New ways of passing encoded data between encoder and decoder.

With this commit frames can be received partition-by-partition
from the encoder and passed partition-by-partition to the
decoder.

At the encoder-side this makes it easier to split encoded
frames at partition boundaries, useful when packetizing
frames. When VPX_CODEC_USE_OUTPUT_PARTITION is enabled,
several VPX_CODEC_CX_FRAME_PKT packets will be returned
from vpx_codec_get_cx_data(), containing one partition
each. The partition_id (starting at 0) specifies the decoding
order of the partitions. All partitions but the last has
the VPX_FRAME_IS_FRAGMENT flag set.

At the decoder this opens up the possibility of decoding partition
N even though partition N-1 was lost (given that independent
partitioning has been enabled in the encoder) if more info
about the missing parts of the stream is available through
external signaling.

Each partition is passed to the decoder through the
vpx_codec_decode() function, with the data pointer pointing
to the start of the partition, and with data_sz equal to the
size of the partition. Missing partitions can be signaled to
the decoder by setting data != NULL and data_sz = 0. When
all partitions have been given to the decoder "end of data"
should be signaled by calling vpx_codec_decode() with
data = NULL and data_sz = 0.

The first partition is the first partition according to the
VP8 bitstream + the uncompressed data chunk + DCT address
offsets if multiple residual partitions are used.

Change-Id: I5bc0682b9e4112e0db77904755c694c3c7ac6e74
This commit is contained in:
Stefan Holmer 2011-06-13 16:42:27 +02:00 committed by John Koleszar
parent b433e12a3d
commit 7296b3f922
10 changed files with 181 additions and 56 deletions

View File

@ -35,6 +35,8 @@ void vp8_initialize_common(void);
#define NUM_YV12_BUFFERS 4
#define MAX_PARTITIONS 9
typedef struct frame_contexts
{
vp8_prob bmode_prob [VP8_BINTRAMODES-1];

View File

@ -33,6 +33,7 @@ extern "C"
int postprocess;
int max_threads;
int error_concealment;
int input_partition;
} VP8D_CONFIG;
typedef enum
{

View File

@ -463,6 +463,40 @@ static unsigned int read_partition_size(const unsigned char *cx_size)
return size;
}
static void setup_token_decoder_partition_input(VP8D_COMP *pbi)
{
vp8_reader *bool_decoder = &pbi->bc2;
int part_idx = 1;
TOKEN_PARTITION multi_token_partition =
(TOKEN_PARTITION)vp8_read_literal(&pbi->bc, 2);
assert(vp8dx_bool_error(&pbi->bc) ||
multi_token_partition == pbi->common.multi_token_partition);
if (pbi->num_partitions > 2)
{
CHECK_MEM_ERROR(pbi->mbc, vpx_malloc((pbi->num_partitions - 1) *
sizeof(vp8_reader)));
bool_decoder = pbi->mbc;
}
for (; part_idx < pbi->num_partitions; ++part_idx)
{
if (vp8dx_start_decode(bool_decoder,
pbi->partitions[part_idx],
pbi->partition_sizes[part_idx]))
vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate bool decoder %d",
part_idx);
bool_decoder++;
}
#if CONFIG_MULTITHREAD
/* Clamp number of decoder threads */
if (pbi->decoding_thread_count > pbi->num_partitions - 1)
pbi->decoding_thread_count = pbi->num_partitions - 1;
#endif
}
static void setup_token_decoder(VP8D_COMP *pbi,
const unsigned char *cx_data)
@ -619,13 +653,19 @@ int vp8_decode_frame(VP8D_COMP *pbi)
VP8_COMMON *const pc = & pbi->common;
MACROBLOCKD *const xd = & pbi->mb;
const unsigned char *data = (const unsigned char *)pbi->Source;
const unsigned char *const data_end = data + pbi->source_sz;
const unsigned char *data_end = data + pbi->source_sz;
ptrdiff_t first_partition_length_in_bytes;
int mb_row;
int i, j, k, l;
const int *const mb_feature_data_bits = vp8_mb_feature_data_bits;
if (pbi->input_partition)
{
data = pbi->partitions[0];
data_end = data + pbi->partition_sizes[0];
}
/* start with no corruption of current frame */
xd->corrupted = 0;
pc->yv12_fb[pc->new_fb_idx].corrupted = 0;
@ -841,7 +881,14 @@ int vp8_decode_frame(VP8D_COMP *pbi)
}
}
setup_token_decoder(pbi, data + first_partition_length_in_bytes);
if (pbi->input_partition)
{
setup_token_decoder_partition_input(pbi);
}
else
{
setup_token_decoder(pbi, data + first_partition_length_in_bytes);
}
xd->current_bc = &pbi->bc2;
/* Read the default quantizers. */
@ -930,10 +977,8 @@ int vp8_decode_frame(VP8D_COMP *pbi)
fclose(z);
}
{
/* read coef probability tree */
for (i = 0; i < BLOCK_TYPES; i++)
for (j = 0; j < COEF_BANDS; j++)
for (k = 0; k < PREV_COEF_CONTEXTS; k++)
@ -1021,7 +1066,6 @@ int vp8_decode_frame(VP8D_COMP *pbi)
}
}
stop_token_decoder(pbi);
/* Collect information about decoder corruption. */

