X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fh264_slice.c;h=49fd89135dc791c1f184db29e4d7126ae579cad5;hb=1da2a20763ae9ca579d5fd20763065871ddf6311;hp=cc1b8965edc628a5bf2803b50643377f6b09b96a;hpb=9d76dd952ffd9fffc7f9fb5cc5af6d093ba54521;p=ffmpeg diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c index cc1b8965edc..49fd89135dc 100644 --- a/libavcodec/h264_slice.c +++ b/libavcodec/h264_slice.c @@ -31,7 +31,6 @@ #include "internal.h" #include "cabac.h" #include "cabac_functions.h" -#include "dsputil.h" #include "error_resilience.h" #include "avcodec.h" #include "h264.h" @@ -144,40 +143,6 @@ static const uint8_t dequant8_coeff_init[6][6] = { { 36, 32, 58, 34, 46, 43 }, }; -static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = { -#if CONFIG_H264_DXVA2_HWACCEL - AV_PIX_FMT_DXVA2_VLD, -#endif -#if CONFIG_H264_VAAPI_HWACCEL - AV_PIX_FMT_VAAPI_VLD, -#endif -#if CONFIG_H264_VDA_HWACCEL - AV_PIX_FMT_VDA_VLD, -#endif -#if CONFIG_H264_VDPAU_HWACCEL - AV_PIX_FMT_VDPAU, -#endif - AV_PIX_FMT_YUV420P, - AV_PIX_FMT_NONE -}; - -static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = { -#if CONFIG_H264_DXVA2_HWACCEL - AV_PIX_FMT_DXVA2_VLD, -#endif -#if CONFIG_H264_VAAPI_HWACCEL - AV_PIX_FMT_VAAPI_VLD, -#endif -#if CONFIG_H264_VDA_HWACCEL - AV_PIX_FMT_VDA_VLD, -#endif -#if CONFIG_H264_VDPAU_HWACCEL - AV_PIX_FMT_VDPAU, -#endif - AV_PIX_FMT_YUVJ420P, - AV_PIX_FMT_NONE -}; - static void release_unused_pictures(H264Context *h, int remove_current) { @@ -185,28 +150,39 @@ static void release_unused_pictures(H264Context *h, int remove_current) /* release non reference frames */ for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) { - if (h->DPB[i].f.buf[0] && !h->DPB[i].reference && + if (h->DPB[i].f->buf[0] && !h->DPB[i].reference && (remove_current || &h->DPB[i] != h->cur_pic_ptr)) { ff_h264_unref_picture(h, &h->DPB[i]); } } } -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); } @@ -224,9 +200,9 @@ static int init_table_pools(H264Context *h) av_buffer_allocz); h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) * sizeof(uint32_t), av_buffer_allocz); - h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) * - sizeof(int16_t), av_buffer_allocz); - h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz); + h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) * + sizeof(int16_t), av_buffer_allocz); + h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz); if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool || !h->ref_index_pool) { @@ -244,22 +220,19 @@ static int alloc_picture(H264Context *h, H264Picture *pic) { int i, ret = 0; - av_assert0(!pic->f.data[0]); + av_assert0(!pic->f->data[0]); - pic->tf.f = &pic->f; + pic->tf.f = pic->f; ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ? AV_GET_BUFFER_FLAG_REF : 0); if (ret < 0) goto fail; - h->linesize = pic->f.linesize[0]; - h->uvlinesize = pic->f.linesize[1]; - if (h->avctx->hwaccel) { const AVHWAccel *hwaccel = h->avctx->hwaccel; av_assert0(!pic->hwaccel_picture_private); - if (hwaccel->priv_data_size) { - pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size); + if (hwaccel->frame_priv_data_size) { + pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size); if (!pic->hwaccel_priv_buf) return AVERROR(ENOMEM); pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data; @@ -298,9 +271,7 @@ fail: static inline int pic_is_unused(H264Context *h, H264Picture *pic) { - if (!pic->f.buf[0]) - return 1; - if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF)) + if (!pic->f->buf[0]) return 1; return 0; } @@ -316,11 +287,6 @@ static int find_unused_picture(H264Context *h) if (i == H264_MAX_PICTURE_COUNT) return AVERROR_INVALIDDATA; - if (h->DPB[i].needs_realloc) { - h->DPB[i].needs_realloc = 0; - ff_h264_unref_picture(h, &h->DPB[i]); - } - return i; } @@ -395,33 +361,8 @@ void h264_init_dequant_tables(H264Context *h) } } -/** - * Mimic alloc_tables(), but for every context thread. - */ -static void clone_tables(H264Context *dst, H264Context *src, int i) -{ - dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride; - dst->non_zero_count = src->non_zero_count; - dst->slice_table = src->slice_table; - dst->cbp_table = src->cbp_table; - dst->mb2b_xy = src->mb2b_xy; - dst->mb2br_xy = src->mb2br_xy; - dst->chroma_pred_mode_table = src->chroma_pred_mode_table; - dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride; - dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride; - dst->direct_table = src->direct_table; - dst->list_counts = src->list_counts; - 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); -} - #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size)))) -#undef REBASE_PICTURE + #define REBASE_PICTURE(pic, new_ctx, old_ctx) \ ((pic && pic >= old_ctx->DPB && \ pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \ @@ -466,14 +407,14 @@ static int copy_parameter_set(void **to, void **from, int count, int size) memcpy(&to->start_field, &from->start_field, \ (char *)&to->end_field - (char *)&to->start_field) -static int h264_slice_header_init(H264Context *h, int reinit); +static int h264_slice_header_init(H264Context *h); int ff_h264_update_thread_context(AVCodecContext *dst, const AVCodecContext *src) { H264Context *h = dst->priv_data, *h1 = src->priv_data; int inited = h->context_initialized, err = 0; - int context_reinitialized = 0; + int need_reinit = 0; int i, ret; if (dst == src || !h1->context_initialized) @@ -487,14 +428,22 @@ int ff_h264_update_thread_context(AVCodecContext *dst, h->sps.bit_depth_luma != h1->sps.bit_depth_luma || h->sps.chroma_format_idc != h1->sps.chroma_format_idc || h->sps.colorspace != h1->sps.colorspace)) { + need_reinit = 1; + } - /* set bits_per_raw_sample to the previous value. the check for changed - * bit depth in h264_set_parameter_from_sps() uses it and sets it to - * the current value */ - h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma; - - av_freep(&h->bipred_scratchpad); + // SPS/PPS + if ((ret = copy_parameter_set((void **)h->sps_buffers, + (void **)h1->sps_buffers, + MAX_SPS_COUNT, sizeof(SPS))) < 0) + return ret; + h->sps = h1->sps; + if ((ret = copy_parameter_set((void **)h->pps_buffers, + (void **)h1->pps_buffers, + MAX_PPS_COUNT, sizeof(PPS))) < 0) + return ret; + h->pps = h1->pps; + if (need_reinit || !inited) { h->width = h1->width; h->height = h1->height; h->mb_height = h1->mb_height; @@ -503,71 +452,15 @@ int ff_h264_update_thread_context(AVCodecContext *dst, h->mb_stride = h1->mb_stride; h->b_stride = h1->b_stride; - if ((err = h264_slice_header_init(h, 1)) < 0) { + if ((err = h264_slice_header_init(h)) < 0) { av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed"); return err; } - context_reinitialized = 1; - - /* update linesize on resize. The decoder doesn't - * necessarily call h264_frame_start in the new thread */ - h->linesize = h1->linesize; - h->uvlinesize = h1->uvlinesize; /* copy block_offset since frame_start may not be called */ memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset)); } - if (!inited) { - for (i = 0; i < MAX_SPS_COUNT; i++) - av_freep(h->sps_buffers + i); - - for (i = 0; i < MAX_PPS_COUNT; i++) - av_freep(h->pps_buffers + i); - - memcpy(h, h1, sizeof(*h1)); - 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->mb, 0, sizeof(h->mb)); - memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc)); - memset(&h->mb_padding, 0, sizeof(h->mb_padding)); - h->context_initialized = 0; - - memset(&h->cur_pic, 0, sizeof(h->cur_pic)); - av_frame_unref(&h->cur_pic.f); - h->cur_pic.tf.f = &h->cur_pic.f; - - h->avctx = dst; - h->DPB = NULL; - h->qscale_table_pool = NULL; - h->mb_type_pool = NULL; - h->ref_index_pool = NULL; - h->motion_val_pool = NULL; - - ret = ff_h264_alloc_tables(h); - if (ret < 0) { - av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n"); - return ret; - } - ret = ff_h264_context_init(h); - if (ret < 0) { - av_log(dst, AV_LOG_ERROR, "context_init() failed.