Pretty printed the C++ code with astyle

Ethan Hugg committed 12 years ago

Commit ff6b6691 parent 063709c225 files · +48457 −51954Browse files

Showing the changes to codec/encoder/core/src/slice_multi_threading.cpp (1 of 225 files). Show all

Unified · Split

codec/encoder/core/src/slice_multi_threading.cpp+1506−1592

@@ -1,1592 +1,1506 @@
1		−/*!
2		− * \copy
3		− * Copyright (c) 2010-2013, Cisco Systems
4		− * All rights reserved.
5		− *
6		− * Redistribution and use in source and binary forms, with or without
7		− * modification, are permitted provided that the following conditions
8		− * are met:
9		− *
10		− * * Redistributions of source code must retain the above copyright
11		− * notice, this list of conditions and the following disclaimer.
12		− *
13		− * * Redistributions in binary form must reproduce the above copyright
14		− * notice, this list of conditions and the following disclaimer in
15		− * the documentation and/or other materials provided with the
16		− * distribution.
17		− *
18		− * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19		− * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20		− * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
21		− * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
22		− * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23		− * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
24		− * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25		− * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
26		− * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27		− * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
28		− * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29		− * POSSIBILITY OF SUCH DAMAGE.
30		− *
31		− *
32		− * \file slice_multi_threading.h
33		− *
34		− * \brief pSlice based multiple threading
35		− *
36		− * \date 04/16/2010 Created
37		− *
38		− *************************************************************************************
39		− */
40		−
41		−#if defined(MT_ENABLED)
42		−
43		−#include <assert.h>
44		−#ifdef __GNUC__
45		−#include <semaphore.h>
46		−#ifndef SEM_NAME_MAX
47		−// length of semaphore name should be system constrained at least on mac 10.7
48		−#define SEM_NAME_MAX 32
49		−#endif//SEM_NAME_MAX
50		−#endif//__GNUC__
51		−#include "slice_multi_threading.h"
52		−#include "mt_defs.h"
53		−#include "nal_encap.h"
54		−#include "utils.h"
55		−#include "encoder.h"
56		−#include "svc_encode_slice.h"
57		−#include "deblocking.h"
58		−#include "svc_enc_golomb.h"
59		−#include "crt_util_safe_x.h" // for safe crt like calls
60		−#include "rc.h"
61		−
62		−#if defined(X86_ASM)
63		−#include "cpu.h"
64		−#endif//X86_ASM
65		−
66		−#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
67		−#include "measure_time.h"
68		−#endif//DYNAMIC_SLICE_ASSIGN
69		−namespace WelsSVCEnc {
70		−void UpdateMbListNeighborParallel( SSliceCtx *pSliceCtx,
71		− SMB *pMbList,
72		− const int32_t uiSliceIdc )
73		−{
74		− const uint8_t *kpMbMap = pSliceCtx->pOverallMbMap;
75		− const int32_t kiMbWidth = pSliceCtx->iMbWidth;
76		− int32_t iIdx = pSliceCtx->pFirstMbInSlice[uiSliceIdc];
77		− const int32_t kiEndMbInSlice = iIdx + pSliceCtx->pCountMbNumInSlice[uiSliceIdc] - 1;
78		−
79		− do {
80		− SMB *pMb = &pMbList[iIdx];
81		− uint32_t uiNeighborAvailFlag = 0;
82		− const int32_t kiMbXY = pMb->iMbXY;
83		− const int32_t kiMbX = pMb->iMbX;
84		− const int32_t kiMbY = pMb->iMbY;
85		− BOOL_T bLeft;
86		− BOOL_T bTop;
87		− BOOL_T bLeftTop;
88		− BOOL_T bRightTop;
89		− int32_t iLeftXY, iTopXY, iLeftTopXY, iRightTopXY;
90		−
91		− iLeftXY = kiMbXY - 1;
92		− iTopXY = kiMbXY - kiMbWidth;
93		− iLeftTopXY = iTopXY - 1;
94		− iRightTopXY = iTopXY + 1;
95		−
96		− bLeft = (kiMbX > 0) && (uiSliceIdc == kpMbMap[iLeftXY]);
97		− bTop = (kiMbY > 0) && (uiSliceIdc == kpMbMap[iTopXY]);
98		− bLeftTop = (kiMbX > 0) && (kiMbY > 0) && (uiSliceIdc == kpMbMap[iLeftTopXY]);
99		− bRightTop = (kiMbX < (kiMbWidth-1)) && (kiMbY > 0) && (uiSliceIdc == kpMbMap[iRightTopXY]);
100		−
101		− if( bLeft ){
102		− uiNeighborAvailFlag \|= LEFT_MB_POS;
103		− }
104		− if( bTop ){
105		− uiNeighborAvailFlag \|= TOP_MB_POS;
106		− }
107		− if( bLeftTop ){
108		− uiNeighborAvailFlag \|= TOPLEFT_MB_POS;
109		− }
110		− if( bRightTop ){
111		− uiNeighborAvailFlag \|= TOPRIGHT_MB_POS;
112		− }
113		− pMb->uiNeighborAvail = (uint8_t)uiNeighborAvailFlag;
114		− pMb->uiSliceIdc = uiSliceIdc;
115		−
116		− ++ iIdx;
117		− } while(iIdx <= kiEndMbInSlice);
118		−}
119		−
120		−void CalcSliceComplexRatio( void pRatio, SSliceCtx pSliceCtx, uint32_t *pSliceConsume )
121		−{
122		− float pRatioList = (float )pRatio;
123		− float fAvI[MAX_SLICES_NUM];
124		− float fSumAv = .0f;
125		− uint32_t pSliceTime = (uint32_t )pSliceConsume;
126		− int32_t pCountMbInSlice = (int32_t )pSliceCtx->pCountMbNumInSlice;
127		− const int32_t kiSliceCount = pSliceCtx->iSliceNumInFrame;
128		− int32_t iSliceIdx = 0;
129		−
130		−#if defined(X86_ASM)
131		− WelsEmms();
132		−#endif //X86_ASM
133		−
134		− while ( iSliceIdx < kiSliceCount )
135		− {
136		− fAvI[iSliceIdx] = 1.0f * pCountMbInSlice[iSliceIdx] / pSliceTime[iSliceIdx];
137		−#if defined(ENABLE_TRACE_MT)
138		− WelsLog(NULL, WELS_LOG_DEBUG, "[MT] CalcSliceComplexRatio(), pSliceConsumeTime[%d]= %d us, slice_run= %d\n", iSliceIdx, pSliceTime[iSliceIdx], pCountMbInSlice[iSliceIdx]);
139		−#endif//ENABLE_TRACE_MT
140		− fSumAv += fAvI[iSliceIdx];
141		−
142		− ++ iSliceIdx;
143		− }
144		− while ( -- iSliceIdx >= 0 )
145		− {
146		− pRatioList[iSliceIdx] = fAvI[iSliceIdx] / fSumAv;
147		− }
148		−}
149		−
150		−#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN) && defined(NOT_ABSOLUTE_BALANCING)
151		−int32_t NeedDynamicAdjust( void *pConsumeTime, const int32_t iSliceNum )
152		−{
153		−#if !defined(USE_RMSE_SLICE_COMPLEXITY_RATIO_FOR_BALANCING)
154		− const float fRatioLower = TOLERANT_BALANCING_RATIO_LOWER( uiSliceNum );
155		− const float fRatioUpper = TOLERANT_BALANCING_RATIO_UPPER( uiSliceNum );
156		−#endif//USE_RMSE_SLICE_COMPLEXITY_RATIO_FOR_BALANCING
157		− uint32_t pSliceConsume = (uint32_t )pConsumeTime;
158		− uint32_t uiTotalConsume = 0;
159		− int32_t iSliceIdx = 0;
160		− int32_t iNeedAdj = false;
161		−
162		−#if defined(X86_ASM)
163		− WelsEmms();
164		−#endif //X86_ASM
165		−
166		− while( iSliceIdx < iSliceNum )
167		− {
168		− uiTotalConsume += pSliceConsume[iSliceIdx] + pSliceConsume[1+iSliceIdx];
169		− iSliceIdx += 2;
170		− }
171		− if (uiTotalConsume == 0)
172		− {
173		−#if defined(ENABLE_TRACE_MT)
174		− WelsLog( NULL, WELS_LOG_DEBUG, "[MT] NeedDynamicAdjust(), herein do no adjust due first picture, iCountSliceNum= %d\n", iSliceNum );
175		−#endif//ENABLE_TRACE_MT
176		− return false;
177		− }
178		−
179		− iSliceIdx = 0;
180		−#if defined(USE_RMSE_SLICE_COMPLEXITY_RATIO_FOR_BALANCING)
181		− float fThr = EPSN; // threshold for various cores cases
182		− float fRmse = .0f; // root mean square error of pSlice consume ratios
183		− const float kfMeanRatio = 1.0f / iSliceNum;
184		− do{
185		− const float fRatio = 1.0f * pSliceConsume[iSliceIdx] / uiTotalConsume;
186		− const float fDiffRatio = fRatio - kfMeanRatio;
187		− fRmse += (fDiffRatio * fDiffRatio);
188		− ++ iSliceIdx;
189		− } while ( iSliceIdx+1 < iSliceNum );
190		− fRmse = sqrtf(fRmse/iSliceNum);
191		− if ( iSliceNum >= 8 )
192		− {
193		− fThr += THRESHOLD_RMSE_CORE8;
194		− }
195		− else if ( iSliceNum >= 4 )
196		− {
197		− fThr += THRESHOLD_RMSE_CORE4;
198		− }
199		− else if ( iSliceNum >= 2 )
200		− {
201		− fThr += THRESHOLD_RMSE_CORE2;
202		− }
203		− else
204		− fThr = 1.0f;
205		− if ( fRmse > fThr )
206		− iNeedAdj = true;
207		−#if defined(ENABLE_TRACE_MT)
208		− WelsLog(NULL, WELS_LOG_DEBUG, "[MT] NeedDynamicAdjust(), herein adjustment decision is made (iNeedAdj= %d) by: fRmse of pSlice complexity ratios %.6f, the corresponding threshold %.6f, iCountSliceNum %d\n",
209		− iNeedAdj, fRmse, fThr, iSliceNum);
210		−#endif//ENABLE_TRACE_MT
211		−#else
212		− do{
213		− const float kfRatio = 1.0f * pSliceConsume[uiSliceIdx] / uiTotalConsume;
214		− if ( kfRatio+EPSN < fRatioLower \|\| kfRatio > ratio_upper+EPSN )
215		− {
216		−#if defined(ENABLE_TRACE_MT)
217		− WelsLog(NULL, WELS_LOG_DEBUG, "[MT] NeedDynamicAdjust(), herein adjustment decision is made by pSlice consume time not balanced at all, uiSliceIdx= %d, comp_ratio= %.6f, pSliceConsumeTime= %d, total_consume_time= %d, iCountSliceNum= %d\n",
218		− uiSliceIdx, kfRatio, pSliceConsume[uiSliceIdx], uiTotalConsume, uiSliceNum);
219		−#endif//ENABLE_TRACE_MT
220		− iNeedAdj = true;
221		− break;
222		− }
223		− ++ uiSliceIdx;
224		− } while ( uiSliceIdx+1 < uiSliceNum );
225		−#endif//USE_RMSE_SLICE_COMPLEXITY_RATIO_FOR_BALANCING
226		−
227		− return iNeedAdj;
228		−}
229		−#endif//..
230		−
231		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
232		−void DynamicAdjustSlicing( sWelsEncCtx *pCtx,
233		− SDqLayer *pCurDqLayer,
234		− void *pComplexRatio,
235		− int32_t iCurDid )
236		−{
237		− SSliceCtx *pSliceCtx = pCurDqLayer->pSliceEncCtx;
238		− const int32_t kiCountSliceNum = pSliceCtx->iSliceNumInFrame;
239		− const int32_t kiCountNumMb = pSliceCtx->iMbNumInFrame;
240		− int32_t iMinimalMbNum = pSliceCtx->iMbWidth; // in theory we need only 1 SMB, here let it as one SMB row required
241		− int32_t iMaximalMbNum = 0; // dynamically assign later
242		− float pSliceComplexRatio = (float )pComplexRatio;
243		− int32_t iMbNumLeft = kiCountNumMb;
244		− int32_t iRunLen[MAX_THREADS_NUM] = {0};
245		− int32_t iSliceIdx = 0;
246		−
247		− int32_t iNumMbInEachGom;
248		− SWelsSvcRc *pWelsSvcRc = &pCtx->pWelsSvcRc[iCurDid];
249		− if(pCtx->pSvcParam->bEnableRc)
250		− {
251		− iNumMbInEachGom = pWelsSvcRc->iNumberMbGom;
252		−
253		− if ( iNumMbInEachGom <= 0 )
254		− {
255		− WelsLog(pCtx, WELS_LOG_ERROR, "[MT] DynamicAdjustSlicing(), invalid iNumMbInEachGom= %d from RC, iDid= %d, iCountNumMb= %d\n", iNumMbInEachGom, iCurDid, kiCountNumMb);
256		− return;
257		− }
258		−
259		− // do not adjust in case no extra iNumMbInEachGom based left for slicing adjustment,
260		− // extra MB of non integrated GOM assigned at the last pSlice in default, keep up on early initial result.
261		− if ( iNumMbInEachGom * kiCountSliceNum >= kiCountNumMb )
262		− {
263		− return;
264		− }
265		− iMinimalMbNum = iNumMbInEachGom;
266		− }
267		−
268		− if ( kiCountSliceNum < 2 \|\| (kiCountSliceNum & 0x01) ) // we need suppose uiSliceNum is even for multiple threading
269		− return;
270		−
271		− iMaximalMbNum = kiCountNumMb - (kiCountSliceNum - 1) * iMinimalMbNum;
272		−
273		−#if defined(X86_ASM)
274		− WelsEmms();
275		−#endif //X86_ASM
276		−
277		−#if defined(ENABLE_TRACE_MT)
278		− WelsLog(pCtx, WELS_LOG_DEBUG, "[MT] DynamicAdjustSlicing(), iDid= %d, iCountNumMb= %d\n", iCurDid, kiCountNumMb);
279		−#endif//ENABLE_TRACE_MT
280		−
281		− iSliceIdx = 0;
282		− while (iSliceIdx+1 < kiCountSliceNum) {
283		− int32_t iNumMbAssigning = (int32_t)(kiCountNumMb * pSliceComplexRatio[iSliceIdx] + EPSN);
284		−
285		− // GOM boundary aligned
286		− if(pCtx->pSvcParam->bEnableRc)
287		− {
288		− iNumMbAssigning=(int32_t)(1.0f * iNumMbAssigning / iNumMbInEachGom + 0.5f + EPSN) * iNumMbInEachGom;
289		− }
290		−
291		− // make sure one GOM at least in each pSlice for safe
292		− if ( iNumMbAssigning < iMinimalMbNum )
293		− iNumMbAssigning = iMinimalMbNum;
294		− else if ( iNumMbAssigning > iMaximalMbNum )
295		− iNumMbAssigning = iMaximalMbNum;
296		−
297		− assert( iNumMbAssigning > 0 );
298		−
299		− iMbNumLeft -= iNumMbAssigning;
300		− if ( iMbNumLeft <= 0 ) // error due to we can not support slice_skip now yet, do not adjust this time
301		− {
302		− assert( 0 );
303		− return;
304		− }
305		− iRunLen[iSliceIdx] = iNumMbAssigning;
306		−#if defined(ENABLE_TRACE_MT)
307		− WelsLog(pCtx, WELS_LOG_DEBUG, "[MT] DynamicAdjustSlicing(), uiSliceIdx= %d, pSliceComplexRatio= %.2f, slice_run_org= %d, slice_run_adj= %d\n",
308		− iSliceIdx, pSliceComplexRatio[iSliceIdx], pSliceCtx->pCountMbNumInSlice[iSliceIdx], iNumMbAssigning);
309		−#endif//ENABLE_TRACE_MT
310		− ++ iSliceIdx;
311		− iMaximalMbNum = iMbNumLeft - (kiCountSliceNum - iSliceIdx - 1) * iMinimalMbNum; // get maximal num_mb in left parts
312		− }
313		− iRunLen[iSliceIdx] = iMbNumLeft;
314		−#if defined(ENABLE_TRACE_MT)
315		− WelsLog(pCtx, WELS_LOG_DEBUG, "[MT] DynamicAdjustSlicing(), iSliceIdx= %d, pSliceComplexRatio= %.2f, slice_run_org= %d, slice_run_adj= %d\n",
316		− iSliceIdx, pSliceComplexRatio[iSliceIdx], pSliceCtx->pCountMbNumInSlice[iSliceIdx], iMbNumLeft);
317		−#endif//ENABLE_TRACE_MT
318		−
319		−
320		− if ( DynamicAdjustSlicePEncCtxAll( pSliceCtx, iRunLen ) == 0 )
321		− {
322		− const int32_t kiThreadNum = pCtx->pSvcParam->iCountThreadsNum;
323		− int32_t iThreadIdx = 0;
324		− do {
325		−#ifdef WIN32
326		− WelsEventSignal( &pCtx->pSliceThreading->pUpdateMbListEvent[iThreadIdx] );
327		−#else
328		− WelsEventSignal( pCtx->pSliceThreading->pUpdateMbListEvent[iThreadIdx] );
329		−#endif//WIN32
330		− ++ iThreadIdx;
331		− } while(iThreadIdx < kiThreadNum);
332		−
333		− WelsMultipleEventsWaitAllBlocking( kiThreadNum, &pCtx->pSliceThreading->pFinUpdateMbListEvent[0] );
334		− }
335		−}
336		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
337		−
338		−#ifdef PACKING_ONE_SLICE_PER_LAYER
339		−void ResetEnvMt( sWelsEncCtx *pCtx)
340		−{
341		− const int16_t kiSliceCount = pCtx->iMaxSliceCount;
342		− int32_t iIdx = 0;
343		−
344		− while ( iIdx < kiSliceCount )
345		− {
346		− SWelsSliceBs *pSliceBs = &pCtx->pSliceBs[iIdx];
347		− pSliceBs->uiBsPos = 0;
348		− ++ iIdx;
349		− }
350		−}
351		−#endif//PACKING_ONE_SLICE_PER_LAYER
352		−
353		−int32_t RequestMtResource( sWelsEncCtx *ppCtx, SWelsSvcCodingParam pCodingParam, const int32_t iCountBsLen, const int32_t iTargetSpatialBsSize )
354		−{
355		− CMemoryAlign *pMa = NULL;
356		− SWelsSvcCodingParam *pPara= NULL;
357		− SSliceThreading *pSmt = NULL;
358		− SWelsSliceBs *pSliceB = NULL;
359		− uint8_t *pBsBase = NULL;
360		− int32_t iNumSpatialLayers = 0;
361		− int32_t iThreadNum = 0;
362		− int32_t iIdx = 0;
363		− int32_t iSliceBsBufferSize= 0;
364		− int16_t iMaxSliceNum = 1;
365		−
366		− if ( NULL == ppCtx \|\| NULL == pCodingParam \|\| NULL == *ppCtx \|\| iCountBsLen <= 0 )
367		− return 1;
368		−
369		− pMa = (*ppCtx)->pMemAlign;
370		− pPara= pCodingParam;
371		− iNumSpatialLayers = pPara->iNumDependencyLayer;
372		− iThreadNum = pPara->iCountThreadsNum;
373		− iMaxSliceNum = (*ppCtx)->iMaxSliceCount;
374		−
375		− pSmt = (SSliceThreading *)pMa->WelsMalloc(sizeof(SSliceThreading), "SSliceThreading");
376		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt), FreeMemorySvc(ppCtx) )
377		− (*ppCtx)->pSliceThreading = pSmt;
378		− pSmt->pThreadPEncCtx = (SSliceThreadPrivateData )pMa->WelsMalloc( sizeof(SSliceThreadPrivateData) iThreadNum, "pThreadPEncCtx" );
379		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pThreadPEncCtx), FreeMemorySvc(ppCtx) )
380		− pSmt->pThreadHandles = (WELS_THREAD_HANDLE )pMa->WelsMalloc( sizeof(WELS_THREAD_HANDLE) iThreadNum, "pThreadHandles" );
381		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pThreadHandles), FreeMemorySvc(ppCtx) )
382		−
383		−#ifdef WIN32
384		− pSmt->pSliceCodedEvent = (WELS_EVENT )pMa->WelsMalloc( sizeof(WELS_EVENT) iThreadNum, "pSliceCodedEvent" );
385		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pSliceCodedEvent), FreeMemorySvc(ppCtx) )
386		− pSmt->pReadySliceCodingEvent = (WELS_EVENT )pMa->WelsMalloc( sizeof(WELS_EVENT) iThreadNum, "pReadySliceCodingEvent" );
387		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pReadySliceCodingEvent), FreeMemorySvc(ppCtx) )
388		− pSmt->pFinSliceCodingEvent = (WELS_EVENT )pMa->WelsMalloc( sizeof(WELS_EVENT) iThreadNum, "pFinSliceCodingEvent" );
389		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pFinSliceCodingEvent), FreeMemorySvc(ppCtx) )
390		−#endif//WIN32
391		−
392		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
393		−#if defined(__GNUC__)
394		− pSmt->pUpdateMbListThrdHandles = (WELS_THREAD_HANDLE )pMa->WelsMalloc( sizeof(WELS_THREAD_HANDLE) iThreadNum, "pUpdateMbListThrdHandles" );
395		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pUpdateMbListThrdHandles), FreeMemorySvc(ppCtx) )
396		−#endif//__GNUC__
397		−#ifdef WIN32
398		− pSmt->pUpdateMbListEvent = (WELS_EVENT )pMa->WelsMalloc( sizeof(WELS_EVENT) iThreadNum, "pUpdateMbListEvent" );
399		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pUpdateMbListEvent), FreeMemorySvc(ppCtx) )
400		− pSmt->pFinUpdateMbListEvent = (WELS_EVENT )pMa->WelsMalloc( sizeof(WELS_EVENT) iThreadNum, "pFinUpdateMbListEvent" );
401		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pFinUpdateMbListEvent), FreeMemorySvc(ppCtx) )
