avcodec/mpeg4_unpack_bframes_bsf: Copy packet props
[ffmpeg.git] / libavcodec / h264_parser.c
index a16366a..ce46c58 100644 (file)
 
 /**
  * @file
- * H.264 / AVC / MPEG4 part10 parser.
+ * H.264 / AVC / MPEG-4 part10 parser.
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
 #define UNCHECKED_BITSTREAM_READER 1
 
-#include "libavutil/attributes.h"
-#include "parser.h"
-#include "h264data.h"
+#include <assert.h>
+#include <stdint.h>
+
+#include "libavutil/avutil.h"
+#include "libavutil/error.h"
+#include "libavutil/log.h"
+#include "libavutil/mem.h"
+#include "libavutil/pixfmt.h"
+
+#include "get_bits.h"
 #include "golomb.h"
+#include "h264.h"
+#include "h264_sei.h"
+#include "h264data.h"
 #include "internal.h"
+#include "mpegutils.h"
+#include "parser.h"
 
-
-static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
-                               int buf_size)
+typedef struct H264ParseContext {
+    ParseContext pc;
+    H264ParamSets ps;
+    H264DSPContext h264dsp;
+    H264POCContext poc;
+    H264SEIContext sei;
+    int is_avc;
+    int nal_length_size;
+    int got_first;
+    int picture_structure;
+    uint8_t parse_history[6];
+    int parse_history_count;
+    int parse_last_mb;
+} H264ParseContext;
+
+
+static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
+                               int buf_size, void *logctx)
 {
     int i, j;
     uint32_t state;
-    ParseContext *pc = &h->parse_context;
-    int next_avc= h->is_avc ? 0 : buf_size;
+    ParseContext *pc = &p->pc;
 
+    int next_avc = p->is_avc ? 0 : buf_size;
 //    mb_addr= pc->mb_addr - 1;
     state = pc->state;
     if (state > 13)
         state = 7;
 
-    if (h->is_avc && !h->nal_length_size)
-        av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
+    if (p->is_avc && !p->nal_length_size)
+        av_log(logctx, AV_LOG_ERROR, "AVC-parser: nal length size invalid\n");
 
     for (i = 0; i < buf_size; i++) {
         if (i >= next_avc) {
             int nalsize = 0;
             i = next_avc;
-            for (j = 0; j < h->nal_length_size; j++)
+            for (j = 0; j < p->nal_length_size; j++)
                 nalsize = (nalsize << 8) | buf[i++];
             if (nalsize <= 0 || nalsize > buf_size - i) {
-                av_log(h->avctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
+                av_log(logctx, AV_LOG_ERROR, "AVC-parser: nal size %d remaining %d\n", nalsize, buf_size - i);
                 return buf_size;
             }
             next_avc = i + nalsize;
@@ -65,7 +92,7 @@ static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
         }
 
         if (state == 7) {
-            i += h->h264dsp.h264_find_start_code_candidate(buf + i, next_avc - i);
+            i += p->h264dsp.startcode_find_candidate(buf + i, next_avc - i);
             if (i < next_avc)
                 state = 2;
         } else if (state <= 2) {
@@ -76,28 +103,29 @@ static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
             else
                 state >>= 1;           // 2->1, 1->0, 0->0
         } else if (state <= 5) {
-            int v = buf[i] & 0x1F;
-            if (v == 6 || v == 7 || v == 8 || v == 9) {
+            int nalu_type = buf[i] & 0x1F;
+            if (nalu_type == NAL_SEI || nalu_type == NAL_SPS ||
+                nalu_type == NAL_PPS || nalu_type == NAL_AUD) {
                 if (pc->frame_start_found) {
                     i++;
                     goto found;
                 }
-            } else if (v == 1 || v == 2 || v == 5) {
+            } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
+                       nalu_type == NAL_IDR_SLICE) {
                 state += 8;
                 continue;
             }
             state = 7;
         } else {
-            h->parse_history[h->parse_history_count++]= buf[i];
-            if (h->parse_history_count>3) {
-                unsigned int mb, last_mb= h->parse_last_mb;
+            p->parse_history[p->parse_history_count++] = buf[i];
+            if (p->parse_history_count > 5) {
+                unsigned int mb, last_mb = p->parse_last_mb;
                 GetBitContext gb;
 
