H264/MPEG frame-level multi-threading.

Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

libavcodec/error_resilience.c

@@ -32,6 +32,7 @@
 #include "mpegvideo.h"
 #include "h264.h"
 #include "rectangle.h"
+#include "thread.h"
 
 /*
  * H264 redefines mb_intra so it is not mistakely used (its uninitialized in h264)
@@ -428,8 +429,7 @@ int score_sum=0;
                     int best_score=256*256*256*64;
                     int best_pred=0;
                     const int mot_index= (mb_x + mb_y*mot_stride) * mot_step;
-                    int prev_x= s->current_picture.motion_val[0][mot_index][0];
-                    int prev_y= s->current_picture.motion_val[0][mot_index][1];
+                    int prev_x, prev_y, prev_ref;
 
                     if((mb_x^mb_y^pass)&1) continue;
 
@@ -527,11 +527,26 @@ skip_mean_and_median:
                     /* zero MV */
                     pred_count++;
 
+                    if (!fixed[mb_xy]) {
+                        if (s->avctx->codec_id == CODEC_ID_H264) {
+                            // FIXME
+                        } else {
+                            ff_thread_await_progress((AVFrame *) s->last_picture_ptr,
+                                                     mb_y, 0);
+                        }
+                        prev_x = s->last_picture.motion_val[0][mot_index][0];
+                        prev_y = s->last_picture.motion_val[0][mot_index][1];
+                        prev_ref = s->last_picture.ref_index[0][4*mb_xy];
+                    } else {
+                        prev_x = s->current_picture.motion_val[0][mot_index][0];
+                        prev_y = s->current_picture.motion_val[0][mot_index][1];
+                        prev_ref = s->current_picture.ref_index[0][4*mb_xy];
+                    }
+
                     /* last MV */
-                    mv_predictor[pred_count][0]= s->current_picture.motion_val[0][mot_index][0];
-                    mv_predictor[pred_count][1]= s->current_picture.motion_val[0][mot_index][1];
-                    ref         [pred_count]   = s->current_picture.ref_index[0][4*mb_xy];
-                    pred_count++;
+                    mv_predictor[pred_count][0]= prev_x;
+                    mv_predictor[pred_count][1]= prev_y;
+                    ref         [pred_count]   = prev_ref;
 
                     s->mv_dir = MV_DIR_FORWARD;
                     s->mb_intra=0;
@@ -662,6 +677,12 @@ static int is_intra_more_likely(MpegEncContext *s){
                 uint8_t *mb_ptr     = s->current_picture.data[0] + mb_x*16 + mb_y*16*s->linesize;
                 uint8_t *last_mb_ptr= s->last_picture.data   [0] + mb_x*16 + mb_y*16*s->linesize;
 
+                if (s->avctx->codec_id == CODEC_ID_H264) {
+                    // FIXME
+                } else {
+                    ff_thread_await_progress((AVFrame *) s->last_picture_ptr,
+                                             mb_y, 0);
+                }
                 is_intra_likely += s->dsp.sad[0](NULL, last_mb_ptr, mb_ptr                    , s->linesize, 16);
                 is_intra_likely -= s->dsp.sad[0](NULL, last_mb_ptr, last_mb_ptr+s->linesize*16, s->linesize, 16);
             }else{
@@ -681,6 +702,7 @@ void ff_er_frame_start(MpegEncContext *s){
 
     memset(s->error_status_table, MV_ERROR|AC_ERROR|DC_ERROR|VP_START|AC_END|DC_END|MV_END, s->mb_stride*s->mb_height*sizeof(uint8_t));
     s->error_count= 3*s->mb_num;
+    s->error_occurred = 0;
 }
 
 /**
@@ -720,7 +742,10 @@ void ff_er_add_slice(MpegEncContext *s, int startx, int starty, int endx, int en
         s->error_count -= end_i - start_i + 1;
     }
 
