/work/vvdec/source/Lib/CommonLib/Quant.cpp
Line | Count | Source |
1 | | /* ----------------------------------------------------------------------------- |
2 | | The copyright in this software is being made available under the Clear BSD |
3 | | License, included below. No patent rights, trademark rights and/or |
4 | | other Intellectual Property Rights other than the copyrights concerning |
5 | | the Software are granted under this license. |
6 | | |
7 | | The Clear BSD License |
8 | | |
9 | | Copyright (c) 2018-2026, Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. & The VVdeC Authors. |
10 | | All rights reserved. |
11 | | |
12 | | Redistribution and use in source and binary forms, with or without modification, |
13 | | are permitted (subject to the limitations in the disclaimer below) provided that |
14 | | the following conditions are met: |
15 | | |
16 | | * Redistributions of source code must retain the above copyright notice, |
17 | | this list of conditions and the following disclaimer. |
18 | | |
19 | | * Redistributions in binary form must reproduce the above copyright |
20 | | notice, this list of conditions and the following disclaimer in the |
21 | | documentation and/or other materials provided with the distribution. |
22 | | |
23 | | * Neither the name of the copyright holder nor the names of its |
24 | | contributors may be used to endorse or promote products derived from this |
25 | | software without specific prior written permission. |
26 | | |
27 | | NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE GRANTED BY |
28 | | THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND |
29 | | CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
30 | | LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A |
31 | | PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR |
32 | | CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
33 | | EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
34 | | PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR |
35 | | BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER |
36 | | IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
37 | | ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
38 | | POSSIBILITY OF SUCH DAMAGE. |
39 | | |
40 | | |
41 | | ------------------------------------------------------------------------------------------- */ |
42 | | |
43 | | /** \file Quant.cpp |
44 | | \brief transform and quantization class |
45 | | */ |
46 | | |
47 | | #include "Quant.h" |
48 | | |
49 | | #include "UnitTools.h" |
50 | | #include "ContextModelling.h" |
51 | | #include "CodingStructure.h" |
52 | | |
53 | | #include "dtrace_buffer.h" |
54 | | |
55 | | #include <stdlib.h> |
56 | | #include <memory.h> |
57 | | |
58 | | namespace vvdec |
59 | | { |
60 | | |
61 | | // ==================================================================================================================== |
62 | | // QpParam constructor |
63 | | // ==================================================================================================================== |
64 | | |
65 | | QpParam::QpParam( const TransformUnit& tu, const ComponentID& compID, const bool allowACTQpoffset ) |
66 | 0 | { |
67 | 0 | const SPS& sps = *tu.cu->sps; |
68 | 0 | const int qpBdOffset = sps.getQpBDOffset(); |
69 | 0 | const bool useJQP = isChroma( compID ) && TU::getICTMode( tu, false ) == 2; |
70 | 0 | const ComponentID jCbCr = useJQP ? JOINT_CbCr : compID; |
71 | |
|
72 | 0 | int baseQp; |
73 | 0 | int qpy = tu.cu->qp; |
74 | | //bool skip = tu.mtsIdx[compID] == MTS_SKIP; |
75 | |
|
76 | 0 | if( isLuma( compID ) ) |
77 | 0 | { |
78 | 0 | baseQp = qpy + qpBdOffset; |
79 | 0 | } |
80 | 0 | else |
81 | 0 | { |
82 | 0 | const PPS &pps = *tu.cu->pps; |
83 | 0 | int |
84 | 0 | chromaQpOffset = pps.getQpOffset ( jCbCr ); |