-                init_get_bits(&gb, h->parse_history, 8*h->parse_history_count);
-                h->parse_history_count=0;
+                init_get_bits(&gb, p->parse_history, 8*p->parse_history_count);
+                p->parse_history_count = 0;
                 mb= get_ue_golomb_long(&gb);
-                last_mb= h->parse_last_mb;
-                h->parse_last_mb= mb;
+                p->parse_last_mb = mb;
                 if (pc->frame_start_found) {
                     if (mb <= last_mb)
                         goto found;
@@ -108,16 +136,114 @@ static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
         }
     }
     pc->state = state;
-    if (h->is_avc)
+    if (p->is_avc)
         return next_avc;
     return END_NOT_FOUND;
 
 found:
     pc->state             = 7;
     pc->frame_start_found = 0;
-    if (h->is_avc)
+    if (p->is_avc)
         return next_avc;
-    return i - (state & 5) - 3 * (state > 7);
+    return i - (state & 5) - 5 * (state > 7);
+}
+
+static int scan_mmco_reset(AVCodecParserContext *s, GetBitContext *gb,
+                           void *logctx)
+{
+    H264PredWeightTable pwt;
+    int slice_type_nos = s->pict_type & 3;
+    H264ParseContext *p = s->priv_data;
+    int list_count, ref_count[2];
+
+
+    if (p->ps.pps->redundant_pic_cnt_present)
+        get_ue_golomb(gb); // redundant_pic_count
+
+    if (slice_type_nos == AV_PICTURE_TYPE_B)
+        get_bits1(gb); // direct_spatial_mv_pred
+
+    if (ff_h264_parse_ref_count(&list_count, ref_count, gb, p->ps.pps,
+                                slice_type_nos, p->picture_structure, logctx) < 0)
+        return AVERROR_INVALIDDATA;
+
+    if (slice_type_nos != AV_PICTURE_TYPE_I) {
+        int list;
+        for (list = 0; list < list_count; list++) {
+            if (get_bits1(gb)) {
+                int index;
+                for (index = 0; ; index++) {
+                    unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(gb);
+
+                    if (reordering_of_pic_nums_idc < 3)
+                        get_ue_golomb_long(gb);
+                    else if (reordering_of_pic_nums_idc > 3) {
+                        av_log(logctx, AV_LOG_ERROR,
+                               "illegal reordering_of_pic_nums_idc %d\n",
+                               reordering_of_pic_nums_idc);
+                        return AVERROR_INVALIDDATA;
+                    } else
+                        break;
+
+                    if (index >= ref_count[list]) {
+                        av_log(logctx, AV_LOG_ERROR,
+                               "reference count %d overflow\n", index);
+                        return AVERROR_INVALIDDATA;
+                    }
+                }
+            }
+        }
+    }
+
+    if ((p->ps.pps->weighted_pred && slice_type_nos == AV_PICTURE_TYPE_P) ||
+        (p->ps.pps->weighted_bipred_idc == 1 && slice_type_nos == AV_PICTURE_TYPE_B))
+        ff_h264_pred_weight_table(gb, p->ps.sps, ref_count, slice_type_nos,
+                                  &pwt, logctx);
+
+    if (get_bits1(gb)) { // adaptive_ref_pic_marking_mode_flag
+        int i;
+        for (i = 0; i < MAX_MMCO_COUNT; i++) {
+            MMCOOpcode opcode = get_ue_golomb_31(gb);
+            if (opcode > (unsigned) MMCO_LONG) {
+                av_log(logctx, AV_LOG_ERROR,
+                       "illegal memory management control operation %d\n",
+                       opcode);
+                return AVERROR_INVALIDDATA;
+            }
+            if (opcode == MMCO_END)
+               return 0;
+            else if (opcode == MMCO_RESET)
+                return 1;
+
+            if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
+                get_ue_golomb_long(gb); // difference_of_pic_nums_minus1
+            if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
+                opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
+                get_ue_golomb_31(gb);
+        }
+    }
+
+    return 0;
+}
+
+static inline int get_avc_nalsize(H264ParseContext *p, const uint8_t *buf,
+                                  int buf_size, int *buf_index, void *logctx)
+{
+    int i, nalsize = 0;
+
+    if (*buf_index >= buf_size - p->nal_length_size) {
+        // the end of the buffer is reached, refill it
+        return AVERROR(EAGAIN);
+    }
+
+    for (i = 0; i < p->nal_length_size; i++)
+        nalsize = ((unsigned)nalsize << 8) | buf[(*buf_index)++];
+    if (nalsize <= 0 || nalsize > buf_size - *buf_index) {
+        av_log(logctx, AV_LOG_ERROR,
+               "AVC: nal size %d\n", nalsize);
+        return AVERROR_INVALIDDATA;
+    }
+    return nalsize;
 }
 
