]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/mpegvideo.c
ape: decode directly to the user-provided AVFrame
[ffmpeg] / libavcodec / mpegvideo.c
index a874548680d3ac45183e3d1d075ecd4f123b2300..4b68fd56a0515031445ab7b9a8e368897f76ea09 100644 (file)
  * The simplest mpeg encoder (well, it was the simplest!).
  */
 
-#include "libavutil/intmath.h"
 #include "libavutil/imgutils.h"
 #include "avcodec.h"
 #include "dsputil.h"
 #include "internal.h"
+#include "mathops.h"
 #include "mpegvideo.h"
-#include "mpegvideo_common.h"
 #include "mjpegenc.h"
 #include "msmpeg4.h"
-#include "faandct.h"
 #include "xvmc_internal.h"
 #include "thread.h"
 #include <limits.h>
 //#include <assert.h>
 
 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale);
+                                   int16_t *block, int n, int qscale);
 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale);
+                                   int16_t *block, int n, int qscale);
 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale);
+                                   int16_t *block, int n, int qscale);
 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale);
+                                   int16_t *block, int n, int qscale);
 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale);
+                                   int16_t *block, int n, int qscale);
 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
-                                  DCTELEM *block, int n, int qscale);
+                                  int16_t *block, int n, int qscale);
 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
-                                  DCTELEM *block, int n, int qscale);
+                                  int16_t *block, int n, int qscale);
 
 
 /* enable all paranoid tests for rounding, overflows, etc... */
@@ -127,17 +125,26 @@ const uint8_t *const ff_mpeg2_dc_scale_table[4] = {
     mpeg2_dc_scale_table3,
 };
 
-const enum PixelFormat ff_pixfmt_list_420[] = {
-    PIX_FMT_YUV420P,
-    PIX_FMT_NONE
+const enum AVPixelFormat ff_pixfmt_list_420[] = {
+    AV_PIX_FMT_YUV420P,
+    AV_PIX_FMT_NONE
 };
 
-const enum PixelFormat ff_hwaccel_pixfmt_list_420[] = {
-    PIX_FMT_DXVA2_VLD,
-    PIX_FMT_VAAPI_VLD,
-    PIX_FMT_VDA_VLD,
-    PIX_FMT_YUV420P,
-    PIX_FMT_NONE
+const enum AVPixelFormat ff_hwaccel_pixfmt_list_420[] = {
+#if CONFIG_DXVA2
+    AV_PIX_FMT_DXVA2_VLD,
+#endif
+#if CONFIG_VAAPI
+    AV_PIX_FMT_VAAPI_VLD,
+#endif
+#if CONFIG_VDA
+    AV_PIX_FMT_VDA_VLD,
+#endif
+#if CONFIG_VDPAU
+    AV_PIX_FMT_VDPAU,
+#endif
+    AV_PIX_FMT_YUV420P,
+    AV_PIX_FMT_NONE
 };
 
 const uint8_t *avpriv_mpv_find_start_code(const uint8_t *restrict p,
@@ -176,7 +183,8 @@ const uint8_t *avpriv_mpv_find_start_code(const uint8_t *restrict p,
 /* init common dct for both encoder and decoder */
 av_cold int ff_dct_common_init(MpegEncContext *s)
 {
-    dsputil_init(&s->dsp, s->avctx);
+    ff_dsputil_init(&s->dsp, s->avctx);
+    ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
 
     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
@@ -187,20 +195,16 @@ av_cold int ff_dct_common_init(MpegEncContext *s)
         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
 
-#if HAVE_MMX
-    MPV_common_init_mmx(s);
+#if ARCH_X86
+    ff_MPV_common_init_x86(s);
 #elif ARCH_ALPHA
-    MPV_common_init_axp(s);
-#elif CONFIG_MLIB
-    MPV_common_init_mlib(s);
-#elif HAVE_MMI
-    MPV_common_init_mmi(s);
+    ff_MPV_common_init_axp(s);
 #elif ARCH_ARM
-    MPV_common_init_arm(s);
+    ff_MPV_common_init_arm(s);
 #elif HAVE_ALTIVEC
-    MPV_common_init_altivec(s);
+    ff_MPV_common_init_altivec(s);
 #elif ARCH_BFIN
-    MPV_common_init_bfin(s);
+    ff_MPV_common_init_bfin(s);
 #endif
 
     /* load & permutate scantables
@@ -230,22 +234,48 @@ void ff_copy_picture(Picture *dst, Picture *src)
  */
 static void free_frame_buffer(MpegEncContext *s, Picture *pic)
 {
-    /* Windows Media Image codecs allocate internal buffers with different
-     * dimensions; ignore user defined callbacks for these
-     */
-    if (s->codec_id != CODEC_ID_WMV3IMAGE && s->codec_id != CODEC_ID_VC1IMAGE)
-        ff_thread_release_buffer(s->avctx, (AVFrame *) pic);
+    /* WM Image / Screen codecs allocate internal buffers with different
+     * dimensions / colorspaces; ignore user-defined callbacks for these. */
+    if (s->codec_id != AV_CODEC_ID_WMV3IMAGE &&
+        s->codec_id != AV_CODEC_ID_VC1IMAGE  &&
+        s->codec_id != AV_CODEC_ID_MSS2)
+        ff_thread_release_buffer(s->avctx, &pic->f);
     else
-        avcodec_default_release_buffer(s->avctx, (AVFrame *) pic);
+        avcodec_default_release_buffer(s->avctx, &pic->f);
     av_freep(&pic->f.hwaccel_picture_private);
 }
 
+int ff_mpv_frame_size_alloc(MpegEncContext *s, int linesize)
+{
+    int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
+
+    // edge emu needs blocksize + filter length - 1
+    // (= 17x17 for  halfpel / 21x21 for  h264)
+    // VC1 computes luma and chroma simultaneously and needs 19X19 + 9x9
+    // at uvlinesize. It supports only YUV420 so 24x24 is enough
+    // linesize * interlaced * MBsize
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->edge_emu_buffer, alloc_size * 2 * 24,
+                      fail);
+
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->me.scratchpad, alloc_size * 2 * 16 * 2,
+                      fail)
+    s->me.temp         = s->me.scratchpad;
+    s->rd_scratchpad   = s->me.scratchpad;
+    s->b_scratchpad    = s->me.scratchpad;
+    s->obmc_scratchpad = s->me.scratchpad + 16;
+
+    return 0;
+fail:
+    av_freep(&s->edge_emu_buffer);
+    return AVERROR(ENOMEM);
+}
+
 /**
  * Allocate a frame buffer
  */
 static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
 {
-    int r;
+    int r, ret;
 
     if (s->avctx->hwaccel) {
         assert(!pic->f.hwaccel_picture_private);
@@ -258,14 +288,16 @@ static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
         }
     }
 
-    if (s->codec_id != CODEC_ID_WMV3IMAGE && s->codec_id != CODEC_ID_VC1IMAGE)
-        r = ff_thread_get_buffer(s->avctx, (AVFrame *) pic);
+    if (s->codec_id != AV_CODEC_ID_WMV3IMAGE &&
+        s->codec_id != AV_CODEC_ID_VC1IMAGE  &&
+        s->codec_id != AV_CODEC_ID_MSS2)
+        r = ff_thread_get_buffer(s->avctx, &pic->f);
     else
-        r = avcodec_default_get_buffer(s->avctx, (AVFrame *) pic);
+        r = avcodec_default_get_buffer(s->avctx, &pic->f);
 
-    if (r < 0 || !pic->f.age || !pic->f.type || !pic->f.data[0]) {
-        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %d %p)\n",
-               r, pic->f.age, pic->f.type, pic->f.data[0]);
+    if (r < 0 || !pic->f.type || !pic->f.data[0]) {
+        av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %p)\n",
+               r, pic->f.type, pic->f.data[0]);
         av_freep(&pic->f.hwaccel_picture_private);
         return -1;
     }
@@ -285,6 +317,14 @@ static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
         return -1;
     }
 