85 | 0 | chromaQpOffset += tu.cu->slice->getSliceChromaQpDelta( jCbCr ); |
86 | 0 | chromaQpOffset += pps.getChromaQpOffsetListEntry( tu.cu->chromaQpAdj ).get( jCbCr ); |
87 | |
|
88 | 0 | int qpi = Clip3( -qpBdOffset, MAX_QP, qpy ); |
89 | 0 | baseQp = sps.getMappedChromaQpValue( jCbCr, qpi ); |
90 | 0 | baseQp = Clip3( 0, MAX_QP + qpBdOffset, baseQp + chromaQpOffset + qpBdOffset ); |
91 | 0 | } |
92 | |
|
93 | 0 | if( allowACTQpoffset && tu.cu->colorTransform() ) |
94 | 0 | { |
95 | 0 | baseQp += DELTA_QP_ACT[jCbCr]; |
96 | 0 | baseQp = Clip3( 0, MAX_QP + qpBdOffset, baseQp ); |
97 | 0 | } |
98 | | |
99 | | // TODO: ensure clip not needed for non-ACT |
100 | | |
101 | | //if( !skip ) |
102 | 0 | { |
103 | 0 | Qps [0] = baseQp; |
104 | 0 | pers[0] = baseQp / 6; |
105 | 0 | rems[0] = baseQp - ( pers[0] << 2 ) - ( pers[0] << 1 ); |
106 | 0 | } |
107 | | //else |
108 | 0 | { |
109 | 0 | int internalMinusInputBitDepth = sps.getInternalMinusInputBitDepth(); |
110 | 0 | int baseQpTS = std::max( baseQp, 4 + 6 * internalMinusInputBitDepth ); |
111 | 0 | Qps [1] = baseQpTS; |
112 | 0 | pers[1] = baseQpTS / 6; |
113 | 0 | rems[1] = baseQpTS - ( pers[1] << 2 ) - ( pers[1] << 1 );; |
114 | 0 | } |
115 | 0 | } |
116 | | |
117 | | // ==================================================================================================================== |
118 | | // Quant class member functions |
119 | | // ==================================================================================================================== |
120 | | |
121 | | template<class T, bool UseScalingList> |
122 | | static void DeQuantImpl( const SizeType width, |
123 | | const int maxX, |
124 | | const int maxY, |
125 | | const int scaleQP, |
126 | | const int* piDequantCoef, // unused if UseScalingList == false |
127 | | const T* const piQCoef, |
128 | | const size_t piQCfStride, |
129 | | TCoeff* const piCoef, |
130 | | const int rightShift, |
131 | | const int inputMaximum, |
132 | | const TCoeff transformMaximum ) |
133 | 0 | { |
134 | 0 | const int inputMinimum = -( inputMaximum + 1 ); |
135 | 0 | const TCoeff transformMinimum = -( transformMaximum + 1 ); |
136 | |
|
137 | 0 | if( rightShift > 0 ) |
138 | 0 | { |
139 | 0 | const Intermediate_Int iAdd = (Intermediate_Int) 1 << ( rightShift - 1 ); |
140 | 0 | for( int y = 0; y <= maxY; y++ ) |
141 | 0 | { |
142 | 0 | int n = y * width; |
143 | 0 | for( int x = 0; x <= maxX; x++, n++ ) |
144 | 0 | { |
145 | 0 | const TCoeff level = piQCoef[x + y * piQCfStride]; |
146 | 0 | if( level ) |
147 | 0 | { |
148 | 0 | const int scale = UseScalingList ? piDequantCoef[n] * scaleQP // |
149 | 0 | : scaleQP; |
150 | 0 | const TCoeff clipQCoef = TCoeff( Clip3<Intermediate_Int>( inputMinimum, inputMaximum, level ) ); |
151 | 0 | Intermediate_Int iCoeffQ = ( Intermediate_Int( clipQCoef ) * scale + iAdd ) >> rightShift; |
152 | |
|
153 | 0 | piCoef[n] = TCoeff( Clip3<Intermediate_Int>( transformMinimum, transformMaximum, iCoeffQ ) ); |
154 | 0 | } |
155 | 0 | } |
156 | 0 | } |
157 | 0 | } |
158 | 0 | else // rightshift <= 0 |
159 | 0 | { |
160 | 0 | const int leftShift = -rightShift; |
161 | 0 | for( int y = 0; y <= maxY; y++ ) |
162 | 0 | { |
163 | 0 | int n = y * width; |
164 | 0 | for( int x = 0; x <= maxX; x++, n++ ) |
165 | 0 | { |
166 | 0 | const TCoeff level = piQCoef[x + y * piQCfStride]; |
167 | 0 | if( level ) |
168 | 0 | { |
169 | 0 | const int scale = UseScalingList ? piDequantCoef[n] * scaleQP // |
170 | 0 | : scaleQP; |
171 | 0 | const TCoeff clipQCoef = TCoeff( Clip3<Intermediate_Int>( inputMinimum, inputMaximum, level ) ); |
172 | 0 | const Intermediate_Int iCoeffQ = ( Intermediate_Int( clipQCoef ) * scale ) * ( 1 << leftShift ); |
173 | |
|
174 | 0 | piCoef[n] = TCoeff( Clip3<Intermediate_Int>( transformMinimum, transformMaximum, iCoeffQ ) ); |
