ref: cc6b409f1268e82c9d0be946d444131d9422471d
dir: /codec/encoder/core/src/slice_multi_threading.cpp/
/*!
* \copy
* Copyright (c) 2010-2013, Cisco Systems
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
*
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
* COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
*
* \file slice_multi_threading.h
*
* \brief pSlice based multiple threading
*
* \date 04/16/2010 Created
*
*************************************************************************************
*/
#include <assert.h>
#if !defined(_WIN32)
#include <semaphore.h>
#include <unistd.h>
#endif//!_WIN32
#ifndef SEM_NAME_MAX
// length of semaphore name should be system constrained at least on mac 10.7
#define SEM_NAME_MAX 32
#endif//SEM_NAME_MAX
#include "slice_multi_threading.h"
#include "mt_defs.h"
#include "nal_encap.h"
#include "utils.h"
#include "encoder.h"
#include "svc_encode_slice.h"
#include "deblocking.h"
#include "svc_enc_golomb.h"
#include "crt_util_safe_x.h" // for safe crt like calls
#include "rc.h"
#include "cpu.h"
#include "measure_time.h"
#include "wels_task_management.h"
#if defined(ENABLE_TRACE_MT)
#define MT_TRACE_LOG(pLog, x, ...) WelsLog(pLog, x, __VA_ARGS__)
#else
#define MT_TRACE_LOG(x, ...)
#endif
#define WELS_THREAD_SIGNAL_AND_BREAK(CodedEventList, CodedMasterEvent, iEventIdx) { \
WelsEventSignal(&CodedEventList[iEventIdx]); \
WelsEventSignal (&CodedMasterEvent); \
break; \
}
namespace WelsEnc {
void UpdateMbListNeighborParallel (SDqLayer* pCurDq,
SMB* pMbList,
const int32_t uiSliceIdc) {
SSliceCtx* pSliceCtx = &pCurDq->sSliceEncCtx;
SSlice* pUpdateSlice = &pCurDq->sLayerInfo.pSliceInLayer[uiSliceIdc];
const uint16_t* kpMbMap = pSliceCtx->pOverallMbMap;
const int32_t kiMbWidth = pSliceCtx->iMbWidth;
int32_t iIdx = pUpdateSlice->sSliceHeaderExt.sSliceHeader.iFirstMbInSlice;
const int32_t kiEndMbInSlice = iIdx + pSliceCtx->pCountMbNumInSlice[uiSliceIdc] - 1;
do {
SMB* pMb = &pMbList[iIdx];
uint32_t uiNeighborAvailFlag = 0;
const int32_t kiMbXY = pMb->iMbXY;
const int32_t kiMbX = pMb->iMbX;
const int32_t kiMbY = pMb->iMbY;
bool bLeft;
bool bTop;
bool bLeftTop;
bool bRightTop;
int32_t iLeftXY, iTopXY, iLeftTopXY, iRightTopXY;
iLeftXY = kiMbXY - 1;
iTopXY = kiMbXY - kiMbWidth;
iLeftTopXY = iTopXY - 1;
iRightTopXY = iTopXY + 1;
bLeft = (kiMbX > 0) && (uiSliceIdc == kpMbMap[iLeftXY]);
bTop = (kiMbY > 0) && (uiSliceIdc == kpMbMap[iTopXY]);
bLeftTop = (kiMbX > 0) && (kiMbY > 0) && (uiSliceIdc == kpMbMap[iLeftTopXY]);
bRightTop = (kiMbX < (kiMbWidth - 1)) && (kiMbY > 0) && (uiSliceIdc == kpMbMap[iRightTopXY]);
if (bLeft) {
uiNeighborAvailFlag |= LEFT_MB_POS;
}
if (bTop) {
uiNeighborAvailFlag |= TOP_MB_POS;
}
if (bLeftTop) {
uiNeighborAvailFlag |= TOPLEFT_MB_POS;
}
if (bRightTop) {
uiNeighborAvailFlag |= TOPRIGHT_MB_POS;
}
pMb->uiNeighborAvail = (uint8_t)uiNeighborAvailFlag;
pMb->uiSliceIdc = uiSliceIdc;
++ iIdx;
} while (iIdx <= kiEndMbInSlice);
}
void CalcSliceComplexRatio (SDqLayer* pCurDq) {
SSliceCtx* pSliceCtx = &pCurDq->sSliceEncCtx;
SSlice* pSliceInLayer = pCurDq->sLayerInfo.pSliceInLayer;
int32_t iSumAv = 0;
int32_t* pCountMbInSlice = (int32_t*)pSliceCtx->pCountMbNumInSlice;
const int32_t kiSliceCount = pSliceCtx->iSliceNumInFrame;
int32_t iSliceIdx = 0;
int32_t iAvI[MAX_SLICES_NUM];
WelsEmms();
while (iSliceIdx < kiSliceCount) {
iAvI[iSliceIdx] = WELS_DIV_ROUND (INT_MULTIPLY * pCountMbInSlice[iSliceIdx], pSliceInLayer[iSliceIdx].uiSliceConsumeTime);
MT_TRACE_LOG (NULL, WELS_LOG_DEBUG, "[MT] CalcSliceComplexRatio(), uiSliceConsumeTime[%d]= %d us, slice_run= %d",
iSliceIdx,
pSliceInLayer[iSliceIdx].uiSliceConsumeTime, pCountMbInSlice[iSliceIdx]);
iSumAv += iAvI[iSliceIdx];
++ iSliceIdx;
}
while (-- iSliceIdx >= 0) {
pSliceInLayer[iSliceIdx].iSliceComplexRatio = WELS_DIV_ROUND (INT_MULTIPLY * iAvI[iSliceIdx], iSumAv);
}
}
int32_t NeedDynamicAdjust (SSlice* pSliceInLayer, const int32_t iSliceNum) {
if ( NULL == pSliceInLayer )
return false;
uint32_t uiTotalConsume = 0;
int32_t iSliceIdx = 0;
int32_t iNeedAdj = false;
WelsEmms();
while (iSliceIdx < iSliceNum) {
uiTotalConsume += pSliceInLayer[iSliceIdx].uiSliceConsumeTime;
iSliceIdx ++;
}
if (uiTotalConsume == 0) {
MT_TRACE_LOG (NULL, WELS_LOG_DEBUG,
"[MT] NeedDynamicAdjust(), herein do no adjust due first picture, iCountSliceNum= %d",
iSliceNum);
return false;
}
iSliceIdx = 0;
float fThr = EPSN; // threshold for various cores cases
float fRmse = .0f; // root mean square error of pSlice consume ratios
const float kfMeanRatio = 1.0f / iSliceNum;
do {
const float fRatio = 1.0f * pSliceInLayer[iSliceIdx].uiSliceConsumeTime / uiTotalConsume;
const float fDiffRatio = fRatio - kfMeanRatio;
fRmse += (fDiffRatio * fDiffRatio);
++ iSliceIdx;
} while (iSliceIdx + 1 < iSliceNum);
fRmse = sqrtf (fRmse / iSliceNum);
if (iSliceNum >= 8) {
fThr += THRESHOLD_RMSE_CORE8;
} else if (iSliceNum >= 4) {
fThr += THRESHOLD_RMSE_CORE4;
} else if (iSliceNum >= 2) {
fThr += THRESHOLD_RMSE_CORE2;
} else
fThr = 1.0f;
if (fRmse > fThr)
iNeedAdj = true;
MT_TRACE_LOG (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",
iNeedAdj, fRmse, fThr, iSliceNum);
return iNeedAdj;
}
void DynamicAdjustSlicing (sWelsEncCtx* pCtx,
SDqLayer* pCurDqLayer,
int32_t iCurDid) {
SSliceCtx* pSliceCtx = &pCurDqLayer->sSliceEncCtx;
SSlice* pSliceInLayer = pCurDqLayer->sLayerInfo.pSliceInLayer;
const int32_t kiCountSliceNum = pSliceCtx->iSliceNumInFrame;
const int32_t kiCountNumMb = pSliceCtx->iMbNumInFrame;
int32_t iMinimalMbNum =
pSliceCtx->iMbWidth; // in theory we need only 1 SMB, here let it as one SMB row required
int32_t iMaximalMbNum = 0; // dynamically assign later
int32_t iMbNumLeft = kiCountNumMb;
int32_t iRunLen[MAX_THREADS_NUM] = {0};
int32_t iSliceIdx = 0;
int32_t iNumMbInEachGom = 0;
SWelsSvcRc* pWelsSvcRc = &pCtx->pWelsSvcRc[iCurDid];
if (pCtx->pSvcParam->iRCMode != RC_OFF_MODE) {
iNumMbInEachGom = pWelsSvcRc->iNumberMbGom;
if (iNumMbInEachGom <= 0) {
WelsLog (& (pCtx->sLogCtx), WELS_LOG_ERROR,
"[MT] DynamicAdjustSlicing(), invalid iNumMbInEachGom= %d from RC, iDid= %d, iCountNumMb= %d", iNumMbInEachGom,
iCurDid, kiCountNumMb);
return;
}
// do not adjust in case no extra iNumMbInEachGom based left for slicing adjustment,
// extra MB of non integrated GOM assigned at the last pSlice in default, keep up on early initial result.