402		−#endif//WIN32
403		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
404		−
405		−#ifdef WIN32
406		− pSmt->pExitEncodeEvent = (WELS_EVENT )pMa->WelsMalloc( sizeof(WELS_EVENT) iThreadNum, "pExitEncodeEvent" );
407		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pExitEncodeEvent), FreeMemorySvc(ppCtx) )
408		−#endif//WIN32
409		−
410		−#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
411		− iIdx = 0;
412		− while ( iIdx < iNumSpatialLayers )
413		− {
414		− SMulSliceOption *pMso = &pPara->sDependencyLayers[iIdx].sMso;
415		− const int32_t kiSliceNum= pMso->sSliceArgument.iSliceNum;
416		− if (pMso->uiSliceMode == SM_FIXEDSLCNUM_SLICE && pPara->iMultipleThreadIdc > 1 && pPara->iMultipleThreadIdc >= kiSliceNum )
417		− {
418		− pSmt->pSliceConsumeTime[iIdx] = (uint32_t )pMa->WelsMallocz( kiSliceNum sizeof(uint32_t), "pSliceConsumeTime[]" );
419		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pSliceConsumeTime[iIdx]), FreeMemorySvc(ppCtx) )
420		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
421		− pSmt->pSliceComplexRatio[iIdx] = (float )pMa->WelsMalloc( kiSliceNum sizeof(float), "pSliceComplexRatio[]" );
422		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pSliceComplexRatio[iIdx]), FreeMemorySvc(ppCtx) )
423		−#endif//TRY_SLICING_BALANCE
424		− }
425		− else
426		− {
427		− pSmt->pSliceConsumeTime[iIdx] = NULL;
428		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
429		− pSmt->pSliceComplexRatio[iIdx] = NULL;
430		−#endif//TRY_SLICING_BALANCE
431		− }
432		− ++ iIdx;
433		− }
434		− // NULL for pSliceConsumeTime[iIdx]: iIdx from iNumSpatialLayers to MAX_DEPENDENCY_LAYERS
435		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
436		−
437		−#ifdef MT_DEBUG
438		− // file handle for MT debug
439		− pSmt->pFSliceDiff = NULL;
440		−
441		− if ( pSmt->pFSliceDiff )
442		− {
443		− fclose( pSmt->pFSliceDiff );
444		− pSmt->pFSliceDiff = NULL;
445		− }
446		−#ifdef WIN32
447		− pSmt->pFSliceDiff = fopen(".\\slice_time.txt", "wt+" );
448		−#else
449		− pSmt->pFSliceDiff = fopen("/tmp/slice_time.txt", "wt+" );
450		−#endif//WIN32
451		−#endif//MT_DEBUG
452		−
453		−#if defined(ENABLE_TRACE_MT)
454		− WelsLog((ppCtx), WELS_LOG_INFO, "encpEncCtx= 0x%p\n", (void )(*ppCtx));
455		−#endif//ENABLE_TRACE_MT
456		−
457		− iIdx = 0;
458		− while ( iIdx < iThreadNum )
459		− {
460		−#ifdef __GNUC__ // for posix threading
461		− str_t name[SEM_NAME_MAX] = {0};
462		− int32_t used_len = 0;
463		− WELS_THREAD_ERROR_CODE err = 0;
464		−#endif//__GNUC__
465		− pSmt->pThreadPEncCtx[iIdx].pWelsPEncCtx = (void )(ppCtx);
466		− pSmt->pThreadPEncCtx[iIdx].iSliceIndex = iIdx;
467		− pSmt->pThreadPEncCtx[iIdx].iThreadIndex = iIdx;
468		− pSmt->pThreadHandles[iIdx] = 0;
469		−
470		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
471		−#ifdef WIN32
472		− WelsEventInit( &pSmt->pUpdateMbListEvent[iIdx] );
473		− WelsEventInit( &pSmt->pFinUpdateMbListEvent[iIdx] );
474		−#else
475		− // length of semaphore name should be system constrained at least on mac 10.7
476		− SNPRINTF( name, SEM_NAME_MAX, "ud%d%p", iIdx, (void )(ppCtx) );
477		− err = WelsEventOpen( &pSmt->pUpdateMbListEvent[iIdx], name );
478		−#if defined(ENABLE_TRACE_MT)
479		− WelsLog((*ppCtx), WELS_LOG_INFO, "[MT] Open pUpdateMbListEvent%d named(%s) ret%d err%d\n", iIdx, name, err, errno);
480		−#endif
481		− used_len = SNPRINTF( name, SEM_NAME_MAX, "fu%d%p", iIdx, (void )(ppCtx) );
482		− name[used_len] = '\0';
483		− err = WelsEventOpen( &pSmt->pFinUpdateMbListEvent[iIdx], name );
484		−#if defined(ENABLE_TRACE_MT)
485		− WelsLog((*ppCtx), WELS_LOG_INFO, "[MT] Open pFinUpdateMbListEvent%d named(%s) ret%d err%d\n", iIdx, name, err, errno);
486		−#endif
487		−#endif//WIN32
488		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
489		−
490		−#ifdef WIN32
491		− WelsEventInit( &pSmt->pSliceCodedEvent[iIdx] );
492		− WelsEventInit( &pSmt->pReadySliceCodingEvent[iIdx] );
493		− WelsEventInit( &pSmt->pFinSliceCodingEvent[iIdx] );
494		− WelsEventInit( &pSmt->pExitEncodeEvent[iIdx] );
495		−#else
496		− used_len = SNPRINTF( name, SEM_NAME_MAX, "sc%d%p", iIdx, (void )(ppCtx) );
497		− name[used_len] = '\0';
498		− err = WelsEventOpen( &pSmt->pSliceCodedEvent[iIdx], name );
499		−#if defined(ENABLE_TRACE_MT)
500		− WelsLog((*ppCtx), WELS_LOG_INFO, "[MT] Open pSliceCodedEvent%d named(%s) ret%d err%d\n", iIdx, name, err, errno);
501		−#endif
502		− used_len = SNPRINTF( name, SEM_NAME_MAX, "rc%d%p", iIdx, (void )(ppCtx) );
503		− name[used_len] = '\0';
504		− err = WelsEventOpen( &pSmt->pReadySliceCodingEvent[iIdx], name );
505		−#if defined(ENABLE_TRACE_MT)
506		− WelsLog((ppCtx), WELS_LOG_INFO, "[MT] Open pReadySliceCodingEvent%d = 0x%p named(%s) ret%d err%d\n", iIdx, (void )pSmt->pReadySliceCodingEvent[iIdx]), (void )(ppCtx), err, errno);
507		−#endif
508		−#endif//WIN32
509		−
510		− ++ iIdx;
511		− }
512		−
513		−#ifdef PACKING_ONE_SLICE_PER_LAYER
514		− pSmt->pCountBsSizeInPartition = (uint32_t )pMa->WelsMalloc( sizeof(uint32_t) iThreadNum, "pCountBsSizeInPartition" );
515		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSmt->pCountBsSizeInPartition), FreeMemorySvc(ppCtx) )
516		−#endif//PACKING_ONE_SLICE_PER_LAYER
517		−
518		− WelsMutexInit( &pSmt->mutexSliceNumUpdate );
519		−
520		− (ppCtx)->pSliceBs = (SWelsSliceBs )pMa->WelsMalloc( sizeof(SWelsSliceBs) * iMaxSliceNum, "pSliceBs" );
521		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == (*ppCtx)->pSliceBs), FreeMemorySvc(ppCtx) )
522		−
523		− pBsBase = (*ppCtx)->pFrameBs + iCountBsLen;
524		− pSliceB = (*ppCtx)->pSliceBs;
525		− iSliceBsBufferSize = iTargetSpatialBsSize;
526		− iIdx = 0;
527		− while ( iIdx < iMaxSliceNum )
528		− {
529		− pSliceB->pBsBuffer = (uint8_t *)pMa->WelsMalloc( iSliceBsBufferSize, "pSliceB->pBsBuffer" );
530		−
531		− WELS_VERIFY_RETURN_PROC_IF( 1, (NULL == pSliceB->pBsBuffer), FreeMemorySvc(ppCtx) )
532		− pSliceB->uiSize = iSliceBsBufferSize;
533		−
534		− if ( iIdx > 0 )
535		− {
536		− pSliceB->pBs = pBsBase;
537		− pSliceB->uiBsPos = 0;
538		− pBsBase += iSliceBsBufferSize;
539		− }
540		− else
541		− {
542		− pSliceB->pBs = NULL;
543		− pSliceB->uiBsPos = 0;
544		− }
545		− ++ pSliceB;
546		− ++ iIdx;
547		− }
548		−
549		−#if defined(ENABLE_TRACE_MT)
550		− WelsLog((*ppCtx), WELS_LOG_INFO, "RequestMtResource(), iThreadNum=%d, iCountSliceNum= %d\n", pPara->iCountThreadsNum, iMaxSliceNum);
551		−#endif
552		−
553		− return 0;
554		−}
555		−
556		−void ReleaseMtResource( sWelsEncCtx **ppCtx )
557		−{
558		− SWelsSliceBs *pSliceB = NULL;
559		− SWelsSvcCodingParam *pCodingParam = NULL;
560		− SSliceThreading *pSmt = NULL;
561		− CMemoryAlign *pMa = NULL;
562		− int32_t iIdx = 0;
563		− int32_t iThreadNum = 0;
564		− int16_t uiSliceNum = 0;
565		−
566		− if ( NULL == ppCtx \|\| NULL == *ppCtx )
567		− return;
568		−
569		− pMa = (*ppCtx)->pMemAlign;
570		− pCodingParam = (*ppCtx)->pSvcParam;
571		− uiSliceNum = (*ppCtx)->iMaxSliceCount;
572		− iThreadNum = (*ppCtx)->pSvcParam->iCountThreadsNum;
573		− pSmt = (*ppCtx)->pSliceThreading;
574		−
575		− if ( NULL == pSmt )
576		− return;
577		−
578		− while ( iIdx < iThreadNum) {
579		−#ifdef WIN32
580		− if ( pSmt->pThreadHandles != NULL && pSmt->pThreadHandles[iIdx] != NULL )
581		− WelsThreadDestroy( &pSmt->pThreadHandles[iIdx] );
582		−
583		− if ( pSmt->pSliceCodedEvent != NULL )
584		− WelsEventDestroy( &pSmt->pSliceCodedEvent[iIdx] );
585		− if ( pSmt->pReadySliceCodingEvent != NULL )
586		− WelsEventDestroy( &pSmt->pReadySliceCodingEvent[iIdx] );
587		− if ( pSmt->pFinSliceCodingEvent != NULL )
588		− WelsEventDestroy( &pSmt->pFinSliceCodingEvent[iIdx] );
589		− if ( pSmt->pExitEncodeEvent != NULL )
590		− WelsEventDestroy( &pSmt->pExitEncodeEvent[iIdx] );
591		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
592		− if ( pSmt->pUpdateMbListEvent != NULL )
593		− WelsEventDestroy( &pSmt->pUpdateMbListEvent[iIdx] );
594		− if ( pSmt->pFinUpdateMbListEvent != NULL )
595		− WelsEventDestroy( &pSmt->pFinUpdateMbListEvent[iIdx] );
596		−#endif//DYNAMIC_SLICE_ASSIGN && TRY_SLICING_BALANCE
597		−#else
598		− str_t ename[SEM_NAME_MAX] = {0};
599		− int32_t used_len = 0;
600		− // length of semaphore name should be system constrained at least on mac 10.7
601		− SNPRINTF( ename, SEM_NAME_MAX, "sc%d%p", iIdx, (void )(ppCtx) );
602		− WelsEventClose( pSmt->pSliceCodedEvent[iIdx], ename );
603		− used_len = SNPRINTF( ename, SEM_NAME_MAX, "rc%d%p", iIdx, (void )(ppCtx) );
604		− ename[used_len] = '\0';
605		− WelsEventClose( pSmt->pReadySliceCodingEvent[iIdx], ename );
606		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
607		− used_len = SNPRINTF( ename, SEM_NAME_MAX, "ud%d%p", iIdx, (void )(ppCtx) );
608		− ename[used_len] = '\0';
609		− WelsEventClose( pSmt->pUpdateMbListEvent[iIdx], ename );
610		− used_len = SNPRINTF( ename, SEM_NAME_MAX, "fu%d%p", iIdx, (void )(ppCtx) );
611		− ename[used_len] = '\0';
612		− WelsEventClose( pSmt->pFinUpdateMbListEvent[iIdx], ename );
613		−#endif//DYNAMIC_SLICE_ASSIGN && TRY_SLICING_BALANCE
614		−#endif//WIN32
615		−
616		− ++ iIdx;
617		− }
618		−
619		−#ifdef WIN32
620		− if ( pSmt->pExitEncodeEvent != NULL )
621		− {
622		− pMa->WelsFree( pSmt->pExitEncodeEvent, "pExitEncodeEvent" );
623		− pSmt->pExitEncodeEvent = NULL;
624		− }
625		− if ( pSmt->pSliceCodedEvent != NULL )
626		− {
627		− pMa->WelsFree( pSmt->pSliceCodedEvent, "pSliceCodedEvent" );
628		− pSmt->pSliceCodedEvent = NULL;
629		− }
630		− if ( pSmt->pReadySliceCodingEvent != NULL )
631		− {
632		− pMa->WelsFree( pSmt->pReadySliceCodingEvent, "pReadySliceCodingEvent" );
633		− pSmt->pReadySliceCodingEvent = NULL;
634		− }
635		− if ( pSmt->pFinSliceCodingEvent != NULL )
636		− {
637		− pMa->WelsFree( pSmt->pFinSliceCodingEvent, "pFinSliceCodingEvent" );
638		− pSmt->pFinSliceCodingEvent = NULL;
639		− }
640		−#endif//WIN32
641		−
642		−#ifdef PACKING_ONE_SLICE_PER_LAYER
643		− if ( NULL != pSmt->pCountBsSizeInPartition )
644		− {
645		− pMa->WelsFree( pSmt->pCountBsSizeInPartition, "pCountBsSizeInPartition" );
646		− pSmt->pCountBsSizeInPartition = NULL;
647		− }
648		−#endif//PACKING_ONE_SLICE_PER_LAYER
649		− WelsMutexDestroy( &pSmt->mutexSliceNumUpdate );
650		−
651		− if ( pSmt->pThreadPEncCtx != NULL )
652		− {
653		− pMa->WelsFree( pSmt->pThreadPEncCtx, "pThreadPEncCtx" );
654		− pSmt->pThreadPEncCtx = NULL;
655		− }
656		− if ( pSmt->pThreadHandles != NULL )
657		− {
658		− pMa->WelsFree( pSmt->pThreadHandles, "pThreadHandles" );
659		− pSmt->pThreadHandles = NULL;
660		− }
661		−
662		− pSliceB = (*ppCtx)->pSliceBs;
663		− iIdx = 0;
664		− while ( pSliceB != NULL && iIdx < uiSliceNum )
665		− {
666		− if ( pSliceB->pBsBuffer )
667		− {
668		− pMa->WelsFree( pSliceB->pBsBuffer, "pSliceB->pBsBuffer" );
669		− pSliceB->pBsBuffer = NULL;
670		− pSliceB->uiSize = 0;
671		− }
672		− ++ iIdx;
673		− ++ pSliceB;
674		− }
675		− if ( (*ppCtx)->pSliceBs != NULL )
676		− {
677		− pMa->WelsFree( (*ppCtx)->pSliceBs, "pSliceBs" );
678		− (*ppCtx)->pSliceBs = NULL;
679		− }
680		−#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
681		− if ( pSmt->pSliceConsumeTime != NULL )
682		− {
683		− iIdx = 0;
684		− while (iIdx < pCodingParam->iNumDependencyLayer)
685		− {
686		− if ( pSmt->pSliceConsumeTime[iIdx] )
687		− {
688		− pMa->WelsFree( pSmt->pSliceConsumeTime[iIdx], "pSliceConsumeTime[]" );
689		− pSmt->pSliceConsumeTime[iIdx] = NULL;
690		− }
691		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
692		− if ( pSmt->pSliceComplexRatio[iIdx] != NULL )
693		− {
694		− pMa->WelsFree( pSmt->pSliceComplexRatio[iIdx], "pSliceComplexRatio[]" );
695		− pSmt->pSliceComplexRatio[iIdx] = NULL;
696		− }
697		−#endif//TRY_SLICING_BALANCE
698		− ++ iIdx;
699		− }
700		− }
701		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
702		−
703		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
704		−
705		−#ifdef WIN32
706		− if ( pSmt->pUpdateMbListEvent != NULL )
707		− {
708		− pMa->WelsFree( pSmt->pUpdateMbListEvent, "pUpdateMbListEvent" );
709		− pSmt->pUpdateMbListEvent = NULL;
710		− }
711		− if ( pSmt->pFinUpdateMbListEvent != NULL )
712		− {
713		− pMa->WelsFree( pSmt->pFinUpdateMbListEvent, "pFinUpdateMbListEvent" );
714		− pSmt->pFinUpdateMbListEvent = NULL;
715		− }
716		−#else
717		− if ( pSmt->pUpdateMbListThrdHandles )
718		− {
719		− pMa->WelsFree( pSmt->pUpdateMbListThrdHandles, "pUpdateMbListThrdHandles" );
720		− pSmt->pUpdateMbListThrdHandles = NULL;
721		− }
722		−#endif//WIN32
723		−
724		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
725		−
726		−#ifdef MT_DEBUG
727		− // file handle for debug
728		− if ( pSmt->pFSliceDiff )
729		− {
730		− fclose( pSmt->pFSliceDiff );
731		− pSmt->pFSliceDiff = NULL;
732		− }
733		−#endif//MT_DEBUG
734		− pMa->WelsFree((*ppCtx)->pSliceThreading, "SSliceThreading");
735		− (*ppCtx)->pSliceThreading = NULL;
736		−}
737		−
738		−int32_t AppendSliceToFrameBs( sWelsEncCtx pCtx, SLayerBSInfo pLbi, const int32_t iSliceCount )
739		−{
740		− SWelsSvcCodingParam *pCodingParam = pCtx->pSvcParam;
741		− SDLayerParam *pDlp = &pCodingParam->sDependencyLayers[pCtx->uiDependencyId];
742		− SWelsSliceBs *pSliceBs = NULL;
743		− const BOOL_T kbIsDynamicSlicingMode = (pDlp->sMso.uiSliceMode == SM_DYN_SLICE);
744		− int32_t iLayerSize = 0;
745		− int32_t iNalIdxBase = pLbi->iNalCount;
746		− int32_t iSliceIdx = 0;
747		−
748		− if ( !kbIsDynamicSlicingMode )
749		− {
750		− pSliceBs = &pCtx->pSliceBs[0];
751		− iLayerSize = pSliceBs->uiBsPos; // assign with base pSlice first
752		− iSliceIdx = 1; // pSlice 0 bs has been written to pFrameBs yet by now, so uiSliceIdx base should be 1
753		− while (iSliceIdx < iSliceCount)
754		− {
755		− ++ pSliceBs;
756		− if ( pSliceBs != NULL && pSliceBs->uiBsPos > 0 )
757		− {
758		− int32_t iNalIdx = 0;
759		− const int32_t iCountNal = pSliceBs->iNalIndex;
760		−
761		−#if MT_DEBUG_BS_WR
762		− assert(pSliceBs->bSliceCodedFlag);
763		−#endif//MT_DEBUG_BS_WR
764		−
765		− memmove(pCtx->pFrameBs + pCtx->iPosBsBuffer, pSliceBs->pBs, pSliceBs->uiBsPos); // confirmed_safe_unsafe_usage
766		− pCtx->iPosBsBuffer += pSliceBs->uiBsPos;
767		−
768		− iLayerSize += pSliceBs->uiBsPos;
769		−
770		− while (iNalIdx < iCountNal)
771		− {
772		− pLbi->iNalLengthInByte[iNalIdxBase+iNalIdx] = pSliceBs->iNalLen[iNalIdx];
773		− ++ iNalIdx;
774		− }
775		− pLbi->iNalCount += iCountNal;
776		− iNalIdxBase += iCountNal;
777		− }
778		− ++ iSliceIdx;
779		− }
780		− }
781		− else // for SM_DYN_SLICE
782		− {
783		− const int32_t kiPartitionCnt = iSliceCount;
784		− int32_t iPartitionIdx = 0;
785		−
786		− // due partition_0 has been written to pFrameBsBuffer
787		− // so iLayerSize need add it
788		− while ( iPartitionIdx < kiPartitionCnt )
789		− {
790		− const int32_t kiCountSlicesCoded = pCtx->pCurDqLayer->pNumSliceCodedOfPartition[iPartitionIdx];
791		− int32_t iIdx = 0;
792		−
793		− iSliceIdx = iPartitionIdx;
794		− while(iIdx < kiCountSlicesCoded)
795		− {
796		− pSliceBs = &pCtx->pSliceBs[iSliceIdx];
797		− if ( pSliceBs != NULL && pSliceBs->uiBsPos > 0 )
798		− {
799		− if ( iPartitionIdx > 0 )
800		− {
801		− int32_t iNalIdx = 0;
802		− const int32_t iCountNal = pSliceBs->iNalIndex;
803		−
804		− memmove(pCtx->pFrameBs + pCtx->iPosBsBuffer, pSliceBs->pBs, pSliceBs->uiBsPos); // confirmed_safe_unsafe_usage
805		− pCtx->iPosBsBuffer += pSliceBs->uiBsPos;
806		−
807		− iLayerSize += pSliceBs->uiBsPos;
808		−
809		− while (iNalIdx < iCountNal)
810		− {
811		− pLbi->iNalLengthInByte[iNalIdxBase+iNalIdx] = pSliceBs->iNalLen[iNalIdx];
812		− ++ iNalIdx;
813		− }
814		− pLbi->iNalCount += iCountNal;
815		− iNalIdxBase += iCountNal;
816		− }
817		− else
818		− {
819		− iLayerSize += pSliceBs->uiBsPos;
820		− }
821		− }
822		−
823		− iSliceIdx += kiPartitionCnt;
824		− ++ iIdx;
825		− }
826		− ++ iPartitionIdx;
827		− }
828		− }
829		−
830		− return iLayerSize;
831		−}
832		−
833		−int32_t WriteSliceToFrameBs( sWelsEncCtx pCtx, SLayerBSInfo pLbi, uint8_t *pFrameBsBuffer, const int32_t iSliceIdx )
834		−{
835		− SWelsSliceBs *pSliceBs = &pCtx->pSliceBs[iSliceIdx];
836		− SNalUnitHeaderExt *pNalHdrExt= &pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt;
837		− uint8_t *pDst = pFrameBsBuffer;
838		− int32_t pNalLen[2];
839		− int32_t iSliceSize = 0;
840		− const int32_t kiNalCnt = pSliceBs->iNalIndex;
841		− int32_t iNalIdx = 0;
842		−#if !defined(PACKING_ONE_SLICE_PER_LAYER)
843		− const int32_t iFirstSlice = (iSliceIdx == 0);
844		− int32_t iNalBase = iFirstSlice ? 0 : pLbi->iNalCount;
845		−#else
846		− int32_t iNalBase = 0;
847		−#endif//!PACKING_ONE_SLICE_PER_LAYER
848		−
849		− while ( iNalIdx < kiNalCnt ) {
850		− iSliceSize += WelsEncodeNalExt( &pSliceBs->sNalList[iNalIdx], pNalHdrExt, pDst, &pNalLen[iNalIdx] );
