Coverage Report

Created: 2022-08-24 06:17

/src/x265/source/common/quant.cpp
Line
Count
Source (jump to first uncovered line)
1
/*****************************************************************************
2
 * Copyright (C) 2013-2020 MulticoreWare, Inc
3
 *
4
 * Authors: Steve Borho <steve@borho.org>
5
 *          Min Chen <chenm003@163.com>
6
 *
7
 * This program is free software; you can redistribute it and/or modify
8
 * it under the terms of the GNU General Public License as published by
9
 * the Free Software Foundation; either version 2 of the License, or
10
 * (at your option) any later version.
11
 *
12
 * This program is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15
 * GNU General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU General Public License
18
 * along with this program; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
20
 *
21
 * This program is also available under a commercial proprietary license.
22
 * For more information, contact us at license @ x265.com.
23
 *****************************************************************************/
24
25
#include "common.h"
26
#include "primitives.h"
27
#include "quant.h"
28
#include "framedata.h"
29
#include "entropy.h"
30
#include "yuv.h"
31
#include "cudata.h"
32
#include "contexts.h"
33
34
using namespace X265_NS;
35
36
0
#define SIGN(x,y) ((x^(y >> 31))-(y >> 31))
37
38
namespace {
39
40
struct coeffGroupRDStats
41
{
42
    int     nnzBeforePos0;     /* indicates coeff other than pos 0 are coded */
43
    int64_t codedLevelAndDist; /* distortion and level cost of coded coefficients */
44
    int64_t uncodedDist;       /* uncoded distortion cost of coded coefficients */
45
    int64_t sigCost;           /* cost of signaling significant coeff bitmap */
46
    int64_t sigCost0;          /* cost of signaling sig coeff bit of coeff 0 */
47
};
48
49
inline int fastMin(int x, int y)
50
0
{
51
0
    return y + ((x - y) & ((x - y) >> (sizeof(int) * CHAR_BIT - 1))); // min(x, y)
52
0
}
53
54
inline int getICRate(uint32_t absLevel, int32_t diffLevel, const int* greaterOneBits, const int* levelAbsBits, const uint32_t absGoRice, const uint32_t maxVlc, const uint32_t c1c2Rate)
55
0
{
56
0
    X265_CHECK(absGoRice <= 4, "absGoRice check failure\n");
57
0
    if (!absLevel)
58
0
    {
59
0
        X265_CHECK(diffLevel < 0, "diffLevel check failure\n");
60
0
        return 0;
61
0
    }
62
0
    int rate = 0;
63
64
0
    if (diffLevel < 0)
65
0
    {
66
0
        X265_CHECK(absLevel <= 2, "absLevel check failure\n");
67
0
        rate += greaterOneBits[(absLevel == 2)];
68
69
0
        if (absLevel == 2)
70
0
            rate += levelAbsBits[0];
71
0
    }
72
0
    else
73
0
    {
74
0
        uint32_t symbol = diffLevel;
75
0
        bool expGolomb = (symbol > maxVlc);
76
77
0
        if (expGolomb)
78
0
        {
79
0
            absLevel = symbol - maxVlc;
80
81
            // NOTE: mapping to x86 hardware instruction BSR
82
0
            unsigned long size;
83
0
            CLZ(size, absLevel);
84
0
            int egs = size * 2 + 1;
85
86
0
            rate += egs << 15;
87
88
            // NOTE: in here, expGolomb=true means (symbol >= maxVlc + 1)
89
0
            X265_CHECK(fastMin(symbol, (maxVlc + 1)) == (int)maxVlc + 1, "min check failure\n");
90
0
            symbol = maxVlc + 1;
91
0
        }
92
93
0
        uint32_t prefLen = (symbol >> absGoRice) + 1;
94
0
        uint32_t numBins = fastMin(prefLen + absGoRice, 8 /* g_goRicePrefixLen[absGoRice] + absGoRice */);
95
96
0
        rate += numBins << 15;
97
0
        rate += c1c2Rate;
98
0
    }
99
0
    return rate;
100
0
}
101
102
#if CHECKED_BUILD || _DEBUG
103
inline int getICRateNegDiff(uint32_t absLevel, const int* greaterOneBits, const int* levelAbsBits)
104
{
105
    X265_CHECK(absLevel <= 2, "absLevel check failure\n");
106
107
    int rate;
108
    if (absLevel == 0)
109
        rate = 0;
110
    else if (absLevel == 2)
111
        rate = greaterOneBits[1] + levelAbsBits[0];
112
    else
113
        rate = greaterOneBits[0];
114
    return rate;
115
}
116
#endif
117
118
inline int getICRateLessVlc(uint32_t absLevel, int32_t diffLevel, const uint32_t absGoRice)
119
0
{
120
0
    X265_CHECK(absGoRice <= 4, "absGoRice check failure\n");
121
0
    if (!absLevel)
122
0
    {
123
0
        X265_CHECK(diffLevel < 0, "diffLevel check failure\n");
124
0
        return 0;
125
0
    }
126
0
    int rate;
127
128
0
    uint32_t symbol = diffLevel;
129
0
    uint32_t prefLen = (symbol >> absGoRice) + 1;
130
0
    uint32_t numBins = fastMin(prefLen + absGoRice, 8 /* g_goRicePrefixLen[absGoRice] + absGoRice */);
131
132
0
    rate = numBins << 15;
133
134
0
    return rate;
135
0
}
136
137
/* Calculates the cost for specific absolute transform level */
138
inline uint32_t getICRateCost(uint32_t absLevel, int32_t diffLevel, const int* greaterOneBits, const int* levelAbsBits, uint32_t absGoRice, const uint32_t c1c2Rate)
139
0
{
140
0
    X265_CHECK(absLevel, "absLevel should not be zero\n");
141
142
0
    if (diffLevel < 0)
143
0
    {
144
0
        X265_CHECK((absLevel == 1) || (absLevel == 2), "absLevel range check failure\n");
145
146
0
        uint32_t rate = greaterOneBits[(absLevel == 2)];
147
0
        if (absLevel == 2)
148
0
            rate += levelAbsBits[0];
149
0
        return rate;
150
0
    }
151
0
    else
152
0
    {
153
0
        uint32_t rate;
154
0
        uint32_t symbol = diffLevel;
155
0
        if ((symbol >> absGoRice) < COEF_REMAIN_BIN_REDUCTION)
156
0
        {
157
0
            uint32_t length = symbol >> absGoRice;
158
0
            rate = (length + 1 + absGoRice) << 15;
159
0
        }
160
0
        else
161
0
        {
162
0
            uint32_t length = 0;
163
0
            symbol = (symbol >> absGoRice) - COEF_REMAIN_BIN_REDUCTION;
164
0
            if (symbol)
165
0
            {
166
0
                unsigned long idx;
167
0
                CLZ(idx, symbol + 1);
168
0
                length = idx;
169
0
            }
170
171
0
            rate = (COEF_REMAIN_BIN_REDUCTION + length + absGoRice + 1 + length) << 15;
172
0
        }
173
0
        rate += c1c2Rate;
174
0
        return rate;
175
0
    }
176
0
}
177
178
}
179
180
Quant::rdoQuant_t Quant::rdoQuant_func[NUM_CU_DEPTH] = {&Quant::rdoQuant<2>, &Quant::rdoQuant<3>, &Quant::rdoQuant<4>, &Quant::rdoQuant<5>};
181
182
Quant::Quant()
183
0
{
184
0
    m_resiDctCoeff = NULL;
185
0
    m_fencDctCoeff = NULL;
186
0
    m_fencShortBuf = NULL;
187
0
    m_frameNr      = NULL;
188
0
    m_nr           = NULL;
189
0
}
190
191
bool Quant::init(double psyScale, const ScalingList& scalingList, Entropy& entropy)
192
0
{
193
0
    m_entropyCoder = &entropy;
194
0
    m_psyRdoqScale = (int32_t)(psyScale * 256.0);
195
0
    X265_CHECK((psyScale * 256.0) < (double)MAX_INT, "psyScale value too large\n");
196
0
    m_scalingList  = &scalingList;
197
0
    m_resiDctCoeff = X265_MALLOC(int16_t, MAX_TR_SIZE * MAX_TR_SIZE * 2);
198
0
    m_fencDctCoeff = m_resiDctCoeff + (MAX_TR_SIZE * MAX_TR_SIZE);
199
0
    m_fencShortBuf = X265_MALLOC(int16_t, MAX_TR_SIZE * MAX_TR_SIZE);
200
201
0
    return m_resiDctCoeff && m_fencShortBuf;
202
0
}
203
204
bool Quant::allocNoiseReduction(const x265_param& param)
205
0
{
206
0
    m_frameNr = X265_MALLOC(NoiseReduction, param.frameNumThreads);
207
0
    if (m_frameNr)
208
0
        memset(m_frameNr, 0, sizeof(NoiseReduction) * param.frameNumThreads);
209
0
    else
210
0
        return false;
211
0
    return true;
212
0
}
213
214
Quant::~Quant()
215
0
{
216
0
    X265_FREE(m_frameNr);
217
0
    X265_FREE(m_resiDctCoeff);
218
0
    X265_FREE(m_fencShortBuf);
219
0
}
220
221
void Quant::setQPforQuant(const CUData& ctu, int qp)
222
0
{
223
0
    m_nr = m_frameNr ? &m_frameNr[ctu.m_encData->m_frameEncoderID] : NULL;
224
0
    m_qpParam[TEXT_LUMA].setQpParam(qp + QP_BD_OFFSET);
225
0
    m_rdoqLevel = ctu.m_encData->m_param->rdoqLevel;
226
0
    if (ctu.m_chromaFormat != X265_CSP_I400)
227
0
    {
228
0
        setChromaQP(qp + ctu.m_slice->m_pps->chromaQpOffset[0] + ctu.m_slice->m_chromaQpOffset[0], TEXT_CHROMA_U, ctu.m_chromaFormat);
229
0
        setChromaQP(qp + ctu.m_slice->m_pps->chromaQpOffset[1] + ctu.m_slice->m_chromaQpOffset[1], TEXT_CHROMA_V, ctu.m_chromaFormat);
230
0
    }
231
0
}
232
233
void Quant::setChromaQP(int qpin, TextType ttype, int chFmt)
234
0
{
235
0
    int qp = x265_clip3(-QP_BD_OFFSET, 57, qpin);
236
0
    if (qp >= 30)
237
0
    {
238
0
