1709 |
|
|
1710 |
if (pEnc->num_threads > 0) { |
if (pEnc->num_threads > 0) { |
1711 |
/* multithreaded motion estimation - dispatch threads */ |
/* multithreaded motion estimation - dispatch threads */ |
1712 |
|
|
1713 |
|
void * status; |
1714 |
int rows_per_thread = (pParam->mb_height + pEnc->num_threads - 1)/pEnc->num_threads; |
int rows_per_thread = (pParam->mb_height + pEnc->num_threads - 1)/pEnc->num_threads; |
1715 |
|
|
1716 |
for (k = 0; k < pEnc->num_threads; k++) { |
for (k = 0; k < pEnc->num_threads; k++) { |
1728 |
pEnc->motionData[k].RefQ = pEnc->vInterH.u + 16*k*pParam->edged_width; |
pEnc->motionData[k].RefQ = pEnc->vInterH.u + 16*k*pParam->edged_width; |
1729 |
} |
} |
1730 |
|
|
1731 |
for (k = 0; k < pEnc->num_threads; k++) { |
for (k = 1; k < pEnc->num_threads; k++) { |
1732 |
pthread_create(&pEnc->motionData[k].handle, NULL, |
pthread_create(&pEnc->motionData[k].handle, NULL, |
1733 |
(void*)MotionEstimateSMP, (void*)&pEnc->motionData[k]); |
(void*)MotionEstimateSMP, (void*)&pEnc->motionData[k]); |
1734 |
} |
} |
1735 |
|
|
1736 |
|
MotionEstimateSMP(&pEnc->motionData[0]); |
1737 |
|
|
1738 |
|
for (k = 1; k < pEnc->num_threads; k++) { |
1739 |
|
pthread_join(pEnc->motionData[k].handle, &status); |
1740 |
|
} |
1741 |
|
|
1742 |
|
current->fcode = 0; |
1743 |
|
for (k = 0; k < pEnc->num_threads; k++) { |
1744 |
|
current->sStat.iMvSum += pEnc->motionData[k].mvSum; |
1745 |
|
current->sStat.iMvCount += pEnc->motionData[k].mvCount; |
1746 |
|
if (pEnc->motionData[k].minfcode > current->fcode) |
1747 |
|
current->fcode = pEnc->motionData[k].minfcode; |
1748 |
|
} |
1749 |
|
|
1750 |
} else { |
} else { |
1751 |
/* regular ME */ |
/* regular ME */ |
1752 |
|
|
1762 |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, current, 1, current->mbs[0].quant); |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, current, 1, current->mbs[0].quant); |
1763 |
|
|
1764 |
for (y = 0; y < mb_height; y++) { |
for (y = 0; y < mb_height; y++) { |
|
|
|
|
if (pEnc->num_threads > 0) { |
|
|
/* in multithreaded encoding, only proceed with a row of macroblocks |
|
|
if ME finished with that row */ |
|
|
while (pEnc->motionData[y%pEnc->num_threads].complete_count_self[y/pEnc->num_threads] != mb_width) |
|
|
sched_yield(); |
|
|
} |
|
|
|
|
1765 |
for (x = 0; x < mb_width; x++) { |
for (x = 0; x < mb_width; x++) { |
1766 |
MACROBLOCK *pMB = ¤t->mbs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = ¤t->mbs[x + y * pParam->mb_width]; |
1767 |
int skip_possible; |
int skip_possible; |
1872 |
} |
} |
1873 |
} |
} |
1874 |
|
|
|
if (pEnc->num_threads > 0) { |
|
|
void * status; |
|
|
for (k = 0; k < pEnc->num_threads; k++) { |
|
|
pthread_join(pEnc->motionData[k].handle, &status); |
|
|
} |
|
|
|
|
|
for (k = 0; k < pEnc->num_threads; k++) { |
|
|
current->sStat.iMvSum += pEnc->motionData[k].mvSum; |
|
|
current->sStat.iMvCount += pEnc->motionData[k].mvCount; |
|
|
} |
|
|
} |
|
|
|
|
1875 |
emms(); |
emms(); |
1876 |
updateFcode(¤t->sStat, pEnc); |
updateFcode(¤t->sStat, pEnc); |
1877 |
|
|
2002 |
|
|
2003 |
frame->fcode = frame->bcode = pEnc->current->fcode; |
frame->fcode = frame->bcode = pEnc->current->fcode; |
2004 |
|
|
2005 |
|
start_timer(); |
2006 |
if (pEnc->num_threads > 0) { |
if (pEnc->num_threads > 0) { |
2007 |
|
void * status; |
2008 |
int k; |
int k; |
2009 |
/* multithreaded motion estimation - dispatch threads */ |
/* multithreaded motion estimation - dispatch threads */ |
2010 |
int rows_per_thread = (pEnc->mbParam.mb_height + pEnc->num_threads - 1)/pEnc->num_threads; |