851		− pDst += pNalLen[iNalIdx];
852		− pLbi->iNalLengthInByte[iNalBase+iNalIdx] = pNalLen[iNalIdx];
853		−
854		− ++ iNalIdx;
855		− }
856		−
857		−#if !defined(PACKING_ONE_SLICE_PER_LAYER)
858		− pSliceBs->uiBsPos = iSliceSize;
859		− if ( iFirstSlice )
860		− {
861		− // pBsBuffer has been updated at coding_slice_0_in_encoder_mother_thread()
862		− pLbi->uiLayerType = VIDEO_CODING_LAYER;
863		− pLbi->uiSpatialId = pNalHdrExt->uiDependencyId;
864		− pLbi->uiTemporalId = pNalHdrExt->uiTemporalId;
865		− pLbi->uiQualityId = 0;
866		− pLbi->uiPriorityId = 0;
867		− pLbi->iNalCount = kiNalCnt;
868		− }
869		− else
870		− {
871		− pLbi->iNalCount += kiNalCnt;
872		− }
873		−#else
874		− pLbi->uiLayerType = VIDEO_CODING_LAYER;
875		− pLbi->uiSpatialId = pNalHdrExt->uiDependencyId;
876		− pLbi->uiTemporalId = pNalHdrExt->uiTemporalId;
877		− pLbi->uiQualityId = 0;
878		− pLbi->uiPriorityId = 0;
879		− pLbi->iNalCount = kiNalCnt;
880		−#endif//PACKING_ONE_SLICE_PER_LAYER
881		−
882		− return iSliceSize;
883		−}
884		−
885		−int32_t WriteSliceBs( sWelsEncCtx pCtx, uint8_t pSliceBsBuf, const int32_t iSliceIdx )
886		−{
887		− SWelsSliceBs *pSliceBs = &pCtx->pSliceBs[iSliceIdx];
888		− SNalUnitHeaderExt *pNalHdrExt= &pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt;
889		− uint8_t *pDst = pSliceBsBuf;
890		− int32_t *pNalLen = &pSliceBs->iNalLen[0];
891		− int32_t iSliceSize = 0;
892		− const int32_t kiNalCnt = pSliceBs->iNalIndex;
893		− int32_t iNalIdx = 0;
894		−
895		− assert( kiNalCnt <= 2 );
896		− if ( kiNalCnt > 2 )
897		− return 0;
898		−
899		− while ( iNalIdx < kiNalCnt ) {
900		− iSliceSize += WelsEncodeNalExt( &pSliceBs->sNalList[iNalIdx], pNalHdrExt, pDst, &pNalLen[iNalIdx] );
901		− pDst += pNalLen[iNalIdx];
902		−
903		− ++ iNalIdx;
904		− }
905		− pSliceBs->uiBsPos = iSliceSize;
906		−
907		− return iSliceSize;
908		−}
909		−
910		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
911		−#if defined(__GNUC__)
912		−WELS_THREAD_ROUTINE_TYPE UpdateMbListThreadProc( void *arg )
913		−{
914		− SSliceThreadPrivateData pPrivateData = (SSliceThreadPrivateData )arg;
915		− sWelsEncCtx *pEncPEncCtx = NULL;
916		− SDqLayer *pCurDq = NULL;
917		− int32_t iSliceIdx = -1;
918		− int32_t iEventIdx = -1;
919		− WELS_THREAD_ERROR_CODE iWaitRet = WELS_THREAD_ERROR_GENERIAL;
920		− uint32_t uiThrdRet = 0;
921		−
922		− if ( NULL == pPrivateData )
923		− WELS_THREAD_ROUTINE_RETURN(1);
924		−
925		− pEncPEncCtx = (sWelsEncCtx *)pPrivateData->pWelsPEncCtx;
926		− iSliceIdx = pPrivateData->iSliceIndex;
927		− iEventIdx = pPrivateData->iThreadIndex;
928		−
929		− do {
930		−#if defined(ENABLE_TRACE_MT)
931		− WelsLog(pEncPEncCtx, WELS_LOG_INFO, "[MT] UpdateMbListThreadProc(), try to wait (pUpdateMbListEvent[%d])!\n", iEventIdx);
932		−#endif
933		− iWaitRet = WelsEventWait( pEncPEncCtx->pSliceThreading->pUpdateMbListEvent[iEventIdx] );
934		− if ( WELS_THREAD_ERROR_WAIT_OBJECT_0 == iWaitRet )
935		− {
936		− pCurDq = pEncPEncCtx->pCurDqLayer;
937		− UpdateMbListNeighborParallel( pCurDq->pSliceEncCtx, pCurDq->sMbDataP, iSliceIdx );
938		− WelsEventSignal( pEncPEncCtx->pSliceThreading->pFinUpdateMbListEvent[iEventIdx] ); // mean finished update pMb list for this pSlice
939		− }
940		− else
941		− {
942		− WelsLog(pEncPEncCtx, WELS_LOG_WARNING, "[MT] UpdateMbListThreadProc(), waiting pUpdateMbListEvent[%d] failed(%d) and thread%d terminated!\n", iEventIdx, iWaitRet, iEventIdx);
943		− uiThrdRet = 1;
944		− break;
945		− }
946		− } while(1);
947		−
948		− WELS_THREAD_ROUTINE_RETURN(uiThrdRet);
949		−}
950		−#endif//__GNUC__
951		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
952		−
953		−// thread process for coding one pSlice
954		−WELS_THREAD_ROUTINE_TYPE CodingSliceThreadProc( void *arg )
955		−{
956		− SSliceThreadPrivateData pPrivateData = (SSliceThreadPrivateData )arg;
957		− sWelsEncCtx *pEncPEncCtx = NULL;
958		− SDqLayer *pCurDq = NULL;
959		− SSlice *pSlice = NULL;
960		− SWelsSliceBs *pSliceBs = NULL;
961		−#ifdef WIN32
962		− WELS_EVENT pEventsList[3];
963		− int32_t iEventCount = 0;
964		−#endif
965		− WELS_THREAD_ERROR_CODE iWaitRet = WELS_THREAD_ERROR_GENERIAL;
966		− uint32_t uiThrdRet = 0;
967		− int32_t iSliceSize = 0;
968		− int32_t iSliceIdx = -1;
969		− int32_t iThreadIdx = -1;
970		− int32_t iEventIdx = -1;
971		− bool_t bNeedPrefix = false;
972		− EWelsNalUnitType eNalType = NAL_UNIT_UNSPEC_0;
973		− EWelsNalRefIdc eNalRefIdc = NRI_PRI_LOWEST;
974		−
975		− if ( NULL == pPrivateData )
976		− WELS_THREAD_ROUTINE_RETURN(1);
977		−
978		− WelsSetThreadCancelable();
979		−
980		− pEncPEncCtx = (sWelsEncCtx *)pPrivateData->pWelsPEncCtx;
981		−
982		− iThreadIdx = pPrivateData->iThreadIndex;
983		− iEventIdx = iThreadIdx;
984		−
985		−#ifdef WIN32
986		− pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pReadySliceCodingEvent[iEventIdx];
987		− pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pExitEncodeEvent[iEventIdx];
988		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
989		− pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pUpdateMbListEvent[iEventIdx];
990		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
991		−#endif//WIN32
992		−
993		− do {
994		−#ifdef WIN32
995		− iWaitRet = WelsMultipleEventsWaitSingleBlocking( iEventCount,
996		− &pEventsList[0],
997		− (uint32_t)-1 ); // blocking until at least one event is
998		−#else
999		−#if defined(ENABLE_TRACE_MT)
1000		− WelsLog(pEncPEncCtx, WELS_LOG_INFO, "[MT] CodingSliceThreadProc(), try to call WelsEventWait(pReadySliceCodingEvent[%d]= 0x%p), pEncPEncCtx= 0x%p!\n", iEventIdx, (void )(pEncPEncCtx->pReadySliceCodingEvent[iEventIdx]), (void )pEncPEncCtx );
1001		−#endif
1002		− iWaitRet = WelsEventWait( pEncPEncCtx->pSliceThreading->pReadySliceCodingEvent[iEventIdx] );
1003		−#endif//WIN32
1004		− if ( WELS_THREAD_ERROR_WAIT_OBJECT_0 == iWaitRet ) // start pSlice coding signal waited
1005		− {
1006		− SLayerBSInfo *pLbi = pPrivateData->pLayerBs;
1007		− const int32_t kiCurDid = pEncPEncCtx->uiDependencyId;
1008		− const int32_t kiCurTid = pEncPEncCtx->uiTemporalId;
1009		− SWelsSvcCodingParam *pCodingParam = pEncPEncCtx->pSvcParam;
1010		− SDLayerParam *pParamD = &pCodingParam->sDependencyLayers[kiCurDid];
1011		−
1012		− pCurDq = pEncPEncCtx->pCurDqLayer;
1013		− eNalType = pEncPEncCtx->eNalType;
1014		− eNalRefIdc = pEncPEncCtx->eNalPriority;
1015		− bNeedPrefix = pEncPEncCtx->bNeedPrefixNalFlag;
1016		−
1017		− if ( pParamD->sMso.uiSliceMode != SM_DYN_SLICE )
1018		− {
1019		− int64_t iSliceStart = 0;
1020		− bool_t bDsaFlag = false;
1021		− iSliceIdx = pPrivateData->iSliceIndex;
1022		− pSlice = &pCurDq->sLayerInfo.pSliceInLayer[iSliceIdx];
1023		− pSliceBs = &pEncPEncCtx->pSliceBs[iSliceIdx];
1024		−
1025		−#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
1026		− bDsaFlag = (pParamD->sMso.uiSliceMode == SM_FIXEDSLCNUM_SLICE &&
1027		− pCodingParam->iMultipleThreadIdc > 1 &&
1028		− pCodingParam->iMultipleThreadIdc >= pParamD->sMso.sSliceArgument.iSliceNum);
1029		− if ( bDsaFlag )
1030		− iSliceStart = WelsTime();
1031		−#endif//DYNAMIC_SLICE_ASSIGN \|\| MT_DEBUG
1032		−
1033		−#if !defined(PACKING_ONE_SLICE_PER_LAYER)
1034		− pSliceBs->uiBsPos = 0;
1035		−#endif//!PACKING_ONE_SLICE_PER_LAYER
1036		− pSliceBs->iNalIndex = 0;
1037		− assert( (void)(&pSliceBs->sBsWrite) == (void)pSlice->pSliceBsa );
1038		− InitBits( &pSliceBs->sBsWrite, pSliceBs->pBsBuffer, pSliceBs->uiSize );
1039		−
1040		−#if MT_DEBUG_BS_WR
1041		− pSliceBs->bSliceCodedFlag = FALSE;
1042		−#endif//MT_DEBUG_BS_WR
1043		−
1044		− if ( bNeedPrefix )
1045		− {
1046		− if ( eNalRefIdc != NRI_PRI_LOWEST )
1047		− {
1048		− WelsLoadNalForSlice( pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc );
1049		− WelsWriteSVCPrefixNal( &pSliceBs->sBsWrite, eNalRefIdc, (NAL_UNIT_CODED_SLICE_IDR == eNalType) );
1050		− WelsUnloadNalForSlice( pSliceBs );
1051		− }
1052		− else // No Prefix NAL Unit RBSP syntax here, but need add NAL Unit Header extension
1053		− {
1054		− WelsLoadNalForSlice( pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc );
1055		− // No need write any syntax of prefix NAL Unit RBSP here
1056		− WelsUnloadNalForSlice( pSliceBs );
1057		− }
1058		− }
1059		−
1060		− WelsLoadNalForSlice( pSliceBs, eNalType, eNalRefIdc );
1061		−
1062		− WelsCodeOneSlice( pEncPEncCtx, iSliceIdx, eNalType );
1063		−
1064		− WelsUnloadNalForSlice( pSliceBs );
1065		−
1066		−#if !defined(PACKING_ONE_SLICE_PER_LAYER)
1067		− if ( 0 == iSliceIdx )
1068		− {
1069		− pLbi->pBsBuf = pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer;
1070		− iSliceSize = WriteSliceToFrameBs( pEncPEncCtx, pLbi, pLbi->pBsBuf, iSliceIdx );
1071		− pEncPEncCtx->iPosBsBuffer += iSliceSize;
1072		− }
1073		− else
1074		− iSliceSize = WriteSliceBs( pEncPEncCtx, pSliceBs->pBs, iSliceIdx );
1075		−#else// PACKING_ONE_SLICE_PER_LAYER
1076		− if ( 0 == iSliceIdx )
1077		− {
1078		− pLbi->pBsBuf = pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer;
1079		− iSliceSize = WriteSliceToFrameBs( pEncPEncCtx, pLbi, pLbi->pBsBuf, iSliceIdx );
1080		− pEncPEncCtx->iPosBsBuffer += iSliceSize;
1081		− }
1082		− else
1083		− {
1084		− pLbi->pBsBuf = pSliceBs->bs + pSliceBs->uiBsPos;
1085		− iSliceSize = WriteSliceToFrameBs( pEncPEncCtx, pLbi, pLbi->pBsBuf, iSliceIdx );
1086		− pSliceBs->uiBsPos += iSliceSize;
1087		− }
1088		−#endif//!PACKING_ONE_SLICE_PER_LAYER
1089		−
1090		− if ( pCurDq->bDeblockingParallelFlag && pSlice->sSliceHeaderExt.sSliceHeader.uiDisableDeblockingFilterIdc != 1
1091		−#if !defined(ENABLE_FRAME_DUMP)
1092		− && ( eNalRefIdc != NRI_PRI_LOWEST ) &&
1093		− ( pParamD->iHighestTemporalId == 0 \|\| kiCurTid < pParamD->iHighestTemporalId )
1094		−#endif// !ENABLE_FRAME_DUMP
1095		− )
1096		− {
1097		− DeblockingFilterSliceAvcbase( pCurDq, pEncPEncCtx->pFuncList, iSliceIdx );
1098		− }
1099		−
1100		−#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
1101		− if ( bDsaFlag )
1102		− {
1103		− pEncPEncCtx->pSliceThreading->pSliceConsumeTime[pEncPEncCtx->uiDependencyId][iSliceIdx] = (uint32_t)(WelsTime() - iSliceStart);
1104		−#if defined(ENABLE_TRACE_MT)
1105		− WelsLog(pEncPEncCtx, WELS_LOG_INFO, "[MT] CodingSliceThreadProc(), coding_idx %d, uiSliceIdx %d, pSliceConsumeTime %d, iSliceSize %d, pFirstMbInSlice %d, count_num_mb_in_slice %d\n",
1106		− pEncPEncCtx->iCodingIndex, iSliceIdx, pEncPEncCtx->pSliceThreading->pSliceConsumeTime[pEncPEncCtx->uiDependencyId][iSliceIdx], iSliceSize, pCurDq->pSliceEncCtx->pFirstMbInSlice[iSliceIdx], pCurDq->pSliceEncCtx->pCountMbNumInSlice[iSliceIdx]);
1107		−#endif//ENABLE_TRACE_MT
1108		− }
1109		−#endif//DYNAMIC_SLICE_ASSIGN \|\| MT_DEBUG
1110		−
1111		−#if defined(SLICE_INFO_OUTPUT)
1112		− fprintf( stderr,
1113		− "@pSlice=%-6d sliceType:%c idc:%d size:%-6d\n",
1114		− iSliceIdx,
1115		− (pEncPEncCtx->eSliceType == P_SLICE ? 'P' : 'I'),
1116		− eNalRefIdc,
1117		− iSliceSize
1118		− );
1119		−#endif//SLICE_INFO_OUTPUT
1120		−
1121		−#if MT_DEBUG_BS_WR
1122		− pSliceBs->bSliceCodedFlag = TRUE;
1123		−#endif//MT_DEBUG_BS_WR
1124		−
1125		−#ifdef WIN32
1126		− WelsEventSignal( &pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx] ); // mean finished coding current pSlice
1127		−#else
1128		− WelsEventSignal( pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx] ); // mean finished coding current pSlice
1129		−#endif//WIN32
1130		− }
1131		− else // for SM_DYN_SLICE parallelization
1132		− {
1133		−#ifdef PACKING_ONE_SLICE_PER_LAYER
1134		− SLayerBSInfo *pLbiPacking = NULL;
1135		−#endif//PACKING_ONE_SLICE_PER_LAYER
1136		− SSliceCtx *pSliceCtx = pCurDq->pSliceEncCtx;
1137		− const int32_t kiPartitionId = iThreadIdx;
1138		− const int32_t kiSliceIdxStep = pEncPEncCtx->iActiveThreadsNum;
1139		− const int32_t kiFirstMbInPartition = pPrivateData->iStartMbIndex; // inclusive
1140		− const int32_t kiEndMbInPartition = pPrivateData->iEndMbIndex; // exclusive
1141		− int32_t iAnyMbLeftInPartition = kiEndMbInPartition - kiFirstMbInPartition;
1142		−
1143		− iSliceIdx = pPrivateData->iSliceIndex;
1144		−
1145		− pSliceCtx->pFirstMbInSlice[iSliceIdx] = kiFirstMbInPartition;
1146		− pCurDq->pNumSliceCodedOfPartition[kiPartitionId] = 1; // one pSlice per partition intialized, dynamic slicing inside
1147		− pCurDq->pLastMbIdxOfPartition[kiPartitionId] = kiEndMbInPartition-1;
1148		−
1149		− pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId] = 0;
1150		−
1151		− while( iAnyMbLeftInPartition > 0 )
1152		− {
1153		− if ( iSliceIdx >= pSliceCtx->iMaxSliceNumConstraint )
1154		− {
1155		− // TODO: need exception handler for not large enough of MAX_SLICES_NUM related memory usage
1156		− // No idea about its solution due MAX_SLICES_NUM is fixed lenght in relevent pData structure
1157		− uiThrdRet = 1;
1158		− break;
1159		− }
1160		−
1161		− pSlice = &pCurDq->sLayerInfo.pSliceInLayer[iSliceIdx];
1162		− pSliceBs = &pEncPEncCtx->pSliceBs[iSliceIdx];
1163		−
1164		−#if !defined(PACKING_ONE_SLICE_PER_LAYER)
1165		− pSliceBs->uiBsPos = 0;
1166		−#endif//!PACKING_ONE_SLICE_PER_LAYER
1167		− pSliceBs->iNalIndex = 0;
1168		− InitBits( &pSliceBs->sBsWrite, pSliceBs->pBsBuffer, pSliceBs->uiSize );
1169		−
1170		− if ( bNeedPrefix )
1171		− {
1172		− if ( eNalRefIdc != NRI_PRI_LOWEST )
1173		− {
1174		− WelsLoadNalForSlice( pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc );
1175		− WelsWriteSVCPrefixNal( &pSliceBs->sBsWrite, eNalRefIdc, (NAL_UNIT_CODED_SLICE_IDR == eNalType) );
1176		− WelsUnloadNalForSlice( pSliceBs );
1177		− }
1178		− else // No Prefix NAL Unit RBSP syntax here, but need add NAL Unit Header extension
1179		− {
1180		− WelsLoadNalForSlice( pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc );
1181		− // No need write any syntax of prefix NAL Unit RBSP here
1182		− WelsUnloadNalForSlice( pSliceBs );
1183		− }
1184		− }
1185		−
1186		− WelsLoadNalForSlice( pSliceBs, eNalType, eNalRefIdc );
1187		−
1188		− WelsCodeOneSlice( pEncPEncCtx, iSliceIdx, eNalType );
1189		−
1190		− WelsUnloadNalForSlice( pSliceBs );
1191		−
1192		−#if !defined(PACKING_ONE_SLICE_PER_LAYER)
1193		− if ( 0 == kiPartitionId )
1194		− {
1195		− if ( 0 == iSliceIdx )
1196		− pLbi->pBsBuf = pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer;
1197		− iSliceSize = WriteSliceToFrameBs( pEncPEncCtx, pLbi, pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer, iSliceIdx );
1198		− pEncPEncCtx->iPosBsBuffer += iSliceSize;
1199		− }
1200		− else
1201		− iSliceSize = WriteSliceBs( pEncPEncCtx, pSliceBs->pBs, iSliceIdx );
1202		−#else// PACKING_ONE_SLICE_PER_LAYER
1203		− pLbiPacking = pLbi + (iSliceIdx - kiPartitionId);
1204		−
1205		− if ( 0 == kiPartitionId )
1206		− {
1207		− pLbiPacking->pBsBuf = pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer;
1208		− iSliceSize = WriteSliceToFrameBs( pEncPEncCtx, pLbiPacking, pLbiPacking->pBsBuf, iSliceIdx );
1209		− pEncPEncCtx->iPosBsBuffer += iSliceSize;
1210		− }
1211		− else
1212		− {
1213		− pLbiPacking->pBsBuf = pSliceBs->bs + pSliceBs->uiBsPos;
1214		− iSliceSize = WriteSliceToFrameBs( pEncPEncCtx, pLbiPacking, pLbiPacking->pBsBuf, iSliceIdx );
1215		− pSliceBs->uiBsPos += iSliceSize;
1216		− }
1217		− pEncPEncCtx->pSliceThreading->pCountBsSizeInPartition[kiPartitionId] += iSliceSize;
1218		−#endif//!PACKING_ONE_SLICE_PER_LAYER
1219		−
1220		− if ( pCurDq->bDeblockingParallelFlag && pSlice->sSliceHeaderExt.sSliceHeader.uiDisableDeblockingFilterIdc != 1
1221		−#if !defined(ENABLE_FRAME_DUMP)
1222		− && ( eNalRefIdc != NRI_PRI_LOWEST ) &&
1223		− ( pParamD->iHighestTemporalId == 0 \|\| kiCurTid < pParamD->iHighestTemporalId )
1224		−#endif// !ENABLE_FRAME_DUMP
1225		− )
1226		− {
1227		− DeblockingFilterSliceAvcbase( pCurDq, pEncPEncCtx->pFuncList, iSliceIdx );
1228		− }
1229		−
1230		−#if defined(SLICE_INFO_OUTPUT)
1231		− fprintf( stderr,
1232		− "@pSlice=%-6d sliceType:%c idc:%d size:%-6d\n",
1233		− iSliceIdx,
1234		− (pEncPEncCtx->eSliceType == P_SLICE ? 'P' : 'I'),
1235		− eNalRefIdc,
1236		− iSliceSize
1237		− );
1238		−#endif//SLICE_INFO_OUTPUT
1239		−
1240		−#if defined(ENABLE_TRACE_MT)
1241		− WelsLog(pEncPEncCtx, WELS_LOG_INFO, "[MT] CodingSliceThreadProc(), coding_idx %d, iPartitionId %d, uiSliceIdx %d, iSliceSize %d, count_mb_slice %d, iEndMbInPartition %d, pCurDq->pLastCodedMbIdxOfPartition[%d] %d\n",
1242		− pEncPEncCtx->iCodingIndex, kiPartitionId, iSliceIdx, iSliceSize, pCurDq->pSliceEncCtx->pCountMbNumInSlice[iSliceIdx], kiEndMbInPartition, kiPartitionId, pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId]);
1243		−#endif//ENABLE_TRACE_MT
1244		−
1245		− iAnyMbLeftInPartition = kiEndMbInPartition - (1 + pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId]);
1246		− iSliceIdx += kiSliceIdxStep;
1247		− }
1248		−
1249		− if ( uiThrdRet ) // any exception??
