |
| 1028 | if (wp[0].bPresentFlag) |
| 1029 | { |
| 1030 | int deltaWeight = (wp[0].inputWeight - (1 << wp[0].log2WeightDenom)); |
| 1031 | WRITE_SVLC(deltaWeight, "delta_luma_weight_lX"); |
| 1032 | WRITE_SVLC(wp[0].inputOffset, "luma_offset_lX"); |
| 1033 | } |
| 1034 | |
| 1035 | if (bChroma) |
| 1036 | { |
| 1037 | if (wp[1].bPresentFlag) |
| 1038 | { |
| 1039 | for (int plane = 1; plane < 3; plane++) |
| 1040 | { |
| 1041 | int deltaWeight = (wp[plane].inputWeight - (1 << wp[1].log2WeightDenom)); |
| 1042 | WRITE_SVLC(deltaWeight, "delta_chroma_weight_lX"); |
| 1043 | |
| 1044 | int pred = (128 - ((128 * wp[plane].inputWeight) >> (wp[plane].log2WeightDenom))); |
| 1045 | int deltaChroma = (wp[plane].inputOffset - pred); |
| 1046 | WRITE_SVLC(deltaChroma, "delta_chroma_offset_lX"); |
| 1047 | } |
| 1048 | } |
| 1049 | } |
| 1050 | } |
| 1051 | } |
| 1052 | |
| 1053 | X265_CHECK(totalSignalledWeightFlags <= 24, "total weights must be <= 24\n"); |
| 1054 | } |
| 1055 | } |
| 1056 | |
| 1057 | void Entropy::writeUnaryMaxSymbol(uint32_t symbol, uint8_t* scmModel, int offset, uint32_t maxSymbol) |
| 1058 | { |
| 1059 | X265_CHECK(maxSymbol > 0, "maxSymbol too small\n"); |
| 1060 | |
| 1061 | encodeBin(symbol ? 1 : 0, scmModel[0]); |
| 1062 | |
| 1063 | if (!symbol) |
| 1064 | return; |
| 1065 | |
| 1066 | bool bCodeLast = (maxSymbol > symbol); |
| 1067 | |
| 1068 | while (--symbol) |
| 1069 | encodeBin(1, scmModel[offset]); |
| 1070 | |
| 1071 | if (bCodeLast) |
| 1072 | encodeBin(0, scmModel[offset]); |
| 1073 | } |
| 1074 | |
| 1075 | void Entropy::writeEpExGolomb(uint32_t symbol, uint32_t count) |
| 1076 | { |
| 1077 | uint32_t bins = 0; |
| 1078 | int numBins = 0; |
| 1079 | |
| 1080 | while (symbol >= (uint32_t)(1 << count)) |
| 1081 | { |
| 1082 | bins = 2 * bins + 1; |
| 1083 | numBins++; |
| 1084 | symbol -= 1 << count; |
| 1085 | count++; |
| 1086 | } |
| 1087 | |
| 1088 | bins = 2 * bins + 0; |
| 1089 | numBins++; |
| 1090 | |
| 1091 | bins = (bins << count) | symbol; |
| 1092 | numBins += count; |
| 1093 | |
| 1094 | X265_CHECK(numBins <= 32, "numBins too large\n"); |
| 1095 | encodeBinsEP(bins, numBins); |
| 1096 | } |
| 1097 | |
| 1098 | /** Coding of coeff_abs_level_minus3 */ |
| 1099 | void Entropy::writeCoefRemainExGolomb(uint32_t codeNumber, uint32_t absGoRice) |
| 1100 | { |
| 1101 | uint32_t length; |
| 1102 | const uint32_t codeRemain = codeNumber & ((1 << absGoRice) - 1); |
| 1103 | |
| 1104 | if ((codeNumber >> absGoRice) < COEF_REMAIN_BIN_REDUCTION) |
| 1105 | { |
| 1106 | length = codeNumber >> absGoRice; |
| 1107 | |
| 1108 | X265_CHECK(codeNumber - (length << absGoRice) == (codeNumber & ((1 << absGoRice) - 1)), "codeNumber failure\n"); |
| 1109 | X265_CHECK(length + 1 + absGoRice < 32, "length failure\n"); |
| 1110 | encodeBinsEP((((1 << (length + 1)) - 2) << absGoRice) + codeRemain, length + 1 + absGoRice); |
| 1111 | } |
| 1112 | else |
| 1113 | { |
| 1114 | length = 0; |
| 1115 | codeNumber = (codeNumber >> absGoRice) - COEF_REMAIN_BIN_REDUCTION; |
| 1116 | if (codeNumber != 0) |
| 1117 | { |
| 1118 | unsigned long idx; |
| 1119 | CLZ32(idx, codeNumber + 1); |
| 1120 | length = idx; |
| 1121 | codeNumber -= (1 << idx) - 1; |
| 1122 | } |
| 1123 | codeNumber = (codeNumber << absGoRice) + codeRemain; |
| 1124 | |
| 1125 | encodeBinsEP((1 << (COEF_REMAIN_BIN_REDUCTION + length + 1)) - 2, COEF_REMAIN_BIN_REDUCTION + length + 1); |
| 1126 | encodeBinsEP(codeNumber, length + absGoRice); |
| 1127 | } |
| 1128 | } |
| 1129 | |
| 1130 | // SBAC RD |
| 1131 | void Entropy::loadIntraDirModeLuma(const Entropy& src) |
| 1132 | { |
| 1133 | X265_CHECK(src.m_valid, "invalid copy source context\n"); |
| 1134 | m_fracBits = src.m_fracBits; |
| 1135 | m_contextState[OFF_ADI_CTX] = src.m_contextState[OFF_ADI_CTX]; |
| 1136 | } |
| 1137 | |
| 1138 | void Entropy::copyFrom(const Entropy& src) |
| 1139 | { |
| 1140 | X265_CHECK(src.m_valid, "invalid copy source context\n"); |
| 1141 | |
| 1142 | copyState(src); |
| 1143 | |
| 1144 | memcpy(m_contextState, src.m_contextState, MAX_OFF_CTX_MOD * sizeof(uint8_t)); |
| 1145 | markValid(); |
| 1146 | } |
| 1147 | |
| 1148 | void Entropy::codeMVPIdx(uint32_t symbol) |
| 1149 | { |
| 1150 | encodeBin(symbol, m_contextState[OFF_MVP_IDX_CTX]); |
| 1151 | } |
| 1152 | |
| 1153 | void Entropy::codePartSize(const CUData& cu, uint32_t absPartIdx, uint32_t depth) |
| 1154 | { |
| 1155 | PartSize partSize = (PartSize)cu.m_partSize[absPartIdx]; |
| 1156 | |
| 1157 | if (cu.isIntra(absPartIdx)) |
| 1158 | { |
| 1159 | if (depth == g_maxCUDepth) |
| 1160 | encodeBin(partSize == SIZE_2Nx2N ? 1 : 0, m_contextState[OFF_PART_SIZE_CTX]); |
| 1161 | return; |
| 1162 | } |
| 1163 | |
| 1164 | switch (partSize) |
| 1165 | { |
| 1166 | case SIZE_2Nx2N: |
| 1167 | encodeBin(1, m_contextState[OFF_PART_SIZE_CTX]); |
| 1168 | break; |
| 1169 | |
| 1170 | case SIZE_2NxN: |
| 1171 | case SIZE_2NxnU: |
| 1172 | case SIZE_2NxnD: |
| 1173 | encodeBin(0, m_contextState[OFF_PART_SIZE_CTX + 0]); |
| 1174 | encodeBin(1, m_contextState[OFF_PART_SIZE_CTX + 1]); |
| 1175 | if (cu.m_slice->m_sps->maxAMPDepth > depth) |
| 1176 | { |
| 1177 | encodeBin((partSize == SIZE_2NxN) ? 1 : 0, m_contextState[OFF_PART_SIZE_CTX + 3]); |
| 1178 | if (partSize != SIZE_2NxN) |
| 1179 | encodeBinEP((partSize == SIZE_2NxnU ? 0 : 1)); |
| 1180 | } |
| 1181 | break; |
| 1182 | |
| 1183 | case SIZE_Nx2N: |
| 1184 | case SIZE_nLx2N: |
| 1185 | case SIZE_nRx2N: |
| 1186 | encodeBin(0, m_contextState[OFF_PART_SIZE_CTX + 0]); |
| 1187 | encodeBin(0, m_contextState[OFF_PART_SIZE_CTX + 1]); |
| 1188 | if (depth == g_maxCUDepth && !(cu.m_log2CUSize[absPartIdx] == 3)) |