 /**
@@ -130,147 +256,239 @@ found:
  */
 static inline int parse_nal_units(AVCodecParserContext *s,
                                   AVCodecContext *avctx,
-                                  const uint8_t *buf, int buf_size)
+                                  const uint8_t * const buf, int buf_size)
 {
-    H264Context *h         = s->priv_data;
-    const uint8_t *buf_end = buf + buf_size;
+    H264ParseContext *p = s->priv_data;
+    H2645NAL nal = { NULL };
+    int buf_index, next_avc;
     unsigned int pps_id;
     unsigned int slice_type;
-    int state = -1;
-    const uint8_t *ptr;
+    int state = -1, got_reset = 0;
     int q264 = buf_size >=4 && !memcmp("Q264", buf, 4);
     int field_poc[2];
+    int ret;
 
     /* set some sane default values */
     s->pict_type         = AV_PICTURE_TYPE_I;
     s->key_frame         = 0;
     s->picture_structure = AV_PICTURE_STRUCTURE_UNKNOWN;
 
-    h->avctx                        = avctx;
-    h->sei_recovery_frame_cnt       = -1;
-    h->sei_dpb_output_delay         = 0;
-    h->sei_cpb_removal_delay        = -1;
-    h->sei_buffering_period_present = 0;
-    h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
+    ff_h264_sei_uninit(&p->sei);
+    p->sei.frame_packing.frame_packing_arrangement_cancel_flag = -1;
 
     if (!buf_size)
         return 0;
 
+    buf_index     = 0;
+    next_avc      = p->is_avc ? 0 : buf_size;
     for (;;) {
-        int src_length, dst_length, consumed, nalsize = 0;
-        if (h->is_avc) {
-            int i;
-            if (h->nal_length_size >= buf_end - buf) break;
-            nalsize = 0;
-            for (i = 0; i < h->nal_length_size; i++)
-                nalsize = (nalsize << 8) | *buf++;
-            if (nalsize <= 0 || nalsize > buf_end - buf) {
-                av_log(h->avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
+        const SPS *sps;
+        int src_length, consumed, nalsize = 0;
+
+        if (buf_index >= next_avc) {
+            nalsize = get_avc_nalsize(p, buf, buf_size, &buf_index, avctx);
+            if (nalsize < 0)
                 break;
-            }
-            src_length = nalsize;
+            next_avc = buf_index + nalsize;
         } else {
-        buf = avpriv_find_start_code(buf, buf_end, &state);
-        if (buf >= buf_end)
-            break;
-        --buf;
-        src_length = buf_end - buf;
+            buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
+            if (buf_index >= buf_size)
+                break;
+            if (buf_index >= next_avc)
+                continue;
         }
+        src_length = next_avc - buf_index;
+
+        state = buf[buf_index];
         switch (state & 0x1f) {
         case NAL_SLICE:
         case NAL_IDR_SLICE:
             // Do not walk the whole buffer just to decode slice header
-            if (src_length > 20)
-                src_length = 20;
+            if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
+                /* IDR or disposable slice
+                 * No need to decode many bytes because MMCOs shall not be present. */
+                if (src_length > 60)
+                    src_length = 60;
+            } else {
+                /* To decode up to MMCOs */
+                if (src_length > 1000)
+                    src_length = 1000;
+            }
             break;
         }
-        ptr = ff_h264_decode_nal(h, buf, &dst_length, &consumed, src_length);
-        if (ptr == NULL || dst_length < 0)
+        consumed = ff_h2645_extract_rbsp(buf + buf_index, src_length, &nal);
+        if (consumed < 0)
             break;
 