if (iNumMbInEachGom * kiCountSliceNum >= kiCountNumMb) {
return;
}
iMinimalMbNum = iNumMbInEachGom;
}
if (kiCountSliceNum < 2 || (kiCountSliceNum & 0x01)) // we need suppose uiSliceNum is even for multiple threading
return;
iMaximalMbNum = kiCountNumMb - (kiCountSliceNum - 1) * iMinimalMbNum;
WelsEmms();
MT_TRACE_LOG (pCtx, WELS_LOG_DEBUG, "[MT] DynamicAdjustSlicing(), iDid= %d, iCountNumMb= %d", iCurDid, kiCountNumMb);
iSliceIdx = 0;
while (iSliceIdx + 1 < kiCountSliceNum) {
int32_t iNumMbAssigning = WELS_DIV_ROUND (kiCountNumMb * pSliceInLayer[iSliceIdx].iSliceComplexRatio, INT_MULTIPLY);
// GOM boundary aligned
if (pCtx->pSvcParam->iRCMode != RC_OFF_MODE) {
iNumMbAssigning = iNumMbAssigning / iNumMbInEachGom * iNumMbInEachGom;
}
// make sure one GOM at least in each pSlice for safe
if (iNumMbAssigning < iMinimalMbNum)
iNumMbAssigning = iMinimalMbNum;
else if (iNumMbAssigning > iMaximalMbNum)
iNumMbAssigning = iMaximalMbNum;
assert (iNumMbAssigning > 0);
iMbNumLeft -= iNumMbAssigning;
if (iMbNumLeft <= 0) { // error due to we can not support slice_skip now yet, do not adjust this time
assert (0);
return;
}
iRunLen[iSliceIdx] = iNumMbAssigning;
MT_TRACE_LOG (pCtx, WELS_LOG_DEBUG,
"[MT] DynamicAdjustSlicing(), uiSliceIdx= %d, iSliceComplexRatio= %.2f, slice_run_org= %d, slice_run_adj= %d",
iSliceIdx, pSliceInLayer[iSliceIdx].iSliceComplexRatio * 1.0f / INT_MULTIPLY,
pSliceCtx->pCountMbNumInSlice[iSliceIdx],
iNumMbAssigning);
++ iSliceIdx;
iMaximalMbNum = iMbNumLeft - (kiCountSliceNum - iSliceIdx - 1) * iMinimalMbNum; // get maximal num_mb in left parts
}
iRunLen[iSliceIdx] = iMbNumLeft;
MT_TRACE_LOG (pCtx, WELS_LOG_DEBUG,
"[MT] DynamicAdjustSlicing(), iSliceIdx= %d, iSliceComplexRatio= %.2f, slice_run_org= %d, slice_run_adj= %d",
iSliceIdx, pSliceInLayer[iSliceIdx].iSliceComplexRatio * 1.0f / INT_MULTIPLY,
pSliceCtx->pCountMbNumInSlice[iSliceIdx], iMbNumLeft);
if (DynamicAdjustSlicePEncCtxAll (pCurDqLayer, iRunLen) == 0) {
const int32_t kiThreadNum = pCtx->pSvcParam->iCountThreadsNum;
int32_t iThreadIdx = 0;
do {
WelsEventSignal (&pCtx->pSliceThreading->pUpdateMbListEvent[iThreadIdx]);
WelsEventSignal (&pCtx->pSliceThreading->pThreadMasterEvent[iThreadIdx]);
++ iThreadIdx;
} while (iThreadIdx < kiThreadNum);
WelsMultipleEventsWaitAllBlocking (kiThreadNum, &pCtx->pSliceThreading->pFinUpdateMbListEvent[0]);
}
}
int32_t SetMultiSliceBuffer (sWelsEncCtx** ppCtx, CMemoryAlign* pMa, SSliceThreading* pSmt,
int32_t iMaxSliceNum, int32_t iSlice1Len, int32_t iSlice0Len, bool bDynamicSlice) {
(*ppCtx)->pSliceBs = (SWelsSliceBs*)pMa->WelsMalloc (sizeof (SWelsSliceBs) * iMaxSliceNum, "pSliceBs");
if (NULL == (*ppCtx)->pSliceBs) {
return ENC_RETURN_MEMALLOCERR;
}
if (iSlice0Len <= 0) {
return ENC_RETURN_UNEXPECTED;
}
//slice 0
(*ppCtx)->pSliceBs[0].uiSize = iSlice0Len;
(*ppCtx)->pSliceBs[0].pBs = (*ppCtx)->pFrameBs;
(*ppCtx)->pSliceBs[0].uiBsPos = 0;
(*ppCtx)->pSliceBs[0].pBsBuffer = pSmt->pThreadBsBuffer[0];
if ((iMaxSliceNum == 1) && (!bDynamicSlice)) {
return ENC_RETURN_SUCCESS;
}
//slice >0
if (iSlice1Len <= 0) {
return ENC_RETURN_UNEXPECTED;
}
if ((*ppCtx)->iFrameBsSize < (iSlice0Len + (iMaxSliceNum - 1)*iSlice1Len)) {
return ENC_RETURN_MEMALLOCERR;
}
for (int32_t k = 1; k < iMaxSliceNum; k++) {
(*ppCtx)->pSliceBs[k].uiSize = iSlice1Len;
(*ppCtx)->pSliceBs[k].pBs = (*ppCtx)->pSliceBs[k - 1].pBs + (*ppCtx)->pSliceBs[k - 1].uiSize;
}
return ENC_RETURN_SUCCESS;
}
int32_t RequestMtResource (sWelsEncCtx** ppCtx, SWelsSvcCodingParam* pCodingParam, const int32_t iCountBsLen,
const int32_t iMaxSliceBufferSize, bool bDynamicSlice) {
CMemoryAlign* pMa = NULL;
SWelsSvcCodingParam* pPara = NULL;
SSliceThreading* pSmt = NULL;
int32_t iNumSpatialLayers = 0;
int32_t iThreadNum = 0;
int32_t iIdx = 0;
int16_t iMaxSliceNum = 1;
int32_t iReturn = ENC_RETURN_SUCCESS;
bool bWillUseTaskManage = false;
if (NULL == ppCtx || NULL == pCodingParam || NULL == *ppCtx || iCountBsLen <= 0)
return 1;
pMa = (*ppCtx)->pMemAlign;
pPara = pCodingParam;
iNumSpatialLayers = pPara->iSpatialLayerNum;
iThreadNum = pPara->iCountThreadsNum;
iMaxSliceNum = (*ppCtx)->iMaxSliceCount;
pSmt = (SSliceThreading*)pMa->WelsMalloc (sizeof (SSliceThreading), "SSliceThreading");
WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt), FreeMemorySvc (ppCtx))
(*ppCtx)->pSliceThreading = pSmt;
pSmt->pThreadPEncCtx = (SSliceThreadPrivateData*)pMa->WelsMalloc (sizeof (SSliceThreadPrivateData) * iThreadNum,
"pThreadPEncCtx");
WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pThreadPEncCtx), FreeMemorySvc (ppCtx))
#ifdef _WIN32
// Dummy event namespace, the windows events don't actually use this
WelsSnprintf (pSmt->eventNamespace, sizeof (pSmt->eventNamespace), "%p", (void*) *ppCtx);
#else