\n"); - return ret; - } - - for (i = 0; i < 2; i++) { - h->rbsp_buffer[i] = NULL; - h->rbsp_buffer_size[i] = 0; - } - h->bipred_scratchpad = NULL; - h->edge_emu_buffer = NULL; - - h->thread_context[0] = h; - - h->context_initialized = 1; - } - h->avctx->coded_height = h1->avctx->coded_height; h->avctx->coded_width = h1->avctx->coded_width; h->avctx->width = h1->avctx->width; @@ -575,46 +468,32 @@ int ff_h264_update_thread_context(AVCodecContext *dst, h->coded_picture_number = h1->coded_picture_number; h->first_field = h1->first_field; h->picture_structure = h1->picture_structure; - h->qscale = h1->qscale; h->droppable = h1->droppable; h->low_delay = h1->low_delay; for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) { ff_h264_unref_picture(h, &h->DPB[i]); - if (h1->DPB[i].f.buf[0] && + if (h1->DPB[i].f->buf[0] && (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0) return ret; } h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1); ff_h264_unref_picture(h, &h->cur_pic); - if ((ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0) - return ret; + if (h1->cur_pic.f->buf[0]) { + ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic); + if (ret < 0) + return ret; + } + h->enable_er = h1->enable_er; h->workaround_bugs = h1->workaround_bugs; h->low_delay = h1->low_delay; h->droppable = h1->droppable; - /* frame_start may not be called for the next thread (if it's decoding - * a bottom field) so this has to be allocated here */ - err = alloc_scratch_buffers(h, h1->linesize); - if (err < 0) - return err; - // extradata/NAL handling h->is_avc = h1->is_avc; - - // SPS/PPS - if ((ret = copy_parameter_set((void **)h->sps_buffers, - (void **)h1->sps_buffers, - MAX_SPS_COUNT, sizeof(SPS))) < 0) - return ret; - h->sps = h1->sps; - if ((ret = copy_parameter_set((void **)h->pps_buffers, - (void **)h1->pps_buffers, - MAX_PPS_COUNT, sizeof(PPS))) < 0) - return ret; - h->pps = h1->pps; + h->nal_length_size = h1->nal_length_size; // Dequantization matrices // FIXME these are big - can they be only copied when PPS changes? @@ -631,10 +510,10 @@ int ff_h264_update_thread_context(AVCodecContext *dst, h->dequant_coeff_pps = h1->dequant_coeff_pps; // POC timing - copy_fields(h, h1, poc_lsb, redundant_pic_count); + copy_fields(h, h1, poc_lsb, default_ref_list); // reference lists - copy_fields(h, h1, short_ref, cabac_init_idc); + copy_fields(h, h1, short_ref, current_slice); copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1); copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1); @@ -643,9 +522,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst, h->last_slice_type = h1->last_slice_type; - if (context_reinitialized) - ff_h264_set_parameter_from_sps(h); - if (!h->cur_pic_ptr) return 0; @@ -656,7 +532,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst, } h->prev_frame_num_offset = h->frame_num_offset; h->prev_frame_num = h->frame_num; - h->outputed_poc = h->next_outputed_poc; h->recovery_frame = h1->recovery_frame; h->frame_recovered = h1->frame_recovered; @@ -681,14 +556,14 @@ static int h264_frame_start(H264Context *h) pic = &h->DPB[i]; pic->reference = h->droppable ? 0 : h->picture_structure; - pic->f.coded_picture_number = h->coded_picture_number++; + pic->f->coded_picture_number = h->coded_picture_number++; pic->field_picture = h->picture_structure != PICT_FRAME; /* * Zero key_frame here; IDR markings per slice in frame or fields are ORed * in later. * See decode_nal_units(). */ - pic->f.key_frame = 0; + pic->f->key_frame = 0; pic->mmco_reset = 0; pic->recovered = 0; @@ -700,31 +575,20 @@ static int h264_frame_start(H264Context *h) 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); - - assert(h->linesize && h->uvlinesize); + if (CONFIG_ERROR_RESILIENCE && h->enable_er) + ff_er_frame_start(&h->slice_ctx[0].er); for (i = 0; i < 16; i++) { - h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3); - h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3); + h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3); + h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3); } for (i = 0; i < 16; i++) { h->block_offset[16 + i] = - h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); + h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3); h->block_offset[48 + 16 + i] = - h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); + h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3); } - /* 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; - } - /* Some macroblocks can be accessed before they're available in case * of lost slices, MBAFF or threading. */ memset(h->slice_table, -1, @@ -732,7 +596,7 @@ static int h264_frame_start(H264Context *h) /* We mark the current picture as non-reference after allocating it, so * that if we break out due to an error it can be released automatically - * in the next ff_MPV_frame_start(). + * in the next ff_mpv_frame_start(). */ h->cur_pic_ptr->reference = 0; @@ -745,7 +609,8 @@ static int h264_frame_start(H264Context *h) return 0; } -static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, +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, int simple) @@ -761,13 +626,13 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, src_cr -= uvlinesize; if (!simple && FRAME_MBAFF(h)) { - if (h->mb_y & 1) { - if (!MB_MBAFF(h)) { - top_border = h->top_borders[0][h->mb_x]; + if (sl->mb_y & 1) { + 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); - if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) { + if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) { if (chroma444) { if (pixel_shift) { AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize); @@ -797,20 +662,20 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, } } } - } else if (MB_MBAFF(h)) { + } else if (MB_MBAFF(sl)) { top_idx = 0; } else return; } - top_border = h->top_borders[top_idx][h->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); if (pixel_shift) AV_COPY128(top_border + 16, src_y + 16 * linesize + 16); - if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) { + if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) { if (chroma444) { if (pixel_shift) { AV_COPY128(top_border + 32, src_cb + 16 * linesize); @@ -846,13 +711,13 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, * @param field 0/1 initialize the weight for interlaced MBAFF * -1 initializes the rest */ -static void implicit_weight_table(H264Context *h, 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; for (i = 0; i < 2; i++) { - h->luma_weight_flag[i] = 0; - h->chroma_weight_flag[i] = 0; + sl->luma_weight_flag[i] = 0; + sl->chroma_weight_flag[i] = 0; } if (field < 0) { @@ -861,36 +726,36 @@ static void implicit_weight_table(H264Context *h, int field) } else { cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1]; } - if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) && - h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) { - h->use_weight = 0; - h->use_weight_chroma = 0; + if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) && + sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) { + sl->use_weight = 0; + sl->use_weight_chroma = 0; return; } ref_start = 0; - ref_count0 = h->ref_count[0]; - ref_count1 = h->ref_count[1]; + ref_count0 = sl->ref_count[0]; + ref_count1 = sl->ref_count[1]; } else { cur_poc = h->cur_pic_ptr->field_poc[field]; ref_start = 16; - ref_count0 = 16 + 2 * h->ref_count[0]; - ref_count1 = 16 + 2 * h->ref_count[1]; + ref_count0 = 16 + 2 * sl->ref_count[0]; + ref_count1 = 16 + 2 * sl->ref_count[1]; } - h->use_weight = 2; - h->use_weight_chroma = 2; - h->luma_log2_weight_denom = 5; - h->chroma_log2_weight_denom = 5; + sl->use_weight = 2; + sl->use_weight_chroma = 2; + sl->luma_log2_weight_denom = 5; + sl->chroma_log2_weight_denom = 5; for (ref0 = ref_start; ref0 < ref_count0; ref0++) { - int poc0 = h->ref_list[0][ref0].poc; + int poc0 = sl->ref_list[0][ref0].poc; for (ref1 = ref_start; ref1 < ref_count1; ref1++) { int w = 32; - if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) { - int poc1 = h->ref_list[1][ref1].poc; - int td = av_clip(poc1 - poc0, -128, 127); + if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) { + int poc1 = sl->ref_list[1][ref1].poc; + int td = av_clip_int8(poc1 - poc0); if (td) { - int tb = av_clip(cur_poc - poc0, -128, 127); + int tb = av_clip_int8(cur_poc - poc0); int tx = (16384 + (FFABS(td) >> 1)) / td; int dist_scale_factor = (tb * tx + 32) >> 8; if (dist_scale_factor >= -64 && dist_scale_factor <= 128) @@ -898,10 +763,10 @@ static void