-    if(status & (AC_ERROR|DC_ERROR|MV_ERROR)) s->error_count= INT_MAX;
+    if(status & (AC_ERROR|DC_ERROR|MV_ERROR)) {
+        s->error_occurred = 1;
+        s->error_count= INT_MAX;
+    }
 
     if(mask == ~0x7F){
         memset(&s->error_status_table[start_xy], 0, (end_xy - start_xy) * sizeof(uint8_t));
@@ -995,6 +1020,12 @@ void ff_er_frame_end(MpegEncContext *s){
                     int time_pp= s->pp_time;
                     int time_pb= s->pb_time;
 
+                    if (s->avctx->codec_id == CODEC_ID_H264) {
+                        //FIXME
+                    } else {
+                        ff_thread_await_progress((AVFrame *) s->next_picture_ptr,
+                                                 mb_y, 0);
+                    }
                     s->mv[0][0][0] = s->next_picture.motion_val[0][xy][0]*time_pb/time_pp;
                     s->mv[0][0][1] = s->next_picture.motion_val[0][xy][1]*time_pb/time_pp;
                     s->mv[1][0][0] = s->next_picture.motion_val[0][xy][0]*(time_pb - time_pp)/time_pp;

libavcodec/h263dec.c

@@ -35,6 +35,7 @@
 #include "mpeg4video_parser.h"
 #include "msmpeg4.h"
 #include "vdpau_internal.h"
+#include "thread.h"
 #include "flv.h"
 #include "mpeg4video.h"
 
@@ -229,6 +230,7 @@ static int decode_slice(MpegEncContext *s){
                     if(++s->mb_x >= s->mb_width){
                         s->mb_x=0;
                         ff_draw_horiz_band(s, s->mb_y*mb_size, mb_size);
+                        MPV_report_decode_progress(s);
                         s->mb_y++;
                     }
                     return 0;
@@ -249,6 +251,7 @@ static int decode_slice(MpegEncContext *s){
         }
 
         ff_draw_horiz_band(s, s->mb_y*mb_size, mb_size);
+        MPV_report_decode_progress(s);
 
         s->mb_x= 0;
     }
@@ -609,6 +612,8 @@ retry:
     if(MPV_frame_start(s, avctx) < 0)
         return -1;
 
+    if (!s->divx_packed) ff_thread_finish_setup(avctx);
+
     if (CONFIG_MPEG4_VDPAU_DECODER && (s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)) {
         ff_vdpau_mpeg4_decode_picture(s, s->gb.buffer, s->gb.buffer_end - s->gb.buffer);
         goto frame_end;
@@ -639,8 +644,11 @@ retry:
             if(s->slice_height==0 || s->mb_x!=0 || (s->mb_y%s->slice_height)!=0 || get_bits_count(&s->gb) > s->gb.size_in_bits)
                 break;
         }else{
+            int prev_x=s->mb_x, prev_y=s->mb_y;
             if(ff_h263_resync(s)<0)
                 break;
+            if (prev_y * s->mb_width + prev_x < s->mb_y * s->mb_width + s->mb_x)
+                s->error_occurred = 1;
         }
 
         if(s->msmpeg4_version<4 && s->h263_pred)

libavcodec/h264.h

@@ -506,7 +506,9 @@ typedef struct H264Context{
     Picture *long_ref[32];
     Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
     Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
+    Picture *next_output_pic;
     int outputed_poc;
+    int next_outputed_poc;
 
     /**
      * memory management control operations buffer.

libavcodec/h264_direct.c

@@ -31,6 +31,7 @@
 #include "mpegvideo.h"
 #include "h264.h"
 #include "rectangle.h"
+#include "thread.h"
 
