yading@10: /* yading@10: * H.26L/H.264/AVC/JVT/14496-10/... decoder yading@10: * Copyright (c) 2003 Michael Niedermayer yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: /** yading@10: * @file yading@10: * H.264 / AVC / MPEG4 part10 codec. yading@10: * @author Michael Niedermayer yading@10: */ yading@10: yading@10: #define UNCHECKED_BITSTREAM_READER 1 yading@10: yading@10: #include "libavutil/imgutils.h" yading@10: #include "libavutil/opt.h" yading@10: #include "internal.h" yading@10: #include "cabac.h" yading@10: #include "cabac_functions.h" yading@10: #include "dsputil.h" yading@10: #include "error_resilience.h" yading@10: #include "avcodec.h" yading@10: #include "mpegvideo.h" yading@10: #include "h264.h" yading@10: #include "h264data.h" yading@10: #include "h264chroma.h" yading@10: #include "h264_mvpred.h" yading@10: #include "golomb.h" yading@10: #include "mathops.h" yading@10: #include "rectangle.h" yading@10: #include "svq3.h" yading@10: #include "thread.h" yading@10: #include "vdpau_internal.h" yading@10: #include "libavutil/avassert.h" yading@10: yading@10: // #undef NDEBUG yading@10: #include yading@10: yading@10: const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 }; yading@10: yading@10: static const uint8_t rem6[QP_MAX_NUM + 1] = { yading@10: 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, yading@10: 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, yading@10: 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, yading@10: 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, yading@10: 0, 1, 2, 3, yading@10: }; yading@10: yading@10: static const uint8_t div6[QP_MAX_NUM + 1] = { yading@10: 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, yading@10: 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, yading@10: 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, yading@10: 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13, yading@10: 14,14,14,14, yading@10: }; yading@10: yading@10: static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = { yading@10: #if CONFIG_H264_DXVA2_HWACCEL yading@10: AV_PIX_FMT_DXVA2_VLD, yading@10: #endif yading@10: #if CONFIG_H264_VAAPI_HWACCEL yading@10: AV_PIX_FMT_VAAPI_VLD, yading@10: #endif yading@10: #if CONFIG_H264_VDA_HWACCEL yading@10: AV_PIX_FMT_VDA_VLD, yading@10: #endif yading@10: #if CONFIG_H264_VDPAU_HWACCEL yading@10: AV_PIX_FMT_VDPAU, yading@10: #endif yading@10: AV_PIX_FMT_YUV420P, yading@10: AV_PIX_FMT_NONE yading@10: }; yading@10: yading@10: static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = { yading@10: #if CONFIG_H264_DXVA2_HWACCEL yading@10: AV_PIX_FMT_DXVA2_VLD, yading@10: #endif yading@10: #if CONFIG_H264_VAAPI_HWACCEL yading@10: AV_PIX_FMT_VAAPI_VLD, yading@10: #endif yading@10: #if CONFIG_H264_VDA_HWACCEL yading@10: AV_PIX_FMT_VDA_VLD, yading@10: #endif yading@10: #if CONFIG_H264_VDPAU_HWACCEL yading@10: AV_PIX_FMT_VDPAU, yading@10: #endif yading@10: AV_PIX_FMT_YUVJ420P, yading@10: AV_PIX_FMT_NONE yading@10: }; yading@10: yading@10: int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx) yading@10: { yading@10: H264Context *h = avctx->priv_data; yading@10: return h ? h->sps.num_reorder_frames : 0; yading@10: } yading@10: yading@10: static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type, yading@10: int (*mv)[2][4][2], yading@10: int mb_x, int mb_y, int mb_intra, int mb_skipped) yading@10: { yading@10: H264Context *h = opaque; yading@10: yading@10: h->mb_x = mb_x; yading@10: h->mb_y = mb_y; yading@10: h->mb_xy = mb_x + mb_y * h->mb_stride; yading@10: memset(h->non_zero_count_cache, 0, sizeof(h->non_zero_count_cache)); yading@10: av_assert1(ref >= 0); yading@10: /* FIXME: It is possible albeit uncommon that slice references yading@10: * differ between slices. We take the easy approach and ignore yading@10: * it for now. If this turns out to have any relevance in yading@10: * practice then correct remapping should be added. */ yading@10: if (ref >= h->ref_count[0]) yading@10: ref = 0; yading@10: if (!h->ref_list[0][ref].f.data[0]) { yading@10: av_log(h->avctx, AV_LOG_DEBUG, "Reference not available for error concealing\n"); yading@10: ref = 0; yading@10: } yading@10: if ((h->ref_list[0][ref].reference&3) != 3) { yading@10: av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n"); yading@10: return; yading@10: } yading@10: fill_rectangle(&h->cur_pic.ref_index[0][4 * h->mb_xy], yading@10: 2, 2, 2, ref, 1); yading@10: fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1); yading@10: fill_rectangle(h->mv_cache[0][scan8[0]], 4, 4, 8, yading@10: pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4); yading@10: h->mb_mbaff = yading@10: h->mb_field_decoding_flag = 0; yading@10: ff_h264_hl_decode_mb(h); yading@10: } yading@10: yading@10: void ff_h264_draw_horiz_band(H264Context *h, int y, int height) yading@10: { yading@10: AVCodecContext *avctx = h->avctx; yading@10: Picture *cur = &h->cur_pic; yading@10: Picture *last = h->ref_list[0][0].f.data[0] ? &h->ref_list[0][0] : NULL; yading@10: const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt); yading@10: int vshift = desc->log2_chroma_h; yading@10: const int field_pic = h->picture_structure != PICT_FRAME; yading@10: if (field_pic) { yading@10: height <<= 1; yading@10: y <<= 1; yading@10: } yading@10: yading@10: height = FFMIN(height, avctx->height - y); yading@10: yading@10: if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD)) yading@10: return; yading@10: yading@10: if (avctx->draw_horiz_band) { yading@10: AVFrame *src; yading@10: int offset[AV_NUM_DATA_POINTERS]; yading@10: int i; yading@10: yading@10: if (cur->f.pict_type == AV_PICTURE_TYPE_B || h->low_delay || yading@10: (avctx->slice_flags & SLICE_FLAG_CODED_ORDER)) yading@10: src = &cur->f; yading@10: else if (last) yading@10: src = &last->f; yading@10: else yading@10: return; yading@10: yading@10: offset[0] = y * src->linesize[0]; yading@10: offset[1] = yading@10: offset[2] = (y >> vshift) * src->linesize[1]; yading@10: for (i = 3; i < AV_NUM_DATA_POINTERS; i++) yading@10: offset[i] = 0; yading@10: yading@10: emms_c(); yading@10: yading@10: avctx->draw_horiz_band(avctx, src, offset, yading@10: y, h->picture_structure, height); yading@10: } yading@10: } yading@10: yading@10: static void unref_picture(H264Context *h, Picture *pic) yading@10: { yading@10: int off = offsetof(Picture, tf) + sizeof(pic->tf); yading@10: int i; yading@10: yading@10: if (!pic->f.data[0]) yading@10: return; yading@10: yading@10: ff_thread_release_buffer(h->avctx, &pic->tf); yading@10: av_buffer_unref(&pic->hwaccel_priv_buf); yading@10: yading@10: av_buffer_unref(&pic->qscale_table_buf); yading@10: av_buffer_unref(&pic->mb_type_buf); yading@10: for (i = 0; i < 2; i++) { yading@10: av_buffer_unref(&pic->motion_val_buf[i]); yading@10: av_buffer_unref(&pic->ref_index_buf[i]); yading@10: } yading@10: yading@10: memset((uint8_t*)pic + off, 0, sizeof(*pic) - off); yading@10: } yading@10: yading@10: static void release_unused_pictures(H264Context *h, int remove_current) yading@10: { yading@10: int i; yading@10: yading@10: /* release non reference frames */ yading@10: for (i = 0; i < MAX_PICTURE_COUNT; i++) { yading@10: if (h->DPB[i].f.data[0] && !h->DPB[i].reference && yading@10: (remove_current || &h->DPB[i] != h->cur_pic_ptr)) { yading@10: unref_picture(h, &h->DPB[i]); yading@10: } yading@10: } yading@10: } yading@10: yading@10: static int ref_picture(H264Context *h, Picture *dst, Picture *src) yading@10: { yading@10: int ret, i; yading@10: yading@10: av_assert0(!dst->f.buf[0]); yading@10: av_assert0(src->f.buf[0]); yading@10: yading@10: src->tf.f = &src->f; yading@10: dst->tf.f = &dst->f; yading@10: ret = ff_thread_ref_frame(&dst->tf, &src->tf); yading@10: if (ret < 0) yading@10: goto fail; yading@10: yading@10: yading@10: dst->qscale_table_buf = av_buffer_ref(src->qscale_table_buf); yading@10: dst->mb_type_buf = av_buffer_ref(src->mb_type_buf); yading@10: if (!dst->qscale_table_buf || !dst->mb_type_buf) yading@10: goto fail; yading@10: dst->qscale_table = src->qscale_table; yading@10: dst->mb_type = src->mb_type; yading@10: yading@10: for (i = 0; i < 2; i ++) { yading@10: dst->motion_val_buf[i] = av_buffer_ref(src->motion_val_buf[i]); yading@10: dst->ref_index_buf[i] = av_buffer_ref(src->ref_index_buf[i]); yading@10: if (!dst->motion_val_buf[i] || !dst->ref_index_buf[i]) yading@10: goto fail; yading@10: dst->motion_val[i] = src->motion_val[i]; yading@10: dst->ref_index[i] = src->ref_index[i]; yading@10: } yading@10: yading@10: if (src->hwaccel_picture_private) { yading@10: dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf); yading@10: if (!dst->hwaccel_priv_buf) yading@10: goto fail; yading@10: dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data; yading@10: } yading@10: yading@10: for (i = 0; i < 2; i++) yading@10: dst->field_poc[i] = src->field_poc[i]; yading@10: yading@10: memcpy(dst->ref_poc, src->ref_poc, sizeof(src->ref_poc)); yading@10: memcpy(dst->ref_count, src->ref_count, sizeof(src->ref_count)); yading@10: yading@10: dst->poc = src->poc; yading@10: dst->frame_num = src->frame_num; yading@10: dst->mmco_reset = src->mmco_reset; yading@10: dst->pic_id = src->pic_id; yading@10: dst->long_ref = src->long_ref; yading@10: dst->mbaff = src->mbaff; yading@10: dst->field_picture = src->field_picture; yading@10: dst->needs_realloc = src->needs_realloc; yading@10: dst->reference = src->reference; yading@10: dst->sync = src->sync; yading@10: yading@10: return 0; yading@10: fail: yading@10: unref_picture(h, dst); yading@10: return ret; yading@10: } yading@10: yading@10: yading@10: static int alloc_scratch_buffers(H264Context *h, int linesize) yading@10: { yading@10: int alloc_size = FFALIGN(FFABS(linesize) + 32, 32); yading@10: yading@10: if (h->bipred_scratchpad) yading@10: return 0; yading@10: yading@10: h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size); yading@10: // edge emu needs blocksize + filter length - 1 yading@10: // (= 21x21 for h264) yading@10: h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21); yading@10: h->me.scratchpad = av_mallocz(alloc_size * 2 * 16 * 2); yading@10: yading@10: if (!h->bipred_scratchpad || !h->edge_emu_buffer || !h->me.scratchpad) { yading@10: av_freep(&h->bipred_scratchpad); yading@10: av_freep(&h->edge_emu_buffer); yading@10: av_freep(&h->me.scratchpad); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: h->me.temp = h->me.scratchpad; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int init_table_pools(H264Context *h) yading@10: { yading@10: const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1; yading@10: const int mb_array_size = h->mb_stride * h->mb_height; yading@10: const int b4_stride = h->mb_width * 4 + 1; yading@10: const int b4_array_size = b4_stride * h->mb_height * 4; yading@10: yading@10: h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride, yading@10: av_buffer_allocz); yading@10: h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) * yading@10: sizeof(uint32_t), av_buffer_allocz); yading@10: h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) * yading@10: sizeof(int16_t), av_buffer_allocz); yading@10: h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz); yading@10: yading@10: if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool || yading@10: !h->ref_index_pool) { yading@10: av_buffer_pool_uninit(&h->qscale_table_pool); yading@10: av_buffer_pool_uninit(&h->mb_type_pool); yading@10: av_buffer_pool_uninit(&h->motion_val_pool); yading@10: av_buffer_pool_uninit(&h->ref_index_pool); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int alloc_picture(H264Context *h, Picture *pic) yading@10: { yading@10: int i, ret = 0; yading@10: yading@10: av_assert0(!pic->f.data[0]); yading@10: yading@10: pic->tf.f = &pic->f; yading@10: ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ? yading@10: AV_GET_BUFFER_FLAG_REF : 0); yading@10: if (ret < 0) yading@10: goto fail; yading@10: yading@10: h->linesize = pic->f.linesize[0]; yading@10: h->uvlinesize = pic->f.linesize[1]; yading@10: yading@10: if (h->avctx->hwaccel) { yading@10: const AVHWAccel *hwaccel = h->avctx->hwaccel; yading@10: av_assert0(!pic->hwaccel_picture_private); yading@10: if (hwaccel->priv_data_size) { yading@10: pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size); yading@10: if (!pic->hwaccel_priv_buf) yading@10: return AVERROR(ENOMEM); yading@10: pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data; yading@10: } yading@10: } yading@10: yading@10: if (!h->qscale_table_pool) { yading@10: ret = init_table_pools(h); yading@10: if (ret < 0) yading@10: goto fail; yading@10: } yading@10: yading@10: pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool); yading@10: pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool); yading@10: if (!pic->qscale_table_buf || !pic->mb_type_buf) yading@10: goto fail; yading@10: yading@10: pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1; yading@10: pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1; yading@10: yading@10: for (i = 0; i < 2; i++) { yading@10: pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool); yading@10: pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool); yading@10: if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) yading@10: goto fail; yading@10: yading@10: pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4; yading@10: pic->ref_index[i] = pic->ref_index_buf[i]->data; yading@10: } yading@10: yading@10: return 0; yading@10: fail: yading@10: unref_picture(h, pic); yading@10: return (ret < 0) ? ret : AVERROR(ENOMEM); yading@10: } yading@10: yading@10: static inline int pic_is_unused(H264Context *h, Picture *pic) yading@10: { yading@10: if (pic->f.data[0] == NULL) yading@10: return 1; yading@10: if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF)) yading@10: return 1; yading@10: return 0; yading@10: } yading@10: yading@10: static int find_unused_picture(H264Context *h) yading@10: { yading@10: int i; yading@10: yading@10: for (i = 0; i < MAX_PICTURE_COUNT; i++) { yading@10: if (pic_is_unused(h, &h->DPB[i])) yading@10: break; yading@10: } yading@10: if (i == MAX_PICTURE_COUNT) yading@10: return AVERROR_INVALIDDATA; yading@10: yading@10: if (h->DPB[i].needs_realloc) { yading@10: h->DPB[i].needs_realloc = 0; yading@10: unref_picture(h, &h->DPB[i]); yading@10: } yading@10: yading@10: return i; yading@10: } yading@10: yading@10: /** yading@10: * Check if the top & left blocks are available if needed and yading@10: * change the dc mode so it only uses the available blocks. yading@10: */ yading@10: int ff_h264_check_intra4x4_pred_mode(H264Context *h) yading@10: { yading@10: static const int8_t top[12] = { yading@10: -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0 yading@10: }; yading@10: static const int8_t left[12] = { yading@10: 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED yading@10: }; yading@10: int i; yading@10: yading@10: if (!(h->top_samples_available & 0x8000)) { yading@10: for (i = 0; i < 4; i++) { yading@10: int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]]; yading@10: if (status < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "top block unavailable for requested intra4x4 mode %d at %d %d\n", yading@10: status, h->mb_x, h->mb_y); yading@10: return -1; yading@10: } else if (status) { yading@10: h->intra4x4_pred_mode_cache[scan8[0] + i] = status; yading@10: } yading@10: } yading@10: } yading@10: yading@10: if ((h->left_samples_available & 0x8888) != 0x8888) { yading@10: static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 }; yading@10: for (i = 0; i < 4; i++) yading@10: if (!(h->left_samples_available & mask[i])) { yading@10: int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]]; yading@10: if (status < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "left block unavailable for requested intra4x4 mode %d at %d %d\n", yading@10: status, h->mb_x, h->mb_y); yading@10: return -1; yading@10: } else if (status) { yading@10: h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status; yading@10: } yading@10: } yading@10: } yading@10: yading@10: return 0; yading@10: } // FIXME cleanup like ff_h264_check_intra_pred_mode yading@10: yading@10: /** yading@10: * Check if the top & left blocks are available if needed and yading@10: * change the dc mode so it only uses the available blocks. yading@10: */ yading@10: int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma) yading@10: { yading@10: static const int8_t top[7] = { LEFT_DC_PRED8x8, 1, -1, -1 }; yading@10: static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 }; yading@10: yading@10: if (mode > 6U) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "out of range intra chroma pred mode at %d %d\n", yading@10: h->mb_x, h->mb_y); yading@10: return -1; yading@10: } yading@10: yading@10: if (!(h->top_samples_available & 0x8000)) { yading@10: mode = top[mode]; yading@10: if (mode < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "top block unavailable for requested intra mode at %d %d\n", yading@10: h->mb_x, h->mb_y); yading@10: return -1; yading@10: } yading@10: } yading@10: yading@10: if ((h->left_samples_available & 0x8080) != 0x8080) { yading@10: mode = left[mode]; yading@10: if (is_chroma && (h->left_samples_available & 0x8080)) { yading@10: // mad cow disease mode, aka MBAFF + constrained_intra_pred yading@10: mode = ALZHEIMER_DC_L0T_PRED8x8 + yading@10: (!(h->left_samples_available & 0x8000)) + yading@10: 2 * (mode == DC_128_PRED8x8); yading@10: } yading@10: if (mode < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "left block unavailable for requested intra mode at %d %d\n", yading@10: h->mb_x, h->mb_y); yading@10: return -1; yading@10: } yading@10: } yading@10: yading@10: return mode; yading@10: } yading@10: yading@10: const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, yading@10: int *dst_length, int *consumed, int length) yading@10: { yading@10: int i, si, di; yading@10: uint8_t *dst; yading@10: int bufidx; yading@10: yading@10: // src[0]&0x80; // forbidden bit yading@10: h->nal_ref_idc = src[0] >> 5; yading@10: h->nal_unit_type = src[0] & 0x1F; yading@10: yading@10: src++; yading@10: length--; yading@10: yading@10: #define STARTCODE_TEST \ yading@10: if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \ yading@10: if (src[i + 2] != 3) { \ yading@10: /* startcode, so we must be past the end */ \ yading@10: length = i; \ yading@10: } \ yading@10: break; \ yading@10: } yading@10: #if HAVE_FAST_UNALIGNED yading@10: #define FIND_FIRST_ZERO \ yading@10: if (i > 0 && !src[i]) \ yading@10: i--; \ yading@10: while (src[i]) \ yading@10: i++ yading@10: #if HAVE_FAST_64BIT yading@10: for (i = 0; i + 1 < length; i += 9) { yading@10: if (!((~AV_RN64A(src + i) & yading@10: (AV_RN64A(src + i) - 0x0100010001000101ULL)) & yading@10: 0x8000800080008080ULL)) yading@10: continue; yading@10: FIND_FIRST_ZERO; yading@10: STARTCODE_TEST; yading@10: i -= 7; yading@10: } yading@10: #else yading@10: for (i = 0; i + 1 < length; i += 5) { yading@10: if (!((~AV_RN32A(src + i) & yading@10: (AV_RN32A(src + i) - 0x01000101U)) & yading@10: 0x80008080U)) yading@10: continue; yading@10: FIND_FIRST_ZERO; yading@10: STARTCODE_TEST; yading@10: i -= 3; yading@10: } yading@10: #endif yading@10: #else yading@10: for (i = 0; i + 1 < length; i += 2) { yading@10: if (src[i]) yading@10: continue; yading@10: if (i > 0 && src[i - 1] == 0) yading@10: i--; yading@10: STARTCODE_TEST; yading@10: } yading@10: #endif yading@10: yading@10: // use second escape buffer for inter data yading@10: bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; yading@10: yading@10: si = h->rbsp_buffer_size[bufidx]; yading@10: av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE); yading@10: dst = h->rbsp_buffer[bufidx]; yading@10: yading@10: if (dst == NULL) yading@10: return NULL; yading@10: yading@10: if(i>=length-1){ //no escaped 0 yading@10: *dst_length= length; yading@10: *consumed= length+1; //+1 for the header yading@10: if(h->avctx->flags2 & CODEC_FLAG2_FAST){ yading@10: return src; yading@10: }else{ yading@10: memcpy(dst, src, length); yading@10: return dst; yading@10: } yading@10: } yading@10: yading@10: memcpy(dst, src, i); yading@10: si = di = i; yading@10: while (si + 2 < length) { yading@10: // remove escapes (very rare 1:2^22) yading@10: if (src[si + 2] > 3) { yading@10: dst[di++] = src[si++]; yading@10: dst[di++] = src[si++]; yading@10: } else if (src[si] == 0 && src[si + 1] == 0) { yading@10: if (src[si + 2] == 3) { // escape yading@10: dst[di++] = 0; yading@10: dst[di++] = 0; yading@10: si += 3; yading@10: continue; yading@10: } else // next start code yading@10: goto nsc; yading@10: } yading@10: yading@10: dst[di++] = src[si++]; yading@10: } yading@10: while (si < length) yading@10: dst[di++] = src[si++]; yading@10: nsc: yading@10: yading@10: memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE); yading@10: yading@10: *dst_length = di; yading@10: *consumed = si + 1; // +1 for the header yading@10: /* FIXME store exact number of bits in the getbitcontext yading@10: * (it is needed for decoding) */ yading@10: return dst; yading@10: } yading@10: yading@10: /** yading@10: * Identify the exact end of the bitstream yading@10: * @return the length of the trailing, or 0 if damaged yading@10: */ yading@10: static int decode_rbsp_trailing(H264Context *h, const uint8_t *src) yading@10: { yading@10: int v = *src; yading@10: int r; yading@10: yading@10: tprintf(h->avctx, "rbsp trailing %X\n", v); yading@10: yading@10: for (r = 1; r < 9; r++) { yading@10: if (v & 1) yading@10: return r; yading@10: v >>= 1; yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, yading@10: int height, int y_offset, int list) yading@10: { yading@10: int raw_my = h->mv_cache[list][scan8[n]][1]; yading@10: int filter_height_down = (raw_my & 3) ? 