void vp8cx_init_mbrthread_data(VP8_COMP *cpi,
MACROBLOCK *x,
MB_ROW_COMP *mbr_ei,
int mb_row,
int count
)
{
VP8_COMMON *const cm = & cpi->common;
MACROBLOCKD *const xd = & x->e_mbd;
int i;
(void) mb_row;
for (i = 0; i < count; i++)
{
MACROBLOCK *mb = & mbr_ei[i].mb;
MACROBLOCKD *mbd = &mb->e_mbd;
mbd->subpixel_predict = xd->subpixel_predict;
mbd->subpixel_predict8x4 = xd->subpixel_predict8x4;
mbd->subpixel_predict8x8 = xd->subpixel_predict8x8;
mbd->subpixel_predict16x16 = xd->subpixel_predict16x16;
mb->gf_active_ptr = x->gf_active_ptr;
vpx_memset(mbr_ei[i].segment_counts, 0, sizeof(mbr_ei[i].segment_counts));
mbr_ei[i].totalrate = 0;
mb->partition_info = x->pi + x->e_mbd.mode_info_stride * (i + 1);
mbd->mode_info_context = cm->mi + x->e_mbd.mode_info_stride * (i + 1);
mbd->mode_info_stride = cm->mode_info_stride;
mbd->frame_type = cm->frame_type;
mb->src = * cpi->Source;
mbd->pre = cm->yv12_fb[cm->lst_fb_idx];
mbd->dst = cm->yv12_fb[cm->new_fb_idx];
mb->src.y_buffer += 16 * x->src.y_stride * (i + 1);
mb->src.u_buffer += 8 * x->src.uv_stride * (i + 1);
mb->src.v_buffer += 8 * x->src.uv_stride * (i + 1);
vp8_build_block_offsets(mb);
vp8_setup_block_dptrs(mbd);
vp8_setup_block_ptrs(mb);
mbd->left_context = &cm->left_context;
mb->mvc = cm->fc.mvc;
setup_mbby_copy(&mbr_ei[i].mb, x);
mbd->fullpixel_mask = 0xffffffff;
if(cm->full_pixel)
mbd->fullpixel_mask = 0xfffffff8;
}
}
int vp8cx_create_encoder_threads(VP8_COMP *cpi)
{
const VP8_COMMON * cm = &cpi->common;
cpi->b_multi_threaded = 0;
cpi->encoding_thread_count = 0;
cpi->b_lpf_running = 0;
if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
{
int ithread;
int th_count = cpi->oxcf.multi_threaded - 1;
int rc = 0;
/* don't allocate more threads than cores available */
if (cpi->oxcf.multi_threaded > cm->processor_core_count)
th_count = cm->processor_core_count - 1;
/* we have th_count + 1 (main) threads processing one row each */
/* no point to have more threads than the sync range allows */
if(th_count > ((cm->mb_cols / cpi->mt_sync_range) - 1))
{
th_count = (cm->mb_cols / cpi->mt_sync_range) - 1;
}
if(th_count == 0)
return 0;
CHECK_MEM_ERROR(cpi->h_encoding_thread,
vpx_malloc(sizeof(pthread_t) * th_count));
CHECK_MEM_ERROR(cpi->h_event_start_encoding,
vpx_malloc(sizeof(sem_t) * th_count));
CHECK_MEM_ERROR(cpi->mb_row_ei,
vpx_memalign(32, sizeof(MB_ROW_COMP) * th_count));
vpx_memset(cpi->mb_row_ei, 0, sizeof(MB_ROW_COMP) * th_count);
CHECK_MEM_ERROR(cpi->en_thread_data,
vpx_malloc(sizeof(ENCODETHREAD_DATA) * th_count));
sem_init(&cpi->h_event_end_encoding, 0, 0);
cpi->b_multi_threaded = 1;
cpi->encoding_thread_count = th_count;
/*
printf("[VP8:] multi_threaded encoding is enabled with %d threads\n\n",
(cpi->encoding_thread_count +1));
*/
for (ithread = 0; ithread < th_count; ithread++)
{
ENCODETHREAD_DATA *ethd = &cpi->en_thread_data[ithread];
/* Setup block ptrs and offsets */
vp8_setup_block_ptrs(&cpi->mb_row_ei[ithread].mb);
vp8_setup_block_dptrs(&cpi->mb_row_ei[ithread].mb.e_mbd);