+    if (!s->edge_emu_buffer &&
+        (ret = ff_mpv_frame_size_alloc(s, pic->f.linesize[0])) < 0) {
+        av_log(s->avctx, AV_LOG_ERROR,
+               "get_buffer() failed to allocate context scratch buffers.\n");
+        free_frame_buffer(s, pic);
+        return ret;
+    }
+
     return 0;
 }
 
@@ -362,22 +402,13 @@ int ff_alloc_picture(MpegEncContext *s, Picture *pic, int shared)
         }
         if (s->avctx->debug&FF_DEBUG_DCT_COEFF) {
             FF_ALLOCZ_OR_GOTO(s->avctx, pic->f.dct_coeff,
-                              64 * mb_array_size * sizeof(DCTELEM) * 6, fail)
+                              64 * mb_array_size * sizeof(int16_t) * 6, fail)
         }
         pic->f.qstride = s->mb_stride;
         FF_ALLOCZ_OR_GOTO(s->avctx, pic->f.pan_scan,
                           1 * sizeof(AVPanScan), fail)
     }
 
-    /* It might be nicer if the application would keep track of these
-     * but it would require an API change. */
-    memmove(s->prev_pict_types + 1, s->prev_pict_types,
-            PREV_PICT_TYPES_BUFFER_SIZE-1);
-    s->prev_pict_types[0] =  s->dropable ? AV_PICTURE_TYPE_B : s->pict_type;
-    if (pic->f.age < PREV_PICT_TYPES_BUFFER_SIZE &&
-        s->prev_pict_types[pic->f.age] == AV_PICTURE_TYPE_B)
-        pic->f.age = INT_MAX; // Skipped MBs in B-frames are quite rare in MPEG-1/2
-                              // and it is a bit tricky to skip them anyway.
     pic->owner2 = s;
 
     return 0;
@@ -403,13 +434,16 @@ static void free_picture(MpegEncContext *s, Picture *pic)
     av_freep(&pic->mb_mean);
     av_freep(&pic->f.mbskip_table);
     av_freep(&pic->qscale_table_base);
+    pic->f.qscale_table = NULL;
     av_freep(&pic->mb_type_base);
+    pic->f.mb_type = NULL;
     av_freep(&pic->f.dct_coeff);
     av_freep(&pic->f.pan_scan);
     pic->f.mb_type = NULL;
     for (i = 0; i < 2; i++) {
         av_freep(&pic->motion_val_base[i]);
         av_freep(&pic->f.ref_index[i]);
+        pic->f.motion_val[i] = NULL;
     }
 
     if (pic->f.type == FF_BUFFER_TYPE_SHARED) {
@@ -421,26 +455,20 @@ static void free_picture(MpegEncContext *s, Picture *pic)
     }
 }
 
-static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base)
+static int init_duplicate_context(MpegEncContext *s)
 {
     int y_size = s->b8_stride * (2 * s->mb_height + 1);
     int c_size = s->mb_stride * (s->mb_height + 1);
     int yc_size = y_size + 2 * c_size;
     int i;
 
-    // edge emu needs blocksize + filter length - 1
-    // (= 17x17 for  halfpel / 21x21 for  h264)
-    FF_ALLOCZ_OR_GOTO(s->avctx, s->edge_emu_buffer,
-                      (s->width + 64) * 2 * 21 * 2, fail);    // (width + edge + align)*interlaced*MBsize*tolerance
+    s->edge_emu_buffer =
+    s->me.scratchpad   =
+    s->me.temp         =
+    s->rd_scratchpad   =
+    s->b_scratchpad    =
+    s->obmc_scratchpad = NULL;
 
-    // FIXME should be linesize instead of s->width * 2
-    // but that is not known before get_buffer()
-    FF_ALLOCZ_OR_GOTO(s->avctx, s->me.scratchpad,
-                      (s->width + 64) * 4 * 16 * 2 * sizeof(uint8_t), fail)
-    s->me.temp         = s->me.scratchpad;
-    s->rd_scratchpad   = s->me.scratchpad;
-    s->b_scratchpad    = s->me.scratchpad;
-    s->obmc_scratchpad = s->me.scratchpad + 16;
     if (s->encoding) {
         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map,
                           ME_MAP_SIZE * sizeof(uint32_t), fail)
@@ -451,7 +479,7 @@ static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base)
                               2 * 64 * sizeof(int), fail)
         }
     }
-    FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(DCTELEM), fail)
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail)
     s->block = s->blocks[0];
 
     for (i = 0; i < 12; i++) {
@@ -469,7 +497,7 @@ static int init_duplicate_context(MpegEncContext *s, MpegEncContext *base)
 
     return 0;
 fail:
-    return -1; // free() through MPV_common_end()
+    return -1; // free() through ff_MPV_common_end()
 }
 
 static void free_duplicate_context(MpegEncContext *s)
@@ -519,10 +547,10 @@ static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
 #undef COPY
 }
 
-void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
+int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
 {
     MpegEncContext bak;
-    int i;
+    int i, ret;
     // FIXME copy only needed parts
     // START_TIMER
     backup_duplicate_context(&bak, dst);
@@ -531,13 +559,21 @@ void ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
     for (i = 0; i < 12; i++) {
         dst->pblocks[i] = &dst->block[i];
     }
+    if (!dst->edge_emu_buffer &&
+        (ret = ff_mpv_frame_size_alloc(dst, dst->linesize)) < 0) {
+        av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context "
+               "scratch buffers.\n");
+        return ret;
+    }
     // STOP_TIMER("update_duplicate_context")
     // about 10k cycles / 0.01 sec for  1000frames on 1ghz with 2 threads
+    return 0;
 }
 
 int ff_mpeg_update_thread_context(AVCodecContext *dst,
                                   const AVCodecContext *src)
 {
+    int i;
     MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
 
     if (dst == src || !s1->context_initialized)
@@ -554,7 +590,16 @@ int ff_mpeg_update_thread_context(AVCodecContext *dst,
         s->bitstream_buffer      = NULL;
         s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
 
-        MPV_common_init(s);
+        ff_MPV_common_init(s);
+    }
+
+    if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
+        int err;
+        s->context_reinit = 0;
+        s->height = s1->height;
+        s->width  = s1->width;
+        if ((err = ff_MPV_common_frame_size_change(s)) < 0)
+            return err;
     }
 
     s->avctx->coded_height  = s1->avctx->coded_height;
@@ -570,13 +615,14 @@ int ff_mpeg_update_thread_context(AVCodecContext *dst,
     memcpy(&s->last_picture, &s1->last_picture,
            (char *) &s1->last_picture_ptr - (char *) &s1->last_picture);
 
+    // reset s->picture[].f.extended_data to s->picture[].f.data
+    for (i = 0; i < s->picture_count; i++)
+        s->picture[i].f.extended_data = s->picture[i].f.data;
+
     s->last_picture_ptr    = REBASE_PICTURE(s1->last_picture_ptr,    s, s1);
     s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
     s->next_picture_ptr    = REBASE_PICTURE(s1->next_picture_ptr,    s, s1);
 
-    memcpy(s->prev_pict_types, s1->prev_pict_types,
-           PREV_PICT_TYPES_BUFFER_SIZE);
-
     // Error/bug resilience
     s->next_p_frame_damaged = s1->next_p_frame_damaged;
     s->workaround_bugs      = s1->workaround_bugs;
@@ -588,7 +634,7 @@ int ff_mpeg_update_thread_context(AVCodecContext *dst,
     // B-frame info
     s->max_b_frames = s1->max_b_frames;
     s->low_delay    = s1->low_delay;
-    s->dropable     = s1->dropable;
+    s->droppable    = s1->droppable;
 
     // DivX handling (doesn't work)
     s->divx_packed  = s1->divx_packed;
@@ -606,6 +652,20 @@ int ff_mpeg_update_thread_context(AVCodecContext *dst,
                FF_INPUT_BUFFER_PADDING_SIZE);
     }
 
