Merge "Delay auto key frame insertion in realtime configuration"
diff --git a/vp8/common/loopfilter.c b/vp8/common/loopfilter.c
index f9d0823..90e1d9c 100644
--- a/vp8/common/loopfilter.c
+++ b/vp8/common/loopfilter.c
@@ -263,43 +263,44 @@
}
-void vp8_adjust_mb_lf_value(MACROBLOCKD *mbd, int *filter_level)
+int vp8_adjust_mb_lf_value(MACROBLOCKD *mbd, int filter_level)
{
MB_MODE_INFO *mbmi = &mbd->mode_info_context->mbmi;
if (mbd->mode_ref_lf_delta_enabled)
{
/* Apply delta for reference frame */
- *filter_level += mbd->ref_lf_deltas[mbmi->ref_frame];
+ filter_level += mbd->ref_lf_deltas[mbmi->ref_frame];
/* Apply delta for mode */
if (mbmi->ref_frame == INTRA_FRAME)
{
/* Only the split mode BPRED has a further special case */
if (mbmi->mode == B_PRED)
- *filter_level += mbd->mode_lf_deltas[0];
+ filter_level += mbd->mode_lf_deltas[0];
}
else
{
/* Zero motion mode */
if (mbmi->mode == ZEROMV)
- *filter_level += mbd->mode_lf_deltas[1];
+ filter_level += mbd->mode_lf_deltas[1];
/* Split MB motion mode */
else if (mbmi->mode == SPLITMV)
- *filter_level += mbd->mode_lf_deltas[3];
+ filter_level += mbd->mode_lf_deltas[3];
/* All other inter motion modes (Nearest, Near, New) */
else
- *filter_level += mbd->mode_lf_deltas[2];
+ filter_level += mbd->mode_lf_deltas[2];
}
/* Range check */
- if (*filter_level > MAX_LOOP_FILTER)
- *filter_level = MAX_LOOP_FILTER;
- else if (*filter_level < 0)
- *filter_level = 0;
+ if (filter_level > MAX_LOOP_FILTER)
+ filter_level = MAX_LOOP_FILTER;
+ else if (filter_level < 0)
+ filter_level = 0;
}
+ return filter_level;
}
@@ -373,7 +374,7 @@
* These specified to 8th pel as they are always compared to values that are in 1/8th pel units
* Apply any context driven MB level adjustment
*/
- vp8_adjust_mb_lf_value(mbd, &filter_level);
+ filter_level = vp8_adjust_mb_lf_value(mbd, filter_level);
if (filter_level)
{
@@ -473,7 +474,7 @@
filter_level = baseline_filter_level[Segment];
/* Apply any context driven MB level adjustment */
- vp8_adjust_mb_lf_value(mbd, &filter_level);
+ filter_level = vp8_adjust_mb_lf_value(mbd, filter_level);
if (filter_level)
{
diff --git a/vp8/common/onyxc_int.h b/vp8/common/onyxc_int.h
index e011ec9..c8c2277 100644
--- a/vp8/common/onyxc_int.h
+++ b/vp8/common/onyxc_int.h
@@ -200,7 +200,7 @@
} VP8_COMMON;
-void vp8_adjust_mb_lf_value(MACROBLOCKD *mbd, int *filter_level);
+int vp8_adjust_mb_lf_value(MACROBLOCKD *mbd, int filter_level);
void vp8_init_loop_filter(VP8_COMMON *cm);
void vp8_frame_init_loop_filter(loop_filter_info *lfi, int frame_type);
extern void vp8_loop_filter_frame(VP8_COMMON *cm, MACROBLOCKD *mbd, int filt_val);
diff --git a/vp8/decoder/threading.c b/vp8/decoder/threading.c
index 930d9b9..ec2cb2b 100644
--- a/vp8/decoder/threading.c
+++ b/vp8/decoder/threading.c
@@ -320,7 +320,7 @@
* These are specified to 8th pel as they are always compared to values that are in 1/8th pel units
* Apply any context driven MB level adjustment
*/
- vp8_adjust_mb_lf_value(xd, &filter_level);
+ filter_level = vp8_adjust_mb_lf_value(xd, filter_level);
}
/* Distance of Mb to the various image edges.