        if (chFmt == X265_CSP_I420)
239
0
            qp = g_chromaScale[qp];
240
0
        else
241
0
            qp = X265_MIN(qp, QP_MAX_SPEC);
242
0
    }
243
0
    m_qpParam[ttype].setQpParam(qp + QP_BD_OFFSET);
244
0
}
245
246
/* To minimize the distortion only. No rate is considered */
247
uint32_t Quant::signBitHidingHDQ(int16_t* coeff, int32_t* deltaU, uint32_t numSig, const TUEntropyCodingParameters &codeParams, uint32_t log2TrSize)
248
0
{
249
0
    uint32_t trSize = 1 << log2TrSize;
250
0
    const uint16_t* scan = codeParams.scan;
251
252
0
    uint8_t coeffNum[MLS_GRP_NUM];      // value range[0, 16]
253
0
    uint16_t coeffSign[MLS_GRP_NUM];    // bit mask map for non-zero coeff sign
254
0
    uint16_t coeffFlag[MLS_GRP_NUM];    // bit mask map for non-zero coeff
255
256
#if CHECKED_BUILD || _DEBUG
257
    // clean output buffer, the asm version of scanPosLast Never output anything after latest non-zero coeff group
258
    memset(coeffNum, 0, sizeof(coeffNum));
259
    memset(coeffSign, 0, sizeof(coeffNum));
260
    memset(coeffFlag, 0, sizeof(coeffNum));
261
#endif
262
0
    const int lastScanPos = primitives.scanPosLast(codeParams.scan, coeff, coeffSign, coeffFlag, coeffNum, numSig, g_scan4x4[codeParams.scanType], trSize);
263
0
    const int cgLastScanPos = (lastScanPos >> LOG2_SCAN_SET_SIZE);
264
0
    unsigned long tmp;
265
266
    // first CG need specially processing
267
0
    const uint32_t correctOffset = 0x0F & (lastScanPos ^ 0xF);
268
0
    coeffFlag[cgLastScanPos] <<= correctOffset;
269
270
0
    for (int cg = cgLastScanPos; cg >= 0; cg--)
271
0
    {
272
0
        int cgStartPos = cg << LOG2_SCAN_SET_SIZE;
273
0
        int n;
274
275
#if CHECKED_BUILD || _DEBUG
276
        for (n = SCAN_SET_SIZE - 1; n >= 0; --n)
277
            if (coeff[scan[n + cgStartPos]])
278
                break;
279
        int lastNZPosInCG0 = n;
280
#endif
281
282
0
        if (coeffNum[cg] == 0)
283
0
        {
284
0
            X265_CHECK(lastNZPosInCG0 < 0, "all zero block check failure\n");
285
0
            continue;
286
0
        }
287
288
#if CHECKED_BUILD || _DEBUG
289
        for (n = 0;; n++)
290
            if (coeff[scan[n + cgStartPos]])
291
                break;
292
293
        int firstNZPosInCG0 = n;
294
#endif
295
296
0
        CLZ(tmp, coeffFlag[cg]);
297
0
        const int firstNZPosInCG = (15 ^ tmp);
298
299
0
        CTZ(tmp, coeffFlag[cg]);
300
0
        const int lastNZPosInCG = (15 ^ tmp);
301
302
0
        X265_CHECK(firstNZPosInCG0 == firstNZPosInCG, "firstNZPosInCG0 check failure\n");
303
0
        X265_CHECK(lastNZPosInCG0 == lastNZPosInCG, "lastNZPosInCG0 check failure\n");
304
305
0
        if (lastNZPosInCG - firstNZPosInCG >= SBH_THRESHOLD)
306
0
        {
307
0
            uint32_t signbit = coeff[scan[cgStartPos + firstNZPosInCG]] > 0 ? 0 : 1;
308
0
            uint32_t absSum = 0;
309
310
0
            for (n = firstNZPosInCG; n <= lastNZPosInCG; n++)
311
0
                absSum += coeff[scan[n + cgStartPos]];
312
313
0
            if (signbit != (absSum & 0x1)) // compare signbit with sum_parity
314
0
            {
315
0
                int minCostInc = MAX_INT,  minPos = -1, curCost = MAX_INT;
316
0
                int32_t finalChange = 0, curChange = 0;
317
0
                uint32_t cgFlags = coeffFlag[cg];
318
0
                if (cg == cgLastScanPos)
319
0
                    cgFlags >>= correctOffset;
320
321
0
                for (n = (cg == cgLastScanPos ? lastNZPosInCG : SCAN_SET_SIZE - 1); n >= 0; --n)
322
0
                {
323
0
                    uint32_t blkPos = scan[n + cgStartPos];
324
0
                    X265_CHECK(!!coeff[blkPos] == !!(cgFlags & 1), "non zero coeff check failure\n");
325
326
0
                    if (cgFlags & 1)
327
0
                    {
328
0
                        if (deltaU[blkPos] > 0)
329
0
                        {
330
0
                            curCost = -deltaU[blkPos];
331
0
                            curChange = 1;
332
0
                        }
333
0
                        else
334
0
                        {
335
0
                            if ((cgFlags == 1) && (abs(coeff[blkPos]) == 1))
336
0
                            {
337
0
                                X265_CHECK(n == firstNZPosInCG, "firstNZPosInCG position check failure\n");
338
0
                                curCost = MAX_INT;
339
0
                            }
340
0
                            else
341
0
                            {
342
0
                                curCost = deltaU[blkPos];
343
0
                                curChange = -1;
344
0
                            }
345
0
                        }
346
0
                    }
347
0
                    else
348
0
                    {
349
0
                        if (cgFlags == 0)
350
0
                        {
351
0
                            X265_CHECK(n < firstNZPosInCG, "firstNZPosInCG position check failure\n");
352
0
                            uint32_t thisSignBit = m_resiDctCoeff[blkPos] >= 0 ? 0 : 1;
353
0
                            if (thisSignBit != signbit)
354
0
                                curCost = MAX_INT;
355
0
                            else
356
0
                            {
357
0
                                curCost = -deltaU[blkPos];
358
0
                                curChange = 1;
359
0
                            }
360
0
                        }
361
0
                        else
362
0
                        {
363
0
                            curCost = -deltaU[blkPos];
364
0
                            curChange = 1;
365
0
                        }
366
0
                    }
367
368
0
                    if (curCost < minCostInc)
369
0
                    {
370
0
                        minCostInc = curCost;
371
0
                        finalChange = curChange;
372
0
                        minPos = blkPos;
373
0
                    }
374
0
                    cgFlags>>=1;
375
0
                }
376
377
                /* do not allow change to violate coeff clamp */
378
0
                if (coeff[minPos] == 32767 || coeff[minPos] == -32768)
379
0
                    finalChange = -1;
380
381
0
                if (!coeff[minPos])
382
0
                    numSig++;
383
0
                else if (finalChange == -1 && abs(coeff[minPos]) == 1)
384
0
                    numSig--;
385
386
0
                {
387
0
                    const int16_t sigMask = ((int16_t)m_resiDctCoeff[minPos]) >> 15;
388
0
                    coeff[minPos] += ((int16_t)finalChange ^ sigMask) - sigMask;
389
0
                }
390
0
            }
391
0
        }
392
0
    }
393
394
0
    return numSig;
395
0
}
396
397
uint32_t Quant::transformNxN(const CUData& cu, const pixel* fenc, uint32_t fencStride, const int16_t* residual, uint32_t resiStride,
398
                             coeff_t* coeff, uint32_t log2TrSize, TextType ttype, uint32_t absPartIdx, bool useTransformSkip)
399
0
{
400
0
    const uint32_t sizeIdx = log2TrSize - 2;
401
402
0
    if (cu.m_tqBypass[0])
403
0
    {
404
0
        X265_CHECK(log2TrSize >= 2 && log2TrSize <= 5, "Block size mistake!\n");
405
0
        return primitives.cu[sizeIdx].copy_cnt(coeff, residual, resiStride);
406
0
    }
407
408
0
    bool isLuma  = ttype == TEXT_LUMA;
409
0
    bool usePsy  = m_psyRdoqScale && isLuma && !useTransformSkip;
410
0
    int transformShift = MAX_TR_DYNAMIC_RANGE - X265_DEPTH - log2TrSize; // Represents scaling through forward transform
411
412
0
    X265_CHECK((cu.m_slice->m_sps->quadtreeTULog2MaxSize >= log2TrSize), "transform size too large\n");
413
0
    if (useTransformSkip)
414
0
    {
415
0
#if X265_DEPTH <= 10
416
0
        X265_CHECK(transformShift >= 0, "invalid transformShift\n");
417
0
        primitives.cu[sizeIdx].cpy2Dto1D_shl(m_resiDctCoeff, residual, resiStride, transformShift);
418
#else
419
        if (transformShift >= 0)
420
            primitives.cu[sizeIdx].cpy2Dto1D_shl(m_resiDctCoeff, residual, resiStride, transformShift);
421
        else
422
            primitives.cu[sizeIdx].cpy2Dto1D_shr(m_resiDctCoeff, residual, resiStride, -transformShift);
423
#endif
424
0
    }
425
0
    else
426
0
    {
427
0
        bool isIntra = cu.isIntra(absPartIdx);
428
429
0
        if (!sizeIdx && isLuma && isIntra)
430
0
            primitives.dst4x4(residual, m_resiDctCoeff, resiStride);
431
0
        else
432
0
            primitives.cu[sizeIdx].dct(residual, m_resiDctCoeff, resiStride);