-        init_get_bits(&h->gb, ptr, 8 * dst_length);
-        switch (h->nal_unit_type) {
+        buf_index += consumed;
+
+        ret = init_get_bits8(&nal.gb, nal.data, nal.size);
+        if (ret < 0)
+            goto fail;
+        get_bits1(&nal.gb);
+        nal.ref_idc = get_bits(&nal.gb, 2);
+        nal.type    = get_bits(&nal.gb, 5);
+
+        switch (nal.type) {
         case NAL_SPS:
-            ff_h264_decode_seq_parameter_set(h);
+            ff_h264_decode_seq_parameter_set(&nal.gb, avctx, &p->ps, 0);
             break;
         case NAL_PPS:
-            ff_h264_decode_picture_parameter_set(h, h->gb.size_in_bits);
+            ff_h264_decode_picture_parameter_set(&nal.gb, avctx, &p->ps,
+                                                 nal.size_bits);
             break;
         case NAL_SEI:
-            ff_h264_decode_sei(h);
+            ff_h264_sei_decode(&p->sei, &nal.gb, &p->ps, avctx);
             break;
         case NAL_IDR_SLICE:
             s->key_frame = 1;
 
-            h->prev_frame_num        = 0;
-            h->prev_frame_num_offset = 0;
-            h->prev_poc_msb          =
-            h->prev_poc_lsb          = 0;
+            p->poc.prev_frame_num        = 0;
+            p->poc.prev_frame_num_offset = 0;
+            p->poc.prev_poc_msb          =
+            p->poc.prev_poc_lsb          = 0;
         /* fall through */
         case NAL_SLICE:
-            get_ue_golomb_long(&h->gb);  // skip first_mb_in_slice
-            slice_type   = get_ue_golomb_31(&h->gb);
-            s->pict_type = golomb_to_pict_type[slice_type % 5];
-            if (h->sei_recovery_frame_cnt >= 0) {
+            get_ue_golomb_long(&nal.gb);  // skip first_mb_in_slice
+            slice_type   = get_ue_golomb_31(&nal.gb);
+            s->pict_type = ff_h264_golomb_to_pict_type[slice_type % 5];
+            if (p->sei.recovery_point.recovery_frame_cnt >= 0) {
                 /* key frame, since recovery_frame_cnt is set */
                 s->key_frame = 1;
             }
-            pps_id = get_ue_golomb(&h->gb);
+            pps_id = get_ue_golomb(&nal.gb);
             if (pps_id >= MAX_PPS_COUNT) {
-                av_log(h->avctx, AV_LOG_ERROR,
-                       "pps_id out of range\n");
-                return -1;
+                av_log(avctx, AV_LOG_ERROR,
+                       "pps_id %u out of range\n", pps_id);
+                goto fail;
             }
-            if (!h->pps_buffers[pps_id]) {
-                av_log(h->avctx, AV_LOG_ERROR,
-                       "non-existing PPS referenced\n");
-                return -1;
+            if (!p->ps.pps_list[pps_id]) {
+                av_log(avctx, AV_LOG_ERROR,
+                       "non-existing PPS %u referenced\n", pps_id);
+                goto fail;
             }
-            h->pps = *h->pps_buffers[pps_id];
-            if (!h->sps_buffers[h->pps.sps_id]) {
-                av_log(h->avctx, AV_LOG_ERROR,
-                       "non-existing SPS referenced\n");
-                return -1;
+
+            av_buffer_unref(&p->ps.pps_ref);
+            av_buffer_unref(&p->ps.sps_ref);
+            p->ps.pps = NULL;
+            p->ps.sps = NULL;
+            p->ps.pps_ref = av_buffer_ref(p->ps.pps_list[pps_id]);
+            if (!p->ps.pps_ref)
+                goto fail;
+            p->ps.pps = (const PPS*)p->ps.pps_ref->data;
+
+            if (!p->ps.sps_list[p->ps.pps->sps_id]) {
+                av_log(avctx, AV_LOG_ERROR,
+                       "non-existing SPS %u referenced\n", p->ps.pps->sps_id);
+                goto fail;
             }
-            h->sps       = *h->sps_buffers[h->pps.sps_id];
-            h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
 