@@ -863,7 +863,7 @@
* These are specified to 8th pel as they are always compared to values that are in 1/8th pel units
* Apply any context driven MB level adjustment
*/
- vp8_adjust_mb_lf_value(xd, &filter_level);
+ filter_level = vp8_adjust_mb_lf_value(xd, filter_level);
}
/* Distance of Mb to the various image edges.
diff --git a/vp8/encoder/encodeframe.c b/vp8/encoder/encodeframe.c
index 08fb6e5..d535894 100644
--- a/vp8/encoder/encodeframe.c
+++ b/vp8/encoder/encodeframe.c
@@ -487,6 +487,16 @@
int recon_uv_stride = cm->yv12_fb[ref_fb_idx].uv_stride;
int seg_map_index = (mb_row * cpi->common.mb_cols);
+#if CONFIG_MULTITHREAD
+ const int nsync = cpi->mt_sync_range;
+ const int rightmost_col = cm->mb_cols - 1;
+ volatile const int *last_row_current_mb_col;
+
+ if ((cpi->b_multi_threaded != 0) && (mb_row != 0))
+ last_row_current_mb_col = &cpi->mt_current_mb_col[mb_row - 1];
+ else
+ last_row_current_mb_col = &rightmost_col;
+#endif
// reset above block coeffs
xd->above_context = cm->above_context;
@@ -532,6 +542,21 @@
x->rddiv = cpi->RDDIV;
x->rdmult = cpi->RDMULT;
+#if CONFIG_MULTITHREAD
+ if ((cpi->b_multi_threaded != 0) && (mb_row != 0))
+ {
+ if ((mb_col & (nsync - 1)) == 0)
+ {
+ while (mb_col > (*last_row_current_mb_col - nsync)
+ && (*last_row_current_mb_col) != (cm->mb_cols - 1))
+ {
+ x86_pause_hint();
+ thread_sleep(0);
+ }
+ }
+ }
+#endif
+
if(cpi->oxcf.tuning == VP8_TUNE_SSIM)
activity_sum += vp8_activity_masking(cpi, x);
@@ -628,7 +653,12 @@
x->partition_info++;
xd->above_context++;
- cpi->current_mb_col_main = mb_col;
+#if CONFIG_MULTITHREAD
+ if (cpi->b_multi_threaded != 0)
+ {
+ cpi->mt_current_mb_col[mb_row] = mb_col;
+ }
+#endif
}
//extend the recon for intra prediction
@@ -642,12 +672,15 @@
xd->mode_info_context++;
x->partition_info++;
x->activity_sum += activity_sum;
+
+#if CONFIG_MULTITHREAD
+ if ((cpi->b_multi_threaded != 0) && (mb_row == cm->mb_rows - 1))
+ {
+ sem_post(&cpi->h_event_end_encoding); /* signal frame encoding end */
+ }
+#endif
}
-
-
-
-
void vp8_encode_frame(VP8_COMP *cpi)
{
int mb_row;
@@ -800,22 +833,16 @@
vp8cx_init_mbrthread_data(cpi, x, cpi->mb_row_ei, 1, cpi->encoding_thread_count);
+ for (i = 0; i < cm->mb_rows; i++)
+ cpi->mt_current_mb_col[i] = 0;
+
+ for (i = 0; i < cpi->encoding_thread_count; i++)
+ {
+ sem_post(&cpi->h_event_start_encoding[i]);
+ }
+
for (mb_row = 0; mb_row < cm->mb_rows; mb_row += (cpi->encoding_thread_count + 1))
{
- cpi->current_mb_col_main = -1;
-
- for (i = 0; i < cpi->encoding_thread_count; i++)
- {
- if ((mb_row + i + 1) >= cm->mb_rows)
- break;
-
- cpi->mb_row_ei[i].mb_row = mb_row + i + 1;
- cpi->mb_row_ei[i].tp = cpi->tok + (mb_row + i + 1) * (cm->mb_cols * 16 * 24);
- cpi->mb_row_ei[i].current_mb_col = -1;
- //SetEvent(cpi->h_event_mbrencoding[i]);
- sem_post(&cpi->h_event_mbrencoding[i]);
- }
-
vp8_zero(cm->left_context)
tp = cpi->tok + mb_row * (cm->mb_cols * 16 * 24);
@@ -830,26 +857,10 @@
xd->mode_info_context += xd->mode_info_stride * cpi->encoding_thread_count;
x->partition_info += xd->mode_info_stride * cpi->encoding_thread_count;
- if (mb_row < cm->mb_rows - 1)
- //WaitForSingleObject(cpi->h_event_main, INFINITE);
- sem_wait(&cpi->h_event_main);
}
- /*
- for( ;mb_row<cm->mb_rows; mb_row ++)
- {
- vp8_zero( cm->left_context)
+ sem_wait(&cpi->h_event_end_encoding); /* wait for other threads to finish */
- tp = cpi->tok + mb_row * (cm->mb_cols * 16 * 24);
-
- encode_mb_row(cpi, cm, mb_row, x, xd, &tp, segment_counts, &totalrate);
- // adjust to the next row of mbs
- x->src.y_buffer += 16 * x->src.y_stride - 16 * cm->mb_cols;
- x->src.u_buffer += 8 * x->src.uv_stride - 8 * cm->mb_cols;
- x->src.v_buffer += 8 * x->src.uv_stride - 8 * cm->mb_cols;
-
- }
- */
cpi->tok_count = 0;
for (mb_row = 0; mb_row < cm->mb_rows; mb_row ++)
@@ -859,7 +870,6 @@
if (xd->segmentation_enabled)
{
-
int i, j;
if (xd->segmentation_enabled)
@@ -871,7 +881,6 @@
segment_counts[j] += cpi->mb_row_ei[i].segment_counts[j];
}
}
-
}
for (i = 0; i < cpi->encoding_thread_count; i++)
diff --git a/vp8/encoder/ethreading.c b/vp8/encoder/ethreading.c
index 7b5cfe0..8da93a6 100644
--- a/vp8/encoder/ethreading.c
+++ b/vp8/encoder/ethreading.c
@@ -8,15 +8,16 @@
* be found in the AUTHORS file in the root of the source tree.