175 | 0 | } |
176 | 0 | } |
177 | 0 | } |
178 | 0 | } |
179 | 0 | } Unexecuted instantiation: Quant.cpp:void vvdec::DeQuantImpl<short, false>(unsigned int, int, int, int, int const*, short const*, unsigned long, int*, int, int, int) Unexecuted instantiation: Quant.cpp:void vvdec::DeQuantImpl<int, false>(unsigned int, int, int, int, int const*, int const*, unsigned long, int*, int, int, int) Unexecuted instantiation: Quant.cpp:void vvdec::DeQuantImpl<short, true>(unsigned int, int, int, int, int const*, short const*, unsigned long, int*, int, int, int) Unexecuted instantiation: Quant.cpp:void vvdec::DeQuantImpl<int, true>(unsigned int, int, int, int, int const*, int const*, unsigned long, int*, int, int, int) |
180 | | |
181 | | template<class T> |
182 | | void Quant::DeQuantScalingCore( const SizeType width, |
183 | | const int maxX, |
184 | | const int maxY, |
185 | | const int scaleQP, |
186 | | const int* piDequantCoef, |
187 | | const T* const piQCoef, |
188 | | const size_t piQCfStride, |
189 | | TCoeff* const piCoef, |
190 | | const int rightShift, |
191 | | const int inputMaximum, |
192 | | const TCoeff transformMaximum ) |
193 | 0 | { |
194 | 0 | DeQuantImpl<T, true>( width, maxX, maxY, scaleQP, piDequantCoef, piQCoef, piQCfStride, piCoef, rightShift, inputMaximum, transformMaximum ); |
195 | 0 | } Unexecuted instantiation: void vvdec::Quant::DeQuantScalingCore<short>(unsigned int, int, int, int, int const*, short const*, unsigned long, int*, int, int, int) Unexecuted instantiation: void vvdec::Quant::DeQuantScalingCore<int>(unsigned int, int, int, int, int const*, int const*, unsigned long, int*, int, int, int) |
196 | | |
197 | | template<class T> |
198 | | void Quant::DeQuantCore( const SizeType width, |
199 | | const int maxX, |
200 | | const int maxY, |
201 | | const int scale, |
202 | | const T* const piQCoef, |
203 | | const size_t piQCfStride, |
204 | | TCoeff* const piCoef, |
205 | | const int rightShift, |
206 | | const int inputMaximum, |
207 | | const TCoeff transformMaximum ) |
208 | 0 | { |
209 | 0 | DeQuantImpl<T, false>( width, maxX, maxY, scale, nullptr, piQCoef, piQCfStride, piCoef, rightShift, inputMaximum, transformMaximum ); |
210 | 0 | } Unexecuted instantiation: void vvdec::Quant::DeQuantCore<short>(unsigned int, int, int, int, short const*, unsigned long, int*, int, int, int) Unexecuted instantiation: void vvdec::Quant::DeQuantCore<int>(unsigned int, int, int, int, int const*, unsigned long, int*, int, int, int) |
211 | | |
212 | | Quant::Quant( const Quant* other, bool enableOpt ) |
213 | 22.7k | : m_dequantCoefBuf( nullptr ) |
214 | 22.7k | , m_ownDequantCoeff( false ) |
215 | 22.7k | { |
216 | 22.7k | xInitScalingList( other ); |
217 | | |
218 | 22.7k | DeQuant = DeQuantCore<TCoeffSig>; |
219 | 22.7k | DeQuantPCM = DeQuantCore<TCoeff>; |
220 | 22.7k | DeQuantScaling = DeQuantScalingCore<TCoeffSig>; |
221 | 22.7k | DeQuantScalingPCM = DeQuantScalingCore<TCoeff>; |
222 | | |
223 | 22.7k | if( enableOpt ) |
224 | 22.7k | { |
225 | | #if ENABLE_SIMD_OPT_QUANT && defined( TARGET_SIMD_X86 ) |
226 | | initQuantX86(); |
227 | | #endif |
228 | 22.7k | } |
229 | 22.7k | } |
230 | | |
231 | | Quant::~Quant() |
232 | 22.7k | { |
233 | 22.7k | xDestroyScalingList(); |
234 | 22.7k | } |
235 | | |
236 | | void invResDPCM( const TransformUnit &tu, const ComponentID &compID, CoeffBuf &dstBuf ) |
237 | 0 | { |
238 | 0 | const CompArea& rect = tu.blocks[compID]; |
239 | 0 | const int wdt = rect.width; |
240 | 0 | const int hgt = rect.height; |
241 | 0 | const CCoeffSigBuf coeffs = tu.cu->cs->getRecoBuf( tu.block( compID ) ); |
242 | |
|
243 | 0 | const int maxLog2TrDynamicRange = tu.cu->sps->getMaxLog2TrDynamicRange(toChannelType(compID)); |
244 | 0 | const TCoeff inputMinimum = -(1 << maxLog2TrDynamicRange); |