+    // linesize dependend scratch buffer allocation
+    if (!s->edge_emu_buffer)
+        if (s1->linesize) {
+            if (ff_mpv_frame_size_alloc(s, s1->linesize) < 0) {
+                av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
+                       "scratch buffers.\n");
+                return AVERROR(ENOMEM);
+            }
+        } else {
+            av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
+                   "be allocated due to unknown size.\n");
+            return AVERROR_BUG;
+        }
+
     // MPEG2/interlacing info
     memcpy(&s->progressive_sequence, &s1->progressive_sequence,
            (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
@@ -629,7 +689,7 @@ int ff_mpeg_update_thread_context(AVCodecContext *dst,
  * The changed fields will not depend upon the
  * prior state of the MpegEncContext.
  */
-void MPV_common_defaults(MpegEncContext *s)
+void ff_MPV_common_defaults(MpegEncContext *s)
 {
     s->y_dc_scale_table      =
     s->c_dc_scale_table      = ff_mpeg1_dc_scale_table;
@@ -649,6 +709,8 @@ void MPV_common_defaults(MpegEncContext *s)
 
     s->picture_range_start   = 0;
     s->picture_range_end     = MAX_PICTURE_COUNT;
+
+    s->slice_context_count   = 1;
 }
 
 /**
@@ -656,42 +718,200 @@ void MPV_common_defaults(MpegEncContext *s)
  * the changed fields will not depend upon
  * the prior state of the MpegEncContext.
  */
-void MPV_decode_defaults(MpegEncContext *s)
+void ff_MPV_decode_defaults(MpegEncContext *s)
 {
-    MPV_common_defaults(s);
+    ff_MPV_common_defaults(s);
+}
+
+/**
+ * Initialize and allocates MpegEncContext fields dependent on the resolution.
+ */
+static int init_context_frame(MpegEncContext *s)
+{
+    int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
+
+    s->mb_width   = (s->width + 15) / 16;
+    s->mb_stride  = s->mb_width + 1;
+    s->b8_stride  = s->mb_width * 2 + 1;
+    s->b4_stride  = s->mb_width * 4 + 1;
+    mb_array_size = s->mb_height * s->mb_stride;
+    mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
+
+    /* set default edge pos, will be overriden
+     * in decode_header if needed */
+    s->h_edge_pos = s->mb_width * 16;
+    s->v_edge_pos = s->mb_height * 16;
+
+    s->mb_num     = s->mb_width * s->mb_height;
+
+    s->block_wrap[0] =
+    s->block_wrap[1] =
+    s->block_wrap[2] =
+    s->block_wrap[3] = s->b8_stride;
+    s->block_wrap[4] =
+    s->block_wrap[5] = s->mb_stride;
+
+    y_size  = s->b8_stride * (2 * s->mb_height + 1);
+    c_size  = s->mb_stride * (s->mb_height + 1);
+    yc_size = y_size + 2   * c_size;
+
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int),
+                      fail); // error ressilience code looks cleaner with this
+    for (y = 0; y < s->mb_height; y++)
+        for (x = 0; x < s->mb_width; x++)
+            s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
+
+    s->mb_index2xy[s->mb_height * s->mb_width] =
+        (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
+
+    if (s->encoding) {
+        /* Allocate MV tables */
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base,
+                          mv_table_size * 2 * sizeof(int16_t), fail);
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base,
+                          mv_table_size * 2 * sizeof(int16_t), fail);
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base,
+                          mv_table_size * 2 * sizeof(int16_t), fail);
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base,
+                          mv_table_size * 2 * sizeof(int16_t), fail);
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base,
+                          mv_table_size * 2 * sizeof(int16_t), fail);
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base,
+                          mv_table_size * 2 * sizeof(int16_t), fail);
+        s->p_mv_table            = s->p_mv_table_base + s->mb_stride + 1;
+        s->b_forw_mv_table       = s->b_forw_mv_table_base + s->mb_stride + 1;
+        s->b_back_mv_table       = s->b_back_mv_table_base + s->mb_stride + 1;
+        s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base +
+                                   s->mb_stride + 1;
+        s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base +
+                                   s->mb_stride + 1;
+        s->b_direct_mv_table     = s->b_direct_mv_table_base + s->mb_stride + 1;
+
+        /* Allocate MB type table */
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size *
+                          sizeof(uint16_t), fail); // needed for encoding
+
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size *
+                          sizeof(int), fail);
+
+        FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
+                         mb_array_size * sizeof(float), fail);
+        FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab,
+                         mb_array_size * sizeof(float), fail);
+
+    }
+
+    FF_ALLOC_OR_GOTO(s->avctx, s->er_temp_buffer,
+                     mb_array_size * sizeof(uint8_t), fail);
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->error_status_table,
+                      mb_array_size * sizeof(uint8_t), fail);
+
+    if (s->codec_id == AV_CODEC_ID_MPEG4 ||
+        (s->flags & CODEC_FLAG_INTERLACED_ME)) {
+        /* interlaced direct mode decoding tables */
+        for (i = 0; i < 2; i++) {
+            int j, k;
+            for (j = 0; j < 2; j++) {
+                for (k = 0; k < 2; k++) {
+                    FF_ALLOCZ_OR_GOTO(s->avctx,
+                                      s->b_field_mv_table_base[i][j][k],
+                                      mv_table_size * 2 * sizeof(int16_t),
+                                      fail);
+                    s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
+                                                   s->mb_stride + 1;
+                }
+                FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j],
+                                  mb_array_size * 2 * sizeof(uint8_t), fail);
+                FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j],
+                                  mv_table_size * 2 * sizeof(int16_t), fail);
+                s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j]
+                                            + s->mb_stride + 1;
+            }
+            FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i],
+                              mb_array_size * 2 * sizeof(uint8_t), fail);
+        }
+    }
+    if (s->out_format == FMT_H263) {
+        /* cbp values */
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size, fail);
+        s->coded_block = s->coded_block_base + s->b8_stride + 1;
+
+        /* cbp, ac_pred, pred_dir */
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table,
+                          mb_array_size * sizeof(uint8_t), fail);
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table,
+                          mb_array_size * sizeof(uint8_t), fail);
+    }
+
+    if (s->h263_pred || s->h263_plus || !s->encoding) {
+        /* dc values */
+        // MN: we need these for  error resilience of intra-frames
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base,
+                          yc_size * sizeof(int16_t), fail);
+        s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
+        s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
+        s->dc_val[2] = s->dc_val[1] + c_size;
+        for (i = 0; i < yc_size; i++)
+            s->dc_val_base[i] = 1024;
+    }
+
+    /* which mb is a intra block */
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
+    memset(s->mbintra_table, 1, mb_array_size);
+
+    /* init macroblock skip table */
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
+    // Note the + 1 is for  a quicker mpeg4 slice_end detection
+
+    if ((s->avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
+        s->avctx->debug_mv) {
+        s->visualization_buffer[0] = av_malloc((s->mb_width * 16 +
+                    2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
+        s->visualization_buffer[1] = av_malloc((s->mb_width * 16 +
+                    2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
+        s->visualization_buffer[2] = av_malloc((s->mb_width * 16 +
+                    2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
+    }
+
+    return 0;
+fail:
+    return AVERROR(ENOMEM);
 }
 
 /**
  * init common structure for both encoder and decoder.
  * this assumes that some variables like width/height are already set
  */
-av_cold int MPV_common_init(MpegEncContext *s)
+av_cold int ff_MPV_common_init(MpegEncContext *s)
 {
-    int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y,
-        threads = (s->encoding ||
-                   (HAVE_THREADS &&
-                    s->avctx->active_thread_type & FF_THREAD_SLICE)) ?
-                  s->avctx->thread_count : 1;
+    int i;
+    int nb_slices = (HAVE_THREADS &&
+                     s->avctx->active_thread_type & FF_THREAD_SLICE) ?
+                    s->avctx->thread_count : 1;
 