*/
-
#include "onyx_int.h"
#include "threading.h"
#include "common.h"
#include "extend.h"
-
-extern int vp8cx_encode_inter_macroblock(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t, int recon_yoffset, int recon_uvoffset);
-extern int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x, TOKENEXTRA **t);
+extern int vp8cx_encode_inter_macroblock(VP8_COMP *cpi, MACROBLOCK *x,
+ TOKENEXTRA **t, int recon_yoffset,
+ int recon_uvoffset);
+extern int vp8cx_encode_intra_macro_block(VP8_COMP *cpi, MACROBLOCK *x,
+ TOKENEXTRA **t);
extern void vp8cx_mb_init_quantizer(VP8_COMP *cpi, MACROBLOCK *x);
extern void vp8_build_block_offsets(MACROBLOCK *x);
extern void vp8_setup_block_ptrs(MACROBLOCK *x);
@@ -26,10 +27,11 @@
{
#if CONFIG_MULTITHREAD
int ithread = ((ENCODETHREAD_DATA *)p_data)->ithread;
- VP8_COMP *cpi = (VP8_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr1);
+ VP8_COMP *cpi = (VP8_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr1);
MB_ROW_COMP *mbri = (MB_ROW_COMP *)(((ENCODETHREAD_DATA *)p_data)->ptr2);
ENTROPY_CONTEXT_PLANES mb_row_left_context;
+ const int nsync = cpi->mt_sync_range;
//printf("Started thread %d\n", ithread);
while (1)
@@ -38,210 +40,209 @@
break;
//if(WaitForSingleObject(cpi->h_event_mbrencoding[ithread], INFINITE) == WAIT_OBJECT_0)
- if (sem_wait(&cpi->h_event_mbrencoding[ithread]) == 0)
+ if (sem_wait(&cpi->h_event_start_encoding[ithread]) == 0)
{
+ VP8_COMMON *cm = &cpi->common;
+ int mb_row;
+ MACROBLOCK *x = &mbri->mb;
+ MACROBLOCKD *xd = &x->e_mbd;
+ TOKENEXTRA *tp ;
+
+ int *segment_counts = mbri->segment_counts;
+ int *totalrate = &mbri->totalrate;
+
if (cpi->b_multi_threaded == FALSE) // we're shutting down
break;
- else
+
+ for (mb_row = ithread + 1; mb_row < cm->mb_rows; mb_row += (cpi->encoding_thread_count + 1))
{
- VP8_COMMON *cm = &cpi->common;
- int mb_row = mbri->mb_row;
- MACROBLOCK *x = &mbri->mb;
- MACROBLOCKD *xd = &x->e_mbd;
- TOKENEXTRA **tp = &mbri->tp;
- int *segment_counts = mbri->segment_counts;
- int *totalrate = &mbri->totalrate;
+ int i;
+ int recon_yoffset, recon_uvoffset;
+ int mb_col;
+ int ref_fb_idx = cm->lst_fb_idx;
+ int dst_fb_idx = cm->new_fb_idx;
+ int recon_y_stride = cm->yv12_fb[ref_fb_idx].y_stride;
+ int recon_uv_stride = cm->yv12_fb[ref_fb_idx].uv_stride;
+ volatile int *last_row_current_mb_col;
+ INT64 activity_sum = 0;
+
+ tp = cpi->tok + (mb_row * (cm->mb_cols * 16 * 24));
+
+ last_row_current_mb_col = &cpi->mt_current_mb_col[mb_row - 1];
+
+ // reset above block coeffs
+ xd->above_context = cm->above_context;
+ xd->left_context = &mb_row_left_context;
+
+ vp8_zero(mb_row_left_context);
+
+ xd->up_available = (mb_row != 0);
+ recon_yoffset = (mb_row * recon_y_stride * 16);
+ recon_uvoffset = (mb_row * recon_uv_stride * 8);
+
+ cpi->tplist[mb_row].start = tp;
+
+ //printf("Thread mb_row = %d\n", mb_row);
+
+ // for each macroblock col in image
+ for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
{
- int i;
- int recon_yoffset, recon_uvoffset;
- int mb_col;
- int ref_fb_idx = cm->lst_fb_idx;
- int dst_fb_idx = cm->new_fb_idx;
- int recon_y_stride = cm->yv12_fb[ref_fb_idx].y_stride;
- int recon_uv_stride = cm->yv12_fb[ref_fb_idx].uv_stride;
- volatile int *last_row_current_mb_col;
- INT64 activity_sum = 0;
+ int seg_map_index = (mb_row * cm->mb_cols);
- if (ithread > 0)
- last_row_current_mb_col = &cpi->mb_row_ei[ithread-1].current_mb_col;
- else
- last_row_current_mb_col = &cpi->current_mb_col_main;
-
- // reset above block coeffs
- xd->above_context = cm->above_context;
- xd->left_context = &mb_row_left_context;
-
- vp8_zero(mb_row_left_context);
-
- xd->up_available = (mb_row != 0);
- recon_yoffset = (mb_row * recon_y_stride * 16);
- recon_uvoffset = (mb_row * recon_uv_stride * 8);
-
-
- cpi->tplist[mb_row].start = *tp;
-
- //printf("Thread mb_row = %d\n", mb_row);
-
- // for each macroblock col in image
- for (mb_col = 0; mb_col < cm->mb_cols; mb_col++)
+ if ((mb_col & (nsync - 1)) == 0)
{
- int seg_map_index = (mb_row * cm->mb_cols);
-
- while (mb_col > (*last_row_current_mb_col - 1) && *last_row_current_mb_col != cm->mb_cols - 1)
+ while (mb_col > (*last_row_current_mb_col - nsync) && *last_row_current_mb_col != cm->mb_cols - 1)
{
x86_pause_hint();
thread_sleep(0);
}
-
- // Distance of Mb to the various image edges.
