}
}
-static int alloc_scratch_buffers(H264Context *h, int linesize)
+static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
{
+ const H264Context *h = sl->h264;
int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
- if (h->bipred_scratchpad)
- return 0;
-
- h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
+ av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
// edge emu needs blocksize + filter length - 1
// (= 21x21 for h264)
- h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
-
- if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
- av_freep(&h->bipred_scratchpad);
- av_freep(&h->edge_emu_buffer);
+ av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
+
+ av_fast_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
+ h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
+ av_fast_malloc(&sl->top_borders[1], &sl->top_borders_allocated[1],
+ h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
+
+ if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
+ !sl->top_borders[0] || !sl->top_borders[1]) {
+ av_freep(&sl->bipred_scratchpad);
+ av_freep(&sl->edge_emu_buffer);
+ av_freep(&sl->top_borders[0]);
+ av_freep(&sl->top_borders[1]);
+
+ sl->bipred_scratchpad_allocated = 0;
+ sl->edge_emu_buffer_allocated = 0;
+ sl->top_borders_allocated[0] = 0;
+ sl->top_borders_allocated[1] = 0;
return AVERROR(ENOMEM);
}
dst->DPB = src->DPB;
dst->cur_pic_ptr = src->cur_pic_ptr;
dst->cur_pic = src->cur_pic;
- dst->bipred_scratchpad = NULL;
- dst->edge_emu_buffer = NULL;
ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
src->sps.chroma_format_idc);
}
* the current value */
h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
- av_freep(&h->bipred_scratchpad);
-
h->width = h1->width;
h->height = h1->height;
h->mb_height = h1->mb_height;
for (i = 0; i < MAX_PPS_COUNT; i++)
av_freep(h->pps_buffers + i);
- av_freep(&h->rbsp_buffer);
ff_h264_unref_picture(h, &h->last_pic_for_ec);
memcpy(h, h1, sizeof(H264Context));
memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
- memset(&h->er, 0, sizeof(h->er));
memset(&h->cur_pic, 0, sizeof(h->cur_pic));
memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
h->slice_ctx = orig_slice_ctx;
+ memset(&h->slice_ctx[0].er, 0, sizeof(h->slice_ctx[0].er));
memset(&h->slice_ctx[0].mb, 0, sizeof(h->slice_ctx[0].mb));
memset(&h->slice_ctx[0].mb_luma_dc, 0, sizeof(h->slice_ctx[0].mb_luma_dc));
memset(&h->slice_ctx[0].mb_padding, 0, sizeof(h->slice_ctx[0].mb_padding));
h->list_counts = NULL;
h->mb2b_xy = NULL;
h->mb2br_xy = NULL;
- h->rbsp_buffer = NULL;
- h->rbsp_buffer_size = 0;
if (h1->context_initialized) {
h->context_initialized = 0;
av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
return ret;
}
- ret = ff_h264_context_init(h);
+ ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
if (ret < 0) {
av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
return ret;
}
}
- h->bipred_scratchpad = NULL;
- h->edge_emu_buffer = NULL;
-
h->thread_context[0] = h;
h->context_initialized = h1->context_initialized;
}
h->cur_pic_ptr = pic;
ff_h264_unref_picture(h, &h->cur_pic);
if (CONFIG_ERROR_RESILIENCE) {
- ff_h264_set_erpic(&h->er.cur_pic, NULL);
+ ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
}
if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
return ret;
if (CONFIG_ERROR_RESILIENCE) {
- ff_er_frame_start(&h->er);
- ff_h264_set_erpic(&h->er.last_pic, NULL);
- ff_h264_set_erpic(&h->er.next_pic, NULL);
+ ff_er_frame_start(&h->slice_ctx[0].er);
+ ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
+ ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
}
assert(h->linesize && h->uvlinesize);
return 0;
}
-static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *sl,
+static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
uint8_t *src_y,
uint8_t *src_cb, uint8_t *src_cr,
int linesize, int uvlinesize,
if (!simple && FRAME_MBAFF(h)) {
if (sl->mb_y & 1) {
- if (!MB_MBAFF(h)) {
- top_border = h->top_borders[0][sl->mb_x];
+ if (!MB_MBAFF(sl)) {
+ top_border = sl->top_borders[0][sl->mb_x];
AV_COPY128(top_border, src_y + 15 * linesize);
if (pixel_shift)
AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
}
}
}
- } else if (MB_MBAFF(h)) {
+ } else if (MB_MBAFF(sl)) {
top_idx = 0;
} else
return;
}
- top_border = h->top_borders[top_idx][sl->mb_x];
+ top_border = sl->top_borders[top_idx][sl->mb_x];
/* There are two lines saved, the line above the top macroblock
* of a pair, and the line above the bottom macroblock. */
AV_COPY128(top_border, src_y + 16 * linesize);
* @param field 0/1 initialize the weight for interlaced MBAFF
* -1 initializes the rest
*/
-static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int field)
+static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
{
int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
}
}
-/**
- * Replicate H264 "master" context to thread contexts.