WelsSnprintf (pSmt->eventNamespace, sizeof (pSmt->eventNamespace), "%p%x", (void*) *ppCtx, getpid());
#endif//!_WIN32
iIdx = 0;
while (iIdx < iNumSpatialLayers) {
SSliceArgument* pSliceArgument = &pPara->sSpatialLayers[iIdx].sSliceArgument;
if (pSliceArgument->uiSliceMode == SM_FIXEDSLCNUM_SLICE || pSliceArgument->uiSliceMode == SM_RASTER_SLICE) {
bWillUseTaskManage = true;
}
++ iIdx;
}
#ifdef MT_DEBUG
// file handle for MT debug
pSmt->pFSliceDiff = NULL;
if (pSmt->pFSliceDiff) {
fclose (pSmt->pFSliceDiff);
pSmt->pFSliceDiff = NULL;
}
pSmt->pFSliceDiff = fopen ("slice_time.txt", "wt+");
#endif//MT_DEBUG
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "encpEncCtx= 0x%p", (void*) *ppCtx);
char name[SEM_NAME_MAX] = {0};
WELS_GCC_UNUSED WELS_THREAD_ERROR_CODE err = 0;
iIdx = 0;
while (iIdx < iThreadNum) {
pSmt->pThreadPEncCtx[iIdx].pWelsPEncCtx = (void*) *ppCtx;
pSmt->pThreadPEncCtx[iIdx].iSliceIndex = iIdx;
pSmt->pThreadPEncCtx[iIdx].iThreadIndex = iIdx;
pSmt->pThreadHandles[iIdx] = 0;
WelsSnprintf (name, SEM_NAME_MAX, "ee%d%s", iIdx, pSmt->eventNamespace);
err = WelsEventOpen (&pSmt->pExitEncodeEvent[iIdx], name);
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "[MT] Open pExitEncodeEvent%d named(%s) ret%d err%d", iIdx, name, err, errno);
WelsSnprintf (name, SEM_NAME_MAX, "tm%d%s", iIdx, pSmt->eventNamespace);
err = WelsEventOpen (&pSmt->pThreadMasterEvent[iIdx], name);
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "[MT] Open pThreadMasterEvent%d named(%s) ret%d err%d", iIdx, name, err, errno);
// length of semaphore name should be system constrained at least on mac 10.7
WelsSnprintf (name, SEM_NAME_MAX, "ud%d%s", iIdx, pSmt->eventNamespace);
err = WelsEventOpen (&pSmt->pUpdateMbListEvent[iIdx], name);
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "[MT] Open pUpdateMbListEvent%d named(%s) ret%d err%d", iIdx, name, err, errno);
WelsSnprintf (name, SEM_NAME_MAX, "fu%d%s", iIdx, pSmt->eventNamespace);
err = WelsEventOpen (&pSmt->pFinUpdateMbListEvent[iIdx], name);
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "[MT] Open pFinUpdateMbListEvent%d named(%s) ret%d err%d", iIdx, name, err,
errno);
WelsSnprintf (name, SEM_NAME_MAX, "sc%d%s", iIdx, pSmt->eventNamespace);
err = WelsEventOpen (&pSmt->pSliceCodedEvent[iIdx], name);
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "[MT] Open pSliceCodedEvent%d named(%s) ret%d err%d", iIdx, name, err, errno);
WelsSnprintf (name, SEM_NAME_MAX, "rc%d%s", iIdx, pSmt->eventNamespace);
err = WelsEventOpen (&pSmt->pReadySliceCodingEvent[iIdx], name);
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "[MT] Open pReadySliceCodingEvent%d = 0x%p named(%s) ret%d err%d", iIdx,
(void*)pSmt->pReadySliceCodingEvent[iIdx], name, err, errno);
pSmt->pThreadBsBuffer[iIdx] = (uint8_t*)pMa->WelsMalloc (iCountBsLen, "pSmt->pThreadBsBuffer");
WELS_VERIFY_RETURN_PROC_IF (1, (NULL == pSmt->pThreadBsBuffer[iIdx]), FreeMemorySvc (ppCtx))
++ iIdx;
}
for (; iIdx < MAX_THREADS_NUM; iIdx++) {
pSmt->pThreadBsBuffer[iIdx] = NULL;
}
//previous conflict
WelsSnprintf (name, SEM_NAME_MAX, "scm%s", pSmt->eventNamespace);
err = WelsEventOpen (&pSmt->pSliceCodedMasterEvent, name);
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "[MT] Open pSliceCodedMasterEvent named(%s) ret%d err%d", name, err, errno);
//previous conflict ends
iReturn = SetMultiSliceBuffer (ppCtx, pMa, pSmt, iMaxSliceNum,
iMaxSliceBufferSize,
iCountBsLen,
bDynamicSlice);
WELS_VERIFY_RETURN_PROC_IF (iReturn, (ENC_RETURN_SUCCESS != iReturn), FreeMemorySvc (ppCtx))
iReturn = WelsMutexInit (&pSmt->mutexSliceNumUpdate);
WELS_VERIFY_RETURN_PROC_IF (1, (WELS_THREAD_ERROR_OK != iReturn), FreeMemorySvc (ppCtx))
if (bWillUseTaskManage) {
(*ppCtx)->pTaskManage = IWelsTaskManage::CreateTaskManage (*ppCtx, iNumSpatialLayers, bDynamicSlice);
WELS_VERIFY_RETURN_PROC_IF (iReturn, (NULL == (*ppCtx)->pTaskManage), FreeMemorySvc (ppCtx))
}
memset (&pSmt->bThreadBsBufferUsage, 0, MAX_THREADS_NUM * sizeof (bool));
iReturn = WelsMutexInit (&pSmt->mutexThreadBsBufferUsage);
WELS_VERIFY_RETURN_PROC_IF (1, (WELS_THREAD_ERROR_OK != iReturn), FreeMemorySvc (ppCtx))
iReturn = WelsMutexInit (& (*ppCtx)->mutexEncoderError);
WELS_VERIFY_RETURN_PROC_IF (1, (WELS_THREAD_ERROR_OK != iReturn), FreeMemorySvc (ppCtx))
MT_TRACE_LOG (*ppCtx, WELS_LOG_INFO, "RequestMtResource(), iThreadNum=%d, iCountSliceNum= %d",
pPara->iCountThreadsNum,
iMaxSliceNum);
return 0;
}
void ReleaseMtResource (sWelsEncCtx** ppCtx) {
SWelsSliceBs* pSliceB = NULL;
SSliceThreading* pSmt = NULL;
CMemoryAlign* pMa = NULL;
int32_t iIdx = 0;
int32_t iThreadNum = 0;
int16_t uiSliceNum = 0;
if (NULL == ppCtx || NULL == *ppCtx)
return;
pMa = (*ppCtx)->pMemAlign;
uiSliceNum = (*ppCtx)->iMaxSliceCount;