-    if (s->codec_id == CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
+    if (s->encoding && s->avctx->slices)
+        nb_slices = s->avctx->slices;
+
+    if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
         s->mb_height = (s->height + 31) / 32 * 2;
-    else if (s->codec_id != CODEC_ID_H264)
+    else if (s->codec_id != AV_CODEC_ID_H264)
         s->mb_height = (s->height + 15) / 16;
 
-    if (s->avctx->pix_fmt == PIX_FMT_NONE) {
+    if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
         av_log(s->avctx, AV_LOG_ERROR,
-               "decoding to PIX_FMT_NONE is not supported.\n");
+               "decoding to AV_PIX_FMT_NONE is not supported.\n");
         return -1;
     }
 
-    if ((s->encoding || (s->avctx->active_thread_type & FF_THREAD_SLICE)) &&
-        (s->avctx->thread_count > MAX_THREADS ||
-         (s->avctx->thread_count > s->mb_height && s->mb_height))) {
-        int max_threads = FFMIN(MAX_THREADS, s->mb_height);
-        av_log(s->avctx, AV_LOG_WARNING,
-               "too many threads (%d), reducing to %d\n",
-               s->avctx->thread_count, max_threads);
-        threads = max_threads;
+    if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
+        int max_slices;
+        if (s->mb_height)
+            max_slices = FFMIN(MAX_THREADS, s->mb_height);
+        else
+            max_slices = MAX_THREADS;
+        av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
+               " reducing to %d\n", nb_slices, max_slices);
+        nb_slices = max_slices;
     }
 
     if ((s->width || s->height) &&
@@ -704,78 +924,19 @@ av_cold int MPV_common_init(MpegEncContext *s)
     s->flags2 = s->avctx->flags2;
 
     if (s->width && s->height) {
-        s->mb_width   = (s->width + 15) / 16;
-        s->mb_stride  = s->mb_width + 1;
-        s->b8_stride  = s->mb_width * 2 + 1;
-        s->b4_stride  = s->mb_width * 4 + 1;
-        mb_array_size = s->mb_height * s->mb_stride;
-        mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
-
         /* set chroma shifts */
-        avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,&(s->chroma_x_shift),
-                                      &(s->chroma_y_shift) );
-
-        /* set default edge pos, will be overriden
-         * in decode_header if needed */
-        s->h_edge_pos = s->mb_width * 16;
-        s->v_edge_pos = s->mb_height * 16;
-
-        s->mb_num     = s->mb_width * s->mb_height;
-
-        s->block_wrap[0] =
-        s->block_wrap[1] =
-        s->block_wrap[2] =
-        s->block_wrap[3] = s->b8_stride;
-        s->block_wrap[4] =
-        s->block_wrap[5] = s->mb_stride;
-
-        y_size  = s->b8_stride * (2 * s->mb_height + 1);
-        c_size  = s->mb_stride * (s->mb_height + 1);
-        yc_size = y_size + 2   * c_size;
+        av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
+                                         &s->chroma_x_shift,
+                                         &s->chroma_y_shift);
 
         /* convert fourcc to upper case */
         s->codec_tag          = avpriv_toupper4(s->avctx->codec_tag);
 
         s->stream_codec_tag   = avpriv_toupper4(s->avctx->stream_codec_tag);
 
-        s->avctx->coded_frame = (AVFrame *)&s->current_picture;
-
-        FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int),
-                          fail); // error ressilience code looks cleaner with this
-        for (y = 0; y < s->mb_height; y++)
-            for (x = 0; x < s->mb_width; x++)
-                s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
-
-        s->mb_index2xy[s->mb_height * s->mb_width] =
-                       (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
+        s->avctx->coded_frame = &s->current_picture.f;
 
         if (s->encoding) {
-            /* Allocate MV tables */
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base,
-                              mv_table_size * 2 * sizeof(int16_t), fail);
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base,
-                              mv_table_size * 2 * sizeof(int16_t), fail);
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base,
-                              mv_table_size * 2 * sizeof(int16_t), fail);
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base,
-                              mv_table_size * 2 * sizeof(int16_t), fail);
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base,
-                              mv_table_size * 2 * sizeof(int16_t), fail);
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base,
-                              mv_table_size * 2 * sizeof(int16_t), fail);
-            s->p_mv_table            = s->p_mv_table_base +
-                                       s->mb_stride + 1;
-            s->b_forw_mv_table       = s->b_forw_mv_table_base +
-                                       s->mb_stride + 1;
-            s->b_back_mv_table       = s->b_back_mv_table_base +
-                                       s->mb_stride + 1;
-            s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base +
-                                       s->mb_stride + 1;
-            s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base +
-                                       s->mb_stride + 1;
-            s->b_direct_mv_table     = s->b_direct_mv_table_base +
-                                       s->mb_stride + 1;
-
             if (s->msmpeg4_version) {
                 FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats,
                                   2 * 2 * (MAX_LEVEL + 1) *
@@ -783,13 +944,6 @@ av_cold int MPV_common_init(MpegEncContext *s)
             }
             FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
 
-            /* Allocate MB type table */
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size *
-                              sizeof(uint16_t), fail); // needed for encoding
-
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size *
-                              sizeof(int), fail);
-
             FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix,
                               64 * 32   * sizeof(int), fail);
             FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix,
@@ -814,140 +968,58 @@ av_cold int MPV_common_init(MpegEncContext *s)
     FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
                       s->picture_count * sizeof(Picture), fail);
     for (i = 0; i < s->picture_count; i++) {
-        avcodec_get_frame_defaults((AVFrame *) &s->picture[i]);
+        avcodec_get_frame_defaults(&s->picture[i].f);
     }
 
     if (s->width && s->height) {
-        FF_ALLOCZ_OR_GOTO(s->avctx, s->error_status_table,
-                          mb_array_size * sizeof(uint8_t), fail);
-
-        if (s->codec_id == CODEC_ID_MPEG4 ||
-            (s->flags & CODEC_FLAG_INTERLACED_ME)) {
-            /* interlaced direct mode decoding tables */
-            for (i = 0; i < 2; i++) {
-                int j, k;
-                for (j = 0; j < 2; j++) {
-                    for (k = 0; k < 2; k++) {
-                        FF_ALLOCZ_OR_GOTO(s->avctx,
-                                          s->b_field_mv_table_base[i][j][k],
-                                          mv_table_size * 2 * sizeof(int16_t),
-                                          fail);
-                        s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
-                                                       s->mb_stride + 1;
-                    }
-                    FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j],
-                                      mb_array_size * 2 * sizeof(uint8_t),
-                                      fail);
-                    FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j],
-                                      mv_table_size * 2 * sizeof(int16_t),
-                                      fail);
-                    s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j]
-                                                + s->mb_stride + 1;
-                }
-                FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i],
-                                  mb_array_size * 2 * sizeof(uint8_t),
-                                  fail);
-            }
-        }
-        if (s->out_format == FMT_H263) {
-            /* cbp values */
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size, fail);
-            s->coded_block = s->coded_block_base + s->b8_stride + 1;
-
-            /* cbp, ac_pred, pred_dir */
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table,
-                              mb_array_size * sizeof(uint8_t), fail);
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table,
-                              mb_array_size * sizeof(uint8_t), fail);
-        }
-
-        if (s->h263_pred || s->h263_plus || !s->encoding) {
-            /* dc values */
-            // MN: we need these for  error resilience of intra-frames
-            FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base,
-                              yc_size * sizeof(int16_t), fail);
-            s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
-            s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
-            s->dc_val[2] = s->dc_val[1] + c_size;
-            for (i = 0; i < yc_size; i++)
-                s->dc_val_base[i] = 1024;
-        }
-
-        /* which mb is a intra block */
-        FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
-        memset(s->mbintra_table, 1, mb_array_size);
-
-        /* init macroblock skip table */
-        FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
-        // Note the + 1 is for  a quicker mpeg4 slice_end detection
-        FF_ALLOCZ_OR_GOTO(s->avctx, s->prev_pict_types,
-                          PREV_PICT_TYPES_BUFFER_SIZE, fail);
+        if (init_context_frame(s))
+            goto fail;
 
         s->parse_context.state = -1;
-        if ((s->avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
-            s->avctx->debug_mv) {
-            s->visualization_buffer[0] = av_malloc((s->mb_width * 16 +
-                        2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
-            s->visualization_buffer[1] = av_malloc((s->mb_width * 16 +
-                        2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
-            s->visualization_buffer[2] = av_malloc((s->mb_width * 16 +
-                        2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
-        }
     }
 
     s->context_initialized = 1;
     s->thread_context[0]   = s;
 
     if (s->width && s->height) {
-        if (s->encoding || (HAVE_THREADS &&
-                            s->avctx->active_thread_type&FF_THREAD_SLICE)) {
-            for (i = 1; i < threads; i++) {
+        if (nb_slices > 1) {
+            for (i = 1; i < nb_slices; i++) {
                 s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
                 memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
             }
 
