h264_parser: move the H264DSPContext to the parser context
[ffmpeg.git] / libavcodec / h264_parser.c
index da2a5f9..d62eb73 100644 (file)
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
-#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 "h264data.h"
 #include "internal.h"
+#include "mpegutils.h"
+#include "parser.h"
 
-#include <assert.h>
+typedef struct H264ParseContext {
+    H264Context h;
+    ParseContext pc;
+    H264ParamSets ps;
+    H264DSPContext h264dsp;
+    int got_first;
+} H264ParseContext;
 
 
-static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
+static int h264_find_frame_end(H264ParseContext *p, const uint8_t *buf,
                                int buf_size)
 {
     int i;
     uint32_t state;
-    ParseContext *pc = &h->parse_context;
+    ParseContext *pc = &p->pc;
 //    mb_addr= pc->mb_addr - 1;
     state = pc->state;
     if (state > 13)
@@ -47,30 +64,9 @@ static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
 
     for (i = 0; i < buf_size; i++) {
         if (state == 7) {
-#if HAVE_FAST_UNALIGNED
-            /* we check i < buf_size instead of i + 3 / 7 because it is
-             * simpler and there must be FF_INPUT_BUFFER_PADDING_SIZE
-             * bytes at the end.
-             */
-#if HAVE_FAST_64BIT
-            while (i < buf_size &&
-                   !((~*(const uint64_t *)(buf + i) &
-                      (*(const uint64_t *)(buf + i) - 0x0101010101010101ULL)) &
-                      0x8080808080808080ULL))
-                i += 8;
-#else
-            while (i < buf_size &&
-                   !((~*(const uint32_t *)(buf + i) &
-                      (*(const uint32_t *)(buf + i) - 0x01010101U)) &
-                      0x80808080U))
-                i += 4;
-#endif
-#endif
-            for (; i < buf_size; i++)
-                if (!buf[i]) {
-                    state = 2;
-                    break;
-                }
+            i += p->h264dsp.startcode_find_candidate(buf + i, buf_size - i);
+            if (i < buf_size)
+                state = 2;
         } else if (state <= 2) {
             if (buf[i] == 1)
                 state ^= 5;            // 2->7, 1->4, 0->5
@@ -79,13 +75,15 @@ 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) {
                 if (pc->frame_start_found) {
                     state += 8;
                     continue;
@@ -94,6 +92,7 @@ static int h264_find_frame_end(H264Context *h, const uint8_t *buf,
             }
             state = 7;
         } else {
+            // first_mb_in_slice is 0, probably the first nal of a new slice
             if (buf[i] & 0x80)
                 goto found;
             state = 7;
@@ -108,6 +107,84 @@ found:
     return i - (state & 5);
 }
 
+static int scan_mmco_reset(AVCodecParserContext *s, GetBitContext *gb)
+{
+    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];
+
+
+    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, h->picture_structure) < 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(gb);
+                    else if (reordering_of_pic_nums_idc > 3) {
+                        av_log(h->avctx, 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(h->avctx, 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);
+
+    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(h->avctx, 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(gb);
+            if (opcode == MMCO_SHORT2LONG || opcode == MMCO_LONG2UNUSED ||
+                opcode == MMCO_LONG || opcode == MMCO_SET_MAX_LONG)
+                get_ue_golomb_31(gb);
+        }
+    }
+
+    return 0;
+}
+
 /**
  * Parse NAL units of found picture and decode some basic information.
  *
@@ -120,30 +197,32 @@ static inline int parse_nal_units(AVCodecParserContext *s,
                                   AVCodecContext *avctx,
                                   const uint8_t *buf, int buf_size)
 {
-    H264Context *h         = s->priv_data;
+    H264ParseContext *p = s->priv_data;
+    H264Context      *h = &p->h;
     const uint8_t *buf_end = buf + buf_size;
+
+    H2645NAL nal = { NULL };
+
     unsigned int pps_id;
     unsigned int slice_type;
-    int state = -1;
-    const uint8_t *ptr;
+    int state = -1, got_reset = 0;
     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->avctx = avctx;
+    ff_h264_reset_sei(h);
 
     if (!buf_size)
         return 0;
 
     for (;;) {
-        int src_length, dst_length, consumed;
+        const SPS *sps;
+        int src_length, consumed;
         buf = avpriv_find_start_code(buf, buf_end, &state);
         if (buf >= buf_end)
             break;
@@ -153,21 +232,41 @@ static inline int parse_nal_units(AVCodecParserContext *s,
         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, src_length, &nal);
+        if (consumed < 0)
             break;
 