| 1189 | encodeBin(1, m_contextState[OFF_PART_SIZE_CTX + 2]); |
| 1190 | if (cu.m_slice->m_sps->maxAMPDepth > depth) |
| 1191 | { |
| 1192 | encodeBin((partSize == SIZE_Nx2N) ? 1 : 0, m_contextState[OFF_PART_SIZE_CTX + 3]); |
| 1193 | if (partSize != SIZE_Nx2N) |
| 1194 | encodeBinEP((partSize == SIZE_nLx2N ? 0 : 1)); |
| 1195 | } |
| 1196 | break; |
| 1197 | default: |
| 1198 | X265_CHECK(0, "invalid CU partition\n"); |
| 1199 | break; |
| 1200 | } |
| 1201 | } |
| 1202 | |
| 1203 | void Entropy::codePredMode(int predMode) |
| 1204 | { |
| 1205 | encodeBin(predMode == MODE_INTER ? 0 : 1, m_contextState[OFF_PRED_MODE_CTX]); |
| 1206 | } |
| 1207 | |
| 1208 | void Entropy::codeCUTransquantBypassFlag(uint32_t symbol) |
| 1209 | { |
| 1210 | encodeBin(symbol, m_contextState[OFF_TQUANT_BYPASS_FLAG_CTX]); |
| 1211 | } |
| 1212 | |
| 1213 | void Entropy::codeSkipFlag(const CUData& cu, uint32_t absPartIdx) |
| 1214 | { |
| 1215 | // get context function is here |
| 1216 | uint32_t symbol = cu.isSkipped(absPartIdx) ? 1 : 0; |
| 1217 | uint32_t ctxSkip = cu.getCtxSkipFlag(absPartIdx); |
| 1218 | |
| 1219 | encodeBin(symbol, m_contextState[OFF_SKIP_FLAG_CTX + ctxSkip]); |
| 1220 | } |
| 1221 | |
| 1222 | void Entropy::codeMergeFlag(const CUData& cu, uint32_t absPartIdx) |
| 1223 | { |
| 1224 | const uint32_t symbol = cu.m_mergeFlag[absPartIdx] ? 1 : 0; |
| 1225 | |
| 1226 | encodeBin(symbol, m_contextState[OFF_MERGE_FLAG_EXT_CTX]); |
| 1227 | } |
| 1228 | |
| 1229 | void Entropy::codeMergeIndex(const CUData& cu, uint32_t absPartIdx) |
| 1230 | { |
| 1231 | uint32_t numCand = cu.m_slice->m_maxNumMergeCand; |
| 1232 | |
| 1233 | if (numCand > 1) |
| 1234 | { |
| 1235 | uint32_t unaryIdx = cu.m_mvpIdx[0][absPartIdx]; // merge candidate index was stored in L0 MVP idx |
| 1236 | encodeBin((unaryIdx != 0), m_contextState[OFF_MERGE_IDX_EXT_CTX]); |
| 1237 | |
| 1238 | X265_CHECK(unaryIdx < numCand, "unaryIdx out of range\n"); |
| 1239 | |
| 1240 | if (unaryIdx != 0) |
| 1241 | { |
| 1242 | uint32_t mask = (1 << unaryIdx) - 2; |
| 1243 | mask >>= (unaryIdx == numCand - 1) ? 1 : 0; |
| 1244 | encodeBinsEP(mask, unaryIdx - (unaryIdx == numCand - 1)); |
| 1245 | } |
| 1246 | } |
| 1247 | } |
| 1248 | |
| 1249 | void Entropy::codeSplitFlag(const CUData& cu, uint32_t absPartIdx, uint32_t depth) |
| 1250 | { |
| 1251 | X265_CHECK(depth < g_maxCUDepth, "invalid depth\n"); |
| 1252 | |
| 1253 | uint32_t ctx = cu.getCtxSplitFlag(absPartIdx, depth); |
| 1254 | uint32_t currSplitFlag = (cu.m_cuDepth[absPartIdx] > depth) ? 1 : 0; |
| 1255 | |
| 1256 | X265_CHECK(ctx < 3, "ctx out of range\n"); |
| 1257 | encodeBin(currSplitFlag, m_contextState[OFF_SPLIT_FLAG_CTX + ctx]); |
| 1258 | } |
| 1259 | |
| 1260 | void Entropy::codeTransformSubdivFlag(uint32_t symbol, uint32_t ctx) |
| 1261 | { |
| 1262 | encodeBin(symbol, m_contextState[OFF_TRANS_SUBDIV_FLAG_CTX + ctx]); |
| 1263 | } |
| 1264 | |
| 1265 | uint32_t Entropy::bitsIntraModeNonMPM() const |
| 1266 | { |
| 1267 | uint32_t mstate = m_contextState[OFF_ADI_CTX]; |
| 1268 | uint32_t bits = ((uint32_t)(m_fracBits & 32767) + sbacGetEntropyBits(mstate, 0)) >> 15; |
| 1269 | return bits + 5; /* fixed cost for encodeBinsEP() */ |
| 1270 | } |
| 1271 | |
| 1272 | uint32_t Entropy::bitsIntraModeMPM(const uint32_t preds[3], uint32_t dir) const |
| 1273 | { |
| 1274 | X265_CHECK(dir == preds[0] || dir == preds[1] || dir == preds[2], "dir must be a most probable mode\n"); |
| 1275 | uint32_t mstate = m_contextState[OFF_ADI_CTX]; |
| 1276 | uint32_t bits = ((uint32_t)(m_fracBits & 32767) + sbacGetEntropyBits(mstate, 1)) >> 15; |
| 1277 | return bits + (dir == preds[0] ? 1 : 2); |
| 1278 | } |
| 1279 | |
| 1280 | void Entropy::codeIntraDirLumaAng(const CUData& cu, uint32_t absPartIdx, bool isMultiple) |
| 1281 | { |
| 1282 | uint32_t dir[4], j; |
| 1283 | uint32_t preds[4][3]; |
| 1284 | int predIdx[4]; |
| 1285 | PartSize mode = (PartSize)cu.m_partSize[absPartIdx]; |
| 1286 | uint32_t partNum = isMultiple ? (mode == SIZE_NxN ? 4 : 1) : 1; |
| 1287 | uint32_t partOffset = (NUM_CU_PARTITIONS >> (cu.m_cuDepth[absPartIdx] << 1)) >> 2; |
| 1288 | |
| 1289 | for (j = 0; j < partNum; j++) |
| 1290 | { |
| 1291 | dir[j] = cu.m_lumaIntraDir[absPartIdx + partOffset * j]; |
| 1292 | cu.getIntraDirLumaPredictor(absPartIdx + partOffset * j, preds[j]); |
| 1293 | predIdx[j] = -1; |
| 1294 | for (uint32_t i = 0; i < 3; i++) |
| 1295 | if (dir[j] == preds[j][i]) |
| 1296 | predIdx[j] = i; |
| 1297 | |
| 1298 | encodeBin((predIdx[j] != -1) ? 1 : 0, m_contextState[OFF_ADI_CTX]); |
| 1299 | } |
| 1300 | |
| 1301 | for (j = 0; j < partNum; j++) |
| 1302 | { |
| 1303 | if (predIdx[j] != -1) |
| 1304 | { |
| 1305 | X265_CHECK((predIdx[j] >= 0) && (predIdx[j] <= 2), "predIdx out of range\n"); |
| 1306 | // NOTE: Mapping |
| 1307 | // 0 = 0 |
| 1308 | // 1 = 10 |
| 1309 | // 2 = 11 |
| 1310 | int nonzero = (!!predIdx[j]); |
| 1311 | encodeBinsEP(predIdx[j] + nonzero, 1 + nonzero); |
| 1312 | } |
| 1313 | else |
| 1314 | { |
| 1315 | if (preds[j][0] > preds[j][1]) |
| 1316 | std::swap(preds[j][0], preds[j][1]); |
| 1317 | |
| 1318 | if (preds[j][0] > preds[j][2]) |
| 1319 | std::swap(preds[j][0], preds[j][2]); |
| 1320 | |
| 1321 | if (preds[j][1] > preds[j][2]) |
| 1322 | std::swap(preds[j][1], preds[j][2]); |
| 1323 | |
| 1324 | dir[j] += (dir[j] > preds[j][2]) ? -1 : 0; |
| 1325 | dir[j] += (dir[j] > preds[j][1]) ? -1 : 0; |
| 1326 | dir[j] += (dir[j] > preds[j][0]) ? -1 : 0; |
| 1327 | |
| 1328 | encodeBinsEP(dir[j], 5); |
| 1329 | } |
| 1330 | } |
| 1331 | } |
| 1332 | |
