avcodec/dump_extradata_bsf: don't allocate a new AVPacket per filtered packet
[ffmpeg.git] / libavcodec / h264_parser.c
index 8eda6b8..5f9a9c4 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 "avcodec.h"
+#include "get_bits.h"
 #include "golomb.h"
+#include "h264.h"
+#include "h264_sei.h"
+#include "h264_ps.h"
+#include "h264data.h"
 #include "internal.h"
 #include "mpegutils.h"
+#include "parser.h"
 
 typedef struct H264ParseContext {
-    H264Context h;
     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;
+    int64_t reference_dts;
+    int last_frame_num, last_picture_structure;
 } H264ParseContext;
 
 
 static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
-                               int buf_size)
+                               int buf_size, void *logctx)
 {
-    H264Context *h = &p->h;
     int i, j;
     uint32_t state;
     ParseContext *pc = &p->pc;
 
-    int next_avc= h->is_avc ? 0 : buf_size;
+    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;
@@ -73,7 +96,7 @@ static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
         }
 
         if (state == 7) {
-            i += h->h264dsp.startcode_find_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) {
@@ -85,83 +108,92 @@ static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
                 state >>= 1;           // 2->1, 1->0, 0->0
         } else if (state <= 5) {
             int nalu_type = buf[i] & 0x1F;
-            if (nalu_type == NAL_SEI || nalu_type == NAL_SPS ||
-                nalu_type == NAL_PPS || nalu_type == NAL_AUD) {
+            if (nalu_type == H264_NAL_SEI || nalu_type == H264_NAL_SPS ||
+                nalu_type == H264_NAL_PPS || nalu_type == H264_NAL_AUD) {
                 if (pc->frame_start_found) {
                     i++;
                     goto found;
                 }
-            } else if (nalu_type == NAL_SLICE || nalu_type == NAL_DPA ||
-                       nalu_type == NAL_IDR_SLICE) {
+            } else if (nalu_type == H264_NAL_SLICE || nalu_type == H264_NAL_DPA ||
+                       nalu_type == H264_NAL_IDR_SLICE) {
                 state += 8;
                 continue;
             }
             state = 7;
         } else {
-            h->parse_history[h->parse_history_count++]= buf[i];
-            if (h->parse_history_count>5) {
-                unsigned int mb, last_mb= h->parse_last_mb;
-                GetBitContext gb;
-
-                init_get_bits(&gb, h->parse_history, 8*h->parse_history_count);
-                h->parse_history_count=0;
-                mb= get_ue_golomb_long(&gb);
-                h->parse_last_mb= mb;
+            unsigned int mb, last_mb = p->parse_last_mb;
+            GetBitContext gb;
+            p->parse_history[p->parse_history_count++] = buf[i];
+
+            init_get_bits(&gb, p->parse_history, 8*p->parse_history_count);
+            mb= get_ue_golomb_long(&gb);
+            if (get_bits_left(&gb) > 0 || p->parse_history_count > 5) {
+                p->parse_last_mb = mb;
                 if (pc->frame_start_found) {
-                    if (mb <= last_mb)
+                    if (mb <= last_mb) {
+                        i -= p->parse_history_count - 1;
+                        p->parse_history_count = 0;
                         goto found;
+                    }
                 } else
                     pc->frame_start_found = 1;
+                p->parse_history_count = 0;
                 state = 7;
             }
         }
     }
     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) - 5 * (state > 7);
+    return i - (state & 5);
 }
 
-static int scan_mmco_reset(AVCodecParserContext *s)
+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;
-    H264Context      *h = &p->h;
+    int list_count, ref_count[2];
 
-    h->slice_type_nos = s->pict_type & 3;
 
-    if (h->pps.redundant_pic_cnt_present)
-        get_ue_golomb(&h->gb); // redundant_pic_count
+    if (p->ps.pps->redundant_pic_cnt_present)
+        get_ue_golomb(gb); // redundant_pic_count
 