433
434
        /* NOTE: if RDOQ is disabled globally, psy-rdoq is also disabled, so
435
         * there is no risk of performing this DCT unnecessarily */
436
0
        if (usePsy)
437
0
        {
438
0
            int trSize = 1 << log2TrSize;
439
            /* perform DCT on source pixels for psy-rdoq */
440
0
            primitives.cu[sizeIdx].copy_ps(m_fencShortBuf, trSize, fenc, fencStride);
441
0
            primitives.cu[sizeIdx].dct(m_fencShortBuf, m_fencDctCoeff, trSize);
442
0
        }
443
444
0
        if (m_nr && m_nr->offset)
445
0
        {
446
            /* denoise is not applied to intra residual, so DST can be ignored */
447
0
            int cat = sizeIdx + 4 * !isLuma + 8 * !isIntra;
448
0
            int numCoeff = 1 << (log2TrSize * 2);
449
0
            primitives.denoiseDct(m_resiDctCoeff, m_nr->residualSum[cat], m_nr->offset[cat], numCoeff);
450
0
            m_nr->count[cat]++;
451
0
        }
452
0
    }
453
454
0
    if (m_rdoqLevel)
455
0
        return (this->*rdoQuant_func[log2TrSize - 2])(cu, coeff, ttype, absPartIdx, usePsy);
456
0
    else
457
0
    {
458
0
        int deltaU[32 * 32];
459
460
0
        int scalingListType = (cu.isIntra(absPartIdx) ? 0 : 3) + ttype;
461
0
        int rem = m_qpParam[ttype].rem;
462
0
        int per = m_qpParam[ttype].per;
463
0
        const int32_t* quantCoeff = m_scalingList->m_quantCoef[log2TrSize - 2][scalingListType][rem];
464
465
0
        int qbits = QUANT_SHIFT + per + transformShift;
466
0
        int add = (cu.m_slice->m_sliceType == I_SLICE ? 171 : 85) << (qbits - 9);
467
0
        int numCoeff = 1 << (log2TrSize * 2);
468
469
0
        uint32_t numSig = primitives.quant(m_resiDctCoeff, quantCoeff, deltaU, coeff, qbits, add, numCoeff);
470
471
0
        if (numSig >= 2 && cu.m_slice->m_pps->bSignHideEnabled)
472
0
        {
473
0
            TUEntropyCodingParameters codeParams;
474
0
            cu.getTUEntropyCodingParameters(codeParams, absPartIdx, log2TrSize, isLuma);
475
0
            return signBitHidingHDQ(coeff, deltaU, numSig, codeParams, log2TrSize);
476
0
        }
477
0
        else
478
0
            return numSig;
479
0
    }
480
0
}
481
482
uint64_t Quant::ssimDistortion(const CUData& cu, const pixel* fenc, uint32_t fStride, const pixel* recon, intptr_t rstride, uint32_t log2TrSize, TextType ttype, uint32_t absPartIdx)
483
0
{
484
0
    static const int ssim_c1 = (int)(.01 * .01 * PIXEL_MAX * PIXEL_MAX * 64 + .5); // 416
485
0
    static const int ssim_c2 = (int)(.03 * .03 * PIXEL_MAX * PIXEL_MAX * 64 * 63 + .5); // 235963
486
0
    int shift = (X265_DEPTH - 8);
487
488
0
    int trSize = 1 << log2TrSize;
489
0
    uint64_t ssDc = 0, ssBlock = 0, ssAc = 0;
490
491
    // Calculation of (X(0) - Y(0)) * (X(0) - Y(0)), DC
492
0
    ssDc = 0;
493
0
    for (int y = 0; y < trSize; y += 4)
494
0
    {
495
0
        for (int x = 0; x < trSize; x += 4)
496
0
        {
497
0
            int temp = fenc[y * fStride + x] - recon[y * rstride + x]; // copy of residual coeff
498
0
            ssDc += temp * temp;
499
0
        }
500
0
    }
501
502
    // Calculation of (X(k) - Y(k)) * (X(k) - Y(k)), AC
503
0
    ssBlock = 0;
504
0
    uint64_t ac_k = 0;
505
0
    primitives.cu[log2TrSize - 2].ssimDist(fenc, fStride, recon, rstride, &ssBlock, shift, &ac_k);
506
0
    ssAc = ssBlock - ssDc;
507
508
    // 1. Calculation of fdc'
509
    // Calculate numerator of dc normalization factor
510
0
    uint64_t fDc_num = 0;
511
512
    // 2. Calculate dc component
513
0
    uint64_t dc_k = 0;
514
0
    for (int block_yy = 0; block_yy < trSize; block_yy += 4)
515
0
    {
516
0
        for (int block_xx = 0; block_xx < trSize; block_xx += 4)
517
0
        {
518
0
            uint32_t temp = fenc[block_yy * fStride + block_xx] >> shift;
519
0
            dc_k += temp * temp;
520
0
        }
521
0
    }
522
523
0
    fDc_num = (2 * dc_k)  + (trSize * trSize * ssim_c1); // 16 pixels -> for each 4x4 block
524
0
    fDc_num /= ((trSize >> 2) * (trSize >> 2));
525
526
    // 1. Calculation of fac'
527
    // Calculate numerator of ac normalization factor
528
0
    uint64_t fAc_num = 0;
529
530
    // 2. Calculate ac component
531
0
    ac_k -= dc_k;
532
533
0
    double s = 1 + 0.005 * cu.m_qp[absPartIdx];
534
535
0
    fAc_num = ac_k + uint64_t(s * ac_k) + ssim_c2;
536
0
    fAc_num /= ((trSize >> 2) * (trSize >> 2));
537
538
    // Calculate dc and ac normalization factor
539
0
    uint64_t ssim_distortion = ((ssDc * cu.m_fDc_den[ttype]) / fDc_num) + ((ssAc * cu.m_fAc_den[ttype]) / fAc_num);
540
0
    return ssim_distortion;
541
0
}
542
543
void Quant::invtransformNxN(const CUData& cu, int16_t* residual, uint32_t resiStride, const coeff_t* coeff,
544
                            uint32_t log2TrSize, TextType ttype, bool bIntra, bool useTransformSkip, uint32_t numSig)
545
0
{
546
0
    const uint32_t sizeIdx = log2TrSize - 2;
547
0
    if (cu.m_tqBypass[0])
548
0
    {
549
0
        primitives.cu[sizeIdx].cpy1Dto2D_shl[resiStride % 64 == 0](residual, coeff, resiStride, 0);
550
0
        return;
551
0
    }
552
    // Values need to pass as input parameter in dequant
553
0
    int rem = m_qpParam[ttype].rem;
554
0
    int per = m_qpParam[ttype].per;
555
0
    int transformShift = MAX_TR_DYNAMIC_RANGE - X265_DEPTH - log2TrSize;
556
0
    int shift = QUANT_IQUANT_SHIFT - QUANT_SHIFT - transformShift;
557
0
    int numCoeff = 1 << (log2TrSize * 2);
558
559
0
    if (m_scalingList->m_bEnabled)
560
0
    {
561
0
        int scalingListType = (bIntra ? 0 : 3) + ttype;
562
0
        const int32_t* dequantCoef = m_scalingList->m_dequantCoef[sizeIdx][scalingListType][rem];
563
0
        primitives.dequant_scaling(coeff, dequantCoef, m_resiDctCoeff, numCoeff, per, shift);
564
0
    }
565
0
    else
566
0
    {
567
0
        int scale = m_scalingList->s_invQuantScales[rem] << per;
568
0
        primitives.dequant_normal(coeff, m_resiDctCoeff, numCoeff, scale, shift);
569
0
    }
570
571
0
    if (useTransformSkip)
572
0
    {
573
0
#if X265_DEPTH <= 10
574
0
        X265_CHECK(transformShift > 0, "invalid transformShift\n");
575
0
        primitives.cu[sizeIdx].cpy1Dto2D_shr(residual, m_resiDctCoeff, resiStride, transformShift);
576
#else
577
        if (transformShift > 0)
578
            primitives.cu[sizeIdx].cpy1Dto2D_shr(residual, m_resiDctCoeff, resiStride, transformShift);
579
        else
580
            primitives.cu[sizeIdx].cpy1Dto2D_shl[resiStride % 64 == 0](residual, m_resiDctCoeff, resiStride, -transformShift);
581
#endif
582
0
    }
583
0
    else
584
0
    {
585
0
        int useDST = !sizeIdx && ttype == TEXT_LUMA && bIntra;
586
0
        X265_CHECK((int)numSig == primitives.cu[log2TrSize - 2].count_nonzero(coeff), "numSig differ\n");
587
        // DC only
588
0
        if (numSig == 1 && coeff[0] != 0 && !useDST)
589
0
        {
590
0
            const int shift_1st = 7 - 6;
591
0
            const int add_1st = 1 << (shift_1st - 1);
592
0
            const int shift_2nd = 12 - (X265_DEPTH - 8) - 3;
593
0
            const int add_2nd = 1 << (shift_2nd - 1);
594
595
0
            int dc_val = (((m_resiDctCoeff[0] * (64 >> 6) + add_1st) >> shift_1st) * (64 >> 3) + add_2nd) >> shift_2nd;
596
0
            primitives.cu[sizeIdx].blockfill_s[resiStride % 64 == 0](residual, resiStride, (int16_t)dc_val);
597
0
            return;
598
0
        }
599
600
0
        if (useDST)
601
0
            primitives.idst4x4(m_resiDctCoeff, residual, resiStride);
602
0
        else
603
0
            primitives.cu[sizeIdx].idct(m_resiDctCoeff, residual, resiStride);
604
0
    }
605
0
}
606
607
/* Rate distortion optimized quantization for entropy coding engines using
608
 * probability models like CABAC */
609
template<uint32_t log2TrSize>
610
uint32_t Quant::rdoQuant(const CUData& cu, int16_t* dstCoeff, TextType ttype, uint32_t absPartIdx, bool usePsy)
611
0
{
612
0
    const int transformShift = MAX_TR_DYNAMIC_RANGE - X265_DEPTH - log2TrSize; /* Represents scaling through forward transform */
613
0
    int scalingListType = (cu.isIntra(absPartIdx) ? 0 : 3) + ttype;
614
0
    const uint32_t usePsyMask = usePsy ? -1 : 0;
615
616
0
    X265_CHECK(scalingListType < 6, "scaling list type out of range\n");
617
618
0
    int rem = m_qpParam[ttype].rem;
619
0
    int per = m_qpParam[ttype].per;
620
0
    int qbits = QUANT_SHIFT + per + transformShift; /* Right shift of non-RDOQ quantizer level = (coeff*Q + offset)>>q_bits */
621
0
