ref: dadc9189ed3e6a1b5efd77333e6037d2b0e2f398
parent: f05feab7b967ffd717d9437b7803bfbcaeced9a4
parent: d02e74a0732896b632880884b3d839a316473ae0
author: John Koleszar <jkoleszar@google.com>
date: Tue Feb 21 05:09:23 EST 2012
Merge changes I0341554f,I64e110c8 * changes: Consolidate C version of token packing functions Multithreaded encoder, late sync loopfilter
--- a/vp8/encoder/bitstream.c
+++ b/vp8/encoder/bitstream.c
@@ -382,221 +382,25 @@
int i;
unsigned char *ptr = cx_data;
unsigned char *ptr_end = cx_data_end;
- unsigned int shift;
- vp8_writer *w;
- ptr = cx_data;
+ vp8_writer * w;
for (i = 0; i < num_part; i++)
{+ int mb_row;
+
w = cpi->bc + i + 1;
+
vp8_start_encode(w, ptr, ptr_end);
- {- unsigned int split;
- int count = w->count;
- unsigned int range = w->range;
- unsigned int lowvalue = w->lowvalue;
- int mb_row;
- for (mb_row = i; mb_row < cpi->common.mb_rows; mb_row += num_part)
- {- TOKENEXTRA *p = cpi->tplist[mb_row].start;
- TOKENEXTRA *stop = cpi->tplist[mb_row].stop;
-
- while (p < stop)
- {- const int t = p->Token;
- vp8_token *const a = vp8_coef_encodings + t;
- const vp8_extra_bit_struct *const b = vp8_extra_bits + t;
- int i = 0;
- const unsigned char *pp = p->context_tree;
- int v = a->value;
- int n = a->Len;
-
- if (p->skip_eob_node)
- {- n--;
- i = 2;
- }
-
- do
- {- const int bb = (v >> --n) & 1;
- split = 1 + (((range - 1) * pp[i>>1]) >> 8);
- i = vp8_coef_tree[i+bb];
-
- if (bb)
- {- lowvalue += split;
- range = range - split;
- }
- else
- {- range = split;
- }
+ for (mb_row = i; mb_row < cpi->common.mb_rows; mb_row += num_part)
+ {+ const TOKENEXTRA *p = cpi->tplist[mb_row].start;
+ const TOKENEXTRA *stop = cpi->tplist[mb_row].stop;
+ int tokens = stop - p;
- shift = vp8_norm[range];
- range <<= shift;
- count += shift;
-
- if (count >= 0)
- {- int offset = shift - count;
-
- if ((lowvalue << (offset - 1)) & 0x80000000)
- {- int x = w->pos - 1;
-
- while (x >= 0 && w->buffer[x] == 0xff)
- {- w->buffer[x] = (unsigned char)0;
- x--;
- }
-
- w->buffer[x] += 1;
- }
-
- validate_buffer(w->buffer + w->pos,
- 1,
- cx_data_end,
- &cpi->common.error);
-
- w->buffer[w->pos++] = (lowvalue >> (24 - offset));
-
- lowvalue <<= offset;
- shift = count;
- lowvalue &= 0xffffff;
- count -= 8 ;
- }
+ pack_tokens_c(w, p, tokens);
+ }
- lowvalue <<= shift;
- }
- while (n);
-
-
- if (b->base_val)
- {- const int e = p->Extra, L = b->Len;
-
- if (L)
- {- const unsigned char *pp = b->prob;
- int v = e >> 1;
- int n = L; /* number of bits in v, assumed nonzero */
- int i = 0;
-
- do
- {- const int bb = (v >> --n) & 1;
- split = 1 + (((range - 1) * pp[i>>1]) >> 8);
- i = b->tree[i+bb];
-
- if (bb)
- {- lowvalue += split;
- range = range - split;
- }
- else
- {- range = split;
- }
-
- shift = vp8_norm[range];
- range <<= shift;
- count += shift;
-
- if (count >= 0)
- {- int offset = shift - count;
-
- if ((lowvalue << (offset - 1)) & 0x80000000)
- {- int x = w->pos - 1;
-
- while (x >= 0 && w->buffer[x] == 0xff)
- {- w->buffer[x] = (unsigned char)0;
- x--;
- }
-
- w->buffer[x] += 1;
- }
-
- validate_buffer(w->buffer + w->pos,
- 1,
- cx_data_end,
- &cpi->common.error);
-
- w->buffer[w->pos++] =
- (lowvalue >> (24 - offset));
-
- lowvalue <<= offset;
- shift = count;
- lowvalue &= 0xffffff;
- count -= 8 ;
- }
-
- lowvalue <<= shift;