-            for (i = 0; i < threads; i++) {
-                if (init_duplicate_context(s->thread_context[i], s) < 0)
+            for (i = 0; i < nb_slices; i++) {
+                if (init_duplicate_context(s->thread_context[i]) < 0)
                     goto fail;
                     s->thread_context[i]->start_mb_y =
-                        (s->mb_height * (i) + s->avctx->thread_count / 2) /
-                        s->avctx->thread_count;
+                        (s->mb_height * (i) + nb_slices / 2) / nb_slices;
                     s->thread_context[i]->end_mb_y   =
-                        (s->mb_height * (i + 1) + s->avctx->thread_count / 2) /
-                        s->avctx->thread_count;
+                        (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
             }
         } else {
-            if (init_duplicate_context(s, s) < 0)
+            if (init_duplicate_context(s) < 0)
                 goto fail;
             s->start_mb_y = 0;
             s->end_mb_y   = s->mb_height;
         }
+        s->slice_context_count = nb_slices;
     }
 
     return 0;
  fail:
-    MPV_common_end(s);
+    ff_MPV_common_end(s);
     return -1;
 }
 
-/* init common structure for both encoder and decoder */
-void MPV_common_end(MpegEncContext *s)
+/**
+ * Frees and resets MpegEncContext fields depending on the resolution.
+ * Is used during resolution changes to avoid a full reinitialization of the
+ * codec.
+ */
+static int free_context_frame(MpegEncContext *s)
 {
     int i, j, k;
 
-    if (s->encoding || (HAVE_THREADS &&
-        s->avctx->active_thread_type & FF_THREAD_SLICE)) {
-        for (i = 0; i < s->avctx->thread_count; i++) {
-            free_duplicate_context(s->thread_context[i]);
-        }
-        for (i = 1; i < s->avctx->thread_count; i++) {
-            av_freep(&s->thread_context[i]);
-        }
-    } else free_duplicate_context(s);
-
-    av_freep(&s->parse_context.buffer);
-    s->parse_context.buffer_size = 0;
-
     av_freep(&s->mb_type);
     av_freep(&s->p_mv_table_base);
     av_freep(&s->b_forw_mv_table_base);
@@ -981,15 +1053,117 @@ void MPV_common_end(MpegEncContext *s)
     av_freep(&s->pred_dir_table);
 
     av_freep(&s->mbskip_table);
-    av_freep(&s->prev_pict_types);
+
+    av_freep(&s->error_status_table);
+    av_freep(&s->er_temp_buffer);
+    av_freep(&s->mb_index2xy);
+    av_freep(&s->lambda_table);
+    av_freep(&s->cplx_tab);
+    av_freep(&s->bits_tab);
+
+    s->linesize = s->uvlinesize = 0;
+
+    for (i = 0; i < 3; i++)
+        av_freep(&s->visualization_buffer[i]);
+
+    return 0;
+}
+
+int ff_MPV_common_frame_size_change(MpegEncContext *s)
+{
+    int i, err = 0;
+
+    if (s->slice_context_count > 1) {
+        for (i = 0; i < s->slice_context_count; i++) {
+            free_duplicate_context(s->thread_context[i]);
+        }
+        for (i = 1; i < s->slice_context_count; i++) {
+            av_freep(&s->thread_context[i]);
+        }
+    } else
+        free_duplicate_context(s);
+
+    free_context_frame(s);
+
+    if (s->picture)
+        for (i = 0; i < s->picture_count; i++) {
+                s->picture[i].needs_realloc = 1;
+        }
+
+    s->last_picture_ptr         =
+    s->next_picture_ptr         =
+    s->current_picture_ptr      = NULL;
+
+    // init
+    if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
+        s->mb_height = (s->height + 31) / 32 * 2;
+    else if (s->codec_id != AV_CODEC_ID_H264)
+        s->mb_height = (s->height + 15) / 16;
+
+    if ((s->width || s->height) &&
+        av_image_check_size(s->width, s->height, 0, s->avctx))
+        return AVERROR_INVALIDDATA;
+
+    if ((err = init_context_frame(s)))
+        goto fail;
+
+    s->thread_context[0]   = s;
+
+    if (s->width && s->height) {
+        int nb_slices = s->slice_context_count;
+        if (nb_slices > 1) {
+            for (i = 1; i < nb_slices; i++) {
+                s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
+                memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
+            }
+
+            for (i = 0; i < nb_slices; i++) {
+                if (init_duplicate_context(s->thread_context[i]) < 0)
+                    goto fail;
+                    s->thread_context[i]->start_mb_y =
+                        (s->mb_height * (i) + nb_slices / 2) / nb_slices;
+                    s->thread_context[i]->end_mb_y   =
+                        (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
+            }
+        } else {
+            if (init_duplicate_context(s) < 0)
+                goto fail;
+            s->start_mb_y = 0;
+            s->end_mb_y   = s->mb_height;
+        }
+        s->slice_context_count = nb_slices;
+    }
+
+    return 0;
+ fail:
+    ff_MPV_common_end(s);
+    return err;
+}
+
+/* init common structure for both encoder and decoder */
+void ff_MPV_common_end(MpegEncContext *s)
+{
+    int i;
+
+    if (s->slice_context_count > 1) {
+        for (i = 0; i < s->slice_context_count; i++) {
+            free_duplicate_context(s->thread_context[i]);
+        }
+        for (i = 1; i < s->slice_context_count; i++) {
+            av_freep(&s->thread_context[i]);
+        }
+        s->slice_context_count = 1;
+    } else free_duplicate_context(s);
+
+    av_freep(&s->parse_context.buffer);
+    s->parse_context.buffer_size = 0;
+
     av_freep(&s->bitstream_buffer);
     s->allocated_bitstream_buffer_size = 0;
 
     av_freep(&s->avctx->stats_out);
     av_freep(&s->ac_stats);
-    av_freep(&s->error_status_table);
-    av_freep(&s->mb_index2xy);
-    av_freep(&s->lambda_table);
+
     av_freep(&s->q_intra_matrix);
     av_freep(&s->q_inter_matrix);
     av_freep(&s->q_intra_matrix16);
@@ -1004,27 +1178,27 @@ void MPV_common_end(MpegEncContext *s)
         }
     }
     av_freep(&s->picture);
+
+    free_context_frame(s);
+
+    if (!(s->avctx->active_thread_type & FF_THREAD_FRAME))
+        avcodec_default_free_buffers(s->avctx);
+
     s->context_initialized      = 0;
     s->last_picture_ptr         =
     s->next_picture_ptr         =
     s->current_picture_ptr      = NULL;
     s->linesize = s->uvlinesize = 0;
-
-    for (i = 0; i < 3; i++)
-        av_freep(&s->visualization_buffer[i]);
-
-    if (!(s->avctx->active_thread_type & FF_THREAD_FRAME))
-        avcodec_default_free_buffers(s->avctx);
 }
 