| 1333 | void Entropy::codeIntraDirChroma(const CUData& cu, uint32_t absPartIdx, uint32_t *chromaDirMode) |
| 1334 | { |
| 1335 | uint32_t intraDirChroma = cu.m_chromaIntraDir[absPartIdx]; |
| 1336 | |
| 1337 | if (intraDirChroma == DM_CHROMA_IDX) |
| 1338 | encodeBin(0, m_contextState[OFF_CHROMA_PRED_CTX]); |
| 1339 | else |
| 1340 | { |
| 1341 | for (int i = 0; i < NUM_CHROMA_MODE - 1; i++) |
| 1342 | { |
| 1343 | if (intraDirChroma == chromaDirMode[i]) |
| 1344 | { |
| 1345 | intraDirChroma = i; |
| 1346 | break; |
| 1347 | } |
| 1348 | } |
| 1349 | |
| 1350 | encodeBin(1, m_contextState[OFF_CHROMA_PRED_CTX]); |
| 1351 | encodeBinsEP(intraDirChroma, 2); |
| 1352 | } |
| 1353 | } |
| 1354 | |
| 1355 | void Entropy::codeInterDir(const CUData& cu, uint32_t absPartIdx) |
| 1356 | { |
| 1357 | const uint32_t interDir = cu.m_interDir[absPartIdx] - 1; |
| 1358 | const uint32_t ctx = cu.m_cuDepth[absPartIdx]; // the context of the inter dir is the depth of the CU |
| 1359 | |
| 1360 | if (cu.m_partSize[absPartIdx] == SIZE_2Nx2N || cu.m_log2CUSize[absPartIdx] != 3) |
| 1361 | encodeBin(interDir == 2 ? 1 : 0, m_contextState[OFF_INTER_DIR_CTX + ctx]); |
| 1362 | if (interDir < 2) |
| 1363 | encodeBin(interDir, m_contextState[OFF_INTER_DIR_CTX + 4]); |
| 1364 | } |
| 1365 | |
| 1366 | void Entropy::codeRefFrmIdx(const CUData& cu, uint32_t absPartIdx, int list) |
| 1367 | { |
| 1368 | uint32_t refFrame = cu.m_refIdx[list][absPartIdx]; |
| 1369 | |
| 1370 | encodeBin(refFrame > 0, m_contextState[OFF_REF_NO_CTX]); |
| 1371 | |
| 1372 | if (refFrame > 0) |
| 1373 | { |
| 1374 | uint32_t refNum = cu.m_slice->m_numRefIdx[list] - 2; |
| 1375 | if (refNum == 0) |
| 1376 | return; |
| 1377 | |
| 1378 | refFrame--; |
| 1379 | encodeBin(refFrame > 0, m_contextState[OFF_REF_NO_CTX + 1]); |
| 1380 | if (refFrame > 0) |
| 1381 | { |
| 1382 | uint32_t mask = (1 << refFrame) - 2; |
| 1383 | mask >>= (refFrame == refNum) ? 1 : 0; |
| 1384 | encodeBinsEP(mask, refFrame - (refFrame == refNum)); |
| 1385 | } |
| 1386 | } |
| 1387 | } |
| 1388 | |
| 1389 | void Entropy::codeMvd(const CUData& cu, uint32_t absPartIdx, int list) |
| 1390 | { |
| 1391 | const MV& mvd = cu.m_mvd[list][absPartIdx]; |
| 1392 | const int hor = mvd.x; |
| 1393 | const int ver = mvd.y; |
| 1394 | |
| 1395 | encodeBin(hor != 0 ? 1 : 0, m_contextState[OFF_MV_RES_CTX]); |
| 1396 | encodeBin(ver != 0 ? 1 : 0, m_contextState[OFF_MV_RES_CTX]); |
| 1397 | |
| 1398 | const bool bHorAbsGr0 = hor != 0; |
| 1399 | const bool bVerAbsGr0 = ver != 0; |
| 1400 | const uint32_t horAbs = 0 > hor ? -hor : hor; |
| 1401 | const uint32_t verAbs = 0 > ver ? -ver : ver; |
| 1402 | |
| 1403 | if (bHorAbsGr0) |
| 1404 | encodeBin(horAbs > 1 ? 1 : 0, m_contextState[OFF_MV_RES_CTX + 1]); |
| 1405 | |
| 1406 | if (bVerAbsGr0) |
| 1407 | encodeBin(verAbs > 1 ? 1 : 0, m_contextState[OFF_MV_RES_CTX + 1]); |
| 1408 | |
| 1409 | if (bHorAbsGr0) |
| 1410 | { |
| 1411 | if (horAbs > 1) |
| 1412 | writeEpExGolomb(horAbs - 2, 1); |
| 1413 | |
| 1414 | encodeBinEP(0 > hor ? 1 : 0); |
| 1415 | } |
| 1416 | |
| 1417 | if (bVerAbsGr0) |
| 1418 | { |
| 1419 | if (verAbs > 1) |
| 1420 | writeEpExGolomb(verAbs - 2, 1); |
| 1421 | |
| 1422 | encodeBinEP(0 > ver ? 1 : 0); |
| 1423 | } |
| 1424 | } |
| 1425 | |
| 1426 | void Entropy::codeDeltaQP(const CUData& cu, uint32_t absPartIdx) |
| 1427 | { |
| 1428 | int dqp = cu.m_qp[absPartIdx] - cu.getRefQP(absPartIdx); |
| 1429 | |
| 1430 | int qpBdOffsetY = QP_BD_OFFSET; |
| 1431 | |
| 1432 | dqp = (dqp + 78 + qpBdOffsetY + (qpBdOffsetY / 2)) % (52 + qpBdOffsetY) - 26 - (qpBdOffsetY / 2); |
| 1433 | |
| 1434 | uint32_t absDQp = (uint32_t)((dqp > 0) ? dqp : (-dqp)); |
| 1435 | uint32_t TUValue = X265_MIN((int)absDQp, CU_DQP_TU_CMAX); |
| 1436 | writeUnaryMaxSymbol(TUValue, &m_contextState[OFF_DELTA_QP_CTX], 1, CU_DQP_TU_CMAX); |
| 1437 | if (absDQp >= CU_DQP_TU_CMAX) |
| 1438 | writeEpExGolomb(absDQp - CU_DQP_TU_CMAX, CU_DQP_EG_k); |
| 1439 | |
| 1440 | if (absDQp > 0) |
| 1441 | { |
| 1442 | uint32_t sign = (dqp > 0 ? 0 : 1); |
| 1443 | encodeBinEP(sign); |
| 1444 | } |
| 1445 | } |
| 1446 | |
| 1447 | void Entropy::codeQtCbf(const CUData& cu, uint32_t absPartIdx, uint32_t absPartIdxStep, uint32_t width, uint32_t height, TextType ttype, uint32_t trDepth, bool lowestLevel) |
| 1448 | { |
| 1449 | uint32_t ctx = ctxCbf[ttype][trDepth]; |
| 1450 | |
| 1451 | bool canQuadSplit = (width >= (MIN_TU_SIZE * 2)) && (height >= (MIN_TU_SIZE * 2)); |
| 1452 | uint32_t lowestTUDepth = trDepth + ((!lowestLevel && !canQuadSplit) ? 1 : 0); // unsplittable TUs inherit their parent's CBF |
| 1453 | |
| 1454 | if ((width != height) && (lowestLevel || !canQuadSplit)) // if sub-TUs are present |
| 1455 | { |
| 1456 | uint32_t subTUDepth = lowestTUDepth + 1; // if this is the lowest level of the TU-tree, the sub-TUs are directly below. |
| 1457 | // Otherwise, this must be the level above the lowest level (as specified above) |
| 1458 | uint32_t partIdxesPerSubTU = absPartIdxStep >> 1; |
| 1459 | |
| 1460 | for (uint32_t subTU = 0; subTU < 2; subTU++) |
| 1461 | { |
| 1462 | uint32_t subTUAbsPartIdx = absPartIdx + (subTU * partIdxesPerSubTU); |
| 1463 | uint32_t cbf = cu.getCbf(subTUAbsPartIdx, ttype, subTUDepth); |
| 1464 | |
| 1465 | encodeBin(cbf, m_contextState[OFF_QT_CBF_CTX + ctx]); |
| 1466 | } |
| 1467 | } |
| 1468 | else |
| 1469 | { |
| 1470 | uint32_t cbf = cu.getCbf(absPartIdx, ttype, lowestTUDepth); |
| 1471 | |
| 1472 | encodeBin(cbf, m_contextState[OFF_QT_CBF_CTX + ctx]); |
| 1473 | } |
| 1474 | } |
| 1475 | |
| 1476 | void Entropy::codeQtCbf(const CUData& cu, uint32_t absPartIdx, TextType ttype, uint32_t trDepth) |
| 1477 | { |