-            avctx->profile = ff_h264_get_profile(&h->sps);
-            avctx->level   = h->sps.level_idc;
+            p->ps.sps_ref = av_buffer_ref(p->ps.sps_list[p->ps.pps->sps_id]);
+            if (!p->ps.sps_ref)
+                goto fail;
+            p->ps.sps = (const SPS*)p->ps.sps_ref->data;
+
+            sps = p->ps.sps;
+
+            // heuristic to detect non marked keyframes
+            if (p->ps.sps->ref_frame_count <= 1 && p->ps.pps->ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
+                s->key_frame = 1;
+
+            p->poc.frame_num = get_bits(&nal.gb, sps->log2_max_frame_num);
+
+            s->coded_width  = 16 * sps->mb_width;
+            s->coded_height = 16 * sps->mb_height;
+            s->width        = s->coded_width  - (sps->crop_right + sps->crop_left);
+            s->height       = s->coded_height - (sps->crop_top   + sps->crop_bottom);
+            if (s->width <= 0 || s->height <= 0) {
+                s->width  = s->coded_width;
+                s->height = s->coded_height;
+            }
 
-            if (h->sps.frame_mbs_only_flag) {
-                h->picture_structure = PICT_FRAME;
+            switch (sps->bit_depth_luma) {
+            case 9:
+                if (sps->chroma_format_idc == 3)      s->format = AV_PIX_FMT_YUV444P9;
+                else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P9;
+                else                                  s->format = AV_PIX_FMT_YUV420P9;
+                break;
+            case 10:
+                if (sps->chroma_format_idc == 3)      s->format = AV_PIX_FMT_YUV444P10;
+                else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P10;
+                else                                  s->format = AV_PIX_FMT_YUV420P10;
+                break;
+            case 8:
+                if (sps->chroma_format_idc == 3)      s->format = AV_PIX_FMT_YUV444P;
+                else if (sps->chroma_format_idc == 2) s->format = AV_PIX_FMT_YUV422P;
+                else                                  s->format = AV_PIX_FMT_YUV420P;
+                break;
+            default:
+                s->format = AV_PIX_FMT_NONE;
+            }
+
+            avctx->profile = ff_h264_get_profile(sps);
+            avctx->level   = sps->level_idc;
+
+            if (sps->frame_mbs_only_flag) {
+                p->picture_structure = PICT_FRAME;
             } else {
-                if (get_bits1(&h->gb)) { // field_pic_flag
-                    h->picture_structure = PICT_TOP_FIELD + get_bits1(&h->gb); // bottom_field_flag
+                if (get_bits1(&nal.gb)) { // field_pic_flag
+                    p->picture_structure = PICT_TOP_FIELD + get_bits1(&nal.gb); // bottom_field_flag
                 } else {
-                    h->picture_structure = PICT_FRAME;
+                    p->picture_structure = PICT_FRAME;
                 }
             }
 