iThreadNum = (*ppCtx)->pSvcParam->iCountThreadsNum;
pSmt = (*ppCtx)->pSliceThreading;
if (NULL == pSmt)
return;
char ename[SEM_NAME_MAX] = {0};
while (iIdx < iThreadNum) {
// length of semaphore name should be system constrained at least on mac 10.7
WelsSnprintf (ename, SEM_NAME_MAX, "ee%d%s", iIdx, pSmt->eventNamespace);
WelsEventClose (&pSmt->pExitEncodeEvent[iIdx], ename);
WelsSnprintf (ename, SEM_NAME_MAX, "tm%d%s", iIdx, pSmt->eventNamespace);
WelsEventClose (&pSmt->pThreadMasterEvent[iIdx], ename);
WelsSnprintf (ename, SEM_NAME_MAX, "sc%d%s", iIdx, pSmt->eventNamespace);
WelsEventClose (&pSmt->pSliceCodedEvent[iIdx], ename);
WelsSnprintf (ename, SEM_NAME_MAX, "rc%d%s", iIdx, pSmt->eventNamespace);
WelsEventClose (&pSmt->pReadySliceCodingEvent[iIdx], ename);
WelsSnprintf (ename, SEM_NAME_MAX, "ud%d%s", iIdx, pSmt->eventNamespace);
WelsEventClose (&pSmt->pUpdateMbListEvent[iIdx], ename);
WelsSnprintf (ename, SEM_NAME_MAX, "fu%d%s", iIdx, pSmt->eventNamespace);
WelsEventClose (&pSmt->pFinUpdateMbListEvent[iIdx], ename);
++ iIdx;
}
WelsSnprintf (ename, SEM_NAME_MAX, "scm%s", pSmt->eventNamespace);
WelsEventClose (&pSmt->pSliceCodedMasterEvent, ename);
WelsMutexDestroy (&pSmt->mutexSliceNumUpdate);
WelsMutexDestroy (&pSmt->mutexThreadBsBufferUsage);
WelsMutexDestroy (& ((*ppCtx)->mutexEncoderError));
if (pSmt->pThreadPEncCtx != NULL) {
pMa->WelsFree (pSmt->pThreadPEncCtx, "pThreadPEncCtx");
pSmt->pThreadPEncCtx = NULL;
}
for (int i = 0; i < MAX_THREADS_NUM; i++) {
if (pSmt->pThreadBsBuffer[i]) {
pMa->WelsFree (pSmt->pThreadBsBuffer[i], "pSmt->pThreadBsBuffer");
pSmt->pThreadBsBuffer[i] = NULL;
}
}
memset (&pSmt->bThreadBsBufferUsage, 0, MAX_THREADS_NUM * sizeof (bool));
pSliceB = (*ppCtx)->pSliceBs;
iIdx = 0;
while (pSliceB != NULL && iIdx < uiSliceNum) {
pSliceB->pBsBuffer = NULL;
pSliceB->uiSize = 0;
pSliceB->uiBsPos = 0;
++ iIdx;
++ pSliceB;
}
if ((*ppCtx)->pSliceBs != NULL) {
pMa->WelsFree ((*ppCtx)->pSliceBs, "pSliceBs");
(*ppCtx)->pSliceBs = NULL;
}
if ((*ppCtx)->pTaskManage != NULL) {
delete (*ppCtx)->pTaskManage;
(*ppCtx)->pTaskManage = NULL;
}
#ifdef MT_DEBUG
// file handle for debug
if (pSmt->pFSliceDiff) {
fclose (pSmt->pFSliceDiff);
pSmt->pFSliceDiff = NULL;
}
#endif//MT_DEBUG
pMa->WelsFree ((*ppCtx)->pSliceThreading, "SSliceThreading");
(*ppCtx)->pSliceThreading = NULL;
}
int32_t AppendSliceToFrameBs (sWelsEncCtx* pCtx, SLayerBSInfo* pLbi, const int32_t iSliceCount) {
SWelsSvcCodingParam* pCodingParam = pCtx->pSvcParam;
SSpatialLayerConfig* pDlp = &pCodingParam->sSpatialLayers[pCtx->uiDependencyId];
SWelsSliceBs* pSliceBs = NULL;
const bool kbIsDynamicSlicingMode = (pDlp->sSliceArgument.uiSliceMode == SM_SIZELIMITED_SLICE);
int32_t iLayerSize = 0;
int32_t iNalIdxBase = pLbi->iNalCount;
int32_t iSliceIdx = 0;
if (!kbIsDynamicSlicingMode) {
pSliceBs = &pCtx->pSliceBs[0];
iLayerSize = pSliceBs->uiBsPos; // assign with base pSlice first
iSliceIdx = 0;
iNalIdxBase = pLbi->iNalCount = 0;
while (iSliceIdx < iSliceCount) {
if (pSliceBs != NULL && pSliceBs->uiBsPos > 0) {
int32_t iNalIdx = 0;
const int32_t iCountNal = pSliceBs->iNalIndex;
#if MT_DEBUG_BS_WR
assert (pSliceBs->bSliceCodedFlag);
#endif//MT_DEBUG_BS_WR
if (iSliceIdx > 0) {
// pSlice 0 bs has been written to pFrameBs yet by now, so uiSliceIdx base should be 1
memmove (pCtx->pFrameBs + pCtx->iPosBsBuffer, pSliceBs->pBs, pSliceBs->uiBsPos); // confirmed_safe_unsafe_usage
pCtx->iPosBsBuffer += pSliceBs->uiBsPos;
iLayerSize += pSliceBs->uiBsPos;
}
while (iNalIdx < iCountNal) {
pLbi->pNalLengthInByte[iNalIdxBase + iNalIdx] = pSliceBs->iNalLen[iNalIdx];
++ iNalIdx;
}
pLbi->iNalCount += iCountNal;
iNalIdxBase += iCountNal;
}
++ iSliceIdx;
++ pSliceBs;
}
} else { // for SM_SIZELIMITED_SLICE
const int32_t kiPartitionCnt = iSliceCount;
int32_t iPartitionIdx = 0;
// due partition_0 has been written to pFrameBsBuffer
// so iLayerSize need add it
while (iPartitionIdx < kiPartitionCnt) {
const int32_t kiCountSlicesCoded = pCtx->pCurDqLayer->pNumSliceCodedOfPartition[iPartitionIdx];
int32_t iIdx = 0;
iSliceIdx = iPartitionIdx;
while (iIdx < kiCountSlicesCoded) {
pSliceBs = &pCtx->pSliceBs[iSliceIdx];
if (pSliceBs != NULL && pSliceBs->uiBsPos > 0) {
if (iSliceIdx > 0) {
int32_t iNalIdx = 0;
const int32_t iCountNal = pSliceBs->iNalIndex;
memmove (pCtx->pFrameBs + pCtx->iPosBsBuffer, pSliceBs->pBs, pSliceBs->uiBsPos); // confirmed_safe_unsafe_usage
pCtx->iPosBsBuffer += pSliceBs->uiBsPos;
iLayerSize += pSliceBs->uiBsPos;
while (iNalIdx < iCountNal) {
pLbi->pNalLengthInByte[iNalIdxBase + iNalIdx] = pSliceBs->iNalLen[iNalIdx];
++ iNalIdx;
}
pLbi->iNalCount += iCountNal;
iNalIdxBase += iCountNal;
} else {
iLayerSize += pSliceBs->uiBsPos;
}
}
iSliceIdx += kiPartitionCnt;
++ iIdx;
}
++ iPartitionIdx;
}
}