- */
-static int clone_slice(H264Context *dst, H264Context *src)
-{
- memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
- dst->cur_pic_ptr = src->cur_pic_ptr;
- dst->cur_pic = src->cur_pic;
- dst->linesize = src->linesize;
- dst->uvlinesize = src->uvlinesize;
- dst->first_field = src->first_field;
-
- dst->prev_poc_msb = src->prev_poc_msb;
- dst->prev_poc_lsb = src->prev_poc_lsb;
- dst->prev_frame_num_offset = src->prev_frame_num_offset;
- dst->prev_frame_num = src->prev_frame_num;
- dst->short_ref_count = src->short_ref_count;
-
- memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
- memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
- memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
-
- memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
- memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
-
- return 0;
-}
-
static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
{
#define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
h->slice_context_count = nb_slices;
if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
- ret = ff_h264_context_init(h);
+ ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
if (ret < 0) {
av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
goto fail;
}
for (i = 0; i < h->slice_context_count; i++)
- if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
+ if ((ret = ff_h264_slice_context_init(h, &h->slice_ctx[i])) < 0) {
av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
goto fail;
}
* This will (re)intialize the decoder and call h264_frame_start() as needed.
*
* @param h h264context
- * @param h0 h264 master context (differs from 'h' when doing sliced based
- * parallel decoding)
*
* @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
*/
-int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Context *h0)
+int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
{
unsigned int first_mb_in_slice;
unsigned int pps_id;
int must_reinit;
int needs_reinit = 0;
int field_pic_flag, bottom_field_flag;
- int first_slice = h == h0 && !h0->current_slice;
+ int first_slice = sl == h->slice_ctx && !h->current_slice;
int frame_num, picture_structure, droppable;
PPS *pps;
first_mb_in_slice = get_ue_golomb_long(&sl->gb);
if (first_mb_in_slice == 0) { // FIXME better field boundary detection
- if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
+ if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
ff_h264_field_end(h, sl, 1);
}
- h0->current_slice = 0;
- if (!h0->first_field) {
+ h->current_slice = 0;
+ if (!h->first_field) {
if (h->cur_pic_ptr && !h->droppable) {
ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
h->picture_structure == PICT_BOTTOM_FIELD);
sl->slice_type_fixed = 0;
slice_type = golomb_to_pict_type[slice_type];
+
sl->slice_type = slice_type;
sl->slice_type_nos = slice_type & 3;
av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
return AVERROR_INVALIDDATA;
}
- if (!h0->pps_buffers[pps_id]) {
+ if (!h->pps_buffers[pps_id]) {
av_log(h->avctx, AV_LOG_ERROR,
"non-existing PPS %u referenced\n",
pps_id);
return AVERROR_INVALIDDATA;
}
- if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
+ if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
av_log(h->avctx, AV_LOG_ERROR,
"PPS change from %d to %d forbidden\n",
- h0->au_pps_id, pps_id);