-            if (h->nal_unit_type == NAL_IDR_SLICE)
-                get_ue_golomb(&h->gb); /* idr_pic_id */
-            if (h->sps.poc_type == 0) {
-                h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
+            if (nal.type == NAL_IDR_SLICE)
+                get_ue_golomb_long(&nal.gb); /* idr_pic_id */
+            if (sps->poc_type == 0) {
+                p->poc.poc_lsb = get_bits(&nal.gb, sps->log2_max_poc_lsb);
 
-                if (h->pps.pic_order_present == 1 &&
-                    h->picture_structure == PICT_FRAME)
-                    h->delta_poc_bottom = get_se_golomb(&h->gb);
+                if (p->ps.pps->pic_order_present == 1 &&
+                    p->picture_structure == PICT_FRAME)
+                    p->poc.delta_poc_bottom = get_se_golomb(&nal.gb);
             }
 
-            if (h->sps.poc_type == 1 &&
-                !h->sps.delta_pic_order_always_zero_flag) {
-                h->delta_poc[0] = get_se_golomb(&h->gb);
+            if (sps->poc_type == 1 &&
+                !sps->delta_pic_order_always_zero_flag) {
+                p->poc.delta_poc[0] = get_se_golomb(&nal.gb);
 
-                if (h->pps.pic_order_present == 1 &&
-                    h->picture_structure == PICT_FRAME)
-                    h->delta_poc[1] = get_se_golomb(&h->gb);
+                if (p->ps.pps->pic_order_present == 1 &&
+                    p->picture_structure == PICT_FRAME)
+                    p->poc.delta_poc[1] = get_se_golomb(&nal.gb);
             }
 
-            ff_init_poc(h, field_poc, &s->output_picture_number);
+            /* Decode POC of this picture.
+             * The prev_ values needed for decoding POC of the next picture are not set here. */
+            field_poc[0] = field_poc[1] = INT_MAX;
+            ff_h264_init_poc(field_poc, &s->output_picture_number, sps,
+                             &p->poc, p->picture_structure, nal.ref_idc);
+
+            /* Continue parsing to check if MMCO_RESET is present.
+             * FIXME: MMCO_RESET could appear in non-first slice.
+             *        Maybe, we should parse all undisposable non-IDR slice of this
+             *        picture until encountering MMCO_RESET in a slice of it. */
+            if (nal.ref_idc && nal.type != NAL_IDR_SLICE) {
+                got_reset = scan_mmco_reset(s, &nal.gb, avctx);
+                if (got_reset < 0)
+                    goto fail;
+            }
 
-            if (h->sps.pic_struct_present_flag) {
-                switch (h->sei_pic_struct) {
+            /* Set up the prev_ values for decoding POC of the next picture. */
+            p->poc.prev_frame_num        = got_reset ? 0 : p->poc.frame_num;
+            p->poc.prev_frame_num_offset = got_reset ? 0 : p->poc.frame_num_offset;
+            if (nal.ref_idc != 0) {
+                if (!got_reset) {
+                    p->poc.prev_poc_msb = p->poc.poc_msb;
+                    p->poc.prev_poc_lsb = p->poc.poc_lsb;
+                } else {
+                    p->poc.prev_poc_msb = 0;
+                    p->poc.prev_poc_lsb =
+                        p->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
+                }
+            }
+
+            if (sps->pic_struct_present_flag) {
+                switch (p->sei.picture_timing.pic_struct) {
                 case SEI_PIC_STRUCT_TOP_FIELD:
                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
                     s->repeat_pict = 0;
@@ -291,17 +509,17 @@ static inline int parse_nal_units(AVCodecParserContext *s,
                     s->repeat_pict = 5;
                     break;
                 default:
-                    s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
+                    s->repeat_pict = p->picture_structure == PICT_FRAME ? 1 : 0;
                     break;
                 }
             } else {
-                s->repeat_pict = h->picture_structure == PICT_FRAME ? 1 : 0;
+                s->repeat_pict = p->picture_structure == PICT_FRAME ? 1 : 0;
             }
 