return iLayerSize;
}
int32_t WriteSliceBs (sWelsEncCtx* pCtx, uint8_t* pDst,
int32_t* pNalLen,
int32_t iTotalLeftLength,
const int32_t iSliceIdx,
int32_t& iSliceSize) {
SWelsSliceBs* pSliceBs = &pCtx->pSliceBs[iSliceIdx];
SNalUnitHeaderExt* pNalHdrExt = &pCtx->pCurDqLayer->sLayerInfo.sNalHeaderExt;
const int32_t kiNalCnt = pSliceBs->iNalIndex;
int32_t iNalIdx = 0;
int32_t iNalSize = 0;
int32_t iReturn = ENC_RETURN_SUCCESS;
iSliceSize = 0;
assert (kiNalCnt <= 2);
if (kiNalCnt > 2)
return 0;
while (iNalIdx < kiNalCnt) {
iNalSize = 0;
iReturn = WelsEncodeNal (&pSliceBs->sNalList[iNalIdx], pNalHdrExt, iTotalLeftLength - iSliceSize,
pDst, &iNalSize);
WELS_VERIFY_RETURN_IFNEQ (iReturn, ENC_RETURN_SUCCESS)
pNalLen[iNalIdx] = iNalSize;
iSliceSize += iNalSize;
pDst += iNalSize;
++ iNalIdx;
}
pSliceBs->uiBsPos = iSliceSize;
return iReturn;
}
// thread process for coding one pSlice
WELS_THREAD_ROUTINE_TYPE CodingSliceThreadProc (void* arg) {
SSliceThreadPrivateData* pPrivateData = (SSliceThreadPrivateData*)arg;
sWelsEncCtx* pEncPEncCtx = NULL;
SDqLayer* pCurDq = NULL;
SSlice* pSlice = NULL;
SWelsSliceBs* pSliceBs = NULL;
WELS_EVENT pEventsList[3];
int32_t iEventCount = 0;
WELS_THREAD_ERROR_CODE iWaitRet = WELS_THREAD_ERROR_GENERAL;
uint32_t uiThrdRet = 0;
int32_t iSliceSize = 0;
int32_t iSliceIdx = -1;
int32_t iThreadIdx = -1;
int32_t iEventIdx = -1;
bool bNeedPrefix = false;
EWelsNalUnitType eNalType = NAL_UNIT_UNSPEC_0;
EWelsNalRefIdc eNalRefIdc = NRI_PRI_LOWEST;
int32_t iReturn = ENC_RETURN_SUCCESS;
if (NULL == pPrivateData)
WELS_THREAD_ROUTINE_RETURN (1);
pEncPEncCtx = (sWelsEncCtx*)pPrivateData->pWelsPEncCtx;
iThreadIdx = pPrivateData->iThreadIndex;
iEventIdx = iThreadIdx;
pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pReadySliceCodingEvent[iEventIdx];
pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pExitEncodeEvent[iEventIdx];
pEventsList[iEventCount++] = pEncPEncCtx->pSliceThreading->pUpdateMbListEvent[iEventIdx];
WelsThreadSetName ("OpenH264Enc_CodingSliceThreadProc");
do {
MT_TRACE_LOG (pEncPEncCtx, WELS_LOG_INFO,
"[MT] CodingSliceThreadProc(), try to call WelsMultipleEventsWaitSingleBlocking(pEventsList= %p %p %p), pEncPEncCtx= %p!",
pEventsList[0], pEventsList[1], pEventsList[1], (void*)pEncPEncCtx);
iWaitRet = WelsMultipleEventsWaitSingleBlocking (iEventCount,
&pEventsList[0],
&pEncPEncCtx->pSliceThreading->pThreadMasterEvent[iEventIdx]); // blocking until at least one event is signalled
if (WELS_THREAD_ERROR_WAIT_OBJECT_0 == iWaitRet) { // start pSlice coding signal waited
int iLayerIndex = pEncPEncCtx->pOut->iLayerBsIndex;
SFrameBSInfo* pFrameBsInfo = pPrivateData->pFrameBsInfo;
SLayerBSInfo* pLbi = &pFrameBsInfo->sLayerInfo [iLayerIndex];
const int32_t kiCurDid = pEncPEncCtx->uiDependencyId;
SWelsSvcCodingParam* pCodingParam = pEncPEncCtx->pSvcParam;
SSpatialLayerConfig* pParamD = &pCodingParam->sSpatialLayers[kiCurDid];
pCurDq = pEncPEncCtx->pCurDqLayer;
eNalType = pEncPEncCtx->eNalType;
eNalRefIdc = pEncPEncCtx->eNalPriority;
bNeedPrefix = pEncPEncCtx->bNeedPrefixNalFlag;
if (pParamD->sSliceArgument.uiSliceMode != SM_SIZELIMITED_SLICE) {
int64_t iSliceStart = 0;
bool bDsaFlag = false;
iSliceIdx = pPrivateData->iSliceIndex;
pSlice = &pCurDq->sLayerInfo.pSliceInLayer[iSliceIdx];
pSliceBs = &pEncPEncCtx->pSliceBs[iSliceIdx];
bDsaFlag = ((pParamD->sSliceArgument.uiSliceMode == SM_FIXEDSLCNUM_SLICE) &&
pCodingParam->iMultipleThreadIdc > 1 &&
pCodingParam->iMultipleThreadIdc >= pParamD->sSliceArgument.uiSliceNum);
if (bDsaFlag)
iSliceStart = WelsTime();
pSliceBs->uiBsPos = 0;
pSliceBs->iNalIndex = 0;
assert ((void*) (&pSliceBs->sBsWrite) == (void*)pSlice->pSliceBsa);
InitBits (&pSliceBs->sBsWrite, pSliceBs->pBsBuffer, pSliceBs->uiSize);
#if MT_DEBUG_BS_WR
pSliceBs->bSliceCodedFlag = false;
#endif//MT_DEBUG_BS_WR
if (bNeedPrefix) {
if (eNalRefIdc != NRI_PRI_LOWEST) {
WelsLoadNalForSlice (pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc);
WelsWriteSVCPrefixNal (&pSliceBs->sBsWrite, eNalRefIdc, (NAL_UNIT_CODED_SLICE_IDR == eNalType));
WelsUnloadNalForSlice (pSliceBs);
} else { // No Prefix NAL Unit RBSP syntax here, but need add NAL Unit Header extension
WelsLoadNalForSlice (pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc);
// No need write any syntax of prefix NAL Unit RBSP here
WelsUnloadNalForSlice (pSliceBs);
}
}
WelsLoadNalForSlice (pSliceBs, eNalType, eNalRefIdc);
iReturn = WelsCodeOneSlice (pEncPEncCtx, iSliceIdx, eNalType);
if (ENC_RETURN_SUCCESS != iReturn) {
uiThrdRet = iReturn;
WELS_THREAD_SIGNAL_AND_BREAK (pEncPEncCtx->pSliceThreading->pSliceCodedEvent,
pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent,
iEventIdx);
}
WelsUnloadNalForSlice (pSliceBs);
int32_t iLeftBufferSize = (iSliceIdx > 0) ?