 //#undef NDEBUG
 #include <assert.h>
@@ -126,7 +127,7 @@ void ff_h264_direct_ref_list_init(H264Context * const h){
         h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
         ref1sidx=sidx= h->col_parity;
     }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){ // FL -> FL & differ parity
-        h->col_fieldoff= s->mb_stride*(2*(h->ref_list[1][0].reference) - 3);
+        h->col_fieldoff= 2*(h->ref_list[1][0].reference) - 3;
     }
 
     if(cur->pict_type != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred)
@@ -140,11 +141,27 @@ void ff_h264_direct_ref_list_init(H264Context * const h){
     }
 }
 
+static void await_reference_mb_row(H264Context * const h, Picture *ref, int mb_y)
+{
+    int ref_field = ref->reference - 1;
+    int ref_field_picture = ref->field_picture;
+    int ref_height = 16*h->s.mb_height >> ref_field_picture;
+
+    if(!HAVE_PTHREADS || !(h->s.avctx->active_thread_type&FF_THREAD_FRAME))
+        return;
+
+    //FIXME it can be safe to access mb stuff
+    //even if pixels aren't deblocked yet
+
+    ff_thread_await_progress((AVFrame*)ref, FFMIN(16*mb_y >> ref_field_picture, ref_height-1),
+                             ref_field_picture && ref_field);
+}
+
 static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
     MpegEncContext * const s = &h->s;
     int b8_stride = 2;
     int b4_stride = h->b_stride;
-    int mb_xy = h->mb_xy;
+    int mb_xy = h->mb_xy, mb_y = s->mb_y;
     int mb_type_col[2];
     const int16_t (*l1mv0)[2], (*l1mv1)[2];
     const int8_t *l1ref0, *l1ref1;
@@ -157,6 +174,8 @@ static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
 
     assert(h->ref_list[1][0].reference&3);
 
+    await_reference_mb_row(h, &h->ref_list[1][0], s->mb_y + !!IS_INTERLACED(*mb_type));
+
 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
 
 
@@ -217,14 +236,17 @@ static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
 
     if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
         if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
+            mb_y = (s->mb_y&~1) + h->col_parity;
             mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
             b8_stride = 0;
         }else{
-            mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
+            mb_y  += h->col_fieldoff;
+            mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
         }
         goto single_col;
     }else{                                               // AFL/AFR/FR/FL -> AFR/FR
         if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
+            mb_y = s->mb_y&~1;
             mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
             mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
@@ -260,6 +282,8 @@ single_col:
         }
     }
 
+    await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
+
     l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
     l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
     l1ref0 = &h->ref_list[1][0].ref_index [0][4*mb_xy];
@@ -384,7 +408,7 @@ static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
     MpegEncContext * const s = &h->s;
     int b8_stride = 2;
     int b4_stride = h->b_stride;
-    int mb_xy = h->mb_xy;
+    int mb_xy = h->mb_xy, mb_y = s->mb_y;
     int mb_type_col[2];
     const int16_t (*l1mv0)[2], (*l1mv1)[2];
     const int8_t *l1ref0, *l1ref1;
@@ -394,16 +418,21 @@ static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
 
     assert(h->ref_list[1][0].reference&3);
 
+    await_reference_mb_row(h, &h->ref_list[1][0], s->mb_y + !!IS_INTERLACED(*mb_type));
+
     if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
         if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
+            mb_y = (s->mb_y&~1) + h->col_parity;
             mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
             b8_stride = 0;
         }else{
-            mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
+            mb_y  += h->col_fieldoff;
+            mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
         }
         goto single_col;
     }else{                                               // AFL/AFR/FR/FL -> AFR/FR
         if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
+            mb_y = s->mb_y&~1;
             mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
             mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
@@ -440,6 +469,8 @@ single_col:
         }
     }
 
+    await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
+
     l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
     l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
     l1ref0 = &h->ref_list[1][0].ref_index [0][4*mb_xy];

libavcodec/mpeg12.c

@@ -37,6 +37,7 @@
 #include "bytestream.h"
 #include "vdpau_internal.h"
 #include "xvmc_internal.h"
+#include "thread.h"
 