-            if (h->picture_structure == PICT_FRAME) {
+            if (p->picture_structure == PICT_FRAME) {
                 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
-                if (h->sps.pic_struct_present_flag) {
-                    switch (h->sei_pic_struct) {
+                if (sps->pic_struct_present_flag) {
+                    switch (p->sei.picture_timing.pic_struct) {
                     case SEI_PIC_STRUCT_TOP_BOTTOM:
                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
                         s->field_order = AV_FIELD_TT;
@@ -323,21 +541,25 @@ static inline int parse_nal_units(AVCodecParserContext *s,
                         s->field_order = AV_FIELD_PROGRESSIVE;
                 }
             } else {
-                if (h->picture_structure == PICT_TOP_FIELD)
+                if (p->picture_structure == PICT_TOP_FIELD)
                     s->picture_structure = AV_PICTURE_STRUCTURE_TOP_FIELD;
                 else
                     s->picture_structure = AV_PICTURE_STRUCTURE_BOTTOM_FIELD;
                 s->field_order = AV_FIELD_UNKNOWN;
             }
 
+            av_freep(&nal.rbsp_buffer);
             return 0; /* no need to evaluate the rest */
         }
-        buf += h->is_avc ? nalsize : consumed;
     }
-    if (q264)
+    if (q264) {
+        av_freep(&nal.rbsp_buffer);
         return 0;
+    }
     /* didn't find a picture! */
-    av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
+    av_log(avctx, AV_LOG_ERROR, "missing picture in access unit with size %d\n", buf_size);
+fail:
+    av_freep(&nal.rbsp_buffer);
     return -1;
 }
 
@@ -346,28 +568,23 @@ static int h264_parse(AVCodecParserContext *s,
                       const uint8_t **poutbuf, int *poutbuf_size,
                       const uint8_t *buf, int buf_size)
 {
-    H264Context *h   = s->priv_data;
-    ParseContext *pc = &h->parse_context;
+    H264ParseContext *p = s->priv_data;
+    ParseContext *pc = &p->pc;
     int next;
 
-    if (!h->got_first) {
-        h->got_first = 1;
+    if (!p->got_first) {
+        p->got_first = 1;
         if (avctx->extradata_size) {
-            h->avctx = avctx;
-            // must be done like in decoder, otherwise opening the parser,
-            // letting it create extradata and then closing and opening again
-            // will cause has_b_frames to be always set.
-            // Note that estimate_timings_from_pts does exactly this.
-            if (!avctx->has_b_frames)
-                h->low_delay = 1;
-            ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size);
+            ff_h264_decode_extradata(avctx->extradata, avctx->extradata_size,
+                                     &p->ps, &p->is_avc, &p->nal_length_size,
+                                     avctx->err_recognition, avctx);
         }
     }
 
     if (s->flags & PARSER_FLAG_COMPLETE_FRAMES) {
         next = buf_size;
     } else {
-        next = h264_find_frame_end(h, buf, buf_size);
+        next = h264_find_frame_end(p, buf, buf_size, avctx);
 
         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
             *poutbuf      = NULL;
@@ -377,16 +594,18 @@ static int h264_parse(AVCodecParserContext *s,
 
         if (next < 0 && next != END_NOT_FOUND) {
             av_assert1(pc->last_index + next >= 0);
-            h264_find_frame_end(h, &pc->buffer[pc->last_index + next], -next); // update state
+            h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next, avctx); // update state
         }
     }
 
     parse_nal_units(s, avctx, buf, buf_size);
 