245 | 0 | const TCoeff inputMaximum = (1 << maxLog2TrDynamicRange) - 1; |
246 | |
|
247 | 0 | const TCoeffSig* coef = &coeffs.buf[0]; |
248 | 0 | TCoeff* dst = &dstBuf.buf[0]; |
249 | |
|
250 | 0 | if( isLuma( compID ) ? tu.cu->bdpcmMode() == 1 : tu.cu->bdpcmModeChroma() == 1 ) |
251 | 0 | { |
252 | 0 | for( int y = 0; y < hgt; y++ ) |
253 | 0 | { |
254 | 0 | dst[0] = coef[0]; |
255 | 0 | for( int x = 1; x < wdt; x++ ) |
256 | 0 | { |
257 | 0 | dst[x] = Clip3(inputMinimum, inputMaximum, dst[x - 1] + coef[x]); |
258 | 0 | } |
259 | 0 | coef += coeffs.stride; |
260 | 0 | dst += dstBuf.stride; |
261 | 0 | } |
262 | 0 | } |
263 | 0 | else |
264 | 0 | { |
265 | 0 | for( int x = 0; x < wdt; x++ ) |
266 | 0 | { |
267 | 0 | dst[x] = coef[x]; |
268 | 0 | } |
269 | 0 | for( int y = 0; y < hgt - 1; y++ ) |
270 | 0 | { |
271 | 0 | for( int x = 0; x < wdt; x++ ) |
272 | 0 | { |
273 | 0 | dst[dstBuf.stride + x] = Clip3(inputMinimum, inputMaximum, dst[x] + coef[coeffs.stride + x]); |
274 | 0 | } |
275 | 0 | coef += coeffs.stride; |
276 | 0 | dst += dstBuf.stride; |
277 | 0 | } |
278 | 0 | } |
279 | 0 | } |
280 | | |
281 | | static inline int getTransformShift( const int channelBitDepth, const Size size, const int maxLog2TrDynamicRange ) |
282 | 0 | { |
283 | 0 | return maxLog2TrDynamicRange - channelBitDepth - ( ( getLog2( size.width ) + getLog2( size.height ) ) >> 1 ); |
284 | 0 | } |
285 | | |
286 | | static inline int getScalingListType( const PredMode predMode, const ComponentID compID ) |
287 | 0 | { |
288 | 0 | return ( predMode == MODE_INTRA ? 0 : MAX_NUM_COMPONENT ) + compID; |
289 | 0 | } |
290 | | |
291 | | |
292 | | void Quant::dequant( const TransformUnit& tu, CoeffBuf& dstCoeff, const ComponentID& compID, const QpParam& cQP ) |
293 | 0 | { |
294 | 0 | const SPS* sps = tu.cu->sps; |
295 | 0 | const CompArea& area = tu.blocks[compID]; |
296 | 0 | const CCoeffSigBuf coeffBuf = tu.cu->cs->getRecoBuf( tu.block( compID ) ); |
297 | 0 | const TCoeffSig* const piQCoef = coeffBuf.buf; |
298 | 0 | const size_t piQCfStride = coeffBuf.stride; |
299 | 0 | TCoeff* const piCoef = dstCoeff.buf; |
300 | 0 | const int maxLog2TrDynamicRange = sps->getMaxLog2TrDynamicRange( toChannelType( compID ) ); |
301 | 0 | const TCoeff transformMaximum = ( 1 << maxLog2TrDynamicRange ) - 1; |
302 | 0 | const bool isTransformSkip = ( tu.mtsIdx( compID ) == MTS_SKIP ); |
303 | 0 | setUseScalingList( tu.cu->slice->getExplicitScalingListUsed() ); |
304 | 0 | const bool disableSMForLFNST = tu.cu->slice->getExplicitScalingListUsed() ? sps->getDisableScalingMatrixForLfnstBlks() : false; |
305 | 0 | const bool isLfnstApplied = tu.cu->lfnstIdx() > 0 && ( CU::isSepTree( *tu.cu ) ? true : isLuma( compID ) ); |
306 | 0 | const bool disableSMForACT = tu.cu->sps->getScalingMatrixForAlternativeColourSpaceDisabledFlag() && tu.cu->sps->getScalingMatrixDesignatedColourSpaceFlag() == tu.cu->colorTransform(); |
307 | 0 | const bool enableScalingLists = getUseScalingList( isTransformSkip, isLfnstApplied, disableSMForLFNST, disableSMForACT ); |
308 | 0 | const int scalingListType = getScalingListType( tu.cu->predMode(), compID ); |
309 | 0 | const int channelBitDepth = sps->getBitDepth(); |
310 | |
|
311 | 0 | int maxX, maxY; |
312 | |
|
313 | 0 | if( ( tu.cu->bdpcmMode() && isLuma(compID) ) || ( tu.cu->bdpcmModeChroma() && isChroma(compID) ) ) |
314 | 0 | { |
315 | 0 | invResDPCM( tu, compID, dstCoeff ); |
316 | 0 | maxX = area.width - 1; |
317 | 0 | maxY = area.height - 1; |
318 | 0 | } |
319 | 0 | else |
320 | 0 | { |
321 | 0 | maxX = tu.maxScanPosX[compID]; |
322 | 0 | maxY = tu.maxScanPosY[compID]; |
323 | 0 | } |
324 | |
|
325 | 0 | CHECK(scalingListType >= SCALING_LIST_NUM, "Invalid scaling list"); |
326 | | |
327 | | // Represents scaling through forward transform |