View File

@ -109,6 +109,8 @@ VP8D_PTR vp8dx_create_decompressor(VP8D_CONFIG *oxcf)
pbi->ec_enabled = 0;
#endif
pbi->input_partition = oxcf->input_partition;
return (VP8D_PTR) pbi;
}
@ -312,69 +314,92 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign
pbi->common.error.error_code = VPX_CODEC_OK;
if (size == 0)
if (pbi->input_partition && !(source == NULL && size == 0))
{
/* This is used to signal that we are missing frames.
* We do not know if the missing frame(s) was supposed to update
* any of the reference buffers, but we act conservative and
* mark only the last buffer as corrupted.
*/
cm->yv12_fb[cm->lst_fb_idx].corrupted = 1;
/* If error concealment is disabled we won't signal missing frames to
* the decoder.
/* Store a pointer to this partition and return. We haven't
* received the complete frame yet, so we will wait with decoding.
*/
if (!pbi->ec_enabled)
pbi->partitions[pbi->num_partitions] = source;
pbi->partition_sizes[pbi->num_partitions] = size;
pbi->source_sz += size;
pbi->num_partitions++;
if (pbi->num_partitions > (1<<pbi->common.multi_token_partition) + 1)
pbi->common.multi_token_partition++;
if (pbi->common.multi_token_partition > EIGHT_PARTITION)
{
/* Signal that we have no frame to show. */
cm->show_frame = 0;
/* Nothing more to do. */
return 0;
pbi->common.error.error_code = VPX_CODEC_UNSUP_BITSTREAM;
pbi->common.error.setjmp = 0;
return -1;
}
return 0;
}
#if HAVE_ARMV7
#if CONFIG_RUNTIME_CPU_DETECT
if (cm->rtcd.flags & HAS_NEON)
#endif
else
{
vp8_push_neon(dx_store_reg);
}
#endif
if (!pbi->input_partition)
{
pbi->Source = source;
pbi->source_sz = size;
}
cm->new_fb_idx = get_free_fb (cm);
if (pbi->source_sz == 0)
{
/* This is used to signal that we are missing frames.
* We do not know if the missing frame(s) was supposed to update
* any of the reference buffers, but we act conservative and
* mark only the last buffer as corrupted.
*/
cm->yv12_fb[cm->lst_fb_idx].corrupted = 1;
/* If error concealment is disabled we won't signal missing frames to
* the decoder.
*/
if (!pbi->ec_enabled)
{
/* Signal that we have no frame to show. */
cm->show_frame = 0;
/* Nothing more to do. */
return 0;
}
}
if (setjmp(pbi->common.error.jmp))
{
#if HAVE_ARMV7
#if CONFIG_RUNTIME_CPU_DETECT
if (cm->rtcd.flags & HAS_NEON)
#endif
{
vp8_pop_neon(dx_store_reg);
vp8_push_neon(dx_store_reg);
}
#endif
pbi->common.error.setjmp = 0;
/* We do not know if the missing frame(s) was supposed to update
* any of the reference buffers, but we act conservative and
* mark only the last buffer as corrupted.
*/
cm->yv12_fb[cm->lst_fb_idx].corrupted = 1;
cm->new_fb_idx = get_free_fb (cm);
if (cm->fb_idx_ref_cnt[cm->new_fb_idx] > 0)
cm->fb_idx_ref_cnt[cm->new_fb_idx]--;
return -1;
if (setjmp(pbi->common.error.jmp))
{
#if HAVE_ARMV7
#if CONFIG_RUNTIME_CPU_DETECT
if (cm->rtcd.flags & HAS_NEON)
#endif
{
vp8_pop_neon(dx_store_reg);
}
#endif
pbi->common.error.setjmp = 0;
/* We do not know if the missing frame(s) was supposed to update
* any of the reference buffers, but we act conservative and
* mark only the last buffer as corrupted.
*/
cm->yv12_fb[cm->lst_fb_idx].corrupted = 1;
if (cm->fb_idx_ref_cnt[cm->new_fb_idx] > 0)
cm->fb_idx_ref_cnt[cm->new_fb_idx]--;
return -1;
}
pbi->common.error.setjmp = 1;
}
pbi->common.error.setjmp = 1;
/*cm->current_video_frame++;*/
pbi->Source = source;
pbi->source_sz = size;
retcode = vp8_decode_frame(pbi);
if (retcode < 0)
@ -473,6 +498,10 @@ int vp8dx_receive_compressed_data(VP8D_PTR ptr, unsigned long size, const unsign
pbi->ready_for_new_data = 0;
pbi->last_time_stamp = time_stamp;
pbi->num_partitions = 0;
if (pbi->input_partition)
pbi->common.multi_token_partition = 0;
pbi->source_sz = 0;
#if 0
{