    int add = (1 << (qbits - 1));
622
0
    const int32_t* qCoef = m_scalingList->m_quantCoef[log2TrSize - 2][scalingListType][rem];
623
624
0
    const int numCoeff = 1 << (log2TrSize * 2);
625
0
    uint32_t numSig = primitives.nquant(m_resiDctCoeff, qCoef, dstCoeff, qbits, add, numCoeff);
626
0
    X265_CHECK((int)numSig == primitives.cu[log2TrSize - 2].count_nonzero(dstCoeff), "numSig differ\n");
627
0
    if (!numSig)
628
0
        return 0;
629
0
    const uint32_t trSize = 1 << log2TrSize;
630
0
    int64_t lambda2 = m_qpParam[ttype].lambda2;
631
0
    int64_t psyScale = ((int64_t)m_psyRdoqScale * m_qpParam[ttype].lambda);
632
    /* unquant constants for measuring distortion. Scaling list quant coefficients have a (1 << 4)
633
     * scale applied that must be removed during unquant. Note that in real dequant there is clipping
634
     * at several stages. We skip the clipping for simplicity when measuring RD cost */
635
0
    const int32_t* unquantScale = m_scalingList->m_dequantCoef[log2TrSize - 2][scalingListType][rem];
636
0
    int unquantShift = QUANT_IQUANT_SHIFT - QUANT_SHIFT - transformShift + (m_scalingList->m_bEnabled ? 4 : 0);
637
0
    int unquantRound = (unquantShift > per) ? 1 << (unquantShift - per - 1) : 0;
638
0
    const int scaleBits = SCALE_BITS - 2 * transformShift;
639
640
0
#define UNQUANT(lvl)    (((lvl) * (unquantScale[blkPos] << per) + unquantRound) >> unquantShift)
641
0
#define SIGCOST(bits)   ((lambda2 * (bits)) >> 8)
642
0
#define RDCOST(d, bits) ((((int64_t)d * d) << scaleBits) + SIGCOST(bits))
643
0
#define PSYVALUE(rec)   ((psyScale * (rec)) >> X265_MAX(0, (2 * transformShift + 1)))
644
645
0
    int64_t costCoeff[trSize * trSize];   /* d*d + lambda * bits */
646
0
    int64_t costUncoded[trSize * trSize]; /* d*d + lambda * 0    */
647
0
    int64_t costSig[trSize * trSize];     /* lambda * bits       */
648
649
0
    int rateIncUp[trSize * trSize];      /* signal overhead of increasing level */
650
0
    int rateIncDown[trSize * trSize];    /* signal overhead of decreasing level */
651
0
    int sigRateDelta[trSize * trSize];   /* signal difference between zero and non-zero */
652
653
0
    int64_t costCoeffGroupSig[MLS_GRP_NUM]; /* lambda * bits of group coding cost */
654
0
    uint64_t sigCoeffGroupFlag64 = 0;
655
656
0
    const uint32_t cgSize = (1 << MLS_CG_SIZE); /* 4x4 num coef = 16 */
657
0
    bool bIsLuma = ttype == TEXT_LUMA;
658
659
    /* total rate distortion cost of transform block, as CBF=0 */
660
0
    int64_t totalUncodedCost = 0;
661
662
    /* Total rate distortion cost of this transform block, counting te distortion of uncoded blocks,
663
     * the distortion and signal cost of coded blocks, and the coding cost of significant
664
     * coefficient and coefficient group bitmaps */
665
0
    int64_t totalRdCost = 0;
666
667
0
    TUEntropyCodingParameters codeParams;
668
0
    cu.getTUEntropyCodingParameters(codeParams, absPartIdx, log2TrSize, bIsLuma);
669
0
    const uint32_t log2TrSizeCG = log2TrSize - 2;
670
0
    const uint32_t cgNum = 1 << (log2TrSizeCG * 2);
671
0
    const uint32_t cgStride = (trSize >> MLS_CG_LOG2_SIZE);
672
673
0
    uint8_t coeffNum[MLS_GRP_NUM];      // value range[0, 16]
674
0
    uint16_t coeffSign[MLS_GRP_NUM];    // bit mask map for non-zero coeff sign
675
0
    uint16_t coeffFlag[MLS_GRP_NUM];    // bit mask map for non-zero coeff
676
677
#if CHECKED_BUILD || _DEBUG
678
    // clean output buffer, the asm version of scanPosLast Never output anything after latest non-zero coeff group
679
    memset(coeffNum, 0, sizeof(coeffNum));
680
    memset(coeffSign, 0, sizeof(coeffNum));
681
    memset(coeffFlag, 0, sizeof(coeffNum));
682
#endif
683
0
    const int lastScanPos = primitives.scanPosLast(codeParams.scan, dstCoeff, coeffSign, coeffFlag, coeffNum, numSig, g_scan4x4[codeParams.scanType], trSize);
684
0
    const int cgLastScanPos = (lastScanPos >> LOG2_SCAN_SET_SIZE);
685
686
687
    /* TODO: update bit estimates if dirty */
688
0
    EstBitsSbac& estBitsSbac = m_entropyCoder->m_estBitsSbac;
689
690
0
    uint32_t scanPos = 0;
691
0
    uint32_t c1 = 1;
692
693
    // process trail all zero Coeff Group
694
695
    /* coefficients after lastNZ have no distortion signal cost */
696
0
    const int zeroCG = cgNum - 1 - cgLastScanPos;
697
0
    memset(&costCoeff[(cgLastScanPos + 1) << MLS_CG_SIZE], 0, zeroCG * MLS_CG_BLK_SIZE * sizeof(int64_t));
698
0
    memset(&costSig[(cgLastScanPos + 1) << MLS_CG_SIZE], 0, zeroCG * MLS_CG_BLK_SIZE * sizeof(int64_t));
699
700
    /* sum zero coeff (uncodec) cost */
701
702
    // TODO: does we need these cost?
703
0
    if (usePsyMask)
704
0
    {
705
0
        for (int cgScanPos = cgLastScanPos + 1; cgScanPos < (int)cgNum ; cgScanPos++)
706
0
        {
707
0
            X265_CHECK(coeffNum[cgScanPos] == 0, "count of coeff failure\n");
708
0
            uint32_t scanPosBase = (cgScanPos << MLS_CG_SIZE);
709
0
            uint32_t blkPos      = codeParams.scan[scanPosBase];
710
0
#if X265_ARCH_X86
711
0
            bool enable512 = detect512();
712
0
            if (enable512)
713
0
                primitives.cu[log2TrSize - 2].psyRdoQuant(m_resiDctCoeff, m_fencDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, &psyScale, blkPos);
714
0
            else
715
0
            {
716
0
                primitives.cu[log2TrSize - 2].psyRdoQuant_1p(m_resiDctCoeff,  costUncoded, &totalUncodedCost, &totalRdCost,blkPos);
717
0
                primitives.cu[log2TrSize - 2].psyRdoQuant_2p(m_resiDctCoeff, m_fencDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, &psyScale, blkPos);
718
0
            }
719
#else
720
            primitives.cu[log2TrSize - 2].psyRdoQuant_1p(m_resiDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, blkPos);
721
            primitives.cu[log2TrSize - 2].psyRdoQuant_2p(m_resiDctCoeff, m_fencDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, &psyScale, blkPos);
722
#endif
723
0
        }
724
0
    }
725
0
    else
726
0
    {
727
        // non-psy path
728
0
        for (int cgScanPos = cgLastScanPos + 1; cgScanPos < (int)cgNum ; cgScanPos++)
729
0
        {
730
0
            X265_CHECK(coeffNum[cgScanPos] == 0, "count of coeff failure\n");
731
0
            uint32_t scanPosBase = (cgScanPos << MLS_CG_SIZE);
732
0
            uint32_t blkPos      = codeParams.scan[scanPosBase];
733
0
            primitives.cu[log2TrSize - 2].nonPsyRdoQuant(m_resiDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, blkPos);
734
0
        }
735
0
    }
736
0
    static const uint8_t table_cnt[5][SCAN_SET_SIZE] =
737
0
    {
738
        // patternSigCtx = 0
739
0
        {
740
0
            2, 1, 1, 0,
741
0
            1, 1, 0, 0,
742
0
            1, 0, 0, 0,
743
0
            0, 0, 0, 0,
744
0
        },
745
        // patternSigCtx = 1
746
0
        {
747
0
            2, 2, 2, 2,
748
0
            1, 1, 1, 1,
749
0
            0, 0, 0, 0,
750
0
            0, 0, 0, 0,
751
0
        },
752
        // patternSigCtx = 2
753
0
        {
754
0
            2, 1, 0, 0,
755
0
            2, 1, 0, 0,
756
0
            2, 1, 0, 0,
757
0
            2, 1, 0, 0,
758
0
        },
759
        // patternSigCtx = 3
760
0
        {
761
0
            2, 2, 2, 2,
762
0
            2, 2, 2, 2,
763
0
            2, 2, 2, 2,
764
0
            2, 2, 2, 2,
765
0
        },
766
        // 4x4
767
0
        {
768
0
            0, 1, 4, 5,
769
0
            2, 3, 4, 5,
770
0
            6, 6, 8, 8,
771
0
            7, 7, 8, 8
772
0
        }
773
0
    };
774
775
    /* iterate over coding groups in reverse scan order */
776
0
    for (int cgScanPos = cgLastScanPos; cgScanPos >= 0; cgScanPos--)
777
0
    {
778
0
        uint32_t ctxSet = (cgScanPos && bIsLuma) ? 2 : 0;
779
0
        const uint32_t cgBlkPos = codeParams.scanCG[cgScanPos];
780
0
        const uint32_t cgPosY   = cgBlkPos >> log2TrSizeCG;
781
0
        const uint32_t cgPosX   = cgBlkPos & ((1 << log2TrSizeCG) - 1);
782
0
        const uint64_t cgBlkPosMask = ((uint64_t)1 << cgBlkPos);
783
0
        const int patternSigCtx = calcPatternSigCtx(sigCoeffGroupFlag64, cgPosX, cgPosY, cgBlkPos, cgStride);
784
0
        const int ctxSigOffset = codeParams.firstSignificanceMapContext + (cgScanPos && bIsLuma ? 3 : 0);
785
786
0
        if (c1 == 0)
787
0
            ctxSet++;
788
0
        c1 = 1;
789
790
0
        if (cgScanPos && (coeffNum[cgScanPos] == 0))
791
0
        {
792
            // TODO: does we need zero-coeff cost?