| 1478 | uint32_t ctx = ctxCbf[ttype][trDepth]; |
| 1479 | uint32_t cbf = cu.getCbf(absPartIdx, ttype, trDepth); |
| 1480 | encodeBin(cbf, m_contextState[OFF_QT_CBF_CTX + ctx]); |
| 1481 | } |
| 1482 | |
| 1483 | void Entropy::codeQtCbf(uint32_t cbf, TextType ttype, uint32_t trDepth) |
| 1484 | { |
| 1485 | uint32_t ctx = ctxCbf[ttype][trDepth]; |
| 1486 | encodeBin(cbf, m_contextState[OFF_QT_CBF_CTX + ctx]); |
| 1487 | } |
| 1488 | |
| 1489 | void Entropy::codeTransformSkipFlags(const CUData& cu, uint32_t absPartIdx, uint32_t trSize, TextType ttype) |
| 1490 | { |
| 1491 | if (cu.m_tqBypass[absPartIdx]) |
| 1492 | return; |
| 1493 | if (trSize != 4) |
| 1494 | return; |
| 1495 | |
| 1496 | uint32_t useTransformSkip = cu.m_transformSkip[ttype][absPartIdx]; |
| 1497 | encodeBin(useTransformSkip, m_contextState[OFF_TRANSFORMSKIP_FLAG_CTX + (ttype ? NUM_TRANSFORMSKIP_FLAG_CTX : 0)]); |
| 1498 | } |
| 1499 | |
| 1500 | void Entropy::codeQtRootCbf(uint32_t cbf) |
| 1501 | { |
| 1502 | encodeBin(cbf, m_contextState[OFF_QT_ROOT_CBF_CTX]); |
| 1503 | } |
| 1504 | |
| 1505 | void Entropy::codeQtCbfZero(TextType ttype, uint32_t trDepth) |
| 1506 | { |
| 1507 | // this function is only used to estimate the bits when cbf is 0 |
| 1508 | // and will never be called when writing the bitsream. |
| 1509 | uint32_t ctx = ctxCbf[ttype][trDepth]; |
| 1510 | encodeBin(0, m_contextState[OFF_QT_CBF_CTX + ctx]); |
| 1511 | } |
| 1512 | |
| 1513 | void Entropy::codeQtRootCbfZero() |
| 1514 | { |
| 1515 | // this function is only used to estimate the bits when cbf is 0 |
| 1516 | // and will never be called when writing the bistream. |
| 1517 | encodeBin(0, m_contextState[OFF_QT_ROOT_CBF_CTX]); |
| 1518 | } |
| 1519 | |
| 1520 | /** Encode (X,Y) position of the last significant coefficient |
| 1521 | * \param posx X component of last coefficient |
| 1522 | * \param posy Y component of last coefficient |
| 1523 | * \param log2TrSize |
| 1524 | * \param bIsLuma |
| 1525 | * \param scanIdx scan type (zig-zag, hor, ver) |
| 1526 | * This method encodes the X and Y component within a block of the last significant coefficient. |
| 1527 | */ |
| 1528 | void Entropy::codeLastSignificantXY(uint32_t posx, uint32_t posy, uint32_t log2TrSize, bool bIsLuma, uint32_t scanIdx) |
| 1529 | { |
| 1530 | // swap |
| 1531 | if (scanIdx == SCAN_VER) |
| 1532 | std::swap(posx, posy); |
| 1533 | |
| 1534 | uint32_t ctxLast; |
| 1535 | uint32_t groupIdxX = getGroupIdx(posx); |
| 1536 | uint32_t groupIdxY = getGroupIdx(posy); |
| 1537 | |
| 1538 | int blkSizeOffset = bIsLuma ? ((log2TrSize - 2) * 3 + ((log2TrSize - 1) >> 2)) : NUM_CTX_LAST_FLAG_XY_LUMA; |
| 1539 | int ctxShift = bIsLuma ? ((log2TrSize + 1) >> 2) : log2TrSize - 2; |
| 1540 | uint32_t maxGroupIdx = log2TrSize * 2 - 1; |
| 1541 | |
| 1542 | // posX |
| 1543 | uint8_t *ctxX = &m_contextState[OFF_CTX_LAST_FLAG_X]; |
| 1544 | for (ctxLast = 0; ctxLast < groupIdxX; ctxLast++) |
| 1545 | encodeBin(1, *(ctxX + blkSizeOffset + (ctxLast >> ctxShift))); |
| 1546 | |
| 1547 | if (groupIdxX < maxGroupIdx) |
| 1548 | encodeBin(0, *(ctxX + blkSizeOffset + (ctxLast >> ctxShift))); |
| 1549 | |
| 1550 | // posY |
| 1551 | uint8_t *ctxY = &m_contextState[OFF_CTX_LAST_FLAG_Y]; |
| 1552 | for (ctxLast = 0; ctxLast < groupIdxY; ctxLast++) |
| 1553 | encodeBin(1, *(ctxY + blkSizeOffset + (ctxLast >> ctxShift))); |
| 1554 | |
| 1555 | if (groupIdxY < maxGroupIdx) |
| 1556 | encodeBin(0, *(ctxY + blkSizeOffset + (ctxLast >> ctxShift))); |
| 1557 | |
| 1558 | if (groupIdxX > 3) |
| 1559 | { |
| 1560 | uint32_t count = (groupIdxX - 2) >> 1; |
| 1561 | posx = posx - g_minInGroup[groupIdxX]; |
| 1562 | encodeBinsEP(posx, count); |
| 1563 | } |
| 1564 | if (groupIdxY > 3) |
| 1565 | { |
| 1566 | uint32_t count = (groupIdxY - 2) >> 1; |
| 1567 | posy = posy - g_minInGroup[groupIdxY]; |
| 1568 | encodeBinsEP(posy, count); |
| 1569 | } |
| 1570 | } |
| 1571 | |
| 1572 | void Entropy::codeCoeffNxN(const CUData& cu, const coeff_t* coeff, uint32_t absPartIdx, uint32_t log2TrSize, TextType ttype) |
| 1573 | { |
| 1574 | uint32_t trSize = 1 << log2TrSize; |
| 1575 | |
| 1576 | // compute number of significant coefficients |
| 1577 | uint32_t numSig = primitives.count_nonzero(coeff, (1 << (log2TrSize << 1))); |
| 1578 | |
| 1579 | X265_CHECK(numSig > 0, "cbf check fail\n"); |
| 1580 | |
| 1581 | bool bHideFirstSign = cu.m_slice->m_pps->bSignHideEnabled && !cu.m_tqBypass[absPartIdx]; |
| 1582 | |
| 1583 | if (cu.m_slice->m_pps->bTransformSkipEnabled) |
| 1584 | codeTransformSkipFlags(cu, absPartIdx, trSize, ttype); |
| 1585 | |
| 1586 | bool bIsLuma = ttype == TEXT_LUMA; |
| 1587 | |
| 1588 | // select scans |
| 1589 | TUEntropyCodingParameters codingParameters; |
| 1590 | cu.getTUEntropyCodingParameters(codingParameters, absPartIdx, log2TrSize, bIsLuma); |
| 1591 | |
| 1592 | //----- encode significance map ----- |
| 1593 | |
| 1594 | // Find position of last coefficient |
| 1595 | int scanPosLast = 0; |
| 1596 | uint32_t posLast; |
| 1597 | uint64_t sigCoeffGroupFlag64 = 0; |
| 1598 | const uint32_t maskPosXY = ((uint32_t)~0 >> (31 - log2TrSize + MLS_CG_LOG2_SIZE)) >> 1; |
| 1599 | assert((uint32_t)((1 << (log2TrSize - MLS_CG_LOG2_SIZE)) - 1) == (((uint32_t)~0 >> (31 - log2TrSize + MLS_CG_LOG2_SIZE)) >> 1)); |
| 1600 | do |
| 1601 | { |
| 1602 | posLast = codingParameters.scan[scanPosLast++]; |
| 1603 | |
| 1604 | const uint32_t isNZCoeff = (coeff[posLast] != 0); |
| 1605 | // get L1 sig map |
| 1606 | // NOTE: the new algorithm is complicated, so I keep reference code here |
| 1607 | //uint32_t posy = posLast >> log2TrSize; |
| 1608 | //uint32_t posx = posLast - (posy << log2TrSize); |