 //#undef NDEBUG
 //#include <assert.h>
@@ -1179,6 +1180,27 @@ static av_cold int mpeg_decode_init(AVCodecContext *avctx)
     return 0;
 }
 
+static int mpeg_decode_update_thread_context(AVCodecContext *avctx, const AVCodecContext *avctx_from)
+{
+    Mpeg1Context *ctx = avctx->priv_data, *ctx_from = avctx_from->priv_data;
+    MpegEncContext *s = &ctx->mpeg_enc_ctx, *s1 = &ctx_from->mpeg_enc_ctx;
+    int err;
+
+    if(avctx == avctx_from || !ctx_from->mpeg_enc_ctx_allocated || !s1->context_initialized)
+        return 0;
+
+    err = ff_mpeg_update_thread_context(avctx, avctx_from);
+    if(err) return err;
+
+    if(!ctx->mpeg_enc_ctx_allocated)
+        memcpy(s + 1, s1 + 1, sizeof(Mpeg1Context) - sizeof(MpegEncContext));
+
+    if(!(s->pict_type == FF_B_TYPE || s->low_delay))
+        s->picture_number++;
+
+    return 0;
+}
+
 static void quant_matrix_rebuild(uint16_t *matrix, const uint8_t *old_perm,
                                      const uint8_t *new_perm){
     uint16_t temp_matrix[64];
@@ -1595,6 +1617,9 @@ static int mpeg_field_start(MpegEncContext *s, const uint8_t *buf, int buf_size)
         }
 
         *s->current_picture_ptr->pan_scan= s1->pan_scan;
+
+        if (HAVE_PTHREADS && (avctx->active_thread_type & FF_THREAD_FRAME))
+            ff_thread_finish_setup(avctx);
     }else{ //second field
             int i;
 
@@ -1769,6 +1794,7 @@ static int mpeg_decode_slice(Mpeg1Context *s1, int mb_y,
             const int mb_size= 16>>s->avctx->lowres;
 
             ff_draw_horiz_band(s, mb_size*(s->mb_y>>field_pic), mb_size);
+            MPV_report_decode_progress(s);
 
             s->mb_x = 0;
             s->mb_y += 1<<field_pic;
@@ -1926,7 +1952,8 @@ static int slice_end(AVCodecContext *avctx, AVFrame *pict)
             *pict= *(AVFrame*)s->current_picture_ptr;
             ff_print_debug_info(s, pict);
         } else {
-            s->picture_number++;
+            if (avctx->active_thread_type & FF_THREAD_FRAME)
+                s->picture_number++;
             /* latency of 1 frame for I- and P-frames */
             /* XXX: use another variable than picture_number */
             if (s->last_picture_ptr != NULL) {
@@ -2262,7 +2289,7 @@ static int decode_chunks(AVCodecContext *avctx,
         buf_ptr = ff_find_start_code(buf_ptr,buf_end, &start_code);
         if (start_code > 0x1ff){
             if(s2->pict_type != AV_PICTURE_TYPE_B || avctx->skip_frame <= AVDISCARD_DEFAULT){
-                if((avctx->active_thread_type & FF_THREAD_SLICE) && avctx->thread_count > 1){
+                if(HAVE_THREADS && (avctx->active_thread_type & FF_THREAD_SLICE)){
                     int i;
 
                     avctx->execute(avctx, slice_decode_thread,  &s2->thread_context[0], NULL, s->slice_count, sizeof(void*));
@@ -2430,7 +2457,7 @@ static int decode_chunks(AVCodecContext *avctx,
                     break;
                 }
 