-    if (h->sei_cpb_removal_delay >= 0) {
-        s->dts_sync_point    = h->sei_buffering_period_present;
-        s->dts_ref_dts_delta = h->sei_cpb_removal_delay;
-        s->pts_dts_delta     = h->sei_dpb_output_delay;
+    if (avctx->framerate.num)
+        avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
+    if (p->sei.picture_timing.cpb_removal_delay >= 0) {
+        s->dts_sync_point    = p->sei.buffering_period.present;
+        s->dts_ref_dts_delta = p->sei.picture_timing.cpb_removal_delay;
+        s->pts_dts_delta     = p->sei.picture_timing.dpb_output_delay;
     } else {
         s->dts_sync_point    = INT_MIN;
         s->dts_ref_dts_delta = INT_MIN;
@@ -405,55 +624,64 @@ static int h264_parse(AVCodecParserContext *s,
 static int h264_split(AVCodecContext *avctx,
                       const uint8_t *buf, int buf_size)
 {
-    int i;
     uint32_t state = -1;
     int has_sps    = 0;
+    int has_pps    = 0;
+    const uint8_t *ptr = buf, *end = buf + buf_size;
+    int nalu_type;
 
-    for (i = 0; i <= buf_size; i++) {
-        if ((state & 0xFFFFFF1F) == 0x107)
+    while (ptr < end) {
+        ptr = avpriv_find_start_code(ptr, end, &state);
+        if ((state & 0xFFFFFF00) != 0x100)
+            break;
+        nalu_type = state & 0x1F;
+        if (nalu_type == NAL_SPS) {
             has_sps = 1;
-        /*  if ((state&0xFFFFFF1F) == 0x101 ||
-         *     (state&0xFFFFFF1F) == 0x102 ||
-         *     (state&0xFFFFFF1F) == 0x105) {
+        } else if (nalu_type == NAL_PPS)
+            has_pps = 1;
+        /* else if (nalu_type == 0x01 ||
+         *     nalu_type == 0x02 ||
+         *     nalu_type == 0x05) {
          *  }
          */
-        if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x107 &&
-            (state & 0xFFFFFF1F) != 0x108 && (state & 0xFFFFFF1F) != 0x109) {
+        else if ((nalu_type != NAL_SEI || has_pps) &&
+                  nalu_type != NAL_AUD && nalu_type != NAL_SPS_EXT &&
+                  nalu_type != 0x0f) {
             if (has_sps) {
-                while (i > 4 && buf[i - 5] == 0)
-                    i--;
-                return i - 4;
+                while (ptr - 4 > buf && ptr[-5] == 0)
+                    ptr--;
+                return ptr - 4 - buf;
             }
         }
-        if (i < buf_size)
-            state = (state << 8) | buf[i];
     }
+
     return 0;
 }
 
-static void close(AVCodecParserContext *s)
+static void h264_close(AVCodecParserContext *s)
 {
-    H264Context *h   = s->priv_data;
-    ParseContext *pc = &h->parse_context;
+    H264ParseContext *p = s->priv_data;
+    ParseContext *pc = &p->pc;
+
+    av_freep(&pc->buffer);
 
-    av_free(pc->buffer);
-    ff_h264_free_context(h);
+    ff_h264_sei_uninit(&p->sei);
+    ff_h264_ps_uninit(&p->ps);
 }
 
 static av_cold int init(AVCodecParserContext *s)
 {
-    H264Context *h = s->priv_data;
-    h->thread_context[0]   = h;
-    h->slice_context_count = 1;
-    ff_h264dsp_init(&h->h264dsp, 8, 1);
+    H264ParseContext *p = s->priv_data;
+
+    ff_h264dsp_init(&p->h264dsp, 8, 1);
     return 0;
 }
 
 AVCodecParser ff_h264_parser = {
     .codec_ids      = { AV_CODEC_ID_H264 },
-    .priv_data_size = sizeof(H264Context),
+    .priv_data_size = sizeof(H264ParseContext),
     .parser_init    = init,
     .parser_parse   = h264_parse,
-    .parser_close   = close,
+    .parser_close   = h264_close,
     .split          = h264_split,
 };