3 : 0; yading@10: int full_my = (raw_my >> 2) + y_offset; yading@10: int bottom = full_my + filter_height_down + height; yading@10: yading@10: av_assert2(height >= 0); yading@10: yading@10: return FFMAX(0, bottom); yading@10: } yading@10: yading@10: static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, yading@10: int height, int y_offset, int list0, yading@10: int list1, int *nrefs) yading@10: { yading@10: int my; yading@10: yading@10: y_offset += 16 * (h->mb_y >> MB_FIELD(h)); yading@10: yading@10: if (list0) { yading@10: int ref_n = h->ref_cache[0][scan8[n]]; yading@10: Picture *ref = &h->ref_list[0][ref_n]; yading@10: yading@10: // Error resilience puts the current picture in the ref list. yading@10: // Don't try to wait on these as it will cause a deadlock. yading@10: // Fields can wait on each other, though. yading@10: if (ref->tf.progress->data != h->cur_pic.tf.progress->data || yading@10: (ref->reference & 3) != h->picture_structure) { yading@10: my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0); yading@10: if (refs[0][ref_n] < 0) yading@10: nrefs[0] += 1; yading@10: refs[0][ref_n] = FFMAX(refs[0][ref_n], my); yading@10: } yading@10: } yading@10: yading@10: if (list1) { yading@10: int ref_n = h->ref_cache[1][scan8[n]]; yading@10: Picture *ref = &h->ref_list[1][ref_n]; yading@10: yading@10: if (ref->tf.progress->data != h->cur_pic.tf.progress->data || yading@10: (ref->reference & 3) != h->picture_structure) { yading@10: my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1); yading@10: if (refs[1][ref_n] < 0) yading@10: nrefs[1] += 1; yading@10: refs[1][ref_n] = FFMAX(refs[1][ref_n], my); yading@10: } yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Wait until all reference frames are available for MC operations. yading@10: * yading@10: * @param h the H264 context yading@10: */ yading@10: static void await_references(H264Context *h) yading@10: { yading@10: const int mb_xy = h->mb_xy; yading@10: const int mb_type = h->cur_pic.mb_type[mb_xy]; yading@10: int refs[2][48]; yading@10: int nrefs[2] = { 0 }; yading@10: int ref, list; yading@10: yading@10: memset(refs, -1, sizeof(refs)); yading@10: yading@10: if (IS_16X16(mb_type)) { yading@10: get_lowest_part_y(h, refs, 0, 16, 0, yading@10: IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs); yading@10: } else if (IS_16X8(mb_type)) { yading@10: get_lowest_part_y(h, refs, 0, 8, 0, yading@10: IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs); yading@10: get_lowest_part_y(h, refs, 8, 8, 8, yading@10: IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs); yading@10: } else if (IS_8X16(mb_type)) { yading@10: get_lowest_part_y(h, refs, 0, 16, 0, yading@10: IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs); yading@10: get_lowest_part_y(h, refs, 4, 16, 0, yading@10: IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs); yading@10: } else { yading@10: int i; yading@10: yading@10: av_assert2(IS_8X8(mb_type)); yading@10: yading@10: for (i = 0; i < 4; i++) { yading@10: const int sub_mb_type = h->sub_mb_type[i]; yading@10: const int n = 4 * i; yading@10: int y_offset = (i & 2) << 2; yading@10: yading@10: if (IS_SUB_8X8(sub_mb_type)) { yading@10: get_lowest_part_y(h, refs, n, 8, y_offset, yading@10: IS_DIR(sub_mb_type, 0, 0), yading@10: IS_DIR(sub_mb_type, 0, 1), yading@10: nrefs); yading@10: } else if (IS_SUB_8X4(sub_mb_type)) { yading@10: get_lowest_part_y(h, refs, n, 4, y_offset, yading@10: IS_DIR(sub_mb_type, 0, 0), yading@10: IS_DIR(sub_mb_type, 0, 1), yading@10: nrefs); yading@10: get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4, yading@10: IS_DIR(sub_mb_type, 0, 0), yading@10: IS_DIR(sub_mb_type, 0, 1), yading@10: nrefs); yading@10: } else if (IS_SUB_4X8(sub_mb_type)) { yading@10: get_lowest_part_y(h, refs, n, 8, y_offset, yading@10: IS_DIR(sub_mb_type, 0, 0), yading@10: IS_DIR(sub_mb_type, 0, 1), yading@10: nrefs); yading@10: get_lowest_part_y(h, refs, n + 1, 8, y_offset, yading@10: IS_DIR(sub_mb_type, 0, 0), yading@10: IS_DIR(sub_mb_type, 0, 1), yading@10: nrefs); yading@10: } else { yading@10: int j; yading@10: av_assert2(IS_SUB_4X4(sub_mb_type)); yading@10: for (j = 0; j < 4; j++) { yading@10: int sub_y_offset = y_offset + 2 * (j & 2); yading@10: get_lowest_part_y(h, refs, n + j, 4, sub_y_offset, yading@10: IS_DIR(sub_mb_type, 0, 0), yading@10: IS_DIR(sub_mb_type, 0, 1), yading@10: nrefs); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: for (list = h->list_count - 1; list >= 0; list--) yading@10: for (ref = 0; ref < 48 && nrefs[list]; ref++) { yading@10: int row = refs[list][ref]; yading@10: if (row >= 0) { yading@10: Picture *ref_pic = &h->ref_list[list][ref]; yading@10: int ref_field = ref_pic->reference - 1; yading@10: int ref_field_picture = ref_pic->field_picture; yading@10: int pic_height = 16 * h->mb_height >> ref_field_picture; yading@10: yading@10: row <<= MB_MBAFF(h); yading@10: nrefs[list]--; yading@10: yading@10: if (!FIELD_PICTURE(h) && ref_field_picture) { // frame referencing two fields yading@10: ff_thread_await_progress(&ref_pic->tf, yading@10: FFMIN((row >> 1) - !(row & 1), yading@10: pic_height - 1), yading@10: 1); yading@10: ff_thread_await_progress(&ref_pic->tf, yading@10: FFMIN((row >> 1), pic_height - 1), yading@10: 0); yading@10: } else if (FIELD_PICTURE(h) && !ref_field_picture) { // field referencing one field of a frame yading@10: ff_thread_await_progress(&ref_pic->tf, yading@10: FFMIN(row * 2 + ref_field, yading@10: pic_height - 1), yading@10: 0); yading@10: } else if (FIELD_PICTURE(h)) { yading@10: ff_thread_await_progress(&ref_pic->tf, yading@10: FFMIN(row, pic_height - 1), yading@10: ref_field); yading@10: } else { yading@10: ff_thread_await_progress(&ref_pic->tf, yading@10: FFMIN(row, pic_height - 1), yading@10: 0); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void mc_dir_part(H264Context *h, Picture *pic, yading@10: int n, int square, int height, yading@10: int delta, int list, yading@10: uint8_t *dest_y, uint8_t *dest_cb, yading@10: uint8_t *dest_cr, yading@10: int src_x_offset, int src_y_offset, yading@10: qpel_mc_func *qpix_op, yading@10: h264_chroma_mc_func chroma_op, yading@10: int pixel_shift, int chroma_idc) yading@10: { yading@10: const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8; yading@10: int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8; yading@10: const int luma_xy = (mx & 3) + ((my & 3) << 2); yading@10: int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize; yading@10: uint8_t *src_y = pic->f.data[0] + offset; yading@10: uint8_t *src_cb, *src_cr; yading@10: int extra_width = 0; yading@10: int extra_height = 0; yading@10: int emu = 0; yading@10: const int full_mx = mx >> 2; yading@10: const int full_my = my >> 2; yading@10: const int pic_width = 16 * h->mb_width; yading@10: const int pic_height = 16 * h->mb_height >> MB_FIELD(h); yading@10: int ysh; yading@10: yading@10: if (mx & 7) yading@10: extra_width -= 3; yading@10: if (my & 7) yading@10: extra_height -= 3; yading@10: yading@10: if (full_mx < 0 - extra_width || yading@10: full_my < 0 - extra_height || yading@10: full_mx + 16 /*FIXME*/ > pic_width + extra_width || yading@10: full_my + 16 /*FIXME*/ > pic_height + extra_height) { yading@10: h->vdsp.emulated_edge_mc(h->edge_emu_buffer, yading@10: src_y - (2 << pixel_shift) - 2 * h->mb_linesize, yading@10: h->mb_linesize, yading@10: 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2, yading@10: full_my - 2, pic_width, pic_height); yading@10: src_y = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize; yading@10: emu = 1; yading@10: } yading@10: yading@10: qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps? yading@10: if (!square) yading@10: qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize); yading@10: yading@10: if (CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY) yading@10: return; yading@10: yading@10: if (chroma_idc == 3 /* yuv444 */) { yading@10: src_cb = pic->f.data[1] + offset; yading@10: if (emu) { yading@10: h->vdsp.emulated_edge_mc(h->edge_emu_buffer, yading@10: src_cb - (2 << pixel_shift) - 2 * h->mb_linesize, yading@10: h->mb_linesize, yading@10: 16 + 5, 16 + 5 /*FIXME*/, yading@10: full_mx - 2, full_my - 2, yading@10: pic_width, pic_height); yading@10: src_cb = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize; yading@10: } yading@10: qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps? yading@10: if (!square) yading@10: qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize); yading@10: yading@10: src_cr = pic->f.data[2] + offset; yading@10: if (emu) { yading@10: h->vdsp.emulated_edge_mc(h->edge_emu_buffer, yading@10: src_cr - (2 << pixel_shift) - 2 * h->mb_linesize, yading@10: h->mb_linesize, yading@10: 16 + 5, 16 + 5 /*FIXME*/, yading@10: full_mx - 2, full_my - 2, yading@10: pic_width, pic_height); yading@10: src_cr = h->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize; yading@10: } yading@10: qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps? yading@10: if (!square) yading@10: qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize); yading@10: return; yading@10: } yading@10: yading@10: ysh = 3 - (chroma_idc == 2 /* yuv422 */); yading@10: if (chroma_idc == 1 /* yuv420 */ && MB_FIELD(h)) { yading@10: // chroma offset when predicting from a field of opposite parity yading@10: my += 2 * ((h->mb_y & 1) - (pic->reference - 1)); yading@10: emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1); yading@10: } yading@10: yading@10: src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + yading@10: (my >> ysh) * h->mb_uvlinesize; yading@10: src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + yading@10: (my >> ysh) * h->mb_uvlinesize; yading@10: yading@10: if (emu) { yading@10: h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->mb_uvlinesize, yading@10: 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh), yading@10: pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */)); yading@10: src_cb = h->edge_emu_buffer; yading@10: } yading@10: chroma_op(dest_cb, src_cb, h->mb_uvlinesize, yading@10: height >> (chroma_idc == 1 /* yuv420 */), yading@10: mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7); yading@10: yading@10: if (emu) { yading@10: h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->mb_uvlinesize, yading@10: 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh), yading@10: pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */)); yading@10: src_cr = h->edge_emu_buffer; yading@10: } yading@10: chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */), yading@10: mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7); yading@10: } yading@10: yading@10: static av_always_inline void mc_part_std(H264Context *h, int n, int square, yading@10: int height, int delta, yading@10: uint8_t *dest_y, uint8_t *dest_cb, yading@10: uint8_t *dest_cr, yading@10: int x_offset, int y_offset, yading@10: qpel_mc_func *qpix_put, yading@10: h264_chroma_mc_func chroma_put, yading@10: qpel_mc_func *qpix_avg, yading@10: h264_chroma_mc_func chroma_avg, yading@10: int list0, int list1, yading@10: int pixel_shift, int chroma_idc) yading@10: { yading@10: qpel_mc_func *qpix_op = qpix_put; yading@10: h264_chroma_mc_func chroma_op = chroma_put; yading@10: yading@10: dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize; yading@10: if (chroma_idc == 3 /* yuv444 */) { yading@10: dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize; yading@10: dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize; yading@10: } else if (chroma_idc == 2 /* yuv422 */) { yading@10: dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize; yading@10: dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize; yading@10: } else { /* yuv420 */ yading@10: dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize; yading@10: dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize; yading@10: } yading@10: x_offset += 8 * h->mb_x; yading@10: y_offset += 8 * (h->mb_y >> MB_FIELD(h)); yading@10: yading@10: if (list0) { yading@10: Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]]; yading@10: mc_dir_part(h, ref, n, square, height, delta, 0, yading@10: dest_y, dest_cb, dest_cr, x_offset, y_offset, yading@10: qpix_op, chroma_op, pixel_shift, chroma_idc); yading@10: yading@10: qpix_op = qpix_avg; yading@10: chroma_op = chroma_avg; yading@10: } yading@10: yading@10: if (list1) { yading@10: Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]]; yading@10: mc_dir_part(h, ref, n, square, height, delta, 1, yading@10: dest_y, dest_cb, dest_cr, x_offset, y_offset, yading@10: qpix_op, chroma_op, pixel_shift, chroma_idc); yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void mc_part_weighted(H264Context *h, int n, int square, yading@10: int height, int delta, yading@10: uint8_t *dest_y, uint8_t *dest_cb, yading@10: uint8_t *dest_cr, yading@10: int x_offset, int y_offset, yading@10: qpel_mc_func *qpix_put, yading@10: h264_chroma_mc_func chroma_put, yading@10: h264_weight_func luma_weight_op, yading@10: h264_weight_func chroma_weight_op, yading@10: h264_biweight_func luma_weight_avg, yading@10: h264_biweight_func chroma_weight_avg, yading@10: int list0, int list1, yading@10: int pixel_shift, int chroma_idc) yading@10: { yading@10: int chroma_height; yading@10: yading@10: dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize; yading@10: if (chroma_idc == 3 /* yuv444 */) { yading@10: chroma_height = height; yading@10: chroma_weight_avg = luma_weight_avg; yading@10: chroma_weight_op = luma_weight_op; yading@10: dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize; yading@10: dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize; yading@10: } else if (chroma_idc == 2 /* yuv422 */) { yading@10: chroma_height = height; yading@10: dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize; yading@10: dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize; yading@10: } else { /* yuv420 */ yading@10: chroma_height = height >> 1; yading@10: dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize; yading@10: dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize; yading@10: } yading@10: x_offset += 8 * h->mb_x; yading@10: y_offset += 8 * (h->mb_y >> MB_FIELD(h)); yading@10: yading@10: if (list0 && list1) { yading@10: /* don't optimize for luma-only case, since B-frames usually yading@10: * use implicit weights => chroma too. */ yading@10: uint8_t *tmp_cb = h->bipred_scratchpad; yading@10: uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift); yading@10: uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize; yading@10: int refn0 = h->ref_cache[0][scan8[n]]; yading@10: int refn1 = h->ref_cache[1][scan8[n]]; yading@10: yading@10: mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0, yading@10: dest_y, dest_cb, dest_cr, yading@10: x_offset, y_offset, qpix_put, chroma_put, yading@10: pixel_shift, chroma_idc); yading@10: mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1, yading@10: tmp_y, tmp_cb, tmp_cr, yading@10: x_offset, y_offset, qpix_put, chroma_put, yading@10: pixel_shift, chroma_idc); yading@10: yading@10: if (h->use_weight == 2) { yading@10: int weight0 = h->implicit_weight[refn0][refn1][h->mb_y & 1]; yading@10: int weight1 = 64 - weight0; yading@10: luma_weight_avg(dest_y, tmp_y, h->mb_linesize, yading@10: height, 5, weight0, weight1, 0); yading@10: chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, yading@10: chroma_height, 5, weight0, weight1, 0); yading@10: chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, yading@10: chroma_height, 5, weight0, weight1, 0); yading@10: } else { yading@10: luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, yading@10: h->luma_log2_weight_denom, yading@10: h->luma_weight[refn0][0][0], yading@10: h->luma_weight[refn1][1][0], yading@10: h->luma_weight[refn0][0][1] + yading@10: h->luma_weight[refn1][1][1]); yading@10: chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, yading@10: h->chroma_log2_weight_denom, yading@10: h->chroma_weight[refn0][0][0][0], yading@10: h->chroma_weight[refn1][1][0][0], yading@10: h->chroma_weight[refn0][0][0][1] + yading@10: h->chroma_weight[refn1][1][0][1]); yading@10: chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, yading@10: h->chroma_log2_weight_denom, yading@10: h->chroma_weight[refn0][0][1][0], yading@10: h->chroma_weight[refn1][1][1][0], yading@10: h->chroma_weight[refn0][0][1][1] + yading@10: h->chroma_weight[refn1][1][1][1]); yading@10: } yading@10: } else { yading@10: int list = list1 ? 1 : 0; yading@10: int refn = h->ref_cache[list][scan8[n]]; yading@10: Picture *ref = &h->ref_list[list][refn]; yading@10: mc_dir_part(h, ref, n, square, height, delta, list, yading@10: dest_y, dest_cb, dest_cr, x_offset, y_offset, yading@10: qpix_put, chroma_put, pixel_shift, chroma_idc); yading@10: yading@10: luma_weight_op(dest_y, h->mb_linesize, height, yading@10: h->luma_log2_weight_denom, yading@10: h->luma_weight[refn][list][0], yading@10: h->luma_weight[refn][list][1]); yading@10: if (h->use_weight_chroma) { yading@10: chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, yading@10: h->chroma_log2_weight_denom, yading@10: h->chroma_weight[refn][list][0][0], yading@10: h->chroma_weight[refn][list][0][1]); yading@10: chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, yading@10: h->chroma_log2_weight_denom, yading@10: h->chroma_weight[refn][list][1][0], yading@10: h->chroma_weight[refn][list][1][1]); yading@10: } yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void prefetch_motion(H264Context *h, int list, yading@10: int pixel_shift, int chroma_idc) yading@10: { yading@10: /* fetch pixels for estimated mv 4 macroblocks ahead yading@10: * optimized for 64byte cache lines */ yading@10: const int refn = h->ref_cache[list][scan8[0]]; yading@10: if (refn >= 0) { yading@10: const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * h->mb_x + 8; yading@10: const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * h->mb_y; yading@10: uint8_t **src = h->ref_list[list][refn].f.data; yading@10: int off = (mx << pixel_shift) + yading@10: (my + (h->mb_x & 3) * 4) * h->mb_linesize + yading@10: (64 << pixel_shift); yading@10: h->vdsp.prefetch(src[0] + off, h->linesize, 4); yading@10: if (chroma_idc == 3 /* yuv444 */) { yading@10: h->vdsp.prefetch(src[1] + off, h->linesize, 4); yading@10: h->vdsp.prefetch(src[2] + off, h->linesize, 4); yading@10: } else { yading@10: off= (((mx>>1)+64)<>1) + (h->mb_x&7))*h->uvlinesize; yading@10: h->vdsp.prefetch(src[1] + off, src[2] - src[1], 2); yading@10: } yading@10: } yading@10: } yading@10: yading@10: static void free_tables(H264Context *h, int free_rbsp) yading@10: { yading@10: int i; yading@10: H264Context *hx; yading@10: yading@10: av_freep(&h->intra4x4_pred_mode); yading@10: av_freep(&h->chroma_pred_mode_table); yading@10: av_freep(&h->cbp_table); yading@10: av_freep(&h->mvd_table[0]); yading@10: av_freep(&h->mvd_table[1]); yading@10: av_freep(&h->direct_table); yading@10: av_freep(&h->non_zero_count); yading@10: av_freep(&h->slice_table_base); yading@10: h->slice_table = NULL; yading@10: av_freep(&h->list_counts); yading@10: yading@10: av_freep(&h->mb2b_xy); yading@10: av_freep(&h->mb2br_xy); yading@10: yading@10: for (i = 0; i < 3; i++) yading@10: av_freep(&h->visualization_buffer[i]); yading@10: yading@10: av_buffer_pool_uninit(&h->qscale_table_pool); yading@10: av_buffer_pool_uninit(&h->mb_type_pool); yading@10: av_buffer_pool_uninit(&h->motion_val_pool); yading@10: av_buffer_pool_uninit(&h->ref_index_pool); yading@10: yading@10: if (free_rbsp && h->DPB) { yading@10: for (i = 0; i < MAX_PICTURE_COUNT; i++) yading@10: unref_picture(h, &h->DPB[i]); yading@10: av_freep(&h->DPB); yading@10: } else if (h->DPB) { yading@10: for (i = 0; i < MAX_PICTURE_COUNT; i++) yading@10: h->DPB[i].needs_realloc = 1; yading@10: } yading@10: yading@10: h->cur_pic_ptr = NULL; yading@10: yading@10: for (i = 0; i < MAX_THREADS; i++) { yading@10: hx = h->thread_context[i]; yading@10: if (!hx) yading@10: continue; yading@10: av_freep(&hx->top_borders[1]); yading@10: av_freep(&hx->top_borders[0]); yading@10: av_freep(&hx->bipred_scratchpad); yading@10: av_freep(&hx->edge_emu_buffer); yading@10: av_freep(&hx->dc_val_base); yading@10: av_freep(&hx->me.scratchpad); yading@10: av_freep(&hx->er.mb_index2xy); yading@10: av_freep(&hx->er.error_status_table); yading@10: av_freep(&hx->er.er_temp_buffer); yading@10: av_freep(&hx->er.mbintra_table); yading@10: av_freep(&hx->er.mbskip_table); yading@10: yading@10: if (free_rbsp) { yading@10: av_freep(&hx->rbsp_buffer[1]); yading@10: av_freep(&hx->rbsp_buffer[0]); yading@10: hx->rbsp_buffer_size[0] = 0; yading@10: hx->rbsp_buffer_size[1] = 0; yading@10: } yading@10: if (i) yading@10: av_freep(&h->thread_context[i]); yading@10: } yading@10: } yading@10: yading@10: static void init_dequant8_coeff_table(H264Context *h) yading@10: { yading@10: int i, j, q, x; yading@10: const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8); yading@10: yading@10: for (i = 0; i < 6; i++) { yading@10: h->dequant8_coeff[i] = h->dequant8_buffer[i]; yading@10: for (j = 0; j < i; j++) yading@10: if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], yading@10: 64 * sizeof(uint8_t))) { yading@10: h->dequant8_coeff[i] = h->dequant8_buffer[j]; yading@10: break; yading@10: } yading@10: if (j < i) yading@10: continue; yading@10: yading@10: for (q = 0; q < max_qp + 1; q++) { yading@10: int shift = div6[q]; yading@10: int idx = rem6[q]; yading@10: for (x = 0; x < 64; x++) yading@10: h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] = yading@10: ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] * yading@10: h->pps.scaling_matrix8[i][x]) << shift; yading@10: } yading@10: } yading@10: } yading@10: yading@10: static void init_dequant4_coeff_table(H264Context *h) yading@10: { yading@10: int i, j, q, x; yading@10: const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8); yading@10: for (i = 0; i < 6; i++) { yading@10: h->dequant4_coeff[i] = h->dequant4_buffer[i]; yading@10: for (j = 0; j < i; j++) yading@10: if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], yading@10: 16 * sizeof(uint8_t))) { yading@10: h->dequant4_coeff[i] = h->dequant4_buffer[j]; yading@10: break; yading@10: } yading@10: if (j < i) yading@10: continue; yading@10: yading@10: for (q = 0; q < max_qp + 1; q++) { yading@10: int shift = div6[q] + 2; yading@10: int idx = rem6[q]; yading@10: for (x = 0; x < 16; x++) yading@10: h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] = yading@10: ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] * yading@10: h->pps.scaling_matrix4[i][x]) << shift; yading@10: } yading@10: } yading@10: } yading@10: yading@10: static void init_dequant_tables(H264Context *h) yading@10: { yading@10: int i, x; yading@10: init_dequant4_coeff_table(h); yading@10: if (h->pps.transform_8x8_mode) yading@10: init_dequant8_coeff_table(h); yading@10: if (h->sps.transform_bypass) { yading@10: for (i = 0; i < 6; i++) yading@10: for (x = 0; x < 16; x++) yading@10: h->dequant4_coeff[i][0][x] = 1 << 6; yading@10: if (h->pps.transform_8x8_mode) yading@10: for (i = 0; i < 6; i++) yading@10: for (x = 0; x < 64; x++) yading@10: h->dequant8_coeff[i][0][x] = 1 << 6; yading@10: } yading@10: } yading@10: yading@10: int ff_h264_alloc_tables(H264Context *h) yading@10: { yading@10: const int big_mb_num = h->mb_stride * (h->mb_height + 1); yading@10: const int row_mb_num = 2*h->mb_stride*FFMAX(h->avctx->thread_count, 1); yading@10: int x, y, i; yading@10: yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode, yading@10: row_mb_num * 8 * sizeof(uint8_t), fail) yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count, yading@10: big_mb_num * 48 * sizeof(uint8_t), fail) yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base, yading@10: (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail) yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table, yading@10: big_mb_num * sizeof(uint16_t), fail) yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table, yading@10: big_mb_num * sizeof(uint8_t), fail) yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0], yading@10: 16 * row_mb_num * sizeof(uint8_t), fail); yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1], yading@10: 16 * row_mb_num * sizeof(uint8_t), fail); yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table, yading@10: 4 * big_mb_num * sizeof(uint8_t), fail); yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts, yading@10: big_mb_num * sizeof(uint8_t), fail) yading@10: yading@10: memset(h->slice_table_base, -1, yading@10: (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base)); yading@10: h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1; yading@10: yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy, yading@10: big_mb_num * sizeof(uint32_t), fail); yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy, yading@10: big_mb_num * sizeof(uint32_t), fail); yading@10: for (y = 0; y < h->mb_height; y++) yading@10: for (x = 0; x < h->mb_width; x++) { yading@10: const int mb_xy = x + y * h->mb_stride; yading@10: const int b_xy = 4 * x + 4 * y * h->b_stride; yading@10: yading@10: h->mb2b_xy[mb_xy] = b_xy; yading@10: h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride))); yading@10: } yading@10: yading@10: if (!h->dequant4_coeff[0]) yading@10: init_dequant_tables(h); yading@10: yading@10: if (!h->DPB) { yading@10: h->DPB = av_mallocz_array(MAX_PICTURE_COUNT, sizeof(*h->DPB)); yading@10: if (!h->DPB) yading@10: return AVERROR(ENOMEM); yading@10: for (i = 0; i < MAX_PICTURE_COUNT; i++) yading@10: avcodec_get_frame_defaults(&h->DPB[i].f); yading@10: avcodec_get_frame_defaults(&h->cur_pic.f); yading@10: } yading@10: yading@10: return 0; yading@10: yading@10: fail: yading@10: free_tables(h, 1); yading@10: return -1; yading@10: } yading@10: yading@10: /** yading@10: * Mimic alloc_tables(), but for every context thread. yading@10: */ yading@10: static void clone_tables(H264Context *dst, H264Context *src, int i) yading@10: { yading@10: dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride; yading@10: dst->non_zero_count = src->non_zero_count; yading@10: dst->slice_table = src->slice_table; yading@10: dst->cbp_table = src->cbp_table; yading@10: dst->mb2b_xy = src->mb2b_xy; yading@10: dst->mb2br_xy = src->mb2br_xy; yading@10: dst->chroma_pred_mode_table = src->chroma_pred_mode_table; yading@10: dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride; yading@10: dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride; yading@10: dst->direct_table = src->direct_table; yading@10: dst->list_counts = src->list_counts; yading@10: dst->DPB = src->DPB; yading@10: dst->cur_pic_ptr = src->cur_pic_ptr; yading@10: dst->cur_pic = src->cur_pic; yading@10: dst->bipred_scratchpad = NULL; yading@10: dst->edge_emu_buffer = NULL; yading@10: dst->me.scratchpad = NULL; yading@10: ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma, yading@10: src->sps.chroma_format_idc); yading@10: } yading@10: yading@10: /** yading@10: * Init context yading@10: * Allocate buffers which are not shared amongst multiple threads. yading@10: */ yading@10: static int context_init(H264Context *h) yading@10: { yading@10: ERContext *er = &h->er; yading@10: int mb_array_size = h->mb_height * h->mb_stride; yading@10: int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1); yading@10: int c_size = h->mb_stride * (h->mb_height + 1); yading@10: int yc_size = y_size + 2 * c_size; yading@10: int x, y, i; yading@10: yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[0], yading@10: h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail) yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->top_borders[1], yading@10: h->mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail) yading@10: yading@10: h->ref_cache[0][scan8[5] + 1] = yading@10: h->ref_cache[0][scan8[7] + 1] = yading@10: h->ref_cache[0][scan8[13] + 1] = yading@10: h->ref_cache[1][scan8[5] + 1] = yading@10: h->ref_cache[1][scan8[7] + 1] = yading@10: h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE; yading@10: yading@10: if (CONFIG_ERROR_RESILIENCE) { yading@10: /* init ER */ yading@10: er->avctx = h->avctx; yading@10: er->dsp = &h->dsp; yading@10: er->decode_mb = h264_er_decode_mb; yading@10: er->opaque = h; yading@10: er->quarter_sample = 1; yading@10: yading@10: er->mb_num = h->mb_num; yading@10: er->mb_width = h->mb_width; yading@10: er->mb_height = h->mb_height; yading@10: er->mb_stride = h->mb_stride; yading@10: er->b8_stride = h->mb_width * 2 + 1; yading@10: yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy, (h->mb_num + 1) * sizeof(int), yading@10: fail); // error ressilience code looks cleaner with this yading@10: for (y = 0; y < h->mb_height; y++) yading@10: for (x = 0; x < h->mb_width; x++) yading@10: er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride; yading@10: yading@10: er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) * yading@10: h->mb_stride + h->mb_width; yading@10: yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table, yading@10: mb_array_size * sizeof(uint8_t), fail); yading@10: yading@10: FF_ALLOC_OR_GOTO(h->avctx, er->mbintra_table, mb_array_size, fail); yading@10: memset(er->mbintra_table, 1, mb_array_size); yading@10: yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, er->mbskip_table, mb_array_size + 2, fail); yading@10: yading@10: FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer, h->mb_height * h->mb_stride, yading@10: fail); yading@10: yading@10: FF_ALLOCZ_OR_GOTO(h->avctx, h->dc_val_base, yc_size * sizeof(int16_t), fail); yading@10: er->dc_val[0] = h->dc_val_base + h->mb_width * 2 + 2; yading@10: er->dc_val[1] = h->dc_val_base + y_size + h->mb_stride + 1; yading@10: er->dc_val[2] = er->dc_val[1] + c_size; yading@10: for (i = 0; i < yc_size; i++) yading@10: h->dc_val_base[i] = 1024; yading@10: } yading@10: yading@10: return 0; yading@10: yading@10: fail: yading@10: return -1; // free_tables will clean up for us yading@10: } yading@10: yading@10: static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size, yading@10: int parse_extradata); yading@10: yading@10: int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size) yading@10: { yading@10: AVCodecContext *avctx = h->avctx; yading@10: yading@10: if (!buf || size <= 0) yading@10: return -1; yading@10: yading@10: if (buf[0] == 1) { yading@10: int i, cnt, nalsize; yading@10: const unsigned char *p = buf; yading@10: yading@10: h->is_avc = 1; yading@10: yading@10: if (size < 7) { yading@10: av_log(avctx, AV_LOG_ERROR, "avcC too short\n"); yading@10: return -1; yading@10: } yading@10: /* sps and pps in the avcC always have length coded with 2 bytes, yading@10: * so put a fake nal_length_size = 2 while parsing them */ yading@10: h->nal_length_size = 2; yading@10: // Decode sps from avcC yading@10: cnt = *(p + 5) & 0x1f; // Number of sps yading@10: p += 6; yading@10: for (i = 0; i < cnt; i++) { yading@10: nalsize = AV_RB16(p) + 2; yading@10: if(nalsize > size - (p-buf)) yading@10: return -1; yading@10: if (decode_nal_units(h, p, nalsize, 1) < 0) { yading@10: av_log(avctx, AV_LOG_ERROR, yading@10: "Decoding sps %d from avcC failed\n", i); yading@10: return -1; yading@10: } yading@10: p += nalsize; yading@10: } yading@10: // Decode pps from avcC yading@10: cnt = *(p++); // Number of pps yading@10: for (i = 0; i < cnt; i++) { yading@10: nalsize = AV_RB16(p) + 2; yading@10: if(nalsize > size - (p-buf)) yading@10: return -1; yading@10: if (decode_nal_units(h, p, nalsize, 1) < 0) { yading@10: av_log(avctx, AV_LOG_ERROR, yading@10: "Decoding pps %d from avcC failed\n", i); yading@10: return -1; yading@10: } yading@10: p += nalsize; yading@10: } yading@10: // Now store right nal length size, that will be used to parse all other nals yading@10: h->nal_length_size = (buf[4] & 0x03) + 1; yading@10: } else { yading@10: h->is_avc = 0; yading@10: if (decode_nal_units(h, buf, size, 1) < 0) yading@10: return -1; yading@10: } yading@10: return size; yading@10: } yading@10: yading@10: av_cold int ff_h264_decode_init(AVCodecContext *avctx) yading@10: { yading@10: H264Context *h = avctx->priv_data; yading@10: int i; yading@10: yading@10: h->avctx = avctx; yading@10: yading@10: h->bit_depth_luma = 8; yading@10: h->chroma_format_idc = 1; yading@10: yading@10: h->avctx->bits_per_raw_sample = 8; yading@10: h->cur_chroma_format_idc = 1; yading@10: yading@10: ff_h264dsp_init(&h->h264dsp, 8, 1); yading@10: av_assert0(h->sps.bit_depth_chroma == 0); yading@10: ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma); yading@10: ff_h264qpel_init(&h->h264qpel, 8); yading@10: ff_h264_pred_init(&h->hpc, h->avctx->codec_id, 8, 1); yading@10: yading@10: h->dequant_coeff_pps = -1; yading@10: yading@10: /* needed so that IDCT permutation is known early */ yading@10: if (CONFIG_ERROR_RESILIENCE) yading@10: ff_dsputil_init(&h->dsp, h->avctx); yading@10: ff_videodsp_init(&h->vdsp, 8); yading@10: yading@10: memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t)); yading@10: memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t)); yading@10: yading@10: h->picture_structure = PICT_FRAME; yading@10: h->slice_context_count = 1; yading@10: h->workaround_bugs = avctx->workaround_bugs; yading@10: h->flags = avctx->flags; yading@10: yading@10: /* set defaults */ yading@10: // s->decode_mb = ff_h263_decode_mb; yading@10: if (!avctx->has_b_frames) yading@10: h->low_delay = 1; yading@10: yading@10: avctx->chroma_sample_location = AVCHROMA_LOC_LEFT; yading@10: yading@10: ff_h264_decode_init_vlc(); yading@10: yading@10: h->pixel_shift = 0; yading@10: h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8; yading@10: yading@10: h->thread_context[0] = h; yading@10: h->outputed_poc = h->next_outputed_poc = INT_MIN; yading@10: for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) yading@10: h->last_pocs[i] = INT_MIN; yading@10: h->prev_poc_msb = 1 << 16; yading@10: h->prev_frame_num = -1; yading@10: h->x264_build = -1; yading@10: ff_h264_reset_sei(h); yading@10: if (avctx->codec_id == AV_CODEC_ID_H264) { yading@10: if (avctx->ticks_per_frame == 1) { yading@10: if(h->avctx->time_base.den < INT_MAX/2) { yading@10: h->avctx->time_base.den *= 2; yading@10: } else yading@10: h->avctx->time_base.num /= 2; yading@10: } yading@10: avctx->ticks_per_frame = 2; yading@10: } yading@10: yading@10: if (avctx->extradata_size > 0 && avctx->extradata && yading@10: ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) { yading@10: ff_h264_free_context(h); yading@10: return -1; yading@10: } yading@10: yading@10: if (h->sps.bitstream_restriction_flag && yading@10: h->avctx->has_b_frames < h->sps.num_reorder_frames) { yading@10: h->avctx->has_b_frames = h->sps.num_reorder_frames; yading@10: h->low_delay = 0; yading@10: } yading@10: yading@10: ff_init_cabac_states(); yading@10: avctx->internal->allocate_progress = 1; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size)))) yading@10: #undef REBASE_PICTURE yading@10: #define REBASE_PICTURE(pic, new_ctx, old_ctx) \ yading@10: ((pic && pic >= old_ctx->DPB && \ yading@10: pic < old_ctx->DPB + MAX_PICTURE_COUNT) ? \ yading@10: &new_ctx->DPB[pic - old_ctx->DPB] : NULL) yading@10: yading@10: static void copy_picture_range(Picture **to, Picture **from, int count, yading@10: H264Context *new_base, yading@10: H264Context *old_base) yading@10: { yading@10: int i; yading@10: yading@10: for (i = 0; i < count; i++) { yading@10: assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) || yading@10: IN_RANGE(from[i], old_base->DPB, yading@10: sizeof(Picture) * MAX_PICTURE_COUNT) || yading@10: !from[i])); yading@10: to[i] = REBASE_PICTURE(from[i], new_base, old_base); yading@10: } yading@10: } yading@10: yading@10: static void copy_parameter_set(void **to, void **from, int count, int size) yading@10: { yading@10: int i; yading@10: yading@10: for (i = 0; i < count; i++) { yading@10: if (to[i] && !from[i]) yading@10: av_freep(&to[i]); yading@10: else if (from[i] && !to[i]) yading@10: to[i] = av_malloc(size); yading@10: yading@10: if (from[i]) yading@10: memcpy(to[i], from[i], size); yading@10: } yading@10: } yading@10: yading@10: static int decode_init_thread_copy(AVCodecContext *avctx) yading@10: { yading@10: H264Context *h = avctx->priv_data; yading@10: yading@10: if (!avctx->internal->is_copy) yading@10: return 0; yading@10: memset(h->sps_buffers, 0, sizeof(h->sps_buffers)); yading@10: memset(h->pps_buffers, 0, sizeof(h->pps_buffers)); yading@10: yading@10: h->context_initialized = 0; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: #define copy_fields(to, from, start_field, end_field) \ yading@10: memcpy(&to->start_field, &from->start_field, \ yading@10: (char *)&to->end_field - (char *)&to->start_field) yading@10: yading@10: static int h264_slice_header_init(H264Context *, int); yading@10: yading@10: static int h264_set_parameter_from_sps(H264Context *h); yading@10: yading@10: static int decode_update_thread_context(AVCodecContext *dst, yading@10: const AVCodecContext *src) yading@10: { yading@10: H264Context *h = dst->priv_data, *h1 = src->priv_data; yading@10: int inited = h->context_initialized, err = 0; yading@10: int context_reinitialized = 0; yading@10: int i, ret; yading@10: yading@10: if (dst == src) yading@10: return 0; yading@10: yading@10: if (inited && yading@10: (h->width != h1->width || yading@10: h->height != h1->height || yading@10: h->mb_width != h1->mb_width || yading@10: h->mb_height != h1->mb_height || yading@10: h->sps.bit_depth_luma != h1->sps.bit_depth_luma || yading@10: h->sps.chroma_format_idc != h1->sps.chroma_format_idc || yading@10: h->sps.colorspace != h1->sps.colorspace)) { yading@10: yading@10: /* set bits_per_raw_sample to the previous value. the check for changed yading@10: * bit depth in h264_set_parameter_from_sps() uses it and sets it to yading@10: * the current value */ yading@10: h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma; yading@10: yading@10: av_freep(&h->bipred_scratchpad); yading@10: yading@10: h->width = h1->width; yading@10: h->height = h1->height; yading@10: h->mb_height = h1->mb_height; yading@10: h->mb_width = h1->mb_width; yading@10: h->mb_num = h1->mb_num; yading@10: h->mb_stride = h1->mb_stride; yading@10: h->b_stride = h1->b_stride; yading@10: // SPS/PPS yading@10: copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers, yading@10: MAX_SPS_COUNT, sizeof(SPS)); yading@10: h->sps = h1->sps; yading@10: copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers, yading@10: MAX_PPS_COUNT, sizeof(PPS)); yading@10: h->pps = h1->pps; yading@10: yading@10: if ((err = h264_slice_header_init(h, 1)) < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed"); yading@10: return err; yading@10: } yading@10: context_reinitialized = 1; yading@10: yading@10: #if 0 yading@10: h264_set_parameter_from_sps(h); yading@10: //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted yading@10: h->cur_chroma_format_idc = h1->cur_chroma_format_idc; yading@10: #endif yading@10: } yading@10: /* update linesize on resize for h264. The h264 decoder doesn't yading@10: * necessarily call ff_MPV_frame_start in the new thread */ yading@10: h->linesize = h1->linesize; yading@10: h->uvlinesize = h1->uvlinesize; yading@10: yading@10: /* copy block_offset since frame_start may not be called */ yading@10: memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset)); yading@10: yading@10: if (!inited) { yading@10: for (i = 0; i < MAX_SPS_COUNT; i++) yading@10: av_freep(h->sps_buffers + i); yading@10: yading@10: for (i = 0; i < MAX_PPS_COUNT; i++) yading@10: av_freep(h->pps_buffers + i); yading@10: yading@10: memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr)); yading@10: memcpy(&h->cabac, &h1->cabac, yading@10: sizeof(H264Context) - offsetof(H264Context, cabac)); yading@10: av_assert0((void*)&h->cabac == &h->mb_padding + 1); yading@10: yading@10: memset(h->sps_buffers, 0, sizeof(h->sps_buffers)); yading@10: memset(h->pps_buffers, 0, sizeof(h->pps_buffers)); yading@10: yading@10: memset(&h->er, 0, sizeof(h->er)); yading@10: memset(&h->me, 0, sizeof(h->me)); yading@10: memset(&h->mb, 0, sizeof(h->mb)); yading@10: memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc)); yading@10: memset(&h->mb_padding, 0, sizeof(h->mb_padding)); yading@10: yading@10: h->avctx = dst; yading@10: h->DPB = NULL; yading@10: h->qscale_table_pool = NULL; yading@10: h->mb_type_pool = NULL; yading@10: h->ref_index_pool = NULL; yading@10: h->motion_val_pool = NULL; yading@10: yading@10: if (h1->context_initialized) { yading@10: h->context_initialized = 0; yading@10: yading@10: memset(&h->cur_pic, 0, sizeof(h->cur_pic)); yading@10: avcodec_get_frame_defaults(&h->cur_pic.f); yading@10: h->cur_pic.tf.f = &h->cur_pic.f; yading@10: yading@10: if (ff_h264_alloc_tables(h) < 0) { yading@10: av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n"); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: context_init(h); yading@10: } yading@10: yading@10: for (i = 0; i < 2; i++) { yading@10: h->rbsp_buffer[i] = NULL; yading@10: h->rbsp_buffer_size[i] = 0; yading@10: } yading@10: h->bipred_scratchpad = NULL; yading@10: h->edge_emu_buffer = NULL; yading@10: yading@10: h->thread_context[0] = h; yading@10: h->context_initialized = h1->context_initialized; yading@10: } yading@10: yading@10: h->avctx->coded_height = h1->avctx->coded_height; yading@10: h->avctx->coded_width = h1->avctx->coded_width; yading@10: h->avctx->width = h1->avctx->width; yading@10: h->avctx->height = h1->avctx->height; yading@10: h->coded_picture_number = h1->coded_picture_number; yading@10: h->first_field = h1->first_field; yading@10: h->picture_structure = h1->picture_structure; yading@10: h->qscale = h1->qscale; yading@10: h->droppable = h1->droppable; yading@10: h->data_partitioning = h1->data_partitioning; yading@10: h->low_delay = h1->low_delay; yading@10: yading@10: for (i = 0; h->DPB && i < MAX_PICTURE_COUNT; i++) { yading@10: unref_picture(h, &h->DPB[i]); yading@10: if (h1->DPB[i].f.data[0] && yading@10: (ret = ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0) yading@10: return ret; yading@10: } yading@10: yading@10: h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1); yading@10: unref_picture(h, &h->cur_pic); yading@10: if (h1->cur_pic.f.buf[0] && (ret = ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0) yading@10: return ret; yading@10: yading@10: h->workaround_bugs = h1->workaround_bugs; yading@10: h->low_delay = h1->low_delay; yading@10: h->droppable = h1->droppable; yading@10: yading@10: // extradata/NAL handling yading@10: h->is_avc = h1->is_avc; yading@10: yading@10: // SPS/PPS yading@10: copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers, yading@10: MAX_SPS_COUNT, sizeof(SPS)); yading@10: h->sps = h1->sps; yading@10: copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers, yading@10: MAX_PPS_COUNT, sizeof(PPS)); yading@10: h->pps = h1->pps; yading@10: yading@10: // Dequantization matrices yading@10: // FIXME these are big - can they be only copied when PPS changes? yading@10: copy_fields(h, h1, dequant4_buffer, dequant4_coeff); yading@10: yading@10: for (i = 0; i < 6; i++) yading@10: h->dequant4_coeff[i] = h->dequant4_buffer[0] + yading@10: (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]); yading@10: yading@10: for (i = 0; i < 6; i++) yading@10: h->dequant8_coeff[i] = h->dequant8_buffer[0] + yading@10: (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]); yading@10: yading@10: h->dequant_coeff_pps = h1->dequant_coeff_pps; yading@10: yading@10: // POC timing yading@10: copy_fields(h, h1, poc_lsb, redundant_pic_count); yading@10: yading@10: // reference lists yading@10: copy_fields(h, h1, short_ref, cabac_init_idc); yading@10: yading@10: copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1); yading@10: copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1); yading@10: copy_picture_range(h->delayed_pic, h1->delayed_pic, yading@10: MAX_DELAYED_PIC_COUNT + 2, h, h1); yading@10: yading@10: h->sync = h1->sync; yading@10: yading@10: if (context_reinitialized) yading@10: h264_set_parameter_from_sps(h); yading@10: yading@10: if (!h->cur_pic_ptr) yading@10: return 0; yading@10: yading@10: if (!h->droppable) { yading@10: err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index); yading@10: h->prev_poc_msb = h->poc_msb; yading@10: h->prev_poc_lsb = h->poc_lsb; yading@10: } yading@10: h->prev_frame_num_offset = h->frame_num_offset; yading@10: h->prev_frame_num = h->frame_num; yading@10: h->outputed_poc = h->next_outputed_poc; yading@10: yading@10: return err; yading@10: } yading@10: yading@10: static int h264_frame_start(H264Context *h) yading@10: { yading@10: Picture *pic; yading@10: int i, ret; yading@10: const int pixel_shift = h->pixel_shift; yading@10: int c[4] = { yading@10: 1<<(h->sps.bit_depth_luma-1), yading@10: 1<<(h->sps.bit_depth_chroma-1), yading@10: 1<<(h->sps.bit_depth_chroma-1), yading@10: -1 yading@10: }; yading@10: yading@10: if (!ff_thread_can_start_frame(h->avctx)) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n"); yading@10: return -1; yading@10: } yading@10: yading@10: release_unused_pictures(h, 1); yading@10: h->cur_pic_ptr = NULL; yading@10: yading@10: i = find_unused_picture(h); yading@10: if (i < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n"); yading@10: return i; yading@10: } yading@10: pic = &h->DPB[i]; yading@10: yading@10: pic->reference = h->droppable ? 0 : h->picture_structure; yading@10: pic->f.coded_picture_number = h->coded_picture_number++; yading@10: pic->field_picture = h->picture_structure != PICT_FRAME; yading@10: yading@10: /* yading@10: * Zero key_frame here; IDR markings per slice in frame or fields are ORed yading@10: * in later. yading@10: * See decode_nal_units(). yading@10: */ yading@10: pic->f.key_frame = 0; yading@10: pic->sync = 0; yading@10: pic->mmco_reset = 0; yading@10: yading@10: if ((ret = alloc_picture(h, pic)) < 0) yading@10: return ret; yading@10: if(!h->sync && !h->avctx->hwaccel && yading@10: !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)) yading@10: avpriv_color_frame(&pic->f, c); yading@10: yading@10: h->cur_pic_ptr = pic; yading@10: unref_picture(h, &h->cur_pic); yading@10: if ((ret = ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0) yading@10: return ret; yading@10: yading@10: if (CONFIG_ERROR_RESILIENCE) { yading@10: ff_er_frame_start(&h->er); yading@10: h->er.last_pic = yading@10: h->er.next_pic = NULL; yading@10: } yading@10: yading@10: assert(h->linesize && h->uvlinesize); yading@10: yading@10: for (i = 0; i < 16; i++) { yading@10: h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3); yading@10: h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3); yading@10: } yading@10: for (i = 0; i < 16; i++) { yading@10: h->block_offset[16 + i] = yading@10: h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); yading@10: h->block_offset[48 + 16 + i] = yading@10: h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3); yading@10: } yading@10: yading@10: // s->decode = (h->flags & CODEC_FLAG_PSNR) || !s->encoding || yading@10: // h->cur_pic.reference /* || h->contains_intra */ || 1; yading@10: yading@10: /* We mark the current picture as non-reference after allocating it, so yading@10: * that if we break out due to an error it can be released automatically yading@10: * in the next ff_MPV_frame_start(). yading@10: */ yading@10: h->cur_pic_ptr->reference = 0; yading@10: yading@10: h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX; yading@10: yading@10: h->next_output_pic = NULL; yading@10: yading@10: assert(h->cur_pic_ptr->long_ref == 0); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Run setup operations that must be run after slice header decoding. yading@10: * This includes finding the next displayed frame. yading@10: * yading@10: * @param h h264 master context yading@10: * @param setup_finished enough NALs have been read that we can call yading@10: * ff_thread_finish_setup() yading@10: */ yading@10: static void decode_postinit(H264Context *h, int setup_finished) yading@10: { yading@10: Picture *out = h->cur_pic_ptr; yading@10: Picture *cur = h->cur_pic_ptr; yading@10: int i, pics, out_of_order, out_idx; yading@10: yading@10: h->cur_pic_ptr->f.pict_type = h->pict_type; yading@10: yading@10: if (h->next_output_pic) yading@10: return; yading@10: yading@10: if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) { yading@10: /* FIXME: if we have two PAFF fields in one packet, we can't start yading@10: * the next thread here. If we have one field per packet, we can. yading@10: * The check in decode_nal_units() is not good enough to find this yading@10: * yet, so we assume the worst for now. */ yading@10: // if (setup_finished) yading@10: // ff_thread_finish_setup(h->avctx); yading@10: return; yading@10: } yading@10: yading@10: cur->f.interlaced_frame = 0; yading@10: cur->f.repeat_pict = 0; yading@10: yading@10: /* Signal interlacing information externally. */ yading@10: /* Prioritize picture timing SEI information over used yading@10: * decoding process if it exists. */ yading@10: yading@10: if (h->sps.pic_struct_present_flag) { yading@10: switch (h->sei_pic_struct) { yading@10: case SEI_PIC_STRUCT_FRAME: yading@10: break; yading@10: case SEI_PIC_STRUCT_TOP_FIELD: yading@10: case SEI_PIC_STRUCT_BOTTOM_FIELD: yading@10: cur->f.interlaced_frame = 1; yading@10: break; yading@10: case SEI_PIC_STRUCT_TOP_BOTTOM: yading@10: case SEI_PIC_STRUCT_BOTTOM_TOP: yading@10: if (FIELD_OR_MBAFF_PICTURE(h)) yading@10: cur->f.interlaced_frame = 1; yading@10: else yading@10: // try to flag soft telecine progressive yading@10: cur->f.interlaced_frame = h->prev_interlaced_frame; yading@10: break; yading@10: case SEI_PIC_STRUCT_TOP_BOTTOM_TOP: yading@10: case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM: yading@10: /* Signal the possibility of telecined film externally yading@10: * (pic_struct 