-                if((avctx->active_thread_type & FF_THREAD_SLICE) && avctx->thread_count > 1){
+                if(HAVE_THREADS && (avctx->active_thread_type & FF_THREAD_SLICE)){
                     int threshold= (s2->mb_height*s->slice_count + avctx->thread_count/2) / avctx->thread_count;
                     if(threshold <= mb_y){
                         MpegEncContext *thread_context= s2->thread_context[s->slice_count];
@@ -2505,6 +2532,7 @@ AVCodec ff_mpeg1video_decoder = {
     .flush= flush,
     .max_lowres= 3,
     .long_name= NULL_IF_CONFIG_SMALL("MPEG-1 video"),
+    .update_thread_context= ONLY_IF_THREADS_ENABLED(mpeg_decode_update_thread_context)
 };
 
 AVCodec ff_mpeg2video_decoder = {
@@ -2541,7 +2569,7 @@ AVCodec ff_mpegvideo_decoder = {
 
 #if CONFIG_MPEG_XVMC_DECODER
 static av_cold int mpeg_mc_decode_init(AVCodecContext *avctx){
-    if((avctx->active_thread_type & FF_THREAD_SLICE) && avctx->thread_count > 1)
+    if( avctx->active_thread_type & FF_THREAD_SLICE )
         return -1;
     if( !(avctx->slice_flags & SLICE_FLAG_CODED_ORDER) )
         return -1;

libavcodec/mpeg4videodec.c

@@ -23,6 +23,7 @@
 #include "mpegvideo.h"
 #include "mpeg4video.h"
 #include "h263.h"
+#include "thread.h"
 
 // The defines below define the number of bits that are read at once for
 // reading vlc values. Changing these may improve speed and data cache needs
@@ -373,7 +374,13 @@ int mpeg4_decode_video_packet_header(MpegEncContext *s)
         return -1;
     }
     if(s->pict_type == AV_PICTURE_TYPE_B){
-        while(s->next_picture.mbskip_table[ s->mb_index2xy[ mb_num ] ]) mb_num++;
+        int mb_x = 0, mb_y = 0;
+
+        while(s->next_picture.mbskip_table[ s->mb_index2xy[ mb_num ] ]) {
+            if (!mb_x) ff_thread_await_progress((AVFrame*)s->next_picture_ptr, mb_y++, 0);
+            mb_num++;
+            if (++mb_x == s->mb_width) mb_x = 0;
+        }
         if(mb_num >= s->mb_num) return -1; // slice contains just skipped MBs which where already decoded
     }
 
@@ -1303,6 +1310,8 @@ static int mpeg4_decode_mb(MpegEncContext *s,
                 s->last_mv[i][1][0]=
                 s->last_mv[i][1][1]= 0;
             }
+
+            ff_thread_await_progress((AVFrame*)s->next_picture_ptr, s->mb_y, 0);
         }
 
         /* if we skipped it in the future P Frame than skip it now too */
@@ -1482,6 +1491,12 @@ end:
     if(s->codec_id==CODEC_ID_MPEG4){
         if(mpeg4_is_resync(s)){
             const int delta= s->mb_x + 1 == s->mb_width ? 2 : 1;
+
+            if(s->pict_type==AV_PICTURE_TYPE_B && s->next_picture.mbskip_table[xy + delta]){
+                ff_thread_await_progress((AVFrame*)s->next_picture_ptr,
+                                        (s->mb_x + delta >= s->mb_width) ? FFMIN(s->mb_y+1, s->mb_height-1) : s->mb_y, 0);
+            }
+
             if(s->pict_type==AV_PICTURE_TYPE_B && s->next_picture.mbskip_table[xy + delta])
                 return SLICE_OK;
             return SLICE_END;
@@ -2235,11 +2250,12 @@ AVCodec ff_mpeg4_decoder = {
     NULL,
     ff_h263_decode_end,
     ff_h263_decode_frame,
-    CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY,
+    CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY | CODEC_CAP_FRAME_THREADS,
     .flush= ff_mpeg_flush,
     .max_lowres= 3,
     .long_name= NULL_IF_CONFIG_SMALL("MPEG-4 part 2"),
     .pix_fmts= ff_hwaccel_pixfmt_list_420,
+    .update_thread_context= ONLY_IF_THREADS_ENABLED(ff_mpeg_update_thread_context)
 };
 