-    if (ff_set_ref_count(h) < 0)
+    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 (h->slice_type_nos != AV_PICTURE_TYPE_I) {
+    if (slice_type_nos != AV_PICTURE_TYPE_I) {
         int list;
-        for (list = 0; list < h->list_count; list++) {
-            if (get_bits1(&h->gb)) {
+        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(&h->gb);
+                    unsigned int reordering_of_pic_nums_idc = get_ue_golomb_31(gb);
 
                     if (reordering_of_pic_nums_idc < 3)
-                        get_ue_golomb(&h->gb);
+                        get_ue_golomb_long(gb);
                     else if (reordering_of_pic_nums_idc > 3) {
-                        av_log(h->avctx, AV_LOG_ERROR,
+                        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 >= h->ref_count[list]) {
-                        av_log(h->avctx, AV_LOG_ERROR,
+                    if (index >= ref_count[list]) {
+                        av_log(logctx, AV_LOG_ERROR,
                                "reference count %d overflow\n", index);
                         return AVERROR_INVALIDDATA;
                     }
@@ -170,16 +202,17 @@ static int scan_mmco_reset(AVCodecParserContext *s)
         }
     }
 
-    if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
-        (h->pps.weighted_bipred_idc == 1 && h->slice_type_nos == AV_PICTURE_TYPE_B))
-        ff_pred_weight_table(h);
+    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, p->picture_structure, logctx);
 
-    if (get_bits1(&h->gb)) { // adaptive_ref_pic_marking_mode_flag
+    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(&h->gb);
+            MMCOOpcode opcode = get_ue_golomb_31(gb);
             if (opcode > (unsigned) MMCO_LONG) {
-                av_log(h->avctx, AV_LOG_ERROR,
+                av_log(logctx, AV_LOG_ERROR,
                        "illegal memory management control operation %d\n",
                        opcode);
                 return AVERROR_INVALIDDATA;
@@ -190,10 +223,10 @@ static int scan_mmco_reset(AVCodecParserContext *s)
                 return 1;
 
             if (opcode == MMCO_SHORT2UNUSED || opcode == MMCO_SHORT2LONG)
-                get_ue_golomb(&h->gb);
+                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(&h->gb);
+                get_ue_golomb_31(gb);
         }
     }
 
@@ -213,34 +246,39 @@ static inline int parse_nal_units(AVCodecParserContext *s,
                                   const uint8_t * const buf, int buf_size)
 {
     H264ParseContext *p = s->priv_data;
-    H264Context      *h = &p->h;
+    H2645RBSP rbsp = { NULL };
+    H2645NAL nal = { NULL };
     int buf_index, next_avc;
     unsigned int pps_id;
     unsigned int slice_type;
     int state = -1, got_reset = 0;
-    const uint8_t *ptr;
     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;
-    ff_h264_reset_sei(h);
-    h->sei_fpa.frame_packing_arrangement_cancel_flag = -1;
+    ff_h264_sei_uninit(&p->sei);
+    p->sei.frame_packing.arrangement_cancel_flag = -1;
 
     if (!buf_size)
         return 0;
 
+    av_fast_padded_malloc(&rbsp.rbsp_buffer, &rbsp.rbsp_buffer_alloc_size, buf_size);
+    if (!rbsp.rbsp_buffer)
+        return AVERROR(ENOMEM);
+
     buf_index     = 0;
-    next_avc      = h->is_avc ? 0 : buf_size;
+    next_avc      = p->is_avc ? 0 : buf_size;
     for (;;) {
-        int src_length, dst_length, consumed, nalsize = 0;
+        const SPS *sps;
+        int src_length, consumed, nalsize = 0;
 
         if (buf_index >= next_avc) {
-            nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
+            nalsize = get_nalsize(p->nal_length_size, buf, buf_size, &buf_index, avctx);
             if (nalsize < 0)
                 break;
             next_avc = buf_index + nalsize;
@@ -255,10 +293,10 @@ static inline int parse_nal_units(AVCodecParserContext *s,
 