793
0
            const uint32_t scanPosBase = (cgScanPos << MLS_CG_SIZE);
794
0
            uint32_t blkPos = codeParams.scan[scanPosBase];
795
0
            if (usePsyMask)
796
0
            {
797
0
#if X265_ARCH_X86
798
0
                bool enable512 = detect512();
799
0
                if (enable512)
800
0
                    primitives.cu[log2TrSize - 2].psyRdoQuant(m_resiDctCoeff, m_fencDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, &psyScale, blkPos);
801
0
                else
802
0
                {
803
0
                    primitives.cu[log2TrSize - 2].psyRdoQuant_1p(m_resiDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, blkPos);
804
0
                    primitives.cu[log2TrSize - 2].psyRdoQuant_2p(m_resiDctCoeff, m_fencDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, &psyScale, blkPos);
805
0
                }
806
#else
807
                primitives.cu[log2TrSize - 2].psyRdoQuant_1p(m_resiDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, blkPos);
808
                primitives.cu[log2TrSize - 2].psyRdoQuant_2p(m_resiDctCoeff, m_fencDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, &psyScale, blkPos);
809
#endif
810
0
                blkPos = codeParams.scan[scanPosBase];
811
0
                for (int y = 0; y < MLS_CG_SIZE; y++)
812
0
                {
813
0
                    for (int x = 0; x < MLS_CG_SIZE; x++)
814
0
                    {
815
0
                        const uint32_t scanPosOffset =  y * MLS_CG_SIZE + x;
816
0
                        const uint32_t ctxSig = table_cnt[patternSigCtx][g_scan4x4[codeParams.scanType][scanPosOffset]] + ctxSigOffset;
817
0
                        X265_CHECK(trSize > 4, "trSize check failure\n");
818
0
                        X265_CHECK(ctxSig == getSigCtxInc(patternSigCtx, log2TrSize, trSize, codeParams.scan[scanPosBase + scanPosOffset], bIsLuma, codeParams.firstSignificanceMapContext), "sigCtx check failure\n");
819
820
0
                        costSig[scanPosBase + scanPosOffset] = SIGCOST(estBitsSbac.significantBits[0][ctxSig]);
821
0
                        costCoeff[scanPosBase + scanPosOffset] = costUncoded[blkPos + x];
822
0
                        sigRateDelta[blkPos + x] = estBitsSbac.significantBits[1][ctxSig] - estBitsSbac.significantBits[0][ctxSig];
823
0
                    }
824
0
                    blkPos += trSize;
825
0
                }
826
0
            }
827
0
            else
828
0
            {
829
                // non-psy path
830
0
                primitives.cu[log2TrSize - 2].nonPsyRdoQuant(m_resiDctCoeff, costUncoded, &totalUncodedCost, &totalRdCost, blkPos);
831
0
                blkPos = codeParams.scan[scanPosBase];
832
0
                for (int y = 0; y < MLS_CG_SIZE; y++)
833
0
                {
834
0
                    for (int x = 0; x < MLS_CG_SIZE; x++)
835
0
                    {
836
0
                        const uint32_t scanPosOffset =  y * MLS_CG_SIZE + x;
837
0
                        const uint32_t ctxSig = table_cnt[patternSigCtx][g_scan4x4[codeParams.scanType][scanPosOffset]] + ctxSigOffset;
838
0
                        X265_CHECK(trSize > 4, "trSize check failure\n");
839
0
                        X265_CHECK(ctxSig == getSigCtxInc(patternSigCtx, log2TrSize, trSize, codeParams.scan[scanPosBase + scanPosOffset], bIsLuma, codeParams.firstSignificanceMapContext), "sigCtx check failure\n");
840
841
0
                        costSig[scanPosBase + scanPosOffset] = SIGCOST(estBitsSbac.significantBits[0][ctxSig]);
842
0
                        costCoeff[scanPosBase + scanPosOffset] = costUncoded[blkPos + x];
843
0
                        sigRateDelta[blkPos + x] = estBitsSbac.significantBits[1][ctxSig] - estBitsSbac.significantBits[0][ctxSig];
844
0
                    }
845
0
                    blkPos += trSize;
846
0
                }
847
0
            }
848
849
            /* there were no coded coefficients in this coefficient group */
850
0
            {
851
0
                uint32_t ctxSig = getSigCoeffGroupCtxInc(sigCoeffGroupFlag64, cgPosX, cgPosY, cgBlkPos, cgStride);
852
0
                costCoeffGroupSig[cgScanPos] = SIGCOST(estBitsSbac.significantCoeffGroupBits[ctxSig][0]);
853
0
                totalRdCost += costCoeffGroupSig[cgScanPos];  /* add cost of 0 bit in significant CG bitmap */
854
0
            }
855
0
            continue;
856
0
        }
857
858
0
        coeffGroupRDStats cgRdStats;
859
0
        memset(&cgRdStats, 0, sizeof(coeffGroupRDStats));
860
861
0
        uint32_t subFlagMask = coeffFlag[cgScanPos];
862
0
        int    c2            = 0;
863
0
        uint32_t goRiceParam = 0;
864
0
        uint32_t levelThreshold = 3;
865
0
        uint32_t c1Idx       = 0;
866
0
        uint32_t c2Idx       = 0;
867
        /* iterate over coefficients in each group in reverse scan order */
868
0
        for (int scanPosinCG = cgSize - 1; scanPosinCG >= 0; scanPosinCG--)
869
0
        {
870
0
            scanPos              = (cgScanPos << MLS_CG_SIZE) + scanPosinCG;
871
0
            uint32_t blkPos      = codeParams.scan[scanPos];
872
0
            uint32_t maxAbsLevel = dstCoeff[blkPos];                  /* abs(quantized coeff) */
873
0
            int signCoef         = m_resiDctCoeff[blkPos];            /* pre-quantization DCT coeff */
874
0
            int predictedCoef    = m_fencDctCoeff[blkPos] - signCoef; /* predicted DCT = source DCT - residual DCT*/
875
876
            /* RDOQ measures distortion as the squared difference between the unquantized coded level
877
             * and the original DCT coefficient. The result is shifted scaleBits to account for the
878
             * FIX15 nature of the CABAC cost tables minus the forward transform scale */
879
880
            /* cost of not coding this coefficient (all distortion, no signal bits) */
881
0
            costUncoded[blkPos] = ((int64_t)signCoef * signCoef) << scaleBits;
882
0
            X265_CHECK((!!scanPos ^ !!blkPos) == 0, "failed on (blkPos=0 && scanPos!=0)\n");
883
0
            if (usePsyMask & scanPos)
884
                /* when no residual coefficient is coded, predicted coef == recon coef */
885
0
                costUncoded[blkPos] -= PSYVALUE(predictedCoef);
886
887
0
            totalUncodedCost += costUncoded[blkPos];
888
889
            // coefficient level estimation
890
0
            const int* greaterOneBits = estBitsSbac.greaterOneBits[4 * ctxSet + c1];
891
            //const uint32_t ctxSig = (blkPos == 0) ? 0 : table_cnt[(trSize == 4) ? 4 : patternSigCtx][g_scan4x4[codeParams.scanType][scanPosinCG]] + ctxSigOffset;
892
0
            static const uint64_t table_cnt64[4] = {0x0000000100110112ULL, 0x0000000011112222ULL, 0x0012001200120012ULL, 0x2222222222222222ULL};
893
0
            uint64_t ctxCnt = (trSize == 4) ? 0x8877886654325410ULL : table_cnt64[patternSigCtx];
894
0
            const uint32_t ctxSig = (blkPos == 0) ? 0 : ((ctxCnt >> (4 * g_scan4x4[codeParams.scanType][scanPosinCG])) & 0xF) + ctxSigOffset;
895
            // NOTE: above equal to 'table_cnt[(trSize == 4) ? 4 : patternSigCtx][g_scan4x4[codeParams.scanType][scanPosinCG]] + ctxSigOffset'
896
0
            X265_CHECK(ctxSig == getSigCtxInc(patternSigCtx, log2TrSize, trSize, blkPos, bIsLuma, codeParams.firstSignificanceMapContext), "sigCtx check failure\n");
897
898
            // before find lastest non-zero coeff
899
0
            if (scanPos > (uint32_t)lastScanPos)
900
0
            {
901
                /* coefficients after lastNZ have no distortion signal cost */
902
0
                costCoeff[scanPos] = 0;
903
0
                costSig[scanPos] = 0;
904
905
                /* No non-zero coefficient yet found, but this does not mean
906
                 * there is no uncoded-cost for this coefficient. Pre-
907
                 * quantization the coefficient may have been non-zero */
908
0
                totalRdCost += costUncoded[blkPos];
909
0
            }
910
0
            else if (!(subFlagMask & 1))
911
0
            {
912
                // fast zero coeff path
913
                /* set default costs to uncoded costs */
914
0
                costSig[scanPos] = SIGCOST(estBitsSbac.significantBits[0][ctxSig]);
915
0
                costCoeff[scanPos] = costUncoded[blkPos] + costSig[scanPos];
916
0
                sigRateDelta[blkPos] = estBitsSbac.significantBits[1][ctxSig] - estBitsSbac.significantBits[0][ctxSig];
917
0
                totalRdCost += costCoeff[scanPos];
918
0
                rateIncUp[blkPos] = greaterOneBits[0];
919
920
0
                subFlagMask >>= 1;
921
0
            }
922
0
            else
923
0
            {
924
0
                subFlagMask >>= 1;
925
926
0
                const uint32_t c1c2idx = ((c1Idx - 8) >> (sizeof(int) * CHAR_BIT - 1)) + (((-(int)c2Idx) >> (sizeof(int) * CHAR_BIT - 1)) + 1) * 2;
927
0
                const uint32_t baseLevel = ((uint32_t)0xD9 >> (c1c2idx * 2)) & 3;  // {1, 2, 1, 3}
928
929
0
                X265_CHECK(!!((int)c1Idx < C1FLAG_NUMBER) == (int)((c1Idx - 8) >> (sizeof(int) * CHAR_BIT - 1)), "scan validation 1\n");
930
0
                X265_CHECK(!!(c2Idx == 0) == ((-(int)c2Idx) >> (sizeof(int) * CHAR_BIT - 1)) + 1, "scan validation 2\n");
931
0
                X265_CHECK((int)baseLevel == ((c1Idx < C1FLAG_NUMBER) ? (2 + (c2Idx == 0)) : 1), "scan validation 3\n");
932
0
                X265_CHECK(c1c2idx <= 3, "c1c2Idx check failure\n");
933
934
                // coefficient level estimation
935
0
                const int* levelAbsBits = estBitsSbac.levelAbsBits[ctxSet + c2];
936
0
                const uint32_t c1c2Rate = ((c1c2idx & 1) ?  greaterOneBits[1] : 0) + ((c1c2idx == 3) ? levelAbsBits[1] : 0);
937
938
0
                uint32_t level = 0;
939
0
                uint32_t sigCoefBits = 0;
940
0
                costCoeff[scanPos] = MAX_INT64;
941
942
0
                if ((int)scanPos == lastScanPos)
943
0