- // These specified to 8th pel as they are always compared to values that are in 1/8th pel units
- xd->mb_to_left_edge = -((mb_col * 16) << 3);
- xd->mb_to_right_edge = ((cm->mb_cols - 1 - mb_col) * 16) << 3;
- xd->mb_to_top_edge = -((mb_row * 16) << 3);
- xd->mb_to_bottom_edge = ((cm->mb_rows - 1 - mb_row) * 16) << 3;
-
- // Set up limit values for motion vectors used to prevent them extending outside the UMV borders
- x->mv_col_min = -((mb_col * 16) + (VP8BORDERINPIXELS - 16));
- x->mv_col_max = ((cm->mb_cols - 1 - mb_col) * 16) + (VP8BORDERINPIXELS - 16);
- x->mv_row_min = -((mb_row * 16) + (VP8BORDERINPIXELS - 16));
- x->mv_row_max = ((cm->mb_rows - 1 - mb_row) * 16) + (VP8BORDERINPIXELS - 16);
-
- xd->dst.y_buffer = cm->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset;
- xd->dst.u_buffer = cm->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset;
- xd->dst.v_buffer = cm->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset;
- xd->left_available = (mb_col != 0);
-
- x->rddiv = cpi->RDDIV;
- x->rdmult = cpi->RDMULT;
-
- if(cpi->oxcf.tuning == VP8_TUNE_SSIM)
- activity_sum += vp8_activity_masking(cpi, x);
-
- // Is segmentation enabled
- // MB level adjutment to quantizer
- if (xd->segmentation_enabled)
- {
- // Code to set segment id in xd->mbmi.segment_id for current MB (with range checking)
- if (cpi->segmentation_map[seg_map_index+mb_col] <= 3)
- xd->mode_info_context->mbmi.segment_id = cpi->segmentation_map[seg_map_index+mb_col];
- else
- xd->mode_info_context->mbmi.segment_id = 0;
-
- vp8cx_mb_init_quantizer(cpi, x);
- }
- else
- xd->mode_info_context->mbmi.segment_id = 0; // Set to Segment 0 by default
-
- x->active_ptr = cpi->active_map + seg_map_index + mb_col;
-
- if (cm->frame_type == KEY_FRAME)
- {
- *totalrate += vp8cx_encode_intra_macro_block(cpi, x, tp);
-#ifdef MODE_STATS
- y_modes[xd->mbmi.mode] ++;
-#endif
- }
- else
- {
- *totalrate += vp8cx_encode_inter_macroblock(cpi, x, tp, recon_yoffset, recon_uvoffset);
-
-#ifdef MODE_STATS
- inter_y_modes[xd->mbmi.mode] ++;
-
- if (xd->mbmi.mode == SPLITMV)
- {
- int b;
-
- for (b = 0; b < xd->mbmi.partition_count; b++)
- {
- inter_b_modes[x->partition->bmi[b].mode] ++;
- }
- }
-
-#endif
-
- // Count of last ref frame 0,0 useage
- if ((xd->mode_info_context->mbmi.mode == ZEROMV) && (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME))
- cpi->inter_zz_count ++;
-
- // Special case code for cyclic refresh
- // If cyclic update enabled then copy xd->mbmi.segment_id; (which may have been updated based on mode
- // during vp8cx_encode_inter_macroblock()) back into the global sgmentation map
- if (cpi->cyclic_refresh_mode_enabled && xd->segmentation_enabled)
- {
- cpi->segmentation_map[seg_map_index+mb_col] = xd->mode_info_context->mbmi.segment_id;
-
- // If the block has been refreshed mark it as clean (the magnitude of the -ve influences how long it will be before we consider another refresh):
- // Else if it was coded (last frame 0,0) and has not already been refreshed then mark it as a candidate for cleanup next time (marked 0)
- // else mark it as dirty (1).