- }
- while (n);
- }
-
- {- split = (range + 1) >> 1;
-
- if (e & 1)
- {- lowvalue += split;
- range = range - split;
- }
- else
- {- range = split;
- }
-
- range <<= 1;
-
- if ((lowvalue & 0x80000000))
- {- int x = w->pos - 1;
-
- while (x >= 0 && w->buffer[x] == 0xff)
- {- w->buffer[x] = (unsigned char)0;
- x--;
- }
-
- w->buffer[x] += 1;
-
- }
-
- lowvalue <<= 1;
-
- if (!++count)
- {- count = -8;
- validate_buffer(w->buffer + w->pos,
- 1,
- cx_data_end,
- &cpi->common.error);
-
- w->buffer[w->pos++] = (lowvalue >> 24);
-
- lowvalue &= 0xffffff;
- }
- }
-
- }
-
- ++p;
- }
- }
-
- w->count = count;
- w->lowvalue = lowvalue;
- w->range = range;
-
- }
-
vp8_stop_encode(w);
ptr += w->pos;
}
@@ -605,208 +409,16 @@
static void pack_mb_row_tokens_c(VP8_COMP *cpi, vp8_writer *w)
{-
- unsigned int split;
- int count = w->count;
- unsigned int range = w->range;
- unsigned int lowvalue = w->lowvalue;
- unsigned int shift;
int mb_row;
for (mb_row = 0; mb_row < cpi->common.mb_rows; mb_row++)
{- TOKENEXTRA *p = cpi->tplist[mb_row].start;
- TOKENEXTRA *stop = cpi->tplist[mb_row].stop;
+ const TOKENEXTRA *p = cpi->tplist[mb_row].start;
+ const TOKENEXTRA *stop = cpi->tplist[mb_row].stop;
+ int tokens = stop - p;
- while (p < stop)
- {- const int t = p->Token;
- vp8_token *const a = vp8_coef_encodings + t;
- const vp8_extra_bit_struct *const b = vp8_extra_bits + t;
- int i = 0;
- const unsigned char *pp = p->context_tree;
- int v = a->value;
- int n = a->Len;
-
- if (p->skip_eob_node)
- {- n--;
- i = 2;
- }
-
- do
- {- const int bb = (v >> --n) & 1;
- split = 1 + (((range - 1) * pp[i>>1]) >> 8);
- i = vp8_coef_tree[i+bb];
-
- if (bb)
- {- lowvalue += split;
- range = range - split;
- }
- else
- {- range = split;
- }
-
- shift = vp8_norm[range];
- range <<= shift;
- count += shift;
-
- if (count >= 0)
- {- int offset = shift - count;
-
- if ((lowvalue << (offset - 1)) & 0x80000000)
- {- int x = w->pos - 1;
-
- while (x >= 0 && w->buffer[x] == 0xff)
- {- w->buffer[x] = (unsigned char)0;
- x--;
- }
-
- w->buffer[x] += 1;
- }
-
- validate_buffer(w->buffer + w->pos,
- 1,
- w->buffer_end,
- w->error);
-
- w->buffer[w->pos++] = (lowvalue >> (24 - offset));
- lowvalue <<= offset;
- shift = count;
- lowvalue &= 0xffffff;
- count -= 8 ;
- }
-
- lowvalue <<= shift;
- }
- while (n);
-
-
- if (b->base_val)
- {- const int e = p->Extra, L = b->Len;
-
- if (L)
- {- const unsigned char *pp = b->prob;
- int v = e >> 1;
- int n = L; /* number of bits in v, assumed nonzero */
- int i = 0;
-
- do
- {- const int bb = (v >> --n) & 1;
- split = 1 + (((range - 1) * pp[i>>1]) >> 8);
- i = b->tree[i+bb];
-
- if (bb)
- {- lowvalue += split;
- range = range - split;
- }
- else
- {- range = split;
- }
-
- shift = vp8_norm[range];
- range <<= shift;
- count += shift;
-
- if (count >= 0)
- {- int offset = shift - count;
-
- if ((lowvalue << (offset - 1)) & 0x80000000)
- {- int x = w->pos - 1;
-
- while (x >= 0 && w->buffer[x] == 0xff)
- {- w->buffer[x] = (unsigned char)0;
- x--;
- }
-
- w->buffer[x] += 1;
- }
-
- validate_buffer(w->buffer + w->pos,
- 1,
- w->buffer_end,
- w->error);
-
- w->buffer[w->pos++] = (lowvalue >> (24 - offset));
- lowvalue <<= offset;
- shift = count;
- lowvalue &= 0xffffff;
- count -= 8 ;
- }
-
- lowvalue <<= shift;
- }
- while (n);
- }
-