                    sigRateDelta[blkPos] = 0;
944
0
                else
945
0
                {
946
0
                    if (maxAbsLevel < 3)
947
0
                    {
948
                        /* set default costs to uncoded costs */
949
0
                        costSig[scanPos] = SIGCOST(estBitsSbac.significantBits[0][ctxSig]);
950
0
                        costCoeff[scanPos] = costUncoded[blkPos] + costSig[scanPos];
951
0
                    }
952
0
                    sigRateDelta[blkPos] = estBitsSbac.significantBits[1][ctxSig] - estBitsSbac.significantBits[0][ctxSig];
953
0
                    sigCoefBits = estBitsSbac.significantBits[1][ctxSig];
954
0
                }
955
956
0
                const uint32_t unQuantLevel = (maxAbsLevel * (unquantScale[blkPos] << per) + unquantRound);
957
                // NOTE: X265_MAX(maxAbsLevel - 1, 1) ==> (X>=2 -> X-1), (X<2 -> 1)  | (0 < X < 2 ==> X=1)
958
0
                if (maxAbsLevel == 1)
959
0
                {
960
0
                    uint32_t levelBits = (c1c2idx & 1) ? greaterOneBits[0] + IEP_RATE : ((1 + goRiceParam) << 15) + IEP_RATE;
961
0
                    X265_CHECK(levelBits == getICRateCost(1, 1 - baseLevel, greaterOneBits, levelAbsBits, goRiceParam, c1c2Rate) + IEP_RATE, "levelBits mistake\n");
962
963
0
                    int unquantAbsLevel = unQuantLevel >> unquantShift;
964
0
                    X265_CHECK(UNQUANT(1) == unquantAbsLevel, "DQuant check failed\n");
965
0
                    int d = abs(signCoef) - unquantAbsLevel;
966
0
                    int64_t curCost = RDCOST(d, sigCoefBits + levelBits);
967
968
                    /* Psy RDOQ: bias in favor of higher AC coefficients in the reconstructed frame */
969
0
                    if (usePsyMask & scanPos)
970
0
                    {
971
0
                        int reconCoef = abs(unquantAbsLevel + SIGN(predictedCoef, signCoef));
972
0
                        curCost -= PSYVALUE(reconCoef);
973
0
                    }
974
975
0
                    if (curCost < costCoeff[scanPos])
976
0
                    {
977
0
                        level = 1;
978
0
                        costCoeff[scanPos] = curCost;
979
0
                        costSig[scanPos] = SIGCOST(sigCoefBits);
980
0
                    }
981
0
                }
982
0
                else if (maxAbsLevel)
983
0
                {
984
0
                    uint32_t levelBits0 = getICRateCost(maxAbsLevel,     maxAbsLevel     - baseLevel, greaterOneBits, levelAbsBits, goRiceParam, c1c2Rate) + IEP_RATE;
985
0
                    uint32_t levelBits1 = getICRateCost(maxAbsLevel - 1, maxAbsLevel - 1 - baseLevel, greaterOneBits, levelAbsBits, goRiceParam, c1c2Rate) + IEP_RATE;
986
987
0
                    const uint32_t preDQuantLevelDiff = (unquantScale[blkPos] << per);
988
989
0
                    const int unquantAbsLevel0 = unQuantLevel >> unquantShift;
990
0
                    X265_CHECK(UNQUANT(maxAbsLevel) == (uint32_t)unquantAbsLevel0, "DQuant check failed\n");
991
0
                    int d0 = abs(signCoef) - unquantAbsLevel0;
992
0
                    int64_t curCost0 = RDCOST(d0, sigCoefBits + levelBits0);
993
994
0
                    const int unquantAbsLevel1 = (unQuantLevel - preDQuantLevelDiff) >> unquantShift;
995
0
                    X265_CHECK(UNQUANT(maxAbsLevel - 1) == (uint32_t)unquantAbsLevel1, "DQuant check failed\n");
996
0
                    int d1 = abs(signCoef) - unquantAbsLevel1;
997
0
                    int64_t curCost1 = RDCOST(d1, sigCoefBits + levelBits1);
998
999
                    /* Psy RDOQ: bias in favor of higher AC coefficients in the reconstructed frame */
1000
0
                    if (usePsyMask & scanPos)
1001
0
                    {
1002
0
                        int reconCoef;
1003
0
                        reconCoef = abs(unquantAbsLevel0 + SIGN(predictedCoef, signCoef));
1004
0
                        curCost0 -= PSYVALUE(reconCoef);
1005
1006
0
                        reconCoef = abs(unquantAbsLevel1 + SIGN(predictedCoef, signCoef));
1007
0
                        curCost1 -= PSYVALUE(reconCoef);
1008
0
                    }
1009
0
                    if (curCost0 < costCoeff[scanPos])
1010
0
                    {
1011
0
                        level = maxAbsLevel;
1012
0
                        costCoeff[scanPos] = curCost0;
1013
0
                        costSig[scanPos] = SIGCOST(sigCoefBits);
1014
0
                    }
1015
0
                    if (curCost1 < costCoeff[scanPos])
1016
0
                    {
1017
0
                        level = maxAbsLevel - 1;
1018
0
                        costCoeff[scanPos] = curCost1;
1019
0
                        costSig[scanPos] = SIGCOST(sigCoefBits);
1020
0
                    }
1021
0
                }
1022
1023
0
                dstCoeff[blkPos] = (int16_t)level;
1024
0
                totalRdCost += costCoeff[scanPos];
1025
1026
                /* record costs for sign-hiding performed at the end */
1027
0
                if ((cu.m_slice->m_pps->bSignHideEnabled ? ~0 : 0) & level)
1028
0
                {
1029
0
                    const int32_t diff0 = level - 1 - baseLevel;
1030
0
                    const int32_t diff2 = level + 1 - baseLevel;
1031
0
                    const int32_t maxVlc = g_goRiceRange[goRiceParam];
1032
0
                    int rate0, rate1, rate2;
1033
1034
0
                    if (diff0 < -2)  // prob (92.9, 86.5, 74.5)%
1035
0
                    {
1036
                        // NOTE: Min: L - 1 - {1,2,1,3} < -2 ==> L < {0,1,0,2}
1037
                        //            additional L > 0, so I got (L > 0 && L < 2) ==> L = 1
1038
0
                        X265_CHECK(level == 1, "absLevel check failure\n");
1039
1040
0
                        const int rateEqual2 = greaterOneBits[1] + levelAbsBits[0];;
1041
0
                        const int rateNotEqual2 = greaterOneBits[0];
1042
1043
0
                        rate0 = 0;
1044
0
                        rate2 = rateEqual2;
1045
0
                        rate1 = rateNotEqual2;
1046
1047
0
                        X265_CHECK(rate1 == getICRateNegDiff(level + 0, greaterOneBits, levelAbsBits), "rate1 check failure!\n");
1048
0
                        X265_CHECK(rate2 == getICRateNegDiff(level + 1, greaterOneBits, levelAbsBits), "rate1 check failure!\n");
1049
0
                        X265_CHECK(rate0 == getICRateNegDiff(level - 1, greaterOneBits, levelAbsBits), "rate1 check failure!\n");
1050
0
                    }
1051
0
                    else if (diff0 >= 0 && diff2 <= maxVlc)     // prob except from above path (98.6, 97.9, 96.9)%
1052
0
                    {
1053
                        // NOTE: no c1c2 correct rate since all of rate include this factor
1054
0
                        rate1 = getICRateLessVlc(level + 0, diff0 + 1, goRiceParam);
1055
0
                        rate2 = getICRateLessVlc(level + 1, diff0 + 2, goRiceParam);
1056
0
                        rate0 = getICRateLessVlc(level - 1, diff0 + 0, goRiceParam);
1057
0
                    }
1058
0
                    else
1059
0
                    {
1060
0
                        rate1 = getICRate(level + 0, diff0 + 1, greaterOneBits, levelAbsBits, goRiceParam, maxVlc, c1c2Rate);
1061
0
                        rate2 = getICRate(level + 1, diff0 + 2, greaterOneBits, levelAbsBits, goRiceParam, maxVlc, c1c2Rate);
1062
0
                        rate0 = getICRate(level - 1, diff0 + 0, greaterOneBits, levelAbsBits, goRiceParam, maxVlc, c1c2Rate);
1063
0
                    }
1064
0
                    rateIncUp[blkPos] = rate2 - rate1;
1065
0
                    rateIncDown[blkPos] = rate0 - rate1;
1066
0
                }
1067
0
                else
1068
0
                {
1069
0
                    rateIncUp[blkPos] = greaterOneBits[0];
1070
0
                    rateIncDown[blkPos] = 0;
1071
0
                }
1072
1073
                /* Update CABAC estimation state */
1074
0
                if ((level >= baseLevel) && (goRiceParam < 4) && (level > levelThreshold))
1075
0
                {
1076
0
                    goRiceParam++;
1077
0
                    levelThreshold <<= 1;
1078
0
                }
1079
1080
0
                const uint32_t isNonZero = (uint32_t)(-(int32_t)level) >> 31;
1081
0
                c1Idx += isNonZero;
1082
1083
                /* update bin model */
1084
0
                if (level > 1)
1085
0
                {
1086
0
                    c1 = 0;
1087
0
                    c2 += (uint32_t)(c2 - 2) >> 31;
1088
0
                    c2Idx++;
1089
0
                }
1090
0
                else if (((c1 == 1) | (c1 == 2)) & isNonZero)
1091
0
                    c1++;
1092
1093
0
                if (dstCoeff[blkPos])
1094
0
                {
1095
0
                    sigCoeffGroupFlag64 |= cgBlkPosMask;
1096
0
                    cgRdStats.codedLevelAndDist += costCoeff[scanPos] - costSig[scanPos];
1097
0
                    cgRdStats.uncodedDist += costUncoded[blkPos];
1098
0
                    cgRdStats.nnzBeforePos0 += scanPosinCG;
1099
0
                }
1100
0
            }
1101
1102
0
            cgRdStats.sigCost += costSig[scanPos];
1103
0
        } /* end for (scanPosinCG) */
1104
1105
0
        X265_CHECK((cgScanPos << MLS_CG_SIZE) == (int)scanPos, "scanPos mistake\n");
1106
0
        cgRdStats.sigCost0 = costSig[scanPos];
1107
1108
0
        costCoeffGroupSig[cgScanPos] = 0;
1109
1110
        /* nothing to do at this case */
1111
0
        X265_CHECK(cgLastScanPos >= 0, "cgLastScanPos check failure\n");
1112
1113
0
        if (!cgScanPos || cgScanPos == cgLastScanPos)
1114
0
        {
1115
            /* coeff group 0 is implied to be present, no signal cost */
1116
            /* coeff group with last NZ is implied to be present, handled below */
1117
0
        }
1118
0
        else if (sigCoeffGroupFlag64 & cgBlkPosMask)
1119
0
        {