View File

@ -83,6 +83,9 @@ typedef struct VP8Decompressor
const unsigned char *Source;
unsigned int source_sz;
const unsigned char *partitions[MAX_PARTITIONS];
unsigned int partition_sizes[MAX_PARTITIONS];
unsigned int num_partitions;
#if CONFIG_MULTITHREAD
/* variable for threading */
@ -137,6 +140,7 @@ typedef struct VP8Decompressor
unsigned int mvs_corrupt_from_mb;
#endif
int ec_enabled;
int input_partition;
} VP8D_COMP;

View File

@ -377,6 +377,7 @@ static void pack_tokens_into_partitions_c(VP8_COMP *cpi, unsigned char *cx_data,
unsigned int shift;
vp8_writer *w = &cpi->bc2;
*size = 3 * (num_part - 1);
cpi->partition_sz[0] += *size;
ptr = cx_data + (*size);
for (i = 0; i < num_part; i++)
@ -573,6 +574,9 @@ static void pack_tokens_into_partitions_c(VP8_COMP *cpi, unsigned char *cx_data,
vp8_stop_encode(w);
*size += w->pos;
/* The first partition size is set earlier */
cpi->partition_sz[i + 1] = w->pos;
if (i < (num_part - 1))
{
write_partition_size(cx_data, w->pos);
@ -1840,6 +1844,7 @@ void vp8_pack_bitstream(VP8_COMP *cpi, unsigned char *dest, unsigned long *size)
}
*size = VP8_HEADER_SIZE + extra_bytes_packed + cpi->bc.pos;
cpi->partition_sz[0] = *size;
if (pc->multi_token_partition != ONE_PARTITION)
{
@ -1865,6 +1870,7 @@ void vp8_pack_bitstream(VP8_COMP *cpi, unsigned char *dest, unsigned long *size)
vp8_stop_encode(&cpi->bc2);
*size += cpi->bc2.pos;
cpi->partition_sz[1] = cpi->bc2.pos;
}
}

View File

@ -501,6 +501,7 @@ typedef struct VP8_COMP
#endif
TOKENLIST *tplist;
unsigned int partition_sz[MAX_PARTITIONS];
// end of multithread data
@ -604,6 +605,8 @@ typedef struct VP8_COMP
unsigned char *gf_active_flags;
int gf_active_count;
int output_partition;
//Store last frame's MV info for next frame MV prediction
int_mv *lfmv;
int *lf_ref_frame_sign_bias;