 

libavcodec/mpegvideo.h

@@ -76,6 +76,8 @@ enum OutputFormat {
 #define EXT_START_CODE          0x000001b5
 #define USER_START_CODE         0x000001b2
 
+struct MpegEncContext;
+
 /**
  * Picture.
  */
@@ -123,6 +125,7 @@ typedef struct Picture{
     int ref_poc[2][2][16];      ///< h264 POCs of the frames used as reference (FIXME need per slice)
     int ref_count[2][2];        ///< number of entries in ref_poc              (FIXME need per slice)
     int mbaff;                  ///< h264 1 -> MBAFF frame 0-> not MBAFF
+    int field_picture;          ///< whether or not the picture was encoded in seperate fields
 
     int mb_var_sum;             ///< sum of MB variance for current frame
     int mc_mb_var_sum;          ///< motion compensated MB variance for current frame
@@ -131,10 +134,9 @@ typedef struct Picture{
     uint8_t *mb_mean;           ///< Table for MB luminance
     int32_t *mb_cmp_score;      ///< Table for MB cmp scores, for mb decision FIXME remove
     int b_frame_score;          /* */
+    struct MpegEncContext *owner2; ///< pointer to the MpegEncContext that allocated this picture
 } Picture;
 
-struct MpegEncContext;
-
 /**
  * Motion estimation context.
  */
@@ -291,6 +293,8 @@ typedef struct MpegEncContext {
     Picture *last_picture_ptr;     ///< pointer to the previous picture.
     Picture *next_picture_ptr;     ///< pointer to the next picture (for bidir pred)
     Picture *current_picture_ptr;  ///< pointer to the current picture
+    int picture_count;             ///< number of allocated pictures (MAX_PICTURE_COUNT * avctx->thread_count)
+    int picture_range_start, picture_range_end; ///< the part of picture that this context can allocate in
     uint8_t *visualization_buffer[3]; //< temporary buffer vor MV visualization
     int last_dc[3];                ///< last DC values for MPEG1
     int16_t *dc_val_base;
@@ -470,7 +474,7 @@ typedef struct MpegEncContext {
     int last_bits; ///< temp var used for calculating the above vars
 
     /* error concealment / resync */
-    int error_count;
+    int error_count, error_occurred;
     uint8_t *error_status_table;       ///< table of the error status of each MB
 #define VP_START            1          ///< current MB is the first after a resync marker
 #define AC_ERROR            2
@@ -677,6 +681,10 @@ typedef struct MpegEncContext {
     void (*denoise_dct)(struct MpegEncContext *s, DCTELEM *block);
 } MpegEncContext;
 
+#define REBASE_PICTURE(pic, new_ctx, old_ctx) (pic ? \
+    (pic >= old_ctx->picture && pic < old_ctx->picture+old_ctx->picture_count ?\
+        &new_ctx->picture[pic - old_ctx->picture] : pic - (Picture*)old_ctx + (Picture*)new_ctx)\
+    : NULL)
 
 void MPV_decode_defaults(MpegEncContext *s);
 int MPV_common_init(MpegEncContext *s);
@@ -699,9 +707,13 @@ void ff_draw_horiz_band(MpegEncContext *s, int y, int h);
 void ff_mpeg_flush(AVCodecContext *avctx);
 void ff_print_debug_info(MpegEncContext *s, AVFrame *pict);
 void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix);
+void ff_release_unused_pictures(MpegEncContext *s, int remove_current);
 int ff_find_unused_picture(MpegEncContext *s, int shared);
 void ff_denoise_dct(MpegEncContext *s, DCTELEM *block);
 void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src);
+int MPV_lowest_referenced_row(MpegEncContext *s, int dir);
+void MPV_report_decode_progress(MpegEncContext *s);
+int ff_mpeg_update_thread_context(AVCodecContext *dst, const AVCodecContext *src);
 const uint8_t *ff_find_start_code(const uint8_t *p, const uint8_t *end, uint32_t *state);
 void ff_set_qscale(MpegEncContext * s, int qscale);
 