-void init_rl(RLTable *rl,
-             uint8_t static_store[2][2 * MAX_RUN + MAX_LEVEL + 3])
+void ff_init_rl(RLTable *rl,
+                uint8_t static_store[2][2 * MAX_RUN + MAX_LEVEL + 3])
 {
     int8_t  max_level[MAX_RUN + 1], max_run[MAX_LEVEL + 1];
     uint8_t index_run[MAX_RUN + 1];
     int last, run, level, start, end, i;
 
-    /* If  table is static, we can quit if rl->max_level[0] is not NULL */
+    /* If table is static, we can quit if rl->max_level[0] is not NULL */
     if (static_store && rl->max_level[0])
         return;
 
@@ -1069,7 +1243,7 @@ void init_rl(RLTable *rl,
     }
 }
 
-void init_vlc_rl(RLTable *rl)
+void ff_init_vlc_rl(RLTable *rl)
 {
     int i, q;
 
@@ -1124,7 +1298,17 @@ void ff_release_unused_pictures(MpegEncContext*s, int remove_current)
     }
 }
 
-int ff_find_unused_picture(MpegEncContext *s, int shared)
+static inline int pic_is_unused(MpegEncContext *s, Picture *pic)
+{
+    if (pic->f.data[0] == NULL)
+        return 1;
+    if (pic->needs_realloc && !(pic->f.reference & DELAYED_PIC_REF))
+        if (!pic->owner2 || pic->owner2 == s)
+            return 1;
+    return 0;
+}
+
+static int find_unused_picture(MpegEncContext *s, int shared)
 {
     int i;
 
@@ -1135,93 +1319,109 @@ int ff_find_unused_picture(MpegEncContext *s, int shared)
         }
     } else {
         for (i = s->picture_range_start; i < s->picture_range_end; i++) {
-            if (s->picture[i].f.data[0] == NULL && s->picture[i].f.type != 0)
+            if (pic_is_unused(s, &s->picture[i]) && s->picture[i].f.type != 0)
                 return i; // FIXME
         }
         for (i = s->picture_range_start; i < s->picture_range_end; i++) {
-            if (s->picture[i].f.data[0] == NULL)
+            if (pic_is_unused(s, &s->picture[i]))
                 return i;
         }
     }
 
-    av_log(s->avctx, AV_LOG_FATAL,
-           "Internal error, picture buffer overflow\n");
-    /* We could return -1, but the codec would crash trying to draw into a
-     * non-existing frame anyway. This is safer than waiting for a random crash.
-     * Also the return of this is never useful, an encoder must only allocate
-     * as much as allowed in the specification. This has no relationship to how
-     * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
-     * enough for such valid streams).
-     * Plus, a decoder has to check stream validity and remove frames if too
-     * many reference frames are around. Waiting for "OOM" is not correct at
-     * all. Similarly, missing reference frames have to be replaced by
-     * interpolated/MC frames, anything else is a bug in the codec ...
-     */
-    abort();
-    return -1;
+    return AVERROR_INVALIDDATA;
+}
+
+int ff_find_unused_picture(MpegEncContext *s, int shared)
+{
+    int ret = find_unused_picture(s, shared);
+
+    if (ret >= 0 && ret < s->picture_range_end) {
+        if (s->picture[ret].needs_realloc) {
+            s->picture[ret].needs_realloc = 0;
+            free_picture(s, &s->picture[ret]);
+            avcodec_get_frame_defaults(&s->picture[ret].f);
+        }
+    }
+    return ret;
 }
 
-static void update_noise_reduction(MpegEncContext *s){
+static void update_noise_reduction(MpegEncContext *s)
+{
     int intra, i;
 
-    for(intra=0; intra<2; intra++){
-        if(s->dct_count[intra] > (1<<16)){
-            for(i=0; i<64; i++){
-                s->dct_error_sum[intra][i] >>=1;
+    for (intra = 0; intra < 2; intra++) {
+        if (s->dct_count[intra] > (1 << 16)) {
+            for (i = 0; i < 64; i++) {
+                s->dct_error_sum[intra][i] >>= 1;
             }
             s->dct_count[intra] >>= 1;
         }
 
-        for(i=0; i<64; i++){
-            s->dct_offset[intra][i]= (s->avctx->noise_reduction * s->dct_count[intra] + s->dct_error_sum[intra][i]/2) / (s->dct_error_sum[intra][i]+1);
+        for (i = 0; i < 64; i++) {
+            s->dct_offset[intra][i] = (s->avctx->noise_reduction *
+                                       s->dct_count[intra] +
+                                       s->dct_error_sum[intra][i] / 2) /
+                                      (s->dct_error_sum[intra][i] + 1);
         }
     }
 }
 
 /**
- * generic function for encode/decode called after coding/decoding the header and before a frame is coded/decoded
+ * generic function for encode/decode called after coding/decoding
+ * the header and before a frame is coded/decoded.
  */
-int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
+int ff_MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
 {
     int i;
     Picture *pic;
     s->mb_skipped = 0;
 
-    assert(s->last_picture_ptr==NULL || s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3);
-
-    /* mark&release old frames */
-    if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr && s->last_picture_ptr != s->next_picture_ptr && s->last_picture_ptr->f.data[0]) {
-      if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){
-          if (s->last_picture_ptr->owner2 == s)
-              free_frame_buffer(s, s->last_picture_ptr);
+    /* mark & release old frames */
+    if (s->out_format != FMT_H264 || s->codec_id == AV_CODEC_ID_SVQ3) {
+        if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
+            s->last_picture_ptr != s->next_picture_ptr &&
+            s->last_picture_ptr->f.data[0]) {
+            if (s->last_picture_ptr->owner2 == s)
+                free_frame_buffer(s, s->last_picture_ptr);
+        }
 
         /* release forgotten pictures */
-        /* if(mpeg124/h263) */
-        if(!s->encoding){
-            for(i=0; i<s->picture_count; i++){
-                if (s->picture[i].owner2 == s && s->picture[i].f.data[0] && &s->picture[i] != s->next_picture_ptr && s->picture[i].f.reference) {
+        /* if (mpeg124/h263) */
+        if (!s->encoding) {
+            for (i = 0; i < s->picture_count; i++) {
+                if (s->picture[i].owner2 == s && s->picture[i].f.data[0] &&
+                    &s->picture[i] != s->last_picture_ptr &&
+                    &s->picture[i] != s->next_picture_ptr &&
+                    s->picture[i].f.reference && !s->picture[i].needs_realloc) {
                     if (!(avctx->active_thread_type & FF_THREAD_FRAME))
-                        av_log(avctx, AV_LOG_ERROR, "releasing zombie picture\n");
+                        av_log(avctx, AV_LOG_ERROR,
+                               "releasing zombie picture\n");
                     free_frame_buffer(s, &s->picture[i]);
                 }
             }
         }
-      }
     }
 
-    if(!s->encoding){
+    if (!s->encoding) {
         ff_release_unused_pictures(s, 1);
 
-        if (s->current_picture_ptr && s->current_picture_ptr->f.data[0] == NULL)
-            pic= s->current_picture_ptr; //we already have a unused image (maybe it was set before reading the header)
-        else{
-            i= ff_find_unused_picture(s, 0);
-            pic= &s->picture[i];
+        if (s->current_picture_ptr &&
+            s->current_picture_ptr->f.data[0] == NULL) {
+            // we already have a unused image
+            // (maybe it was set before reading the header)
+            pic = s->current_picture_ptr;
+        } else {
+            i   = ff_find_unused_picture(s, 0);
+            if (i < 0) {
+                av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
+                return i;
+            }
+            pic = &s->picture[i];
         }
 
         pic->f.reference = 0;
-        if (!s->dropable){
-            if (s->codec_id == CODEC_ID_H264)
+        if (!s->droppable) {
+            if (s->codec_id == AV_CODEC_ID_H264)
                 pic->f.reference = s->picture_structure;
             else if (s->pict_type != AV_PICTURE_TYPE_B)
                 pic->f.reference = 3;
@@ -1229,75 +1429,123 @@ int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
 
         pic->f.coded_picture_number = s->coded_picture_number++;
 