View File

@ -731,6 +731,9 @@ static vpx_codec_err_t vp8e_encode(vpx_codec_alg_priv_t *ctx,
if (ctx->base.init_flags & VPX_CODEC_USE_PSNR)
((VP8_COMP *)ctx->cpi)->b_calculate_psnr = 1;
if (ctx->base.init_flags & VPX_CODEC_USE_OUTPUT_PARTITION)
((VP8_COMP *)ctx->cpi)->output_partition = 1;
/* Convert API flags to internal codec lib flags */
lib_flags = (flags & VPX_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
@ -770,8 +773,6 @@ static vpx_codec_err_t vp8e_encode(vpx_codec_alg_priv_t *ctx,
round = 1000000 * ctx->cfg.g_timebase.num / 2 - 1;
delta = (dst_end_time_stamp - dst_time_stamp);
pkt.kind = VPX_CODEC_CX_FRAME_PKT;
pkt.data.frame.buf = cx_data;
pkt.data.frame.sz = size;
pkt.data.frame.pts =
(dst_time_stamp * ctx->cfg.g_timebase.den + round)
/ ctx->cfg.g_timebase.num / 10000000;
@ -797,11 +798,35 @@ static vpx_codec_err_t vp8e_encode(vpx_codec_alg_priv_t *ctx,
pkt.data.frame.duration = 0;
}
vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt);
if (cpi->output_partition)
{
int i;
const int num_partitions =
(1 << cpi->common.multi_token_partition) + 1;
for (i = 0; i < num_partitions; ++i)
{
pkt.data.frame.buf = cx_data;
pkt.data.frame.sz = cpi->partition_sz[i];
pkt.data.frame.partition_id = i;
/* don't set the fragment bit for the last partition */
if (i < num_partitions - 1)
pkt.data.frame.flags |= VPX_FRAME_IS_FRAGMENT;
vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt);
cx_data += cpi->partition_sz[i];
cx_data_sz -= cpi->partition_sz[i];
}
}
else
{
pkt.data.frame.buf = cx_data;
pkt.data.frame.sz = size;
pkt.data.frame.partition_id = -1;
vpx_codec_pkt_list_add(&ctx->pkt_list.head, &pkt);
cx_data += size;
cx_data_sz -= size;
}
//printf("timestamp: %lld, duration: %d\n", pkt->data.frame.pts, pkt->data.frame.duration);
cx_data += size;
cx_data_sz -= size;
}
}
}
@ -1121,7 +1146,8 @@ CODEC_INTERFACE(vpx_codec_vp8_cx) =
{
"WebM Project VP8 Encoder" VERSION_STRING,
VPX_CODEC_INTERNAL_ABI_VERSION,
VPX_CODEC_CAP_ENCODER | VPX_CODEC_CAP_PSNR,
VPX_CODEC_CAP_ENCODER | VPX_CODEC_CAP_PSNR |
VPX_CODEC_CAP_OUTPUT_PARTITION,
/* vpx_codec_caps_t caps; */
vp8e_init, /* vpx_codec_init_fn_t init; */
vp8e_destroy, /* vpx_codec_destroy_fn_t destroy; */

View File

@ -368,6 +368,8 @@ static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t *ctx,
oxcf.max_threads = ctx->cfg.threads;
oxcf.error_concealment =
(ctx->base.init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT);
oxcf.input_partition =
(ctx->base.init_flags & VPX_CODEC_USE_INPUT_PARTITION);
optr = vp8dx_create_decompressor(&oxcf);
@ -721,7 +723,8 @@ CODEC_INTERFACE(vpx_codec_vp8_dx) =
{
"WebM Project VP8 Decoder" VERSION_STRING,
VPX_CODEC_INTERNAL_ABI_VERSION,
VPX_CODEC_CAP_DECODER | VP8_CAP_POSTPROC | VP8_CAP_ERROR_CONCEALMENT,
VPX_CODEC_CAP_DECODER | VP8_CAP_POSTPROC | VP8_CAP_ERROR_CONCEALMENT |
VPX_CODEC_CAP_INPUT_PARTITION,
/* vpx_codec_caps_t caps; */
vp8_init, /* vpx_codec_init_fn_t init; */
vp8_destroy, /* vpx_codec_destroy_fn_t destroy; */

View File

@ -163,6 +163,13 @@ extern "C" {
unsigned long duration; /**< duration to show frame
(in timebase units) */
vpx_codec_frame_flags_t flags; /**< flags for this frame */
int partition_id; /**< the partition id
defines the decoding order
of the partitions. Only
applicable when "output partition"
mode is enabled. First partition
has id 0.*/
} frame; /**< data for compressed frame packet */
struct vpx_fixed_buf twopass_stats; /**< data for two-pass packet */
struct vpx_psnr_pkt