| 1609 | //uint32_t blkIdx0 = ((posy >> MLS_CG_LOG2_SIZE) << codingParameters.log2TrSizeCG) + (posx >> MLS_CG_LOG2_SIZE); |
| 1610 | const uint32_t blkIdx = ((posLast >> (2 * MLS_CG_LOG2_SIZE)) & ~maskPosXY) + ((posLast >> MLS_CG_LOG2_SIZE) & maskPosXY); |
| 1611 | sigCoeffGroupFlag64 |= ((uint64_t)isNZCoeff << blkIdx); |
| 1612 | numSig -= isNZCoeff; |
| 1613 | } |
| 1614 | while (numSig > 0); |
| 1615 | scanPosLast--; |
| 1616 | |
| 1617 | // Code position of last coefficient |
| 1618 | int posLastY = posLast >> log2TrSize; |
| 1619 | int posLastX = posLast & (trSize - 1); |
| 1620 | codeLastSignificantXY(posLastX, posLastY, log2TrSize, bIsLuma, codingParameters.scanType); |
| 1621 | |
| 1622 | //===== code significance flag ===== |
| 1623 | uint8_t * const baseCoeffGroupCtx = &m_contextState[OFF_SIG_CG_FLAG_CTX + (bIsLuma ? 0 : NUM_SIG_CG_FLAG_CTX)]; |
| 1624 | uint8_t * const baseCtx = bIsLuma ? &m_contextState[OFF_SIG_FLAG_CTX] : &m_contextState[OFF_SIG_FLAG_CTX + NUM_SIG_FLAG_CTX_LUMA]; |
| 1625 | const int lastScanSet = scanPosLast >> MLS_CG_SIZE; |
| 1626 | uint32_t c1 = 1; |
| 1627 | uint32_t goRiceParam = 0; |
| 1628 | int scanPosSig = scanPosLast; |
| 1629 | |
| 1630 | for (int subSet = lastScanSet; subSet >= 0; subSet--) |
| 1631 | { |
| 1632 | int numNonZero = 0; |
| 1633 | int subPos = subSet << MLS_CG_SIZE; |
| 1634 | goRiceParam = 0; |
| 1635 | int absCoeff[1 << MLS_CG_SIZE]; |
| 1636 | uint32_t coeffSigns = 0; |
| 1637 | int lastNZPosInCG = -1; |
| 1638 | int firstNZPosInCG = 1 << MLS_CG_SIZE; |
| 1639 | if (scanPosSig == scanPosLast) |
| 1640 | { |
| 1641 | absCoeff[0] = int(abs(coeff[posLast])); |
| 1642 | coeffSigns = (coeff[posLast] < 0); |
| 1643 | numNonZero = 1; |
| 1644 | lastNZPosInCG = scanPosSig; |
| 1645 | firstNZPosInCG = scanPosSig; |
| 1646 | scanPosSig--; |
| 1647 | } |
| 1648 | // encode significant_coeffgroup_flag |
| 1649 | const int cgBlkPos = codingParameters.scanCG[subSet]; |
| 1650 | const int cgPosY = cgBlkPos >> codingParameters.log2TrSizeCG; |
| 1651 | const int cgPosX = cgBlkPos - (cgPosY << codingParameters.log2TrSizeCG); |
| 1652 | const uint64_t cgBlkPosMask = ((uint64_t)1 << cgBlkPos); |
| 1653 | |
| 1654 | if (subSet == lastScanSet || !subSet) |
| 1655 | sigCoeffGroupFlag64 |= cgBlkPosMask; |
| 1656 | else |
| 1657 | { |
| 1658 | uint32_t sigCoeffGroup = ((sigCoeffGroupFlag64 & cgBlkPosMask) != 0); |
| 1659 | uint32_t ctxSig = Quant::getSigCoeffGroupCtxInc(sigCoeffGroupFlag64, cgPosX, cgPosY, codingParameters.log2TrSizeCG); |
| 1660 | encodeBin(sigCoeffGroup, baseCoeffGroupCtx[ctxSig]); |
| 1661 | } |
| 1662 | |
| 1663 | // encode significant_coeff_flag |
| 1664 | if (sigCoeffGroupFlag64 & cgBlkPosMask) |
| 1665 | { |
| 1666 | const int patternSigCtx = Quant::calcPatternSigCtx(sigCoeffGroupFlag64, cgPosX, cgPosY, codingParameters.log2TrSizeCG); |
| 1667 | uint32_t blkPos, sig, ctxSig; |
| 1668 | for (; scanPosSig >= subPos; scanPosSig--) |
| 1669 | { |
| 1670 | blkPos = codingParameters.scan[scanPosSig]; |
| 1671 | sig = (coeff[blkPos] != 0); |
| 1672 | if (scanPosSig > subPos || subSet == 0 || numNonZero) |
| 1673 | { |
| 1674 | ctxSig = Quant::getSigCtxInc(patternSigCtx, log2TrSize, trSize, blkPos, bIsLuma, codingParameters.firstSignificanceMapContext); |
| 1675 | encodeBin(sig, baseCtx[ctxSig]); |
| 1676 | } |
| 1677 | if (sig) |
| 1678 | { |
| 1679 | absCoeff[numNonZero] = int(abs(coeff[blkPos])); |
| 1680 | coeffSigns = 2 * coeffSigns + ((uint32_t)coeff[blkPos] >> 31); |
| 1681 | numNonZero++; |
| 1682 | if (lastNZPosInCG < 0) |
| 1683 | lastNZPosInCG = scanPosSig; |
| 1684 | firstNZPosInCG = scanPosSig; |
| 1685 | } |
| 1686 | } |
| 1687 | } |
| 1688 | else |
| 1689 | scanPosSig = subPos - 1; |
| 1690 | |
| 1691 | if (numNonZero > 0) |
| 1692 | { |
| 1693 | bool signHidden = (lastNZPosInCG - firstNZPosInCG >= SBH_THRESHOLD); |
| 1694 | uint32_t ctxSet = (subSet > 0 && bIsLuma) ? 2 : 0; |
| 1695 | |
| 1696 | if (c1 == 0) |
| 1697 | ctxSet++; |
| 1698 | |
| 1699 | c1 = 1; |
| 1700 | uint8_t *baseCtxMod = bIsLuma ? &m_contextState[OFF_ONE_FLAG_CTX + 4 * ctxSet] : &m_contextState[OFF_ONE_FLAG_CTX + NUM_ONE_FLAG_CTX_LUMA + 4 * ctxSet]; |
| 1701 | |
| 1702 | int numC1Flag = X265_MIN(numNonZero, C1FLAG_NUMBER); |
| 1703 | int firstC2FlagIdx = -1; |
| 1704 | for (int idx = 0; idx < numC1Flag; idx++) |
| 1705 | { |
| 1706 | uint32_t symbol = absCoeff[idx] > 1; |
| 1707 | encodeBin(symbol, baseCtxMod[c1]); |
| 1708 | if (symbol) |
| 1709 | { |
| 1710 | c1 = 0; |
| 1711 | |
| 1712 | if (firstC2FlagIdx == -1) |
| 1713 | firstC2FlagIdx = idx; |
| 1714 | } |
| 1715 | else if ((c1 < 3) && (c1 > 0)) |
| 1716 | c1++; |
| 1717 | } |
| 1718 | |
| 1719 | if (!c1) |
| 1720 | { |
| 1721 | baseCtxMod = bIsLuma ? &m_contextState[OFF_ABS_FLAG_CTX + ctxSet] : &m_contextState[OFF_ABS_FLAG_CTX + NUM_ABS_FLAG_CTX_LUMA + ctxSet]; |
| 1722 | if (firstC2FlagIdx != -1) |
| 1723 | { |
| 1724 | uint32_t symbol = absCoeff[firstC2FlagIdx] > 2; |
| 1725 | encodeBin(symbol, baseCtxMod[0]); |
| 1726 | } |
| 1727 | } |
| 1728 | |
| 1729 | if (bHideFirstSign && signHidden) |
| 1730 | encodeBinsEP((coeffSigns >> 1), numNonZero - 1); |
| 1731 | else |
| 1732 | encodeBinsEP(coeffSigns, numNonZero); |
| 1733 | |
| 1734 | int firstCoeff2 = 1; |
| 1735 | if (!c1 || numNonZero > C1FLAG_NUMBER) |
| 1736 | { |
| 1737 | for (int idx = 0; idx < numNonZero; idx++) |
| 1738 | { |
| 1739 | int baseLevel = (idx < C1FLAG_NUMBER) ? (2 + firstCoeff2) : 1; |
| 1740 | |
| 1741 | if (absCoeff[idx] >= baseLevel) |
| 1742 | { |
| 1743 | writeCoefRemainExGolomb(absCoeff[idx] - baseLevel, goRiceParam); |