-        init_get_bits(&h->gb, ptr, 8 * dst_length);
+        ret = init_get_bits(&nal.gb, nal.data, nal.size * 8);
+        if (ret < 0)
+            goto fail;
+        get_bits1(&nal.gb);
+        nal.ref_idc = get_bits(&nal.gb, 2);
+        nal.type    = get_bits(&nal.gb, 5);
+
+        h->gb            = nal.gb;
+        h->nal_ref_idc   = nal.ref_idc;
+        h->nal_unit_type = nal.type;
+
         switch (h->nal_unit_type) {
         case NAL_SPS:
-            ff_h264_decode_seq_parameter_set(h);
+            ff_h264_decode_seq_parameter_set(&nal.gb, avctx, &p->ps);
             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);
@@ -181,68 +280,129 @@ static inline int parse_nal_units(AVCodecParserContext *s,
             h->prev_poc_lsb          = 0;
         /* fall through */
         case NAL_SLICE:
-            get_ue_golomb(&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];
+            get_ue_golomb(&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 (h->sei_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;
+                       "pps_id %u out of range\n", pps_id);
+                goto fail;
             }
-            if (!h->pps_buffers[pps_id]) {
+            if (!p->ps.pps_list[pps_id]) {
                 av_log(h->avctx, AV_LOG_ERROR,
-                       "non-existing PPS referenced\n");
-                return -1;
+                       "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]) {
+            p->ps.pps = (const PPS*)p->ps.pps_list[pps_id]->data;
+            if (!p->ps.sps_list[p->ps.pps->sps_id]) {
                 av_log(h->avctx, AV_LOG_ERROR,
-                       "non-existing SPS referenced\n");
-                return -1;
+                       "non-existing SPS %u referenced\n", p->ps.pps->sps_id);
+                goto fail;
+            }
+            p->ps.sps = (SPS*)p->ps.sps_list[p->ps.pps->sps_id]->data;
+
+            h->ps.sps = p->ps.sps;
+            h->ps.pps = p->ps.pps;
+            sps = p->ps.sps;
+
+            h->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 (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;
+                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;
+                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;
+                break;
+            default:
+                s->format = AV_PIX_FMT_NONE;
             }
-            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;
+            avctx->profile = ff_h264_get_profile(sps);
+            avctx->level   = sps->level_idc;
 
-            if (h->sps.frame_mbs_only_flag) {
+            if (sps->frame_mbs_only_flag) {
                 h->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
+                    h->picture_structure = PICT_TOP_FIELD + get_bits1(&nal.gb); // bottom_field_flag
                 } else {
                     h->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);
+                get_ue_golomb(&nal.gb); /* idr_pic_id */
+            if (sps->poc_type == 0) {
+                h->poc_lsb = get_bits(&nal.gb, sps->log2_max_poc_lsb);
 
-                if (h->pps.pic_order_present == 1 &&
+                if (p->ps.pps->pic_order_present == 1 &&
                     h->picture_structure == PICT_FRAME)
-                    h->delta_poc_bottom = get_se_golomb(&h->gb);
+                    h->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) {
+                h->delta_poc[0] = get_se_golomb(&nal.gb);
 
-                if (h->pps.pic_order_present == 1 &&
+                if (p->ps.pps->pic_order_present == 1 &&
                     h->picture_structure == PICT_FRAME)
-                    h->delta_poc[1] = get_se_golomb(&h->gb);
+                    h->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);
 