5,6). From these hints, let the applications yading@10: * decide if they apply deinterlacing. */ yading@10: cur->f.repeat_pict = 1; yading@10: break; yading@10: case SEI_PIC_STRUCT_FRAME_DOUBLING: yading@10: cur->f.repeat_pict = 2; yading@10: break; yading@10: case SEI_PIC_STRUCT_FRAME_TRIPLING: yading@10: cur->f.repeat_pict = 4; yading@10: break; yading@10: } yading@10: yading@10: if ((h->sei_ct_type & 3) && yading@10: h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP) yading@10: cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0; yading@10: } else { yading@10: /* Derive interlacing flag from used decoding process. */ yading@10: cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE(h); yading@10: } yading@10: h->prev_interlaced_frame = cur->f.interlaced_frame; yading@10: yading@10: if (cur->field_poc[0] != cur->field_poc[1]) { yading@10: /* Derive top_field_first from field pocs. */ yading@10: cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1]; yading@10: } else { yading@10: if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) { yading@10: /* Use picture timing SEI information. Even if it is a yading@10: * information of a past frame, better than nothing. */ yading@10: if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM || yading@10: h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP) yading@10: cur->f.top_field_first = 1; yading@10: else yading@10: cur->f.top_field_first = 0; yading@10: } else { yading@10: /* Most likely progressive */ yading@10: cur->f.top_field_first = 0; yading@10: } yading@10: } yading@10: yading@10: cur->mmco_reset = h->mmco_reset; yading@10: h->mmco_reset = 0; yading@10: // FIXME do something with unavailable reference frames yading@10: yading@10: /* Sort B-frames into display order */ yading@10: yading@10: if (h->sps.bitstream_restriction_flag && yading@10: h->avctx->has_b_frames < h->sps.num_reorder_frames) { yading@10: h->avctx->has_b_frames = h->sps.num_reorder_frames; yading@10: h->low_delay = 0; yading@10: } yading@10: yading@10: if (h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT && yading@10: !h->sps.bitstream_restriction_flag) { yading@10: h->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1; yading@10: h->low_delay = 0; yading@10: } yading@10: yading@10: for (i = 0; 1; i++) { yading@10: if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){ yading@10: if(i) yading@10: h->last_pocs[i-1] = cur->poc; yading@10: break; yading@10: } else if(i) { yading@10: h->last_pocs[i-1]= h->last_pocs[i]; yading@10: } yading@10: } yading@10: out_of_order = MAX_DELAYED_PIC_COUNT - i; yading@10: if( cur->f.pict_type == AV_PICTURE_TYPE_B yading@10: || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2)) yading@10: out_of_order = FFMAX(out_of_order, 1); yading@10: if (out_of_order == MAX_DELAYED_PIC_COUNT) { yading@10: av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]); yading@10: for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++) yading@10: h->last_pocs[i] = INT_MIN; yading@10: h->last_pocs[0] = cur->poc; yading@10: cur->mmco_reset = 1; yading@10: } else if(h->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){ yading@10: av_log(h->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order); yading@10: h->avctx->has_b_frames = out_of_order; yading@10: h->low_delay = 0; yading@10: } yading@10: yading@10: pics = 0; yading@10: while (h->delayed_pic[pics]) yading@10: pics++; yading@10: yading@10: av_assert0(pics <= MAX_DELAYED_PIC_COUNT); yading@10: yading@10: h->delayed_pic[pics++] = cur; yading@10: if (cur->reference == 0) yading@10: cur->reference = DELAYED_PIC_REF; yading@10: yading@10: out = h->delayed_pic[0]; yading@10: out_idx = 0; yading@10: for (i = 1; h->delayed_pic[i] && yading@10: !h->delayed_pic[i]->f.key_frame && yading@10: !h->delayed_pic[i]->mmco_reset; yading@10: i++) yading@10: if (h->delayed_pic[i]->poc < out->poc) { yading@10: out = h->delayed_pic[i]; yading@10: out_idx = i; yading@10: } yading@10: if (h->avctx->has_b_frames == 0 && yading@10: (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) yading@10: h->next_outputed_poc = INT_MIN; yading@10: out_of_order = out->poc < h->next_outputed_poc; yading@10: yading@10: if (out_of_order || pics > h->avctx->has_b_frames) { yading@10: out->reference &= ~DELAYED_PIC_REF; yading@10: // for frame threading, the owner must be the second field's thread or yading@10: // else the first thread can release the picture and reuse it unsafely yading@10: for (i = out_idx; h->delayed_pic[i]; i++) yading@10: h->delayed_pic[i] = h->delayed_pic[i + 1]; yading@10: } yading@10: if (!out_of_order && pics > h->avctx->has_b_frames) { yading@10: h->next_output_pic = out; yading@10: if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) { yading@10: h->next_outputed_poc = INT_MIN; yading@10: } else yading@10: h->next_outputed_poc = out->poc; yading@10: } else { yading@10: av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : ""); yading@10: } yading@10: yading@10: if (h->next_output_pic && h->next_output_pic->sync) { yading@10: h->sync |= 2; yading@10: } yading@10: yading@10: if (setup_finished && !h->avctx->hwaccel) yading@10: ff_thread_finish_setup(h->avctx); yading@10: } yading@10: yading@10: static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, yading@10: uint8_t *src_cb, uint8_t *src_cr, yading@10: int linesize, int uvlinesize, yading@10: int simple) yading@10: { yading@10: uint8_t *top_border; yading@10: int top_idx = 1; yading@10: const int pixel_shift = h->pixel_shift; yading@10: int chroma444 = CHROMA444(h); yading@10: int chroma422 = CHROMA422(h); yading@10: yading@10: src_y -= linesize; yading@10: src_cb -= uvlinesize; yading@10: src_cr -= uvlinesize; yading@10: yading@10: if (!simple && FRAME_MBAFF(h)) { yading@10: if (h->mb_y & 1) { yading@10: if (!MB_MBAFF(h)) { yading@10: top_border = h->top_borders[0][h->mb_x]; yading@10: AV_COPY128(top_border, src_y + 15 * linesize); yading@10: if (pixel_shift) yading@10: AV_COPY128(top_border + 16, src_y + 15 * linesize + 16); yading@10: if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) { yading@10: if (chroma444) { yading@10: if (pixel_shift) { yading@10: AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize); yading@10: AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16); yading@10: AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize); yading@10: AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16); yading@10: } else { yading@10: AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize); yading@10: AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize); yading@10: } yading@10: } else if (chroma422) { yading@10: if (pixel_shift) { yading@10: AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize); yading@10: AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize); yading@10: } else { yading@10: AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize); yading@10: AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize); yading@10: } yading@10: } else { yading@10: if (pixel_shift) { yading@10: AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize); yading@10: AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize); yading@10: } else { yading@10: AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize); yading@10: AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } else if (MB_MBAFF(h)) { yading@10: top_idx = 0; yading@10: } else yading@10: return; yading@10: } yading@10: yading@10: top_border = h->top_borders[top_idx][h->mb_x]; yading@10: /* There are two lines saved, the line above the top macroblock yading@10: * of a pair, and the line above the bottom macroblock. */ yading@10: AV_COPY128(top_border, src_y + 16 * linesize); yading@10: if (pixel_shift) yading@10: AV_COPY128(top_border + 16, src_y + 16 * linesize + 16); yading@10: yading@10: if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) { yading@10: if (chroma444) { yading@10: if (pixel_shift) { yading@10: AV_COPY128(top_border + 32, src_cb + 16 * linesize); yading@10: AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16); yading@10: AV_COPY128(top_border + 64, src_cr + 16 * linesize); yading@10: AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16); yading@10: } else { yading@10: AV_COPY128(top_border + 16, src_cb + 16 * linesize); yading@10: AV_COPY128(top_border + 32, src_cr + 16 * linesize); yading@10: } yading@10: } else if (chroma422) { yading@10: if (pixel_shift) { yading@10: AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize); yading@10: AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize); yading@10: } else { yading@10: AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize); yading@10: AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize); yading@10: } yading@10: } else { yading@10: if (pixel_shift) { yading@10: AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize); yading@10: AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize); yading@10: } else { yading@10: AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize); yading@10: AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y, yading@10: uint8_t *src_cb, uint8_t *src_cr, yading@10: int linesize, int uvlinesize, yading@10: int xchg, int chroma444, yading@10: int simple, int pixel_shift) yading@10: { yading@10: int deblock_topleft; yading@10: int deblock_top; yading@10: int top_idx = 1; yading@10: uint8_t *top_border_m1; yading@10: uint8_t *top_border; yading@10: yading@10: if (!simple && FRAME_MBAFF(h)) { yading@10: if (h->mb_y & 1) { yading@10: if (!MB_MBAFF(h)) yading@10: return; yading@10: } else { yading@10: top_idx = MB_MBAFF(h) ? 0 : 1; yading@10: } yading@10: } yading@10: yading@10: if (h->deblocking_filter == 2) { yading@10: deblock_topleft = h->slice_table[h->mb_xy - 1 - h->mb_stride] == h->slice_num; yading@10: deblock_top = h->top_type; yading@10: } else { yading@10: deblock_topleft = (h->mb_x > 0); yading@10: deblock_top = (h->mb_y > !!MB_FIELD(h)); yading@10: } yading@10: yading@10: src_y -= linesize + 1 + pixel_shift; yading@10: src_cb -= uvlinesize + 1 + pixel_shift; yading@10: src_cr -= uvlinesize + 1 + pixel_shift; yading@10: yading@10: top_border_m1 = h->top_borders[top_idx][h->mb_x - 1]; yading@10: top_border = h->top_borders[top_idx][h->mb_x]; yading@10: yading@10: #define XCHG(a, b, xchg) \ yading@10: if (pixel_shift) { \ yading@10: if (xchg) { \ yading@10: AV_SWAP64(b + 0, a + 0); \ yading@10: AV_SWAP64(b + 8, a + 8); \ yading@10: } else { \ yading@10: AV_COPY128(b, a); \ yading@10: } \ yading@10: } else if (xchg) \ yading@10: AV_SWAP64(b, a); \ yading@10: else \ yading@10: AV_COPY64(b, a); yading@10: yading@10: if (deblock_top) { yading@10: if (deblock_topleft) { yading@10: XCHG(top_border_m1 + (8 << pixel_shift), yading@10: src_y - (7 << pixel_shift), 1); yading@10: } yading@10: XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg); yading@10: XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1); yading@10: if (h->mb_x + 1 < h->mb_width) { yading@10: XCHG(h->top_borders[top_idx][h->mb_x + 1], yading@10: src_y + (17 << pixel_shift), 1); yading@10: } yading@10: } yading@10: if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) { yading@10: if (chroma444) { yading@10: if (deblock_topleft) { yading@10: XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1); yading@10: XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1); yading@10: } yading@10: XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg); yading@10: XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1); yading@10: XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg); yading@10: XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1); yading@10: if (h->mb_x + 1 < h->mb_width) { yading@10: XCHG(h->top_borders[top_idx][h->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1); yading@10: XCHG(h->top_borders[top_idx][h->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1); yading@10: } yading@10: } else { yading@10: if (deblock_top) { yading@10: if (deblock_topleft) { yading@10: XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1); yading@10: XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1); yading@10: } yading@10: XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1); yading@10: XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: static av_always_inline int dctcoef_get(int16_t *mb, int high_bit_depth, yading@10: int index) yading@10: { yading@10: if (high_bit_depth) { yading@10: return AV_RN32A(((int32_t *)mb) + index); yading@10: } else yading@10: return AV_RN16A(mb + index); yading@10: } yading@10: yading@10: static av_always_inline void dctcoef_set(int16_t *mb, int high_bit_depth, yading@10: int index, int value) yading@10: { yading@10: if (high_bit_depth) { yading@10: AV_WN32A(((int32_t *)mb) + index, value); yading@10: } else yading@10: AV_WN16A(mb + index, value); yading@10: } yading@10: yading@10: static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, yading@10: int mb_type, int is_h264, yading@10: int simple, yading@10: int transform_bypass, yading@10: int pixel_shift, yading@10: int *block_offset, yading@10: int linesize, yading@10: uint8_t *dest_y, int p) yading@10: { yading@10: void (*idct_add)(uint8_t *dst, int16_t *block, int stride); yading@10: void (*idct_dc_add)(uint8_t *dst, int16_t *block, int stride); yading@10: int i; yading@10: int qscale = p == 0 ? h->qscale : h->chroma_qp[p - 1]; yading@10: block_offset += 16 * p; yading@10: if (IS_INTRA4x4(mb_type)) { yading@10: if (IS_8x8DCT(mb_type)) { yading@10: if (transform_bypass) { yading@10: idct_dc_add = yading@10: idct_add = h->h264dsp.h264_add_pixels8_clear; yading@10: } else { yading@10: idct_dc_add = h->h264dsp.h264_idct8_dc_add; yading@10: idct_add = h->h264dsp.h264_idct8_add; yading@10: } yading@10: for (i = 0; i < 16; i += 4) { yading@10: uint8_t *const ptr = dest_y + block_offset[i]; yading@10: const int dir = h->intra4x4_pred_mode_cache[scan8[i]]; yading@10: if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) { yading@10: h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize); yading@10: } else { yading@10: const int nnz = h->non_zero_count_cache[scan8[i + p * 16]]; yading@10: h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000, yading@10: (h->topright_samples_available << i) & 0x4000, linesize); yading@10: if (nnz) { yading@10: if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256)) yading@10: idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize); yading@10: else yading@10: idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize); yading@10: } yading@10: } yading@10: } yading@10: } else { yading@10: if (transform_bypass) { yading@10: idct_dc_add = yading@10: idct_add = h->h264dsp.h264_add_pixels4_clear; yading@10: } else { yading@10: idct_dc_add = h->h264dsp.h264_idct_dc_add; yading@10: idct_add = h->h264dsp.h264_idct_add; yading@10: } yading@10: for (i = 0; i < 16; i++) { yading@10: uint8_t *const ptr = dest_y + block_offset[i]; yading@10: const int dir = h->intra4x4_pred_mode_cache[scan8[i]]; yading@10: yading@10: if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) { yading@10: h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize); yading@10: } else { yading@10: uint8_t *topright; yading@10: int nnz, tr; yading@10: uint64_t tr_high; yading@10: if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) { yading@10: const int topright_avail = (h->topright_samples_available << i) & 0x8000; yading@10: av_assert2(h->mb_y || linesize <= block_offset[i]); yading@10: if (!topright_avail) { yading@10: if (pixel_shift) { yading@10: tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL; yading@10: topright = (uint8_t *)&tr_high; yading@10: } else { yading@10: tr = ptr[3 - linesize] * 0x01010101u; yading@10: topright = (uint8_t *)&tr; yading@10: } yading@10: } else yading@10: topright = ptr + (4 << pixel_shift) - linesize; yading@10: } else yading@10: topright = NULL; yading@10: yading@10: h->hpc.pred4x4[dir](ptr, topright, linesize); yading@10: nnz = h->non_zero_count_cache[scan8[i + p * 16]]; yading@10: if (nnz) { yading@10: if (is_h264) { yading@10: if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256)) yading@10: idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize); yading@10: else yading@10: idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize); yading@10: } else if (CONFIG_SVQ3_DECODER) yading@10: ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } else { yading@10: h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize); yading@10: if (is_h264) { yading@10: if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) { yading@10: if (!transform_bypass) yading@10: h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift), yading@10: h->mb_luma_dc[p], yading@10: h->dequant4_coeff[p][qscale][0]); yading@10: else { yading@10: static const uint8_t dc_mapping[16] = { yading@10: 0 * 16, 1 * 16, 4 * 16, 5 * 16, yading@10: 2 * 16, 3 * 16, 6 * 16, 7 * 16, yading@10: 8 * 16, 9 * 16, 12 * 16, 13 * 16, yading@10: 10 * 16, 11 * 16, 14 * 16, 15 * 16 }; yading@10: for (i = 0; i < 16; i++) yading@10: dctcoef_set(h->mb + (p * 256 << pixel_shift), yading@10: pixel_shift, dc_mapping[i], yading@10: dctcoef_get(h->mb_luma_dc[p], yading@10: pixel_shift, i)); yading@10: } yading@10: } yading@10: } else if (CONFIG_SVQ3_DECODER) yading@10: ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256, yading@10: h->mb_luma_dc[p], qscale); yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, yading@10: int is_h264, int simple, yading@10: int transform_bypass, yading@10: int pixel_shift, yading@10: int *block_offset, yading@10: int linesize, yading@10: uint8_t *dest_y, int p) yading@10: { yading@10: void (*idct_add)(uint8_t *dst, int16_t *block, int stride); yading@10: int i; yading@10: block_offset += 16 * p; yading@10: if (!IS_INTRA4x4(mb_type)) { yading@10: if (is_h264) { yading@10: if (IS_INTRA16x16(mb_type)) { yading@10: if (transform_bypass) { yading@10: if (h->sps.profile_idc == 244 && yading@10: (h->intra16x16_pred_mode == VERT_PRED8x8 || yading@10: h->intra16x16_pred_mode == HOR_PRED8x8)) { yading@10: h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, yading@10: h->mb + (p * 256 << pixel_shift), yading@10: linesize); yading@10: } else { yading@10: for (i = 0; i < 16; i++) yading@10: if (h->non_zero_count_cache[scan8[i + p * 16]] || yading@10: dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256)) yading@10: h->h264dsp.h264_add_pixels4_clear(dest_y + block_offset[i], yading@10: h->mb + (i * 16 + p * 256 << pixel_shift), yading@10: linesize); yading@10: } yading@10: } else { yading@10: h->h264dsp.h264_idct_add16intra(dest_y, block_offset, yading@10: h->mb + (p * 256 << pixel_shift), yading@10: linesize, yading@10: h->non_zero_count_cache + p * 5 * 8); yading@10: } yading@10: } else if (h->cbp & 15) { yading@10: if (transform_bypass) { yading@10: const int di = IS_8x8DCT(mb_type) ? 4 : 1; yading@10: idct_add = IS_8x8DCT(mb_type) ? h->h264dsp.h264_add_pixels8_clear yading@10: : h->h264dsp.h264_add_pixels4_clear; yading@10: for (i = 0; i < 16; i += di) yading@10: if (h->non_zero_count_cache[scan8[i + p * 16]]) yading@10: idct_add(dest_y + block_offset[i], yading@10: h->mb + (i * 16 + p * 256 << pixel_shift), yading@10: linesize); yading@10: } else { yading@10: if (IS_8x8DCT(mb_type)) yading@10: h->h264dsp.h264_idct8_add4(dest_y, block_offset, yading@10: h->mb + (p * 256 << pixel_shift), yading@10: linesize, yading@10: h->non_zero_count_cache + p * 5 * 8); yading@10: else yading@10: h->h264dsp.h264_idct_add16(dest_y, block_offset, yading@10: h->mb + (p * 256 << pixel_shift), yading@10: linesize, yading@10: h->non_zero_count_cache + p * 5 * 8); yading@10: } yading@10: } yading@10: } else if (CONFIG_SVQ3_DECODER) { yading@10: for (i = 0; i < 16; i++) yading@10: if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) { yading@10: // FIXME benchmark weird rule, & below yading@10: uint8_t *const ptr = dest_y + block_offset[i]; yading@10: ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, yading@10: h->qscale, IS_INTRA(mb_type) ? 1 : 0); yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: #define BITS 8 yading@10: #define SIMPLE 1 yading@10: #include "h264_mb_template.c" yading@10: yading@10: #undef BITS yading@10: #define BITS 16 yading@10: #include "h264_mb_template.c" yading@10: yading@10: #undef SIMPLE yading@10: #define SIMPLE 0 yading@10: #include "h264_mb_template.c" yading@10: yading@10: void ff_h264_hl_decode_mb(H264Context *h) yading@10: { yading@10: const int mb_xy = h->mb_xy; yading@10: const int mb_type = h->cur_pic.mb_type[mb_xy]; yading@10: int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || h->qscale == 0; yading@10: yading@10: if (CHROMA444(h)) { yading@10: if (is_complex || h->pixel_shift) yading@10: hl_decode_mb_444_complex(h); yading@10: else yading@10: hl_decode_mb_444_simple_8(h); yading@10: } else if (is_complex) { yading@10: hl_decode_mb_complex(h); yading@10: } else if (h->pixel_shift) { yading@10: hl_decode_mb_simple_16(h); yading@10: } else yading@10: hl_decode_mb_simple_8(h); yading@10: } yading@10: yading@10: static int pred_weight_table(H264Context *h) yading@10: { yading@10: int list, i; yading@10: int luma_def, chroma_def; yading@10: yading@10: h->use_weight = 0; yading@10: h->use_weight_chroma = 0; yading@10: h->luma_log2_weight_denom = get_ue_golomb(&h->gb); yading@10: if (h->sps.chroma_format_idc) yading@10: h->chroma_log2_weight_denom = get_ue_golomb(&h->gb); yading@10: luma_def = 1 << h->luma_log2_weight_denom; yading@10: chroma_def = 1 << h->chroma_log2_weight_denom; yading@10: yading@10: for (list = 0; list < 2; list++) { yading@10: h->luma_weight_flag[list] = 0; yading@10: h->chroma_weight_flag[list] = 0; yading@10: for (i = 0; i < h->ref_count[list]; i++) { yading@10: int luma_weight_flag, chroma_weight_flag; yading@10: yading@10: luma_weight_flag = get_bits1(&h->gb); yading@10: if (luma_weight_flag) { yading@10: h->luma_weight[i][list][0] = get_se_golomb(&h->gb); yading@10: h->luma_weight[i][list][1] = get_se_golomb(&h->gb); yading@10: if (h->luma_weight[i][list][0] != luma_def || yading@10: h->luma_weight[i][list][1] != 0) { yading@10: h->use_weight = 1; yading@10: h->luma_weight_flag[list] = 1; yading@10: } yading@10: } else { yading@10: h->luma_weight[i][list][0] = luma_def; yading@10: h->luma_weight[i][list][1] = 0; yading@10: } yading@10: yading@10: if (h->sps.chroma_format_idc) { yading@10: chroma_weight_flag = get_bits1(&h->gb); yading@10: if (chroma_weight_flag) { yading@10: int j; yading@10: for (j = 0; j < 2; j++) { yading@10: h->chroma_weight[i][list][j][0] = get_se_golomb(&h->gb); yading@10: h->chroma_weight[i][list][j][1] = get_se_golomb(&h->gb); yading@10: if (h->chroma_weight[i][list][j][0] != chroma_def || yading@10: h->chroma_weight[i][list][j][1] != 0) { yading@10: h->use_weight_chroma = 1; yading@10: h->chroma_weight_flag[list] = 1; yading@10: } yading@10: } yading@10: } else { yading@10: int j; yading@10: for (j = 0; j < 2; j++) { yading@10: h->chroma_weight[i][list][j][0] = chroma_def; yading@10: h->chroma_weight[i][list][j][1] = 0; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: if (h->slice_type_nos != AV_PICTURE_TYPE_B) yading@10: break; yading@10: } yading@10: h->use_weight = h->use_weight || h->use_weight_chroma; yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Initialize implicit_weight table. yading@10: * @param field 0/1 initialize the weight for interlaced MBAFF yading@10: * -1 initializes the rest yading@10: */ yading@10: static void implicit_weight_table(H264Context *h, int field) yading@10: { yading@10: int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1; yading@10: yading@10: for (i = 0; i < 2; i++) { yading@10: h->luma_weight_flag[i] = 0; yading@10: h->chroma_weight_flag[i] = 0; yading@10: } yading@10: yading@10: if (field < 0) { yading@10: if (h->picture_structure == PICT_FRAME) { yading@10: cur_poc = h->cur_pic_ptr->poc; yading@10: } else { yading@10: cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1]; yading@10: } yading@10: if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) && yading@10: h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) { yading@10: h->use_weight = 0; yading@10: h->use_weight_chroma = 0; yading@10: return; yading@10: } yading@10: ref_start = 0; yading@10: ref_count0 = h->ref_count[0]; yading@10: ref_count1 = h->ref_count[1]; yading@10: } else { yading@10: cur_poc = h->cur_pic_ptr->field_poc[field]; yading@10: ref_start = 16; yading@10: ref_count0 = 16 + 2 * h->ref_count[0]; yading@10: ref_count1 = 16 + 2 * h->ref_count[1]; yading@10: } yading@10: yading@10: h->use_weight = 2; yading@10: h->use_weight_chroma = 2; yading@10: h->luma_log2_weight_denom = 5; yading@10: h->chroma_log2_weight_denom = 5; yading@10: yading@10: for (ref0 = ref_start; ref0 < ref_count0; ref0++) { yading@10: int poc0 = h->ref_list[0][ref0].poc; yading@10: for (ref1 = ref_start; ref1 < ref_count1; ref1++) { yading@10: int w = 32; yading@10: if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) { yading@10: int poc1 = h->ref_list[1][ref1].poc; yading@10: int td = av_clip(poc1 - poc0, -128, 127); yading@10: if (td) { yading@10: int tb = av_clip(cur_poc - poc0, -128, 127); yading@10: int tx = (16384 + (FFABS(td) >> 1)) / td; yading@10: int dist_scale_factor = (tb * tx + 32) >> 8; yading@10: if (dist_scale_factor >= -64 && dist_scale_factor <= 128) yading@10: w = 64 - dist_scale_factor; yading@10: } yading@10: } yading@10: if (field < 0) { yading@10: h->implicit_weight[ref0][ref1][0] = yading@10: h->implicit_weight[ref0][ref1][1] = w; yading@10: } else { yading@10: h->implicit_weight[ref0][ref1][field] = w; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * instantaneous decoder refresh. yading@10: */ yading@10: static void idr(H264Context *h) yading@10: { yading@10: int i; yading@10: ff_h264_remove_all_refs(h); yading@10: h->prev_frame_num = 0; yading@10: h->prev_frame_num_offset = 0; yading@10: h->prev_poc_msb = 1<<16; yading@10: h->prev_poc_lsb = 0; yading@10: for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) yading@10: h->last_pocs[i] = INT_MIN; yading@10: } yading@10: yading@10: /* forget old pics after a seek */ yading@10: static void flush_change(H264Context *h) yading@10: { yading@10: int i, j; yading@10: yading@10: h->outputed_poc = h->next_outputed_poc = INT_MIN; yading@10: h->prev_interlaced_frame = 1; yading@10: idr(h); yading@10: yading@10: h->prev_frame_num = -1; yading@10: if (h->cur_pic_ptr) { yading@10: h->cur_pic_ptr->reference = 0; yading@10: for (j=i=0; h->delayed_pic[i]; i++) yading@10: if (h->delayed_pic[i] != h->cur_pic_ptr) yading@10: h->delayed_pic[j++] = h->delayed_pic[i]; yading@10: h->delayed_pic[j] = NULL; yading@10: } yading@10: h->first_field = 0; yading@10: memset(h->ref_list[0], 0, sizeof(h->ref_list[0])); yading@10: memset(h->ref_list[1], 0, sizeof(h->ref_list[1])); yading@10: memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0])); yading@10: memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1])); yading@10: ff_h264_reset_sei(h); yading@10: h->recovery_frame= -1; yading@10: h->sync= 0; yading@10: h->list_count = 0; yading@10: h->current_slice = 0; yading@10: } yading@10: yading@10: /* forget old pics after a seek */ yading@10: static void flush_dpb(AVCodecContext *avctx) yading@10: { yading@10: H264Context *h = avctx->priv_data; yading@10: int i; yading@10: yading@10: for (i = 0; i <= MAX_DELAYED_PIC_COUNT; i++) { yading@10: if (h->delayed_pic[i]) yading@10: h->delayed_pic[i]->reference = 0; yading@10: h->delayed_pic[i] = NULL; yading@10: } yading@10: yading@10: flush_change(h); yading@10: yading@10: if (h->DPB) yading@10: for (i = 0; i < MAX_PICTURE_COUNT; i++) yading@10: unref_picture(h, &h->DPB[i]); yading@10: h->cur_pic_ptr = NULL; yading@10: unref_picture(h, &h->cur_pic); yading@10: yading@10: h->mb_x = h->mb_y = 0; yading@10: yading@10: h->parse_context.state = -1; yading@10: h->parse_context.frame_start_found = 0; yading@10: h->parse_context.overread = 0; yading@10: h->parse_context.overread_index = 0; yading@10: h->parse_context.index = 0; yading@10: h->parse_context.last_index = 0; yading@10: } yading@10: yading@10: static int init_poc(H264Context *h) yading@10: { yading@10: const int max_frame_num = 1 << h->sps.log2_max_frame_num; yading@10: int field_poc[2]; yading@10: Picture *cur = h->cur_pic_ptr; yading@10: yading@10: h->frame_num_offset = h->prev_frame_num_offset; yading@10: if (h->frame_num < h->prev_frame_num) yading@10: h->frame_num_offset += max_frame_num; yading@10: yading@10: if (h->sps.poc_type == 0) { yading@10: const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb; yading@10: yading@10: if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2) yading@10: h->poc_msb = h->prev_poc_msb + max_poc_lsb; yading@10: else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2) yading@10: h->poc_msb = h->prev_poc_msb - max_poc_lsb; yading@10: else yading@10: h->poc_msb = h->prev_poc_msb; yading@10: field_poc[0] = yading@10: field_poc[1] = h->poc_msb + h->poc_lsb; yading@10: if (h->picture_structure == PICT_FRAME) yading@10: field_poc[1] += h->delta_poc_bottom; yading@10: } else if (h->sps.poc_type == 1) { yading@10: int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc; yading@10: int i; yading@10: yading@10: if (h->sps.poc_cycle_length != 0) yading@10: abs_frame_num = h->frame_num_offset + h->frame_num; yading@10: else yading@10: abs_frame_num = 0; yading@10: yading@10: if (h->nal_ref_idc == 0 && abs_frame_num > 0) yading@10: abs_frame_num--; yading@10: yading@10: expected_delta_per_poc_cycle = 0; yading@10: for (i = 0; i < h->sps.poc_cycle_length; i++) yading@10: // FIXME integrate during sps parse yading@10: expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i]; yading@10: yading@10: if (abs_frame_num > 0) { yading@10: int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length; yading@10: int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length; yading@10: yading@10: expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle; yading@10: for (i = 0; i <= frame_num_in_poc_cycle; i++) yading@10: expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i]; yading@10: } else yading@10: expectedpoc = 0; yading@10: yading@10: if (h->nal_ref_idc == 0) yading@10: expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic; yading@10: yading@10: field_poc[0] = expectedpoc + h->delta_poc[0]; yading@10: field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field; yading@10: yading@10: if (h->picture_structure == PICT_FRAME) yading@10: field_poc[1] += h->delta_poc[1]; yading@10: } else { yading@10: int poc = 2 * (h->frame_num_offset + h->frame_num); yading@10: yading@10: if (!h->nal_ref_idc) yading@10: poc--; yading@10: yading@10: field_poc[0] = poc; yading@10: field_poc[1] = poc; yading@10: } yading@10: yading@10: if (h->picture_structure != PICT_BOTTOM_FIELD) yading@10: h->cur_pic_ptr->field_poc[0] = field_poc[0]; yading@10: if (h->picture_structure != PICT_TOP_FIELD) yading@10: h->cur_pic_ptr->field_poc[1] = field_poc[1]; yading@10: cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * initialize scan tables yading@10: */ yading@10: static void init_scan_tables(H264Context *h) yading@10: { yading@10: int i; yading@10: for (i = 0; i < 16; i++) { yading@10: #define T(x) (x >> 2) | ((x << 2) & 0xF) yading@10: h->zigzag_scan[i] = T(zigzag_scan[i]); yading@10: h->field_scan[i] = T(field_scan[i]); yading@10: #undef T yading@10: } yading@10: for (i = 0; i < 64; i++) { yading@10: #define T(x) (x >> 3) | ((x & 7) << 3) yading@10: h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]); yading@10: h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]); yading@10: h->field_scan8x8[i] = T(field_scan8x8[i]); yading@10: h->field_scan8x8_cavlc[i] = T(field_scan8x8_cavlc[i]); yading@10: #undef T yading@10: } yading@10: if (h->sps.transform_bypass) { // FIXME same ugly yading@10: memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 )); yading@10: memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 )); yading@10: memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0)); yading@10: memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 )); yading@10: memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 )); yading@10: memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 )); yading@10: } else { yading@10: memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 )); yading@10: memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 )); yading@10: memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0)); yading@10: memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 )); yading@10: memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 )); yading@10: memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 )); yading@10: } yading@10: } yading@10: yading@10: static int field_end(H264Context *h, int in_setup) yading@10: { yading@10: AVCodecContext *const avctx = h->avctx; yading@10: int err = 0; yading@10: h->mb_y = 0; yading@10: yading@10: if (CONFIG_H264_VDPAU_DECODER && yading@10: h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) yading@10: ff_vdpau_h264_set_reference_frames(h); yading@10: yading@10: if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) { yading@10: if (!h->droppable) { yading@10: err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index); yading@10: h->prev_poc_msb = h->poc_msb; yading@10: h->prev_poc_lsb = h->poc_lsb; yading@10: } yading@10: h->prev_frame_num_offset = h->frame_num_offset; yading@10: h->prev_frame_num = h->frame_num; yading@10: h->outputed_poc = h->next_outputed_poc; yading@10: } yading@10: yading@10: if (avctx->hwaccel) { yading@10: if (avctx->hwaccel->end_frame(avctx) < 0) yading@10: av_log(avctx, AV_LOG_ERROR, yading@10: "hardware accelerator failed to decode picture\n"); yading@10: } yading@10: yading@10: if (CONFIG_H264_VDPAU_DECODER && yading@10: h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) yading@10: ff_vdpau_h264_picture_complete(h); yading@10: yading@10: /* yading@10: * FIXME: Error handling code does not seem to support interlaced yading@10: * when slices span multiple rows yading@10: * The ff_er_add_slice calls don't work right for bottom yading@10: * fields; they cause massive erroneous error concealing yading@10: * Error marking covers both fields (top and bottom). yading@10: * This causes a mismatched s->error_count yading@10: * and a bad error table. Further, the error count goes to yading@10: * INT_MAX when called for bottom field, because mb_y is yading@10: * past end by one (callers fault) and resync_mb_y != 0 yading@10: * causes problems for the first MB line, too. yading@10: */ yading@10: if (CONFIG_ERROR_RESILIENCE && yading@10: !FIELD_PICTURE(h) && h->current_slice && !h->sps.new) { yading@10: h->er.cur_pic = h->cur_pic_ptr; yading@10: ff_er_frame_end(&h->er); yading@10: } yading@10: if (!in_setup && !h->droppable) yading@10: ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, yading@10: h->picture_structure == PICT_BOTTOM_FIELD); yading@10: emms_c(); yading@10: yading@10: h->current_slice = 0; yading@10: yading@10: return err; yading@10: } yading@10: yading@10: /** yading@10: * Replicate H264 "master" context to thread contexts. yading@10: */ yading@10: static int clone_slice(H264Context *dst, H264Context *src) yading@10: { yading@10: memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset)); yading@10: dst->cur_pic_ptr = src->cur_pic_ptr; yading@10: dst->cur_pic = src->cur_pic; yading@10: dst->linesize = src->linesize; yading@10: dst->uvlinesize = src->uvlinesize; yading@10: dst->first_field = src->first_field; yading@10: yading@10: dst->prev_poc_msb = src->prev_poc_msb; yading@10: dst->prev_poc_lsb = src->prev_poc_lsb; yading@10: dst->prev_frame_num_offset = src->prev_frame_num_offset; yading@10: dst->prev_frame_num = src->prev_frame_num; yading@10: dst->short_ref_count = src->short_ref_count; yading@10: yading@10: memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref)); yading@10: memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref)); yading@10: memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list)); yading@10: yading@10: memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff)); yading@10: memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff)); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Compute profile from profile_idc and constraint_set?_flags. yading@10: * yading@10: * @param sps SPS yading@10: * yading@10: * @return profile as defined by FF_PROFILE_H264_* yading@10: */ yading@10: int ff_h264_get_profile(SPS *sps) yading@10: { yading@10: int profile = sps->profile_idc; yading@10: yading@10: switch (sps->profile_idc) { yading@10: case FF_PROFILE_H264_BASELINE: yading@10: // constraint_set1_flag set to 1 yading@10: profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0; yading@10: break; yading@10: case FF_PROFILE_H264_HIGH_10: yading@10: case FF_PROFILE_H264_HIGH_422: yading@10: case FF_PROFILE_H264_HIGH_444_PREDICTIVE: yading@10: // constraint_set3_flag set to 1 yading@10: profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0; yading@10: break; yading@10: } yading@10: yading@10: return profile; yading@10: } yading@10: yading@10: static int h264_set_parameter_from_sps(H264Context *h) yading@10: { yading@10: if (h->flags & CODEC_FLAG_LOW_DELAY || yading@10: (h->sps.bitstream_restriction_flag && yading@10: !h->sps.num_reorder_frames)) { yading@10: if (h->avctx->has_b_frames > 1 || h->delayed_pic[0]) yading@10: av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. " yading@10: "Reenabling low delay requires a codec flush.\n"); yading@10: else yading@10: h->low_delay = 1; yading@10: } yading@10: yading@10: if (h->avctx->has_b_frames < 2) yading@10: h->avctx->has_b_frames = !h->low_delay; yading@10: yading@10: if (h->sps.bit_depth_luma != h->sps.bit_depth_chroma) { yading@10: avpriv_request_sample(h->avctx, yading@10: "Different chroma and luma bit depth"); yading@10: return AVERROR_PATCHWELCOME; yading@10: } yading@10: yading@10: if (h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma || yading@10: h->cur_chroma_format_idc != h->sps.chroma_format_idc) { yading@10: if (h->avctx->codec && yading@10: h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU && yading@10: (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "VDPAU decoding does not support video colorspace.\n"); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 && yading@10: h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13) { yading@10: h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma; yading@10: h->cur_chroma_format_idc = h->sps.chroma_format_idc; yading@10: h->pixel_shift = h->sps.bit_depth_luma > 8; yading@10: yading@10: ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, yading@10: h->sps.chroma_format_idc); yading@10: ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma); yading@10: ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma); yading@10: ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma, yading@10: h->sps.chroma_format_idc); yading@10: yading@10: if (CONFIG_ERROR_RESILIENCE) yading@10: ff_dsputil_init(&h->dsp, h->avctx); yading@10: ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma); yading@10: } else { yading@10: av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n", yading@10: h->sps.bit_depth_luma); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback) yading@10: { yading@10: switch (h->sps.bit_depth_luma) { yading@10: case 9: yading@10: if (CHROMA444(h)) { yading@10: if (h->avctx->colorspace == AVCOL_SPC_RGB) { yading@10: return AV_PIX_FMT_GBRP9; yading@10: } else yading@10: return AV_PIX_FMT_YUV444P9; yading@10: } else if (CHROMA422(h)) yading@10: return AV_PIX_FMT_YUV422P9; yading@10: else yading@10: return AV_PIX_FMT_YUV420P9; yading@10: break; yading@10: case 10: yading@10: if (CHROMA444(h)) { yading@10: if (h->avctx->colorspace == AVCOL_SPC_RGB) { yading@10: return AV_PIX_FMT_GBRP10; yading@10: } else yading@10: return AV_PIX_FMT_YUV444P10; yading@10: } else if (CHROMA422(h)) yading@10: return AV_PIX_FMT_YUV422P10; yading@10: else yading@10: return AV_PIX_FMT_YUV420P10; yading@10: break; yading@10: case 12: yading@10: if (CHROMA444(h)) { yading@10: if (h->avctx->colorspace == AVCOL_SPC_RGB) { yading@10: return AV_PIX_FMT_GBRP12; yading@10: } else yading@10: return AV_PIX_FMT_YUV444P12; yading@10: } else if (CHROMA422(h)) yading@10: return AV_PIX_FMT_YUV422P12; yading@10: else yading@10: return AV_PIX_FMT_YUV420P12; yading@10: break; yading@10: case 14: yading@10: if (CHROMA444(h)) { yading@10: if (h->avctx->colorspace == AVCOL_SPC_RGB) { yading@10: return AV_PIX_FMT_GBRP14; yading@10: } else yading@10: return AV_PIX_FMT_YUV444P14; yading@10: } else if (CHROMA422(h)) yading@10: return AV_PIX_FMT_YUV422P14; yading@10: else yading@10: return AV_PIX_FMT_YUV420P14; yading@10: break; yading@10: case 8: yading@10: if (CHROMA444(h)) { yading@10: if (h->avctx->colorspace == AVCOL_SPC_RGB) { yading@10: av_log(h->avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n"); yading@10: return AV_PIX_FMT_GBR24P; yading@10: } else if (h->avctx->colorspace == AVCOL_SPC_YCGCO) { yading@10: av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n"); yading@10: } yading@10: return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P yading@10: : AV_PIX_FMT_YUV444P; yading@10: } else if (CHROMA422(h)) { yading@10: return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P yading@10: : AV_PIX_FMT_YUV422P; yading@10: } else { yading@10: int i; yading@10: const enum AVPixelFormat * fmt = h->avctx->codec->pix_fmts ? yading@10: h->avctx->codec->pix_fmts : yading@10: h->avctx->color_range == AVCOL_RANGE_JPEG ? yading@10: h264_hwaccel_pixfmt_list_jpeg_420 : yading@10: h264_hwaccel_pixfmt_list_420; yading@10: yading@10: for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++) yading@10: if (fmt[i] == h->avctx->pix_fmt && !force_callback) yading@10: return fmt[i]; yading@10: return ff_thread_get_format(h->avctx, fmt); yading@10: } yading@10: break; yading@10: default: yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "Unsupported bit depth: %d\n", h->sps.bit_depth_luma); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: } yading@10: yading@10: /* export coded and cropped frame dimensions to AVCodecContext */ yading@10: static int init_dimensions(H264Context *h) yading@10: { yading@10: int width = h->width - (h->sps.crop_right + h->sps.crop_left); yading@10: int height = h->height - (h->sps.crop_top + h->sps.crop_bottom); yading@10: yading@10: /* handle container cropping */ yading@10: if (!h->sps.crop && yading@10: FFALIGN(h->avctx->width, 16) == h->width && yading@10: FFALIGN(h->avctx->height, 16) == h->height) { yading@10: width = h->avctx->width; yading@10: height = h->avctx->height; yading@10: } yading@10: yading@10: if (width <= 0 || height <= 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n", yading@10: width, height); yading@10: if (h->avctx->err_recognition & AV_EF_EXPLODE) yading@10: return AVERROR_INVALIDDATA; yading@10: yading@10: av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n"); yading@10: h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0; yading@10: h->sps.crop = 0; yading@10: yading@10: width = h->width; yading@10: height = h->height; yading@10: } yading@10: yading@10: h->avctx->coded_width = h->width; yading@10: h->avctx->coded_height = h->height; yading@10: h->avctx->width = width; yading@10: h->avctx->height = height; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int h264_slice_header_init(H264Context *h, int reinit) yading@10: { yading@10: int nb_slices = (HAVE_THREADS && yading@10: h->avctx->active_thread_type & FF_THREAD_SLICE) ? yading@10: h->avctx->thread_count : 1; yading@10: int i; yading@10: yading@10: h->avctx->sample_aspect_ratio = h->sps.sar; yading@10: av_assert0(h->avctx->sample_aspect_ratio.den); yading@10: av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt, yading@10: &h->chroma_x_shift, &h->chroma_y_shift); yading@10: yading@10: if (h->sps.timing_info_present_flag) { yading@10: int64_t den = h->sps.time_scale; yading@10: if (h->x264_build < 44U) yading@10: den *= 2; yading@10: av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den, yading@10: h->sps.num_units_in_tick, den, 1 << 30); yading@10: } yading@10: yading@10: h->avctx->hwaccel = ff_find_hwaccel(h->avctx->codec->id, h->avctx->pix_fmt); yading@10: yading@10: if (reinit) yading@10: free_tables(h, 0); yading@10: h->first_field = 0; yading@10: h->prev_interlaced_frame = 1; yading@10: yading@10: init_scan_tables(h); yading@10: if (ff_h264_alloc_tables(h) < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "Could not allocate memory for h264\n"); yading@10: return AVERROR(ENOMEM); yading@10: } yading@10: yading@10: if (nb_slices > MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) { yading@10: int max_slices; yading@10: if (h->mb_height) yading@10: max_slices = FFMIN(MAX_THREADS, h->mb_height); yading@10: else yading@10: max_slices = MAX_THREADS; yading@10: av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices (%d)," yading@10: " reducing to %d\n", nb_slices, max_slices); yading@10: nb_slices = max_slices; yading@10: } yading@10: h->slice_context_count = nb_slices; yading@10: yading@10: if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) { yading@10: if (context_init(h) < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n"); yading@10: return -1; yading@10: } yading@10: } else { yading@10: for (i = 1; i < h->slice_context_count; i++) { yading@10: H264Context *c; yading@10: c = h->thread_context[i] = av_mallocz(sizeof(H264Context)); yading@10: c->avctx = h->avctx; yading@10: if (CONFIG_ERROR_RESILIENCE) { yading@10: c->dsp = h->dsp; yading@10: } yading@10: c->vdsp = h->vdsp; yading@10: c->h264dsp = h->h264dsp; yading@10: c->h264qpel = h->h264qpel; yading@10: c->h264chroma = h->h264chroma; yading@10: c->sps = h->sps; yading@10: c->pps = h->pps; yading@10: c->pixel_shift = h->pixel_shift; yading@10: c->cur_chroma_format_idc = h->cur_chroma_format_idc; yading@10: c->width = h->width; yading@10: c->height = h->height; yading@10: c->linesize = h->linesize; yading@10: c->uvlinesize = h->uvlinesize; yading@10: c->chroma_x_shift = h->chroma_x_shift; yading@10: c->chroma_y_shift = h->chroma_y_shift; yading@10: c->qscale = h->qscale; yading@10: c->droppable = h->droppable; yading@10: c->data_partitioning = h->data_partitioning; yading@10: c->low_delay = h->low_delay; yading@10: c->mb_width = h->mb_width; yading@10: c->mb_height = h->mb_height; yading@10: c->mb_stride = h->mb_stride; yading@10: c->mb_num = h->mb_num; yading@10: c->flags = h->flags; yading@10: c->workaround_bugs = h->workaround_bugs; yading@10: c->pict_type = h->pict_type; yading@10: yading@10: init_scan_tables(c); yading@10: clone_tables(c, h, i); yading@10: c->context_initialized = 1; yading@10: } yading@10: yading@10: for (i = 0; i < h->slice_context_count; i++) yading@10: if (context_init(h->thread_context[i]) < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n"); yading@10: return -1; yading@10: } yading@10: } yading@10: yading@10: h->context_initialized = 1; yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: /** yading@10: * Decode a slice header. yading@10: * This will also call ff_MPV_common_init() and frame_start() as needed. yading@10: * yading@10: * @param h h264context yading@10: * @param h0 h264 master context (differs from 'h' when doing sliced based yading@10: * parallel decoding) yading@10: * yading@10: * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded yading@10: */ yading@10: static int decode_slice_header(H264Context *h, H264Context *h0) yading@10: { yading@10: unsigned int first_mb_in_slice; yading@10: unsigned int pps_id; yading@10: int num_ref_idx_active_override_flag, ret; yading@10: unsigned int slice_type, tmp, i, j; yading@10: int last_pic_structure, last_pic_droppable; yading@10: int must_reinit; yading@10: int needs_reinit = 0; yading@10: yading@10: h->me.qpel_put = h->h264qpel.put_h264_qpel_pixels_tab; yading@10: h->me.qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab; yading@10: yading@10: first_mb_in_slice = get_ue_golomb_long(&h->gb); yading@10: yading@10: if (first_mb_in_slice == 0) { // FIXME better field boundary detection yading@10: if (h0->current_slice && FIELD_PICTURE(h)) { yading@10: field_end(h, 1); yading@10: } yading@10: yading@10: h0->current_slice = 0; yading@10: if (!h0->first_field) { yading@10: if (h->cur_pic_ptr && !h->droppable) { yading@10: ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, yading@10: h->picture_structure == PICT_BOTTOM_FIELD); yading@10: } yading@10: h->cur_pic_ptr = NULL; yading@10: } yading@10: } yading@10: yading@10: slice_type = get_ue_golomb_31(&h->gb); yading@10: if (slice_type > 9) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "slice type too large (%d) at %d %d\n", yading@10: slice_type, h->mb_x, h->mb_y); yading@10: return -1; yading@10: } yading@10: if (slice_type > 4) { yading@10: slice_type -= 5; yading@10: h->slice_type_fixed = 1; yading@10: } else yading@10: h->slice_type_fixed = 0; yading@10: yading@10: slice_type = golomb_to_pict_type[slice_type]; yading@10: h->slice_type = slice_type; yading@10: h->slice_type_nos = slice_type & 3; yading@10: yading@10: // to make a few old functions happy, it's wrong though yading@10: h->pict_type = h->slice_type; yading@10: yading@10: pps_id = get_ue_golomb(&h->gb); yading@10: if (pps_id >= MAX_PPS_COUNT) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id); yading@10: return -1; yading@10: } yading@10: if (!h0->pps_buffers[pps_id]) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "non-existing PPS %u referenced\n", yading@10: pps_id); yading@10: return -1; yading@10: } yading@10: h->pps = *h0->pps_buffers[pps_id]; yading@10: yading@10: if (!h0->sps_buffers[h->pps.sps_id]) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "non-existing SPS %u referenced\n", yading@10: h->pps.sps_id); yading@10: return -1; yading@10: } yading@10: yading@10: if (h->pps.sps_id != h->current_sps_id || yading@10: h0->sps_buffers[h->pps.sps_id]->new) { yading@10: h0->sps_buffers[h->pps.sps_id]->new = 0; yading@10: yading@10: h->current_sps_id = h->pps.sps_id; yading@10: h->sps = *h0->sps_buffers[h->pps.sps_id]; yading@10: yading@10: if (h->mb_width != h->sps.mb_width || yading@10: h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) || yading@10: h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma || yading@10: h->cur_chroma_format_idc != h->sps.chroma_format_idc yading@10: ) yading@10: needs_reinit = 1; yading@10: yading@10: if (h->bit_depth_luma != h->sps.bit_depth_luma || yading@10: h->chroma_format_idc != h->sps.chroma_format_idc) { yading@10: h->bit_depth_luma = h->sps.bit_depth_luma; yading@10: h->chroma_format_idc = h->sps.chroma_format_idc; yading@10: needs_reinit = 1; yading@10: } yading@10: if ((ret = h264_set_parameter_from_sps(h)) < 0) yading@10: return ret; yading@10: } yading@10: yading@10: h->avctx->profile = ff_h264_get_profile(&h->sps); yading@10: h->avctx->level = h->sps.level_idc; yading@10: h->avctx->refs = h->sps.ref_frame_count; yading@10: yading@10: must_reinit = (h->context_initialized && yading@10: ( 16*h->sps.mb_width != h->avctx->coded_width yading@10: || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height yading@10: || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma yading@10: || h->cur_chroma_format_idc != h->sps.chroma_format_idc yading@10: || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))); yading@10: if (h0->avctx->pix_fmt != get_pixel_format(h0, 0)) yading@10: must_reinit = 1; yading@10: yading@10: h->mb_width = h->sps.mb_width; yading@10: h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag); yading@10: h->mb_num = h->mb_width * h->mb_height; yading@10: h->mb_stride = h->mb_width + 1; yading@10: yading@10: h->b_stride = h->mb_width * 4; yading@10: yading@10: h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p yading@10: yading@10: h->width = 16 * h->mb_width; yading@10: h->height = 16 * h->mb_height; yading@10: yading@10: ret = init_dimensions(h); yading@10: if (ret < 0) yading@10: return ret; yading@10: yading@10: if (h->sps.video_signal_type_present_flag) { yading@10: h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG yading@10: : AVCOL_RANGE_MPEG; yading@10: if (h->sps.colour_description_present_flag) { yading@10: if (h->avctx->colorspace != h->sps.colorspace) yading@10: needs_reinit = 1; yading@10: h->avctx->color_primaries = h->sps.color_primaries; yading@10: h->avctx->color_trc = h->sps.color_trc; yading@10: h->avctx->colorspace = h->sps.colorspace; yading@10: } yading@10: } yading@10: yading@10: if (h->context_initialized && yading@10: (h->width != h->avctx->coded_width || yading@10: h->height != h->avctx->coded_height || yading@10: must_reinit || yading@10: needs_reinit)) { yading@10: yading@10: if (h != h0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "changing width/height on " yading@10: "slice %d\n", h0->current_slice + 1); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: flush_change(h); yading@10: yading@10: if ((ret = get_pixel_format(h, 1)) < 0) yading@10: return ret; yading@10: h->avctx->pix_fmt = ret; yading@10: yading@10: av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, " yading@10: "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt); yading@10: yading@10: if ((ret = h264_slice_header_init(h, 1)) < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "h264_slice_header_init() failed\n"); yading@10: return ret; yading@10: } yading@10: } yading@10: if (!h->context_initialized) { yading@10: if (h != h0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "Cannot (re-)initialize context during parallel decoding.\n"); yading@10: return -1; yading@10: } yading@10: yading@10: if ((ret = get_pixel_format(h, 1)) < 0) yading@10: return ret; yading@10: h->avctx->pix_fmt = ret; yading@10: yading@10: if ((ret = h264_slice_header_init(h, 0)) < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "h264_slice_header_init() failed\n"); yading@10: return ret; yading@10: } yading@10: } yading@10: yading@10: if (h == h0 && h->dequant_coeff_pps != pps_id) { yading@10: h->dequant_coeff_pps = pps_id; yading@10: init_dequant_tables(h); yading@10: } yading@10: yading@10: h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num); yading@10: yading@10: h->mb_mbaff = 0; yading@10: h->mb_aff_frame = 0; yading@10: last_pic_structure = h0->picture_structure; yading@10: last_pic_droppable = h0->droppable; yading@10: h->droppable = h->nal_ref_idc == 0; yading@10: if (h->sps.frame_mbs_only_flag) { yading@10: h->picture_structure = PICT_FRAME; yading@10: } else { yading@10: if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n"); yading@10: return -1; yading@10: } yading@10: if (get_bits1(&h->gb)) { // field_pic_flag yading@10: h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag yading@10: } else { yading@10: h->picture_structure = PICT_FRAME; yading@10: h->mb_aff_frame = h->sps.mb_aff; yading@10: } yading@10: } yading@10: h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME; yading@10: yading@10: if (h0->current_slice != 0) { yading@10: if (last_pic_structure != h->picture_structure || yading@10: last_pic_droppable != h->droppable) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "Changing field mode (%d -> %d) between slices is not allowed\n", yading@10: last_pic_structure, h->picture_structure); yading@10: h->picture_structure = last_pic_structure; yading@10: h->droppable = last_pic_droppable; yading@10: return AVERROR_INVALIDDATA; yading@10: } else if (!h0->cur_pic_ptr) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "unset cur_pic_ptr on %d. slice\n", yading@10: h0->current_slice + 1); yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: } else { yading@10: /* Shorten frame num gaps so we don't have to allocate reference yading@10: * frames just to throw them away */ yading@10: if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) { yading@10: int unwrap_prev_frame_num = h->prev_frame_num; yading@10: int max_frame_num = 1 << h->sps.log2_max_frame_num; yading@10: yading@10: if (unwrap_prev_frame_num > h->frame_num) yading@10: unwrap_prev_frame_num -= max_frame_num; yading@10: yading@10: if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) { yading@10: unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1; yading@10: if (unwrap_prev_frame_num < 0) yading@10: unwrap_prev_frame_num += max_frame_num; yading@10: yading@10: h->prev_frame_num = unwrap_prev_frame_num; yading@10: } yading@10: } yading@10: yading@10: /* See if we have a decoded first field looking for a pair... yading@10: * Here, we're using that to see if we should mark previously yading@10: * decode frames as "finished". yading@10: * We have to do that before the "dummy" in-between frame allocation, yading@10: * since that can modify h->cur_pic_ptr. */ yading@10: if (h0->first_field) { yading@10: assert(h0->cur_pic_ptr); yading@10: assert(h0->cur_pic_ptr->f.data[0]); yading@10: assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF); yading@10: yading@10: /* Mark old field/frame as completed */ yading@10: if (!last_pic_droppable && h0->cur_pic_ptr->tf.owner == h0->avctx) { yading@10: ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX, yading@10: last_pic_structure == PICT_BOTTOM_FIELD); yading@10: } yading@10: yading@10: /* figure out if we have a complementary field pair */ yading@10: if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) { yading@10: /* Previous field is unmatched. Don't display it, but let it yading@10: * remain for reference if marked as such. */ yading@10: if (!last_pic_droppable && last_pic_structure != PICT_FRAME) { yading@10: ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX, yading@10: last_pic_structure == PICT_TOP_FIELD); yading@10: } yading@10: } else { yading@10: if (h0->cur_pic_ptr->frame_num != h->frame_num) { yading@10: /* This and previous field were reference, but had yading@10: * different frame_nums. Consider this field first in yading@10: * pair. Throw away previous field except for reference yading@10: * purposes. */ yading@10: if (!last_pic_droppable && last_pic_structure != PICT_FRAME) { yading@10: ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX, yading@10: last_pic_structure == PICT_TOP_FIELD); yading@10: } yading@10: } else { yading@10: /* Second field in complementary pair */ yading@10: if (!((last_pic_structure == PICT_TOP_FIELD && yading@10: h->picture_structure == PICT_BOTTOM_FIELD) || yading@10: (last_pic_structure == PICT_BOTTOM_FIELD && yading@10: h->picture_structure == PICT_TOP_FIELD))) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "Invalid field mode combination %d/%d\n", yading@10: last_pic_structure, h->picture_structure); yading@10: h->picture_structure = last_pic_structure; yading@10: h->droppable = last_pic_droppable; yading@10: return AVERROR_INVALIDDATA; yading@10: } else if (last_pic_droppable != h->droppable) { yading@10: avpriv_request_sample(h->avctx, yading@10: "Found reference and non-reference fields in the same frame, which"); yading@10: h->picture_structure = last_pic_structure; yading@10: h->droppable = last_pic_droppable; yading@10: return AVERROR_PATCHWELCOME; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 && !h0->first_field && yading@10: h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) { yading@10: Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL; yading@10: av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", yading@10: h->frame_num, h->prev_frame_num); yading@10: if (!h->sps.gaps_in_frame_num_allowed_flag) yading@10: for(i=0; ilast_pocs); i++) yading@10: h->last_pocs[i] = INT_MIN; yading@10: if (h264_frame_start(h) < 0) yading@10: return -1; yading@10: h->prev_frame_num++; yading@10: h->prev_frame_num %= 1 << h->sps.log2_max_frame_num; yading@10: h->cur_pic_ptr->frame_num = h->prev_frame_num; yading@10: ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0); yading@10: ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1); yading@10: if ((ret = ff_generate_sliding_window_mmcos(h, 1)) < 0 && yading@10: h->avctx->err_recognition & AV_EF_EXPLODE) yading@10: return ret; yading@10: if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 && yading@10: (h->avctx->err_recognition & AV_EF_EXPLODE)) yading@10: return AVERROR_INVALIDDATA; yading@10: /* Error concealment: if a ref is missing, copy the previous ref in its place. yading@10: * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions yading@10: * about there being no actual duplicates. yading@10: * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're yading@10: * concealing a lost frame, this probably isn't noticeable by comparison, but it should yading@10: * be fixed. */ yading@10: if (h->short_ref_count) { yading@10: if (prev) { yading@10: av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize, yading@10: (const uint8_t **)prev->f.data, prev->f.linesize, yading@10: h->avctx->pix_fmt, h->mb_width * 16, h->mb_height * 16); yading@10: h->short_ref[0]->poc = prev->poc + 2; yading@10: } yading@10: h->short_ref[0]->frame_num = h->prev_frame_num; yading@10: } yading@10: } yading@10: yading@10: /* See if we have a decoded first field looking for a pair... yading@10: * We're using that to see whether to continue decoding in that yading@10: * frame, or to allocate a new one. */ yading@10: if (h0->first_field) { yading@10: assert(h0->cur_pic_ptr); yading@10: assert(h0->cur_pic_ptr->f.data[0]); yading@10: assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF); yading@10: yading@10: /* figure out if we have a complementary field pair */ yading@10: if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) { yading@10: /* Previous field is unmatched. Don't display it, but let it yading@10: * remain for reference if marked as such. */ yading@10: h0->cur_pic_ptr = NULL; yading@10: h0->first_field = FIELD_PICTURE(h); yading@10: } else { yading@10: if (h0->cur_pic_ptr->frame_num != h->frame_num) { yading@10: ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX, yading@10: h0->picture_structure==PICT_BOTTOM_FIELD); yading@10: /* This and the previous field had different frame_nums. yading@10: * Consider this field first in pair. Throw away previous yading@10: * one except for reference purposes. */ yading@10: h0->first_field = 1; yading@10: h0->cur_pic_ptr = NULL; yading@10: } else { yading@10: /* Second field in complementary pair */ yading@10: h0->first_field = 0; yading@10: } yading@10: } yading@10: } else { yading@10: /* Frame or first field in a potentially complementary pair */ yading@10: h0->first_field = FIELD_PICTURE(h); yading@10: } yading@10: yading@10: if (!FIELD_PICTURE(h) || h0->first_field) { yading@10: if (h264_frame_start(h) < 0) { yading@10: h0->first_field = 0; yading@10: return -1; yading@10: } yading@10: } else { yading@10: release_unused_pictures(h, 0); yading@10: } yading@10: /* Some macroblocks can be accessed before they're available in case yading@10: * of lost slices, MBAFF or threading. */ yading@10: if (FIELD_PICTURE(h)) { yading@10: for(i = (h->picture_structure == PICT_BOTTOM_FIELD); imb_height; i++) yading@10: memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table)); yading@10: } else { yading@10: memset(h->slice_table, -1, yading@10: (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table)); yading@10: } yading@10: h0->last_slice_type = -1; yading@10: } yading@10: if (h != h0 && (ret = clone_slice(h, h0)) < 0) yading@10: return ret; yading@10: yading@10: /* can't be in alloc_tables because linesize isn't known there. yading@10: * FIXME: redo bipred weight to not require extra buffer? */ yading@10: for (i = 0; i < h->slice_context_count; i++) yading@10: if (h->thread_context[i]) { yading@10: ret = alloc_scratch_buffers(h->thread_context[i], h->linesize); yading@10: if (ret < 0) yading@10: return ret; yading@10: } yading@10: yading@10: h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup yading@10: yading@10: av_assert1(h->mb_num == h->mb_width * h->mb_height); yading@10: if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num || yading@10: first_mb_in_slice >= h->mb_num) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n"); yading@10: return -1; yading@10: } yading@10: h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width; yading@10: h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) << FIELD_OR_MBAFF_PICTURE(h); yading@10: if (h->picture_structure == PICT_BOTTOM_FIELD) yading@10: h->resync_mb_y = h->mb_y = h->mb_y + 1; yading@10: av_assert1(h->mb_y < h->mb_height); yading@10: yading@10: if (h->picture_structure == PICT_FRAME) { yading@10: h->curr_pic_num = h->frame_num; yading@10: h->max_pic_num = 1 << h->sps.log2_max_frame_num; yading@10: } else { yading@10: h->curr_pic_num = 2 * h->frame_num + 1; yading@10: h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1); yading@10: } yading@10: yading@10: if (h->nal_unit_type == NAL_IDR_SLICE) yading@10: get_ue_golomb(&h->gb); /* idr_pic_id */ yading@10: yading@10: if (h->sps.poc_type == 0) { yading@10: h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb); yading@10: yading@10: if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) yading@10: h->delta_poc_bottom = get_se_golomb(&h->gb); yading@10: } yading@10: yading@10: if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) { yading@10: h->delta_poc[0] = get_se_golomb(&h->gb); yading@10: yading@10: if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) yading@10: h->delta_poc[1] = get_se_golomb(&h->gb); yading@10: } yading@10: yading@10: init_poc(h); yading@10: yading@10: if (h->pps.redundant_pic_cnt_present) yading@10: h->redundant_pic_count = get_ue_golomb(&h->gb); yading@10: yading@10: // set defaults, might be overridden a few lines later yading@10: h->ref_count[0] = h->pps.ref_count[0]; yading@10: h->ref_count[1] = h->pps.ref_count[1]; yading@10: yading@10: if (h->slice_type_nos != AV_PICTURE_TYPE_I) { yading@10: unsigned max[2]; yading@10: max[0] = max[1] = h->picture_structure == PICT_FRAME ? 15 : 31; yading@10: yading@10: if (h->slice_type_nos == AV_PICTURE_TYPE_B) yading@10: h->direct_spatial_mv_pred = get_bits1(&h->gb); yading@10: num_ref_idx_active_override_flag = get_bits1(&h->gb); yading@10: yading@10: if (num_ref_idx_active_override_flag) { yading@10: h->ref_count[0] = get_ue_golomb(&h->gb) + 1; yading@10: if (h->slice_type_nos == AV_PICTURE_TYPE_B) { yading@10: h->ref_count[1] = get_ue_golomb(&h->gb) + 1; yading@10: } else yading@10: // full range is spec-ok in this case, even for frames yading@10: h->ref_count[1] = 1; yading@10: } yading@10: yading@10: if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){ yading@10: av_log(h->avctx, AV_LOG_ERROR, "reference overflow %u > %u or %u > %u\n", h->ref_count[0]-1, max[0], h->ref_count[1]-1, max[1]); yading@10: h->ref_count[0] = h->ref_count[1] = 0; yading@10: return AVERROR_INVALIDDATA; yading@10: } yading@10: yading@10: if (h->slice_type_nos == AV_PICTURE_TYPE_B) yading@10: h->list_count = 2; yading@10: else yading@10: h->list_count = 1; yading@10: } else { yading@10: h->list_count = 0; yading@10: h->ref_count[0] = h->ref_count[1] = 0; yading@10: } yading@10: if (slice_type != AV_PICTURE_TYPE_I && yading@10: (h0->current_slice == 0 || yading@10: slice_type != h0->last_slice_type || yading@10: memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) { yading@10: ff_h264_fill_default_ref_list(h); yading@10: } yading@10: yading@10: if (h->slice_type_nos != AV_PICTURE_TYPE_I && yading@10: ff_h264_decode_ref_pic_list_reordering(h) < 0) { yading@10: h->ref_count[1] = h->ref_count[0] = 0; yading@10: return -1; yading@10: } yading@10: yading@10: if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) || yading@10: (h->pps.weighted_bipred_idc == 1 && yading@10: h->slice_type_nos == AV_PICTURE_TYPE_B)) yading@10: pred_weight_table(h); yading@10: else if (h->pps.weighted_bipred_idc == 2 && yading@10: h->slice_type_nos == AV_PICTURE_TYPE_B) { yading@10: implicit_weight_table(h, -1); yading@10: } else { yading@10: h->use_weight = 0; yading@10: for (i = 0; i < 2; i++) { yading@10: h->luma_weight_flag[i] = 0; yading@10: h->chroma_weight_flag[i] = 0; yading@10: } yading@10: } yading@10: yading@10: // If frame-mt is enabled, only update mmco tables for the first slice yading@10: // in a field. Subsequent slices can temporarily clobber h->mmco_index yading@10: // or h->mmco, which will cause ref list mix-ups and decoding errors yading@10: // further down the line. This may break decoding if the first slice is yading@10: // corrupt, thus we only do this if frame-mt is enabled. yading@10: if (h->nal_ref_idc && yading@10: ff_h264_decode_ref_pic_marking(h0, &h->gb, yading@10: !(h->avctx->active_thread_type & FF_THREAD_FRAME) || yading@10: h0->current_slice == 0) < 0 && yading@10: (h->avctx->err_recognition & AV_EF_EXPLODE)) yading@10: return AVERROR_INVALIDDATA; yading@10: yading@10: if (FRAME_MBAFF(h)) { yading@10: ff_h264_fill_mbaff_ref_list(h); yading@10: yading@10: if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) { yading@10: implicit_weight_table(h, 0); yading@10: implicit_weight_table(h, 1); yading@10: } yading@10: } yading@10: yading@10: if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred) yading@10: ff_h264_direct_dist_scale_factor(h); yading@10: ff_h264_direct_ref_list_init(h); yading@10: yading@10: if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) { yading@10: tmp = get_ue_golomb_31(&h->gb); yading@10: if (tmp > 2) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n"); yading@10: return -1; yading@10: } yading@10: h->cabac_init_idc = tmp; yading@10: } yading@10: yading@10: h->last_qscale_diff = 0; yading@10: tmp = h->pps.init_qp + get_se_golomb(&h->gb); yading@10: if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) { yading@10: av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp); yading@10: return -1; yading@10: } yading@10: h->qscale = tmp; yading@10: h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale); yading@10: h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale); yading@10: // FIXME qscale / qp ... stuff yading@10: if (h->slice_type == AV_PICTURE_TYPE_SP) yading@10: get_bits1(&h->gb); /* sp_for_switch_flag */ yading@10: if (h->slice_type == AV_PICTURE_TYPE_SP || yading@10: h->slice_type == AV_PICTURE_TYPE_SI) yading@10: get_se_golomb(&h->gb); /* slice_qs_delta */ yading@10: yading@10: h->deblocking_filter = 1; yading@10: h->slice_alpha_c0_offset = 52; yading@10: h->slice_beta_offset = 52; yading@10: if (h->pps.deblocking_filter_parameters_present) { yading@10: tmp = get_ue_golomb_31(&h->gb); yading@10: if (tmp > 2) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "deblocking_filter_idc %u out of range\n", tmp); yading@10: return -1; yading@10: } yading@10: h->deblocking_filter = tmp; yading@10: if (h->deblocking_filter < 2) yading@10: h->deblocking_filter ^= 1; // 1<->0 yading@10: yading@10: if (h->deblocking_filter) { yading@10: h->slice_alpha_c0_offset += get_se_golomb(&h->gb) << 1; yading@10: h->slice_beta_offset += get_se_golomb(&h->gb) << 1; yading@10: if (h->slice_alpha_c0_offset > 104U || yading@10: h->slice_beta_offset > 104U) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "deblocking filter parameters %d %d out of range\n", yading@10: h->slice_alpha_c0_offset, h->slice_beta_offset); yading@10: return -1; yading@10: } yading@10: } yading@10: } yading@10: yading@10: if (h->avctx->skip_loop_filter >= AVDISCARD_ALL || yading@10: (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY && yading@10: h->slice_type_nos != AV_PICTURE_TYPE_I) || yading@10: (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR && yading@10: h->slice_type_nos == AV_PICTURE_TYPE_B) || yading@10: (h->avctx->skip_loop_filter >= AVDISCARD_NONREF && yading@10: h->nal_ref_idc == 0)) yading@10: h->deblocking_filter = 0; yading@10: yading@10: if (h->deblocking_filter == 1 && h0->max_contexts > 1) { yading@10: if (h->avctx->flags2 & CODEC_FLAG2_FAST) { yading@10: /* Cheat slightly for speed: yading@10: * Do not bother to deblock across slices. */ yading@10: h->deblocking_filter = 2; yading@10: } else { yading@10: h0->max_contexts = 1; yading@10: if (!h0->single_decode_warning) { yading@10: av_log(h->avctx, AV_LOG_INFO, yading@10: "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n"); yading@10: h0->single_decode_warning = 1; yading@10: } yading@10: if (h != h0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "Deblocking switched inside frame.