328 | 0 | const bool bClipTransformShiftTo0 = false; // tu.mtsIdx[compID] != 1 && sps->getSpsRangeExtension().getExtendedPrecisionProcessingFlag(); |
329 | 0 | const int originalTransformShift = getTransformShift( channelBitDepth, area.size(), maxLog2TrDynamicRange ); |
330 | 0 | const bool needSqrtAdjustment = TU::needsBlockSizeTrafoScale( tu, compID ); |
331 | 0 | const int iTransformShift = ( bClipTransformShiftTo0 ? std::max<int>( 0, originalTransformShift ) : originalTransformShift ) // |
332 | 0 | + ( needSqrtAdjustment ? -1 : 0 ); |
333 | 0 | const bool depQuant = tu.cu->slice->getDepQuantEnabledFlag() && ( tu.mtsIdx( compID ) != MTS_SKIP ); |
334 | 0 | const int QP_per = depQuant ? ( ( cQP.Qp( isTransformSkip ) + 1 ) / 6 ) : cQP.per( isTransformSkip ); |
335 | 0 | const int QP_rem = depQuant ? ( cQP.Qp( isTransformSkip ) + 1 - 6 * QP_per ) : cQP.rem( isTransformSkip ); |
336 | 0 | const int rightShift = IQUANT_SHIFT + ( depQuant ? 1 : 0 ) // |
337 | 0 | - ( ( isTransformSkip ? 0 : iTransformShift ) + QP_per ) // |
338 | 0 | + ( enableScalingLists ? LOG2_SCALING_LIST_NEUTRAL_VALUE : 0 ); |
339 | |
|
340 | 0 | const int scaleQP = g_InvQuantScales[needSqrtAdjustment ? 1 : 0][QP_rem]; |
341 | 0 | const int scaleBits = ( IQUANT_SHIFT + 1 ); |
342 | | |
343 | | // from the dequantisation equation: |
344 | | // iCoeffQ = Intermediate_Int((int64_t(clipQCoef) * scale + iAdd) >> rightShift); |
345 | | //(sizeof(Intermediate_Int) * 8) = inputBitDepth + scaleBits - rightShift |
346 | 0 | const uint32_t targetInputBitDepth = std::min<uint32_t>( maxLog2TrDynamicRange + 1, sizeof( Intermediate_Int ) * 8 + rightShift - scaleBits ); |
347 | 0 | const Intermediate_Int inputMaximum = ( 1 << ( targetInputBitDepth - 1 ) ) - 1; |
348 | |
|
349 | 0 | if( !enableScalingLists ) |
350 | 0 | { |
351 | 0 | if( ( tu.cu->bdpcmMode() && isLuma( compID ) ) || ( tu.cu->bdpcmModeChroma() && isChroma( compID ) ) ) |
352 | 0 | { |
353 | 0 | TCoeff* dst = &dstCoeff.buf[0]; |
354 | 0 | DeQuantPCM( area.width, maxX, maxY, scaleQP, dst, dstCoeff.stride, piCoef, rightShift, inputMaximum, transformMaximum ); |
355 | 0 | } |
356 | 0 | else |
357 | 0 | { |
358 | 0 | DeQuant( area.width, maxX, maxY, scaleQP, piQCoef, piQCfStride, piCoef, rightShift, inputMaximum, transformMaximum ); |
359 | 0 | } |
360 | 0 | } |
361 | 0 | else // Scaling Lists |
362 | 0 | { |
363 | 0 | const uint32_t uiLog2TrWidth = getLog2( area.width ); |
364 | 0 | const uint32_t uiLog2TrHeight = getLog2( area.height ); |
365 | 0 | const int* piDequantCoef = getDequantCoeff( scalingListType, uiLog2TrWidth, uiLog2TrHeight ); |
366 | 0 | if( ( tu.cu->bdpcmMode() && isLuma( compID ) ) || ( tu.cu->bdpcmModeChroma() && isChroma( compID ) ) ) |
367 | 0 | { |
368 | 0 | TCoeff* dst = &dstCoeff.buf[0]; |
369 | 0 | DeQuantScalingPCM( area.width, maxX, maxY, scaleQP, piDequantCoef, dst, dstCoeff.stride, piCoef, rightShift, inputMaximum, transformMaximum ); |
370 | 0 | } |
371 | 0 | else |
372 | 0 | { |
373 | 0 | DeQuantScaling( area.width, maxX, maxY, scaleQP, piDequantCoef, piQCoef, piQCfStride, piCoef, rightShift, inputMaximum, transformMaximum ); |
374 | 0 | } |
375 | 0 | } |
376 | 0 | } |
377 | | |
378 | | /** set quantized matrix coefficient for decode |
379 | | * \param scalingList quantized matrix address |
380 | | * \param format chroma format |
381 | | */ |
382 | | void Quant::setScalingListDec( const ScalingList& scalingList ) |
383 | 0 | { |
384 | 0 | int scalingListId = 0; |
385 | 0 | int recScalingListId = 0; |
386 | 0 | bool anyChange = false; |
387 | 0 | for( uint32_t size = SCALING_LIST_FIRST_CODED; size <= SCALING_LIST_LAST_CODED; size++ ) |
388 | 0 | { |
389 | 0 | for( uint32_t list = 0; list < SCALING_LIST_NUM; list++ ) |
390 | 0 | { |
391 | 0 | if( size == SCALING_LIST_2x2 && list < 4 ) // skip 2x2 luma |