| 1744 | if (absCoeff[idx] > 3 * (1 << goRiceParam)) |
| 1745 | goRiceParam = std::min<uint32_t>(goRiceParam + 1, 4); |
| 1746 | } |
| 1747 | if (absCoeff[idx] >= 2) |
| 1748 | firstCoeff2 = 0; |
| 1749 | } |
| 1750 | } |
| 1751 | } |
| 1752 | } |
| 1753 | } |
| 1754 | |
| 1755 | void Entropy::codeSaoMaxUvlc(uint32_t code, uint32_t maxSymbol) |
| 1756 | { |
| 1757 | X265_CHECK(maxSymbol > 0, "maxSymbol too small\n"); |
| 1758 | |
| 1759 | uint32_t isCodeNonZero = !!code; |
| 1760 | |
| 1761 | encodeBinEP(isCodeNonZero); |
| 1762 | if (isCodeNonZero) |
| 1763 | { |
| 1764 | uint32_t isCodeLast = (maxSymbol > code); |
| 1765 | uint32_t mask = (1 << (code - 1)) - 1; |
| 1766 | uint32_t len = code - 1 + isCodeLast; |
| 1767 | mask <<= isCodeLast; |
| 1768 | |
| 1769 | encodeBinsEP(mask, len); |
| 1770 | } |
| 1771 | } |
| 1772 | |
| 1773 | /* estimate bit cost for CBP, significant map and significant coefficients */ |
| 1774 | void Entropy::estBit(EstBitsSbac& estBitsSbac, uint32_t log2TrSize, bool bIsLuma) const |
| 1775 | { |
| 1776 | estCBFBit(estBitsSbac); |
| 1777 | |
| 1778 | estSignificantCoeffGroupMapBit(estBitsSbac, bIsLuma); |
| 1779 | |
| 1780 | // encode significance map |
| 1781 | estSignificantMapBit(estBitsSbac, log2TrSize, bIsLuma); |
| 1782 | |
| 1783 | // encode significant coefficients |
| 1784 | estSignificantCoefficientsBit(estBitsSbac, bIsLuma); |
| 1785 | } |
| 1786 | |
| 1787 | /* estimate bit cost for each CBP bit */ |
| 1788 | void Entropy::estCBFBit(EstBitsSbac& estBitsSbac) const |
| 1789 | { |
| 1790 | const uint8_t *ctx = &m_contextState[OFF_QT_CBF_CTX]; |
| 1791 | |
| 1792 | for (uint32_t ctxInc = 0; ctxInc < NUM_QT_CBF_CTX; ctxInc++) |
| 1793 | { |
| 1794 | estBitsSbac.blockCbpBits[ctxInc][0] = sbacGetEntropyBits(ctx[ctxInc], 0); |
| 1795 | estBitsSbac.blockCbpBits[ctxInc][1] = sbacGetEntropyBits(ctx[ctxInc], 1); |
| 1796 | } |
| 1797 | |
| 1798 | ctx = &m_contextState[OFF_QT_ROOT_CBF_CTX]; |
| 1799 | |
| 1800 | estBitsSbac.blockRootCbpBits[0] = sbacGetEntropyBits(ctx[0], 0); |
| 1801 | estBitsSbac.blockRootCbpBits[1] = sbacGetEntropyBits(ctx[0], 1); |
| 1802 | } |
| 1803 | |
| 1804 | /* estimate SAMBAC bit cost for significant coefficient group map */ |
| 1805 | void Entropy::estSignificantCoeffGroupMapBit(EstBitsSbac& estBitsSbac, bool bIsLuma) const |
| 1806 | { |
| 1807 | int firstCtx = 0, numCtx = NUM_SIG_CG_FLAG_CTX; |
| 1808 | |
| 1809 | for (int ctxIdx = firstCtx; ctxIdx < firstCtx + numCtx; ctxIdx++) |
| 1810 | for (uint32_t bin = 0; bin < 2; bin++) |
| 1811 | estBitsSbac.significantCoeffGroupBits[ctxIdx][bin] = sbacGetEntropyBits(m_contextState[OFF_SIG_CG_FLAG_CTX + ((bIsLuma ? 0 : NUM_SIG_CG_FLAG_CTX) + ctxIdx)], bin); |
| 1812 | } |
| 1813 | |
| 1814 | /* estimate SAMBAC bit cost for significant coefficient map */ |
| 1815 | void Entropy::estSignificantMapBit(EstBitsSbac& estBitsSbac, uint32_t log2TrSize, bool bIsLuma) const |
| 1816 | { |
| 1817 | int firstCtx = 1, numCtx = 8; |
| 1818 | |
| 1819 | if (log2TrSize >= 4) |
| 1820 | { |
| 1821 | firstCtx = bIsLuma ? 21 : 12; |
| 1822 | numCtx = bIsLuma ? 6 : 3; |
| 1823 | } |
| 1824 | else if (log2TrSize == 3) |
| 1825 | { |
| 1826 | firstCtx = 9; |
| 1827 | numCtx = bIsLuma ? 12 : 3; |
| 1828 | } |
| 1829 | |
| 1830 | if (bIsLuma) |
| 1831 | { |
| 1832 | for (uint32_t bin = 0; bin < 2; bin++) |
| 1833 | estBitsSbac.significantBits[0][bin] = sbacGetEntropyBits(m_contextState[OFF_SIG_FLAG_CTX], bin); |
| 1834 | |
| 1835 | for (int ctxIdx = firstCtx; ctxIdx < firstCtx + numCtx; ctxIdx++) |
| 1836 | for (uint32_t bin = 0; bin < 2; bin++) |
| 1837 | estBitsSbac.significantBits[ctxIdx][bin] = sbacGetEntropyBits(m_contextState[OFF_SIG_FLAG_CTX + ctxIdx], bin); |
| 1838 | } |
| 1839 | else |
| 1840 | { |
| 1841 | for (uint32_t bin = 0; bin < 2; bin++) |
| 1842 | estBitsSbac.significantBits[0][bin] = sbacGetEntropyBits(m_contextState[OFF_SIG_FLAG_CTX + (NUM_SIG_FLAG_CTX_LUMA + 0)], bin); |
| 1843 | |
| 1844 | for (int ctxIdx = firstCtx; ctxIdx < firstCtx + numCtx; ctxIdx++) |
| 1845 | for (uint32_t bin = 0; bin < 2; bin++) |
| 1846 | estBitsSbac.significantBits[ctxIdx][bin] = sbacGetEntropyBits(m_contextState[OFF_SIG_FLAG_CTX + (NUM_SIG_FLAG_CTX_LUMA + ctxIdx)], bin); |
| 1847 | } |
| 1848 | int bitsX = 0, bitsY = 0; |
| 1849 | |
| 1850 | int blkSizeOffset = bIsLuma ? ((log2TrSize - 2) * 3 + ((log2TrSize - 1) >> 2)) : NUM_CTX_LAST_FLAG_XY_LUMA; |
| 1851 | int ctxShift = bIsLuma ? ((log2TrSize + 1) >> 2) : log2TrSize - 2; |
| 1852 | uint32_t maxGroupIdx = log2TrSize * 2 - 1; |
| 1853 | |
| 1854 | uint32_t ctx; |
| 1855 | const uint8_t *ctxX = &m_contextState[OFF_CTX_LAST_FLAG_X]; |
| 1856 | for (ctx = 0; ctx < maxGroupIdx; ctx++) |
| 1857 | { |
| 1858 | int ctxOffset = blkSizeOffset + (ctx >> ctxShift); |
| 1859 | estBitsSbac.lastXBits[ctx] = bitsX + sbacGetEntropyBits(ctxX[ctxOffset], 0); |
| 1860 | bitsX += sbacGetEntropyBits(ctxX[ctxOffset], 1); |
| 1861 | } |
| 1862 | |
| 1863 | estBitsSbac.lastXBits[ctx] = bitsX; |
| 1864 | |
| 1865 | const uint8_t *ctxY = &m_contextState[OFF_CTX_LAST_FLAG_Y]; |
| 1866 | for (ctx = 0; ctx < maxGroupIdx; ctx++) |
| 1867 | { |
| 1868 | int ctxOffset = blkSizeOffset + (ctx >> ctxShift); |
| 1869 | estBitsSbac.lastYBits[ctx] = bitsY + sbacGetEntropyBits(ctxY[ctxOffset], 0); |
| 1870 | bitsY += sbacGetEntropyBits(ctxY[ctxOffset], 1); |
| 1871 | } |
| 1872 | |
| 1873 | estBitsSbac.lastYBits[ctx] = bitsY; |
| 1874 | } |
| 1875 | |
| 1876 | /* estimate bit cost of significant coefficient */ |
| 1877 | void Entropy::estSignificantCoefficientsBit(EstBitsSbac& estBitsSbac, bool bIsLuma) const |
| 1878 | { |
| 1879 | if (bIsLuma) |