- if (xd->mode_info_context->mbmi.segment_id)
- cpi->cyclic_refresh_map[seg_map_index+mb_col] = -1;
- else if ((xd->mode_info_context->mbmi.mode == ZEROMV) && (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME))
- {
- if (cpi->cyclic_refresh_map[seg_map_index+mb_col] == 1)
- cpi->cyclic_refresh_map[seg_map_index+mb_col] = 0;
- }
- else
- cpi->cyclic_refresh_map[seg_map_index+mb_col] = 1;
-
- }
- }
- cpi->tplist[mb_row].stop = *tp;
-
- x->gf_active_ptr++; // Increment pointer into gf useage flags structure for next mb
-
- for (i = 0; i < 16; i++)
- vpx_memcpy(&xd->mode_info_context->bmi[i], &xd->block[i].bmi, sizeof(xd->block[i].bmi));
-
- // adjust to the next column of macroblocks
- x->src.y_buffer += 16;
- x->src.u_buffer += 8;
- x->src.v_buffer += 8;
-
- recon_yoffset += 16;
- recon_uvoffset += 8;
-
- // Keep track of segment useage
- segment_counts[xd->mode_info_context->mbmi.segment_id] ++;
-
- // skip to next mb
- xd->mode_info_context++;
- x->partition_info++;
-
- xd->above_context++;
-
- cpi->mb_row_ei[ithread].current_mb_col = mb_col;
-
}
- //extend the recon for intra prediction
- vp8_extend_mb_row(
- &cm->yv12_fb[dst_fb_idx],
- xd->dst.y_buffer + 16,
- xd->dst.u_buffer + 8,
- xd->dst.v_buffer + 8);
+ // Distance of Mb to the various image edges.
+ // These specified to 8th pel as they are always compared to values that are in 1/8th pel units
+ xd->mb_to_left_edge = -((mb_col * 16) << 3);
+ xd->mb_to_right_edge = ((cm->mb_cols - 1 - mb_col) * 16) << 3;
+ xd->mb_to_top_edge = -((mb_row * 16) << 3);
+ xd->mb_to_bottom_edge = ((cm->mb_rows - 1 - mb_row) * 16) << 3;
- // this is to account for the border
+ // Set up limit values for motion vectors used to prevent them extending outside the UMV borders
+ x->mv_col_min = -((mb_col * 16) + (VP8BORDERINPIXELS - 16));
+ x->mv_col_max = ((cm->mb_cols - 1 - mb_col) * 16) + (VP8BORDERINPIXELS - 16);
+ x->mv_row_min = -((mb_row * 16) + (VP8BORDERINPIXELS - 16));
+ x->mv_row_max = ((cm->mb_rows - 1 - mb_row) * 16) + (VP8BORDERINPIXELS - 16);
+
+ xd->dst.y_buffer = cm->yv12_fb[dst_fb_idx].y_buffer + recon_yoffset;
+ xd->dst.u_buffer = cm->yv12_fb[dst_fb_idx].u_buffer + recon_uvoffset;
+ xd->dst.v_buffer = cm->yv12_fb[dst_fb_idx].v_buffer + recon_uvoffset;
+ xd->left_available = (mb_col != 0);
+
+ x->rddiv = cpi->RDDIV;
+ x->rdmult = cpi->RDMULT;
+
+ if (cpi->oxcf.tuning == VP8_TUNE_SSIM)
+ activity_sum += vp8_activity_masking(cpi, x);
+
+ // Is segmentation enabled
+ // MB level adjutment to quantizer
+ if (xd->segmentation_enabled)
+ {
+ // Code to set segment id in xd->mbmi.segment_id for current MB (with range checking)
+ if (cpi->segmentation_map[seg_map_index + mb_col] <= 3)
+ xd->mode_info_context->mbmi.segment_id = cpi->segmentation_map[seg_map_index + mb_col];
+ else
+ xd->mode_info_context->mbmi.segment_id = 0;
+
+ vp8cx_mb_init_quantizer(cpi, x);
+ }
+ else
+ xd->mode_info_context->mbmi.segment_id = 0; // Set to Segment 0 by default
+
+ x->active_ptr = cpi->active_map + seg_map_index + mb_col;
+
+ if (cm->frame_type == KEY_FRAME)
+ {
+ *totalrate += vp8cx_encode_intra_macro_block(cpi, x, &tp);
+#ifdef MODE_STATS
+ y_modes[xd->mbmi.mode] ++;
+#endif
+ }
+ else
+ {
+ *totalrate += vp8cx_encode_inter_macroblock(cpi, x, &tp, recon_yoffset, recon_uvoffset);
+
+#ifdef MODE_STATS
+ inter_y_modes[xd->mbmi.mode] ++;
+
+ if (xd->mbmi.mode == SPLITMV)
+ {
+ int b;
+
+ for (b = 0; b < xd->mbmi.partition_count; b++)
+ {
+ inter_b_modes[x->partition->bmi[b].mode] ++;
+ }
+ }
+
+#endif
+
+ // Count of last ref frame 0,0 useage
+ if ((xd->mode_info_context->mbmi.mode == ZEROMV) && (xd->mode_info_context->mbmi.ref_frame == LAST_FRAME))
+ cpi->inter_zz_count++;
+
+ // Special case code for cyclic refresh
+ // If cyclic update enabled then copy xd->mbmi.segment_id; (which may have been updated based on mode
+ // during vp8cx_encode_inter_macroblock()) back into the global sgmentation map
+ if (cpi->cyclic_refresh_mode_enabled && xd->segmentation_enabled)
+ {
+ const MB_MODE_INFO * mbmi = &xd->mode_info_context->mbmi;
+ cpi->segmentation_map[seg_map_index + mb_col] = mbmi->segment_id;
+
+ // If the block has been refreshed mark it as clean (the magnitude of the -ve influences how long it will be before we consider another refresh):
+ // Else if it was coded (last frame 0,0) and has not already been refreshed then mark it as a candidate for cleanup next time (marked 0)
+ // else mark it as dirty (1).