implicit_weight_table(H264Context *h, int field) } } if (field < 0) { - h->implicit_weight[ref0][ref1][0] = - h->implicit_weight[ref0][ref1][1] = w; + sl->implicit_weight[ref0][ref1][0] = + sl->implicit_weight[ref0][ref1][1] = w; } else { - h->implicit_weight[ref0][ref1][field] = w; + sl->implicit_weight[ref0][ref1][field] = w; } } } @@ -944,75 +809,75 @@ static void init_scan_tables(H264Context *h) } } -/** - * 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) { +#define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \ + CONFIG_H264_D3D11VA_HWACCEL + \ + CONFIG_H264_VAAPI_HWACCEL + \ + (CONFIG_H264_VDA_HWACCEL * 2) + \ + CONFIG_H264_VDPAU_HWACCEL) + enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts; + const enum AVPixelFormat *choices = pix_fmts; + switch (h->sps.bit_depth_luma) { case 9: if (CHROMA444(h)) { if (h->avctx->colorspace == AVCOL_SPC_RGB) { - return AV_PIX_FMT_GBRP9; + *fmt++ = AV_PIX_FMT_GBRP9; } else - return AV_PIX_FMT_YUV444P9; + *fmt++ = AV_PIX_FMT_YUV444P9; } else if (CHROMA422(h)) - return AV_PIX_FMT_YUV422P9; + *fmt++ = AV_PIX_FMT_YUV422P9; else - return AV_PIX_FMT_YUV420P9; + *fmt++ = AV_PIX_FMT_YUV420P9; break; case 10: if (CHROMA444(h)) { if (h->avctx->colorspace == AVCOL_SPC_RGB) { - return AV_PIX_FMT_GBRP10; + *fmt++ = AV_PIX_FMT_GBRP10; } else - return AV_PIX_FMT_YUV444P10; + *fmt++ = AV_PIX_FMT_YUV444P10; } else if (CHROMA422(h)) - return AV_PIX_FMT_YUV422P10; + *fmt++ = AV_PIX_FMT_YUV422P10; else - return AV_PIX_FMT_YUV420P10; + *fmt++ = AV_PIX_FMT_YUV420P10; break; case 8: +#if CONFIG_H264_VDPAU_HWACCEL + *fmt++ = AV_PIX_FMT_VDPAU; +#endif if (CHROMA444(h)) { - if (h->avctx->colorspace == AVCOL_SPC_RGB) { - return AV_PIX_FMT_GBRP; - } else - return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P - : AV_PIX_FMT_YUV444P; + if (h->avctx->colorspace == AVCOL_SPC_RGB) + *fmt++ = AV_PIX_FMT_GBRP; + else if (h->avctx->color_range == AVCOL_RANGE_JPEG) + *fmt++ = AV_PIX_FMT_YUVJ444P; + else + *fmt++ = AV_PIX_FMT_YUV444P; } else if (CHROMA422(h)) { - return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P - : AV_PIX_FMT_YUV422P; + if (h->avctx->color_range == AVCOL_RANGE_JPEG) + *fmt++ = AV_PIX_FMT_YUVJ422P; + else + *fmt++ = AV_PIX_FMT_YUV422P; } else { - return h->avctx->get_format(h->avctx, h->avctx->codec->pix_fmts ? - h->avctx->codec->pix_fmts : - h->avctx->color_range == AVCOL_RANGE_JPEG ? - h264_hwaccel_pixfmt_list_jpeg_420 : - h264_hwaccel_pixfmt_list_420); +#if CONFIG_H264_DXVA2_HWACCEL + *fmt++ = AV_PIX_FMT_DXVA2_VLD; +#endif +#if CONFIG_H264_D3D11VA_HWACCEL + *fmt++ = AV_PIX_FMT_D3D11VA_VLD; +#endif +#if CONFIG_H264_VAAPI_HWACCEL + *fmt++ = AV_PIX_FMT_VAAPI_VLD; +#endif +#if CONFIG_H264_VDA_HWACCEL + *fmt++ = AV_PIX_FMT_VDA_VLD; + *fmt++ = AV_PIX_FMT_VDA; +#endif + if (h->avctx->codec->pix_fmts) + choices = h->avctx->codec->pix_fmts; + else if (h->avctx->color_range == AVCOL_RANGE_JPEG) + *fmt++ = AV_PIX_FMT_YUVJ420P; + else + *fmt++ = AV_PIX_FMT_YUV420P; } break; default: @@ -1020,6 +885,10 @@ static enum AVPixelFormat get_pixel_format(H264Context *h) "Unsupported bit depth %d\n", h->sps.bit_depth_luma); return AVERROR_INVALIDDATA; } + + *fmt = AV_PIX_FMT_NONE; + + return ff_get_format(h->avctx, choices); } /* export coded and cropped frame dimensions to AVCodecContext */ @@ -1029,9 +898,8 @@ static int init_dimensions(H264Context *h) int height = h->height - (h->sps.crop_top + h->sps.crop_bottom); /* handle container cropping */ - if (!h->sps.crop && - FFALIGN(h->avctx->width, 16) == h->width && - FFALIGN(h->avctx->height, 16) == h->height) { + if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) && + FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16)) { width = h->avctx->width; height = h->avctx->height; } @@ -1043,7 +911,10 @@ static int init_dimensions(H264Context *h) return AVERROR_INVALIDDATA; av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n"); - h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0; + h->sps.crop_bottom = + h->sps.crop_top = + h->sps.crop_right = + h->sps.crop_left = h->sps.crop = 0; width = h->width; @@ -1058,15 +929,14 @@ static int init_dimensions(H264Context *h) return 0; } -static int h264_slice_header_init(H264Context *h, int reinit) +static int h264_slice_header_init(H264Context *h) { int nb_slices = (HAVE_THREADS && h->avctx->active_thread_type & FF_THREAD_SLICE) ? h->avctx->thread_count : 1; int i, ret; - h->avctx->sample_aspect_ratio = h->sps.sar; - av_assert0(h->avctx->sample_aspect_ratio.den); + ff_set_sar(h->avctx, h->sps.sar); av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt, &h->chroma_x_shift, &h->chroma_y_shift); @@ -1074,14 +944,12 @@ static int h264_slice_header_init(H264Context *h, int reinit) int64_t den = h->sps.time_scale; if (h->x264_build < 44U) den *= 2; - av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den, + av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num, h->sps.num_units_in_tick, den, 1 << 30); } - h->avctx->hwaccel = ff_find_hwaccel(h->avctx); + ff_h264_free_tables(h); - if (reinit) - ff_h264_free_tables(h, 0); h->first_field = 0; h->prev_interlaced_frame = 1; @@ -1092,6 +960,25 @@ static int h264_slice_header_init(H264Context *h, int reinit) return ret; } + if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 10) { + av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n", + h->sps.bit_depth_luma); + return AVERROR_INVALIDDATA; + } + + h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma; + h->pixel_shift = h->sps.bit_depth_luma > 8; + h->chroma_format_idc = h->sps.chroma_format_idc; + h->bit_depth_luma = h->sps.bit_depth_luma; + + ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, + h->sps.chroma_format_idc); + ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma); + ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma); + ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma, + h->sps.chroma_format_idc); + ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma); + if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) { int max_slices; if (h->mb_height) @@ -1105,54 +992,25 @@ static int h264_slice_header_init(H264Context *h, int reinit) 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"); return ret; } } else { - for (i = 1; i < h->slice_context_count; i++) { - H264Context *c; - c = h->thread_context[i] = av_mallocz(sizeof(H264Context)); - if (!c) - return AVERROR(ENOMEM); - c->avctx = h->avctx; - c->dsp = h->dsp; - c->vdsp = h->vdsp; - c->h264dsp = h->h264dsp; - c->h264qpel = h->h264qpel; - c->h264chroma = h->h264chroma; - c->sps = h->sps; - c->pps = h->pps; - c->pixel_shift = h->pixel_shift; - c->width = h->width; - c->height = h->height; - c->linesize = h->linesize; - c->uvlinesize = h->uvlinesize; - c->chroma_x_shift = h->chroma_x_shift; - c->chroma_y_shift = h->chroma_y_shift; - c->qscale = h->qscale; - c->droppable = h->droppable; - c->data_partitioning = h->data_partitioning; - c->low_delay = h->low_delay; - c->mb_width = h->mb_width; - c->mb_height = h->mb_height; - c->mb_stride = h->mb_stride; - c->mb_num = h->mb_num; - c->flags = h->flags; - c->workaround_bugs = h->workaround_bugs; - c->pict_type = h->pict_type; - - init_scan_tables(c); - clone_tables(c, h, i); - c->context_initialized = 1; - } + for (i = 0; i < h->slice_context_count; i++) { + H264SliceContext *sl = &h->slice_ctx[i]; + + sl->h264 = h; + sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride; + sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride; + sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride; - 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, sl)) < 0) { av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n"); return ret; } + } } h->context_initialized = 1; @@ -1165,12 +1023,10 @@ static int h264_slice_header_init(H264Context *h, int reinit) * 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, H264Context *h0) +int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl) { unsigned int first_mb_in_slice; unsigned int pps_id; @@ -1180,19 +1036,21 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) int last_pic_structure, last_pic_droppable; int needs_reinit = 0; int field_pic_flag, bottom_field_flag; + int frame_num, droppable, picture_structure; + int mb_aff_frame = 0; h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab; h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab; - first_mb_in_slice = get_ue_golomb(&h->gb); + first_mb_in_slice = get_ue_golomb(&sl->gb); if (first_mb_in_slice == 0) { // FIXME better field boundary detection - if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) { - ff_h264_field_end(h, 1); + 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); @@ -1201,50 +1059,56 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) } } - slice_type = get_ue_golomb_31(&h->gb); + slice_type = get_ue_golomb_31(&sl->gb); if (slice_type > 9) { av_log(h->avctx, AV_LOG_ERROR, - "slice type %d too large at %d %d\n", - slice_type, h->mb_x, h->mb_y); + "slice type %d too large at %d\n", + slice_type, first_mb_in_slice); return AVERROR_INVALIDDATA; } if (slice_type > 4) { slice_type -= 5; - h->slice_type_fixed = 1; + sl->slice_type_fixed = 1; } else - h->slice_type_fixed = 0; + sl->slice_type_fixed = 0; slice_type = golomb_to_pict_type[slice_type]; if (slice_type == AV_PICTURE_TYPE_I || - (h0->current_slice != 0 && slice_type == h0->last_slice_type)) { + (h->current_slice != 0 && slice_type == h->last_slice_type)) { default_ref_list_done = 1; } - h->slice_type = slice_type; - h->slice_type_nos = slice_type & 3; + sl->slice_type = slice_type; + sl->slice_type_nos = slice_type & 3; if (h->nal_unit_type == NAL_IDR_SLICE && - h->slice_type_nos != AV_PICTURE_TYPE_I) { + sl->slice_type_nos != AV_PICTURE_TYPE_I) { av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n"); return AVERROR_INVALIDDATA; } // to make a few old functions happy, it's wrong though - h->pict_type = h->slice_type; + if (!h->setup_finished) + h->pict_type = sl->slice_type; - pps_id = get_ue_golomb(&h->gb); + pps_id = get_ue_golomb(&sl->gb); if (pps_id >= MAX_PPS_COUNT) { 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; } - h->pps = *h0->pps_buffers[pps_id]; + if (!h->setup_finished) { + h->pps = *h->pps_buffers[pps_id]; + } else if (h->dequant_coeff_pps != pps_id) { + av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n"); + return AVERROR_INVALIDDATA; + } - if (!h0->sps_buffers[h->pps.sps_id]) { + if (!h->sps_buffers[h->pps.sps_id]) { av_log(h->avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id); @@ -1252,65 +1116,77 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) } if (h->pps.sps_id != h->sps.sps_id || - h0->sps_buffers[h->pps.sps_id]->new) { - h0->sps_buffers[h->pps.sps_id]->new = 0; + h->sps_buffers[h->pps.sps_id]->new) { + h->sps_buffers[h->pps.sps_id]->new = 0; - h->sps = *h0->sps_buffers[h->pps.sps_id]; + h->sps = *h->sps_buffers[h->pps.sps_id]; if (h->bit_depth_luma != h->sps.bit_depth_luma || - h->chroma_format_idc != h->sps.chroma_format_idc) { - h->bit_depth_luma = h->sps.bit_depth_luma; - h->chroma_format_idc = h->sps.chroma_format_idc; + h->chroma_format_idc != h->sps.chroma_format_idc) needs_reinit = 1; + + if (h->flags & AV_CODEC_FLAG_LOW_DELAY || + (h->sps.bitstream_restriction_flag && + !h->sps.num_reorder_frames)) { + if (h->avctx->has_b_frames > 1 || h->delayed_pic[0]) + av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. " + "Reenabling low delay requires a codec flush.\n"); + else + h->low_delay = 1; } - if ((ret = ff_h264_set_parameter_from_sps(h)) < 0) - return ret; + + if (h->avctx->has_b_frames < 2) + h->avctx->has_b_frames = !h->low_delay; + } - h->avctx->profile = ff_h264_get_profile(&h->sps); - h->avctx->level = h->sps.level_idc; - h->avctx->refs = h->sps.ref_frame_count; + if (!h->setup_finished) { + h->avctx->profile = ff_h264_get_profile(&h->sps); + h->avctx->level = h->sps.level_idc; + h->avctx->refs = h->sps.ref_frame_count; - if (h->mb_width != h->sps.mb_width || - h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)) - needs_reinit = 1; + if (h->mb_width != h->sps.mb_width || + h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)) + needs_reinit = 1; - h->mb_width = h->sps.mb_width; - h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag); - h->mb_num = h->mb_width * h->mb_height; - h->mb_stride = h->mb_width + 1; + h->mb_width = h->sps.mb_width; + h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag); + h->mb_num = h->mb_width * h->mb_height; + h->mb_stride = h->mb_width + 1; - h->b_stride = h->mb_width * 4; + h->b_stride = h->mb_width * 4; - h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p + h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p - h->width = 16 * h->mb_width; - h->height = 16 * h->mb_height; + h->width = 16 * h->mb_width; + h->height = 16 * h->mb_height; - ret = init_dimensions(h); - if (ret < 0) - return ret; + ret = init_dimensions(h); + if (ret < 0) + return ret; - if (h->sps.video_signal_type_present_flag) { - h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG - : AVCOL_RANGE_MPEG; - if (h->sps.colour_description_present_flag) { - if (h->avctx->colorspace != h->sps.colorspace) - needs_reinit = 1; - h->avctx->color_primaries = h->sps.color_primaries; - h->avctx->color_trc = h->sps.color_trc; - h->avctx->colorspace = h->sps.colorspace; + if (h->sps.video_signal_type_present_flag) { + h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG + : AVCOL_RANGE_MPEG; + if (h->sps.colour_description_present_flag) { + if (h->avctx->colorspace != h->sps.colorspace) + needs_reinit = 1; + h->avctx->color_primaries = h->sps.color_primaries; + h->avctx->color_trc = h->sps.color_trc; + h->avctx->colorspace = h->sps.colorspace; + } } } if (h->context_initialized && needs_reinit) { - if (h != h0) { + h->context_initialized = 0; + 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; } @@ -1323,14 +1199,14 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, " "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt); - if ((ret = h264_slice_header_init(h, 1)) < 0) { + if ((ret = h264_slice_header_init(h)) < 0) { av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n"); return ret; } } 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; @@ -1340,52 +1216,58 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) return ret; h->avctx->pix_fmt = ret; - if ((ret = h264_slice_header_init(h, 0)) < 0) { + if ((ret = h264_slice_header_init(h)) < 0) { av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n"); return ret; } } - if (h == h0 && h->dequant_coeff_pps != pps_id) { + if (sl == h->slice_ctx && h->dequant_coeff_pps != pps_id) { h->dequant_coeff_pps = pps_id; h264_init_dequant_tables(h); } - h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num); + frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num); + if (!h->setup_finished) + h->frame_num = frame_num; + + sl->mb_mbaff = 0; - h->mb_mbaff = 0; - h->mb_aff_frame = 0; - last_pic_structure = h0->picture_structure; - last_pic_droppable = h0->droppable; - h->droppable = h->nal_ref_idc == 0; + last_pic_structure = h->picture_structure; + last_pic_droppable = h->droppable; + + droppable = h->nal_ref_idc == 0; if (h->sps.frame_mbs_only_flag) { - h->picture_structure = PICT_FRAME; + picture_structure = PICT_FRAME; } else { - field_pic_flag = get_bits1(&h->gb); + field_pic_flag = get_bits1(&sl->gb); if (field_pic_flag) { - bottom_field_flag = get_bits1(&h->gb); - h->picture_structure = PICT_TOP_FIELD + bottom_field_flag; + bottom_field_flag = get_bits1(&sl->gb); + picture_structure = PICT_TOP_FIELD + bottom_field_flag; } else { - h->picture_structure = PICT_FRAME; - h->mb_aff_frame = h->sps.mb_aff; + picture_structure = PICT_FRAME; + mb_aff_frame = h->sps.mb_aff; } } - h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME; + if (!h->setup_finished) { + h->droppable = droppable; + h->picture_structure = picture_structure; + h->mb_aff_frame = mb_aff_frame; + } + sl->mb_field_decoding_flag = h->picture_structure != PICT_FRAME; - if (h0->current_slice != 0) { - if (last_pic_structure != h->picture_structure || - last_pic_droppable != h->droppable) { + if (h->current_slice != 0) { + 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); - h->picture_structure = last_pic_structure; - h->droppable = last_pic_droppable; 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; } } else { @@ -1412,27 +1294,27 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) * decode frames as "finished". * We have to do that before the "dummy" in-between frame allocation, * since that can modify s->current_picture_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); /* 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. */ if (!last_pic_droppable && 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_droppable && 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 { @@ -1465,7 +1347,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) h->frame_num, h->prev_frame_num); ret = h264_frame_start(h); if (ret < 0) { - h0->first_field = 0; + h->first_field = 0; return ret; } @@ -1488,12 +1370,15 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) * vectors. Given we are concealing a lost frame, this probably * is not noticeable by comparison, but it should be fixed. */ if (h->short_ref_count) { - if (prev) { - av_image_copy(h->short_ref[0]->f.data, - h->short_ref[0]->f.linesize, - (const uint8_t **)prev->f.data, - prev->f.linesize, - h->avctx->pix_fmt, + if (prev && + h->short_ref[0]->f->width == prev->f->width && + h->short_ref[0]->f->height == prev->f->height && + h->short_ref[0]->f->format == prev->f->format) { + av_image_copy(h->short_ref[0]->f->data, + h->short_ref[0]->f->linesize, + (const uint8_t **)prev->f->data, + prev->f->linesize, + prev->f->format, h->mb_width * 16, h->mb_height * 16); h->short_ref[0]->poc = prev->poc + 2; @@ -1505,47 +1390,46 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) /* 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->cur_pic_ptr = NULL; - h0->first_field = FIELD_PICTURE(h); + h->cur_pic_ptr = NULL; + h->first_field = FIELD_PICTURE(h); } else { - if (h0->cur_pic_ptr->frame_num != h->frame_num) { + if (h->cur_pic_ptr->frame_num != h->frame_num) { /* 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 { release_unused_pictures(h, 0); } } - if (h != h0 && (ret = clone_slice(h, h0)) < 0) - return ret; - h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup + if (!h->setup_finished) + h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup assert(h->mb_num == h->mb_width * h->mb_height); if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num || @@ -1553,12 +1437,12 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n"); return AVERROR_INVALIDDATA; } - h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width; - h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) << - FIELD_OR_MBAFF_PICTURE(h); + sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width; + sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) << + FIELD_OR_MBAFF_PICTURE(h); if (h->picture_structure == PICT_BOTTOM_FIELD) - h->resync_mb_y = h->mb_y = h->mb_y + 1; - assert(h->mb_y < h->mb_height); + sl->resync_mb_y = sl->mb_y = sl->mb_y + 1; + assert(sl->mb_y < h->mb_height); if (h->picture_structure == PICT_FRAME) { h->curr_pic_num = h->frame_num; @@ -1569,56 +1453,70 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) } if (h->nal_unit_type == NAL_IDR_SLICE) - get_ue_golomb(&h->gb); /* idr_pic_id */ + get_ue_golomb(&sl->gb); /* idr_pic_id */ if (h->sps.poc_type == 0) { - h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb); + int poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb); - if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) - h->delta_poc_bottom = get_se_golomb(&h->gb); + if (!h->setup_finished) + h->poc_lsb = poc_lsb; + + if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) { + int delta_poc_bottom = get_se_golomb(&sl->gb); + if (!h->setup_finished) + h->delta_poc_bottom = delta_poc_bottom; + } } if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) { - h->delta_poc[0] = get_se_golomb(&h->gb); + int delta_poc = get_se_golomb(&sl->gb); - if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) - h->delta_poc[1] = get_se_golomb(&h->gb); + if (!h->setup_finished) + h->delta_poc[0] = delta_poc; + + if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) { + delta_poc = get_se_golomb(&sl->gb); + + if (!h->setup_finished) + h->delta_poc[1] = delta_poc; + } } - ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc); + if (!h->setup_finished) + ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc); if (h->pps.redundant_pic_cnt_present) - h->redundant_pic_count = get_ue_golomb(&h->gb); + sl->redundant_pic_count = get_ue_golomb(&sl->gb); - ret = ff_set_ref_count(h); + ret = ff_set_ref_count(h, sl); if (ret < 0) return ret; else if (ret == 1) default_ref_list_done = 0; if (!default_ref_list_done) - ff_h264_fill_default_ref_list(h); + ff_h264_fill_default_ref_list(h, sl); - if (h->slice_type_nos != AV_PICTURE_TYPE_I) { - ret = ff_h264_decode_ref_pic_list_reordering(h); + if (sl->slice_type_nos != AV_PICTURE_TYPE_I) { + ret = ff_h264_decode_ref_pic_list_reordering(h, sl); if (ret < 0) { - h->ref_count[1] = h->ref_count[0] = 0; + sl->ref_count[1] = sl->ref_count[0] = 0; return ret; } } - if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) || + if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) || (h->pps.weighted_bipred_idc == 1 && - h->slice_type_nos == AV_PICTURE_TYPE_B)) - ff_pred_weight_table(h); + sl->slice_type_nos == AV_PICTURE_TYPE_B)) + ff_pred_weight_table(h, sl); else if (h->pps.weighted_bipred_idc == 2 && - h->slice_type_nos == AV_PICTURE_TYPE_B) { - implicit_weight_table(h, -1); + sl->slice_type_nos == AV_PICTURE_TYPE_B) { + implicit_weight_table(h, sl, -1); } else { - h->use_weight = 0; + sl->use_weight = 0; for (i = 0; i < 2; i++) { - h->luma_weight_flag[i] = 0; - h->chroma_weight_flag[i] = 0; + sl->luma_weight_flag[i] = 0; + sl->chroma_weight_flag[i] = 0; } } @@ -1628,75 +1526,75 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) // 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, &h->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; } if (FRAME_MBAFF(h)) { - ff_h264_fill_mbaff_ref_list(h); + ff_h264_fill_mbaff_ref_list(h, sl); - if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) { - implicit_weight_table(h, 0); - implicit_weight_table(h, 1); + if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) { + implicit_weight_table(h, sl, 0); + implicit_weight_table(h, sl, 1); } } - if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred) - ff_h264_direct_dist_scale_factor(h); - ff_h264_direct_ref_list_init(h); + if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred) + ff_h264_direct_dist_scale_factor(h, sl); + ff_h264_direct_ref_list_init(h, sl); - if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) { - tmp = get_ue_golomb_31(&h->gb); + if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) { + tmp = get_ue_golomb_31(&sl->gb); if (tmp > 2) { av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp); return AVERROR_INVALIDDATA; } - h->cabac_init_idc = tmp; + sl->cabac_init_idc = tmp; } - h->last_qscale_diff = 0; - tmp = h->pps.init_qp + get_se_golomb(&h->gb); + sl->last_qscale_diff = 0; + tmp = h->pps.init_qp + get_se_golomb(&sl->gb); if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) { av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp); return AVERROR_INVALIDDATA; } - h->qscale = tmp; - h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale); - h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale); + sl->qscale = tmp; + sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale); + sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale); // FIXME qscale / qp ... stuff - if (h->slice_type == AV_PICTURE_TYPE_SP) - get_bits1(&h->gb); /* sp_for_switch_flag */ - if (h->slice_type == AV_PICTURE_TYPE_SP || - h->slice_type == AV_PICTURE_TYPE_SI) - get_se_golomb(&h->gb); /* slice_qs_delta */ - - h->deblocking_filter = 1; - h->slice_alpha_c0_offset = 0; - h->slice_beta_offset = 0; + if (sl->slice_type == AV_PICTURE_TYPE_SP) + get_bits1(&sl->gb); /* sp_for_switch_flag */ + if (sl->slice_type == AV_PICTURE_TYPE_SP || + sl->slice_type == AV_PICTURE_TYPE_SI) + get_se_golomb(&sl->gb); /* slice_qs_delta */ + + sl->deblocking_filter = 1; + sl->slice_alpha_c0_offset = 0; + sl->slice_beta_offset = 0; if (h->pps.deblocking_filter_parameters_present) { - tmp = get_ue_golomb_31(&h->gb); + tmp = get_ue_golomb_31(&sl->gb); if (tmp > 2) { av_log(h->avctx, AV_LOG_ERROR, "deblocking_filter_idc %u out of range\n", tmp); return