int rows_per_thread = (pEnc->mbParam.mb_height + pEnc->num_threads - 1)/pEnc->num_threads; |
2030 |
pEnc->motionData[k].RefQ = pEnc->vInterH.u + 16*k*pEnc->mbParam.edged_width; |
pEnc->motionData[k].RefQ = pEnc->vInterH.u + 16*k*pEnc->mbParam.edged_width; |
2031 |
} |
} |
2032 |
|
|
2033 |
for (k = 0; k < pEnc->num_threads; k++) { |
for (k = 1; k < pEnc->num_threads; k++) { |
2034 |
pthread_create(&pEnc->motionData[k].handle, NULL, |
pthread_create(&pEnc->motionData[k].handle, NULL, |
2035 |
(void*)SMPMotionEstimationBVOP, (void*)&pEnc->motionData[k]); |
(void*)SMPMotionEstimationBVOP, (void*)&pEnc->motionData[k]); |
2036 |
} |
} |
2037 |
|
|
2038 |
|
SMPMotionEstimationBVOP(&pEnc->motionData[0]); |
2039 |
|
|
2040 |
|
for (k = 1; k < pEnc->num_threads; k++) { |
2041 |
|
pthread_join(pEnc->motionData[k].handle, &status); |
2042 |
|
} |
2043 |
|
|
2044 |
|
frame->fcode = frame->bcode = 0; |
2045 |
|
for (k = 0; k < pEnc->num_threads; k++) { |
2046 |
|
if (pEnc->motionData[k].minfcode > frame->fcode) |
2047 |
|
frame->fcode = pEnc->motionData[k].minfcode; |
2048 |
|
if (pEnc->motionData[k].minbcode > frame->bcode) |
2049 |
|
frame->bcode = pEnc->motionData[k].minbcode; |
2050 |
|
} |
2051 |
} else { |
} else { |
|
start_timer(); |
|
2052 |
MotionEstimationBVOP(&pEnc->mbParam, frame, |
MotionEstimationBVOP(&pEnc->mbParam, frame, |
2053 |
((int32_t)(pEnc->current->stamp - frame->stamp)), /* time_bp */ |
((int32_t)(pEnc->current->stamp - frame->stamp)), /* time_bp */ |
2054 |
((int32_t)(pEnc->current->stamp - pEnc->reference->stamp)), /* time_pp */ |
((int32_t)(pEnc->current->stamp - pEnc->reference->stamp)), /* time_pp */ |
2056 |
&pEnc->f_refh, &pEnc->f_refv, &pEnc->f_refhv, |
&pEnc->f_refh, &pEnc->f_refv, &pEnc->f_refhv, |
2057 |
pEnc->current, b_ref, &pEnc->vInterH, |
pEnc->current, b_ref, &pEnc->vInterH, |
2058 |
&pEnc->vInterV, &pEnc->vInterHV); |
&pEnc->vInterV, &pEnc->vInterHV); |
|
stop_motion_timer(); |
|
2059 |
} |
} |
2060 |
|
stop_motion_timer(); |
2061 |
|
|
2062 |
set_timecodes(frame, pEnc->reference,pEnc->mbParam.fbase); |
set_timecodes(frame, pEnc->reference,pEnc->mbParam.fbase); |
2063 |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, frame, 1, frame->quant); |
BitstreamWriteVopHeader(bs, &pEnc->mbParam, frame, 1, frame->quant); |
2071 |
frame->sStat.kblks = frame->sStat.ublks = 0; |
frame->sStat.kblks = frame->sStat.ublks = 0; |
2072 |
|
|
2073 |
for (y = 0; y < pEnc->mbParam.mb_height; y++) { |
for (y = 0; y < pEnc->mbParam.mb_height; y++) { |
|
|
|
|
if (pEnc->num_threads > 0) { |
|
|
/* in multithreaded encoding, only proceed with a row of macroblocks |
|
|
if ME finished with that row */ |
|
|
while (pEnc->motionData[y%pEnc->num_threads].complete_count_self[y/pEnc->num_threads] != pEnc->mbParam.mb_width) |
|
|
sched_yield(); |
|
|
} |
|
|
|
|
2074 |
for (x = 0; x < pEnc->mbParam.mb_width; x++) { |
for (x = 0; x < pEnc->mbParam.mb_width; x++) { |
2075 |
MACROBLOCK * const mb = &frame->mbs[x + y * pEnc->mbParam.mb_width]; |
MACROBLOCK * const mb = &frame->mbs[x + y * pEnc->mbParam.mb_width]; |
2076 |
|
|
2114 |
stop_coding_timer(); |
stop_coding_timer(); |
2115 |
} |
} |
2116 |
} |
} |
|
|
|
|
if (pEnc->num_threads > 0) { |
|
|
void * status; |
|
|
int k; |
|
|
for (k = 0; k < pEnc->num_threads; k++) { |
|
|
pthread_join(pEnc->motionData[k].handle, &status); |
|
|
} |
|
|
} |
|
|
|
|
2117 |
emms(); |
emms(); |
2118 |
|
|
2119 |
BitstreamPadAlways(bs); /* next_start_code() at the end of VideoObjectPlane() */ |
BitstreamPadAlways(bs); /* next_start_code() at the end of VideoObjectPlane() */ |