+ if (mbmi->segment_id)
+ cpi->cyclic_refresh_map[seg_map_index + mb_col] = -1;
+ else if ((mbmi->mode == ZEROMV) && (mbmi->ref_frame == LAST_FRAME))
+ {
+ if (cpi->cyclic_refresh_map[seg_map_index + mb_col] == 1)
+ cpi->cyclic_refresh_map[seg_map_index + mb_col] = 0;
+ }
+ else
+ cpi->cyclic_refresh_map[seg_map_index + mb_col] = 1;
+
+ }
+ }
+ cpi->tplist[mb_row].stop = tp;
+
+ x->gf_active_ptr++; // Increment pointer into gf useage flags structure for next mb
+
+ for (i = 0; i < 16; i++)
+ vpx_memcpy(&xd->mode_info_context->bmi[i], &xd->block[i].bmi, sizeof(xd->block[i].bmi));
+
+ // adjust to the next column of macroblocks
+ x->src.y_buffer += 16;
+ x->src.u_buffer += 8;
+ x->src.v_buffer += 8;
+
+ recon_yoffset += 16;
+ recon_uvoffset += 8;
+
+ // Keep track of segment useage
+ segment_counts[xd->mode_info_context->mbmi.segment_id]++;
+
+ // skip to next mb
xd->mode_info_context++;
x->partition_info++;
- x->activity_sum += activity_sum;
+ xd->above_context++;
- x->src.y_buffer += 16 * x->src.y_stride * (cpi->encoding_thread_count + 1) - 16 * cm->mb_cols;
- x->src.u_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
- x->src.v_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
-
- xd->mode_info_context += xd->mode_info_stride * cpi->encoding_thread_count;
- x->partition_info += xd->mode_info_stride * cpi->encoding_thread_count;
-
- if (ithread == (cpi->encoding_thread_count - 1) || mb_row == cm->mb_rows - 1)
- {
- //SetEvent(cpi->h_event_main);
- sem_post(&cpi->h_event_main);
- }
-
+ cpi->mt_current_mb_col[mb_row] = mb_col;
}
+ //extend the recon for intra prediction
+ vp8_extend_mb_row(
+ &cm->yv12_fb[dst_fb_idx],
+ xd->dst.y_buffer + 16,
+ xd->dst.u_buffer + 8,
+ xd->dst.v_buffer + 8);
+
+ // this is to account for the border
+ xd->mode_info_context++;
+ x->partition_info++;
+ x->activity_sum += activity_sum;
+
+ x->src.y_buffer += 16 * x->src.y_stride * (cpi->encoding_thread_count + 1) - 16 * cm->mb_cols;
+ x->src.u_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
+ x->src.v_buffer += 8 * x->src.uv_stride * (cpi->encoding_thread_count + 1) - 8 * cm->mb_cols;
+
+ xd->mode_info_context += xd->mode_info_stride * cpi->encoding_thread_count;
+ x->partition_info += xd->mode_info_stride * cpi->encoding_thread_count;
+
+ if (mb_row == cm->mb_rows - 1)
+ {
+ //SetEvent(cpi->h_event_main);
+ sem_post(&cpi->h_event_end_encoding); /* signal frame encoding end */
+ }
}
}
}
@@ -363,7 +364,6 @@
}
}
-
void vp8cx_init_mbrthread_data(VP8_COMP *cpi,
MACROBLOCK *x,
MB_ROW_COMP *mbr_ei,
@@ -414,7 +414,6 @@
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);
@@ -431,7 +430,6 @@
}
}
-
void vp8cx_create_encoder_threads(VP8_COMP *cpi)
{
cpi->b_multi_threaded = 0;
@@ -451,14 +449,15 @@
else
cpi->encoding_thread_count = cpi->oxcf.multi_threaded - 1;
-