1250		− break;
1251		−
1252		−#ifdef WIN32
1253		− WelsEventSignal( &pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx] ); // mean finished coding current pSlice
1254		−#else
1255		− WelsEventSignal( pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx] ); // mean finished coding current pSlice
1256		−#endif//WIN32
1257		− }
1258		− }
1259		−#ifdef WIN32
1260		− else if ( WELS_THREAD_ERROR_WAIT_OBJECT_0+1 == iWaitRet ) // exit thread signal
1261		− {
1262		− uiThrdRet = 0;
1263		− break;
1264		− }
1265		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
1266		− else if ( WELS_THREAD_ERROR_WAIT_OBJECT_0+2 == iWaitRet ) // update pMb list singal
1267		− {
1268		− iSliceIdx = iEventIdx; // pPrivateData->iSliceIndex; old threads can not be terminated, pPrivateData is not correct for applicable
1269		− pCurDq = pEncPEncCtx->pCurDqLayer;
1270		− UpdateMbListNeighborParallel( pCurDq->pSliceEncCtx, pCurDq->sMbDataP, iSliceIdx );
1271		− WelsEventSignal( &pEncPEncCtx->pSliceThreading->pFinUpdateMbListEvent[iEventIdx] ); // mean finished update pMb list for this pSlice
1272		− }
1273		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
1274		−#endif//WIN32
1275		− else // WELS_THREAD_ERROR_WAIT_TIMEOUT, or WELS_THREAD_ERROR_WAIT_FAILED
1276		− {
1277		− WelsLog(pEncPEncCtx, WELS_LOG_WARNING, "[MT] CodingSliceThreadProc(), waiting pReadySliceCodingEvent[%d] failed(%d) and thread%d terminated!\n", iEventIdx, iWaitRet, iThreadIdx);
1278		− uiThrdRet = 1;
1279		− break;
1280		− }
1281		− } while( 1 );
1282		−
1283		−#ifdef WIN32
1284		− WelsEventSignal( &pEncPEncCtx->pSliceThreading->pFinSliceCodingEvent[iEventIdx] ); // notify to mother encoding threading
1285		−#endif//WIN32
1286		−
1287		− WELS_THREAD_ROUTINE_RETURN(uiThrdRet);
1288		−}
1289		−
1290		−int32_t CreateSliceThreads( sWelsEncCtx *pCtx )
1291		−{
1292		− const int32_t kiThreadCount = pCtx->pSvcParam->iCountThreadsNum;
1293		− int32_t iIdx = 0;
1294		−#if defined(WIN32) && defined(BIND_CPU_CORES_TO_THREADS)
1295		− DWORD dwProcessAffinity;
1296		− DWORD dwSystemAffinity;
1297		− GetProcessAffinityMask(GetCurrentProcess(), &dwProcessAffinity, &dwSystemAffinity);
1298		−#endif//WIN32 && BIND_CPU_CORES_TO_THREADS
1299		−
1300		− while ( iIdx < kiThreadCount ) {
1301		− WelsThreadCreate( &pCtx->pSliceThreading->pThreadHandles[iIdx], CodingSliceThreadProc, &pCtx->pSliceThreading->pThreadPEncCtx[iIdx], 0);
1302		−#if defined(WIN32) && defined(BIND_CPU_CORES_TO_THREADS)
1303		− if ( dwProcessAffinity > 1 && pCtx->pSliceThreading->pThreadHandles[iIdx] != NULL ) // multiple cores and thread created successfully
1304		− {
1305		− DWORD dw = 0;
1306		− DWORD dwAffinityMask = 1 << iIdx;
1307		− if (dwAffinityMask & dwProcessAffinity) // check if cpu is available
1308		− {
1309		− dw = SetThreadAffinityMask( pCtx->pSliceThreading->pThreadHandles[iIdx], dwAffinityMask ); //1 << iIdx
1310		− if ( dw == 0)
1311		− {
1312		− str_t str[64] = {0};
1313		− SNPRINTF(str, 64, "SetThreadAffinityMask iIdx:%d", iIdx);
1314		− }
1315		− }
1316		− }
1317		−#endif//WIN32 && BIND_CPU_CORES_TO_THREADS
1318		− // We need extra threads for update_mb_list_proc on __GNUC__ like OS (mac/linux)
1319		− // due to WelsMultipleEventsWaitSingleBlocking implememtation can not work well
1320		− // in case waiting pUpdateMbListEvent and pReadySliceCodingEvent events at the same time
1321		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
1322		−#if defined(__GNUC__)
1323		− WelsThreadCreate( &pCtx->pSliceThreading->pUpdateMbListThrdHandles[iIdx], UpdateMbListThreadProc, &pCtx->pSliceThreading->pThreadPEncCtx[iIdx], 0);
1324		−#endif//__GNUC__
1325		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
1326		− ++ iIdx;
1327		− }
1328		−#if defined(ENABLE_TRACE_MT)
1329		− WelsLog(pCtx, WELS_LOG_INFO, "CreateSliceThreads() exit..\n");
1330		−#endif
1331		− return 0;
1332		−}
1333		−
1334		−#ifdef PACKING_ONE_SLICE_PER_LAYER
1335		−void ResetCountBsSizeInPartitions( uint32_t *pCountBsSizeList, const int32_t iPartitionCnt )
1336		−{
1337		− if ( pCountBsSizeList != NULL && iPartitionCnt > 0 )
1338		− {
1339		− memset(pCountBsSizeList, 0, sizeof(pCountBsSizeList[0]) * iPartitionCnt );
1340		− }
1341		−}
1342		−#endif//PACKING_ONE_SLICE_PER_LAYER
1343		−
1344		−#ifdef WIN32
1345		−int32_t FiredSliceThreads( SSliceThreadPrivateData pPriData, WELS_EVENT pEventsList, SLayerBSInfo pLbi, const uint32_t uiNumThreads, SSliceCtx pSliceCtx, const BOOL_T bIsDynamicSlicingMode )
1346		−#else
1347		−int32_t FiredSliceThreads( SSliceThreadPrivateData pPriData, WELS_EVENT pEventsList, SLayerBSInfo pLbi, const uint32_t uiNumThreads, SSliceCtx *pSliceCtx, const BOOL_T bIsDynamicSlicingMode )
1348		−#endif//WIN32
1349		−{
1350		− int32_t iEndMbIdx = 0;
1351		− int32_t iIdx = 0;
1352		− const int32_t kiEventCnt = uiNumThreads;
1353		−
1354		− if ( pPriData == NULL \|\| pLbi == NULL \|\| kiEventCnt <= 0 \|\| pEventsList == NULL )
1355		− {
1356		− WelsLog( NULL, WELS_LOG_ERROR, "FiredSliceThreads(), fail due pPriData == %p \|\| pLbi == %p \|\| iEventCnt(%d) <= 0 \|\| pEventsList == %p!!\n", (void )pPriData, (void )pLbi, uiNumThreads, (void *)pEventsList);
1357		− return 1;
1358		− }
1359		−
1360		−#if defined(PACKING_ONE_SLICE_PER_LAYER)
1361		− ////////////////////////////////////////
1362		− if ( bIsDynamicSlicingMode )
1363		− {
1364		− iEndMbIdx = pSliceCtx->iMbNumInFrame;
1365		− for (iIdx = kiEventCnt-1; iIdx >= 0; --iIdx)
1366		− {
1367		− const int32_t kiFirstMbIdx = pSliceCtx->pFirstMbInSlice[iIdx];
1368		− pPriData[iIdx].iStartMbIndex = kiFirstMbIdx;
1369		− pPriData[iIdx].iEndMbIndex = iEndMbIdx;
1370		− iEndMbIdx = kiFirstMbIdx;
1371		− }
1372		− }
1373		−
1374		− iIdx = 0;
1375		− while (iIdx < kiEventCnt) {
1376		− pPriData[iIdx].pLayerBs = pLbi;
1377		− pPriData[iIdx].iSliceIndex = iIdx;
1378		−#ifdef WIN32
1379		− if ( pEventsList[iIdx] )
1380		− WelsEventSignal( &pEventsList[iIdx] );
1381		−#else
1382		− WelsEventSignal( pEventsList[iIdx] );
1383		−#endif//WIN32
1384		− ++ pLbi;
1385		− ++ iIdx;
1386		− }
1387		− ////////////////////////////////////////
1388		−#else
1389		− ////////////////////////////////////////
1390		− if ( bIsDynamicSlicingMode )
1391		− {
1392		− iEndMbIdx = pSliceCtx->iMbNumInFrame;
1393		− for (iIdx = kiEventCnt-1; iIdx >= 0; --iIdx)
1394		− {
1395		− const int32_t iFirstMbIdx = pSliceCtx->pFirstMbInSlice[iIdx];
1396		− pPriData[iIdx].iStartMbIndex = iFirstMbIdx;
1397		− pPriData[iIdx].iEndMbIndex = iEndMbIdx;
1398		− iEndMbIdx = iFirstMbIdx;
1399		− }
1400		− }
1401		−
1402		− iIdx = 0;
1403		− while (iIdx < kiEventCnt) {
1404		− pPriData[iIdx].pLayerBs = pLbi;
1405		− pPriData[iIdx].iSliceIndex = iIdx;
1406		−#ifdef WIN32
1407		− if ( pEventsList[iIdx] )
1408		− WelsEventSignal( &pEventsList[iIdx] );
1409		−#else
1410		− WelsEventSignal( pEventsList[iIdx] );
1411		−#endif//WIN32
1412		− ++ iIdx;
1413		− }
1414		− ////////////////////////////////////////
1415		−#endif//PACKING_ONE_SLICE_PER_LAYER
1416		−
1417		− return 0;
1418		−}
1419		−
1420		−int32_t DynamicDetectCpuCores()
1421		−{
1422		− WelsLogicalProcessInfo info;
1423		− WelsQueryLogicalProcessInfo(&info);
1424		− return info.ProcessorCount;
1425		−}
1426		−
1427		−#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN)
1428		−
1429		−int32_t AdjustBaseLayer( sWelsEncCtx *pCtx )
1430		−{
1431		− SDqLayer *pCurDq = pCtx->ppDqLayerList[0];
1432		− int32_t iNeedAdj = 1;
1433		−#ifdef MT_DEBUG
1434		− int64_t iT0 = WelsTime();
1435		−#endif//MT_DEBUG
1436		−#ifdef TRY_SLICING_BALANCE
1437		−
1438		− pCtx->pCurDqLayer = pCurDq;
1439		−
1440		−#ifdef NOT_ABSOLUTE_BALANCING
1441		− // do not need adjust due to not different at both slices of consumed time
1442		− iNeedAdj = NeedDynamicAdjust( pCtx->pSliceThreading->pSliceConsumeTime[0], pCurDq->pSliceEncCtx->iSliceNumInFrame );
1443		− if ( iNeedAdj )
1444		−#endif//NOT_ABSOLUTE_BALANCING
1445		− DynamicAdjustSlicing( pCtx,
1446		− pCurDq,
1447		− pCtx->pSliceThreading->pSliceComplexRatio[0],
1448		− 0 );
1449		−#endif//TRY_SLICING_BALANCE
1450		−#ifdef MT_DEBUG
1451		− iT0 = WelsTime() - iT0;
1452		− if ( pCtx->pSliceThreading->pFSliceDiff )
1453		− {
1454		− fprintf( pCtx->pSliceThreading->pFSliceDiff,
1455		−#ifdef WIN32
1456		− "%6I64d us adjust time at base spatial layer, iNeedAdj %d, DynamicAdjustSlicing()\n",
1457		−#else
1458		− "%6lld us adjust time at base spatial layer, iNeedAdj %d, DynamicAdjustSlicing()\n",
1459		−#endif//WIN32
1460		− iT0, iNeedAdj );
1461		− }
1462		−#endif//MT_DEBUG
1463		−
1464		− return iNeedAdj;
1465		−}
1466		−
1467		−int32_t AdjustEnhanceLayer( sWelsEncCtx *pCtx, int32_t iCurDid )
1468		−{
1469		−#ifdef MT_DEBUG
1470		− int64_t iT1 = WelsTime();
1471		−#endif//MT_DEBUG
1472		− int32_t iNeedAdj = 1;
1473		− // uiSliceMode of referencing spatial should be SM_FIXEDSLCNUM_SLICE
1474		− // if using spatial base layer for complexity estimation
1475		− const BOOL_T kbModelingFromSpatial = (pCtx->pCurDqLayer->pRefLayer != NULL && iCurDid > 0)
1476		− && (pCtx->pSvcParam->sDependencyLayers[iCurDid-1].sMso.uiSliceMode == SM_FIXEDSLCNUM_SLICE && pCtx->pSvcParam->iMultipleThreadIdc >= pCtx->pSvcParam->sDependencyLayers[iCurDid-1].sMso.sSliceArgument.iSliceNum);
1477		−
1478		− if ( kbModelingFromSpatial ) // using spatial base layer for complexity estimation
1479		− {
1480		−#ifdef TRY_SLICING_BALANCE
1481		−#ifdef NOT_ABSOLUTE_BALANCING
1482		− // do not need adjust due to not different at both slices of consumed time
1483		− iNeedAdj = NeedDynamicAdjust( pCtx->pSliceThreading->pSliceConsumeTime[iCurDid-1], pCtx->pCurDqLayer->pSliceEncCtx->iSliceNumInFrame );
1484		− if ( iNeedAdj )
1485		−#endif//NOT_ABSOLUTE_BALANCING
1486		− DynamicAdjustSlicing( pCtx,
1487		− pCtx->pCurDqLayer,
1488		− pCtx->pSliceThreading->pSliceComplexRatio[iCurDid-1],
1489		− iCurDid
1490		− );
1491		−#endif//TRY_SLICING_BALANCE
1492		− }
1493		− else // use temporal layer for complexity estimation
1494		− {
1495		−#ifdef TRY_SLICING_BALANCE
1496		−#ifdef NOT_ABSOLUTE_BALANCING
1497		− // do not need adjust due to not different at both slices of consumed time
1498		− iNeedAdj = NeedDynamicAdjust( pCtx->pSliceThreading->pSliceConsumeTime[iCurDid], pCtx->pCurDqLayer->pSliceEncCtx->iSliceNumInFrame );
1499		− if ( iNeedAdj )
1500		−#endif//NOT_ABSOLUTE_BALANCING
1501		− DynamicAdjustSlicing( pCtx,
1502		− pCtx->pCurDqLayer,
1503		− pCtx->pSliceThreading->pSliceComplexRatio[iCurDid],
1504		− iCurDid
1505		− );
1506		−#endif//TRY_SLICING_BALANCE
1507		− }
1508		−
1509		−#ifdef MT_DEBUG
1510		− iT1 = WelsTime() - iT1;
1511		− if ( pCtx->pSliceThreading->pFSliceDiff )
1512		− {
1513		− fprintf( pCtx->pSliceThreading->pFSliceDiff,
1514		−#ifdef WIN32
1515		− "%6I64d us adjust time at spatial layer %d, iNeedAdj %d, DynamicAdjustSlicing()\n",
1516		−#else
1517		− "%6lld us adjust time at spatial layer %d, iNeedAdj %d, DynamicAdjustSlicing()\n",
1518		−#endif//WIN32
1519		− iT1, iCurDid, iNeedAdj );
1520		− }
1521		−#endif//MT_DEBUG
1522		−
1523		− return iNeedAdj;
1524		−}
1525		−
1526		−#endif//#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN)
1527		−
1528		−#if defined(MT_ENABLED)
1529		−
1530		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE) && defined(MT_DEBUG)
1531		−void TrackSliceComplexities( sWelsEncCtx *pCtx, const int32_t iCurDid )
1532		−{
1533		− const int32_t kiCountSliceNum = pCtx->pCurDqLayer->pSliceEncCtx->iSliceNumInFrame;
1534		− if ( kiCountSliceNum > 0 )
1535		− {
1536		− int32_t iSliceIdx = 0;
1537		− do {
1538		− fprintf( pCtx->pSliceThreading->pFSliceDiff, "%6.3f complexity pRatio at iDid %d pSlice %d\n", pCtx->pSliceThreading->pSliceComplexRatio[iCurDid][iSliceIdx], iCurDid, iSliceIdx );
1539		− ++ iSliceIdx;
1540		− } while(iSliceIdx < kiCountSliceNum);
1541		− }
1542		−}
1543		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
1544		−
1545		−#if defined(DYNAMIC_SLICE_ASSIGN) && defined(MT_DEBUG)
1546		−void TrackSliceConsumeTime( sWelsEncCtx pCtx, int32_t pDidList, const int32_t iSpatialNum )
1547		−{
1548		− SWelsSvcCodingParam *pPara = NULL;
1549		− int32_t iSpatialIdx = 0;
1550		−
1551		− if ( iSpatialNum > MAX_DEPENDENCY_LAYER )
1552		− return;
1553		−
1554		− pPara = pCtx->pSvcParam;
1555		− while ( iSpatialIdx < iSpatialNum )
1556		− {
1557		− const int32_t kiDid = pDidList[iSpatialIdx];
1558		− SDLayerParam *pDlp = &pPara->sDependencyLayers[kiDid];
1559		− SMulSliceOption *pMso = &pDlp->sMso;
1560		− SDqLayer *pCurDq = pCtx->ppDqLayerList[kiDid];
1561		− SSliceCtx *pSliceCtx= pCurDq->pSliceEncCtx;
1562		− const uint32_t kuiCountSliceNum = pSliceCtx->iSliceNumInFrame;
1563		− if(pCtx->pSliceThreading)
1564		− {
1565		− if ( pCtx->pSliceThreading->pFSliceDiff && pMso->uiSliceMode == SM_FIXEDSLCNUM_SLICE && pPara->iMultipleThreadIdc > 1 && pPara->iMultipleThreadIdc >= kuiCountSliceNum )
1566		− {
1567		− uint32_t i = 0;
1568		− uint32_t uiMaxT = 0;
1569		− int32_t iMaxI = 0;
1570		− while (i < kuiCountSliceNum) {
1571		− if ( pCtx->pSliceThreading->pSliceConsumeTime[kiDid] != NULL )
1572		− fprintf( pCtx->pSliceThreading->pFSliceDiff, "%6d us consume_time coding_idx %d iDid %d pSlice %d\n",
1573		− pCtx->pSliceThreading->pSliceConsumeTime[kiDid][i], pCtx->iCodingIndex, kiDid, i // 1000/);
1574		− if (pCtx->pSliceThreading->pSliceConsumeTime[kiDid][i] > uiMaxT)
1575		− {
1576		− uiMaxT = pCtx->pSliceThreading->pSliceConsumeTime[kiDid][i];
1577		− iMaxI = i;
1578		− }
1579		− ++ i;
1580		− }
1581		− fprintf( pCtx->pSliceThreading->pFSliceDiff, "%6d us consume_time_max coding_idx %d iDid %d pSlice %d\n", uiMaxT, pCtx->iCodingIndex, kiDid, iMaxI // 1000/);
1582		− }
1583		− }
1584		− ++ iSpatialIdx;
1585		− }
1586		−}
1587		−#endif//#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
1588		−
1589		−#endif//MT_ENABLED
1590		−}
1591		−#endif//MT_ENABLED
1592		−
	1	+/*!
	2	+ * \copy
	3	+ * Copyright (c) 2010-2013, Cisco Systems
	4	+ * All rights reserved.
	5	+ *
	6	+ * Redistribution and use in source and binary forms, with or without
	7	+ * modification, are permitted provided that the following conditions
	8	+ * are met:
	9	+ *
	10	+ * * Redistributions of source code must retain the above copyright
	11	+ * notice, this list of conditions and the following disclaimer.
	12	+ *
	13	+ * * Redistributions in binary form must reproduce the above copyright
	14	+ * notice, this list of conditions and the following disclaimer in
	15	+ * the documentation and/or other materials provided with the
	16	+ * distribution.
	17	+ *
	18	+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
	19	+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
	20	+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
	21	+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
	22	+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
	23	+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
	24	+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
	25	+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
	26	+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
	27	+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
	28	+ * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
	29	+ * POSSIBILITY OF SUCH DAMAGE.