- {- split = (range + 1) >> 1;
-
- if (e & 1)
- {- lowvalue += split;
- range = range - split;
- }
- else
- {- range = split;
- }
-
- range <<= 1;
-
- if ((lowvalue & 0x80000000))
- {- int x = w->pos - 1;
-
- while (x >= 0 && w->buffer[x] == 0xff)
- {- w->buffer[x] = (unsigned char)0;
- x--;
- }
-
- w->buffer[x] += 1;
-
- }
-
- lowvalue <<= 1;
-
- if (!++count)
- {- count = -8;
-
- validate_buffer(w->buffer + w->pos,
- 1,
- w->buffer_end,
- w->error);
-
- w->buffer[w->pos++] = (lowvalue >> 24);
- lowvalue &= 0xffffff;
- }
- }
-
- }
-
- ++p;
- }
- }
-
- w->count = count;
- w->lowvalue = lowvalue;
- w->range = range;
+ pack_tokens_c(w, p, tokens);
+ }
}
--- a/vp8/encoder/ethreading.c
+++ b/vp8/encoder/ethreading.c
@@ -24,9 +24,9 @@
extern void vp8_build_block_offsets(MACROBLOCK *x);
extern void vp8_setup_block_ptrs(MACROBLOCK *x);
-extern void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm);
+extern void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm);
-static THREAD_FUNCTION loopfilter_thread(void *p_data)
+static THREAD_FUNCTION thread_loopfilter(void *p_data)
{VP8_COMP *cpi = (VP8_COMP *)(((LPFTHREAD_DATA *)p_data)->ptr1);
VP8_COMMON *cm = &cpi->common;
@@ -41,7 +41,7 @@
if (cpi->b_multi_threaded == 0) // we're shutting down
break;
- loopfilter_frame(cpi, cm);
+ vp8_loopfilter_frame(cpi, cm);
sem_post(&cpi->h_event_end_lpf);
}
@@ -468,6 +468,7 @@
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)
{@@ -526,7 +527,7 @@
sem_init(&cpi->h_event_end_lpf, 0, 0);
lpfthd->ptr1 = (void *)cpi;
- pthread_create(&cpi->h_filter_thread, 0, loopfilter_thread, lpfthd);
+ pthread_create(&cpi->h_filter_thread, 0, thread_loopfilter, lpfthd);
}
}
--- a/vp8/encoder/onyx_if.c
+++ b/vp8/encoder/onyx_if.c
@@ -3086,7 +3086,7 @@
}
}
-void loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
+void vp8_loopfilter_frame(VP8_COMP *cpi, VP8_COMMON *cm)
{if (cm->no_lpf)
{@@ -3716,6 +3716,15 @@
vp8_setup_key_frame(cpi);
}
+#if CONFIG_MULTITHREAD
+ /* wait for the last picture loopfilter thread done */
+ if (cpi->b_lpf_running)
+ {+ sem_wait(&cpi->h_event_end_lpf);
+ cpi->b_lpf_running = 0;
+ }
+#endif
+
// transform / motion compensation build reconstruction frame
vp8_encode_frame(cpi);
@@ -4074,11 +4083,12 @@
if (cpi->b_multi_threaded)
{sem_post(&cpi->h_event_start_lpf); /* start loopfilter in separate thread */
+ cpi->b_lpf_running = 1;
}
else
#endif
{- loopfilter_frame(cpi, cm);
+ vp8_loopfilter_frame(cpi, cm);
}
update_reference_frames(cm);
@@ -4098,10 +4108,11 @@
vp8_pack_bitstream(cpi, dest, dest_end, size);
#if CONFIG_MULTITHREAD
- /* wait for loopfilter thread done */
- if (cpi->b_multi_threaded)
+ /* if PSNR packets are generated we have to wait for the lpf */
+ if (cpi->b_lpf_running && cpi->b_calculate_psnr)
{sem_wait(&cpi->h_event_end_lpf);
+ cpi->b_lpf_running = 0;
}
#endif
@@ -5107,6 +5118,15 @@
else
{int ret;
+
+#if CONFIG_MULTITHREAD
+ if(cpi->b_lpf_running)
+ {+ sem_wait(&cpi->h_event_end_lpf);
+ cpi->b_lpf_running = 0;
+ }
+#endif
+
#if CONFIG_POSTPROC
ret = vp8_post_proc_frame(&cpi->common, dest, flags);
#else
--- a/vp8/encoder/onyx_int.h
+++ b/vp8/encoder/onyx_int.h
@@ -525,6 +525,7 @@
int mt_sync_range;
int b_multi_threaded;
int encoding_thread_count;
+ int b_lpf_running;
pthread_t *h_encoding_thread;
pthread_t h_filter_thread;
--
⑨