CHECK_MEM_ERROR(cpi->h_encoding_thread, vpx_malloc(sizeof(pthread_t) * cpi->encoding_thread_count));
- CHECK_MEM_ERROR(cpi->h_event_mbrencoding, vpx_malloc(sizeof(sem_t) * cpi->encoding_thread_count));
+ CHECK_MEM_ERROR(cpi->h_event_start_encoding, vpx_malloc(sizeof(sem_t) * cpi->encoding_thread_count));
CHECK_MEM_ERROR(cpi->mb_row_ei, vpx_memalign(32, sizeof(MB_ROW_COMP) * cpi->encoding_thread_count));
vpx_memset(cpi->mb_row_ei, 0, sizeof(MB_ROW_COMP) * cpi->encoding_thread_count);
CHECK_MEM_ERROR(cpi->en_thread_data, vpx_malloc(sizeof(ENCODETHREAD_DATA) * cpi->encoding_thread_count));
+ CHECK_MEM_ERROR(cpi->mt_current_mb_col, vpx_malloc(sizeof(*cpi->mt_current_mb_col) * cpi->common.mb_rows));
+
//cpi->h_event_main = CreateEvent(NULL, FALSE, FALSE, NULL);
- sem_init(&cpi->h_event_main, 0, 0);
+ sem_init(&cpi->h_event_end_encoding, 0, 0);
cpi->b_multi_threaded = 1;
@@ -466,11 +465,13 @@
for (ithread = 0; ithread < cpi->encoding_thread_count; ithread++)
{
+ ENCODETHREAD_DATA * ethd = &cpi->en_thread_data[ithread];
+
//cpi->h_event_mbrencoding[ithread] = CreateEvent(NULL, FALSE, FALSE, NULL);
- sem_init(&cpi->h_event_mbrencoding[ithread], 0, 0);
- cpi->en_thread_data[ithread].ithread = ithread;
- cpi->en_thread_data[ithread].ptr1 = (void *)cpi;
- cpi->en_thread_data[ithread].ptr2 = (void *)&cpi->mb_row_ei[ithread];
+ 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];
//printf(" call begin thread %d \n", ithread);
@@ -482,8 +483,7 @@
// 0,
// NULL);
- pthread_create(&cpi->h_encoding_thread[ithread], 0, thread_encoding_proc, (&cpi->en_thread_data[ithread]));
-
+ pthread_create(&cpi->h_encoding_thread[ithread], 0, thread_encoding_proc, ethd);
}
}
@@ -505,18 +505,21 @@
for (i = 0; i < cpi->encoding_thread_count; i++)
{
//SetEvent(cpi->h_event_mbrencoding[i]);
- sem_post(&cpi->h_event_mbrencoding[i]);
+ sem_post(&cpi->h_event_start_encoding[i]);
pthread_join(cpi->h_encoding_thread[i], 0);
- }
- for (i = 0; i < cpi->encoding_thread_count; i++)
- sem_destroy(&cpi->h_event_mbrencoding[i]);
+ sem_destroy(&cpi->h_event_start_encoding[i]);
+ }
}
+
+ sem_destroy(&cpi->h_event_end_encoding);
+
//free thread related resources
- vpx_free(cpi->h_event_mbrencoding);
+ 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);
+ vpx_free(cpi->mt_current_mb_col);
}
#endif
diff --git a/vp8/encoder/onyx_if.c b/vp8/encoder/onyx_if.c
index d967031..df1a6f5 100644
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -1465,6 +1465,17 @@
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate firstpass stats");
#endif
+
+#if CONFIG_MULTITHREAD
+ if (width < 640)
+ cpi->mt_sync_range = 1;
+ else if (width <= 1280)
+ cpi->mt_sync_range = 4;
+ else if (width <= 2560)
+ cpi->mt_sync_range = 8;
+ else
+ cpi->mt_sync_range = 16;
+#endif
}
@@ -4433,7 +4444,7 @@
}
// Update the GF useage maps.
- // Update the GF useage maps.