	30	+ *
	31	+ *
	32	+ * \file slice_multi_threading.h
	33	+ *
	34	+ * \brief pSlice based multiple threading
	35	+ *
	36	+ * \date 04/16/2010 Created
	37	+ *
	38	+ *************************************************************************************
	39	+ */
	40	+
	41	+#if defined(MT_ENABLED)
	42	+
	43	+#include <assert.h>
	44	+#ifdef __GNUC__
	45	+#include <semaphore.h>
	46	+#ifndef SEM_NAME_MAX
	47	+// length of semaphore name should be system constrained at least on mac 10.7
	48	+#define SEM_NAME_MAX 32
	49	+#endif//SEM_NAME_MAX
	50	+#endif//__GNUC__
	51	+#include "slice_multi_threading.h"
	52	+#include "mt_defs.h"
	53	+#include "nal_encap.h"
	54	+#include "utils.h"
	55	+#include "encoder.h"
	56	+#include "svc_encode_slice.h"
	57	+#include "deblocking.h"
	58	+#include "svc_enc_golomb.h"
	59	+#include "crt_util_safe_x.h" // for safe crt like calls
	60	+#include "rc.h"
	61	+
	62	+#if defined(X86_ASM)
	63	+#include "cpu.h"
	64	+#endif//X86_ASM
	65	+
	66	+#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
	67	+#include "measure_time.h"
	68	+#endif//DYNAMIC_SLICE_ASSIGN
	69	+namespace WelsSVCEnc {
	70	+void UpdateMbListNeighborParallel (SSliceCtx* pSliceCtx,
	71	+ SMB* pMbList,
	72	+ const int32_t uiSliceIdc) {
	73	+ const uint8_t* kpMbMap = pSliceCtx->pOverallMbMap;
	74	+ const int32_t kiMbWidth = pSliceCtx->iMbWidth;
	75	+ int32_t iIdx = pSliceCtx->pFirstMbInSlice[uiSliceIdc];
	76	+ const int32_t kiEndMbInSlice = iIdx + pSliceCtx->pCountMbNumInSlice[uiSliceIdc] - 1;
	77	+
	78	+ do {
	79	+ SMB* pMb = &pMbList[iIdx];
	80	+ uint32_t uiNeighborAvailFlag = 0;
	81	+ const int32_t kiMbXY = pMb->iMbXY;
	82	+ const int32_t kiMbX = pMb->iMbX;
	83	+ const int32_t kiMbY = pMb->iMbY;
	84	+ BOOL_T bLeft;
	85	+ BOOL_T bTop;
	86	+ BOOL_T bLeftTop;
	87	+ BOOL_T bRightTop;
	88	+ int32_t iLeftXY, iTopXY, iLeftTopXY, iRightTopXY;
	89	+
	90	+ iLeftXY = kiMbXY - 1;
	91	+ iTopXY = kiMbXY - kiMbWidth;
	92	+ iLeftTopXY = iTopXY - 1;
	93	+ iRightTopXY = iTopXY + 1;
	94	+
	95	+ bLeft = (kiMbX > 0) && (uiSliceIdc == kpMbMap[iLeftXY]);
	96	+ bTop = (kiMbY > 0) && (uiSliceIdc == kpMbMap[iTopXY]);
	97	+ bLeftTop = (kiMbX > 0) && (kiMbY > 0) && (uiSliceIdc == kpMbMap[iLeftTopXY]);
	98	+ bRightTop = (kiMbX < (kiMbWidth - 1)) && (kiMbY > 0) && (uiSliceIdc == kpMbMap[iRightTopXY]);
	99	+
	100	+ if (bLeft) {
	101	+ uiNeighborAvailFlag \|= LEFT_MB_POS;
	102	+ }
	103	+ if (bTop) {
	104	+ uiNeighborAvailFlag \|= TOP_MB_POS;
	105	+ }
	106	+ if (bLeftTop) {
	107	+ uiNeighborAvailFlag \|= TOPLEFT_MB_POS;
	108	+ }
	109	+ if (bRightTop) {
	110	+ uiNeighborAvailFlag \|= TOPRIGHT_MB_POS;
	111	+ }
	112	+ pMb->uiNeighborAvail = (uint8_t)uiNeighborAvailFlag;
	113	+ pMb->uiSliceIdc = uiSliceIdc;
	114	+
	115	+ ++ iIdx;
	116	+ } while (iIdx <= kiEndMbInSlice);
	117	+}
	118	+
	119	+void CalcSliceComplexRatio (void* pRatio, SSliceCtx* pSliceCtx, uint32_t* pSliceConsume) {
	120	+ float* pRatioList = (float*)pRatio;
	121	+ float fAvI[MAX_SLICES_NUM];
	122	+ float fSumAv = .0f;
	123	+ uint32_t* pSliceTime = (uint32_t*)pSliceConsume;
	124	+ int32_t* pCountMbInSlice = (int32_t*)pSliceCtx->pCountMbNumInSlice;
	125	+ const int32_t kiSliceCount = pSliceCtx->iSliceNumInFrame;
	126	+ int32_t iSliceIdx = 0;
	127	+
	128	+#if defined(X86_ASM)
	129	+ WelsEmms();
	130	+#endif //X86_ASM
	131	+
	132	+ while (iSliceIdx < kiSliceCount) {
	133	+ fAvI[iSliceIdx] = 1.0f * pCountMbInSlice[iSliceIdx] / pSliceTime[iSliceIdx];
	134	+#if defined(ENABLE_TRACE_MT)
	135	+ WelsLog (NULL, WELS_LOG_DEBUG, "[MT] CalcSliceComplexRatio(), pSliceConsumeTime[%d]= %d us, slice_run= %d\n", iSliceIdx,
	136	+ pSliceTime[iSliceIdx], pCountMbInSlice[iSliceIdx]);
	137	+#endif//ENABLE_TRACE_MT
	138	+ fSumAv += fAvI[iSliceIdx];
	139	+
	140	+ ++ iSliceIdx;
	141	+ }
	142	+ while (-- iSliceIdx >= 0) {
	143	+ pRatioList[iSliceIdx] = fAvI[iSliceIdx] / fSumAv;
	144	+ }
	145	+}
	146	+
	147	+#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN) && defined(NOT_ABSOLUTE_BALANCING)
	148	+int32_t NeedDynamicAdjust (void* pConsumeTime, const int32_t iSliceNum) {
	149	+#if !defined(USE_RMSE_SLICE_COMPLEXITY_RATIO_FOR_BALANCING)
	150	+ const float fRatioLower = TOLERANT_BALANCING_RATIO_LOWER (uiSliceNum);
	151	+ const float fRatioUpper = TOLERANT_BALANCING_RATIO_UPPER (uiSliceNum);
	152	+#endif//USE_RMSE_SLICE_COMPLEXITY_RATIO_FOR_BALANCING
	153	+ uint32_t* pSliceConsume = (uint32_t*)pConsumeTime;
	154	+ uint32_t uiTotalConsume = 0;
	155	+ int32_t iSliceIdx = 0;
	156	+ int32_t iNeedAdj = false;
	157	+
	158	+#if defined(X86_ASM)
	159	+ WelsEmms();
	160	+#endif //X86_ASM
	161	+
	162	+ while (iSliceIdx < iSliceNum) {
	163	+ uiTotalConsume += pSliceConsume[iSliceIdx] + pSliceConsume[1 + iSliceIdx];
	164	+ iSliceIdx += 2;
	165	+ }
	166	+ if (uiTotalConsume == 0) {
	167	+#if defined(ENABLE_TRACE_MT)
	168	+ WelsLog (NULL, WELS_LOG_DEBUG, "[MT] NeedDynamicAdjust(), herein do no adjust due first picture, iCountSliceNum= %d\n",
	169	+ iSliceNum);
	170	+#endif//ENABLE_TRACE_MT
	171	+ return false;
	172	+ }
	173	+
	174	+ iSliceIdx = 0;
	175	+#if defined(USE_RMSE_SLICE_COMPLEXITY_RATIO_FOR_BALANCING)
	176	+ float fThr = EPSN; // threshold for various cores cases
	177	+ float fRmse = .0f; // root mean square error of pSlice consume ratios
	178	+ const float kfMeanRatio = 1.0f / iSliceNum;
	179	+ do {
	180	+ const float fRatio = 1.0f * pSliceConsume[iSliceIdx] / uiTotalConsume;
	181	+ const float fDiffRatio = fRatio - kfMeanRatio;
	182	+ fRmse += (fDiffRatio * fDiffRatio);
	183	+ ++ iSliceIdx;
	184	+ } while (iSliceIdx + 1 < iSliceNum);
	185	+ fRmse = sqrtf (fRmse / iSliceNum);
	186	+ if (iSliceNum >= 8) {
	187	+ fThr += THRESHOLD_RMSE_CORE8;
	188	+ } else if (iSliceNum >= 4) {
	189	+ fThr += THRESHOLD_RMSE_CORE4;
	190	+ } else if (iSliceNum >= 2) {
	191	+ fThr += THRESHOLD_RMSE_CORE2;
	192	+ } else
	193	+ fThr = 1.0f;
	194	+ if (fRmse > fThr)
	195	+ iNeedAdj = true;
	196	+#if defined(ENABLE_TRACE_MT)
	197	+ WelsLog (NULL, WELS_LOG_DEBUG,
	198	+ "[MT] NeedDynamicAdjust(), herein adjustment decision is made (iNeedAdj= %d) by: fRmse of pSlice complexity ratios %.6f, the corresponding threshold %.6f, iCountSliceNum %d\n",
	199	+ iNeedAdj, fRmse, fThr, iSliceNum);
	200	+#endif//ENABLE_TRACE_MT
	201	+#else
	202	+ do {
	203	+ const float kfRatio = 1.0f * pSliceConsume[uiSliceIdx] / uiTotalConsume;
	204	+ if (kfRatio + EPSN < fRatioLower \|\| kfRatio > ratio_upper + EPSN) {
	205	+#if defined(ENABLE_TRACE_MT)
	206	+ WelsLog (NULL, WELS_LOG_DEBUG,
	207	+ "[MT] NeedDynamicAdjust(), herein adjustment decision is made by pSlice consume time not balanced at all, uiSliceIdx= %d, comp_ratio= %.6f, pSliceConsumeTime= %d, total_consume_time= %d, iCountSliceNum= %d\n",
	208	+ uiSliceIdx, kfRatio, pSliceConsume[uiSliceIdx], uiTotalConsume, uiSliceNum);
	209	+#endif//ENABLE_TRACE_MT
	210	+ iNeedAdj = true;
	211	+ break;
	212	+ }
	213	+ ++ uiSliceIdx;
	214	+ } while (uiSliceIdx + 1 < uiSliceNum);
	215	+#endif//USE_RMSE_SLICE_COMPLEXITY_RATIO_FOR_BALANCING
	216	+
	217	+ return iNeedAdj;
	218	+}
	219	+#endif//..
	220	+
	221	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	222	+void DynamicAdjustSlicing (sWelsEncCtx* pCtx,
	223	+ SDqLayer* pCurDqLayer,
	224	+ void* pComplexRatio,
	225	+ int32_t iCurDid) {
	226	+ SSliceCtx* pSliceCtx = pCurDqLayer->pSliceEncCtx;
	227	+ const int32_t kiCountSliceNum = pSliceCtx->iSliceNumInFrame;
	228	+ const int32_t kiCountNumMb = pSliceCtx->iMbNumInFrame;
	229	+ int32_t iMinimalMbNum = pSliceCtx->iMbWidth; // in theory we need only 1 SMB, here let it as one SMB row required
	230	+ int32_t iMaximalMbNum = 0; // dynamically assign later
	231	+ float* pSliceComplexRatio = (float*)pComplexRatio;
	232	+ int32_t iMbNumLeft = kiCountNumMb;
	233	+ int32_t iRunLen[MAX_THREADS_NUM] = {0};
	234	+ int32_t iSliceIdx = 0;
	235	+
	236	+ int32_t iNumMbInEachGom;
	237	+ SWelsSvcRc* pWelsSvcRc = &pCtx->pWelsSvcRc[iCurDid];
	238	+ if (pCtx->pSvcParam->bEnableRc) {
	239	+ iNumMbInEachGom = pWelsSvcRc->iNumberMbGom;
	240	+
	241	+ if (iNumMbInEachGom <= 0) {
	242	+ WelsLog (pCtx, WELS_LOG_ERROR,
	243	+ "[MT] DynamicAdjustSlicing(), invalid iNumMbInEachGom= %d from RC, iDid= %d, iCountNumMb= %d\n", iNumMbInEachGom,
	244	+ iCurDid, kiCountNumMb);
	245	+ return;
	246	+ }
	247	+
	248	+ // do not adjust in case no extra iNumMbInEachGom based left for slicing adjustment,
	249	+ // extra MB of non integrated GOM assigned at the last pSlice in default, keep up on early initial result.
	250	+ if (iNumMbInEachGom * kiCountSliceNum >= kiCountNumMb) {
	251	+ return;
	252	+ }
	253	+ iMinimalMbNum = iNumMbInEachGom;
	254	+ }
	255	+
	256	+ if (kiCountSliceNum < 2 \|\| (kiCountSliceNum & 0x01)) // we need suppose uiSliceNum is even for multiple threading
	257	+ return;
	258	+
	259	+ iMaximalMbNum = kiCountNumMb - (kiCountSliceNum - 1) * iMinimalMbNum;
	260	+
	261	+#if defined(X86_ASM)
	262	+ WelsEmms();
	263	+#endif //X86_ASM
	264	+
	265	+#if defined(ENABLE_TRACE_MT)
	266	+ WelsLog (pCtx, WELS_LOG_DEBUG, "[MT] DynamicAdjustSlicing(), iDid= %d, iCountNumMb= %d\n", iCurDid, kiCountNumMb);
	267	+#endif//ENABLE_TRACE_MT
	268	+
	269	+ iSliceIdx = 0;
	270	+ while (iSliceIdx + 1 < kiCountSliceNum) {
	271	+ int32_t iNumMbAssigning = (int32_t) (kiCountNumMb * pSliceComplexRatio[iSliceIdx] + EPSN);
	272	+
	273	+ // GOM boundary aligned
	274	+ if (pCtx->pSvcParam->bEnableRc) {
	275	+ iNumMbAssigning = (int32_t) (1.0f * iNumMbAssigning / iNumMbInEachGom + 0.5f + EPSN) * iNumMbInEachGom;
	276	+ }
	277	+
	278	+ // make sure one GOM at least in each pSlice for safe
	279	+ if (iNumMbAssigning < iMinimalMbNum)
	280	+ iNumMbAssigning = iMinimalMbNum;
	281	+ else if (iNumMbAssigning > iMaximalMbNum)
	282	+ iNumMbAssigning = iMaximalMbNum;
	283	+
	284	+ assert (iNumMbAssigning > 0);
	285	+
	286	+ iMbNumLeft -= iNumMbAssigning;
	287	+ if (iMbNumLeft <= 0) { // error due to we can not support slice_skip now yet, do not adjust this time
	288	+ assert (0);
	289	+ return;
	290	+ }
	291	+ iRunLen[iSliceIdx] = iNumMbAssigning;
	292	+#if defined(ENABLE_TRACE_MT)
	293	+ WelsLog (pCtx, WELS_LOG_DEBUG,
	294	+ "[MT] DynamicAdjustSlicing(), uiSliceIdx= %d, pSliceComplexRatio= %.2f, slice_run_org= %d, slice_run_adj= %d\n",
	295	+ iSliceIdx, pSliceComplexRatio[iSliceIdx], pSliceCtx->pCountMbNumInSlice[iSliceIdx], iNumMbAssigning);
	296	+#endif//ENABLE_TRACE_MT
	297	+ ++ iSliceIdx;
	298	+ iMaximalMbNum = iMbNumLeft - (kiCountSliceNum - iSliceIdx - 1) * iMinimalMbNum; // get maximal num_mb in left parts
	299	+ }
	300	+ iRunLen[iSliceIdx] = iMbNumLeft;
	301	+#if defined(ENABLE_TRACE_MT)
	302	+ WelsLog (pCtx, WELS_LOG_DEBUG,
	303	+ "[MT] DynamicAdjustSlicing(), iSliceIdx= %d, pSliceComplexRatio= %.2f, slice_run_org= %d, slice_run_adj= %d\n",
	304	+ iSliceIdx, pSliceComplexRatio[iSliceIdx], pSliceCtx->pCountMbNumInSlice[iSliceIdx], iMbNumLeft);
	305	+#endif//ENABLE_TRACE_MT
	306	+
	307	+
	308	+ if (DynamicAdjustSlicePEncCtxAll (pSliceCtx, iRunLen) == 0) {
	309	+ const int32_t kiThreadNum = pCtx->pSvcParam->iCountThreadsNum;
	310	+ int32_t iThreadIdx = 0;
	311	+ do {
	312	+#ifdef WIN32
	313	+ WelsEventSignal (&pCtx->pSliceThreading->pUpdateMbListEvent[iThreadIdx]);
	314	+#else
	315	+ WelsEventSignal (pCtx->pSliceThreading->pUpdateMbListEvent[iThreadIdx]);
	316	+#endif//WIN32
	317	+ ++ iThreadIdx;
	318	+ } while (iThreadIdx < kiThreadNum);
	319	+
	320	+ WelsMultipleEventsWaitAllBlocking (kiThreadNum, &pCtx->pSliceThreading->pFinUpdateMbListEvent[0]);
	321	+ }
	322	+}
	323	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	324	+
	325	+#ifdef PACKING_ONE_SLICE_PER_LAYER
	326	+void ResetEnvMt (sWelsEncCtx* pCtx) {
	327	+ const int16_t kiSliceCount = pCtx->iMaxSliceCount;
	328	+ int32_t iIdx = 0;
	329	+
	330	+ while (iIdx < kiSliceCount) {
	331	+ SWelsSliceBs* pSliceBs = &pCtx->pSliceBs[iIdx];
	332	+ pSliceBs->uiBsPos = 0;
	333	+ ++ iIdx;
	334	+ }
	335	+}
	336	+#endif//PACKING_ONE_SLICE_PER_LAYER
	337	+
	338	+int32_t RequestMtResource (sWelsEncCtx** ppCtx, SWelsSvcCodingParam* pCodingParam, const int32_t iCountBsLen,
	339	+ const int32_t iTargetSpatialBsSize) {
	340	+ CMemoryAlign* pMa = NULL;
	341	+ SWelsSvcCodingParam* pPara = NULL;
	342	+ SSliceThreading* pSmt = NULL;
	343	+ SWelsSliceBs* pSliceB = NULL;
	344	+ uint8_t* pBsBase = NULL;
	345	+ int32_t iNumSpatialLayers = 0;
	346	+ int32_t iThreadNum = 0;
	347	+ int32_t iIdx = 0;
	348	+ int32_t iSliceBsBufferSize = 0;
	349	+ int16_t iMaxSliceNum = 1;
	350	+
	351	+ if (NULL == ppCtx \|\| NULL == pCodingParam \|\| NULL == *ppCtx \|\| iCountBsLen <= 0)
	352	+ return 1;
	353	+
	354	+ pMa = (*ppCtx)->pMemAlign;
	355	+ pPara = pCodingParam;
	356	+ iNumSpatialLayers = pPara->iNumDependencyLayer;
	357	+ iThreadNum = pPara->iCountThreadsNum;
	358	+ iMaxSliceNum = (*ppCtx)->iMaxSliceCount;
	359	+
	360	+ pSmt = (SSliceThreading*)pMa->WelsMalloc (sizeof (SSliceThreading), "SSliceThreading");
	361	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt), FreeMemorySvc (ppCtx))
	362	+ (*ppCtx)->pSliceThreading = pSmt;
	363	+ pSmt->pThreadPEncCtx = (SSliceThreadPrivateData)pMa->WelsMalloc (sizeof (SSliceThreadPrivateData) iThreadNum,
	364	+ "pThreadPEncCtx");
	365	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pThreadPEncCtx), FreeMemorySvc (ppCtx))
	366	+ pSmt->pThreadHandles = (WELS_THREAD_HANDLE)pMa->WelsMalloc (sizeof (WELS_THREAD_HANDLE) iThreadNum,
	367	+ "pThreadHandles");
	368	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pThreadHandles), FreeMemorySvc (ppCtx))
	369	+
	370	+#ifdef WIN32
	371	+ pSmt->pSliceCodedEvent = (WELS_EVENT)pMa->WelsMalloc (sizeof (WELS_EVENT) iThreadNum, "pSliceCodedEvent");
	372	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pSliceCodedEvent), FreeMemorySvc (ppCtx))
	373	+ pSmt->pReadySliceCodingEvent = (WELS_EVENT)pMa->WelsMalloc (sizeof (WELS_EVENT) iThreadNum,
	374	+ "pReadySliceCodingEvent");
	375	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pReadySliceCodingEvent), FreeMemorySvc (ppCtx))
	376	+ pSmt->pFinSliceCodingEvent = (WELS_EVENT)pMa->WelsMalloc (sizeof (WELS_EVENT) iThreadNum, "pFinSliceCodingEvent");
	377	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pFinSliceCodingEvent), FreeMemorySvc (ppCtx))
	378	+#endif//WIN32
	379	+
	380	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	381	+#if defined(__GNUC__)
	382	+ pSmt->pUpdateMbListThrdHandles = (WELS_THREAD_HANDLE)pMa->WelsMalloc (sizeof (WELS_THREAD_HANDLE) iThreadNum,
	383	+ "pUpdateMbListThrdHandles");
	384	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pUpdateMbListThrdHandles), FreeMemorySvc (ppCtx))
	385	+#endif//__GNUC__
	386	+#ifdef WIN32
	387	+ pSmt->pUpdateMbListEvent = (WELS_EVENT)pMa->WelsMalloc (sizeof (WELS_EVENT) iThreadNum, "pUpdateMbListEvent");
	388	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pUpdateMbListEvent), FreeMemorySvc (ppCtx))
	389	+ pSmt->pFinUpdateMbListEvent = (WELS_EVENT)pMa->WelsMalloc (sizeof (WELS_EVENT) iThreadNum, "pFinUpdateMbListEvent");
	390	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pFinUpdateMbListEvent), FreeMemorySvc (ppCtx))
	391	+#endif//WIN32
	392	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	393	+
	394	+#ifdef WIN32
	395	+ pSmt->pExitEncodeEvent = (WELS_EVENT)pMa->WelsMalloc (sizeof (WELS_EVENT) iThreadNum, "pExitEncodeEvent");
	396	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pExitEncodeEvent), FreeMemorySvc (ppCtx))
	397	+#endif//WIN32
	398	+
	399	+#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
	400	+ iIdx = 0;
	401	+ while (iIdx < iNumSpatialLayers) {
	402	+ SMulSliceOption* pMso = &pPara->sDependencyLayers[iIdx].sMso;
	403	+ const int32_t kiSliceNum = pMso->sSliceArgument.iSliceNum;
	404	+ if (pMso->uiSliceMode == SM_FIXEDSLCNUM_SLICE && pPara->iMultipleThreadIdc > 1
	405	+ && pPara->iMultipleThreadIdc >= kiSliceNum) {
	406	+ pSmt->pSliceConsumeTime[iIdx] = (uint32_t)pMa->WelsMallocz (kiSliceNum sizeof (uint32_t), "pSliceConsumeTime[]");
	407	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pSliceConsumeTime[iIdx]), FreeMemorySvc (ppCtx))