+ h->au_pps_id, pps_id);
return AVERROR_INVALIDDATA;
}
- pps = h0->pps_buffers[pps_id];
+ pps = h->pps_buffers[pps_id];
- if (!h0->sps_buffers[pps->sps_id]) {
+ if (!h->sps_buffers[pps->sps_id]) {
av_log(h->avctx, AV_LOG_ERROR,
"non-existing SPS %u referenced\n",
h->pps.sps_id);
return AVERROR_INVALIDDATA;
}
if (first_slice)
- h->pps = *h0->pps_buffers[pps_id];
+ h->pps = *h->pps_buffers[pps_id];
if (pps->sps_id != h->sps.sps_id ||
pps->sps_id != h->current_sps_id ||
- h0->sps_buffers[pps->sps_id]->new) {
+ h->sps_buffers[pps->sps_id]->new) {
if (!first_slice) {
av_log(h->avctx, AV_LOG_ERROR,
return AVERROR_INVALIDDATA;
}
- h->sps = *h0->sps_buffers[h->pps.sps_id];
+ h->sps = *h->sps_buffers[h->pps.sps_id];
if (h->mb_width != h->sps.mb_width ||
h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
|| h->mb_width != h->sps.mb_width
|| h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
));
- if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
+ if (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0)))
must_reinit = 1;
if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
if (h->context_initialized &&
(must_reinit || needs_reinit)) {
- if (h != h0) {
+ if (sl != h->slice_ctx) {
av_log(h->avctx, AV_LOG_ERROR,
"changing width %d -> %d / height %d -> %d on "
"slice %d\n",
h->width, h->avctx->coded_width,
h->height, h->avctx->coded_height,
- h0->current_slice + 1);
+ h->current_slice + 1);
return AVERROR_INVALIDDATA;
}
}
}
if (!h->context_initialized) {
- if (h != h0) {
+ if (sl != h->slice_ctx) {
av_log(h->avctx, AV_LOG_ERROR,
"Cannot (re-)initialize context during parallel decoding.\n");
return AVERROR_PATCHWELCOME;
frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
if (!first_slice) {
- if (h0->frame_num != frame_num) {
+ if (h->frame_num != frame_num) {
av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
- h0->frame_num, frame_num);
+ h->frame_num, frame_num);
return AVERROR_INVALIDDATA;
}
}
- h->mb_mbaff = 0;
+ sl->mb_mbaff = 0;
h->mb_aff_frame = 0;
- last_pic_structure = h0->picture_structure;
- last_pic_droppable = h0->droppable;
+ last_pic_structure = h->picture_structure;
+ last_pic_droppable = h->droppable;
droppable = h->nal_ref_idc == 0;
if (h->sps.frame_mbs_only_flag) {
picture_structure = PICT_FRAME;
h->mb_aff_frame = h->sps.mb_aff;
}
}
- if (h0->current_slice) {
+ if (h->current_slice) {
if (last_pic_structure != picture_structure ||
last_pic_droppable != droppable) {
av_log(h->avctx, AV_LOG_ERROR,
"Changing field mode (%d -> %d) between slices is not allowed\n",
last_pic_structure, h->picture_structure);
return AVERROR_INVALIDDATA;
- } else if (!h0->cur_pic_ptr) {
+ } else if (!h->cur_pic_ptr) {
av_log(h->avctx, AV_LOG_ERROR,
"unset cur_pic_ptr on slice %d\n",
- h0->current_slice + 1);
+ h->current_slice + 1);
return AVERROR_INVALIDDATA;
}
}
h->picture_structure = picture_structure;
h->droppable = droppable;
h->frame_num = frame_num;
- h->mb_field_decoding_flag = picture_structure != PICT_FRAME;
+ sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
- if (h0->current_slice == 0) {
+ if (h->current_slice == 0) {
/* Shorten frame num gaps so we don't have to allocate reference
* frames just to throw them away */
if (h->frame_num != h->prev_frame_num) {
* decode frames as "finished".