-        if(ff_alloc_picture(s, pic, 0) < 0)
+        if (ff_alloc_picture(s, pic, 0) < 0)
             return -1;
 
-        s->current_picture_ptr= pic;
-        //FIXME use only the vars from current_pic
+        s->current_picture_ptr = pic;
+        // FIXME use only the vars from current_pic
         s->current_picture_ptr->f.top_field_first = s->top_field_first;
-        if(s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO) {
-            if(s->picture_structure != PICT_FRAME)
-                s->current_picture_ptr->f.top_field_first = (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
+        if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
+            s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
+            if (s->picture_structure != PICT_FRAME)
+                s->current_picture_ptr->f.top_field_first =
+                    (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
         }
-        s->current_picture_ptr->f.interlaced_frame = !s->progressive_frame && !s->progressive_sequence;
-        s->current_picture_ptr->field_picture = s->picture_structure != PICT_FRAME;
+        s->current_picture_ptr->f.interlaced_frame = !s->progressive_frame &&
+                                                     !s->progressive_sequence;
+        s->current_picture_ptr->field_picture      =  s->picture_structure != PICT_FRAME;
     }
 
     s->current_picture_ptr->f.pict_type = s->pict_type;
-//    if(s->flags && CODEC_FLAG_QSCALE)
-  //      s->current_picture_ptr->quality= s->new_picture_ptr->quality;
+    // if (s->flags && CODEC_FLAG_QSCALE)
+    //     s->current_picture_ptr->quality = s->new_picture_ptr->quality;
     s->current_picture_ptr->f.key_frame = s->pict_type == AV_PICTURE_TYPE_I;
 
     ff_copy_picture(&s->current_picture, s->current_picture_ptr);
 
     if (s->pict_type != AV_PICTURE_TYPE_B) {
-        s->last_picture_ptr= s->next_picture_ptr;
-        if(!s->dropable)
-            s->next_picture_ptr= s->current_picture_ptr;
-    }
-/*    av_log(s->avctx, AV_LOG_DEBUG, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n", s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
-        s->last_picture_ptr    ? s->last_picture_ptr->f.data[0]    : NULL,
-        s->next_picture_ptr    ? s->next_picture_ptr->f.data[0]    : NULL,
-        s->current_picture_ptr ? s->current_picture_ptr->f.data[0] : NULL,
-        s->pict_type, s->dropable);*/
-
-    if(s->codec_id != CODEC_ID_H264){
-        if ((s->last_picture_ptr == NULL || s->last_picture_ptr->f.data[0] == NULL) &&
-           (s->pict_type!=AV_PICTURE_TYPE_I || s->picture_structure != PICT_FRAME)){
+        s->last_picture_ptr = s->next_picture_ptr;
+        if (!s->droppable)
+            s->next_picture_ptr = s->current_picture_ptr;
+    }
+    av_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
+            s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
+            s->last_picture_ptr    ? s->last_picture_ptr->f.data[0]    : NULL,
+            s->next_picture_ptr    ? s->next_picture_ptr->f.data[0]    : NULL,
+            s->current_picture_ptr ? s->current_picture_ptr->f.data[0] : NULL,
+            s->pict_type, s->droppable);
+
+    if (s->codec_id != AV_CODEC_ID_H264) {
+        if ((s->last_picture_ptr == NULL ||
+             s->last_picture_ptr->f.data[0] == NULL) &&
+            (s->pict_type != AV_PICTURE_TYPE_I ||
+             s->picture_structure != PICT_FRAME)) {
+            int h_chroma_shift, v_chroma_shift;
+            av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
+                                             &h_chroma_shift, &v_chroma_shift);
             if (s->pict_type != AV_PICTURE_TYPE_I)
-                av_log(avctx, AV_LOG_ERROR, "warning: first frame is no keyframe\n");
+                av_log(avctx, AV_LOG_ERROR,
+                       "warning: first frame is no keyframe\n");
             else if (s->picture_structure != PICT_FRAME)
-                av_log(avctx, AV_LOG_INFO, "allocate dummy last picture for field based first keyframe\n");
+                av_log(avctx, AV_LOG_INFO,
+                       "allocate dummy last picture for field based first keyframe\n");
 
             /* Allocate a dummy frame */
-            i= ff_find_unused_picture(s, 0);
-            s->last_picture_ptr= &s->picture[i];
-            if(ff_alloc_picture(s, s->last_picture_ptr, 0) < 0)
+            i = ff_find_unused_picture(s, 0);
+            if (i < 0) {
+                av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
+                return i;
+            }
+            s->last_picture_ptr = &s->picture[i];
+            if (ff_alloc_picture(s, s->last_picture_ptr, 0) < 0) {
+                s->last_picture_ptr = NULL;
                 return -1;
-            ff_thread_report_progress((AVFrame*)s->last_picture_ptr, INT_MAX, 0);
-            ff_thread_report_progress((AVFrame*)s->last_picture_ptr, INT_MAX, 1);
+            }
+
+            memset(s->last_picture_ptr->f.data[0], 0,
+                   avctx->height * s->last_picture_ptr->f.linesize[0]);
+            memset(s->last_picture_ptr->f.data[1], 0x80,
+                   (avctx->height >> v_chroma_shift) *
+                   s->last_picture_ptr->f.linesize[1]);
+            memset(s->last_picture_ptr->f.data[2], 0x80,
+                   (avctx->height >> v_chroma_shift) *
+                   s->last_picture_ptr->f.linesize[2]);
+
+            ff_thread_report_progress(&s->last_picture_ptr->f, INT_MAX, 0);
+            ff_thread_report_progress(&s->last_picture_ptr->f, INT_MAX, 1);
+            s->last_picture_ptr->f.reference = 3;
         }
-        if ((s->next_picture_ptr == NULL || s->next_picture_ptr->f.data[0] == NULL) && s->pict_type == AV_PICTURE_TYPE_B) {
+        if ((s->next_picture_ptr == NULL ||
+             s->next_picture_ptr->f.data[0] == NULL) &&
+            s->pict_type == AV_PICTURE_TYPE_B) {
             /* Allocate a dummy frame */
-            i= ff_find_unused_picture(s, 0);
-            s->next_picture_ptr= &s->picture[i];
-            if(ff_alloc_picture(s, s->next_picture_ptr, 0) < 0)
+            i = ff_find_unused_picture(s, 0);
+            if (i < 0) {
+                av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
+                return i;
+            }
+            s->next_picture_ptr = &s->picture[i];
+            if (ff_alloc_picture(s, s->next_picture_ptr, 0) < 0) {
+                s->next_picture_ptr = NULL;
                 return -1;
-            ff_thread_report_progress((AVFrame*)s->next_picture_ptr, INT_MAX, 0);
-            ff_thread_report_progress((AVFrame*)s->next_picture_ptr, INT_MAX, 1);
+            }
+            ff_thread_report_progress(&s->next_picture_ptr->f, INT_MAX, 0);
+            ff_thread_report_progress(&s->next_picture_ptr->f, INT_MAX, 1);
+            s->next_picture_ptr->f.reference = 3;
         }
     }
 
-    if(s->last_picture_ptr) ff_copy_picture(&s->last_picture, s->last_picture_ptr);
-    if(s->next_picture_ptr) ff_copy_picture(&s->next_picture, s->next_picture_ptr);
+    if (s->last_picture_ptr)
+        ff_copy_picture(&s->last_picture, s->last_picture_ptr);
+    if (s->next_picture_ptr)
+        ff_copy_picture(&s->next_picture, s->next_picture_ptr);
+
+    if (HAVE_THREADS && (avctx->active_thread_type & FF_THREAD_FRAME)) {
+        if (s->next_picture_ptr)
+            s->next_picture_ptr->owner2 = s;
+        if (s->last_picture_ptr)
+            s->last_picture_ptr->owner2 = s;
+    }
 