AVERROR_INVALIDDATA; } - h->deblocking_filter = tmp; - if (h->deblocking_filter < 2) - h->deblocking_filter ^= 1; // 1<->0 - - if (h->deblocking_filter) { - h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2; - h->slice_beta_offset = get_se_golomb(&h->gb) * 2; - if (h->slice_alpha_c0_offset > 12 || - h->slice_alpha_c0_offset < -12 || - h->slice_beta_offset > 12 || - h->slice_beta_offset < -12) { + sl->deblocking_filter = tmp; + if (sl->deblocking_filter < 2) + sl->deblocking_filter ^= 1; // 1<->0 + + if (sl->deblocking_filter) { + sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2; + sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2; + if (sl->slice_alpha_c0_offset > 12 || + sl->slice_alpha_c0_offset < -12 || + sl->slice_beta_offset > 12 || + sl->slice_beta_offset < -12) { av_log(h->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", - h->slice_alpha_c0_offset, h->slice_beta_offset); + sl->slice_alpha_c0_offset, sl->slice_beta_offset); return AVERROR_INVALIDDATA; } } @@ -1704,62 +1602,62 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) if (h->avctx->skip_loop_filter >= AVDISCARD_ALL || (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY && - h->slice_type_nos != AV_PICTURE_TYPE_I) || + sl->slice_type_nos != AV_PICTURE_TYPE_I) || (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR && - h->slice_type_nos == AV_PICTURE_TYPE_B) || + sl->slice_type_nos == AV_PICTURE_TYPE_B) || (h->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0)) - h->deblocking_filter = 0; + sl->deblocking_filter = 0; - if (h->deblocking_filter == 1 && h0->max_contexts > 1) { - if (h->avctx->flags2 & CODEC_FLAG2_FAST) { + if (sl->deblocking_filter == 1 && h->max_contexts > 1) { + if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) { /* Cheat slightly for speed: * Do not bother to deblock across slices. */ - h->deblocking_filter = 2; + 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 deblocking type 1, decoding such frames in sequential order\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 1; } } } - h->qp_thresh = 15 - - FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - + sl->qp_thresh = 15 - + FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) - FFMAX3(0, h->pps.chroma_qp_index_offset[0], h->pps.chroma_qp_index_offset[1]) + 6 * (h->sps.bit_depth_luma - 8); - h0->last_slice_type = slice_type; - h->slice_num = ++h0->current_slice; - if (h->slice_num >= MAX_SLICES) { + h->last_slice_type = slice_type; + sl->slice_num = ++h->current_slice; + if (sl->slice_num >= MAX_SLICES) { av_log(h->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n"); } for (j = 0; j < 2; j++) { int id_list[16]; - int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j]; + int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j]; for (i = 0; i < 16; i++) { id_list[i] = 60; - if (j < h->list_count && i < h->ref_count[j] && - h->ref_list[j][i].f.buf[0]) { + if (j < sl->list_count && i < sl->ref_count[j] && + sl->ref_list[j][i].parent->f->buf[0]) { int k; - AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer; + AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer; for (k = 0; k < h->short_ref_count; k++) - if (h->short_ref[k]->f.buf[0]->buffer == buf) { + if (h->short_ref[k]->f->buf[0]->buffer == buf) { id_list[i] = k; break; } for (k = 0; k < h->long_ref_count; k++) - if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) { + if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) { id_list[i] = h->short_ref_count + k; break; } @@ -1769,41 +1667,41 @@ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0) ref2frm[0] = ref2frm[1] = -1; for (i = 0; i < 16; i++) - ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3); + ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3); ref2frm[18 + 0] = ref2frm[18 + 1] = -1; for (i = 16; i < 48; i++) ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] + - (h->ref_list[j][i].reference & 3); + (sl->ref_list[j][i].reference & 3); } if (h->avctx->debug & FF_DEBUG_PICT_INFO) { av_log(h->avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n", - h->slice_num, + sl->slice_num, (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"), first_mb_in_slice, - av_get_picture_type_char(h->slice_type), - h->slice_type_fixed ? " fix" : "", + av_get_picture_type_char(sl->slice_type), + sl->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "", pps_id, h->frame_num, h->cur_pic_ptr->field_poc[0], h->cur_pic_ptr->field_poc[1], - h->ref_count[0], h->ref_count[1], - h->qscale, - h->deblocking_filter, - h->slice_alpha_c0_offset, h->slice_beta_offset, - h->use_weight, - h->use_weight == 1 && h->use_weight_chroma ? "c" : "", - h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""); + sl->ref_count[0], sl->ref_count[1], + sl->qscale, + sl->deblocking_filter, + sl->slice_alpha_c0_offset, sl->slice_beta_offset, + sl->use_weight, + sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "", + sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""); } return 0; } -int ff_h264_get_slice_type(const H264Context *h) +int ff_h264_get_slice_type(const H264SliceContext *sl) { - switch (h->slice_type) { + switch (sl->slice_type) { case AV_PICTURE_TYPE_P: return 0; case AV_PICTURE_TYPE_B: @@ -1819,7 +1717,8 @@ int ff_h264_get_slice_type(const H264Context *h) } } -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], int top_type, @@ -1827,13 +1726,13 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h, int mb_xy, int list) { int b_stride = h->b_stride; - int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]]; - int8_t *ref_cache = &h->ref_cache[list][scan8[0]]; + int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]]; + int8_t *ref_cache = &sl->ref_cache[list][scan8[0]]; if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) { 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] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2); + int (*ref2frm)[64] = 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]]; @@ -1848,7 +1747,7 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h, 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] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2); + int (*ref2frm)[64] = 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]); @@ -1881,7 +1780,7 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h, { int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy]; - int (*ref2frm)[64] = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2); + int (*ref2frm)[64] = 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); @@ -1891,7 +1790,7 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h, } { - int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride]; + int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride]; AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride); AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride); AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride); @@ -1903,15 +1802,15 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h, * * @return non zero if the loop filter can be skipped */ -static int fill_filter_caches(H264Context *h, int mb_type) +static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type) { - const int mb_xy = h->mb_xy; + const int mb_xy = sl->mb_xy; int top_xy, left_xy[LEFT_MBS]; int top_type, left_type[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. */ @@ -1920,7 +1819,7 @@ static int fill_filter_caches(H264Context *h, int mb_type) if (FRAME_MBAFF(h)) { const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]); const int curr_mb_field_flag = IS_INTERLACED(mb_type); - if (h->mb_y & 1) { + if (sl->mb_y & 1) { if (left_mb_field_flag != curr_mb_field_flag) left_xy[LTOP] -= h->mb_stride; } else { @@ -1932,14 +1831,14 @@ static int fill_filter_caches(H264Context *h, int mb_type) } } - h->top_mb_xy = top_xy; - h->left_mb_xy[LTOP] = left_xy[LTOP]; - h->left_mb_xy[LBOT] = left_xy[LBOT]; + sl->top_mb_xy = top_xy; + sl->left_mb_xy[LTOP] = left_xy[LTOP]; + sl->left_mb_xy[LBOT] = left_xy[LBOT]; { /* For sufficiently low qp, filtering wouldn't do anything. * This is a conservative estimate: could also check beta_offset * and more accurate chroma_qp. */ - int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice + int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice int qp = h->cur_pic.qscale_table[mb_xy]; if (qp <= qp_thresh && (left_xy[LTOP] < 0 || @@ -1959,10 +1858,10 @@ static int fill_filter_caches(H264Context *h, int mb_type) top_type = h->cur_pic.mb_type[top_xy]; left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]]; left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]]; - if (h->deblocking_filter == 2) { - if (h->slice_table[top_xy] != h->slice_num) + if (sl->deblocking_filter == 2) { + if (h->slice_table[top_xy] != sl->slice_num) top_type = 0; - if (h->slice_table[left_xy[LBOT]] != h->slice_num) + if (h->slice_table[left_xy[LBOT]] != sl->slice_num) left_type[LTOP] = left_type[LBOT] = 0; } else { if (h->slice_table[top_xy] == 0xFFFF) @@ -1970,26 +1869,26 @@ static int fill_filter_caches(H264Context *h, int mb_type) if (h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP] = left_type[LBOT] = 0; } - h->top_type = top_type; - h->left_type[LTOP] = left_type[LTOP]; - h->left_type[LBOT] = left_type[LBOT]; + sl->top_type = top_type; + sl->left_type[LTOP] = left_type[LTOP]; + sl->left_type[LBOT] = left_type[LBOT]; if (IS_INTRA(mb_type)) return 0; - fill_filter_caches_inter(h, mb_type, top_xy, left_xy, + fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0); - if (h->list_count == 2) - fill_filter_caches_inter(h, mb_type, top_xy, left_xy, + if (sl->list_count == 2) + fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1); nnz = h->non_zero_count[mb_xy]; - nnz_cache = h->non_zero_count_cache; + nnz_cache = sl->non_zero_count_cache; AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]); AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]); AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]); AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]); - h->cbp = h->cbp_table[mb_xy]; + sl->cbp = h->cbp_table[mb_xy]; if (top_type) { nnz = h->non_zero_count[top_xy]; @@ -2026,118 +1925,118 @@ static int fill_filter_caches(H264Context *h, int mb_type) nnz_cache[scan8[0]] = nnz_cache[scan8[1]] = nnz_cache[scan8[2]] = - nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12; + nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12; nnz_cache[scan8[0 + 4]] = nnz_cache[scan8[1 + 4]] = nnz_cache[scan8[2 + 4]] = - nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12; + nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12; nnz_cache[scan8[0 + 8]] = nnz_cache[scan8[1 + 8]] = nnz_cache[scan8[2 + 8]] = - nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12; + nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12; nnz_cache[scan8[0 + 12]] = nnz_cache[scan8[1 + 12]] = nnz_cache[scan8[2 + 12]] = - nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12; + nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12; } } return 0; } -static void loop_filter(H264Context *h, 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; - const int end_mb_y = h->mb_y + FRAME_MBAFF(h); - const int old_slice_type = h->slice_type; + const int end_mb_y = sl->mb_y + FRAME_MBAFF(h); + const int old_slice_type = sl->slice_type; const int pixel_shift = h->pixel_shift; const int block_h = 16 >> h->chroma_y_shift; - if (h->deblocking_filter) { + if (sl->deblocking_filter) { for (mb_x = start_x; mb_x < end_x; mb_x++) for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) { int mb_xy, mb_type; - mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride; - h->slice_num = h->slice_table[mb_xy]; + mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride; + sl->slice_num = h->slice_table[mb_xy]; mb_type = h->cur_pic.mb_type[mb_xy]; - h->list_count = h->list_counts[mb_xy]; + sl->list_count = h->list_counts[mb_xy]; if (FRAME_MBAFF(h)) - h->mb_mbaff = - h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type); - - h->mb_x = mb_x; - h->mb_y = mb_y; - dest_y = h->cur_pic.f.data[0] + - ((mb_x << pixel_shift) + mb_y * h->linesize) * 16; - dest_cb = h->cur_pic.f.data[1] + + sl->mb_mbaff = + sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type); + + sl->mb_x = mb_x; + sl->mb_y = mb_y; + dest_y = h->cur_pic.f->data[0] + + ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16; + dest_cb = h->cur_pic.f->data[1] + (mb_x << pixel_shift) * (8 << CHROMA444(h)) + - mb_y * h->uvlinesize * block_h; - dest_cr = h->cur_pic.f.data[2] + + mb_y * sl->uvlinesize * block_h; + dest_cr = h->cur_pic.f->data[2] + (mb_x << pixel_shift) * (8 << CHROMA444(h)) + - mb_y * h->uvlinesize * block_h; + mb_y * sl->uvlinesize * block_h; // FIXME simplify above - if (MB_FIELD(h)) { - linesize = h->mb_linesize = h->linesize * 2; - uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2; + if (MB_FIELD(sl)) { + linesize = sl->mb_linesize = sl->linesize * 2; + uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2; if (mb_y & 1) { // FIXME move out of this function? - dest_y -= h->linesize * 15; - dest_cb -= h->uvlinesize * (block_h - 1); - dest_cr -= h->uvlinesize * (block_h - 1); + dest_y -= sl->linesize * 15; + dest_cb -= sl->uvlinesize * (block_h - 1); + dest_cr -= sl->uvlinesize * (block_h - 1); } } else { - linesize = h->mb_linesize = h->linesize; - uvlinesize = h->mb_uvlinesize = h->uvlinesize; + linesize = sl->mb_linesize = sl->linesize; + uvlinesize = sl->mb_uvlinesize = sl->uvlinesize; } - backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, + backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0); - if (fill_filter_caches(h, mb_type)) + if (fill_filter_caches(h, sl, mb_type)) continue; - h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]); - h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]); + sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]); + sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]); if (FRAME_MBAFF(h)) { - ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, + ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize); } else { - ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, + ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize); } } } - h->slice_type = old_slice_type; - h->mb_x = end_x; - h->mb_y = end_mb_y - FRAME_MBAFF(h); - h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale); - h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale); + sl->slice_type = old_slice_type; + sl->mb_x = end_x; + sl->mb_y = end_mb_y - FRAME_MBAFF(h); + sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale); + sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale); } -static void predict_field_decoding_flag(H264Context *h) +static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl) { - const int mb_xy = h->mb_x + h->mb_y * h->mb_stride; - int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ? + 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] == h->slice_num) ? + (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) +static void decode_finish_row(const H264Context *h, H264SliceContext *sl) { - int top = 16 * (h->mb_y >> FIELD_PICTURE(h)); + int top = 16 * (sl->mb_y >> FIELD_PICTURE(h)); int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h); int height = 16 << FRAME_MBAFF(h); int deblock_border = (16 + 4) << FRAME_MBAFF(h); - if (h->deblocking_filter) { + if (sl->deblocking_filter) { if ((top + height) >= pic_height) height += deblock_border; top -= deblock_border; @@ -2152,7 +2051,7 @@ static void decode_finish_row(H264Context *h) top = 0; } - ff_h264_draw_horiz_band(h, top, height); + ff_h264_draw_horiz_band(h, sl, top, height); if (h->droppable) return; @@ -2161,170 +2060,195 @@ static void decode_finish_row(H264Context *h) h->picture_structure == PICT_BOTTOM_FIELD); } -static void er_add_slice(H264Context *h, int startx, int starty, +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; + + if (!sl->h264->enable_er) + return; - er->ref_count = h->ref_count[0]; + er->ref_count = sl->ref_count[0]; ff_er_add_slice(er, startx, starty, endx, endy, status); #endif } static int decode_slice(struct AVCodecContext *avctx, void *arg) { - H264Context *h = *(void **)arg; - int lf_x_start = h->mb_x; + H264SliceContext *sl = arg; + const H264Context *h = sl->h264; + int lf_x_start = sl->mb_x; + int ret; + + sl->linesize = h->cur_pic_ptr->f->linesize[0]; + sl->uvlinesize = h->cur_pic_ptr->f->linesize[1]; + + ret = alloc_scratch_buffers(sl, sl->linesize); + if (ret < 0) + return ret; - h->mb_skip_run = -1; + sl->mb_skip_run = -1; - h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME || - avctx->codec_id != AV_CODEC_ID_H264 || - (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY)); + sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME || + avctx->codec_id != AV_CODEC_ID_H264 || + (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY)); if (h->pps.cabac) { /* realign */ - align_get_bits(&h->gb); + align_get_bits(&sl->gb); /* init cabac */ - ff_init_cabac_decoder(&h->cabac, - h->gb.buffer + get_bits_count(&h->gb) / 8, - (get_bits_left(&h->gb) + 7) / 8); + ff_init_cabac_decoder(&sl->cabac, + sl->gb.buffer + get_bits_count(&sl->gb) / 8, + (get_bits_left(&sl->gb) + 7) / 8); - ff_h264_init_cabac_states(h); + ff_h264_init_cabac_states(h, sl); for (;;) { // START_TIMER - int ret = ff_h264_decode_mb_cabac(h); - int eos; + int ret, eos; + + if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) { + av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n", + sl->next_slice_idx); + return AVERROR_INVALIDDATA; + } + + ret = ff_h264_decode_mb_cabac(h, sl); // STOP_TIMER("decode_mb_cabac") if (ret >= 0) - ff_h264_hl_decode_mb(h); + ff_h264_hl_decode_mb(h, sl); // FIXME optimal? or let mb_decode decode 16x32 ? if (ret >= 0 && FRAME_MBAFF(h)) { - h->mb_y++; + sl->mb_y++; - ret = ff_h264_decode_mb_cabac(h); + ret = ff_h264_decode_mb_cabac(h, sl); if (ret >= 0) - ff_h264_hl_decode_mb(h); - h->mb_y--; + ff_h264_hl_decode_mb(h, sl); + sl->mb_y--; } - eos = get_cabac_terminate(&h->cabac); + eos = get_cabac_terminate(&sl->cabac); if ((h->workaround_bugs & FF_BUG_TRUNCATED) && - h->cabac.bytestream > h->cabac.bytestream_end + 2) { - er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1, - h->mb_y, ER_MB_END); - if (h->mb_x >= lf_x_start) - loop_filter(h, lf_x_start, h->mb_x + 1); + sl->cabac.bytestream > sl->cabac.bytestream_end + 2) { + 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); return 0; } - if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) { + if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) { av_log(h->avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream %td\n", - h->mb_x, h->mb_y, - h->cabac.bytestream_end - h->cabac.bytestream); - er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x, - h->mb_y, ER_MB_ERROR); + sl->mb_x, sl->mb_y, + sl->cabac.bytestream_end - sl->cabac.bytestream); + 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 (++h->mb_x >= h->mb_width) { - loop_filter(h, lf_x_start, h->mb_x); - h->mb_x = lf_x_start = 0; - decode_finish_row(h); - ++h->mb_y; + if (++sl->mb_x >= h->mb_width) { + loop_filter(h, sl, lf_x_start, sl->mb_x); + sl->mb_x = lf_x_start = 0; + decode_finish_row(h, sl); + ++sl->mb_y; if (FIELD_OR_MBAFF_PICTURE(h)) { - ++h->mb_y; - if (FRAME_MBAFF(h) && h->mb_y < h->mb_height) - predict_field_decoding_flag(h); + ++sl->mb_y; + if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height) + predict_field_decoding_flag(h, sl); } } - if (eos || h->mb_y >= h->mb_height) { - tprintf(h->avctx, "slice end %d %d\n", - get_bits_count(&h->gb), h->gb.size_in_bits); - er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1, - h->mb_y, ER_MB_END); - if (h->mb_x > lf_x_start) - loop_filter(h, lf_x_start, h->mb_x); + if (eos || sl->mb_y >= h->mb_height) { + ff_tlog(h->avctx, "slice end %d %d\n", + get_bits_count(&sl->gb), sl->gb.size_in_bits); + 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 { for (;;) { - int ret = ff_h264_decode_mb_cavlc(h); + int ret; + + if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) { + av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n", + sl->next_slice_idx); + return AVERROR_INVALIDDATA; + } + + ret = ff_h264_decode_mb_cavlc(h, sl); if (ret >= 0) - ff_h264_hl_decode_mb(h); + ff_h264_hl_decode_mb(h, sl); // FIXME optimal? or let mb_decode decode 16x32 ? if (ret >= 0 && FRAME_MBAFF(h)) { - h->mb_y++; - ret = ff_h264_decode_mb_cavlc(h); + sl->mb_y++; + ret = ff_h264_decode_mb_cavlc(h, sl); if (ret >= 0) - ff_h264_hl_decode_mb(h); - h->mb_y--; + ff_h264_hl_decode_mb(h, sl); + sl->mb_y--; } if (ret < 0) { av_log(h->avctx, AV_LOG_ERROR, - "error while decoding MB %d %d\n", h->mb_x, h->mb_y); - er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x, - h->mb_y, ER_MB_ERROR); + "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y); + er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x, + sl->mb_y, ER_MB_ERROR); return ret; } - if (++h->mb_x >= h->mb_width) { - loop_filter(h, lf_x_start, h->mb_x); - h->mb_x = lf_x_start = 0; - decode_finish_row(h); - ++h->mb_y; + if (++sl->mb_x >= h->mb_width) { + loop_filter(h, sl, lf_x_start, sl->mb_x); + sl->mb_x = lf_x_start = 0; + decode_finish_row(h, sl); + ++sl->mb_y; if (FIELD_OR_MBAFF_PICTURE(h)) { - ++h->mb_y; - if (FRAME_MBAFF(h) && h->mb_y < h->mb_height) - predict_field_decoding_flag(h); + ++sl->mb_y; + if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height) + predict_field_decoding_flag(h, sl); } - if (h->mb_y >= h->mb_height) { - tprintf(h->avctx, "slice end %d %d\n", - get_bits_count(&h->gb), h->gb.size_in_bits); + if (sl->mb_y >= h->mb_height) { + ff_tlog(h->avctx, "slice end %d %d\n", + get_bits_count(&sl->gb), sl->gb.size_in_bits); - if (get_bits_left(&h->gb) == 0) { - er_add_slice(h, h->resync_mb_x, h->resync_mb_y, - h->mb_x - 1, h->mb_y, - ER_MB_END); + if (get_bits_left(&sl->gb) == 0) { + 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, h->resync_mb_x, h->resync_mb_y, - h->mb_x - 1, h->mb_y, - ER_MB_END); + er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, + sl->mb_x - 1, sl->mb_y, ER_MB_END); return AVERROR_INVALIDDATA; } } } - if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) { - tprintf(h->avctx, "slice end %d %d\n", - get_bits_count(&h->gb), h->gb.size_in_bits); + if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) { + ff_tlog(h->avctx, "slice end %d %d\n", + get_bits_count(&sl->gb), sl->gb.size_in_bits); - if (get_bits_left(&h->gb) == 0) { - er_add_slice(h, h->resync_mb_x, h->resync_mb_y, - h->mb_x - 1, h->mb_y, - ER_MB_END); - if (h->mb_x > lf_x_start) - loop_filter(h, lf_x_start, h->mb_x); + if (get_bits_left(&sl->gb) == 0) { + 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, h->resync_mb_x, h->resync_mb_y, h->mb_x, - h->mb_y, ER_MB_ERROR); + er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x, + sl->mb_y, ER_MB_ERROR); return AVERROR_INVALIDDATA; } @@ -2342,36 +2266,48 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg) int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count) { AVCodecContext *const avctx = h->avctx; - H264Context *hx; - int i; - - if (h->mb_y >= h->mb_height) { - av_log(h->avctx, AV_LOG_ERROR, - "Input contains more MB rows than the frame height.\n"); - return AVERROR_INVALIDDATA; - } + H264SliceContext *sl; + int i, j; if (h->avctx->hwaccel) return 0; if (context_count == 1) { - return decode_slice(avctx, &h); + int ret; + + h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height; + + ret = decode_slice(avctx, &h->slice_ctx[0]); + h->mb_y = h->slice_ctx[0].mb_y; + return ret; } else { - for (i = 1; i < context_count; i++) { - hx = h->thread_context[i]; - hx->er.error_count = 0; + for (i = 0; i < context_count; i++) { + int next_slice_idx = h->mb_width * h->mb_height; + int slice_idx; + + sl = &h->slice_ctx[i]; + sl->er.error_count = 0; + + /* make sure none of those slices overlap */ + slice_idx = sl->mb_y * h->mb_width + sl->mb_x; + for (j = 0; j < context_count; j++) { + H264SliceContext *sl2 = &h->slice_ctx[j]; + int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x; + + if (i == j || slice_idx2 < slice_idx) + continue; + next_slice_idx = FFMIN(next_slice_idx, slice_idx2); + } + sl->next_slice_idx = next_slice_idx; } - avctx->execute(avctx, decode_slice, h->thread_context, - NULL, context_count, sizeof(void *)); + 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]; - h->mb_x = hx->mb_x; - h->mb_y = hx->mb_y; - h->droppable = hx->droppable; - h->picture_structure = hx->picture_structure; + sl = &h->slice_ctx[context_count - 1]; + h->mb_y = sl->mb_y; 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; } return 0;