         state = buf[buf_index];
         switch (state & 0x1f) {
-        case NAL_SLICE:
-        case NAL_IDR_SLICE:
+        case H264_NAL_SLICE:
+        case H264_NAL_IDR_SLICE:
             // Do not walk the whole buffer just to decode slice header
-            if ((state & 0x1f) == NAL_IDR_SLICE || ((state >> 5) & 0x3) == 0) {
+            if ((state & 0x1f) == H264_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)
@@ -270,192 +308,218 @@ static inline int parse_nal_units(AVCodecParserContext *s,
             }
             break;
         }
-        ptr = ff_h264_decode_nal(h, buf + buf_index, &dst_length,
-                                 &consumed, src_length);
-        if (!ptr || dst_length < 0)
+        consumed = ff_h2645_extract_rbsp(buf + buf_index, src_length, &rbsp, &nal, 1);
+        if (consumed < 0)
             break;
 
         buf_index += consumed;
 
-        init_get_bits(&h->gb, ptr, 8 * dst_length);
-        switch (h->nal_unit_type) {
-        case NAL_SPS:
-            ff_h264_decode_seq_parameter_set(h);
+        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 H264_NAL_SPS:
+            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);
+        case H264_NAL_PPS:
+            ff_h264_decode_picture_parameter_set(&nal.gb, avctx, &p->ps,
+                                                 nal.size_bits);
             break;
-        case NAL_SEI:
-            ff_h264_decode_sei(h);
+        case H264_NAL_SEI:
+            ff_h264_sei_decode(&p->sei, &nal.gb, &p->ps, avctx);
             break;
-        case NAL_IDR_SLICE:
+        case H264_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) {
+        case H264_NAL_SLICE:
+            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,
+                av_log(avctx, AV_LOG_ERROR,
                        "pps_id %u out of range\n", pps_id);
-                return -1;
+                goto fail;
             }
-            if (!h->pps_buffers[pps_id]) {
-                av_log(h->avctx, AV_LOG_ERROR,
+            if (!p->ps.pps_list[pps_id]) {
+                av_log(avctx, AV_LOG_ERROR,
                        "non-existing PPS %u referenced\n", pps_id);
-                return -1;
+                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 %u referenced\n", h->pps.sps_id);
-                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);
 
-            if(h->sps.ref_frame_count <= 1 && h->pps.ref_count[0] <= 1 && s->pict_type == AV_PICTURE_TYPE_I)
+            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;
 
-            s->coded_width  = 16 * h->sps.mb_width;
-            s->coded_height = 16 * h->sps.mb_height;
-            s->width        = s->coded_width  - (h->sps.crop_right + h->sps.crop_left);
-            s->height       = s->coded_height - (h->sps.crop_top   + h->sps.crop_bottom);
+            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;
             }
 
-            switch (h->sps.bit_depth_luma) {
+            switch (sps->bit_depth_luma) {
             case 9:
-                if (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P9;
-                else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P9;
-                else                   s->format = AV_PIX_FMT_YUV420P9;
+                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 (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P10;
-                else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P10;
-                else                   s->format = AV_PIX_FMT_YUV420P10;
+                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 (CHROMA444(h))      s->format = AV_PIX_FMT_YUV444P;
-                else if (CHROMA422(h)) s->format = AV_PIX_FMT_YUV422P;
-                else                   s->format = AV_PIX_FMT_YUV420P;
+                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(&h->sps);
-            avctx->level   = h->sps.level_idc;
+            avctx->profile = ff_h264_get_profile(sps);
+            avctx->level   = sps->level_idc;
 
-            if (h->sps.frame_mbs_only_flag) {
-                h->picture_structure = PICT_FRAME;
+            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 == H264_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);
             }
 
             /* 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_init_poc(h, field_poc, &s->output_picture_number);
+            ret = ff_h264_init_poc(field_poc, &s->output_picture_number, sps,
+                             &p->poc, p->picture_structure, nal.ref_idc);
+            if (ret < 0)
+                goto fail;
 
             /* 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 (h->nal_ref_idc && h->nal_unit_type != NAL_IDR_SLICE) {
-                got_reset = scan_mmco_reset(s);
+            if (nal.ref_idc && nal.type != H264_NAL_IDR_SLICE) {
+                got_reset = scan_mmco_reset(s, &nal.gb, avctx);
                 if (got_reset < 0)
-                    return got_reset;
+                    goto fail;
             }
 