(pSliceBs->uiSize - pSliceBs->uiBsPos)
: (pEncPEncCtx->iFrameBsSize - pEncPEncCtx->iPosBsBuffer);
iReturn = WriteSliceBs (pEncPEncCtx, pSliceBs->pBs,
&pSliceBs->iNalLen[0],
iLeftBufferSize,
iSliceIdx, iSliceSize);
if (ENC_RETURN_SUCCESS != iReturn) {
uiThrdRet = iReturn;
WELS_THREAD_SIGNAL_AND_BREAK (pEncPEncCtx->pSliceThreading->pSliceCodedEvent,
pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent,
iEventIdx);
}
if (0 == iSliceIdx) {
pEncPEncCtx->iPosBsBuffer += iSliceSize;
}
pEncPEncCtx->pFuncList->pfDeblocking.pfDeblockingFilterSlice (pCurDq, pEncPEncCtx->pFuncList, iSliceIdx);
if (bDsaFlag) {
pEncPEncCtx->pCurDqLayer->sLayerInfo.pSliceInLayer[iSliceIdx].uiSliceConsumeTime = (uint32_t) (
WelsTime() - iSliceStart);
MT_TRACE_LOG (& (pEncPEncCtx->sLogCtx), WELS_LOG_INFO,
"[MT] CodingSliceThreadProc(), coding_idx %d, uiSliceIdx %d, uiSliceConsumeTime %d, iSliceSize %d, iFirstMbInSlice %d, count_num_mb_in_slice %d",
pEncPEncCtx->iCodingIndex, iSliceIdx,
pEncPEncCtx->pCurDqLayer->sLayerInfo.pSliceInLayer[iSliceIdx].uiSliceConsumeTime, iSliceSize,
pCurDq->sLayerInfo.pSliceInLayer[iSliceIdx].sSliceHeaderExt.sSliceHeader.iFirstMbInSlice,
pCurDq->sSliceEncCtx.pCountMbNumInSlice[iSliceIdx]);
}
#if defined(SLICE_INFO_OUTPUT)
fprintf (stderr,
"@pSlice=%-6d sliceType:%c idc:%d size:%-6d\n",
iSliceIdx,
(pEncPEncCtx->eSliceType == P_SLICE ? 'P' : 'I'),
eNalRefIdc,
iSliceSize
);
#endif//SLICE_INFO_OUTPUT
#if MT_DEBUG_BS_WR
pSliceBs->bSliceCodedFlag = true;
#endif//MT_DEBUG_BS_WR
WelsEventSignal (
&pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx]); // mean finished coding current pSlice
WelsEventSignal (
&pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent);
} else { // for SM_SIZELIMITED_SLICE parallelization
SSliceCtx* pSliceCtx = &pCurDq->sSliceEncCtx;
const int32_t kiPartitionId = iThreadIdx;
const int32_t kiSliceIdxStep = pEncPEncCtx->iActiveThreadsNum;
const int32_t kiFirstMbInPartition = pPrivateData->iStartMbIndex; // inclusive
const int32_t kiEndMbInPartition = pPrivateData->iEndMbIndex; // exclusive
int32_t iAnyMbLeftInPartition = kiEndMbInPartition - kiFirstMbInPartition;
SSliceHeaderExt* pStartSliceHeaderExt = &pCurDq->sLayerInfo.pSliceInLayer[iSliceIdx].sSliceHeaderExt;
iSliceIdx = pPrivateData->iSliceIndex;
pStartSliceHeaderExt->sSliceHeader.iFirstMbInSlice = kiFirstMbInPartition;
pCurDq->pNumSliceCodedOfPartition[kiPartitionId] =
1; // one pSlice per partition intialized, dynamic slicing inside
pCurDq->pLastMbIdxOfPartition[kiPartitionId] = kiEndMbInPartition - 1;
pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId] = 0;
while (iAnyMbLeftInPartition > 0) {
if (iSliceIdx >= pSliceCtx->iMaxSliceNumConstraint) {
// TODO: need exception handler for not large enough of MAX_SLICES_NUM related memory usage
// No idea about its solution due MAX_SLICES_NUM is fixed lenght in relevent pData structure
uiThrdRet = 1;
WELS_THREAD_SIGNAL_AND_BREAK (pEncPEncCtx->pSliceThreading->pSliceCodedEvent,
pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent,
iEventIdx);
}
SetOneSliceBsBufferUnderMultithread (pEncPEncCtx, kiPartitionId, iSliceIdx);
pSlice = &pCurDq->sLayerInfo.pSliceInLayer[iSliceIdx];
pSliceBs = &pEncPEncCtx->pSliceBs[iSliceIdx];
pSliceBs->uiBsPos = 0;
pSliceBs->iNalIndex = 0;
InitBits (&pSliceBs->sBsWrite, pSliceBs->pBsBuffer, pSliceBs->uiSize);
if (bNeedPrefix) {
if (eNalRefIdc != NRI_PRI_LOWEST) {
WelsLoadNalForSlice (pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc);
WelsWriteSVCPrefixNal (&pSliceBs->sBsWrite, eNalRefIdc, (NAL_UNIT_CODED_SLICE_IDR == eNalType));
WelsUnloadNalForSlice (pSliceBs);
} else { // No Prefix NAL Unit RBSP syntax here, but need add NAL Unit Header extension
WelsLoadNalForSlice (pSliceBs, NAL_UNIT_PREFIX, eNalRefIdc);
// No need write any syntax of prefix NAL Unit RBSP here
WelsUnloadNalForSlice (pSliceBs);
}
}
WelsLoadNalForSlice (pSliceBs, eNalType, eNalRefIdc);
iReturn = WelsCodeOneSlice (pEncPEncCtx, iSliceIdx, eNalType);
if (ENC_RETURN_SUCCESS != iReturn) {
uiThrdRet = iReturn;
WELS_THREAD_SIGNAL_AND_BREAK (pEncPEncCtx->pSliceThreading->pSliceCodedEvent,
pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent,
iEventIdx);
}
WelsUnloadNalForSlice (pSliceBs);
if (0 == iSliceIdx) {
iReturn = WriteSliceBs (pEncPEncCtx, pLbi->pBsBuf,
&pLbi->pNalLengthInByte[pLbi->iNalCount],
pEncPEncCtx->iFrameBsSize - pEncPEncCtx->iPosBsBuffer,
iSliceIdx, iSliceSize);
pLbi->iNalCount += pSliceBs->iNalIndex;
if (ENC_RETURN_SUCCESS != iReturn) {
uiThrdRet = iReturn;
WELS_THREAD_SIGNAL_AND_BREAK (pEncPEncCtx->pSliceThreading->pSliceCodedEvent,
pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent,
iEventIdx);
}
pEncPEncCtx->iPosBsBuffer += iSliceSize;
} else {
iReturn = WriteSliceBs (pEncPEncCtx, pSliceBs->pBs, &pSliceBs->iNalLen[0],
pSliceBs->uiSize - pSliceBs->uiBsPos,
iSliceIdx, iSliceSize);
if (ENC_RETURN_SUCCESS != iReturn) {
uiThrdRet = iReturn;
WELS_THREAD_SIGNAL_AND_BREAK (pEncPEncCtx->pSliceThreading->pSliceCodedEvent,
pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent,
iEventIdx);
}
}
pEncPEncCtx->pFuncList->pfDeblocking.pfDeblockingFilterSlice (pCurDq, pEncPEncCtx->pFuncList, iSliceIdx);
#if defined(SLICE_INFO_OUTPUT)
fprintf (stderr,
"@pSlice=%-6d sliceType:%c idc:%d size:%-6d\n",
iSliceIdx,
(pEncPEncCtx->eSliceType == P_SLICE ? 'P' : 'I'),
eNalRefIdc,
iSliceSize
);
#endif//SLICE_INFO_OUTPUT
MT_TRACE_LOG (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",
pEncPEncCtx->iCodingIndex, kiPartitionId, iSliceIdx, iSliceSize, pCurDq->sSliceEncCtx.pCountMbNumInSlice[iSliceIdx],
kiEndMbInPartition, kiPartitionId, pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId]);
iAnyMbLeftInPartition = kiEndMbInPartition - (1 + pCurDq->pLastCodedMbIdxOfPartition[kiPartitionId]);
iSliceIdx += kiSliceIdxStep;
}
if (uiThrdRet) { // any exception??