392 | 0 | { |
393 | 0 | continue; |
394 | 0 | } |
395 | 0 | scalingListId = g_scalingListId[size][list]; |
396 | 0 | anyChange |= xSetScalingListDec( scalingList, list, size, scalingListId ); |
397 | 0 | } |
398 | 0 | } |
399 | 0 | if( !anyChange ) return; |
400 | | // based on square result and apply downsample technology |
401 | 0 | for( uint32_t sizew = 0; sizew <= SCALING_LIST_LAST_CODED; sizew++ ) // 7 |
402 | 0 | { |
403 | 0 | for( uint32_t sizeh = 0; sizeh <= SCALING_LIST_LAST_CODED; sizeh++ ) // 7 |
404 | 0 | { |
405 | 0 | if( sizew == sizeh || ( sizew == SCALING_LIST_1x1 && sizeh < SCALING_LIST_4x4 ) || ( sizeh == SCALING_LIST_1x1 && sizew < SCALING_LIST_4x4 ) ) |
406 | 0 | { |
407 | 0 | continue; |
408 | 0 | } |
409 | 0 | for( uint32_t list = 0; list < SCALING_LIST_NUM; list++ ) // 9 |
410 | 0 | { |
411 | 0 | int largerSide = ( sizew > sizeh ) ? sizew : sizeh; |
412 | 0 | CHECK( largerSide < SCALING_LIST_4x4, "Rectangle Error!" ); |
413 | 0 | recScalingListId = g_scalingListId[largerSide][list]; |
414 | 0 | xSetRecScalingListDec( scalingList, list, sizew, sizeh, recScalingListId ); |
415 | 0 | } |
416 | 0 | } |
417 | 0 | } |
418 | 0 | } |
419 | | |
420 | | /** set quantized matrix coefficient for decode |
421 | | * \param scalingList quantaized matrix address |
422 | | * \param listId List index |
423 | | * \param sizeId size index |
424 | | * \param qp Quantization parameter |
425 | | * \param format chroma format |
426 | | */ |
427 | | bool Quant::xSetScalingListDec(const ScalingList &scalingList, uint32_t listId, uint32_t sizeId, uint32_t scalingListId) |
428 | 0 | { |
429 | 0 | const uint32_t width = g_vvcScalingListSizeX[sizeId]; |
430 | 0 | const uint32_t height = g_vvcScalingListSizeX[sizeId]; |
431 | | #if defined( __SANITIZE_ADDRESS__ ) // work around a bug in GCC address-sanitizer, when building with -fsanitize=address, but without -fsanitize=undefined |
432 | | volatile |
433 | | #endif |
434 | 0 | const uint32_t ratio = g_vvcScalingListSizeX[sizeId]/std::min(MAX_MATRIX_SIZE_NUM,(int)g_vvcScalingListSizeX[sizeId]); |
435 | |
|
436 | 0 | const int *coeff = scalingList.getScalingListAddress(scalingListId); |
437 | 0 | int *dequantcoeff = getDequantCoeff(listId, sizeId, sizeId); |
438 | |
|
439 | 0 | return |
440 | 0 | processScalingListDec(coeff, |
441 | 0 | dequantcoeff, |
442 | 0 | height, width, ratio, |
443 | 0 | std::min(MAX_MATRIX_SIZE_NUM, (int)g_vvcScalingListSizeX[sizeId]), |
444 | 0 | scalingList.getScalingListDC(scalingListId)); |
445 | 0 | } |
446 | | |
447 | | /** set quantized matrix coefficient for decode |
448 | | * \param scalingList quantaized matrix address |
449 | | * \param listId List index |
450 | | * \param sizeId size index |
451 | | * \param qp Quantization parameter |
452 | | * \param format chroma format |
453 | | */ |
454 | | void Quant::xSetRecScalingListDec(const ScalingList &scalingList, uint32_t listId, uint32_t sizeIdw, uint32_t sizeIdh, uint32_t scalingListId) |
455 | 0 | { |
456 | 0 | if (sizeIdw == sizeIdh) return; |
457 | | |
458 | 0 | const uint32_t width = g_vvcScalingListSizeX[sizeIdw]; |
459 | 0 | const uint32_t height = g_vvcScalingListSizeX[sizeIdh]; |
460 | 0 | const uint32_t largeSideId = (sizeIdw > sizeIdh) ? sizeIdw : sizeIdh; //16 |
461 | |
|
462 | 0 | const int *coeff = scalingList.getScalingListAddress(scalingListId); |
463 | 0 | int *dequantcoeff = getDequantCoeff(listId, sizeIdw, sizeIdh); |
464 | |
|
465 | 0 | processScalingListDec(coeff, |
466 | 0 | dequantcoeff, |
467 | 0 | height, width, (largeSideId>3) ? 2 : 1, |
468 | 0 | (largeSideId >= 3 ? 8 : 4), |
469 | 0 | scalingList.getScalingListDC(scalingListId)); |
470 | 0 | } |
471 | | |
472 | | /** set quantized matrix coefficient for decode |
473 | | * \param coeff quantaized matrix address |