	408	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	409	+ pSmt->pSliceComplexRatio[iIdx] = (float)pMa->WelsMalloc (kiSliceNum sizeof (float), "pSliceComplexRatio[]");
	410	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pSliceComplexRatio[iIdx]), FreeMemorySvc (ppCtx))
	411	+#endif//TRY_SLICING_BALANCE
	412	+ } else {
	413	+ pSmt->pSliceConsumeTime[iIdx] = NULL;
	414	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	415	+ pSmt->pSliceComplexRatio[iIdx] = NULL;
	416	+#endif//TRY_SLICING_BALANCE
	417	+ }
	418	+ ++ iIdx;
	419	+ }
	420	+ // NULL for pSliceConsumeTime[iIdx]: iIdx from iNumSpatialLayers to MAX_DEPENDENCY_LAYERS
	421	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
	422	+
	423	+#ifdef MT_DEBUG
	424	+ // file handle for MT debug
	425	+ pSmt->pFSliceDiff = NULL;
	426	+
	427	+ if (pSmt->pFSliceDiff) {
	428	+ fclose (pSmt->pFSliceDiff);
	429	+ pSmt->pFSliceDiff = NULL;
	430	+ }
	431	+#ifdef WIN32
	432	+ pSmt->pFSliceDiff = fopen (".\\slice_time.txt", "wt+");
	433	+#else
	434	+ pSmt->pFSliceDiff = fopen ("/tmp/slice_time.txt", "wt+");
	435	+#endif//WIN32
	436	+#endif//MT_DEBUG
	437	+
	438	+#if defined(ENABLE_TRACE_MT)
	439	+ WelsLog ((ppCtx), WELS_LOG_INFO, "encpEncCtx= 0x%p\n", (void) (*ppCtx));
	440	+#endif//ENABLE_TRACE_MT
	441	+
	442	+ iIdx = 0;
	443	+ while (iIdx < iThreadNum) {
	444	+#ifdef __GNUC__ // for posix threading
	445	+ str_t name[SEM_NAME_MAX] = {0};
	446	+ int32_t used_len = 0;
	447	+ WELS_THREAD_ERROR_CODE err = 0;
	448	+#endif//__GNUC__
	449	+ pSmt->pThreadPEncCtx[iIdx].pWelsPEncCtx = (void) (ppCtx);
	450	+ pSmt->pThreadPEncCtx[iIdx].iSliceIndex = iIdx;
	451	+ pSmt->pThreadPEncCtx[iIdx].iThreadIndex = iIdx;
	452	+ pSmt->pThreadHandles[iIdx] = 0;
	453	+
	454	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	455	+#ifdef WIN32
	456	+ WelsEventInit (&pSmt->pUpdateMbListEvent[iIdx]);
	457	+ WelsEventInit (&pSmt->pFinUpdateMbListEvent[iIdx]);
	458	+#else
	459	+ // length of semaphore name should be system constrained at least on mac 10.7
	460	+ SNPRINTF (name, SEM_NAME_MAX, "ud%d%p", iIdx, (void) (ppCtx));
	461	+ err = WelsEventOpen (&pSmt->pUpdateMbListEvent[iIdx], name);
	462	+#if defined(ENABLE_TRACE_MT)
	463	+ WelsLog ((*ppCtx), WELS_LOG_INFO, "[MT] Open pUpdateMbListEvent%d named(%s) ret%d err%d\n", iIdx, name, err, errno);
	464	+#endif
	465	+ used_len = SNPRINTF (name, SEM_NAME_MAX, "fu%d%p", iIdx, (void) (ppCtx));
	466	+ name[used_len] = '\0';
	467	+ err = WelsEventOpen (&pSmt->pFinUpdateMbListEvent[iIdx], name);
	468	+#if defined(ENABLE_TRACE_MT)
	469	+ WelsLog ((*ppCtx), WELS_LOG_INFO, "[MT] Open pFinUpdateMbListEvent%d named(%s) ret%d err%d\n", iIdx, name, err, errno);
	470	+#endif
	471	+#endif//WIN32
	472	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	473	+
	474	+#ifdef WIN32
	475	+ WelsEventInit (&pSmt->pSliceCodedEvent[iIdx]);
	476	+ WelsEventInit (&pSmt->pReadySliceCodingEvent[iIdx]);
	477	+ WelsEventInit (&pSmt->pFinSliceCodingEvent[iIdx]);
	478	+ WelsEventInit (&pSmt->pExitEncodeEvent[iIdx]);
	479	+#else
	480	+ used_len = SNPRINTF (name, SEM_NAME_MAX, "sc%d%p", iIdx, (void) (ppCtx));
	481	+ name[used_len] = '\0';
	482	+ err = WelsEventOpen (&pSmt->pSliceCodedEvent[iIdx], name);
	483	+#if defined(ENABLE_TRACE_MT)
	484	+ WelsLog ((*ppCtx), WELS_LOG_INFO, "[MT] Open pSliceCodedEvent%d named(%s) ret%d err%d\n", iIdx, name, err, errno);
	485	+#endif
	486	+ used_len = SNPRINTF (name, SEM_NAME_MAX, "rc%d%p", iIdx, (void) (ppCtx));
	487	+ name[used_len] = '\0';
	488	+ err = WelsEventOpen (&pSmt->pReadySliceCodingEvent[iIdx], name);
	489	+#if defined(ENABLE_TRACE_MT)
	490	+ WelsLog ((*ppCtx), WELS_LOG_INFO, "[MT] Open pReadySliceCodingEvent%d = 0x%p named(%s) ret%d err%d\n", iIdx,
	491	+ (void)pSmt->pReadySliceCodingEvent[iIdx]), (void) (*ppCtx), err, errno);
	492	+#endif
	493	+#endif//WIN32
	494	+
	495	+ ++ iIdx;
	496	+ }
	497	+
	498	+#ifdef PACKING_ONE_SLICE_PER_LAYER
	499	+ pSmt->pCountBsSizeInPartition = (uint32_t)pMa->WelsMalloc (sizeof (uint32_t) iThreadNum, "pCountBsSizeInPartition");
	500	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pCountBsSizeInPartition), FreeMemorySvc (ppCtx))
	501	+#endif//PACKING_ONE_SLICE_PER_LAYER
	502	+
	503	+ WelsMutexInit (&pSmt->mutexSliceNumUpdate);
	504	+
	505	+ (ppCtx)->pSliceBs = (SWelsSliceBs)pMa->WelsMalloc (sizeof (SWelsSliceBs) * iMaxSliceNum, "pSliceBs");
	506	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == (*ppCtx)->pSliceBs), FreeMemorySvc (ppCtx))
	507	+
	508	+ pBsBase = (*ppCtx)->pFrameBs + iCountBsLen;
	509	+ pSliceB = (*ppCtx)->pSliceBs;
	510	+ iSliceBsBufferSize = iTargetSpatialBsSize;
	511	+ iIdx = 0;
	512	+ while (iIdx < iMaxSliceNum) {
	513	+ pSliceB->pBsBuffer = (uint8_t*)pMa->WelsMalloc (iSliceBsBufferSize, "pSliceB->pBsBuffer");
	514	+
	515	+ WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSliceB->pBsBuffer), FreeMemorySvc (ppCtx))
	516	+ pSliceB->uiSize = iSliceBsBufferSize;
	517	+
	518	+ if (iIdx > 0) {
	519	+ pSliceB->pBs = pBsBase;
	520	+ pSliceB->uiBsPos = 0;
	521	+ pBsBase += iSliceBsBufferSize;
	522	+ } else {
	523	+ pSliceB->pBs = NULL;
	524	+ pSliceB->uiBsPos = 0;
	525	+ }
	526	+ ++ pSliceB;
	527	+ ++ iIdx;
	528	+ }
	529	+
	530	+#if defined(ENABLE_TRACE_MT)
	531	+ WelsLog ((*ppCtx), WELS_LOG_INFO, "RequestMtResource(), iThreadNum=%d, iCountSliceNum= %d\n", pPara->iCountThreadsNum,
	532	+ iMaxSliceNum);
	533	+#endif
	534	+
	535	+ return 0;
	536	+}
	537	+
	538	+void ReleaseMtResource (sWelsEncCtx** ppCtx) {
	539	+ SWelsSliceBs* pSliceB = NULL;
	540	+ SWelsSvcCodingParam* pCodingParam = NULL;
	541	+ SSliceThreading* pSmt = NULL;
	542	+ CMemoryAlign* pMa = NULL;
	543	+ int32_t iIdx = 0;
	544	+ int32_t iThreadNum = 0;
	545	+ int16_t uiSliceNum = 0;
	546	+
	547	+ if (NULL == ppCtx \|\| NULL == *ppCtx)
	548	+ return;
	549	+
	550	+ pMa = (*ppCtx)->pMemAlign;
	551	+ pCodingParam = (*ppCtx)->pSvcParam;
	552	+ uiSliceNum = (*ppCtx)->iMaxSliceCount;
	553	+ iThreadNum = (*ppCtx)->pSvcParam->iCountThreadsNum;
	554	+ pSmt = (*ppCtx)->pSliceThreading;
	555	+
	556	+ if (NULL == pSmt)
	557	+ return;
	558	+
	559	+ while (iIdx < iThreadNum) {
	560	+#ifdef WIN32
	561	+ if (pSmt->pThreadHandles != NULL && pSmt->pThreadHandles[iIdx] != NULL)
	562	+ WelsThreadDestroy (&pSmt->pThreadHandles[iIdx]);
	563	+
	564	+ if (pSmt->pSliceCodedEvent != NULL)
	565	+ WelsEventDestroy (&pSmt->pSliceCodedEvent[iIdx]);
	566	+ if (pSmt->pReadySliceCodingEvent != NULL)
	567	+ WelsEventDestroy (&pSmt->pReadySliceCodingEvent[iIdx]);
	568	+ if (pSmt->pFinSliceCodingEvent != NULL)
	569	+ WelsEventDestroy (&pSmt->pFinSliceCodingEvent[iIdx]);
	570	+ if (pSmt->pExitEncodeEvent != NULL)
	571	+ WelsEventDestroy (&pSmt->pExitEncodeEvent[iIdx]);
	572	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	573	+ if (pSmt->pUpdateMbListEvent != NULL)
	574	+ WelsEventDestroy (&pSmt->pUpdateMbListEvent[iIdx]);
	575	+ if (pSmt->pFinUpdateMbListEvent != NULL)
	576	+ WelsEventDestroy (&pSmt->pFinUpdateMbListEvent[iIdx]);
	577	+#endif//DYNAMIC_SLICE_ASSIGN && TRY_SLICING_BALANCE
	578	+#else
	579	+ str_t ename[SEM_NAME_MAX] = {0};
	580	+ int32_t used_len = 0;
	581	+ // length of semaphore name should be system constrained at least on mac 10.7
	582	+ SNPRINTF (ename, SEM_NAME_MAX, "sc%d%p", iIdx, (void) (ppCtx));
	583	+ WelsEventClose (pSmt->pSliceCodedEvent[iIdx], ename);
	584	+ used_len = SNPRINTF (ename, SEM_NAME_MAX, "rc%d%p", iIdx, (void) (ppCtx));
	585	+ ename[used_len] = '\0';
	586	+ WelsEventClose (pSmt->pReadySliceCodingEvent[iIdx], ename);
	587	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	588	+ used_len = SNPRINTF (ename, SEM_NAME_MAX, "ud%d%p", iIdx, (void) (ppCtx));
	589	+ ename[used_len] = '\0';
	590	+ WelsEventClose (pSmt->pUpdateMbListEvent[iIdx], ename);
	591	+ used_len = SNPRINTF (ename, SEM_NAME_MAX, "fu%d%p", iIdx, (void) (ppCtx));
	592	+ ename[used_len] = '\0';
	593	+ WelsEventClose (pSmt->pFinUpdateMbListEvent[iIdx], ename);
	594	+#endif//DYNAMIC_SLICE_ASSIGN && TRY_SLICING_BALANCE
	595	+#endif//WIN32
	596	+
	597	+ ++ iIdx;
	598	+ }
	599	+
	600	+#ifdef WIN32
	601	+ if (pSmt->pExitEncodeEvent != NULL) {
	602	+ pMa->WelsFree (pSmt->pExitEncodeEvent, "pExitEncodeEvent");
	603	+ pSmt->pExitEncodeEvent = NULL;
	604	+ }
	605	+ if (pSmt->pSliceCodedEvent != NULL) {
	606	+ pMa->WelsFree (pSmt->pSliceCodedEvent, "pSliceCodedEvent");
	607	+ pSmt->pSliceCodedEvent = NULL;
	608	+ }
	609	+ if (pSmt->pReadySliceCodingEvent != NULL) {
	610	+ pMa->WelsFree (pSmt->pReadySliceCodingEvent, "pReadySliceCodingEvent");
	611	+ pSmt->pReadySliceCodingEvent = NULL;
	612	+ }
	613	+ if (pSmt->pFinSliceCodingEvent != NULL) {
	614	+ pMa->WelsFree (pSmt->pFinSliceCodingEvent, "pFinSliceCodingEvent");
	615	+ pSmt->pFinSliceCodingEvent = NULL;
	616	+ }
	617	+#endif//WIN32
	618	+
	619	+#ifdef PACKING_ONE_SLICE_PER_LAYER
	620	+ if (NULL != pSmt->pCountBsSizeInPartition) {
	621	+ pMa->WelsFree (pSmt->pCountBsSizeInPartition, "pCountBsSizeInPartition");
	622	+ pSmt->pCountBsSizeInPartition = NULL;
	623	+ }
	624	+#endif//PACKING_ONE_SLICE_PER_LAYER
	625	+ WelsMutexDestroy (&pSmt->mutexSliceNumUpdate);
	626	+
	627	+ if (pSmt->pThreadPEncCtx != NULL) {
	628	+ pMa->WelsFree (pSmt->pThreadPEncCtx, "pThreadPEncCtx");
	629	+ pSmt->pThreadPEncCtx = NULL;
	630	+ }
	631	+ if (pSmt->pThreadHandles != NULL) {
	632	+ pMa->WelsFree (pSmt->pThreadHandles, "pThreadHandles");
	633	+ pSmt->pThreadHandles = NULL;
	634	+ }
	635	+
	636	+ pSliceB = (*ppCtx)->pSliceBs;
	637	+ iIdx = 0;
	638	+ while (pSliceB != NULL && iIdx < uiSliceNum) {
	639	+ if (pSliceB->pBsBuffer) {
	640	+ pMa->WelsFree (pSliceB->pBsBuffer, "pSliceB->pBsBuffer");
	641	+ pSliceB->pBsBuffer = NULL;
	642	+ pSliceB->uiSize = 0;
	643	+ }
	644	+ ++ iIdx;
	645	+ ++ pSliceB;
	646	+ }
	647	+ if ((*ppCtx)->pSliceBs != NULL) {
	648	+ pMa->WelsFree ((*ppCtx)->pSliceBs, "pSliceBs");
	649	+ (*ppCtx)->pSliceBs = NULL;
	650	+ }
	651	+#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
	652	+ if (pSmt->pSliceConsumeTime != NULL) {
	653	+ iIdx = 0;
	654	+ while (iIdx < pCodingParam->iNumDependencyLayer) {
	655	+ if (pSmt->pSliceConsumeTime[iIdx]) {
	656	+ pMa->WelsFree (pSmt->pSliceConsumeTime[iIdx], "pSliceConsumeTime[]");
	657	+ pSmt->pSliceConsumeTime[iIdx] = NULL;
	658	+ }
	659	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	660	+ if (pSmt->pSliceComplexRatio[iIdx] != NULL) {
	661	+ pMa->WelsFree (pSmt->pSliceComplexRatio[iIdx], "pSliceComplexRatio[]");
	662	+ pSmt->pSliceComplexRatio[iIdx] = NULL;
	663	+ }
	664	+#endif//TRY_SLICING_BALANCE
	665	+ ++ iIdx;
	666	+ }
	667	+ }
	668	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
	669	+
	670	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	671	+
	672	+#ifdef WIN32
	673	+ if (pSmt->pUpdateMbListEvent != NULL) {
	674	+ pMa->WelsFree (pSmt->pUpdateMbListEvent, "pUpdateMbListEvent");
	675	+ pSmt->pUpdateMbListEvent = NULL;
	676	+ }
	677	+ if (pSmt->pFinUpdateMbListEvent != NULL) {
	678	+ pMa->WelsFree (pSmt->pFinUpdateMbListEvent, "pFinUpdateMbListEvent");
	679	+ pSmt->pFinUpdateMbListEvent = NULL;
	680	+ }
	681	+#else
	682	+ if (pSmt->pUpdateMbListThrdHandles) {
	683	+ pMa->WelsFree (pSmt->pUpdateMbListThrdHandles, "pUpdateMbListThrdHandles");
	684	+ pSmt->pUpdateMbListThrdHandles = NULL;
	685	+ }
	686	+#endif//WIN32
	687	+
	688	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	689	+
	690	+#ifdef MT_DEBUG
	691	+ // file handle for debug
	692	+ if (pSmt->pFSliceDiff) {
	693	+ fclose (pSmt->pFSliceDiff);
	694	+ pSmt->pFSliceDiff = NULL;
	695	+ }
	696	+#endif//MT_DEBUG
	697	+ pMa->WelsFree ((*ppCtx)->pSliceThreading, "SSliceThreading");
	698	+ (*ppCtx)->pSliceThreading = NULL;
	699	+}
	700	+
	701	+int32_t AppendSliceToFrameBs (sWelsEncCtx* pCtx, SLayerBSInfo* pLbi, const int32_t iSliceCount) {
	702	+ SWelsSvcCodingParam* pCodingParam = pCtx->pSvcParam;
	703	+ SDLayerParam* pDlp = &pCodingParam->sDependencyLayers[pCtx->uiDependencyId];
	704	+ SWelsSliceBs* pSliceBs = NULL;
	705	+ const BOOL_T kbIsDynamicSlicingMode = (pDlp->sMso.uiSliceMode == SM_DYN_SLICE);
	706	+ int32_t iLayerSize = 0;
	707	+ int32_t iNalIdxBase = pLbi->iNalCount;
	708	+ int32_t iSliceIdx = 0;
	709	+
	710	+ if (!kbIsDynamicSlicingMode) {
	711	+ pSliceBs = &pCtx->pSliceBs[0];
	712	+ iLayerSize = pSliceBs->uiBsPos; // assign with base pSlice first
	713	+ iSliceIdx = 1; // pSlice 0 bs has been written to pFrameBs yet by now, so uiSliceIdx base should be 1
	714	+ while (iSliceIdx < iSliceCount) {
	715	+ ++ pSliceBs;
	716	+ if (pSliceBs != NULL && pSliceBs->uiBsPos > 0) {
	717	+ int32_t iNalIdx = 0;
	718	+ const int32_t iCountNal = pSliceBs->iNalIndex;
	719	+
	720	+#if MT_DEBUG_BS_WR
	721	+ assert (pSliceBs->bSliceCodedFlag);
	722	+#endif//MT_DEBUG_BS_WR
	723	+
	724	+ memmove (pCtx->pFrameBs + pCtx->iPosBsBuffer, pSliceBs->pBs, pSliceBs->uiBsPos); // confirmed_safe_unsafe_usage
	725	+ pCtx->iPosBsBuffer += pSliceBs->uiBsPos;
	726	+
	727	+ iLayerSize += pSliceBs->uiBsPos;
	728	+
	729	+ while (iNalIdx < iCountNal) {
	730	+ pLbi->iNalLengthInByte[iNalIdxBase + iNalIdx] = pSliceBs->iNalLen[iNalIdx];
	731	+ ++ iNalIdx;
	732	+ }
	733	+ pLbi->iNalCount += iCountNal;
	734	+ iNalIdxBase += iCountNal;
	735	+ }
	736	+ ++ iSliceIdx;
	737	+ }
	738	+ } else { // for SM_DYN_SLICE
	739	+ const int32_t kiPartitionCnt = iSliceCount;
	740	+ int32_t iPartitionIdx = 0;
	741	+
	742	+ // due partition_0 has been written to pFrameBsBuffer
	743	+ // so iLayerSize need add it
	744	+ while (iPartitionIdx < kiPartitionCnt) {
	745	+ const int32_t kiCountSlicesCoded = pCtx->pCurDqLayer->pNumSliceCodedOfPartition[iPartitionIdx];
	746	+ int32_t iIdx = 0;
	747	+
	748	+ iSliceIdx = iPartitionIdx;
	749	+ while (iIdx < kiCountSlicesCoded) {
	750	+ pSliceBs = &pCtx->pSliceBs[iSliceIdx];
	751	+ if (pSliceBs != NULL && pSliceBs->uiBsPos > 0) {
	752	+ if (iPartitionIdx > 0) {
	753	+ int32_t iNalIdx = 0;
	754	+ const int32_t iCountNal = pSliceBs->iNalIndex;
	755	+
	756	+ memmove (pCtx->pFrameBs + pCtx->iPosBsBuffer, pSliceBs->pBs, pSliceBs->uiBsPos); // confirmed_safe_unsafe_usage
	757	+ pCtx->iPosBsBuffer += pSliceBs->uiBsPos;
	758	+
	759	+ iLayerSize += pSliceBs->uiBsPos;
	760	+
	761	+ while (iNalIdx < iCountNal) {
	762	+ pLbi->iNalLengthInByte[iNalIdxBase + iNalIdx] = pSliceBs->iNalLen[iNalIdx];
	763	+ ++ iNalIdx;
	764	+ }
	765	+ pLbi->iNalCount += iCountNal;
	766	+ iNalIdxBase += iCountNal;
	767	+ } else {
	768	+ iLayerSize += pSliceBs->uiBsPos;
	769	+ }
	770	+ }
	771	+
	772	+ iSliceIdx += kiPartitionCnt;
	773	+ ++ iIdx;
	774	+ }
	775	+ ++ iPartitionIdx;
	776	+ }
	777	+ }
	778	+
	779	+ return iLayerSize;
	780	+}
	781	+
	782	+int32_t WriteSliceToFrameBs (sWelsEncCtx* pCtx, SLayerBSInfo* pLbi, uint8_t* pFrameBsBuffer, const int32_t iSliceIdx) {
	783	+ SWelsSliceBs* pSliceBs = &pCtx->pSliceBs[iSliceIdx];
	784	+ SNalUnitHeaderExt* pNalHdrExt = &pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt;
	785	+ uint8_t* pDst = pFrameBsBuffer;
	786	+ int32_t pNalLen[2];
	787	+ int32_t iSliceSize = 0;
	788	+ const int32_t kiNalCnt = pSliceBs->iNalIndex;
	789	+ int32_t iNalIdx = 0;
	790	+#if !defined(PACKING_ONE_SLICE_PER_LAYER)
	791	+ const int32_t iFirstSlice = (iSliceIdx == 0);
	792	+ int32_t iNalBase = iFirstSlice ? 0 : pLbi->iNalCount;
	793	+#else
	794	+ int32_t iNalBase = 0;
	795	+#endif//!PACKING_ONE_SLICE_PER_LAYER
	796	+
	797	+ while (iNalIdx < kiNalCnt) {
	798	+ iSliceSize += WelsEncodeNalExt (&pSliceBs->sNalList[iNalIdx], pNalHdrExt, pDst, &pNalLen[iNalIdx]);
	799	+ pDst += pNalLen[iNalIdx];
	800	+ pLbi->iNalLengthInByte[iNalBase + iNalIdx] = pNalLen[iNalIdx];
	801	+
	802	+ ++ iNalIdx;
	803	+ }
	804	+
	805	+#if !defined(PACKING_ONE_SLICE_PER_LAYER)
	806	+ pSliceBs->uiBsPos = iSliceSize;
	807	+ if (iFirstSlice) {
	808	+ // pBsBuffer has been updated at coding_slice_0_in_encoder_mother_thread()
	809	+ pLbi->uiLayerType = VIDEO_CODING_LAYER;
	810	+ pLbi->uiSpatialId = pNalHdrExt->uiDependencyId;
	811	+ pLbi->uiTemporalId = pNalHdrExt->uiTemporalId;
	812	+ pLbi->uiQualityId = 0;