* We have to do that before the "dummy" in-between frame allocation,
* since that can modify h->cur_pic_ptr. */
- if (h0->first_field) {
- assert(h0->cur_pic_ptr);
- assert(h0->cur_pic_ptr->f.buf[0]);
- assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
+ if (h->first_field) {
+ assert(h->cur_pic_ptr);
+ assert(h->cur_pic_ptr->f.buf[0]);
+ assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
/* Mark old field/frame as completed */
- if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
- ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
+ if (h->cur_pic_ptr->tf.owner == h->avctx) {
+ ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
last_pic_structure == PICT_BOTTOM_FIELD);
}
/* Previous field is unmatched. Don't display it, but let it
* remain for reference if marked as such. */
if (last_pic_structure != PICT_FRAME) {
- ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
+ ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
last_pic_structure == PICT_TOP_FIELD);
}
} else {
- if (h0->cur_pic_ptr->frame_num != h->frame_num) {
+ if (h->cur_pic_ptr->frame_num != h->frame_num) {
/* This and previous field were reference, but had
* different frame_nums. Consider this field first in
* pair. Throw away previous field except for reference
* purposes. */
if (last_pic_structure != PICT_FRAME) {
- ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
+ ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
last_pic_structure == PICT_TOP_FIELD);
}
} else {
}
}
- while (h->frame_num != h->prev_frame_num && !h0->first_field &&
+ while (h->frame_num != h->prev_frame_num && !h->first_field &&
h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
h->last_pocs[i] = INT_MIN;
ret = h264_frame_start(h);
if (ret < 0) {
- h0->first_field = 0;
+ h->first_field = 0;
return ret;
}
/* See if we have a decoded first field looking for a pair...
* We're using that to see whether to continue decoding in that
* frame, or to allocate a new one. */
- if (h0->first_field) {
- assert(h0->cur_pic_ptr);
- assert(h0->cur_pic_ptr->f.buf[0]);
- assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
+ if (h->first_field) {
+ assert(h->cur_pic_ptr);
+ assert(h->cur_pic_ptr->f.buf[0]);
+ assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
/* figure out if we have a complementary field pair */
if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
/* Previous field is unmatched. Don't display it, but let it
* remain for reference if marked as such. */
- h0->missing_fields ++;
- h0->cur_pic_ptr = NULL;
- h0->first_field = FIELD_PICTURE(h);
+ h->missing_fields ++;
+ h->cur_pic_ptr = NULL;
+ h->first_field = FIELD_PICTURE(h);
} else {
- h0->missing_fields = 0;
- if (h0->cur_pic_ptr->frame_num != h->frame_num) {
- ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
- h0->picture_structure==PICT_BOTTOM_FIELD);
+ h->missing_fields = 0;
+ if (h->cur_pic_ptr->frame_num != h->frame_num) {
+ ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
+ h->picture_structure==PICT_BOTTOM_FIELD);
/* This and the previous field had different frame_nums.
* Consider this field first in pair. Throw away previous
* one except for reference purposes. */
- h0->first_field = 1;
- h0->cur_pic_ptr = NULL;
+ h->first_field = 1;
+ h->cur_pic_ptr = NULL;
} else {
/* Second field in complementary pair */
- h0->first_field = 0;
+ h->first_field = 0;
}
}
} else {
/* Frame or first field in a potentially complementary pair */
- h0->first_field = FIELD_PICTURE(h);
+ h->first_field = FIELD_PICTURE(h);
}
- if (!FIELD_PICTURE(h) || h0->first_field) {
+ if (!FIELD_PICTURE(h) || h->first_field) {
if (h264_frame_start(h) < 0) {
- h0->first_field = 0;
+ h->first_field = 0;
return AVERROR_INVALIDDATA;
}
} else {
memset(h->slice_table, -1,
(h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
}
- h0->last_slice_type = -1;
+ h->last_slice_type = -1;
}
- if (h != h0 && (ret = clone_slice(h, h0)) < 0)
- return ret;
- /* can't be in alloc_tables because linesize isn't known there.
- * FIXME: redo bipred weight to not require extra buffer? */
- for (i = 0; i < h->slice_context_count; i++)
- if (h->thread_context[i]) {
- ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
- if (ret < 0)
- return ret;
- }
h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
return ret;
if (slice_type != AV_PICTURE_TYPE_I &&
- (h0->current_slice == 0 ||
- slice_type != h0->last_slice_type ||
- memcmp(h0->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
+ (h->current_slice == 0 ||
+ slice_type != h->last_slice_type ||
+ memcmp(h->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
ff_h264_fill_default_ref_list(h, sl);
}
// further down the line. This may break decoding if the first slice is
// corrupt, thus we only do this if frame-mt is enabled.