+ // This is done after completing the compression of a frame when all modes etc. are finalized but before loop filter
// This is done after completing the compression of a frame when all modes etc. are finalized but before loop filter
vp8_update_gf_useage_maps(cpi, cm, &cpi->mb);
diff --git a/vp8/encoder/onyx_int.h b/vp8/encoder/onyx_int.h
index 50f8158..e8a452d 100644
--- a/vp8/encoder/onyx_int.h
+++ b/vp8/encoder/onyx_int.h
@@ -191,11 +191,8 @@
typedef struct
{
MACROBLOCK mb;
- int mb_row;
- TOKENEXTRA *tp;
int segment_counts[MAX_MB_SEGMENTS];
int totalrate;
- int current_mb_col;
} MB_ROW_COMP;
typedef struct
@@ -593,7 +590,8 @@
signed char *cyclic_refresh_map;
// multithread data
- int current_mb_col_main;
+ int * mt_current_mb_col;
+ int mt_sync_range;
int processor_core_count;
int b_multi_threaded;
int encoding_thread_count;
@@ -606,8 +604,8 @@
#if CONFIG_MULTITHREAD
//events
- sem_t *h_event_mbrencoding;
- sem_t h_event_main;
+ sem_t *h_event_start_encoding;
+ sem_t h_event_end_encoding;
#endif
TOKENLIST *tplist;
diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index bffe433..248d9c4 100644
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -1953,15 +1953,9 @@
switch (this_mode)
{
case B_PRED:
-
- for (i = 0; i < 16; i++)
- {
- vpx_memset(&x->e_mbd.block[i].bmi, 0, sizeof(B_MODE_INFO));
- }
// Note the rate value returned here includes the cost of coding the BPRED mode : x->mbmode_cost[x->e_mbd.frame_type][BPRED];
vp8_rd_pick_intra4x4mby_modes(cpi, x, &rate, &rate_y, &distortion);
rate2 += rate;
-
distortion2 += distortion;
rate2 += uv_intra_rate;
rate_uv = uv_intra_rate_tokenonly;
@@ -2003,22 +1997,16 @@
case V_PRED:
case H_PRED:
case TM_PRED:
- for (i = 0; i < 16; i++)
- {
- vpx_memset(&x->e_mbd.block[i].bmi, 0, sizeof(B_MODE_INFO));
- }
x->e_mbd.mode_info_context->mbmi.ref_frame = INTRA_FRAME;
vp8_build_intra_predictors_mby_ptr(&x->e_mbd);
- {
- macro_block_yrd(x, &rate_y, &distortion, IF_RTCD(&cpi->rtcd.encodemb)) ;
- rate2 += rate_y;
- distortion2 += distortion;
- rate2 += x->mbmode_cost[x->e_mbd.frame_type][x->e_mbd.mode_info_context->mbmi.mode];
- rate2 += uv_intra_rate;
- rate_uv = uv_intra_rate_tokenonly;
- distortion2 += uv_intra_distortion;
- distortion_uv = uv_intra_distortion;
- }
+ macro_block_yrd(x, &rate_y, &distortion, IF_RTCD(&cpi->rtcd.encodemb)) ;
+ rate2 += rate_y;
+ distortion2 += distortion;
+ rate2 += x->mbmode_cost[x->e_mbd.frame_type][x->e_mbd.mode_info_context->mbmi.mode];
+ rate2 += uv_intra_rate;
+ rate_uv = uv_intra_rate_tokenonly;
+ distortion2 += uv_intra_distortion;
+ distortion_uv = uv_intra_distortion;
break;
case NEWMV:
@@ -2461,6 +2449,15 @@
}
+ if(best_mbmode.mode <= B_PRED)
+ {
+ int i;
+ for (i = 0; i < 16; i++)
+ {
+ best_bmodes[i].mv.as_int = 0;
+ }
+ }
+
// macroblock modes
vpx_memcpy(&x->e_mbd.mode_info_context->mbmi, &best_mbmode, sizeof(MB_MODE_INFO));
vpx_memcpy(x->partition_info, &best_partition, sizeof(PARTITION_INFO));
diff --git a/vpx/internal/vpx_codec_internal.h b/vpx/internal/vpx_codec_internal.h
index dcb451d..9cde426 100644
--- a/vpx/internal/vpx_codec_internal.h
+++ b/vpx/internal/vpx_codec_internal.h
@@ -321,7 +321,7 @@
{
vpx_codec_put_frame_cb_fn_t put_frame;
vpx_codec_put_slice_cb_fn_t put_slice;
- };
+ } u;
void *user_priv;
} vpx_codec_priv_cb_pair_t;
diff --git a/vpx/src/vpx_decoder.c b/vpx/src/vpx_decoder.c
index 27049a5..7b78e5c 100644
--- a/vpx/src/vpx_decoder.c
+++ b/vpx/src/vpx_decoder.c
@@ -160,7 +160,7 @@
res = VPX_CODEC_ERROR;
else
{
- ctx->priv->dec.put_frame_cb.put_frame = cb;
+ ctx->priv->dec.put_frame_cb.u.put_frame = cb;
ctx->priv->dec.put_frame_cb.user_priv = user_priv;
res = VPX_CODEC_OK;
}
@@ -182,7 +182,7 @@
res = VPX_CODEC_ERROR;
else
{
- ctx->priv->dec.put_slice_cb.put_slice = cb;
+ ctx->priv->dec.put_slice_cb.u.put_slice = cb;
ctx->priv->dec.put_slice_cb.user_priv = user_priv;
res = VPX_CODEC_OK;
}