             /* Set up the prev_ values for decoding POC of the next picture. */
-            h->prev_frame_num        = got_reset ? 0 : h->frame_num;
-            h->prev_frame_num_offset = got_reset ? 0 : h->frame_num_offset;
-            if (h->nal_ref_idc != 0) {
+            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) {
-                    h->prev_poc_msb = h->poc_msb;
-                    h->prev_poc_lsb = h->poc_lsb;
+                    p->poc.prev_poc_msb = p->poc.poc_msb;
+                    p->poc.prev_poc_lsb = p->poc.poc_lsb;
                 } else {
-                    h->prev_poc_msb = 0;
-                    h->prev_poc_lsb =
-                        h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
+                    p->poc.prev_poc_msb = 0;
+                    p->poc.prev_poc_lsb =
+                        p->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
                 }
             }
 
-            if (h->sps.pic_struct_present_flag) {
-                switch (h->sei_pic_struct) {
-                case SEI_PIC_STRUCT_TOP_FIELD:
-                case SEI_PIC_STRUCT_BOTTOM_FIELD:
+            if (sps->pic_struct_present_flag && p->sei.picture_timing.present) {
+                switch (p->sei.picture_timing.pic_struct) {
+                case H264_SEI_PIC_STRUCT_TOP_FIELD:
+                case H264_SEI_PIC_STRUCT_BOTTOM_FIELD:
                     s->repeat_pict = 0;
                     break;
-                case SEI_PIC_STRUCT_FRAME:
-                case SEI_PIC_STRUCT_TOP_BOTTOM:
-                case SEI_PIC_STRUCT_BOTTOM_TOP:
+                case H264_SEI_PIC_STRUCT_FRAME:
+                case H264_SEI_PIC_STRUCT_TOP_BOTTOM:
+                case H264_SEI_PIC_STRUCT_BOTTOM_TOP:
                     s->repeat_pict = 1;
                     break;
-                case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
-                case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
+                case H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
+                case H264_SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
                     s->repeat_pict = 2;
                     break;
-                case SEI_PIC_STRUCT_FRAME_DOUBLING:
+                case H264_SEI_PIC_STRUCT_FRAME_DOUBLING:
                     s->repeat_pict = 3;
                     break;
-                case SEI_PIC_STRUCT_FRAME_TRIPLING:
+                case H264_SEI_PIC_STRUCT_FRAME_TRIPLING:
                     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) {
-                    case SEI_PIC_STRUCT_TOP_BOTTOM:
-                    case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
+                if (sps->pic_struct_present_flag && p->sei.picture_timing.present) {
+                    switch (p->sei.picture_timing.pic_struct) {
+                    case H264_SEI_PIC_STRUCT_TOP_BOTTOM:
+                    case H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
                         s->field_order = AV_FIELD_TT;
                         break;
-                    case SEI_PIC_STRUCT_BOTTOM_TOP:
-                    case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
+                    case H264_SEI_PIC_STRUCT_BOTTOM_TOP:
+                    case H264_SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
                         s->field_order = AV_FIELD_BB;
                         break;
                     default:
@@ -471,20 +535,37 @@ 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;
+                if (p->poc.frame_num == p->last_frame_num &&
+                    p->last_picture_structure != AV_PICTURE_STRUCTURE_UNKNOWN &&
+                    p->last_picture_structure != AV_PICTURE_STRUCTURE_FRAME &&
+                    p->last_picture_structure != s->picture_structure) {
+                    if (p->last_picture_structure == AV_PICTURE_STRUCTURE_TOP_FIELD)
+                        s->field_order = AV_FIELD_TT;
+                    else
+                        s->field_order = AV_FIELD_BB;
+                } else {
+                    s->field_order = AV_FIELD_UNKNOWN;
+                }
+                p->last_picture_structure = s->picture_structure;
+                p->last_frame_num = p->poc.frame_num;
             }
 
+            av_freep(&rbsp.rbsp_buffer);
             return 0; /* no need to evaluate the rest */
         }
     }
-    if (q264)
+    if (q264) {
+        av_freep(&rbsp.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(&rbsp.rbsp_buffer);
     return -1;
 }
 