if (h->nal_ref_idc) {
- ret = ff_h264_decode_ref_pic_marking(h0, &sl->gb,
+ ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
!(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
- h0->current_slice == 0);
+ h->current_slice == 0);
if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
return AVERROR_INVALIDDATA;
}
h->nal_ref_idc == 0))
sl->deblocking_filter = 0;
- if (sl->deblocking_filter == 1 && h0->max_contexts > 1) {
+ if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
/* Cheat slightly for speed:
* Do not bother to deblock across slices. */
sl->deblocking_filter = 2;
} else {
- h0->max_contexts = 1;
- if (!h0->single_decode_warning) {
+ h->max_contexts = 1;
+ if (!h->single_decode_warning) {
av_log(h->avctx, AV_LOG_INFO,
"Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
"To parallelize slice decoding you need video encoded with disable_deblocking_filter_idc set to 2 (deblock only edges that do not cross slices).\n"
"Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
"but will generate non-standard-compliant output.\n");
- h0->single_decode_warning = 1;
+ h->single_decode_warning = 1;
}
- if (h != h0) {
+ if (sl != h->slice_ctx) {
av_log(h->avctx, AV_LOG_ERROR,
"Deblocking switched inside frame.\n");
return SLICE_SINGLETHREAD;
h->pps.chroma_qp_index_offset[1]) +
6 * (h->sps.bit_depth_luma - 8);
- h0->last_slice_type = slice_type;
- memcpy(h0->last_ref_count, sl->ref_count, sizeof(h0->last_ref_count));
- sl->slice_num = ++h0->current_slice;
+ h->last_slice_type = slice_type;
+ memcpy(h->last_ref_count, sl->ref_count, sizeof(h->last_ref_count));
+ sl->slice_num = ++h->current_slice;
if (sl->slice_num)
- h0->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
- if ( h0->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
- && h0->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
+ h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
+ if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
+ && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
&& sl->slice_num >= MAX_SLICES) {
//in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", sl->slice_num, MAX_SLICES);
(sl->ref_list[j][i].reference & 3);
}
- h0->au_pps_id = pps_id;
+ h->au_pps_id = pps_id;
h->sps.new =
- h0->sps_buffers[h->pps.sps_id]->new = 0;
+ h->sps_buffers[h->pps.sps_id]->new = 0;
h->current_sps_id = h->pps.sps_id;
if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
}
}
-static av_always_inline void fill_filter_caches_inter(H264Context *h,
+static av_always_inline void fill_filter_caches_inter(const H264Context *h,
H264SliceContext *sl,
int mb_type, int top_xy,
int left_xy[LEFT_MBS],
if (USES_LIST(top_type, list)) {
const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
const int b8_xy = 4 * top_xy + 2;
- int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+ int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
ref_cache[0 - 1 * 8] =
ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
if (USES_LIST(left_type[LTOP], list)) {
const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
const int b8_xy = 4 * left_xy[LTOP] + 1;
- int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+ int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
{
int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
- int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+ int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
AV_WN32A(&ref_cache[0 * 8], ref01);
*
* @return non zero if the loop filter can be skipped
*/
-static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
+static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
{
const int mb_xy = sl->mb_xy;
int top_xy, left_xy[LEFT_MBS];
uint8_t *nnz;
uint8_t *nnz_cache;
- top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
+ top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
/* Wow, what a mess, why didn't they simplify the interlacing & intra
* stuff, I can't imagine that these complex rules are worth it. */
return 0;
}
-static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int end_x)
+static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
{
uint8_t *dest_y, *dest_cb, *dest_cr;
int linesize, uvlinesize, mb_x, mb_y;
sl->list_count = h->list_counts[mb_xy];
if (FRAME_MBAFF(h))
- h->mb_mbaff =
- h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
+ sl->mb_mbaff =
+ sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
sl->mb_x = mb_x;
sl->mb_y = mb_y;
mb_y * h->uvlinesize * block_h;
// FIXME simplify above
- if (MB_FIELD(h)) {
+ if (MB_FIELD(sl)) {
linesize = sl->mb_linesize = h->linesize * 2;
uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2;
if (mb_y & 1) { // FIXME move out of this function?
sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
}
-static void predict_field_decoding_flag(H264Context *h, H264SliceContext *sl)
+static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
{
const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
h->cur_pic.mb_type[mb_xy - 1] :
(h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
- h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
+ sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
}
/**
* Draw edges and report progress for the last MB row.