libavcodec/mpegvideo_enc.c

@@ -36,6 +36,7 @@
 #include "mjpegenc.h"
 #include "msmpeg4.h"
 #include "faandct.h"
+#include "thread.h"
 #include "aandcttab.h"
 #include "flv.h"
 #include "mpeg4video.h"
@@ -1225,9 +1226,9 @@ int MPV_encode_picture(AVCodecContext *avctx,
 {
     MpegEncContext *s = avctx->priv_data;
     AVFrame *pic_arg = data;
-    int i, stuffing_count;
+    int i, stuffing_count, context_count = avctx->active_thread_type&FF_THREAD_SLICE ? avctx->thread_count : 1;
 
-    for(i=0; i<avctx->thread_count; i++){
+    for(i=0; i<context_count; i++){
         int start_y= s->thread_context[i]->start_mb_y;
         int   end_y= s->thread_context[i]->  end_mb_y;
         int h= s->mb_height;
@@ -1291,7 +1292,7 @@ vbv_retry:
                     s->last_non_b_time= s->time - s->pp_time;
                 }
 //                av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
-                for(i=0; i<avctx->thread_count; i++){
+                for(i=0; i<context_count; i++){
                     PutBitContext *pb= &s->thread_context[i]->pb;
                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
                 }
@@ -2758,6 +2759,7 @@ static int encode_picture(MpegEncContext *s, int picture_number)
 {
     int i;
     int bits;
+    int context_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
 
     s->picture_number = picture_number;
 
@@ -2797,7 +2799,7 @@ static int encode_picture(MpegEncContext *s, int picture_number)
     }
 
     s->mb_intra=0; //for the rate distortion & bit compare functions
-    for(i=1; i<s->avctx->thread_count; i++){
+    for(i=1; i<context_count; i++){
         ff_update_duplicate_context(s->thread_context[i], s);
     }
 
@@ -2810,11 +2812,11 @@ static int encode_picture(MpegEncContext *s, int picture_number)
         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
         if(s->pict_type != AV_PICTURE_TYPE_B && s->avctx->me_threshold==0){
             if((s->avctx->pre_me && s->last_non_b_pict_type==AV_PICTURE_TYPE_I) || s->avctx->pre_me==2){
-                s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
+                s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
             }
         }
 
-        s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
+        s->avctx->execute(s->avctx, estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
     }else /* if(s->pict_type == AV_PICTURE_TYPE_I) */{
         /* I-Frame */
         for(i=0; i<s->mb_stride*s->mb_height; i++)
@@ -2822,10 +2824,10 @@ static int encode_picture(MpegEncContext *s, int picture_number)
 
         if(!s->fixed_qscale){
             /* finding spatial complexity for I-frame rate control */
-            s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
+            s->avctx->execute(s->avctx, mb_var_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
         }
     }
-    for(i=1; i<s->avctx->thread_count; i++){
+    for(i=1; i<context_count; i++){
         merge_context_after_me(s, s->thread_context[i]);
     }
     s->current_picture.mc_mb_var_sum= s->current_picture_ptr->mc_mb_var_sum= s->me.mc_mb_var_sum_temp;
@@ -2961,11 +2963,11 @@ static int encode_picture(MpegEncContext *s, int picture_number)
     bits= put_bits_count(&s->pb);
     s->header_bits= bits - s->last_bits;
 