-            if (h->sps.pic_struct_present_flag) {
+            /* 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, &nal.gb);
+                if (got_reset < 0)
+                    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) {
+                if (!got_reset) {
+                    h->prev_poc_msb = h->poc_msb;
+                    h->prev_poc_lsb = h->poc_lsb;
+                } else {
+                    h->prev_poc_msb = 0;
+                    h->prev_poc_lsb =
+                        h->picture_structure == PICT_BOTTOM_FIELD ? 0 : field_poc[0];
+                }
+            }
+
+            if (sps->pic_struct_present_flag) {
                 switch (h->sei_pic_struct) {
                 case SEI_PIC_STRUCT_TOP_FIELD:
                 case SEI_PIC_STRUCT_BOTTOM_FIELD:
@@ -273,7 +433,7 @@ static inline int parse_nal_units(AVCodecParserContext *s,
 
             if (h->picture_structure == PICT_FRAME) {
                 s->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
-                if (h->sps.pic_struct_present_flag) {
+                if (sps->pic_struct_present_flag) {
                     switch (h->sei_pic_struct) {
                     case SEI_PIC_STRUCT_TOP_BOTTOM:
                     case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
@@ -303,12 +463,15 @@ static inline int parse_nal_units(AVCodecParserContext *s,
                 s->field_order = AV_FIELD_UNKNOWN;
             }
 
+            av_freep(&nal.rbsp_buffer);
             return 0; /* no need to evaluate the rest */
         }
         buf += consumed;
     }
     /* didn't find a picture! */
     av_log(h->avctx, AV_LOG_ERROR, "missing picture in access unit\n");
+fail:
+    av_freep(&nal.rbsp_buffer);
     return -1;
 }
 
@@ -317,12 +480,13 @@ 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;
+    H264Context      *h = &p->h;
+    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 the decoder.
@@ -339,7 +503,7 @@ static int h264_parse(AVCodecParserContext *s,
     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);
 
         if (ff_combine_frame(pc, next, &buf, &buf_size) < 0) {
             *poutbuf      = NULL;
@@ -349,7 +513,7 @@ static int h264_parse(AVCodecParserContext *s,
 
         if (next < 0 && next != END_NOT_FOUND) {
             assert(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); // update state
         }
     }
 
@@ -389,8 +553,10 @@ static int h264_split(AVCodecContext *avctx,
          *     (state&0xFFFFFF1F) == 0x105) {
          *  }
          */
-        if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x107 &&
-            (state & 0xFFFFFF1F) != 0x108 && (state & 0xFFFFFF1F) != 0x109) {
+        if ((state & 0xFFFFFF00) == 0x100 && (state & 0xFFFFFF1F) != 0x106 &&
+            (state & 0xFFFFFF1F) != 0x107 && (state & 0xFFFFFF1F) != 0x108 &&
+            (state & 0xFFFFFF1F) != 0x109 && (state & 0xFFFFFF1F) != 0x10d &&
+            (state & 0xFFFFFF1F) != 0x10f) {
             if (has_sps) {
                 while (i > 4 && buf[i - 5] == 0)
                     i--;
@@ -403,29 +569,43 @@ static int h264_split(AVCodecContext *avctx,
     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;
+    H264Context      *h = &p->h;
+    ParseContext *pc = &p->pc;
+    int i;
 
     av_free(pc->buffer);
     ff_h264_free_context(h);
+
+    for (i = 0; i < FF_ARRAY_ELEMS(p->ps.sps_list); i++)
+        av_buffer_unref(&p->ps.sps_list[i]);
+
+    for (i = 0; i < FF_ARRAY_ELEMS(p->ps.pps_list); i++)
+        av_buffer_unref(&p->ps.pps_list[i]);
 }
 
 static av_cold int init(AVCodecParserContext *s)
 {
-    H264Context *h = s->priv_data;
-    h->thread_context[0]   = h;
+    H264ParseContext *p = s->priv_data;
+    H264Context      *h = &p->h;
+
+    h->slice_ctx = av_mallocz(sizeof(*h->slice_ctx));
+    if (!h->slice_ctx)
+        return 0;
+    h->nb_slice_ctx = 1;
+
     h->slice_context_count = 1;
-    ff_h264dsp_init(&h->h264dsp, 8, 1);
+    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,
 };