	813	+ pLbi->uiPriorityId = 0;
	814	+ pLbi->iNalCount = kiNalCnt;
	815	+ } else {
	816	+ pLbi->iNalCount += kiNalCnt;
	817	+ }
	818	+#else
	819	+ pLbi->uiLayerType = VIDEO_CODING_LAYER;
	820	+ pLbi->uiSpatialId = pNalHdrExt->uiDependencyId;
	821	+ pLbi->uiTemporalId = pNalHdrExt->uiTemporalId;
	822	+ pLbi->uiQualityId = 0;
	823	+ pLbi->uiPriorityId = 0;
	824	+ pLbi->iNalCount = kiNalCnt;
	825	+#endif//PACKING_ONE_SLICE_PER_LAYER
	826	+
	827	+ return iSliceSize;
	828	+}
	829	+
	830	+int32_t WriteSliceBs (sWelsEncCtx* pCtx, uint8_t* pSliceBsBuf, const int32_t iSliceIdx) {
	831	+ SWelsSliceBs* pSliceBs = &pCtx->pSliceBs[iSliceIdx];
	832	+ SNalUnitHeaderExt* pNalHdrExt = &pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt;
	833	+ uint8_t* pDst = pSliceBsBuf;
	834	+ int32_t* pNalLen = &pSliceBs->iNalLen[0];
	835	+ int32_t iSliceSize = 0;
	836	+ const int32_t kiNalCnt = pSliceBs->iNalIndex;
	837	+ int32_t iNalIdx = 0;
	838	+
	839	+ assert (kiNalCnt <= 2);
	840	+ if (kiNalCnt > 2)
	841	+ return 0;
	842	+
	843	+ while (iNalIdx < kiNalCnt) {
	844	+ iSliceSize += WelsEncodeNalExt (&pSliceBs->sNalList[iNalIdx], pNalHdrExt, pDst, &pNalLen[iNalIdx]);
	845	+ pDst += pNalLen[iNalIdx];
	846	+
	847	+ ++ iNalIdx;
	848	+ }
	849	+ pSliceBs->uiBsPos = iSliceSize;
	850	+
	851	+ return iSliceSize;
	852	+}
	853	+
	854	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	855	+#if defined(__GNUC__)
	856	+WELS_THREAD_ROUTINE_TYPE UpdateMbListThreadProc (void* arg) {
	857	+ SSliceThreadPrivateData* pPrivateData = (SSliceThreadPrivateData*)arg;
	858	+ sWelsEncCtx* pEncPEncCtx = NULL;
	859	+ SDqLayer* pCurDq = NULL;
	860	+ int32_t iSliceIdx = -1;
	861	+ int32_t iEventIdx = -1;
	862	+ WELS_THREAD_ERROR_CODE iWaitRet = WELS_THREAD_ERROR_GENERIAL;
	863	+ uint32_t uiThrdRet = 0;
	864	+
	865	+ if (NULL == pPrivateData)
	866	+ WELS_THREAD_ROUTINE_RETURN (1);
	867	+
	868	+ pEncPEncCtx = (sWelsEncCtx*)pPrivateData->pWelsPEncCtx;
	869	+ iSliceIdx = pPrivateData->iSliceIndex;
	870	+ iEventIdx = pPrivateData->iThreadIndex;
	871	+
	872	+ do {
	873	+#if defined(ENABLE_TRACE_MT)
	874	+ WelsLog (pEncPEncCtx, WELS_LOG_INFO, "[MT] UpdateMbListThreadProc(), try to wait (pUpdateMbListEvent[%d])!\n",
	875	+ iEventIdx);
	876	+#endif
	877	+ iWaitRet = WelsEventWait (pEncPEncCtx->pSliceThreading->pUpdateMbListEvent[iEventIdx]);
	878	+ if (WELS_THREAD_ERROR_WAIT_OBJECT_0 == iWaitRet) {
	879	+ pCurDq = pEncPEncCtx->pCurDqLayer;
	880	+ UpdateMbListNeighborParallel (pCurDq->pSliceEncCtx, pCurDq->sMbDataP, iSliceIdx);
	881	+ WelsEventSignal (
	882	+ pEncPEncCtx->pSliceThreading->pFinUpdateMbListEvent[iEventIdx]); // mean finished update pMb list for this pSlice
	883	+ } else {
	884	+ WelsLog (pEncPEncCtx, WELS_LOG_WARNING,
	885	+ "[MT] UpdateMbListThreadProc(), waiting pUpdateMbListEvent[%d] failed(%d) and thread%d terminated!\n", iEventIdx,
	886	+ iWaitRet, iEventIdx);
	887	+ uiThrdRet = 1;
	888	+ break;
	889	+ }
	890	+ } while (1);
	891	+
	892	+ WELS_THREAD_ROUTINE_RETURN (uiThrdRet);
	893	+}
	894	+#endif//__GNUC__
	895	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	896	+
	897	+// thread process for coding one pSlice
	898	+WELS_THREAD_ROUTINE_TYPE CodingSliceThreadProc (void* arg) {
	899	+ SSliceThreadPrivateData* pPrivateData = (SSliceThreadPrivateData*)arg;
	900	+ sWelsEncCtx* pEncPEncCtx = NULL;
	901	+ SDqLayer* pCurDq = NULL;
	902	+ SSlice* pSlice = NULL;
	903	+ SWelsSliceBs* pSliceBs = NULL;
	904	+#ifdef WIN32
	905	+ WELS_EVENT pEventsList[3];
	906	+ int32_t iEventCount = 0;
	907	+#endif
	908	+ WELS_THREAD_ERROR_CODE iWaitRet = WELS_THREAD_ERROR_GENERIAL;
	909	+ uint32_t uiThrdRet = 0;
	910	+ int32_t iSliceSize = 0;
	911	+ int32_t iSliceIdx = -1;
	912	+ int32_t iThreadIdx = -1;
	913	+ int32_t iEventIdx = -1;
	914	+ bool_t bNeedPrefix = false;
	915	+ EWelsNalUnitType eNalType = NAL_UNIT_UNSPEC_0;
	916	+ EWelsNalRefIdc eNalRefIdc = NRI_PRI_LOWEST;
	917	+
	918	+ if (NULL == pPrivateData)
	919	+ WELS_THREAD_ROUTINE_RETURN (1);
	920	+
	921	+ WelsSetThreadCancelable();
	922	+
	923	+ pEncPEncCtx = (sWelsEncCtx*)pPrivateData->pWelsPEncCtx;
	924	+
	925	+ iThreadIdx = pPrivateData->iThreadIndex;
	926	+ iEventIdx = iThreadIdx;
	927	+
	928	+#ifdef WIN32
	929	+ pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pReadySliceCodingEvent[iEventIdx];
	930	+ pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pExitEncodeEvent[iEventIdx];
	931	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	932	+ pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pUpdateMbListEvent[iEventIdx];
	933	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	934	+#endif//WIN32
	935	+
	936	+ do {
	937	+#ifdef WIN32
	938	+ iWaitRet = WelsMultipleEventsWaitSingleBlocking (iEventCount,
	939	+ &pEventsList[0],
	940	+ (uint32_t) - 1); // blocking until at least one event is
	941	+#else
	942	+#if defined(ENABLE_TRACE_MT)
	943	+ WelsLog (pEncPEncCtx, WELS_LOG_INFO,
	944	+ "[MT] CodingSliceThreadProc(), try to call WelsEventWait(pReadySliceCodingEvent[%d]= 0x%p), pEncPEncCtx= 0x%p!\n",
	945	+ iEventIdx, (void) (pEncPEncCtx->pReadySliceCodingEvent[iEventIdx]), (void)pEncPEncCtx);
	946	+#endif
	947	+ iWaitRet = WelsEventWait (pEncPEncCtx->pSliceThreading->pReadySliceCodingEvent[iEventIdx]);
	948	+#endif//WIN32
	949	+ if (WELS_THREAD_ERROR_WAIT_OBJECT_0 == iWaitRet) { // start pSlice coding signal waited
	950	+ SLayerBSInfo* pLbi = pPrivateData->pLayerBs;
	951	+ const int32_t kiCurDid = pEncPEncCtx->uiDependencyId;
	952	+ const int32_t kiCurTid = pEncPEncCtx->uiTemporalId;
	953	+ SWelsSvcCodingParam* pCodingParam = pEncPEncCtx->pSvcParam;
	954	+ SDLayerParam* pParamD = &pCodingParam->sDependencyLayers[kiCurDid];
	955	+
	956	+ pCurDq = pEncPEncCtx->pCurDqLayer;
	957	+ eNalType = pEncPEncCtx->eNalType;
	958	+ eNalRefIdc = pEncPEncCtx->eNalPriority;
	959	+ bNeedPrefix = pEncPEncCtx->bNeedPrefixNalFlag;
	960	+
	961	+ if (pParamD->sMso.uiSliceMode != SM_DYN_SLICE) {
	962	+ int64_t iSliceStart = 0;
	963	+ bool_t bDsaFlag = false;
	964	+ iSliceIdx = pPrivateData->iSliceIndex;
	965	+ pSlice = &pCurDq->sLayerInfo.pSliceInLayer[iSliceIdx];
	966	+ pSliceBs = &pEncPEncCtx->pSliceBs[iSliceIdx];
	967	+
	968	+#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
	969	+ bDsaFlag = (pParamD->sMso.uiSliceMode == SM_FIXEDSLCNUM_SLICE &&
	970	+ pCodingParam->iMultipleThreadIdc > 1 &&
	971	+ pCodingParam->iMultipleThreadIdc >= pParamD->sMso.sSliceArgument.iSliceNum);
	972	+ if (bDsaFlag)
	973	+ iSliceStart = WelsTime();
	974	+#endif//DYNAMIC_SLICE_ASSIGN \|\| MT_DEBUG
	975	+
	976	+#if !defined(PACKING_ONE_SLICE_PER_LAYER)
	977	+ pSliceBs->uiBsPos = 0;
	978	+#endif//!PACKING_ONE_SLICE_PER_LAYER
	979	+ pSliceBs->iNalIndex = 0;
	980	+ assert ((void) (&pSliceBs->sBsWrite) == (void)pSlice->pSliceBsa);
	981	+ InitBits (&pSliceBs->sBsWrite, pSliceBs->pBsBuffer, pSliceBs->uiSize);
	982	+
	983	+#if MT_DEBUG_BS_WR
	984	+ pSliceBs->bSliceCodedFlag = FALSE;
	985	+#endif//MT_DEBUG_BS_WR
	986	+
	987	+ if (bNeedPrefix) {
	988	+ if (eNalRefIdc != NRI_PRI_LOWEST) {
	989	+ WelsLoadNalForSlice (pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc);
	990	+ WelsWriteSVCPrefixNal (&pSliceBs->sBsWrite, eNalRefIdc, (NAL_UNIT_CODED_SLICE_IDR == eNalType));
	991	+ WelsUnloadNalForSlice (pSliceBs);
	992	+ } else { // No Prefix NAL Unit RBSP syntax here, but need add NAL Unit Header extension
	993	+ WelsLoadNalForSlice (pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc);
	994	+ // No need write any syntax of prefix NAL Unit RBSP here
	995	+ WelsUnloadNalForSlice (pSliceBs);
	996	+ }
	997	+ }
	998	+
	999	+ WelsLoadNalForSlice (pSliceBs, eNalType, eNalRefIdc);
	1000	+
	1001	+ WelsCodeOneSlice (pEncPEncCtx, iSliceIdx, eNalType);
	1002	+
	1003	+ WelsUnloadNalForSlice (pSliceBs);
	1004	+
	1005	+#if !defined(PACKING_ONE_SLICE_PER_LAYER)
	1006	+ if (0 == iSliceIdx) {
	1007	+ pLbi->pBsBuf = pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer;
	1008	+ iSliceSize = WriteSliceToFrameBs (pEncPEncCtx, pLbi, pLbi->pBsBuf, iSliceIdx);
	1009	+ pEncPEncCtx->iPosBsBuffer += iSliceSize;
	1010	+ } else
	1011	+ iSliceSize = WriteSliceBs (pEncPEncCtx, pSliceBs->pBs, iSliceIdx);
	1012	+#else// PACKING_ONE_SLICE_PER_LAYER
	1013	+ if (0 == iSliceIdx) {
	1014	+ pLbi->pBsBuf = pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer;
	1015	+ iSliceSize = WriteSliceToFrameBs (pEncPEncCtx, pLbi, pLbi->pBsBuf, iSliceIdx);
	1016	+ pEncPEncCtx->iPosBsBuffer += iSliceSize;
	1017	+ } else {
	1018	+ pLbi->pBsBuf = pSliceBs->bs + pSliceBs->uiBsPos;
	1019	+ iSliceSize = WriteSliceToFrameBs (pEncPEncCtx, pLbi, pLbi->pBsBuf, iSliceIdx);
	1020	+ pSliceBs->uiBsPos += iSliceSize;
	1021	+ }
	1022	+#endif//!PACKING_ONE_SLICE_PER_LAYER
	1023	+
	1024	+ if (pCurDq->bDeblockingParallelFlag && pSlice->sSliceHeaderExt.sSliceHeader.uiDisableDeblockingFilterIdc != 1
	1025	+#if !defined(ENABLE_FRAME_DUMP)
	1026	+ && (eNalRefIdc != NRI_PRI_LOWEST) &&
	1027	+ (pParamD->iHighestTemporalId == 0 \|\| kiCurTid < pParamD->iHighestTemporalId)
	1028	+#endif// !ENABLE_FRAME_DUMP
	1029	+ ) {
	1030	+ DeblockingFilterSliceAvcbase (pCurDq, pEncPEncCtx->pFuncList, iSliceIdx);
	1031	+ }
	1032	+
	1033	+#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
	1034	+ if (bDsaFlag) {
	1035	+ pEncPEncCtx->pSliceThreading->pSliceConsumeTime[pEncPEncCtx->uiDependencyId][iSliceIdx] = (uint32_t) (
	1036	+ WelsTime() - iSliceStart);
	1037	+#if defined(ENABLE_TRACE_MT)
	1038	+ WelsLog (pEncPEncCtx, WELS_LOG_INFO,
	1039	+ "[MT] CodingSliceThreadProc(), coding_idx %d, uiSliceIdx %d, pSliceConsumeTime %d, iSliceSize %d, pFirstMbInSlice %d, count_num_mb_in_slice %d\n",
	1040	+ pEncPEncCtx->iCodingIndex, iSliceIdx,
	1041	+ pEncPEncCtx->pSliceThreading->pSliceConsumeTime[pEncPEncCtx->uiDependencyId][iSliceIdx], iSliceSize,
	1042	+ pCurDq->pSliceEncCtx->pFirstMbInSlice[iSliceIdx], pCurDq->pSliceEncCtx->pCountMbNumInSlice[iSliceIdx]);
	1043	+#endif//ENABLE_TRACE_MT
	1044	+ }
	1045	+#endif//DYNAMIC_SLICE_ASSIGN \|\| MT_DEBUG
	1046	+
	1047	+#if defined(SLICE_INFO_OUTPUT)
	1048	+ fprintf (stderr,
	1049	+ "@pSlice=%-6d sliceType:%c idc:%d size:%-6d\n",
	1050	+ iSliceIdx,
	1051	+ (pEncPEncCtx->eSliceType == P_SLICE ? 'P' : 'I'),
	1052	+ eNalRefIdc,
	1053	+ iSliceSize
	1054	+ );
	1055	+#endif//SLICE_INFO_OUTPUT
	1056	+
	1057	+#if MT_DEBUG_BS_WR
	1058	+ pSliceBs->bSliceCodedFlag = TRUE;
	1059	+#endif//MT_DEBUG_BS_WR
	1060	+
	1061	+#ifdef WIN32
	1062	+ WelsEventSignal (
	1063	+ &pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx]); // mean finished coding current pSlice
	1064	+#else
	1065	+ WelsEventSignal (pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx]); // mean finished coding current pSlice
	1066	+#endif//WIN32
	1067	+ } else { // for SM_DYN_SLICE parallelization
	1068	+#ifdef PACKING_ONE_SLICE_PER_LAYER
	1069	+ SLayerBSInfo* pLbiPacking = NULL;
	1070	+#endif//PACKING_ONE_SLICE_PER_LAYER
	1071	+ SSliceCtx* pSliceCtx = pCurDq->pSliceEncCtx;
	1072	+ const int32_t kiPartitionId = iThreadIdx;
	1073	+ const int32_t kiSliceIdxStep = pEncPEncCtx->iActiveThreadsNum;
	1074	+ const int32_t kiFirstMbInPartition = pPrivateData->iStartMbIndex; // inclusive
	1075	+ const int32_t kiEndMbInPartition = pPrivateData->iEndMbIndex; // exclusive
	1076	+ int32_t iAnyMbLeftInPartition = kiEndMbInPartition - kiFirstMbInPartition;
	1077	+
	1078	+ iSliceIdx = pPrivateData->iSliceIndex;
	1079	+
	1080	+ pSliceCtx->pFirstMbInSlice[iSliceIdx] = kiFirstMbInPartition;
	1081	+ pCurDq->pNumSliceCodedOfPartition[kiPartitionId] = 1; // one pSlice per partition intialized, dynamic slicing inside
	1082	+ pCurDq->pLastMbIdxOfPartition[kiPartitionId] = kiEndMbInPartition - 1;
	1083	+
	1084	+ pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId] = 0;
	1085	+
	1086	+ while (iAnyMbLeftInPartition > 0) {
	1087	+ if (iSliceIdx >= pSliceCtx->iMaxSliceNumConstraint) {
	1088	+ // TODO: need exception handler for not large enough of MAX_SLICES_NUM related memory usage
	1089	+ // No idea about its solution due MAX_SLICES_NUM is fixed lenght in relevent pData structure
	1090	+ uiThrdRet = 1;
	1091	+ break;
	1092	+ }
	1093	+
	1094	+ pSlice = &pCurDq->sLayerInfo.pSliceInLayer[iSliceIdx];
	1095	+ pSliceBs = &pEncPEncCtx->pSliceBs[iSliceIdx];
	1096	+
	1097	+#if !defined(PACKING_ONE_SLICE_PER_LAYER)
	1098	+ pSliceBs->uiBsPos = 0;
	1099	+#endif//!PACKING_ONE_SLICE_PER_LAYER
	1100	+ pSliceBs->iNalIndex = 0;
	1101	+ InitBits (&pSliceBs->sBsWrite, pSliceBs->pBsBuffer, pSliceBs->uiSize);
	1102	+
	1103	+ if (bNeedPrefix) {
	1104	+ if (eNalRefIdc != NRI_PRI_LOWEST) {
	1105	+ WelsLoadNalForSlice (pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc);
	1106	+ WelsWriteSVCPrefixNal (&pSliceBs->sBsWrite, eNalRefIdc, (NAL_UNIT_CODED_SLICE_IDR == eNalType));
	1107	+ WelsUnloadNalForSlice (pSliceBs);
	1108	+ } else { // No Prefix NAL Unit RBSP syntax here, but need add NAL Unit Header extension
	1109	+ WelsLoadNalForSlice (pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc);
	1110	+ // No need write any syntax of prefix NAL Unit RBSP here
	1111	+ WelsUnloadNalForSlice (pSliceBs);
	1112	+ }
	1113	+ }
	1114	+
	1115	+ WelsLoadNalForSlice (pSliceBs, eNalType, eNalRefIdc);
	1116	+
	1117	+ WelsCodeOneSlice (pEncPEncCtx, iSliceIdx, eNalType);
	1118	+
	1119	+ WelsUnloadNalForSlice (pSliceBs);
	1120	+
	1121	+#if !defined(PACKING_ONE_SLICE_PER_LAYER)
	1122	+ if (0 == kiPartitionId) {
	1123	+ if (0 == iSliceIdx)
	1124	+ pLbi->pBsBuf = pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer;
	1125	+ iSliceSize = WriteSliceToFrameBs (pEncPEncCtx, pLbi, pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer, iSliceIdx);
	1126	+ pEncPEncCtx->iPosBsBuffer += iSliceSize;
	1127	+ } else
	1128	+ iSliceSize = WriteSliceBs (pEncPEncCtx, pSliceBs->pBs, iSliceIdx);
	1129	+#else// PACKING_ONE_SLICE_PER_LAYER
	1130	+ pLbiPacking = pLbi + (iSliceIdx - kiPartitionId);
	1131	+
	1132	+ if (0 == kiPartitionId) {
	1133	+ pLbiPacking->pBsBuf = pEncPEncCtx->pFrameBs + pEncPEncCtx->iPosBsBuffer;
	1134	+ iSliceSize = WriteSliceToFrameBs (pEncPEncCtx, pLbiPacking, pLbiPacking->pBsBuf, iSliceIdx);
	1135	+ pEncPEncCtx->iPosBsBuffer += iSliceSize;
	1136	+ } else {
	1137	+ pLbiPacking->pBsBuf = pSliceBs->bs + pSliceBs->uiBsPos;
	1138	+ iSliceSize = WriteSliceToFrameBs (pEncPEncCtx, pLbiPacking, pLbiPacking->pBsBuf, iSliceIdx);
	1139	+ pSliceBs->uiBsPos += iSliceSize;
	1140	+ }
	1141	+ pEncPEncCtx->pSliceThreading->pCountBsSizeInPartition[kiPartitionId] += iSliceSize;
	1142	+#endif//!PACKING_ONE_SLICE_PER_LAYER
	1143	+
	1144	+ if (pCurDq->bDeblockingParallelFlag && pSlice->sSliceHeaderExt.sSliceHeader.uiDisableDeblockingFilterIdc != 1
	1145	+#if !defined(ENABLE_FRAME_DUMP)
	1146	+ && (eNalRefIdc != NRI_PRI_LOWEST) &&
	1147	+ (pParamD->iHighestTemporalId == 0 \|\| kiCurTid < pParamD->iHighestTemporalId)
	1148	+#endif// !ENABLE_FRAME_DUMP
	1149	+ ) {
	1150	+ DeblockingFilterSliceAvcbase (pCurDq, pEncPEncCtx->pFuncList, iSliceIdx);
	1151	+ }
	1152	+
	1153	+#if defined(SLICE_INFO_OUTPUT)
	1154	+ fprintf (stderr,
	1155	+ "@pSlice=%-6d sliceType:%c idc:%d size:%-6d\n",
	1156	+ iSliceIdx,
	1157	+ (pEncPEncCtx->eSliceType == P_SLICE ? 'P' : 'I'),
	1158	+ eNalRefIdc,
	1159	+ iSliceSize
	1160	+ );
	1161	+#endif//SLICE_INFO_OUTPUT
	1162	+
	1163	+#if defined(ENABLE_TRACE_MT)
	1164	+ WelsLog (pEncPEncCtx, WELS_LOG_INFO,
	1165	+ "[MT] CodingSliceThreadProc(), coding_idx %d, iPartitionId %d, uiSliceIdx %d, iSliceSize %d, count_mb_slice %d, iEndMbInPartition %d, pCurDq->pLastCodedMbIdxOfPartition[%d] %d\n",
	1166	+ pEncPEncCtx->iCodingIndex, kiPartitionId, iSliceIdx, iSliceSize, pCurDq->pSliceEncCtx->pCountMbNumInSlice[iSliceIdx],
	1167	+ kiEndMbInPartition, kiPartitionId, pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId]);
	1168	+#endif//ENABLE_TRACE_MT
	1169	+
	1170	+ iAnyMbLeftInPartition = kiEndMbInPartition - (1 + pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId]);
	1171	+ iSliceIdx += kiSliceIdxStep;
	1172	+ }
	1173	+
	1174	+ if (uiThrdRet) // any exception??