WELS_THREAD_SIGNAL_AND_BREAK (pEncPEncCtx->pSliceThreading->pSliceCodedEvent,
pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent,
iEventIdx);
}
WelsEventSignal (&pEncPEncCtx->pSliceThreading->pSliceCodedEvent[iEventIdx]); // mean finished coding current pSlice
WelsEventSignal (&pEncPEncCtx->pSliceThreading->pSliceCodedMasterEvent);
}
} else if (WELS_THREAD_ERROR_WAIT_OBJECT_0 + 1 == iWaitRet) { // exit thread signal
uiThrdRet = 0;
break;
} else if (WELS_THREAD_ERROR_WAIT_OBJECT_0 + 2 == iWaitRet) { // update pMb list singal
iSliceIdx =
iEventIdx; // pPrivateData->iSliceIndex; old threads can not be terminated, pPrivateData is not correct for applicable
pCurDq = pEncPEncCtx->pCurDqLayer;
UpdateMbListNeighborParallel (pCurDq, pCurDq->sMbDataP, iSliceIdx);
WelsEventSignal (
&pEncPEncCtx->pSliceThreading->pFinUpdateMbListEvent[iEventIdx]); // mean finished update pMb list for this pSlice
} else { // WELS_THREAD_ERROR_WAIT_TIMEOUT, or WELS_THREAD_ERROR_WAIT_FAILED
WelsLog (& (pEncPEncCtx->sLogCtx), WELS_LOG_WARNING,
"[MT] CodingSliceThreadProc(), waiting pReadySliceCodingEvent[%d] failed(%d) and thread%d terminated!", iEventIdx,
iWaitRet, iThreadIdx);
uiThrdRet = 1;
break;
}
} while (1);
//sync multi-threading error
WelsMutexLock (&pEncPEncCtx->mutexEncoderError);
if (uiThrdRet) pEncPEncCtx->iEncoderError |= uiThrdRet;
WelsMutexUnlock (&pEncPEncCtx->mutexEncoderError);
WELS_THREAD_ROUTINE_RETURN (uiThrdRet);
}
int32_t CreateSliceThreads (sWelsEncCtx* pCtx) {
const int32_t kiThreadCount = pCtx->pSvcParam->iCountThreadsNum;
int32_t iIdx = 0;
while (iIdx < kiThreadCount) {
if (WelsThreadCreate (&pCtx->pSliceThreading->pThreadHandles[iIdx], CodingSliceThreadProc,
&pCtx->pSliceThreading->pThreadPEncCtx[iIdx], 0)) {
return 1;
}
++ iIdx;
}
MT_TRACE_LOG (pCtx, WELS_LOG_INFO, "CreateSliceThreads() exit..");
return 0;
}
int32_t FiredSliceThreads (sWelsEncCtx* pCtx, SSliceThreadPrivateData* pPriData, WELS_EVENT* pEventsList,
WELS_EVENT* pMasterEventsList, SFrameBSInfo* pFrameBsInfo,
const uint32_t uiNumThreads, SSliceCtx* pSliceCtx, const bool bIsDynamicSlicingMode) {
int32_t iEndMbIdx = 0;
int32_t iIdx = 0;
const int32_t kiEventCnt = uiNumThreads;
int32_t iLayerBsIdx = pCtx->pOut->iLayerBsIndex;
SLayerBSInfo* pLbi = &pFrameBsInfo->sLayerInfo[iLayerBsIdx];
SSlice* pSliceInLayer = pCtx->pCurDqLayer->sLayerInfo.pSliceInLayer;
if (pPriData == NULL || pFrameBsInfo == NULL || pLbi == NULL || kiEventCnt <= 0 || pEventsList == NULL) {
WelsLog (& (pCtx->sLogCtx), WELS_LOG_ERROR,
"FiredSliceThreads(), fail due pPriData == %p ||pFrameBsInfo == %p || pLbi == %p || iEventCnt(%d) <= 0 || pEventsList == %p!!",
(void*)pPriData, (void*)pFrameBsInfo, (void*)pLbi, uiNumThreads, (void*)pEventsList);
return 1;
}
////////////////////////////////////////
if (bIsDynamicSlicingMode) {
iEndMbIdx = pSliceCtx->iMbNumInFrame;
for (iIdx = kiEventCnt - 1; iIdx >= 0; --iIdx) {
const int32_t iFirstMbIdx = pSliceInLayer[iIdx].sSliceHeaderExt.sSliceHeader.iFirstMbInSlice;
pPriData[iIdx].iStartMbIndex = iFirstMbIdx;
pPriData[iIdx].iEndMbIndex = iEndMbIdx;
iEndMbIdx = iFirstMbIdx;
}
}
pLbi->pBsBuf = pCtx->pFrameBs + pCtx->iPosBsBuffer;
pLbi->uiLayerType = VIDEO_CODING_LAYER;
pLbi->uiSpatialId = pCtx->uiDependencyId;
pLbi->uiTemporalId = pCtx->uiTemporalId;
pLbi->uiQualityId = 0;
pLbi->iNalCount = 0;
pCtx->pSliceBs[0].pBs = pLbi->pBsBuf;
iIdx = 0;
while (iIdx < kiEventCnt) {
pPriData[iIdx].pFrameBsInfo = pFrameBsInfo;
pPriData[iIdx].iSliceIndex = iIdx;
SetOneSliceBsBufferUnderMultithread (pCtx, iIdx, iIdx);
if (pEventsList[iIdx])
WelsEventSignal (&pEventsList[iIdx]);
if (pMasterEventsList[iIdx])
WelsEventSignal (&pMasterEventsList[iIdx]);
++ iIdx;
}
return 0;
}
int32_t DynamicDetectCpuCores() {
WelsLogicalProcessInfo info;
WelsQueryLogicalProcessInfo (&info);
return info.ProcessorCount;
}
int32_t AdjustBaseLayer (sWelsEncCtx* pCtx) {
SDqLayer* pCurDq = pCtx->ppDqLayerList[0];
int32_t iNeedAdj = 1;
#ifdef MT_DEBUG
int64_t iT0 = WelsTime();
#endif//MT_DEBUG
pCtx->pCurDqLayer = pCurDq;
// do not need adjust due to not different at both slices of consumed time
iNeedAdj = NeedDynamicAdjust (pCtx->ppDqLayerList[0]->sLayerInfo.pSliceInLayer, pCurDq->sSliceEncCtx.iSliceNumInFrame);
if (iNeedAdj)
DynamicAdjustSlicing (pCtx,
pCurDq,
0);