\n"); yading@10: return 1; yading@10: } yading@10: } yading@10: } yading@10: h->qp_thresh = 15 + 52 - yading@10: FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) - yading@10: FFMAX3(0, yading@10: h->pps.chroma_qp_index_offset[0], yading@10: h->pps.chroma_qp_index_offset[1]) + yading@10: 6 * (h->sps.bit_depth_luma - 8); yading@10: yading@10: h0->last_slice_type = slice_type; yading@10: memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count)); yading@10: h->slice_num = ++h0->current_slice; yading@10: yading@10: if (h->slice_num) yading@10: h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y; yading@10: if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y yading@10: && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y yading@10: && h->slice_num >= MAX_SLICES) { yading@10: //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case yading@10: av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES); yading@10: } yading@10: yading@10: for (j = 0; j < 2; j++) { yading@10: int id_list[16]; yading@10: int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j]; yading@10: for (i = 0; i < 16; i++) { yading@10: id_list[i] = 60; yading@10: if (j < h->list_count && i < h->ref_count[j] && h->ref_list[j][i].f.buf[0]) { yading@10: int k; yading@10: AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer; yading@10: for (k = 0; k < h->short_ref_count; k++) yading@10: if (h->short_ref[k]->f.buf[0]->buffer == buf) { yading@10: id_list[i] = k; yading@10: break; yading@10: } yading@10: for (k = 0; k < h->long_ref_count; k++) yading@10: if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) { yading@10: id_list[i] = h->short_ref_count + k; yading@10: break; yading@10: } yading@10: } yading@10: } yading@10: yading@10: ref2frm[0] = yading@10: ref2frm[1] = -1; yading@10: for (i = 0; i < 16; i++) yading@10: ref2frm[i + 2] = 4 * id_list[i] + yading@10: (h->ref_list[j][i].reference & 3); yading@10: ref2frm[18 + 0] = yading@10: ref2frm[18 + 1] = -1; yading@10: for (i = 16; i < 48; i++) yading@10: ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] + yading@10: (h->ref_list[j][i].reference & 3); yading@10: } yading@10: yading@10: if (h->ref_count[0]) h->er.last_pic = &h->ref_list[0][0]; yading@10: if (h->ref_count[1]) h->er.next_pic = &h->ref_list[1][0]; yading@10: yading@10: if (h->avctx->debug & FF_DEBUG_PICT_INFO) { yading@10: av_log(h->avctx, AV_LOG_DEBUG, yading@10: "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", yading@10: h->slice_num, yading@10: (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"), yading@10: first_mb_in_slice, yading@10: av_get_picture_type_char(h->slice_type), yading@10: h->slice_type_fixed ? " fix" : "", yading@10: h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "", yading@10: pps_id, h->frame_num, yading@10: h->cur_pic_ptr->field_poc[0], yading@10: h->cur_pic_ptr->field_poc[1], yading@10: h->ref_count[0], h->ref_count[1], yading@10: h->qscale, yading@10: h->deblocking_filter, yading@10: h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26, yading@10: h->use_weight, yading@10: h->use_weight == 1 && h->use_weight_chroma ? "c" : "", yading@10: h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""); yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: int ff_h264_get_slice_type(const H264Context *h) yading@10: { yading@10: switch (h->slice_type) { yading@10: case AV_PICTURE_TYPE_P: yading@10: return 0; yading@10: case AV_PICTURE_TYPE_B: yading@10: return 1; yading@10: case AV_PICTURE_TYPE_I: yading@10: return 2; yading@10: case AV_PICTURE_TYPE_SP: yading@10: return 3; yading@10: case AV_PICTURE_TYPE_SI: yading@10: return 4; yading@10: default: yading@10: return -1; yading@10: } yading@10: } yading@10: yading@10: static av_always_inline void fill_filter_caches_inter(H264Context *h, yading@10: int mb_type, int top_xy, yading@10: int left_xy[LEFT_MBS], yading@10: int top_type, yading@10: int left_type[LEFT_MBS], yading@10: int mb_xy, int list) yading@10: { yading@10: int b_stride = h->b_stride; yading@10: int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]]; yading@10: int8_t *ref_cache = &h->ref_cache[list][scan8[0]]; yading@10: if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) { yading@10: if (USES_LIST(top_type, list)) { yading@10: const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride; yading@10: const int b8_xy = 4 * top_xy + 2; yading@10: int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2)); yading@10: AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]); yading@10: ref_cache[0 - 1 * 8] = yading@10: ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]]; yading@10: ref_cache[2 - 1 * 8] = yading@10: ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]]; yading@10: } else { yading@10: AV_ZERO128(mv_dst - 1 * 8); yading@10: AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u); yading@10: } yading@10: yading@10: if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) { yading@10: if (USES_LIST(left_type[LTOP], list)) { yading@10: const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3; yading@10: const int b8_xy = 4 * left_xy[LTOP] + 1; yading@10: int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2)); yading@10: AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]); yading@10: AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]); yading@10: AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]); yading@10: AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]); yading@10: ref_cache[-1 + 0] = yading@10: ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]]; yading@10: ref_cache[-1 + 16] = yading@10: ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]]; yading@10: } else { yading@10: AV_ZERO32(mv_dst - 1 + 0); yading@10: AV_ZERO32(mv_dst - 1 + 8); yading@10: AV_ZERO32(mv_dst - 1 + 16); yading@10: AV_ZERO32(mv_dst - 1 + 24); yading@10: ref_cache[-1 + 0] = yading@10: ref_cache[-1 + 8] = yading@10: ref_cache[-1 + 16] = yading@10: ref_cache[-1 + 24] = LIST_NOT_USED; yading@10: } yading@10: } yading@10: } yading@10: yading@10: if (!USES_LIST(mb_type, list)) { yading@10: fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4); yading@10: AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u); yading@10: AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u); yading@10: AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u); yading@10: AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u); yading@10: return; yading@10: } yading@10: yading@10: { yading@10: int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy]; yading@10: int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2)); yading@10: uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101; yading@10: uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101; yading@10: AV_WN32A(&ref_cache[0 * 8], ref01); yading@10: AV_WN32A(&ref_cache[1 * 8], ref01); yading@10: AV_WN32A(&ref_cache[2 * 8], ref23); yading@10: AV_WN32A(&ref_cache[3 * 8], ref23); yading@10: } yading@10: yading@10: { yading@10: int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride]; yading@10: AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride); yading@10: AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride); yading@10: AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride); yading@10: AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride); yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * yading@10: * @return non zero if the loop filter can be skipped yading@10: */ yading@10: static int fill_filter_caches(H264Context *h, int mb_type) yading@10: { yading@10: const int mb_xy = h->mb_xy; yading@10: int top_xy, left_xy[LEFT_MBS]; yading@10: int top_type, left_type[LEFT_MBS]; yading@10: uint8_t *nnz; yading@10: uint8_t *nnz_cache; yading@10: yading@10: top_xy = mb_xy - (h->mb_stride << MB_FIELD(h)); yading@10: yading@10: /* Wow, what a mess, why didn't they simplify the interlacing & intra yading@10: * stuff, I can't imagine that these complex rules are worth it. */ yading@10: yading@10: left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1; yading@10: if (FRAME_MBAFF(h)) { yading@10: const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]); yading@10: const int curr_mb_field_flag = IS_INTERLACED(mb_type); yading@10: if (h->mb_y & 1) { yading@10: if (left_mb_field_flag != curr_mb_field_flag) yading@10: left_xy[LTOP] -= h->mb_stride; yading@10: } else { yading@10: if (curr_mb_field_flag) yading@10: top_xy += h->mb_stride & yading@10: (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1); yading@10: if (left_mb_field_flag != curr_mb_field_flag) yading@10: left_xy[LBOT] += h->mb_stride; yading@10: } yading@10: } yading@10: yading@10: h->top_mb_xy = top_xy; yading@10: h->left_mb_xy[LTOP] = left_xy[LTOP]; yading@10: h->left_mb_xy[LBOT] = left_xy[LBOT]; yading@10: { yading@10: /* For sufficiently low qp, filtering wouldn't do anything. yading@10: * This is a conservative estimate: could also check beta_offset yading@10: * and more accurate chroma_qp. */ yading@10: int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice yading@10: int qp = h->cur_pic.qscale_table[mb_xy]; yading@10: if (qp <= qp_thresh && yading@10: (left_xy[LTOP] < 0 || yading@10: ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) && yading@10: (top_xy < 0 || yading@10: ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) { yading@10: if (!FRAME_MBAFF(h)) yading@10: return 1; yading@10: if ((left_xy[LTOP] < 0 || yading@10: ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) && yading@10: (top_xy < h->mb_stride || yading@10: ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh)) yading@10: return 1; yading@10: } yading@10: } yading@10: yading@10: top_type = h->cur_pic.mb_type[top_xy]; yading@10: left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]]; yading@10: left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]]; yading@10: if (h->deblocking_filter == 2) { yading@10: if (h->slice_table[top_xy] != h->slice_num) yading@10: top_type = 0; yading@10: if (h->slice_table[left_xy[LBOT]] != h->slice_num) yading@10: left_type[LTOP] = left_type[LBOT] = 0; yading@10: } else { yading@10: if (h->slice_table[top_xy] == 0xFFFF) yading@10: top_type = 0; yading@10: if (h->slice_table[left_xy[LBOT]] == 0xFFFF) yading@10: left_type[LTOP] = left_type[LBOT] = 0; yading@10: } yading@10: h->top_type = top_type; yading@10: h->left_type[LTOP] = left_type[LTOP]; yading@10: h->left_type[LBOT] = left_type[LBOT]; yading@10: yading@10: if (IS_INTRA(mb_type)) yading@10: return 0; yading@10: yading@10: fill_filter_caches_inter(h, mb_type, top_xy, left_xy, yading@10: top_type, left_type, mb_xy, 0); yading@10: if (h->list_count == 2) yading@10: fill_filter_caches_inter(h, mb_type, top_xy, left_xy, yading@10: top_type, left_type, mb_xy, 1); yading@10: yading@10: nnz = h->non_zero_count[mb_xy]; yading@10: nnz_cache = h->non_zero_count_cache; yading@10: AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]); yading@10: AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]); yading@10: AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]); yading@10: AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]); yading@10: h->cbp = h->cbp_table[mb_xy]; yading@10: yading@10: if (top_type) { yading@10: nnz = h->non_zero_count[top_xy]; yading@10: AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]); yading@10: } yading@10: yading@10: if (left_type[LTOP]) { yading@10: nnz = h->non_zero_count[left_xy[LTOP]]; yading@10: nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4]; yading@10: nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4]; yading@10: nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4]; yading@10: nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4]; yading@10: } yading@10: yading@10: /* CAVLC 8x8dct requires NNZ values for residual decoding that differ yading@10: * from what the loop filter needs */ yading@10: if (!CABAC(h) && h->pps.transform_8x8_mode) { yading@10: if (IS_8x8DCT(top_type)) { yading@10: nnz_cache[4 + 8 * 0] = yading@10: nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12; yading@10: nnz_cache[6 + 8 * 0] = yading@10: nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12; yading@10: } yading@10: if (IS_8x8DCT(left_type[LTOP])) { yading@10: nnz_cache[3 + 8 * 1] = yading@10: nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF yading@10: } yading@10: if (IS_8x8DCT(left_type[LBOT])) { yading@10: nnz_cache[3 + 8 * 3] = yading@10: nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF yading@10: } yading@10: yading@10: if (IS_8x8DCT(mb_type)) { yading@10: nnz_cache[scan8[0]] = yading@10: nnz_cache[scan8[1]] = yading@10: nnz_cache[scan8[2]] = yading@10: nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12; yading@10: yading@10: nnz_cache[scan8[0 + 4]] = yading@10: nnz_cache[scan8[1 + 4]] = yading@10: nnz_cache[scan8[2 + 4]] = yading@10: nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12; yading@10: yading@10: nnz_cache[scan8[0 + 8]] = yading@10: nnz_cache[scan8[1 + 8]] = yading@10: nnz_cache[scan8[2 + 8]] = yading@10: nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12; yading@10: yading@10: nnz_cache[scan8[0 + 12]] = yading@10: nnz_cache[scan8[1 + 12]] = yading@10: nnz_cache[scan8[2 + 12]] = yading@10: nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12; yading@10: } yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static void loop_filter(H264Context *h, int start_x, int end_x) yading@10: { yading@10: uint8_t *dest_y, *dest_cb, *dest_cr; yading@10: int linesize, uvlinesize, mb_x, mb_y; yading@10: const int end_mb_y = h->mb_y + FRAME_MBAFF(h); yading@10: const int old_slice_type = h->slice_type; yading@10: const int pixel_shift = h->pixel_shift; yading@10: const int block_h = 16 >> h->chroma_y_shift; yading@10: yading@10: if (h->deblocking_filter) { yading@10: for (mb_x = start_x; mb_x < end_x; mb_x++) yading@10: for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) { yading@10: int mb_xy, mb_type; yading@10: mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride; yading@10: h->slice_num = h->slice_table[mb_xy]; yading@10: mb_type = h->cur_pic.mb_type[mb_xy]; yading@10: h->list_count = h->list_counts[mb_xy]; yading@10: yading@10: if (FRAME_MBAFF(h)) yading@10: h->mb_mbaff = yading@10: h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type); yading@10: yading@10: h->mb_x = mb_x; yading@10: h->mb_y = mb_y; yading@10: dest_y = h->cur_pic.f.data[0] + yading@10: ((mb_x << pixel_shift) + mb_y * h->linesize) * 16; yading@10: dest_cb = h->cur_pic.f.data[1] + yading@10: (mb_x << pixel_shift) * (8 << CHROMA444(h)) + yading@10: mb_y * h->uvlinesize * block_h; yading@10: dest_cr = h->cur_pic.f.data[2] + yading@10: (mb_x << pixel_shift) * (8 << CHROMA444(h)) + yading@10: mb_y * h->uvlinesize * block_h; yading@10: // FIXME simplify above yading@10: yading@10: if (MB_FIELD(h)) { yading@10: linesize = h->mb_linesize = h->linesize * 2; yading@10: uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2; yading@10: if (mb_y & 1) { // FIXME move out of this function? yading@10: dest_y -= h->linesize * 15; yading@10: dest_cb -= h->uvlinesize * (block_h - 1); yading@10: dest_cr -= h->uvlinesize * (block_h - 1); yading@10: } yading@10: } else { yading@10: linesize = h->mb_linesize = h->linesize; yading@10: uvlinesize = h->mb_uvlinesize = h->uvlinesize; yading@10: } yading@10: backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, yading@10: uvlinesize, 0); yading@10: if (fill_filter_caches(h, mb_type)) yading@10: continue; yading@10: h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]); yading@10: h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]); yading@10: yading@10: if (FRAME_MBAFF(h)) { yading@10: ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, yading@10: linesize, uvlinesize); yading@10: } else { yading@10: ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, yading@10: dest_cr, linesize, uvlinesize); yading@10: } yading@10: } yading@10: } yading@10: h->slice_type = old_slice_type; yading@10: h->mb_x = end_x; yading@10: h->mb_y = end_mb_y - FRAME_MBAFF(h); yading@10: h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale); yading@10: h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale); yading@10: } yading@10: yading@10: static void predict_field_decoding_flag(H264Context *h) yading@10: { yading@10: const int mb_xy = h->mb_x + h->mb_y * h->mb_stride; yading@10: int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ? yading@10: h->cur_pic.mb_type[mb_xy - 1] : yading@10: (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ? yading@10: h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0; yading@10: h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0; yading@10: } yading@10: yading@10: /** yading@10: * Draw edges and report progress for the last MB row. yading@10: */ yading@10: static void decode_finish_row(H264Context *h) yading@10: { yading@10: int top = 16 * (h->mb_y >> FIELD_PICTURE(h)); yading@10: int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h); yading@10: int height = 16 << FRAME_MBAFF(h); yading@10: int deblock_border = (16 + 4) << FRAME_MBAFF(h); yading@10: yading@10: if (h->deblocking_filter) { yading@10: if ((top + height) >= pic_height) yading@10: height += deblock_border; yading@10: top -= deblock_border; yading@10: } yading@10: yading@10: if (top >= pic_height || (top + height) < 0) yading@10: return; yading@10: yading@10: height = FFMIN(height, pic_height - top); yading@10: if (top < 0) { yading@10: height = top + height; yading@10: top = 0; yading@10: } yading@10: yading@10: ff_h264_draw_horiz_band(h, top, height); yading@10: yading@10: if (h->droppable || h->er.error_occurred) yading@10: return; yading@10: yading@10: ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1, yading@10: h->picture_structure == PICT_BOTTOM_FIELD); yading@10: } yading@10: yading@10: static void er_add_slice(H264Context *h, int startx, int starty, yading@10: int endx, int endy, int status) yading@10: { yading@10: if (CONFIG_ERROR_RESILIENCE) { yading@10: ERContext *er = &h->er; yading@10: yading@10: er->ref_count = h->ref_count[0]; yading@10: ff_er_add_slice(er, startx, starty, endx, endy, status); yading@10: } yading@10: } yading@10: yading@10: static int decode_slice(struct AVCodecContext *avctx, void *arg) yading@10: { yading@10: H264Context *h = *(void **)arg; yading@10: int lf_x_start = h->mb_x; yading@10: yading@10: h->mb_skip_run = -1; yading@10: yading@10: av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3)); yading@10: yading@10: h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME || yading@10: avctx->codec_id != AV_CODEC_ID_H264 || yading@10: (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY)); yading@10: yading@10: if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME) { yading@10: const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1); yading@10: if (start_i) { yading@10: int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]]; yading@10: prev_status &= ~ VP_START; yading@10: if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END)) yading@10: h->er.error_occurred = 1; yading@10: } yading@10: } yading@10: yading@10: if (h->pps.cabac) { yading@10: /* realign */ yading@10: align_get_bits(&h->gb); yading@10: yading@10: /* init cabac */ yading@10: ff_init_cabac_decoder(&h->cabac, yading@10: h->gb.buffer + get_bits_count(&h->gb) / 8, yading@10: (get_bits_left(&h->gb) + 7) / 8); yading@10: yading@10: ff_h264_init_cabac_states(h); yading@10: yading@10: for (;;) { yading@10: // START_TIMER yading@10: int ret = ff_h264_decode_mb_cabac(h); yading@10: int eos; yading@10: // STOP_TIMER("decode_mb_cabac") yading@10: yading@10: if (ret >= 0) yading@10: ff_h264_hl_decode_mb(h); yading@10: yading@10: // FIXME optimal? or let mb_decode decode 16x32 ? yading@10: if (ret >= 0 && FRAME_MBAFF(h)) { yading@10: h->mb_y++; yading@10: yading@10: ret = ff_h264_decode_mb_cabac(h); yading@10: yading@10: if (ret >= 0) yading@10: ff_h264_hl_decode_mb(h); yading@10: h->mb_y--; yading@10: } yading@10: eos = get_cabac_terminate(&h->cabac); yading@10: yading@10: if ((h->workaround_bugs & FF_BUG_TRUNCATED) && yading@10: h->cabac.bytestream > h->cabac.bytestream_end + 2) { yading@10: er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1, yading@10: h->mb_y, ER_MB_END); yading@10: if (h->mb_x >= lf_x_start) yading@10: loop_filter(h, lf_x_start, h->mb_x + 1); yading@10: return 0; yading@10: } yading@10: if (h->cabac.bytestream > h->cabac.bytestream_end + 2 ) yading@10: av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream); yading@10: if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "error while decoding MB %d %d, bytestream (%td)\n", yading@10: h->mb_x, h->mb_y, yading@10: h->cabac.bytestream_end - h->cabac.bytestream); yading@10: er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x, yading@10: h->mb_y, ER_MB_ERROR); yading@10: return -1; yading@10: } yading@10: yading@10: if (++h->mb_x >= h->mb_width) { yading@10: loop_filter(h, lf_x_start, h->mb_x); yading@10: h->mb_x = lf_x_start = 0; yading@10: decode_finish_row(h); yading@10: ++h->mb_y; yading@10: if (FIELD_OR_MBAFF_PICTURE(h)) { yading@10: ++h->mb_y; yading@10: if (FRAME_MBAFF(h) && h->mb_y < h->mb_height) yading@10: predict_field_decoding_flag(h); yading@10: } yading@10: } yading@10: yading@10: if (eos || h->mb_y >= h->mb_height) { yading@10: tprintf(h->avctx, "slice end %d %d\n", yading@10: get_bits_count(&h->gb), h->gb.size_in_bits); yading@10: er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1, yading@10: h->mb_y, ER_MB_END); yading@10: if (h->mb_x > lf_x_start) yading@10: loop_filter(h, lf_x_start, h->mb_x); yading@10: return 0; yading@10: } yading@10: } yading@10: } else { yading@10: for (;;) { yading@10: int ret = ff_h264_decode_mb_cavlc(h); yading@10: yading@10: if (ret >= 0) yading@10: ff_h264_hl_decode_mb(h); yading@10: yading@10: // FIXME optimal? or let mb_decode decode 16x32 ? yading@10: if (ret >= 0 && FRAME_MBAFF(h)) { yading@10: h->mb_y++; yading@10: ret = ff_h264_decode_mb_cavlc(h); yading@10: yading@10: if (ret >= 0) yading@10: ff_h264_hl_decode_mb(h); yading@10: h->mb_y--; yading@10: } yading@10: yading@10: if (ret < 0) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "error while decoding MB %d %d\n", h->mb_x, h->mb_y); yading@10: er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x, yading@10: h->mb_y, ER_MB_ERROR); yading@10: return -1; yading@10: } yading@10: yading@10: if (++h->mb_x >= h->mb_width) { yading@10: loop_filter(h, lf_x_start, h->mb_x); yading@10: h->mb_x = lf_x_start = 0; yading@10: decode_finish_row(h); yading@10: ++h->mb_y; yading@10: if (FIELD_OR_MBAFF_PICTURE(h)) { yading@10: ++h->mb_y; yading@10: if (FRAME_MBAFF(h) && h->mb_y < h->mb_height) yading@10: predict_field_decoding_flag(h); yading@10: } yading@10: if (h->mb_y >= h->mb_height) { yading@10: tprintf(h->avctx, "slice end %d %d\n", yading@10: get_bits_count(&h->gb), h->gb.size_in_bits); yading@10: yading@10: if ( get_bits_left(&h->gb) == 0 yading@10: || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) { yading@10: er_add_slice(h, h->resync_mb_x, h->resync_mb_y, yading@10: h->mb_x - 1, h->mb_y, yading@10: ER_MB_END); yading@10: yading@10: return 0; yading@10: } else { yading@10: er_add_slice(h, h->resync_mb_x, h->resync_mb_y, yading@10: h->mb_x, h->mb_y, yading@10: ER_MB_END); yading@10: yading@10: return -1; yading@10: } yading@10: } yading@10: } yading@10: yading@10: if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) { yading@10: tprintf(h->avctx, "slice end %d %d\n", yading@10: get_bits_count(&h->gb), h->gb.size_in_bits); yading@10: if (get_bits_left(&h->gb) == 0) { yading@10: er_add_slice(h, h->resync_mb_x, h->resync_mb_y, yading@10: h->mb_x - 1, h->mb_y, yading@10: ER_MB_END); yading@10: if (h->mb_x > lf_x_start) yading@10: loop_filter(h, lf_x_start, h->mb_x); yading@10: yading@10: return 0; yading@10: } else { yading@10: er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x, yading@10: h->mb_y, ER_MB_ERROR); yading@10: yading@10: return -1; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: /** yading@10: * Call decode_slice() for each context. yading@10: * yading@10: * @param h h264 master context yading@10: * @param context_count number of contexts to execute yading@10: */ yading@10: static int execute_decode_slices(H264Context *h, int context_count) yading@10: { yading@10: AVCodecContext *const avctx = h->avctx; yading@10: H264Context *hx; yading@10: int i; yading@10: yading@10: if (h->avctx->hwaccel || yading@10: h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) yading@10: return 0; yading@10: if (context_count == 1) { yading@10: return decode_slice(avctx, &h); yading@10: } else { yading@10: av_assert0(context_count > 0); yading@10: for (i = 1; i < context_count; i++) { yading@10: hx = h->thread_context[i]; yading@10: if (CONFIG_ERROR_RESILIENCE) { yading@10: hx->er.error_count = 0; yading@10: } yading@10: hx->x264_build = h->x264_build; yading@10: } yading@10: yading@10: avctx->execute(avctx, decode_slice, h->thread_context, yading@10: NULL, context_count, sizeof(void *)); yading@10: yading@10: /* pull back stuff from slices to master context */ yading@10: hx = h->thread_context[context_count - 1]; yading@10: h->mb_x = hx->mb_x; yading@10: h->mb_y = hx->mb_y; yading@10: h->droppable = hx->droppable; yading@10: h->picture_structure = hx->picture_structure; yading@10: if (CONFIG_ERROR_RESILIENCE) { yading@10: for (i = 1; i < context_count; i++) yading@10: h->er.error_count += h->thread_context[i]->er.error_count; yading@10: } yading@10: } yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size, yading@10: int parse_extradata) yading@10: { yading@10: AVCodecContext *const avctx = h->avctx; yading@10: H264Context *hx; ///< thread context yading@10: int buf_index; yading@10: int context_count; yading@10: int next_avc; yading@10: int pass = !