*/
-static void decode_finish_row(H264Context *h, H264SliceContext *sl)
+static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
{
int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
ff_h264_draw_horiz_band(h, sl, top, height);
- if (h->droppable || h->er.error_occurred)
+ if (h->droppable || sl->er.error_occurred)
return;
ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
h->picture_structure == PICT_BOTTOM_FIELD);
}
-static void er_add_slice(H264Context *h, H264SliceContext *sl,
+static void er_add_slice(H264SliceContext *sl,
int startx, int starty,
int endx, int endy, int status)
{
if (CONFIG_ERROR_RESILIENCE) {
- ERContext *er = &h->er;
+ ERContext *er = &sl->er;
ff_er_add_slice(er, startx, starty, endx, endy, status);
}
static int decode_slice(struct AVCodecContext *avctx, void *arg)
{
H264SliceContext *sl = arg;
- H264Context *h = sl->h264;
+ const H264Context *h = avctx->priv_data;
int lf_x_start = sl->mb_x;
+ int ret;
+
+ ret = alloc_scratch_buffers(sl, h->linesize);
+ if (ret < 0)
+ return ret;
sl->mb_skip_run = -1;
avctx->codec_id != AV_CODEC_ID_H264 ||
(CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
- if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
+ if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && sl->er.error_status_table) {
const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
if (start_i) {
- int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
+ int prev_status = sl->er.error_status_table[sl->er.mb_index2xy[start_i - 1]];
prev_status &= ~ VP_START;
if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
- h->er.error_occurred = 1;
+ sl->er.error_occurred = 1;
}
}
if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
- er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
+ er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
sl->mb_y, ER_MB_END);
if (sl->mb_x >= lf_x_start)
loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
"error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
sl->mb_x, sl->mb_y,
sl->cabac.bytestream_end - sl->cabac.bytestream);
- er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+ er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
sl->mb_y, ER_MB_ERROR);
return AVERROR_INVALIDDATA;
}
if (eos || sl->mb_y >= h->mb_height) {
tprintf(h->avctx, "slice end %d %d\n",
get_bits_count(&sl->gb), sl->gb.size_in_bits);
- er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
+ er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
sl->mb_y, ER_MB_END);
if (sl->mb_x > lf_x_start)
loop_filter(h, sl, lf_x_start, sl->mb_x);
if (ret < 0) {
av_log(h->avctx, AV_LOG_ERROR,
"error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
- er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+ er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
sl->mb_y, ER_MB_ERROR);
return ret;
}
if ( get_bits_left(&sl->gb) == 0
|| get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
- er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
+ er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
sl->mb_x - 1, sl->mb_y, ER_MB_END);
return 0;
} else {
- er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
+ er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
sl->mb_x, sl->mb_y, ER_MB_END);
return AVERROR_INVALIDDATA;
get_bits_count(&sl->gb), sl->gb.size_in_bits);
if (get_bits_left(&sl->gb) == 0) {
- er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
+ er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
sl->mb_x - 1, sl->mb_y, ER_MB_END);
if (sl->mb_x > lf_x_start)
loop_filter(h, sl, lf_x_start, sl->mb_x);
return 0;
} else {
- er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+ er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
sl->mb_y, ER_MB_ERROR);
return AVERROR_INVALIDDATA;
int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
{
AVCodecContext *const avctx = h->avctx;
- H264Context *hx;
H264SliceContext *sl;
int i;
} else {
av_assert0(context_count > 0);
for (i = 1; i < context_count; i++) {
- hx = h->thread_context[i];
+ sl = &h->slice_ctx[i];
if (CONFIG_ERROR_RESILIENCE) {
- hx->er.error_count = 0;
+ sl->er.error_count = 0;
}
- hx->x264_build = h->x264_build;
}
avctx->execute(avctx, decode_slice, h->slice_ctx,
NULL, context_count, sizeof(h->slice_ctx[0]));
/* pull back stuff from slices to master context */
- hx = h->thread_context[context_count - 1];
sl = &h->slice_ctx[context_count - 1];
h->mb_y = sl->mb_y;
- h->droppable = hx->droppable;
- h->picture_structure = hx->picture_structure;
if (CONFIG_ERROR_RESILIENCE) {
for (i = 1; i < context_count; i++)
- h->er.error_count += h->thread_context[i]->er.error_count;
+ h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
}
}