sem_init(&cpi->h_event_start_encoding[ithread], 0, 0);
ethd->ithread = ithread;
ethd->ptr1 = (void *)cpi;
ethd->ptr2 = (void *)&cpi->mb_row_ei[ithread];
rc = pthread_create(&cpi->h_encoding_thread[ithread], 0,
thread_encoding_proc, ethd);
if(rc)
break;
}
if(rc)
{
/* shutdown other threads */
cpi->b_multi_threaded = 0;
for(--ithread; ithread >= 0; ithread--)
{
pthread_join(cpi->h_encoding_thread[ithread], 0);
sem_destroy(&cpi->h_event_start_encoding[ithread]);
}
sem_destroy(&cpi->h_event_end_encoding);
/* free thread related resources */
vpx_free(cpi->h_event_start_encoding);
vpx_free(cpi->h_encoding_thread);
vpx_free(cpi->mb_row_ei);
vpx_free(cpi->en_thread_data);
return -1;
}
{
LPFTHREAD_DATA * lpfthd = &cpi->lpf_thread_data;
sem_init(&cpi->h_event_start_lpf, 0, 0);
sem_init(&cpi->h_event_end_lpf, 0, 0);
lpfthd->ptr1 = (void *)cpi;
rc = pthread_create(&cpi->h_filter_thread, 0, thread_loopfilter,
lpfthd);
if(rc)
{
/* shutdown other threads */
cpi->b_multi_threaded = 0;
for(--ithread; ithread >= 0; ithread--)
{
sem_post(&cpi->h_event_start_encoding[ithread]);
pthread_join(cpi->h_encoding_thread[ithread], 0);
sem_destroy(&cpi->h_event_start_encoding[ithread]);
}
sem_destroy(&cpi->h_event_end_encoding);
sem_destroy(&cpi->h_event_end_lpf);
sem_destroy(&cpi->h_event_start_lpf);
/* free thread related resources */
vpx_free(cpi->h_event_start_encoding);
vpx_free(cpi->h_encoding_thread);
vpx_free(cpi->mb_row_ei);
vpx_free(cpi->en_thread_data);
return -2;
}
}
}
return 0;
}
void vp8cx_init_mbrthread_data(VP8_COMP *cpi,
MACROBLOCK *x,
MB_ROW_COMP *mbr_ei,
int mb_row,
int count
)
{
VP8_COMMON *const cm = & cpi->common;
MACROBLOCKD *const xd = & x->e_mbd;
int i;
(void) mb_row;
for (i = 0; i < count; i++)
{
MACROBLOCK *mb = & mbr_ei[i].mb;
MACROBLOCKD *mbd = &mb->e_mbd;
mbd->subpixel_predict = xd->subpixel_predict;
mbd->subpixel_predict8x4 = xd->subpixel_predict8x4;
mbd->subpixel_predict8x8 = xd->subpixel_predict8x8;
mbd->subpixel_predict16x16 = xd->subpixel_predict16x16;
#if CONFIG_RUNTIME_CPU_DETECT
mbd->rtcd = xd->rtcd;
#endif
mb->gf_active_ptr = x->gf_active_ptr;
mb->vector_range = 32;
vpx_memset(mbr_ei[i].segment_counts, 0, sizeof(mbr_ei[i].segment_counts));
mbr_ei[i].totalrate = 0;
mb->partition_info = x->pi + x->e_mbd.mode_info_stride * (i + 1);
mbd->mode_info_context = cm->mi + x->e_mbd.mode_info_stride * (i + 1);
mbd->mode_info_stride = cm->mode_info_stride;
mbd->frame_type = cm->frame_type;
mbd->frames_since_golden = cm->frames_since_golden;
mbd->frames_till_alt_ref_frame = cm->frames_till_alt_ref_frame;
mb->src = * cpi->Source;
mbd->pre = cm->yv12_fb[cm->lst_fb_idx];
mbd->dst = cm->yv12_fb[cm->new_fb_idx];
mb->src.y_buffer += 16 * x->src.y_stride * (i + 1);
mb->src.u_buffer += 8 * x->src.uv_stride * (i + 1);
mb->src.v_buffer += 8 * x->src.uv_stride * (i + 1);
vp8_build_block_offsets(mb);
vp8_setup_block_dptrs(mbd);
vp8_setup_block_ptrs(mb);
mb->rddiv = cpi->RDDIV;
mb->rdmult = cpi->RDMULT;
mbd->left_context = &cm->left_context;
mb->mvc = cm->fc.mvc;
setup_mbby_copy(&mbr_ei[i].mb, x);
}
}