(avctx->active_thread_type & FF_THREAD_FRAME); yading@10: int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts yading@10: int nal_index; yading@10: int idr_cleared=0; yading@10: int first_slice = 0; yading@10: yading@10: h->nal_unit_type= 0; yading@10: yading@10: if(!h->slice_context_count) yading@10: h->slice_context_count= 1; yading@10: h->max_contexts = h->slice_context_count; yading@10: if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS)) { yading@10: h->current_slice = 0; yading@10: if (!h->first_field) yading@10: h->cur_pic_ptr = NULL; yading@10: ff_h264_reset_sei(h); yading@10: } yading@10: yading@10: if (h->nal_length_size == 4) { yading@10: if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) { yading@10: h->is_avc = 0; yading@10: }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size) yading@10: h->is_avc = 1; yading@10: } yading@10: yading@10: for (; pass <= 1; pass++) { yading@10: buf_index = 0; yading@10: context_count = 0; yading@10: next_avc = h->is_avc ? 0 : buf_size; yading@10: nal_index = 0; yading@10: for (;;) { yading@10: int consumed; yading@10: int dst_length; yading@10: int bit_length; yading@10: const uint8_t *ptr; yading@10: int i, nalsize = 0; yading@10: int err; yading@10: yading@10: if (buf_index >= next_avc) { yading@10: if (buf_index >= buf_size - h->nal_length_size) yading@10: break; yading@10: nalsize = 0; yading@10: for (i = 0; i < h->nal_length_size; i++) yading@10: nalsize = (nalsize << 8) | buf[buf_index++]; yading@10: if (nalsize <= 0 || nalsize > buf_size - buf_index) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "AVC: nal size %d\n", nalsize); yading@10: break; yading@10: } yading@10: next_avc = buf_index + nalsize; yading@10: } else { yading@10: // start code prefix search yading@10: for (; buf_index + 3 < next_avc; buf_index++) yading@10: // This should always succeed in the first iteration. yading@10: if (buf[buf_index] == 0 && yading@10: buf[buf_index + 1] == 0 && yading@10: buf[buf_index + 2] == 1) yading@10: break; yading@10: yading@10: if (buf_index + 3 >= buf_size) { yading@10: buf_index = buf_size; yading@10: break; yading@10: } yading@10: yading@10: buf_index += 3; yading@10: if (buf_index >= next_avc) yading@10: continue; yading@10: } yading@10: yading@10: hx = h->thread_context[context_count]; yading@10: yading@10: ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length, yading@10: &consumed, next_avc - buf_index); yading@10: if (ptr == NULL || dst_length < 0) { yading@10: buf_index = -1; yading@10: goto end; yading@10: } yading@10: i = buf_index + consumed; yading@10: if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc && yading@10: buf[i] == 0x00 && buf[i + 1] == 0x00 && yading@10: buf[i + 2] == 0x01 && buf[i + 3] == 0xE0) yading@10: h->workaround_bugs |= FF_BUG_TRUNCATED; yading@10: yading@10: if (!(h->workaround_bugs & FF_BUG_TRUNCATED)) yading@10: while(dst_length > 0 && ptr[dst_length - 1] == 0) yading@10: dst_length--; yading@10: bit_length = !dst_length ? 0 yading@10: : (8 * dst_length - yading@10: decode_rbsp_trailing(h, ptr + dst_length - 1)); yading@10: yading@10: if (h->avctx->debug & FF_DEBUG_STARTCODE) yading@10: av_log(h->avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d pass %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length, pass); yading@10: yading@10: if (h->is_avc && (nalsize != consumed) && nalsize) yading@10: av_log(h->avctx, AV_LOG_DEBUG, yading@10: "AVC: Consumed only %d bytes instead of %d\n", yading@10: consumed, nalsize); yading@10: yading@10: buf_index += consumed; yading@10: nal_index++; yading@10: yading@10: if (pass == 0) { yading@10: /* packets can sometimes contain multiple PPS/SPS, yading@10: * e.g. two PAFF field pictures in one packet, or a demuxer yading@10: * which splits NALs strangely if so, when frame threading we yading@10: * can't start the next thread until we've read all of them */ yading@10: switch (hx->nal_unit_type) { yading@10: case NAL_SPS: yading@10: case NAL_PPS: yading@10: nals_needed = nal_index; yading@10: break; yading@10: case NAL_DPA: yading@10: case NAL_IDR_SLICE: yading@10: case NAL_SLICE: yading@10: init_get_bits(&hx->gb, ptr, bit_length); yading@10: if (!get_ue_golomb(&hx->gb) || !first_slice) yading@10: nals_needed = nal_index; yading@10: if (!first_slice) yading@10: first_slice = hx->nal_unit_type; yading@10: } yading@10: continue; yading@10: } yading@10: yading@10: if (!first_slice) yading@10: switch (hx->nal_unit_type) { yading@10: case NAL_DPA: yading@10: case NAL_IDR_SLICE: yading@10: case NAL_SLICE: yading@10: first_slice = hx->nal_unit_type; yading@10: } yading@10: yading@10: // FIXME do not discard SEI id yading@10: if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0) yading@10: continue; yading@10: yading@10: again: yading@10: /* Ignore per frame NAL unit type during extradata yading@10: * parsing. Decoding slices is not possible in codec init yading@10: * with frame-mt */ yading@10: if (parse_extradata) { yading@10: switch (hx->nal_unit_type) { yading@10: case NAL_IDR_SLICE: yading@10: case NAL_SLICE: yading@10: case NAL_DPA: yading@10: case NAL_DPB: yading@10: case NAL_DPC: yading@10: case NAL_AUXILIARY_SLICE: yading@10: av_log(h->avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header/extradata\n", hx->nal_unit_type); yading@10: hx->nal_unit_type = NAL_FF_IGNORE; yading@10: } yading@10: } yading@10: yading@10: err = 0; yading@10: yading@10: switch (hx->nal_unit_type) { yading@10: case NAL_IDR_SLICE: yading@10: if (first_slice != NAL_IDR_SLICE) { yading@10: av_log(h->avctx, AV_LOG_ERROR, yading@10: "Invalid mix of idr and non-idr slices\n"); yading@10: buf_index = -1; yading@10: goto end; yading@10: } yading@10: if(!idr_cleared) yading@10: idr(h); // FIXME ensure we don't lose some frames if there is reordering yading@10: idr_cleared = 1; yading@10: case NAL_SLICE: yading@10: init_get_bits(&hx->gb, ptr, bit_length); yading@10: hx->intra_gb_ptr = yading@10: hx->inter_gb_ptr = &hx->gb; yading@10: hx->data_partitioning = 0; yading@10: yading@10: if ((err = decode_slice_header(hx, h))) yading@10: break; yading@10: yading@10: if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I)) yading@10: h->valid_recovery_point = 1; yading@10: yading@10: if ( h->sei_recovery_frame_cnt >= 0 yading@10: && ( h->recovery_frame<0 yading@10: || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) { yading@10: h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) % yading@10: (1 << h->sps.log2_max_frame_num); yading@10: yading@10: if (!h->valid_recovery_point) yading@10: h->recovery_frame = h->frame_num; yading@10: } yading@10: yading@10: h->cur_pic_ptr->f.key_frame |= yading@10: (hx->nal_unit_type == NAL_IDR_SLICE); yading@10: yading@10: if (h->recovery_frame == h->frame_num) { yading@10: h->cur_pic_ptr->sync |= 1; yading@10: h->recovery_frame = -1; yading@10: } yading@10: yading@10: h->sync |= !!h->cur_pic_ptr->f.key_frame; yading@10: h->sync |= 3*!!(avctx->flags2 & CODEC_FLAG2_SHOW_ALL); yading@10: h->cur_pic_ptr->sync |= h->sync; yading@10: yading@10: if (h->current_slice == 1) { yading@10: if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS)) yading@10: decode_postinit(h, nal_index >= nals_needed); yading@10: yading@10: if (h->avctx->hwaccel && yading@10: h->avctx->hwaccel->start_frame(h->avctx, NULL, 0) < 0) yading@10: return -1; yading@10: if (CONFIG_H264_VDPAU_DECODER && yading@10: h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) yading@10: ff_vdpau_h264_picture_start(h); yading@10: } yading@10: yading@10: if (hx->redundant_pic_count == 0 && yading@10: (avctx->skip_frame < AVDISCARD_NONREF || yading@10: hx->nal_ref_idc) && yading@10: (avctx->skip_frame < AVDISCARD_BIDIR || yading@10: hx->slice_type_nos != AV_PICTURE_TYPE_B) && yading@10: (avctx->skip_frame < AVDISCARD_NONKEY || yading@10: hx->slice_type_nos == AV_PICTURE_TYPE_I) && yading@10: avctx->skip_frame < AVDISCARD_ALL) { yading@10: if (avctx->hwaccel) { yading@10: if (avctx->hwaccel->decode_slice(avctx, yading@10: &buf[buf_index - consumed], yading@10: consumed) < 0) yading@10: return -1; yading@10: } else if (CONFIG_H264_VDPAU_DECODER && yading@10: h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) { yading@10: static const uint8_t start_code[] = { yading@10: 0x00, 0x00, 0x01 }; yading@10: ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0], start_code, yading@10: sizeof(start_code)); yading@10: ff_vdpau_add_data_chunk(h->cur_pic_ptr->f.data[0], &buf[buf_index - consumed], yading@10: consumed); yading@10: } else yading@10: context_count++; yading@10: } yading@10: break; yading@10: case NAL_DPA: yading@10: init_get_bits(&hx->gb, ptr, bit_length); yading@10: hx->intra_gb_ptr = yading@10: hx->inter_gb_ptr = NULL; yading@10: yading@10: if ((err = decode_slice_header(hx, h)) < 0) yading@10: break; yading@10: yading@10: hx->data_partitioning = 1; yading@10: break; yading@10: case NAL_DPB: yading@10: init_get_bits(&hx->intra_gb, ptr, bit_length); yading@10: hx->intra_gb_ptr = &hx->intra_gb; yading@10: break; yading@10: case NAL_DPC: yading@10: init_get_bits(&hx->inter_gb, ptr, bit_length); yading@10: hx->inter_gb_ptr = &hx->inter_gb; yading@10: yading@10: av_log(h->avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n"); yading@10: break; yading@10: yading@10: if (hx->redundant_pic_count == 0 && yading@10: hx->intra_gb_ptr && yading@10: hx->data_partitioning && yading@10: h->cur_pic_ptr && h->context_initialized && yading@10: (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) && yading@10: (avctx->skip_frame < AVDISCARD_BIDIR || yading@10: hx->slice_type_nos != AV_PICTURE_TYPE_B) && yading@10: (avctx->skip_frame < AVDISCARD_NONKEY || yading@10: hx->slice_type_nos == AV_PICTURE_TYPE_I) && yading@10: avctx->skip_frame < AVDISCARD_ALL) yading@10: context_count++; yading@10: break; yading@10: case NAL_SEI: yading@10: init_get_bits(&h->gb, ptr, bit_length); yading@10: ff_h264_decode_sei(h); yading@10: break; yading@10: case NAL_SPS: yading@10: init_get_bits(&h->gb, ptr, bit_length); yading@10: if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) { yading@10: av_log(h->avctx, AV_LOG_DEBUG, yading@10: "SPS decoding failure, trying again with the complete NAL\n"); yading@10: if (h->is_avc) yading@10: av_assert0(next_avc - buf_index + consumed == nalsize); yading@10: if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8) yading@10: break; yading@10: init_get_bits(&h->gb, &buf[buf_index + 1 - consumed], yading@10: 8*(next_avc - buf_index + consumed - 1)); yading@10: ff_h264_decode_seq_parameter_set(h); yading@10: } yading@10: yading@10: break; yading@10: case NAL_PPS: yading@10: init_get_bits(&h->gb, ptr, bit_length); yading@10: ff_h264_decode_picture_parameter_set(h, bit_length); yading@10: break; yading@10: case NAL_AUD: yading@10: case NAL_END_SEQUENCE: yading@10: case NAL_END_STREAM: yading@10: case NAL_FILLER_DATA: yading@10: case NAL_SPS_EXT: yading@10: case NAL_AUXILIARY_SLICE: yading@10: break; yading@10: case NAL_FF_IGNORE: yading@10: break; yading@10: default: yading@10: av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", yading@10: hx->nal_unit_type, bit_length); yading@10: } yading@10: yading@10: if (context_count == h->max_contexts) { yading@10: execute_decode_slices(h, context_count); yading@10: context_count = 0; yading@10: } yading@10: yading@10: if (err < 0) yading@10: av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n"); yading@10: else if (err == 1) { yading@10: /* Slice could not be decoded in parallel mode, copy down yading@10: * NAL unit stuff to context 0 and restart. Note that yading@10: * rbsp_buffer is not transferred, but since we no longer yading@10: * run in parallel mode this should not be an issue. */ yading@10: h->nal_unit_type = hx->nal_unit_type; yading@10: h->nal_ref_idc = hx->nal_ref_idc; yading@10: hx = h; yading@10: goto again; yading@10: } yading@10: } yading@10: } yading@10: if (context_count) yading@10: execute_decode_slices(h, context_count); yading@10: yading@10: end: yading@10: /* clean up */ yading@10: if (h->cur_pic_ptr && !h->droppable) { yading@10: ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, yading@10: h->picture_structure == PICT_BOTTOM_FIELD); yading@10: } yading@10: yading@10: return buf_index; yading@10: } yading@10: yading@10: /** yading@10: * Return the number of bytes consumed for building the current frame. yading@10: */ yading@10: static int get_consumed_bytes(int pos, int buf_size) yading@10: { yading@10: if (pos == 0) yading@10: pos = 1; // avoid infinite loops (i doubt that is needed but ...) yading@10: if (pos + 10 > buf_size) yading@10: pos = buf_size; // oops ;) yading@10: yading@10: return pos; yading@10: } yading@10: yading@10: static int output_frame(H264Context *h, AVFrame *dst, AVFrame *src) yading@10: { yading@10: int i; yading@10: int ret = av_frame_ref(dst, src); yading@10: if (ret < 0) yading@10: return ret; yading@10: yading@10: if (!h->sps.crop) yading@10: return 0; yading@10: yading@10: for (i = 0; i < 3; i++) { yading@10: int hshift = (i > 0) ? h->chroma_x_shift : 0; yading@10: int vshift = (i > 0) ? h->chroma_y_shift : 0; yading@10: int off = ((h->sps.crop_left >> hshift) << h->pixel_shift) + yading@10: (h->sps.crop_top >> vshift) * dst->linesize[i]; yading@10: dst->data[i] += off; yading@10: } yading@10: return 0; yading@10: } yading@10: yading@10: static int decode_frame(AVCodecContext *avctx, void *data, yading@10: int *got_frame, AVPacket *avpkt) yading@10: { yading@10: const uint8_t *buf = avpkt->data; yading@10: int buf_size = avpkt->size; yading@10: H264Context *h = avctx->priv_data; yading@10: AVFrame *pict = data; yading@10: int buf_index = 0; yading@10: Picture *out; yading@10: int i, out_idx; yading@10: int ret; yading@10: yading@10: h->flags = avctx->flags; yading@10: yading@10: /* end of stream, output what is still in the buffers */ yading@10: if (buf_size == 0) { yading@10: out: yading@10: yading@10: h->cur_pic_ptr = NULL; yading@10: h->first_field = 0; yading@10: yading@10: // FIXME factorize this with the output code below yading@10: out = h->delayed_pic[0]; yading@10: out_idx = 0; yading@10: for (i = 1; yading@10: h->delayed_pic[i] && yading@10: !h->delayed_pic[i]->f.key_frame && yading@10: !h->delayed_pic[i]->mmco_reset; yading@10: i++) yading@10: if (h->delayed_pic[i]->poc < out->poc) { yading@10: out = h->delayed_pic[i]; yading@10: out_idx = i; yading@10: } yading@10: yading@10: for (i = out_idx; h->delayed_pic[i]; i++) yading@10: h->delayed_pic[i] = h->delayed_pic[i + 1]; yading@10: yading@10: if (out) { yading@10: out->reference &= ~DELAYED_PIC_REF; yading@10: ret = output_frame(h, pict, &out->f); yading@10: if (ret < 0) yading@10: return ret; yading@10: *got_frame = 1; yading@10: } yading@10: yading@10: return buf_index; yading@10: } yading@10: if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){ yading@10: int cnt= buf[5]&0x1f; yading@10: const uint8_t *p= buf+6; yading@10: while(cnt--){ yading@10: int nalsize= AV_RB16(p) + 2; yading@10: if(nalsize > buf_size - (p-buf) || p[2]!=0x67) yading@10: goto not_extra; yading@10: p += nalsize; yading@10: } yading@10: cnt = *(p++); yading@10: if(!cnt) yading@10: goto not_extra; yading@10: while(cnt--){ yading@10: int nalsize= AV_RB16(p) + 2; yading@10: if(nalsize > buf_size - (p-buf) || p[2]!=0x68) yading@10: goto not_extra; yading@10: p += nalsize; yading@10: } yading@10: yading@10: return ff_h264_decode_extradata(h, buf, buf_size); yading@10: } yading@10: not_extra: yading@10: yading@10: buf_index = decode_nal_units(h, buf, buf_size, 0); yading@10: if (buf_index < 0) yading@10: return -1; yading@10: yading@10: if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) { yading@10: av_assert0(buf_index <= buf_size); yading@10: goto out; yading@10: } yading@10: yading@10: if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) { yading@10: if (avctx->skip_frame >= AVDISCARD_NONREF || yading@10: buf_size >= 4 && !memcmp("Q264", buf, 4)) yading@10: return buf_size; yading@10: av_log(avctx, AV_LOG_ERROR, "no frame!\n"); yading@10: return -1; yading@10: } yading@10: yading@10: if (!(avctx->flags2 & CODEC_FLAG2_CHUNKS) || yading@10: (h->mb_y >= h->mb_height && h->mb_height)) { yading@10: if (avctx->flags2 & CODEC_FLAG2_CHUNKS) yading@10: decode_postinit(h, 1); yading@10: yading@10: field_end(h, 0); yading@10: yading@10: /* Wait for second field. */ yading@10: *got_frame = 0; yading@10: if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) { yading@10: ret = output_frame(h, pict, &h->next_output_pic->f); yading@10: if (ret < 0) yading@10: return ret; yading@10: *got_frame = 1; yading@10: if (CONFIG_MPEGVIDEO) { yading@10: ff_print_debug_info2(h->avctx, h->next_output_pic, pict, h->er.mbskip_table, yading@10: &h->low_delay, yading@10: h->mb_width, h->mb_height, h->mb_stride, 1); yading@10: } yading@10: } yading@10: } yading@10: yading@10: assert(pict->data[0] || !*got_frame); yading@10: yading@10: return get_consumed_bytes(buf_index, buf_size); yading@10: } yading@10: yading@10: av_cold void ff_h264_free_context(H264Context *h) yading@10: { yading@10: int i; yading@10: yading@10: free_tables(h, 1); // FIXME cleanup init stuff perhaps yading@10: yading@10: for (i = 0; i < MAX_SPS_COUNT; i++) yading@10: av_freep(h->sps_buffers + i); yading@10: yading@10: for (i = 0; i < MAX_PPS_COUNT; i++) yading@10: av_freep(h->pps_buffers + i); yading@10: } yading@10: yading@10: static av_cold int h264_decode_end(AVCodecContext *avctx) yading@10: { yading@10: H264Context *h = avctx->priv_data; yading@10: yading@10: ff_h264_remove_all_refs(h); yading@10: ff_h264_free_context(h); yading@10: yading@10: unref_picture(h, &h->cur_pic); yading@10: yading@10: return 0; yading@10: } yading@10: yading@10: static const AVProfile profiles[] = { yading@10: { FF_PROFILE_H264_BASELINE, "Baseline" }, yading@10: { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" }, yading@10: { FF_PROFILE_H264_MAIN, "Main" }, yading@10: { FF_PROFILE_H264_EXTENDED, "Extended" }, yading@10: { FF_PROFILE_H264_HIGH, "High" }, yading@10: { FF_PROFILE_H264_HIGH_10, "High 10" }, yading@10: { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" }, yading@10: { FF_PROFILE_H264_HIGH_422, "High 4:2:2" }, yading@10: { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" }, yading@10: { FF_PROFILE_H264_HIGH_444, "High 4:4:4" }, yading@10: { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" }, yading@10: { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" }, yading@10: { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" }, yading@10: { FF_PROFILE_UNKNOWN }, yading@10: }; yading@10: yading@10: static const AVOption h264_options[] = { yading@10: {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0}, yading@10: {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0}, yading@10: {NULL} yading@10: }; yading@10: yading@10: static const AVClass h264_class = { yading@10: .class_name = "H264 Decoder", yading@10: .item_name = av_default_item_name, yading@10: .option = h264_options, yading@10: .version = LIBAVUTIL_VERSION_INT, yading@10: }; yading@10: yading@10: static const AVClass h264_vdpau_class = { yading@10: .class_name = "H264 VDPAU Decoder", yading@10: .item_name = av_default_item_name, yading@10: .option = h264_options, yading@10: .version = LIBAVUTIL_VERSION_INT, yading@10: }; yading@10: yading@10: AVCodec ff_h264_decoder = { yading@10: .name = "h264", yading@10: .type = AVMEDIA_TYPE_VIDEO, yading@10: .id = AV_CODEC_ID_H264, yading@10: .priv_data_size = sizeof(H264Context), yading@10: .init = ff_h264_decode_init, yading@10: .close = h264_decode_end, yading@10: .decode = decode_frame, yading@10: .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | yading@10: CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS | yading@10: CODEC_CAP_FRAME_THREADS, yading@10: .flush = flush_dpb, yading@10: .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"), yading@10: .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy), yading@10: .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context), yading@10: .profiles = NULL_IF_CONFIG_SMALL(profiles), yading@10: .priv_class = &h264_class, yading@10: }; yading@10: yading@10: #if CONFIG_H264_VDPAU_DECODER yading@10: AVCodec ff_h264_vdpau_decoder = { yading@10: .name = "h264_vdpau", yading@10: .type = AVMEDIA_TYPE_VIDEO, yading@10: .id = AV_CODEC_ID_H264, yading@10: .priv_data_size = sizeof(H264Context), yading@10: .init = ff_h264_decode_init, yading@10: .close = h264_decode_end, yading@10: .decode = decode_frame, yading@10: .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU, yading@10: .flush = flush_dpb, yading@10: .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"), yading@10: .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264, yading@10: AV_PIX_FMT_NONE}, yading@10: .profiles = NULL_IF_CONFIG_SMALL(profiles), yading@10: .priv_class = &h264_vdpau_class, yading@10: }; yading@10: #endif