474 | | * \param dequantcoeff quantaized matrix address |
475 | | * \param invQuantScales IQ(QP%6)) |
476 | | * \param height height |
477 | | * \param width width |
478 | | * \param ratio ratio for upscale |
479 | | * \param sizuNum matrix size |
480 | | * \param dc dc parameter |
481 | | */ |
482 | | bool Quant::processScalingListDec( const int *coeff, int *dequantcoeff, uint32_t height, uint32_t width, uint32_t ratio, int sizuNum, uint32_t dc) |
483 | 0 | { |
484 | 0 | if (height != width) |
485 | 0 | { |
486 | 0 | const int hl2 = getLog2( height ); |
487 | 0 | const int wl2 = getLog2( width ); |
488 | 0 | const int sl2 = getLog2( sizuNum ); |
489 | |
|
490 | 0 | const int loopH = std::min<int>( height, JVET_C0024_ZERO_OUT_TH ); |
491 | 0 | const int loopW = std::min<int>( width, JVET_C0024_ZERO_OUT_TH ); |
492 | |
|
493 | 0 | const int ratioWH = height > width ? hl2 - wl2 : wl2 - hl2; |
494 | 0 | const int ratioH = height / sizuNum ? hl2 - sl2 : sl2 - hl2; |
495 | 0 | const int ratioW = width / sizuNum ? wl2 - sl2 : sl2 - wl2; |
496 | |
|
497 | 0 | if( height > width ) |
498 | 0 | { |
499 | 0 | for( uint32_t j = 0; j < loopH; j += ( 1 << ratioH ) ) |
500 | 0 | { |
501 | 0 | for( uint32_t i = 0; i < loopW; i++ ) |
502 | 0 | { |
503 | 0 | dequantcoeff[j * width + i] = coeff[sizuNum * ( j >> ratioH ) + ( ( i << ratioWH ) >> ratioH )]; |
504 | 0 | } |
505 | |
|
506 | 0 | const int* src = &dequantcoeff[j * width]; |
507 | 0 | for( int jj = 1; jj < ( 1 << ratioH ); jj++ ) |
508 | 0 | { |
509 | 0 | memcpy( &dequantcoeff[( j + jj ) * width], src, loopW * sizeof( int ) ); |
510 | 0 | } |
511 | 0 | } |
512 | 0 | } |
513 | 0 | else |
514 | 0 | { |
515 | 0 | for( uint32_t j = 0; j < loopH; j++ ) |
516 | 0 | { |
517 | 0 | for( uint32_t i = 0; i < loopW; i += ( 1 << ratioW ) ) |
518 | 0 | { |
519 | 0 | const int coeffi = coeff[sizuNum * ( ( j << ratioWH ) >> ratioW ) + ( i >> ratioW )]; |
520 | 0 | for( uint32_t ii = 0; ii < ( 1 << ratioW ); ii++ ) |
521 | 0 | { |
522 | 0 | dequantcoeff[j * width + i + ii] = coeffi; |
523 | 0 | } |
524 | 0 | } |
525 | 0 | } |
526 | 0 | } |
527 | |
|
528 | 0 | const int largeOne = std::max( width, height ); |
529 | 0 | if( largeOne > 8 ) |
530 | 0 | { |
531 | 0 | dequantcoeff[0] = dc; |
532 | 0 | } |
533 | 0 | return true; |
534 | 0 | } |
535 | | |
536 | 0 | bool anyChange = false; |
537 | |
|
538 | 0 | const int rl2 = getLog2( ratio ); |
539 | 0 | const int loopH = std::min<int>( height, JVET_C0024_ZERO_OUT_TH ); |
540 | 0 | const int loopW = std::min<int>( width, JVET_C0024_ZERO_OUT_TH ); |
541 | |
|
542 | 0 | for( uint32_t j = 0; j < loopH; j += ( 1 << rl2 ) ) |
543 | 0 | { |
544 | 0 | for( uint32_t i = 0; i < loopW; i += ( 1 << rl2 ) ) |
545 | 0 | { |
546 | 0 | const int coeffi = coeff[sizuNum * ( j >> rl2 ) + ( i >> rl2 )]; |
547 | 0 | anyChange |= coeffi != dequantcoeff[j * width + i]; |
548 | 0 | for( uint32_t ii = 0; anyChange && ii < ( 1 << rl2 ); ii++ ) |
549 | 0 | { |
550 | 0 | dequantcoeff[j * width + i + ii] = coeffi; |
551 | 0 | } |
552 | 0 | } |
553 | |
|
554 | 0 | const int* src = &dequantcoeff[j * width]; |
555 | 0 | for( int jj = 1; jj < ( 1 << rl2 ); jj++ ) |
556 | 0 | { |
557 | 0 | memcpy( &dequantcoeff[( j + jj ) * width], src, loopW * sizeof( int ) ); |
558 | 0 | } |
559 | 0 | } |
560 | |
|
561 | 0 | if( ratio > 1 ) |
562 | 0 | { |
563 | 0 | anyChange |= dequantcoeff[0] != dc; |
564 | 0 | dequantcoeff[0] = dc; |
565 | 0 | } |
566 | |
|
567 | 0 | return anyChange; |
568 | 0 | } |
569 | | |
570 | | static constexpr int g_numScalingListCoeffs = 96774; |
571 | | |
572 | | /** initialization process of scaling list array |
573 | | */ |
574 | | void Quant::xInitScalingList( const Quant* other ) |
575 | 22.7k | { |
576 | 22.7k | if( other ) |
577 | 22.0k | { |
578 | 22.0k | m_dequantCoefBuf = other->m_dequantCoefBuf; |