| 1880 | { |
| 1881 | const uint8_t *ctxOne = &m_contextState[OFF_ONE_FLAG_CTX]; |
| 1882 | const uint8_t *ctxAbs = &m_contextState[OFF_ABS_FLAG_CTX]; |
| 1883 | |
| 1884 | for (int ctxIdx = 0; ctxIdx < NUM_ONE_FLAG_CTX_LUMA; ctxIdx++) |
| 1885 | { |
| 1886 | estBitsSbac.greaterOneBits[ctxIdx][0] = sbacGetEntropyBits(ctxOne[ctxIdx], 0); |
| 1887 | estBitsSbac.greaterOneBits[ctxIdx][1] = sbacGetEntropyBits(ctxOne[ctxIdx], 1); |
| 1888 | } |
| 1889 | |
| 1890 | for (int ctxIdx = 0; ctxIdx < NUM_ABS_FLAG_CTX_LUMA; ctxIdx++) |
| 1891 | { |
| 1892 | estBitsSbac.levelAbsBits[ctxIdx][0] = sbacGetEntropyBits(ctxAbs[ctxIdx], 0); |
| 1893 | estBitsSbac.levelAbsBits[ctxIdx][1] = sbacGetEntropyBits(ctxAbs[ctxIdx], 1); |
| 1894 | } |
| 1895 | } |
| 1896 | else |
| 1897 | { |
| 1898 | const uint8_t *ctxOne = &m_contextState[OFF_ONE_FLAG_CTX + NUM_ONE_FLAG_CTX_LUMA]; |
| 1899 | const uint8_t *ctxAbs = &m_contextState[OFF_ABS_FLAG_CTX + NUM_ABS_FLAG_CTX_LUMA]; |
| 1900 | |
| 1901 | for (int ctxIdx = 0; ctxIdx < NUM_ONE_FLAG_CTX_CHROMA; ctxIdx++) |
| 1902 | { |
| 1903 | estBitsSbac.greaterOneBits[ctxIdx][0] = sbacGetEntropyBits(ctxOne[ctxIdx], 0); |
| 1904 | estBitsSbac.greaterOneBits[ctxIdx][1] = sbacGetEntropyBits(ctxOne[ctxIdx], 1); |
| 1905 | } |
| 1906 | |
| 1907 | for (int ctxIdx = 0; ctxIdx < NUM_ABS_FLAG_CTX_CHROMA; ctxIdx++) |
| 1908 | { |
| 1909 | estBitsSbac.levelAbsBits[ctxIdx][0] = sbacGetEntropyBits(ctxAbs[ctxIdx], 0); |
| 1910 | estBitsSbac.levelAbsBits[ctxIdx][1] = sbacGetEntropyBits(ctxAbs[ctxIdx], 1); |
| 1911 | } |
| 1912 | } |
| 1913 | } |
| 1914 | |
| 1915 | /* Initialize our context information from the nominated source */ |
| 1916 | void Entropy::copyContextsFrom(const Entropy& src) |
| 1917 | { |
| 1918 | X265_CHECK(src.m_valid, "invalid copy source context\n"); |
| 1919 | |
| 1920 | memcpy(m_contextState, src.m_contextState, MAX_OFF_CTX_MOD * sizeof(m_contextState[0])); |
| 1921 | markValid(); |
| 1922 | } |
| 1923 | |
| 1924 | void Entropy::start() |
| 1925 | { |
| 1926 | m_low = 0; |
| 1927 | m_range = 510; |
| 1928 | m_bitsLeft = -12; |
| 1929 | m_numBufferedBytes = 0; |
| 1930 | m_bufferedByte = 0xff; |
| 1931 | } |
| 1932 | |
| 1933 | void Entropy::finish() |
| 1934 | { |
| 1935 | if (m_low >> (21 + m_bitsLeft)) |
| 1936 | { |
| 1937 | m_bitIf->writeByte(m_bufferedByte + 1); |
| 1938 | while (m_numBufferedBytes > 1) |
| 1939 | { |
| 1940 | m_bitIf->writeByte(0x00); |
| 1941 | m_numBufferedBytes--; |
| 1942 | } |
| 1943 | |
| 1944 | m_low -= 1 << (21 + m_bitsLeft); |
| 1945 | } |
| 1946 | else |
| 1947 | { |
| 1948 | if (m_numBufferedBytes > 0) |
| 1949 | m_bitIf->writeByte(m_bufferedByte); |
| 1950 | |
| 1951 | while (m_numBufferedBytes > 1) |
| 1952 | { |
| 1953 | m_bitIf->writeByte(0xff); |
| 1954 | m_numBufferedBytes--; |
| 1955 | } |
| 1956 | } |
| 1957 | m_bitIf->write(m_low >> 8, 13 + m_bitsLeft); |
| 1958 | } |
| 1959 | |
| 1960 | void Entropy::copyState(const Entropy& other) |
| 1961 | { |
| 1962 | m_low = other.m_low; |
| 1963 | m_range = other.m_range; |
| 1964 | m_bitsLeft = other.m_bitsLeft; |
| 1965 | m_bufferedByte = other.m_bufferedByte; |
| 1966 | m_numBufferedBytes = other.m_numBufferedBytes; |
| 1967 | m_fracBits = other.m_fracBits; |
| 1968 | } |
| 1969 | |
| 1970 | void Entropy::resetBits() |
| 1971 | { |
| 1972 | m_low = 0; |
| 1973 | m_bitsLeft = -12; |
| 1974 | m_numBufferedBytes = 0; |
| 1975 | m_bufferedByte = 0xff; |
| 1976 | m_fracBits &= 32767; |
| 1977 | if (m_bitIf) |
| 1978 | m_bitIf->resetBits(); |
| 1979 | } |
| 1980 | |
| 1981 | /** Encode bin */ |
| 1982 | void Entropy::encodeBin(uint32_t binValue, uint8_t &ctxModel) |
| 1983 | { |
| 1984 | uint32_t mstate = ctxModel; |
| 1985 | |
| 1986 | ctxModel = sbacNext(mstate, binValue); |
| 1987 | |
| 1988 | if (!m_bitIf) |
| 1989 | { |
| 1990 | m_fracBits += sbacGetEntropyBits(mstate, binValue); |
| 1991 | return; |
| 1992 | } |
| 1993 | |
| 1994 | uint32_t range = m_range; |
| 1995 | uint32_t state = sbacGetState(mstate); |
| 1996 | uint32_t lps = g_lpsTable[state][((uint8_t)range >> 6)]; |
| 1997 | range -= lps; |
| 1998 | |
| 1999 | X265_CHECK(lps >= 2, "lps is too small\n"); |
| 2000 | |
| 2001 | int numBits = (uint32_t)(range - 256) >> 31; |
| 2002 | uint32_t low = m_low; |
| 2003 | |
| 2004 | // NOTE: MPS must be LOWEST bit in mstate |
| 2005 | X265_CHECK((uint32_t)((binValue ^ mstate) & 1) == (uint32_t)(binValue != sbacGetMps(mstate)), "binValue failure\n"); |
| 2006 | if ((binValue ^ mstate) & 1) |
| 2007 | { |
| 2008 | // NOTE: lps is non-zero and the maximum of idx is 8 because lps less than 256 |
| 2009 | //numBits = g_renormTable[lps >> 3]; |
| 2010 | unsigned long idx; |
| 2011 | CLZ32(idx, lps); |
| 2012 | X265_CHECK(state != 63 || idx == 1, "state failure\n"); |
| 2013 | |
| 2014 | numBits = 8 - idx; |
| 2015 | if (state >= 63) |
| 2016 | numBits = 6; |
| 2017 | X265_CHECK(numBits <= 6, "numBits failure\n"); |
| 2018 | |
| 2019 | low += range; |
| 2020 | range = lps; |
| 2021 | } |
| 2022 | m_low = (low << numBits); |
| 2023 | m_range = (range << numBits); |
| 2024 | m_bitsLeft += numBits; |
| 2025 | |
| 2026 | if (m_bitsLeft >= 0) |
| 2027 | writeOut(); |
| 2028 | } |
| 2029 | |
| 2030 | /** Encode equiprobable bin */ |
| 2031 | void Entropy::encodeBinEP(uint32_t binValue) |
| 2032 | { |
| 2033 | if (!m_bitIf) |
| 2034 | { |
| 2035 | m_fracBits += 32768; |
| 2036 | return; |
| 2037 | } |
| 2038 | m_low <<= 1; |
| 2039 | if (binValue) |
| 2040 | m_low += m_range; |
| 2041 | m_bitsLeft++; |
| 2042 | |
| 2043 | if (m_bitsLeft >= 0) |
| 2044 | writeOut(); |
| 2045 | } |
| 2046 | |
| 2047 | /** Encode equiprobable bins */ |