#ifdef MT_DEBUG
iT0 = WelsTime() - iT0;
if (pCtx->pSliceThreading->pFSliceDiff) {
fprintf (pCtx->pSliceThreading->pFSliceDiff,
"%6"PRId64" us adjust time at base spatial layer, iNeedAdj %d, DynamicAdjustSlicing()\n",
iT0, iNeedAdj);
}
#endif//MT_DEBUG
return iNeedAdj;
}
int32_t AdjustEnhanceLayer (sWelsEncCtx* pCtx, int32_t iCurDid) {
#ifdef MT_DEBUG
int64_t iT1 = WelsTime();
#endif//MT_DEBUG
int32_t iNeedAdj = 1;
// uiSliceMode of referencing spatial should be SM_FIXEDSLCNUM_SLICE
// if using spatial base layer for complexity estimation
const bool kbModelingFromSpatial = (pCtx->pCurDqLayer->pRefLayer != NULL && iCurDid > 0)
&& (pCtx->pSvcParam->sSpatialLayers[iCurDid - 1].sSliceArgument.uiSliceMode == SM_FIXEDSLCNUM_SLICE
&& pCtx->pSvcParam->iMultipleThreadIdc >= pCtx->pSvcParam->sSpatialLayers[iCurDid -
1].sSliceArgument.uiSliceNum);
if (kbModelingFromSpatial) { // using spatial base layer for complexity estimation
// do not need adjust due to not different at both slices of consumed time
iNeedAdj = NeedDynamicAdjust (pCtx->ppDqLayerList[iCurDid - 1]->sLayerInfo.pSliceInLayer,
pCtx->pCurDqLayer->sSliceEncCtx.iSliceNumInFrame);
if (iNeedAdj)
DynamicAdjustSlicing (pCtx,
pCtx->pCurDqLayer,
iCurDid
);
} else { // use temporal layer for complexity estimation
// do not need adjust due to not different at both slices of consumed time
iNeedAdj = NeedDynamicAdjust (pCtx->ppDqLayerList[iCurDid]->sLayerInfo.pSliceInLayer,
pCtx->pCurDqLayer->sSliceEncCtx.iSliceNumInFrame);
if (iNeedAdj)
DynamicAdjustSlicing (pCtx,
pCtx->pCurDqLayer,
iCurDid
);
}
#ifdef MT_DEBUG
iT1 = WelsTime() - iT1;
if (pCtx->pSliceThreading->pFSliceDiff) {
fprintf (pCtx->pSliceThreading->pFSliceDiff,
"%6"PRId64" us adjust time at spatial layer %d, iNeedAdj %d, DynamicAdjustSlicing()\n",
iT1, iCurDid, iNeedAdj);
}
#endif//MT_DEBUG
return iNeedAdj;
}
#if defined(MT_DEBUG)
void TrackSliceComplexities (sWelsEncCtx* pCtx, const int32_t iCurDid) {
const int32_t kiCountSliceNum = pCtx->pCurDqLayer->sSliceEncCtx.iSliceNumInFrame;
SSlice* pSliceInLayer = pCtx->pCurDqLayer->sLayerInfo.pSliceInLayer;
if (kiCountSliceNum > 0) {
int32_t iSliceIdx = 0;
do {
fprintf (pCtx->pSliceThreading->pFSliceDiff, "%6.3f complexity pRatio at iDid %d pSlice %d\n",
pSliceInLayer[iSliceIdx].iSliceComplexRatio, iCurDid, iSliceIdx);
++ iSliceIdx;
} while (iSliceIdx < kiCountSliceNum);
}
}
#endif
#if defined(MT_DEBUG)
void TrackSliceConsumeTime (sWelsEncCtx* pCtx, int32_t* pDidList, const int32_t iSpatialNum) {
SWelsSvcCodingParam* pPara = NULL;
int32_t iSpatialIdx = 0;
if (iSpatialNum > MAX_DEPENDENCY_LAYER)
return;
pPara = pCtx->pSvcParam;
while (iSpatialIdx < iSpatialNum) {
const int32_t kiDid = pDidList[iSpatialIdx];
SSliceConfig* pSliceArgument = &pPara->sDependencyLayers[kiDid].sSliceArgument;
SDqLayer* pCurDq = pCtx->ppDqLayerList[kiDid];
SSlice* pSliceInLayer = pCurDq->sLayerInfo.pSliceInLayer;
SSliceCtx* pSliceCtx = &pCurDq->sSliceEncCtx;
const uint32_t kuiCountSliceNum = pSliceCtx->iSliceNumInFrame;
if (pCtx->pSliceThreading) {
if (pCtx->pSliceThreading->pFSliceDiff
&& ((pSliceArgument->uiSliceMode == SM_FIXEDSLCNUM_SLICE) || (pSliceArgument->uiSliceMode == SM_AUTO_SLICE))
&& pPara->iMultipleThreadIdc > 1
&& pPara->iMultipleThreadIdc >= kuiCountSliceNum) {
uint32_t i = 0;
uint32_t uiMaxT = 0;
int32_t iMaxI = 0;
while (i < kuiCountSliceNum) {
if (pSliceInLayer[i] != NULL)
fprintf (pCtx->pSliceThreading->pFSliceDiff, "%6d us consume_time coding_idx %d iDid %d pSlice %d\n",
pSliceInLayer[i].uiSliceConsumeTime, pCtx->iCodingIndex, kiDid, i /*/ 1000*/);
if (pSliceInLayer[i].uiSliceConsumeTime > uiMaxT) {
uiMaxT = pSliceInLayer[i].uiSliceConsumeTime;
iMaxI = i;
}
++ i;
}
fprintf (pCtx->pSliceThreading->pFSliceDiff, "%6d us consume_time_max coding_idx %d iDid %d pSlice %d\n", uiMaxT,
pCtx->iCodingIndex, kiDid, iMaxI /*/ 1000*/);
}
}
++ iSpatialIdx;
}
}
#endif//#if defined(MT_DEBUG)
void SetOneSliceBsBufferUnderMultithread (sWelsEncCtx* pCtx, const int32_t kiThreadIdx, const int32_t iSliceIdx) {
pCtx->pSliceBs[iSliceIdx].pBsBuffer = pCtx->pSliceThreading->pThreadBsBuffer[kiThreadIdx];
pCtx->pSliceBs[iSliceIdx].uiBsPos = 0;
//printf("SetOneSliceBsBufferUnderMultithread, thread %d, slice %d, buffer=%x\n", kiThreadIdx, iSliceIdx, pCtx->pSliceBs[iSliceIdx].pBsBuffer);
}
}