	1175	+ break;
	1176	+
	1177	+#ifdef WIN32
	1178	+ WelsEventSignal (&pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx]); // mean finished coding current pSlice
	1179	+#else
	1180	+ WelsEventSignal (pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx]); // mean finished coding current pSlice
	1181	+#endif//WIN32
	1182	+ }
	1183	+ }
	1184	+#ifdef WIN32
	1185	+ else if (WELS_THREAD_ERROR_WAIT_OBJECT_0 + 1 == iWaitRet) { // exit thread signal
	1186	+ uiThrdRet = 0;
	1187	+ break;
	1188	+ }
	1189	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	1190	+ else if (WELS_THREAD_ERROR_WAIT_OBJECT_0 + 2 == iWaitRet) { // update pMb list singal
	1191	+ iSliceIdx =
	1192	+ iEventIdx; // pPrivateData->iSliceIndex; old threads can not be terminated, pPrivateData is not correct for applicable
	1193	+ pCurDq = pEncPEncCtx->pCurDqLayer;
	1194	+ UpdateMbListNeighborParallel (pCurDq->pSliceEncCtx, pCurDq->sMbDataP, iSliceIdx);
	1195	+ WelsEventSignal (
	1196	+ &pEncPEncCtx->pSliceThreading->pFinUpdateMbListEvent[iEventIdx]); // mean finished update pMb list for this pSlice
	1197	+ }
	1198	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	1199	+#endif//WIN32
	1200	+ else { // WELS_THREAD_ERROR_WAIT_TIMEOUT, or WELS_THREAD_ERROR_WAIT_FAILED
	1201	+ WelsLog (pEncPEncCtx, WELS_LOG_WARNING,
	1202	+ "[MT] CodingSliceThreadProc(), waiting pReadySliceCodingEvent[%d] failed(%d) and thread%d terminated!\n", iEventIdx,
	1203	+ iWaitRet, iThreadIdx);
	1204	+ uiThrdRet = 1;
	1205	+ break;
	1206	+ }
	1207	+ } while (1);
	1208	+
	1209	+#ifdef WIN32
	1210	+ WelsEventSignal (&pEncPEncCtx->pSliceThreading->pFinSliceCodingEvent[iEventIdx]); // notify to mother encoding threading
	1211	+#endif//WIN32
	1212	+
	1213	+ WELS_THREAD_ROUTINE_RETURN (uiThrdRet);
	1214	+}
	1215	+
	1216	+int32_t CreateSliceThreads (sWelsEncCtx* pCtx) {
	1217	+ const int32_t kiThreadCount = pCtx->pSvcParam->iCountThreadsNum;
	1218	+ int32_t iIdx = 0;
	1219	+#if defined(WIN32) && defined(BIND_CPU_CORES_TO_THREADS)
	1220	+ DWORD dwProcessAffinity;
	1221	+ DWORD dwSystemAffinity;
	1222	+ GetProcessAffinityMask (GetCurrentProcess(), &dwProcessAffinity, &dwSystemAffinity);
	1223	+#endif//WIN32 && BIND_CPU_CORES_TO_THREADS
	1224	+
	1225	+ while (iIdx < kiThreadCount) {
	1226	+ WelsThreadCreate (&pCtx->pSliceThreading->pThreadHandles[iIdx], CodingSliceThreadProc,
	1227	+ &pCtx->pSliceThreading->pThreadPEncCtx[iIdx], 0);
	1228	+#if defined(WIN32) && defined(BIND_CPU_CORES_TO_THREADS)
	1229	+ if (dwProcessAffinity > 1
	1230	+ && pCtx->pSliceThreading->pThreadHandles[iIdx] != NULL) { // multiple cores and thread created successfully
	1231	+ DWORD dw = 0;
	1232	+ DWORD dwAffinityMask = 1 << iIdx;
	1233	+ if (dwAffinityMask & dwProcessAffinity) { // check if cpu is available
	1234	+ dw = SetThreadAffinityMask (pCtx->pSliceThreading->pThreadHandles[iIdx], dwAffinityMask); //1 << iIdx
	1235	+ if (dw == 0) {
	1236	+ str_t str[64] = {0};
	1237	+ SNPRINTF (str, 64, "SetThreadAffinityMask iIdx:%d", iIdx);
	1238	+ }
	1239	+ }
	1240	+ }
	1241	+#endif//WIN32 && BIND_CPU_CORES_TO_THREADS
	1242	+ // We need extra threads for update_mb_list_proc on __GNUC__ like OS (mac/linux)
	1243	+ // due to WelsMultipleEventsWaitSingleBlocking implememtation can not work well
	1244	+ // in case waiting pUpdateMbListEvent and pReadySliceCodingEvent events at the same time
	1245	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	1246	+#if defined(__GNUC__)
	1247	+ WelsThreadCreate (&pCtx->pSliceThreading->pUpdateMbListThrdHandles[iIdx], UpdateMbListThreadProc,
	1248	+ &pCtx->pSliceThreading->pThreadPEncCtx[iIdx], 0);
	1249	+#endif//__GNUC__
	1250	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	1251	+ ++ iIdx;
	1252	+ }
	1253	+#if defined(ENABLE_TRACE_MT)
	1254	+ WelsLog (pCtx, WELS_LOG_INFO, "CreateSliceThreads() exit..\n");
	1255	+#endif
	1256	+ return 0;
	1257	+}
	1258	+
	1259	+#ifdef PACKING_ONE_SLICE_PER_LAYER
	1260	+void ResetCountBsSizeInPartitions (uint32_t* pCountBsSizeList, const int32_t iPartitionCnt) {
	1261	+ if (pCountBsSizeList != NULL && iPartitionCnt > 0) {
	1262	+ memset (pCountBsSizeList, 0, sizeof (pCountBsSizeList[0]) * iPartitionCnt);
	1263	+ }
	1264	+}
	1265	+#endif//PACKING_ONE_SLICE_PER_LAYER
	1266	+
	1267	+#ifdef WIN32
	1268	+int32_t FiredSliceThreads (SSliceThreadPrivateData* pPriData, WELS_EVENT* pEventsList, SLayerBSInfo* pLbi,
	1269	+ const uint32_t uiNumThreads, SSliceCtx* pSliceCtx, const BOOL_T bIsDynamicSlicingMode)
	1270	+#else
	1271	+int32_t FiredSliceThreads (SSliceThreadPrivateData* pPriData, WELS_EVENT** pEventsList, SLayerBSInfo* pLbi,
	1272	+ const uint32_t uiNumThreads, SSliceCtx* pSliceCtx, const BOOL_T bIsDynamicSlicingMode)
	1273	+#endif//WIN32
	1274	+{
	1275	+ int32_t iEndMbIdx = 0;
	1276	+ int32_t iIdx = 0;
	1277	+ const int32_t kiEventCnt = uiNumThreads;
	1278	+
	1279	+ if (pPriData == NULL \|\| pLbi == NULL \|\| kiEventCnt <= 0 \|\| pEventsList == NULL) {
	1280	+ WelsLog (NULL, WELS_LOG_ERROR,
	1281	+ "FiredSliceThreads(), fail due pPriData == %p \|\| pLbi == %p \|\| iEventCnt(%d) <= 0 \|\| pEventsList == %p!!\n",
	1282	+ (void)pPriData, (void)pLbi, uiNumThreads, (void*)pEventsList);
	1283	+ return 1;
	1284	+ }
	1285	+
	1286	+#if defined(PACKING_ONE_SLICE_PER_LAYER)
	1287	+ ////////////////////////////////////////
	1288	+ if (bIsDynamicSlicingMode) {
	1289	+ iEndMbIdx = pSliceCtx->iMbNumInFrame;
	1290	+ for (iIdx = kiEventCnt - 1; iIdx >= 0; --iIdx) {
	1291	+ const int32_t kiFirstMbIdx = pSliceCtx->pFirstMbInSlice[iIdx];
	1292	+ pPriData[iIdx].iStartMbIndex = kiFirstMbIdx;
	1293	+ pPriData[iIdx].iEndMbIndex = iEndMbIdx;
	1294	+ iEndMbIdx = kiFirstMbIdx;
	1295	+ }
	1296	+ }
	1297	+
	1298	+ iIdx = 0;
	1299	+ while (iIdx < kiEventCnt) {
	1300	+ pPriData[iIdx].pLayerBs = pLbi;
	1301	+ pPriData[iIdx].iSliceIndex = iIdx;
	1302	+#ifdef WIN32
	1303	+ if (pEventsList[iIdx])
	1304	+ WelsEventSignal (&pEventsList[iIdx]);
	1305	+#else
	1306	+ WelsEventSignal (pEventsList[iIdx]);
	1307	+#endif//WIN32
	1308	+ ++ pLbi;
	1309	+ ++ iIdx;
	1310	+ }
	1311	+ ////////////////////////////////////////
	1312	+#else
	1313	+ ////////////////////////////////////////
	1314	+ if (bIsDynamicSlicingMode) {
	1315	+ iEndMbIdx = pSliceCtx->iMbNumInFrame;
	1316	+ for (iIdx = kiEventCnt - 1; iIdx >= 0; --iIdx) {
	1317	+ const int32_t iFirstMbIdx = pSliceCtx->pFirstMbInSlice[iIdx];
	1318	+ pPriData[iIdx].iStartMbIndex = iFirstMbIdx;
	1319	+ pPriData[iIdx].iEndMbIndex = iEndMbIdx;
	1320	+ iEndMbIdx = iFirstMbIdx;
	1321	+ }
	1322	+ }
	1323	+
	1324	+ iIdx = 0;
	1325	+ while (iIdx < kiEventCnt) {
	1326	+ pPriData[iIdx].pLayerBs = pLbi;
	1327	+ pPriData[iIdx].iSliceIndex = iIdx;
	1328	+#ifdef WIN32
	1329	+ if (pEventsList[iIdx])
	1330	+ WelsEventSignal (&pEventsList[iIdx]);
	1331	+#else
	1332	+ WelsEventSignal (pEventsList[iIdx]);
	1333	+#endif//WIN32
	1334	+ ++ iIdx;
	1335	+ }
	1336	+ ////////////////////////////////////////
	1337	+#endif//PACKING_ONE_SLICE_PER_LAYER
	1338	+
	1339	+ return 0;
	1340	+}
	1341	+
	1342	+int32_t DynamicDetectCpuCores() {
	1343	+ WelsLogicalProcessInfo info;
	1344	+ WelsQueryLogicalProcessInfo (&info);
	1345	+ return info.ProcessorCount;
	1346	+}
	1347	+
	1348	+#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN)
	1349	+
	1350	+int32_t AdjustBaseLayer (sWelsEncCtx* pCtx) {
	1351	+ SDqLayer* pCurDq = pCtx->ppDqLayerList[0];
	1352	+ int32_t iNeedAdj = 1;
	1353	+#ifdef MT_DEBUG
	1354	+ int64_t iT0 = WelsTime();
	1355	+#endif//MT_DEBUG
	1356	+#ifdef TRY_SLICING_BALANCE
	1357	+
	1358	+ pCtx->pCurDqLayer = pCurDq;
	1359	+
	1360	+#ifdef NOT_ABSOLUTE_BALANCING
	1361	+ // do not need adjust due to not different at both slices of consumed time
	1362	+ iNeedAdj = NeedDynamicAdjust (pCtx->pSliceThreading->pSliceConsumeTime[0], pCurDq->pSliceEncCtx->iSliceNumInFrame);
	1363	+ if (iNeedAdj)
	1364	+#endif//NOT_ABSOLUTE_BALANCING
	1365	+ DynamicAdjustSlicing (pCtx,
	1366	+ pCurDq,
	1367	+ pCtx->pSliceThreading->pSliceComplexRatio[0],
	1368	+ 0);
	1369	+#endif//TRY_SLICING_BALANCE
	1370	+#ifdef MT_DEBUG
	1371	+ iT0 = WelsTime() - iT0;
	1372	+ if (pCtx->pSliceThreading->pFSliceDiff) {
	1373	+ fprintf (pCtx->pSliceThreading->pFSliceDiff,
	1374	+#ifdef WIN32
	1375	+ "%6I64d us adjust time at base spatial layer, iNeedAdj %d, DynamicAdjustSlicing()\n",
	1376	+#else
	1377	+ "%6lld us adjust time at base spatial layer, iNeedAdj %d, DynamicAdjustSlicing()\n",
	1378	+#endif//WIN32
	1379	+ iT0, iNeedAdj);
	1380	+ }
	1381	+#endif//MT_DEBUG
	1382	+
	1383	+ return iNeedAdj;
	1384	+}
	1385	+
	1386	+int32_t AdjustEnhanceLayer (sWelsEncCtx* pCtx, int32_t iCurDid) {
	1387	+#ifdef MT_DEBUG
	1388	+ int64_t iT1 = WelsTime();
	1389	+#endif//MT_DEBUG
	1390	+ int32_t iNeedAdj = 1;
	1391	+ // uiSliceMode of referencing spatial should be SM_FIXEDSLCNUM_SLICE
	1392	+ // if using spatial base layer for complexity estimation
	1393	+ const BOOL_T kbModelingFromSpatial = (pCtx->pCurDqLayer->pRefLayer != NULL && iCurDid > 0)
	1394	+ && (pCtx->pSvcParam->sDependencyLayers[iCurDid - 1].sMso.uiSliceMode == SM_FIXEDSLCNUM_SLICE
	1395	+ && pCtx->pSvcParam->iMultipleThreadIdc >= pCtx->pSvcParam->sDependencyLayers[iCurDid -
	1396	+ 1].sMso.sSliceArgument.iSliceNum);
	1397	+
	1398	+ if (kbModelingFromSpatial) { // using spatial base layer for complexity estimation
	1399	+#ifdef TRY_SLICING_BALANCE
	1400	+#ifdef NOT_ABSOLUTE_BALANCING
	1401	+ // do not need adjust due to not different at both slices of consumed time
	1402	+ iNeedAdj = NeedDynamicAdjust (pCtx->pSliceThreading->pSliceConsumeTime[iCurDid - 1],
	1403	+ pCtx->pCurDqLayer->pSliceEncCtx->iSliceNumInFrame);
	1404	+ if (iNeedAdj)
	1405	+#endif//NOT_ABSOLUTE_BALANCING
	1406	+ DynamicAdjustSlicing (pCtx,
	1407	+ pCtx->pCurDqLayer,
	1408	+ pCtx->pSliceThreading->pSliceComplexRatio[iCurDid - 1],
	1409	+ iCurDid
	1410	+ );
	1411	+#endif//TRY_SLICING_BALANCE
	1412	+ } else { // use temporal layer for complexity estimation
	1413	+#ifdef TRY_SLICING_BALANCE
	1414	+#ifdef NOT_ABSOLUTE_BALANCING
	1415	+ // do not need adjust due to not different at both slices of consumed time
	1416	+ iNeedAdj = NeedDynamicAdjust (pCtx->pSliceThreading->pSliceConsumeTime[iCurDid],
	1417	+ pCtx->pCurDqLayer->pSliceEncCtx->iSliceNumInFrame);
	1418	+ if (iNeedAdj)
	1419	+#endif//NOT_ABSOLUTE_BALANCING
	1420	+ DynamicAdjustSlicing (pCtx,
	1421	+ pCtx->pCurDqLayer,
	1422	+ pCtx->pSliceThreading->pSliceComplexRatio[iCurDid],
	1423	+ iCurDid
	1424	+ );
	1425	+#endif//TRY_SLICING_BALANCE
	1426	+ }
	1427	+
	1428	+#ifdef MT_DEBUG
	1429	+ iT1 = WelsTime() - iT1;
	1430	+ if (pCtx->pSliceThreading->pFSliceDiff) {
	1431	+ fprintf (pCtx->pSliceThreading->pFSliceDiff,
	1432	+#ifdef WIN32
	1433	+ "%6I64d us adjust time at spatial layer %d, iNeedAdj %d, DynamicAdjustSlicing()\n",
	1434	+#else
	1435	+ "%6lld us adjust time at spatial layer %d, iNeedAdj %d, DynamicAdjustSlicing()\n",
	1436	+#endif//WIN32
	1437	+ iT1, iCurDid, iNeedAdj);
	1438	+ }
	1439	+#endif//MT_DEBUG
	1440	+
	1441	+ return iNeedAdj;
	1442	+}
	1443	+
	1444	+#endif//#if defined(MT_ENABLED) && defined(DYNAMIC_SLICE_ASSIGN)
	1445	+
	1446	+#if defined(MT_ENABLED)
	1447	+
	1448	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE) && defined(MT_DEBUG)
	1449	+void TrackSliceComplexities (sWelsEncCtx* pCtx, const int32_t iCurDid) {
	1450	+ const int32_t kiCountSliceNum = pCtx->pCurDqLayer->pSliceEncCtx->iSliceNumInFrame;
	1451	+ if (kiCountSliceNum > 0) {
	1452	+ int32_t iSliceIdx = 0;
	1453	+ do {
	1454	+ fprintf (pCtx->pSliceThreading->pFSliceDiff, "%6.3f complexity pRatio at iDid %d pSlice %d\n",
	1455	+ pCtx->pSliceThreading->pSliceComplexRatio[iCurDid][iSliceIdx], iCurDid, iSliceIdx);
	1456	+ ++ iSliceIdx;
	1457	+ } while (iSliceIdx < kiCountSliceNum);
	1458	+ }
	1459	+}
	1460	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) && defined(TRY_SLICING_BALANCE)
	1461	+
	1462	+#if defined(DYNAMIC_SLICE_ASSIGN) && defined(MT_DEBUG)
	1463	+void TrackSliceConsumeTime (sWelsEncCtx* pCtx, int32_t* pDidList, const int32_t iSpatialNum) {
	1464	+ SWelsSvcCodingParam* pPara = NULL;
	1465	+ int32_t iSpatialIdx = 0;
	1466	+
	1467	+ if (iSpatialNum > MAX_DEPENDENCY_LAYER)
	1468	+ return;
	1469	+
	1470	+ pPara = pCtx->pSvcParam;
	1471	+ while (iSpatialIdx < iSpatialNum) {
	1472	+ const int32_t kiDid = pDidList[iSpatialIdx];
	1473	+ SDLayerParam* pDlp = &pPara->sDependencyLayers[kiDid];
	1474	+ SMulSliceOption* pMso = &pDlp->sMso;
	1475	+ SDqLayer* pCurDq = pCtx->ppDqLayerList[kiDid];
	1476	+ SSliceCtx* pSliceCtx = pCurDq->pSliceEncCtx;
	1477	+ const uint32_t kuiCountSliceNum = pSliceCtx->iSliceNumInFrame;
	1478	+ if (pCtx->pSliceThreading) {
	1479	+ if (pCtx->pSliceThreading->pFSliceDiff && pMso->uiSliceMode == SM_FIXEDSLCNUM_SLICE && pPara->iMultipleThreadIdc > 1
	1480	+ && pPara->iMultipleThreadIdc >= kuiCountSliceNum) {
	1481	+ uint32_t i = 0;
	1482	+ uint32_t uiMaxT = 0;
	1483	+ int32_t iMaxI = 0;
	1484	+ while (i < kuiCountSliceNum) {
	1485	+ if (pCtx->pSliceThreading->pSliceConsumeTime[kiDid] != NULL)
	1486	+ fprintf (pCtx->pSliceThreading->pFSliceDiff, "%6d us consume_time coding_idx %d iDid %d pSlice %d\n",
	1487	+ pCtx->pSliceThreading->pSliceConsumeTime[kiDid][i], pCtx->iCodingIndex, kiDid, i // 1000/);
	1488	+ if (pCtx->pSliceThreading->pSliceConsumeTime[kiDid][i] > uiMaxT) {
	1489	+ uiMaxT = pCtx->pSliceThreading->pSliceConsumeTime[kiDid][i];
	1490	+ iMaxI = i;
	1491	+ }
	1492	+ ++ i;
	1493	+ }
	1494	+ fprintf (pCtx->pSliceThreading->pFSliceDiff, "%6d us consume_time_max coding_idx %d iDid %d pSlice %d\n", uiMaxT,
	1495	+ pCtx->iCodingIndex, kiDid, iMaxI // 1000/);
	1496	+ }
	1497	+ }
	1498	+ ++ iSpatialIdx;
	1499	+ }
	1500	+}
	1501	+#endif//#if defined(DYNAMIC_SLICE_ASSIGN) \|\| defined(MT_DEBUG)
	1502	+
	1503	+#endif//MT_ENABLED
	1504	+}
	1505	+#endif//MT_ENABLED
	1506	+

Pretty printed the C++ code with astyle

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code