| 2048 | void Entropy::encodeBinsEP(uint32_t binValues, int numBins) |
| 2049 | { |
| 2050 | if (!m_bitIf) |
| 2051 | { |
| 2052 | m_fracBits += 32768 * numBins; |
| 2053 | return; |
| 2054 | } |
| 2055 | |
| 2056 | while (numBins > 8) |
| 2057 | { |
| 2058 | numBins -= 8; |
| 2059 | uint32_t pattern = binValues >> numBins; |
| 2060 | m_low <<= 8; |
| 2061 | m_low += m_range * pattern; |
| 2062 | binValues -= pattern << numBins; |
| 2063 | m_bitsLeft += 8; |
| 2064 | |
| 2065 | if (m_bitsLeft >= 0) |
| 2066 | writeOut(); |
| 2067 | } |
| 2068 | |
| 2069 | m_low <<= numBins; |
| 2070 | m_low += m_range * binValues; |
| 2071 | m_bitsLeft += numBins; |
| 2072 | |
| 2073 | if (m_bitsLeft >= 0) |
| 2074 | writeOut(); |
| 2075 | } |
| 2076 | |
| 2077 | /** Encode terminating bin */ |
| 2078 | void Entropy::encodeBinTrm(uint32_t binValue) |
| 2079 | { |
| 2080 | if (!m_bitIf) |
| 2081 | { |
| 2082 | m_fracBits += sbacGetEntropyBitsTrm(binValue); |
| 2083 | return; |
| 2084 | } |
| 2085 | |
| 2086 | m_range -= 2; |
| 2087 | if (binValue) |
| 2088 | { |
| 2089 | m_low += m_range; |
| 2090 | m_low <<= 7; |
| 2091 | m_range = 2 << 7; |
| 2092 | m_bitsLeft += 7; |
| 2093 | } |
| 2094 | else if (m_range >= 256) |
| 2095 | return; |
| 2096 | else |
| 2097 | { |
| 2098 | m_low <<= 1; |
| 2099 | m_range <<= 1; |
| 2100 | m_bitsLeft++; |
| 2101 | } |
| 2102 | |
| 2103 | if (m_bitsLeft >= 0) |
| 2104 | writeOut(); |
| 2105 | } |
| 2106 | |
| 2107 | /** Move bits from register into bitstream */ |
| 2108 | void Entropy::writeOut() |
| 2109 | { |
| 2110 | uint32_t leadByte = m_low >> (13 + m_bitsLeft); |
| 2111 | uint32_t low_mask = (uint32_t)(~0) >> (11 + 8 - m_bitsLeft); |
| 2112 | |
| 2113 | m_bitsLeft -= 8; |
| 2114 | m_low &= low_mask; |
| 2115 | |
| 2116 | if (leadByte == 0xff) |
| 2117 | m_numBufferedBytes++; |
| 2118 | else |
| 2119 | { |
| 2120 | uint32_t numBufferedBytes = m_numBufferedBytes; |
| 2121 | if (numBufferedBytes > 0) |
| 2122 | { |
| 2123 | uint32_t carry = leadByte >> 8; |
| 2124 | uint32_t byteTowrite = m_bufferedByte + carry; |
| 2125 | m_bitIf->writeByte(byteTowrite); |
| 2126 | |
| 2127 | byteTowrite = (0xff + carry) & 0xff; |
| 2128 | while (numBufferedBytes > 1) |
| 2129 | { |
| 2130 | m_bitIf->writeByte(byteTowrite); |
| 2131 | numBufferedBytes--; |
| 2132 | } |
| 2133 | } |
| 2134 | m_numBufferedBytes = 1; |
| 2135 | m_bufferedByte = (uint8_t)leadByte; |
| 2136 | } |
| 2137 | } |
| 2138 | |
| 2139 | const uint32_t g_entropyBits[128] = |
| 2140 | { |
| 2141 | // Corrected table, most notably for last state |
| 2142 | 0x07b23, 0x085f9, 0x074a0, 0x08cbc, 0x06ee4, 0x09354, 0x067f4, 0x09c1b, 0x060b0, 0x0a62a, 0x05a9c, 0x0af5b, 0x0548d, 0x0b955, 0x04f56, 0x0c2a9, |
| 2143 | 0x04a87, 0x0cbf7, 0x045d6, 0x0d5c3, 0x04144, 0x0e01b, 0x03d88, 0x0e937, 0x039e0, 0x0f2cd, 0x03663, 0x0fc9e, 0x03347, 0x10600, 0x03050, 0x10f95, |
| 2144 | 0x02d4d, 0x11a02, 0x02ad3, 0x12333, 0x0286e, 0x12cad, 0x02604, 0x136df, 0x02425, 0x13f48, 0x021f4, 0x149c4, 0x0203e, 0x1527b, 0x01e4d, 0x15d00, |
| 2145 | 0x01c99, 0x166de, 0x01b18, 0x17017, 0x019a5, 0x17988, 0x01841, 0x18327, 0x016df, 0x18d50, 0x015d9, 0x19547, 0x0147c, 0x1a083, 0x0138e, 0x1a8a3, |
| 2146 | 0x01251, 0x1b418, 0x01166, 0x1bd27, 0x01068, 0x1c77b, 0x00f7f, 0x1d18e, 0x00eda, 0x1d91a, 0x00e19, 0x1e254, 0x00d4f, 0x1ec9a, 0x00c90, 0x1f6e0, |
| 2147 | 0x00c01, 0x1fef8, 0x00b5f, 0x208b1, 0x00ab6, 0x21362, 0x00a15, 0x21e46, 0x00988, 0x2285d, 0x00934, 0x22ea8, 0x008a8, 0x239b2, 0x0081d, 0x24577, |
| 2148 | 0x007c9, 0x24ce6, 0x00763, 0x25663, 0x00710, 0x25e8f, 0x006a0, 0x26a26, 0x00672, 0x26f23, 0x005e8, 0x27ef8, 0x005ba, 0x284b5, 0x0055e, 0x29057, |
| 2149 | 0x0050c, 0x29bab, 0x004c1, 0x2a674, 0x004a7, 0x2aa5e, 0x0046f, 0x2b32f, 0x0041f, 0x2c0ad, 0x003e7, 0x2ca8d, 0x003ba, 0x2d323, 0x0010c, 0x3bfbb |
| 2150 | }; |
| 2151 | |
| 2152 | const uint8_t g_nextState[128][2] = |
| 2153 | { |
| 2154 | { 2, 1 }, { 0, 3 }, { 4, 0 }, { 1, 5 }, { 6, 2 }, { 3, 7 }, { 8, 4 }, { 5, 9 }, |
| 2155 | { 10, 4 }, { 5, 11 }, { 12, 8 }, { 9, 13 }, { 14, 8 }, { 9, 15 }, { 16, 10 }, { 11, 17 }, |
| 2156 | { 18, 12 }, { 13, 19 }, { 20, 14 }, { 15, 21 }, { 22, 16 }, { 17, 23 }, { 24, 18 }, { 19, 25 }, |
| 2157 | { 26, 18 }, { 19, 27 }, { 28, 22 }, { 23, 29 }, { 30, 22 }, { 23, 31 }, { 32, 24 }, { 25, 33 }, |
| 2158 | { 34, 26 }, { 27, 35 }, { 36, 26 }, { 27, 37 }, { 38, 30 }, { 31, 39 }, { 40, 30 }, { 31, 41 }, |
| 2159 | { 42, 32 }, { 33, 43 }, { 44, 32 }, { 33, 45 }, { 46, 36 }, { 37, 47 }, { 48, 36 }, { 37, 49 }, |
| 2160 | { 50, 38 }, { 39, 51 }, { 52, 38 }, { 39, 53 }, { 54, 42 }, { 43, 55 }, { 56, 42 }, { 43, 57 }, |
| 2161 | { 58, 44 }, { 45, 59 }, { 60, 44 }, { 45, 61 }, { 62, 46 }, { 47, 63 }, { 64, 48 }, { 49, 65 }, |
| 2162 | { 66, 48 }, { 49, 67 }, { 68, 50 }, { 51, 69 }, { 70, 52 }, { 53, 71 }, { 72, 52 }, { 53, 73 }, |
| 2163 | { 74, 54 }, { 55, 75 }, { 76, 54 }, { 55, 77 }, { 78, 56 }, { 57, 79 }, { 80, 58 }, { 59, 81 }, |
| 2164 | { 82, 58 }, { 59, 83 }, { 84, 60 }, { 61, 85 }, { 86, 60 }, { 61, 87 }, { 88, 60 }, { 61, 89 }, |
| 2165 | { 90, 62 }, { 63, 91 }, { 92, 64 }, { 65, 93 }, { 94, 64 }, { 65, 95 }, { 96, 66 }, { 67, 97 }, |
| 2166 | { 98, 66 }, { 67, 99 }, { 100, 66 }, { 67, 101 }, { 102, 68 }, { 69, 103 }, { 104, 68 }, { 69, 105 }, |
| 2167 | { 106, 70 }, { 71, 107 }, { 108, 70 }, { 71, 109 }, { 110, 70 }, { 71, 111 }, { 112, 72 }, { 73, 113 }, |
| 2168 | { 114, 72 }, { 73, 115 }, { 116, 72 }, { 73, 117 }, { 118, 74 }, { 75, 119 }, { 120, 74 }, { 75, 121 }, |
| 2169 | { 122, 74 }, { 75, 123 }, { 124, 76 }, { 77, 125 }, { 124, 76 }, { 77, 125 }, { 126, 126 }, { 127, 127 } |
| 2170 | }; |
| 2171 | |
| 2172 | } |