1120
0
            if (!cgRdStats.nnzBeforePos0)
1121
0
            {
1122
                /* if only coeff 0 in this CG is coded, its significant coeff bit is implied */
1123
0
                totalRdCost -= cgRdStats.sigCost0;
1124
0
                cgRdStats.sigCost -= cgRdStats.sigCost0;
1125
0
            }
1126
1127
            /* there are coded coefficients in this group, but now we include the signaling cost
1128
             * of the significant coefficient group flag and evaluate whether the RD cost of the
1129
             * coded group is more than the RD cost of the uncoded group */
1130
1131
0
            uint32_t sigCtx = getSigCoeffGroupCtxInc(sigCoeffGroupFlag64, cgPosX, cgPosY, cgBlkPos, cgStride);
1132
1133
0
            int64_t costZeroCG = totalRdCost + SIGCOST(estBitsSbac.significantCoeffGroupBits[sigCtx][0]);
1134
0
            costZeroCG += cgRdStats.uncodedDist;       /* add distortion for resetting non-zero levels to zero levels */
1135
0
            costZeroCG -= cgRdStats.codedLevelAndDist; /* remove distortion and level cost of coded coefficients */
1136
0
            costZeroCG -= cgRdStats.sigCost;           /* remove signaling cost of significant coeff bitmap */
1137
1138
0
            costCoeffGroupSig[cgScanPos] = SIGCOST(estBitsSbac.significantCoeffGroupBits[sigCtx][1]);
1139
0
            totalRdCost += costCoeffGroupSig[cgScanPos];  /* add the cost of 1 bit in significant CG bitmap */
1140
1141
0
            if (costZeroCG < totalRdCost && m_rdoqLevel > 1)
1142
0
            {
1143
0
                sigCoeffGroupFlag64 &= ~cgBlkPosMask;
1144
0
                totalRdCost = costZeroCG;
1145
0
                costCoeffGroupSig[cgScanPos] = SIGCOST(estBitsSbac.significantCoeffGroupBits[sigCtx][0]);
1146
1147
                /* reset all coeffs to 0. UNCODE THIS COEFF GROUP! */
1148
0
                const uint32_t blkPos = codeParams.scan[cgScanPos * cgSize];
1149
0
                memset(&dstCoeff[blkPos + 0 * trSize], 0, 4 * sizeof(*dstCoeff));
1150
0
                memset(&dstCoeff[blkPos + 1 * trSize], 0, 4 * sizeof(*dstCoeff));
1151
0
                memset(&dstCoeff[blkPos + 2 * trSize], 0, 4 * sizeof(*dstCoeff));
1152
0
                memset(&dstCoeff[blkPos + 3 * trSize], 0, 4 * sizeof(*dstCoeff));
1153
0
            }
1154
0
        }
1155
0
        else
1156
0
        {
1157
            /* there were no coded coefficients in this coefficient group */
1158
0
            uint32_t ctxSig = getSigCoeffGroupCtxInc(sigCoeffGroupFlag64, cgPosX, cgPosY, cgBlkPos, cgStride);
1159
0
            costCoeffGroupSig[cgScanPos] = SIGCOST(estBitsSbac.significantCoeffGroupBits[ctxSig][0]);
1160
0
            totalRdCost += costCoeffGroupSig[cgScanPos];  /* add cost of 0 bit in significant CG bitmap */
1161
0
            totalRdCost -= cgRdStats.sigCost;             /* remove cost of significant coefficient bitmap */
1162
0
        }
1163
0
    } /* end for (cgScanPos) */
1164
1165
0
    X265_CHECK(lastScanPos >= 0, "numSig non zero, but no coded CG\n");
1166
1167
    /* calculate RD cost of uncoded block CBF=0, and add cost of CBF=1 to total */
1168
0
    int64_t bestCost;
1169
0
    if (!cu.isIntra(absPartIdx) && bIsLuma && !cu.m_tuDepth[absPartIdx])
1170
0
    {
1171
0
        bestCost = totalUncodedCost + SIGCOST(estBitsSbac.blockRootCbpBits[0]);
1172
0
        totalRdCost += SIGCOST(estBitsSbac.blockRootCbpBits[1]);
1173
0
    }
1174
0
    else
1175
0
    {
1176
0
        int ctx = ctxCbf[ttype][cu.m_tuDepth[absPartIdx]];
1177
0
        bestCost = totalUncodedCost + SIGCOST(estBitsSbac.blockCbpBits[ctx][0]);
1178
0
        totalRdCost += SIGCOST(estBitsSbac.blockCbpBits[ctx][1]);
1179
0
    }
1180
1181
    /* This loop starts with the last non-zero found in the first loop and then refines this last
1182
     * non-zero by measuring the true RD cost of the last NZ at this position, and then the RD costs
1183
     * at all previous coefficients until a coefficient greater than 1 is encountered or we run out
1184
     * of coefficients to evaluate.  This will factor in the cost of coding empty groups and empty
1185
     * coeff prior to the last NZ. The base best cost is the RD cost of CBF=0 */
1186
0
    int  bestLastIdx = 0;
1187
0
    bool foundLast = false;
1188
0
    for (int cgScanPos = cgLastScanPos; cgScanPos >= 0 && !foundLast; cgScanPos--)
1189
0
    {
1190
0
        if (!cgScanPos || cgScanPos == cgLastScanPos)
1191
0
        {
1192
            /* the presence of these coefficient groups are inferred, they have no bit in
1193
             * sigCoeffGroupFlag64 and no saved costCoeffGroupSig[] cost */
1194
0
        }
1195
0
        else if (sigCoeffGroupFlag64 & (1ULL << codeParams.scanCG[cgScanPos]))
1196
0
        {
1197
            /* remove cost of significant coeff group flag, the group's presence would be inferred
1198
             * from lastNZ if it were present in this group */
1199
0
            totalRdCost -= costCoeffGroupSig[cgScanPos];
1200
0
        }
1201
0
        else
1202
0
        {
1203
            /* remove cost of signaling this empty group as not present */
1204
0
            totalRdCost -= costCoeffGroupSig[cgScanPos];
1205
0
            continue;
1206
0
        }
1207
1208
0
        for (int scanPosinCG = cgSize - 1; scanPosinCG >= 0; scanPosinCG--)
1209
0
        {
1210
0
            scanPos = cgScanPos * cgSize + scanPosinCG;
1211
0
            if ((int)scanPos > lastScanPos)
1212
0
                continue;
1213
1214
            /* if the coefficient was coded, measure the RD cost of it as the last non-zero and then
1215
             * continue as if it were uncoded. If the coefficient was already uncoded, remove the
1216
             * cost of signaling it as not-significant */
1217
0
            uint32_t blkPos = codeParams.scan[scanPos];
1218
0
            if (dstCoeff[blkPos])
1219
0
            {
1220
                // Calculates the cost of signaling the last significant coefficient in the block 
1221
0
                uint32_t pos[2] = { (blkPos & (trSize - 1)), (blkPos >> log2TrSize) };
1222
0
                if (codeParams.scanType == SCAN_VER)
1223
0
                    std::swap(pos[0], pos[1]);
1224
0
                uint32_t bitsLastNZ = 0;
1225
1226
0
                for (int i = 0; i < 2; i++)
1227
0
                {
1228
0
                    int temp = g_lastCoeffTable[pos[i]];
1229
0
                    int prefixOnes = temp & 15;
1230
0
                    int suffixLen = temp >> 4;
1231
1232
0
                    bitsLastNZ += m_entropyCoder->m_estBitsSbac.lastBits[i][prefixOnes];
1233
0
                    bitsLastNZ += IEP_RATE * suffixLen;
1234
0
                }
1235
1236
0
                int64_t costAsLast = totalRdCost - costSig[scanPos] + SIGCOST(bitsLastNZ);
1237
1238
0
                if (costAsLast < bestCost)
1239
0
                {
1240
0
                    bestLastIdx = scanPos + 1;
1241
0
                    bestCost = costAsLast;
1242
0
                }
1243
0
                if (dstCoeff[blkPos] > 1 || m_rdoqLevel == 1)
1244
0
                {
1245
0
                    foundLast = true;
1246
0
                    break;
1247
0
                }
1248
1249
0
                totalRdCost -= costCoeff[scanPos];
1250
0
                totalRdCost += costUncoded[blkPos];
1251
0
            }
1252
0
            else
1253
0
                totalRdCost -= costSig[scanPos];
1254
0
        }
1255
0
    }
1256
1257
    /* recount non-zero coefficients and re-apply sign of DCT coef */
1258
0
    numSig = 0;
1259
0
    for (int pos = 0; pos < bestLastIdx; pos++)
1260
0
    {
1261
0
        int blkPos = codeParams.scan[pos];
1262
0
        int level  = dstCoeff[blkPos];
1263
0
        numSig += (level != 0);
1264
1265
0
        uint32_t mask = (int32_t)m_resiDctCoeff[blkPos] >> 31;
1266
0
        dstCoeff[blkPos] = (int16_t)((level ^ mask) - mask);
1267
0
    }
1268
1269
    // Average 49.62 pixels
1270
    /* clean uncoded coefficients */
1271
0
    X265_CHECK((uint32_t)(fastMin(lastScanPos, bestLastIdx) | (SCAN_SET_SIZE - 1)) < trSize * trSize, "array beyond bound\n");
1272
0
    for (int pos = bestLastIdx; pos <= (fastMin(lastScanPos, bestLastIdx) | (SCAN_SET_SIZE - 1)); pos++)
1273
0
    {
1274
0
        dstCoeff[codeParams.scan[pos]] = 0;
1275
0
    }
1276
0
    for (int pos = (bestLastIdx & ~(SCAN_SET_SIZE - 1)) + SCAN_SET_SIZE; pos <= lastScanPos; pos += SCAN_SET_SIZE)
1277
0
    {
1278
0
        const uint32_t blkPos = codeParams.scan[pos];
1279
0
        memset(&dstCoeff[blkPos + 0 * trSize], 0, 4 * sizeof(*dstCoeff));
1280
0
        memset(&dstCoeff[blkPos + 1 * trSize], 0, 4 * sizeof(*dstCoeff));
1281
0
        memset(&dstCoeff[blkPos + 2 * trSize], 0, 4 * sizeof(*dstCoeff));
1282
0
        memset(&dstCoeff[blkPos + 3 * trSize], 0, 4 * sizeof(*dstCoeff));
1283
0
    }
1284
1285
    /* rate-distortion based sign-hiding */
1286
0
    if (cu.m_slice->m_pps->bSignHideEnabled && numSig >= 2)
1287
0
    {
1288
0
        const int realLastScanPos = (bestLastIdx - 1) >> LOG2_SCAN_SET_SIZE;
1289
0
        int lastCG = 1;
1290
1291
0
        for (int subSet = realLastScanPos; subSet >= 0; subSet--)
1292
0
        {
1293
0
            int subPos = subSet << LOG2_SCAN_SET_SIZE;
1294
0
            int n;
1295
1296
0
            if (!(sigCoeffGroupFlag64 & (1ULL << codeParams.scanCG[subSet])))
1297
0
                continue;
1298
1299
            /* measure distance between first and last non-zero coef in this
1300
             * coding group */
1301
0
            const uint32_t posFirstLast = primitives.findPosFirstLast(&dstCoeff[codeParams.scan[subPos]], trSize, g_scan4x4[codeParams.scanType]);
1302
0
            const int firstNZPosInCG = (uint8_t)posFirstLast;
1303
0
            const int lastNZPosInCG = (int8_t)(posFirstLast >> 8);
1304
0
            const uint32_t absSumSign = posFirstLast;
1305
1306
0