579 | 22.0k | m_ownDequantCoeff = false; |
580 | 22.0k | } |
581 | 712 | else |
582 | 712 | { |
583 | 712 | const size_t numScalingCoeffAlloc = g_numScalingListCoeffs + 4; // +4 to prevent out of bounds read in SIMD code |
584 | 712 | m_dequantCoefBuf = new int[numScalingCoeffAlloc]; |
585 | 712 | m_ownDequantCoeff = true; |
586 | 712 | std::fill_n( m_dequantCoefBuf, numScalingCoeffAlloc, 0 ); |
587 | 712 | } |
588 | | |
589 | 22.7k | size_t numQuants = 0; |
590 | 182k | for(uint32_t sizeIdX = 0; sizeIdX < SCALING_LIST_SIZE_NUM; sizeIdX++) |
591 | 159k | { |
592 | 1.27M | for(uint32_t sizeIdY = 0; sizeIdY < SCALING_LIST_SIZE_NUM; sizeIdY++) |
593 | 1.11M | { |
594 | 7.81M | for(uint32_t listId = 0; listId < SCALING_LIST_NUM; listId++) |
595 | 6.69M | { |
596 | 6.69M | m_dequantCoef [sizeIdX][sizeIdY][listId] = &m_dequantCoefBuf[numQuants]; |
597 | 6.69M | numQuants += g_vvcScalingListSizeX[sizeIdX] * g_vvcScalingListSizeX[sizeIdY]; |
598 | 6.69M | } // listID loop |
599 | 1.11M | } |
600 | 159k | } |
601 | | |
602 | 22.7k | CHECK( numQuants != g_numScalingListCoeffs, "Incorrect size of scaling list entries number!" ); |
603 | 22.7k | } |
604 | | |
605 | | /** destroy quantization matrix array |
606 | | */ |
607 | | void Quant::xDestroyScalingList() |
608 | 22.7k | { |
609 | 22.7k | if( m_ownDequantCoeff ) |
610 | 712 | { |
611 | 712 | delete[] m_dequantCoefBuf; |
612 | 712 | } |
613 | | |
614 | 22.7k | m_ownDequantCoeff = false; |
615 | 22.7k | m_dequantCoefBuf = nullptr; |
616 | 22.7k | } |
617 | | |
618 | | void Quant::init( const Picture *pic ) |
619 | 0 | { |
620 | 0 | const Slice* scalingListSlice = nullptr; |
621 | |
|
622 | 0 | for( const Slice* slice : pic->slices ) |
623 | 0 | { |
624 | 0 | if( slice->getExplicitScalingListUsed() ) |
625 | 0 | { |
626 | 0 | scalingListSlice = slice; |
627 | 0 | break; |
628 | 0 | } |
629 | 0 | } |
630 | |
|
631 | 0 | const Slice* slice = scalingListSlice; |
632 | |
|
633 | 0 | if( slice && slice->getExplicitScalingListUsed() ) |
634 | 0 | { |
635 | 0 | const std::shared_ptr<const APS> scalingListAPS = slice->getPicHeader()->getScalingListAPS(); |
636 | 0 | if( slice->getNalUnitLayerId() != scalingListAPS->getLayerId() ) |
637 | 0 | { |
638 | 0 | CHECK( scalingListAPS->getLayerId() > slice->getNalUnitLayerId(), "Layer Id of APS cannot be greater than layer Id of VCL NAL unit the refer to it" ); |
639 | 0 | CHECK( slice->getSPS()->getVPSId() == 0, "VPSId of the referred SPS cannot be 0 when layer Id of APS and layer Id of current slice are different" ); |
640 | 0 | for( int i = 0; i < slice->getVPS()->getNumOutputLayerSets(); i++ ) |
641 | 0 | { |
642 | 0 | bool isCurrLayerInOls = false; |
643 | 0 | bool isRefLayerInOls = false; |
644 | 0 | for( int j = slice->getVPS()->getNumLayersInOls(i) - 1; j >= 0; j-- ) |
645 | 0 | { |
646 | 0 | if( slice->getVPS()->getLayerIdInOls(i, j) == slice->getNalUnitLayerId() ) |
647 | 0 | { |
648 | 0 | isCurrLayerInOls = true; |
649 | 0 | } |
650 | 0 | if( slice->getVPS()->getLayerIdInOls(i, j) == scalingListAPS->getLayerId() ) |
651 | 0 | { |
652 | 0 | isRefLayerInOls = true; |
653 | 0 | } |
654 | 0 | } |
655 | 0 | CHECK( isCurrLayerInOls && !isRefLayerInOls, "When VCL NAl unit in layer A refers to APS in layer B, all OLS that contains layer A shall also contains layer B" ); |
656 | 0 | } |
657 | 0 | } |
658 | 0 | const ScalingList& scalingList = scalingListAPS->getScalingList(); |
659 | 0 | if( m_ownDequantCoeff ) |
660 | 0 | { |
661 | 0 | memset( m_dequantCoefBuf, 0, sizeof( int ) * g_numScalingListCoeffs ); |
662 | 0 | setScalingListDec( scalingList ); |
663 | 0 | } |
664 | 0 | setUseScalingList(true); |
665 | 0 | } |
666 | 0 | else |
667 | 0 | { |
668 | 0 | setUseScalingList( false ); |
669 | 0 | } |
670 | 0 | } |
671 | | |
672 | | } |