-    for(i=1; i<s->avctx->thread_count; i++){
+    for(i=1; i<context_count; i++){
         update_duplicate_context_after_me(s->thread_context[i], s);
     }
-    s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, s->avctx->thread_count, sizeof(void*));
-    for(i=1; i<s->avctx->thread_count; i++){
+    s->avctx->execute(s->avctx, encode_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
+    for(i=1; i<context_count; i++){
         merge_context_after_encode(s, s->thread_context[i]);
     }
     emms_c();

libavcodec/pthread.c

@@ -55,7 +55,7 @@ typedef struct ThreadContext {
 } ThreadContext;
 
 /// Max number of frame buffers that can be allocated when using frame threads.
-#define MAX_BUFFERS 32
+#define MAX_BUFFERS (32+1)
 
 /**
  * Context used by codec threads and stored in their AVCodecContext thread_opaque.

libavcodec/utils.c

@@ -115,7 +115,7 @@ typedef struct InternalBuffer{
     enum PixelFormat pix_fmt;
 }InternalBuffer;
 
-#define INTERNAL_BUFFER_SIZE 32
+#define INTERNAL_BUFFER_SIZE (32+1)
 
 void avcodec_align_dimensions2(AVCodecContext *s, int *width, int *height, int linesize_align[4]){
     int w_align= 1;
@@ -360,6 +360,7 @@ void avcodec_default_release_buffer(AVCodecContext *s, AVFrame *pic){
     assert(pic->type==FF_BUFFER_TYPE_INTERNAL);
     assert(s->internal_buffer_count);
 
+    if(s->internal_buffer){
     buf = NULL; /* avoids warning */
     for(i=0; i<s->internal_buffer_count; i++){ //just 3-5 checks so is not worth to optimize
         buf= &((InternalBuffer*)s->internal_buffer)[i];
@@ -371,6 +372,7 @@ void avcodec_default_release_buffer(AVCodecContext *s, AVFrame *pic){
     last = &((InternalBuffer*)s->internal_buffer)[s->internal_buffer_count];
 
     FFSWAP(InternalBuffer, *buf, *last);
+    }
 
     for(i=0; i<4; i++){
         pic->data[i]=NULL;

tests/ref/vsynth1/error

 7416dfd319f04044d4575dc9d1b406e1 *./tests/data/vsynth1/error-mpeg4-adv.avi
-756836 ./tests/data/vsynth1/error-mpeg4-adv.avi
-ef8bfcd6e0883daba95d0f32486ebe2d *./tests/data/error.vsynth1.out.yuv
-stddev:   18.05 PSNR: 23.00 MAXDIFF:  245 bytes:  7603200/  7603200
+  756836 ./tests/data/vsynth1/error-mpeg4-adv.avi
+54342963593ba08bcde95244a011efe5 *./tests/data/error.vsynth1.out.yuv
+stddev:   17.59 PSNR: 23.22 MAXDIFF:  240 bytes:  7603200/  7603200

tests/ref/vsynth2/error

 90e65096aa9ebafa3fe3f44a5a47cdc4 *./tests/data/vsynth2/error-mpeg4-adv.avi
-176588 ./tests/data/vsynth2/error-mpeg4-adv.avi
-9fe1082179f80179439953c7397a46ef *./tests/data/error.vsynth2.out.yuv
-stddev:    9.00 PSNR: 29.04 MAXDIFF:  168 bytes:  7603200/  7603200
+  176588 ./tests/data/vsynth2/error-mpeg4-adv.avi
+ce12aa852126f2740838dd2da9e21a03 *./tests/data/error.vsynth2.out.yuv
+stddev:   10.06 PSNR: 28.08 MAXDIFF:  193 bytes:  7603200/  7603200

tests/regression-funcs.sh

@@ -102,7 +102,7 @@ do_ffmpeg_crc()
 
 do_video_decoding()
 {
-    do_ffmpeg $raw_dst $DEC_OPTS $1 -i $target_path/$file -f rawvideo $ENC_OPTS $2
+    do_ffmpeg $raw_dst $DEC_OPTS $1 -i $target_path/$file -f rawvideo $ENC_OPTS -vsync 0 $2
 }
 
 do_video_encoding()