            if (lastNZPosInCG - firstNZPosInCG >= SBH_THRESHOLD)
1307
0
            {
1308
0
                const int32_t signbit = ((int32_t)dstCoeff[codeParams.scan[subPos + firstNZPosInCG]]);
1309
1310
#if CHECKED_BUILD || _DEBUG
1311
                int32_t absSum_dummy = 0;
1312
                for (n = firstNZPosInCG; n <= lastNZPosInCG; n++)
1313
                    absSum_dummy += dstCoeff[codeParams.scan[n + subPos]];
1314
                X265_CHECK(((uint32_t)absSum_dummy & 1) == (absSumSign >> 31), "absSumSign check failure\n");
1315
#endif
1316
1317
                //if (signbit != absSumSign)
1318
0
                if (((int32_t)(signbit ^ absSumSign)) < 0)
1319
0
                {
1320
                    /* We must find a coeff to toggle up or down so the sign bit of the first non-zero coeff
1321
                     * is properly implied. Note dstCoeff[] are signed by this point but curChange and
1322
                     * finalChange imply absolute levels (+1 is away from zero, -1 is towards zero) */
1323
1324
0
                    int64_t minCostInc = MAX_INT64, curCost = MAX_INT64;
1325
0
                    uint32_t minPos = 0;
1326
0
                    int8_t finalChange = 0;
1327
0
                    int curChange = 0;
1328
0
                    uint32_t lastCoeffAdjust = (lastCG & (abs(dstCoeff[codeParams.scan[lastNZPosInCG + subPos]]) == 1)) * 4 * IEP_RATE;
1329
1330
0
                    for (n = (lastCG ? lastNZPosInCG : SCAN_SET_SIZE - 1); n >= 0; --n)
1331
0
                    {
1332
0
                        const uint32_t blkPos = codeParams.scan[n + subPos];
1333
0
                        const int32_t signCoef = m_resiDctCoeff[blkPos]; /* pre-quantization DCT coeff */
1334
0
                        const int absLevel = abs(dstCoeff[blkPos]);
1335
                        // TODO: this is constant in non-scaling mode
1336
0
                        const uint32_t preDQuantLevelDiff = (unquantScale[blkPos] << per);
1337
0
                        const uint32_t unQuantLevel = (absLevel * (unquantScale[blkPos] << per) + unquantRound);
1338
1339
0
                        int d = abs(signCoef) - (unQuantLevel >> unquantShift);
1340
0
                        X265_CHECK((uint32_t)UNQUANT(absLevel) == (unQuantLevel >> unquantShift), "dquant check failed\n");
1341
1342
0
                        const int64_t origDist = (((int64_t)d * d));
1343
1344
0
#define DELTARDCOST(d0, d, deltabits) ((((int64_t)d * d - d0) << scaleBits) + ((lambda2 * (int64_t)(deltabits)) >> 8))
1345
1346
0
                        const uint32_t isOne = (absLevel == 1);
1347
0
                        if (dstCoeff[blkPos])
1348
0
                        {
1349
0
                            d = abs(signCoef) - ((unQuantLevel + preDQuantLevelDiff) >> unquantShift);
1350
0
                            X265_CHECK((uint32_t)UNQUANT(absLevel + 1) == ((unQuantLevel + preDQuantLevelDiff) >> unquantShift), "dquant check failed\n");
1351
0
                            int64_t costUp = DELTARDCOST(origDist, d, rateIncUp[blkPos]);
1352
1353
                            /* if decrementing would make the coeff 0, we can include the
1354
                             * significant coeff flag cost savings */
1355
0
                            d = abs(signCoef) - ((unQuantLevel - preDQuantLevelDiff) >> unquantShift);
1356
0
                            X265_CHECK((uint32_t)UNQUANT(absLevel - 1) == ((unQuantLevel - preDQuantLevelDiff) >> unquantShift), "dquant check failed\n");
1357
0
                            int downBits = rateIncDown[blkPos] - (isOne ? (IEP_RATE + sigRateDelta[blkPos]) : 0);
1358
0
                            int64_t costDown = DELTARDCOST(origDist, d, downBits);
1359
1360
0
                            costDown -= lastCoeffAdjust;
1361
0
                            curCost = ((n == firstNZPosInCG) & isOne) ? MAX_INT64 : costDown;
1362
1363
0
                            curChange = 2 * (costUp < costDown) - 1;
1364
0
                            curCost = (costUp < costDown) ? costUp : curCost;
1365
0
                        }
1366
                        //else if ((n < firstNZPosInCG) & (signbit != ((uint32_t)signCoef >> 31)))
1367
0
                        else if ((n < firstNZPosInCG) & ((signbit ^ signCoef) < 0))
1368
0
                        {
1369
                            /* don't try to make a new coded coeff before the first coeff if its
1370
                             * sign would be different than the first coeff, the inferred sign would
1371
                             * still be wrong and we'd have to do this again. */
1372
0
                            curCost = MAX_INT64;
1373
0
                        }
1374
0
                        else
1375
0
                        {
1376
                            /* evaluate changing an uncoded coeff 0 to a coded coeff +/-1 */
1377
0
                            d = abs(signCoef) - ((preDQuantLevelDiff + unquantRound) >> unquantShift);
1378
0
                            X265_CHECK((uint32_t)UNQUANT(1) == ((preDQuantLevelDiff + unquantRound) >> unquantShift), "dquant check failed\n");
1379
0
                            curCost = DELTARDCOST(origDist, d, rateIncUp[blkPos] + IEP_RATE + sigRateDelta[blkPos]);
1380
0
                            curChange = 1;
1381
0
                        }
1382
1383
0
                        if (curCost < minCostInc)
1384
0
                        {
1385
0
                            minCostInc = curCost;
1386
0
                            finalChange = (int8_t)curChange;
1387
0
                            minPos = blkPos + (absLevel << 16);
1388
0
                        }
1389
0
                        lastCoeffAdjust = 0;
1390
0
                    }
1391
1392
0
                    const int absInMinPos = (minPos >> 16);
1393
0
                    minPos = (uint16_t)minPos;
1394
1395
                    // if (dstCoeff[minPos] == 32767 || dstCoeff[minPos] == -32768)
1396
0
                    if (absInMinPos >= 32767)
1397
                        /* don't allow sign hiding to violate the SPEC range */
1398
0
                        finalChange = -1;
1399
1400
                    // NOTE: Reference code
1401
                    //if (dstCoeff[minPos] == 0)
1402
                    //    numSig++;
1403
                    //else if (finalChange == -1 && abs(dstCoeff[minPos]) == 1)
1404
                    //    numSig--;
1405
0
                    numSig += (absInMinPos == 0) - ((finalChange == -1) & (absInMinPos == 1));
1406
1407
1408
                    // NOTE: Reference code
1409
                    //if (m_resiDctCoeff[minPos] >= 0)
1410
                    //    dstCoeff[minPos] += finalChange;
1411
                    //else
1412
                    //    dstCoeff[minPos] -= finalChange;
1413
0
                    const int16_t resiCoeffSign = ((int16_t)m_resiDctCoeff[minPos] >> 16);
1414
0
                    dstCoeff[minPos] += (((int16_t)finalChange ^ resiCoeffSign) - resiCoeffSign);
1415
0
                }
1416
0
            }
1417
1418
0
            lastCG = 0;
1419
0
        }
1420
0
    }
1421
1422
0
    return numSig;
1423
0
}
Unexecuted instantiation: unsigned int x265::Quant::rdoQuant<2u>(x265::CUData const&, short*, x265::TextType, unsigned int, bool)
Unexecuted instantiation: unsigned int x265::Quant::rdoQuant<3u>(x265::CUData const&, short*, x265::TextType, unsigned int, bool)
Unexecuted instantiation: unsigned int x265::Quant::rdoQuant<4u>(x265::CUData const&, short*, x265::TextType, unsigned int, bool)
Unexecuted instantiation: unsigned int x265::Quant::rdoQuant<5u>(x265::CUData const&, short*, x265::TextType, unsigned int, bool)
1424
1425
/* Context derivation process of coeff_abs_significant_flag */
1426
uint32_t Quant::getSigCtxInc(uint32_t patternSigCtx, uint32_t log2TrSize, uint32_t trSize, uint32_t blkPos, bool bIsLuma,
1427
                             uint32_t firstSignificanceMapContext)
1428
0
{
1429
0
    static const uint8_t ctxIndMap[16] =
1430
0
    {
1431
0
        0, 1, 4, 5,
1432
0
        2, 3, 4, 5,
1433
0
        6, 6, 8, 8,
1434
0
        7, 7, 8, 8
1435
0
    };
1436
1437
0
    if (!blkPos) // special case for the DC context variable
1438
0
        return 0;
1439
1440
0
    if (log2TrSize == 2) // 4x4
1441
0
        return ctxIndMap[blkPos];
1442
1443
0
    const uint32_t posY = blkPos >> log2TrSize;
1444
0
    const uint32_t posX = blkPos & (trSize - 1);
1445
0
    X265_CHECK((blkPos - (posY << log2TrSize)) == posX, "block pos check failed\n");
1446
1447
0
    int posXinSubset = blkPos & 3;
1448
0
    X265_CHECK((posX & 3) == (blkPos & 3), "pos alignment fail\n");
1449
0
    int posYinSubset = posY & 3;
1450
1451
    // NOTE: [patternSigCtx][posXinSubset][posYinSubset]
1452
0
    static const uint8_t table_cnt[4][4][4] =
1453
0
    {
1454
        // patternSigCtx = 0
1455
0
        {
1456
0
            { 2, 1, 1, 0 },
1457
0
            { 1, 1, 0, 0 },
1458
0
            { 1, 0, 0, 0 },
1459
0
            { 0, 0, 0, 0 },
1460
0
        },
1461
        // patternSigCtx = 1
1462
0
        {
1463
0
            { 2, 1, 0, 0 },
1464
0
            { 2, 1, 0, 0 },
1465
0
            { 2, 1, 0, 0 },
1466
0
            { 2, 1, 0, 0 },
1467
0
        },
1468
        // patternSigCtx = 2
1469
0
        {
1470
0
            { 2, 2, 2, 2 },
1471
0
            { 1, 1, 1, 1 },
1472
0
            { 0, 0, 0, 0 },
1473
0
            { 0, 0, 0, 0 },
1474
0
        },
1475
        // patternSigCtx = 3
1476
0
        {
1477
0
            { 2, 2, 2, 2 },
1478
0
            { 2, 2, 2, 2 },
1479
0
            { 2, 2, 2, 2 },
1480
0
            { 2, 2, 2, 2 },
1481
0
        }
1482
0
    };
1483
1484
0
    int cnt = table_cnt[patternSigCtx][posXinSubset][posYinSubset];
1485
0
    int offset = firstSignificanceMapContext;
1486
1487
0
    offset += cnt;
1488
1489
0
    return (bIsLuma && (posX | posY) >= 4) ? 3 + offset : offset;
1490
0
}
1491