@@ -494,28 +575,22 @@ static int h264_parse(AVCodecParserContext *s,
                       const uint8_t *buf, int buf_size)
 {
     H264ParseContext *p = s->priv_data;
-    H264Context      *h = &p->h;
     ParseContext *pc = &p->pc;
     int next;
 
     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(p, 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;
@@ -525,7 +600,7 @@ static int h264_parse(AVCodecParserContext *s,
 
         if (next < 0 && next != END_NOT_FOUND) {
             av_assert1(pc->last_index + next >= 0);
-            h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next); // update state
+            h264_find_frame_end(p, &pc->buffer[pc->last_index + next], -next, avctx); // update state
         }
     }
 
@@ -533,10 +608,10 @@ static int h264_parse(AVCodecParserContext *s,
 
     if (avctx->framerate.num)
         avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
-    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 (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;
@@ -547,6 +622,26 @@ static int h264_parse(AVCodecParserContext *s,
         s->flags &= PARSER_FLAG_COMPLETE_FRAMES;
     }
 
+    if (s->dts_sync_point >= 0) {
+        int64_t den = avctx->time_base.den * (int64_t)avctx->pkt_timebase.num;
+        if (den > 0) {
+            int64_t num = avctx->time_base.num * (int64_t)avctx->pkt_timebase.den;
+            if (s->dts != AV_NOPTS_VALUE) {
+                // got DTS from the stream, update reference timestamp
+                p->reference_dts = s->dts - av_rescale(s->dts_ref_dts_delta, num, den);
+            } else if (p->reference_dts != AV_NOPTS_VALUE) {
+                // compute DTS based on reference timestamp
+                s->dts = p->reference_dts + av_rescale(s->dts_ref_dts_delta, num, den);
+            }
+
+            if (p->reference_dts != AV_NOPTS_VALUE && s->pts == AV_NOPTS_VALUE)
+                s->pts = s->dts + av_rescale(s->pts_dts_delta, num, den);
+
+            if (s->dts_sync_point > 0)
+                p->reference_dts = s->dts; // new reference
+        }
+    }
+
     *poutbuf      = buf;
     *poutbuf_size = buf_size;
     return next;
@@ -566,17 +661,17 @@ static int h264_split(AVCodecContext *avctx,
         if ((state & 0xFFFFFF00) != 0x100)
             break;
         nalu_type = state & 0x1F;
-        if (nalu_type == NAL_SPS) {
+        if (nalu_type == H264_NAL_SPS) {
             has_sps = 1;
-        } else if (nalu_type == NAL_PPS)
+        } else if (nalu_type == H264_NAL_PPS)
             has_pps = 1;
         /* else if (nalu_type == 0x01 ||
          *     nalu_type == 0x02 ||
          *     nalu_type == 0x05) {
          *  }
          */
-        else if ((nalu_type != NAL_SEI || has_pps) &&
-                  nalu_type != NAL_AUD && nalu_type != NAL_SPS_EXT &&
+        else if ((nalu_type != H264_NAL_SEI || has_pps) &&
+                  nalu_type != H264_NAL_AUD && nalu_type != H264_NAL_SPS_EXT &&
                   nalu_type != 0x0f) {
             if (has_sps) {
                 while (ptr - 4 > buf && ptr[-5] == 0)
@@ -592,20 +687,21 @@ static int h264_split(AVCodecContext *avctx,
 static void h264_close(AVCodecParserContext *s)
 {
     H264ParseContext *p = s->priv_data;
-    H264Context      *h = &p->h;
     ParseContext *pc = &p->pc;
 
     av_freep(&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)
 {
     H264ParseContext *p = s->priv_data;
-    H264Context      *h = &p->h;
-    h->thread_context[0]   = h;
-    h->slice_context_count = 1;
-    ff_h264dsp_init(&h->h264dsp, 8, 1);
+
+    p->reference_dts = AV_NOPTS_VALUE;
+    p->last_frame_num = INT_MAX;
+    ff_h264dsp_init(&p->h264dsp, 8, 1);
     return 0;
 }