-    assert(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr && s->last_picture_ptr->f.data[0]));
+    assert(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
+                                                 s->last_picture_ptr->f.data[0]));
 
-    if(s->picture_structure!=PICT_FRAME && s->out_format != FMT_H264){
+    if (s->picture_structure!= PICT_FRAME && s->out_format != FMT_H264) {
         int i;
-        for(i=0; i<4; i++){
-            if(s->picture_structure == PICT_BOTTOM_FIELD){
-                 s->current_picture.f.data[i] += s->current_picture.f.linesize[i];
+        for (i = 0; i < 4; i++) {
+            if (s->picture_structure == PICT_BOTTOM_FIELD) {
+                s->current_picture.f.data[i] +=
+                    s->current_picture.f.linesize[i];
             }
             s->current_picture.f.linesize[i] *= 2;
             s->last_picture.f.linesize[i]    *= 2;
@@ -1307,95 +1555,101 @@ int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
 
     s->err_recognition = avctx->err_recognition;
 
-    /* set dequantizer, we can't do it during init as it might change for mpeg4
-       and we can't do it in the header decode as init is not called for mpeg4 there yet */
-    if(s->mpeg_quant || s->codec_id == CODEC_ID_MPEG2VIDEO){
+    /* set dequantizer, we can't do it during init as
+     * it might change for mpeg4 and we can't do it in the header
+     * decode as init is not called for mpeg4 there yet */
+    if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
-    }else if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
+    } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
-    }else{
+    } else {
         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
     }
 
-    if(s->dct_error_sum){
+    if (s->dct_error_sum) {
         assert(s->avctx->noise_reduction && s->encoding);
-
         update_noise_reduction(s);
     }
 
-    if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration)
+    if (CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration)
         return ff_xvmc_field_start(s, avctx);
 
     return 0;
 }
 
-/* generic function for encode/decode called after a frame has been coded/decoded */
-void MPV_frame_end(MpegEncContext *s)
+/* generic function for encode/decode called after a
+ * frame has been coded/decoded. */
+void ff_MPV_frame_end(MpegEncContext *s)
 {
     int i;
     /* redraw edges for the frame if decoding didn't complete */
-    //just to make sure that all data is rendered.
-    if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
+    // just to make sure that all data is rendered.
+    if (CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration) {
         ff_xvmc_field_end(s);
-   }else if((s->error_count || s->encoding)
-       && !s->avctx->hwaccel
-       && !(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
-       && s->unrestricted_mv
-       && s->current_picture.f.reference
-       && !s->intra_only
-       && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
-            int hshift = av_pix_fmt_descriptors[s->avctx->pix_fmt].log2_chroma_w;
-            int vshift = av_pix_fmt_descriptors[s->avctx->pix_fmt].log2_chroma_h;
-            s->dsp.draw_edges(s->current_picture.f.data[0], s->linesize,
-                              s->h_edge_pos             , s->v_edge_pos,
-                              EDGE_WIDTH        , EDGE_WIDTH        , EDGE_TOP | EDGE_BOTTOM);
-            s->dsp.draw_edges(s->current_picture.f.data[1], s->uvlinesize,
-                              s->h_edge_pos>>hshift, s->v_edge_pos>>vshift,
-                              EDGE_WIDTH>>hshift, EDGE_WIDTH>>vshift, EDGE_TOP | EDGE_BOTTOM);
-            s->dsp.draw_edges(s->current_picture.f.data[2], s->uvlinesize,
-                              s->h_edge_pos>>hshift, s->v_edge_pos>>vshift,
-                              EDGE_WIDTH>>hshift, EDGE_WIDTH>>vshift, EDGE_TOP | EDGE_BOTTOM);
+   } else if ((s->error_count || s->encoding) &&
+              !s->avctx->hwaccel &&
+              !(s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) &&
+              s->unrestricted_mv &&
+              s->current_picture.f.reference &&
+              !s->intra_only &&
+              !(s->flags & CODEC_FLAG_EMU_EDGE)) {
+       const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
+       int hshift = desc->log2_chroma_w;
+       int vshift = desc->log2_chroma_h;
+       s->dsp.draw_edges(s->current_picture.f.data[0], s->linesize,
+                         s->h_edge_pos, s->v_edge_pos,
+                         EDGE_WIDTH, EDGE_WIDTH,
+                         EDGE_TOP | EDGE_BOTTOM);
+       s->dsp.draw_edges(s->current_picture.f.data[1], s->uvlinesize,
+                         s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
+                         EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
+                         EDGE_TOP | EDGE_BOTTOM);
+       s->dsp.draw_edges(s->current_picture.f.data[2], s->uvlinesize,
+                         s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
+                         EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
+                         EDGE_TOP | EDGE_BOTTOM);
     }
 
     emms_c();
 
-    s->last_pict_type    = s->pict_type;
-    s->last_lambda_for[s->pict_type] = s->current_picture_ptr->f.quality;
-    if(s->pict_type!=AV_PICTURE_TYPE_B){
-        s->last_non_b_pict_type= s->pict_type;
+    s->last_pict_type                 = s->pict_type;
+    s->last_lambda_for [s->pict_type] = s->current_picture_ptr->f.quality;
+    if (s->pict_type!= AV_PICTURE_TYPE_B) {
+        s->last_non_b_pict_type = s->pict_type;
     }
 #if 0
-        /* copy back current_picture variables */
-    for(i=0; i<MAX_PICTURE_COUNT; i++){
-        if(s->picture[i].f.data[0] == s->current_picture.f.data[0]){
-            s->picture[i]= s->current_picture;
+    /* copy back current_picture variables */
+    for (i = 0; i < MAX_PICTURE_COUNT; i++) {
+        if (s->picture[i].f.data[0] == s->current_picture.f.data[0]) {
+            s->picture[i] = s->current_picture;
             break;
         }
     }
-    assert(i<MAX_PICTURE_COUNT);
+    assert(i < MAX_PICTURE_COUNT);
 #endif
 
-    if(s->encoding){
+    if (s->encoding) {
         /* release non-reference frames */
-        for(i=0; i<s->picture_count; i++){
-            if (s->picture[i].f.data[0] && !s->picture[i].f.reference /*&& s->picture[i].type != FF_BUFFER_TYPE_SHARED*/) {
+        for (i = 0; i < s->picture_count; i++) {
+            if (s->picture[i].f.data[0] && !s->picture[i].f.reference
+                /* && s->picture[i].type != FF_BUFFER_TYPE_SHARED */) {
                 free_frame_buffer(s, &s->picture[i]);
             }
         }
     }
     // clear copies, to avoid confusion
 #if 0
-    memset(&s->last_picture, 0, sizeof(Picture));
-    memset(&s->next_picture, 0, sizeof(Picture));
+    memset(&s->last_picture,    0, sizeof(Picture));
+    memset(&s->next_picture,    0, sizeof(Picture));
     memset(&s->current_picture, 0, sizeof(Picture));
 #endif
-    s->avctx->coded_frame= (AVFrame*)s->current_picture_ptr;
+    s->avctx->coded_frame = &s->current_picture_ptr->f;
 
-    if (s->codec_id != CODEC_ID_H264 && s->current_picture.f.reference) {
-        ff_thread_report_progress((AVFrame*)s->current_picture_ptr, s->mb_height-1, 0);
+    if (s->codec_id != AV_CODEC_ID_H264 && s->current_picture.f.reference) {
+        ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
     }
 }
 
@@ -1406,44 +1660,48 @@ void MPV_frame_end(MpegEncContext *s)
  * @param stride stride/linesize of the image
  * @param color color of the arrow
  */
-static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
+static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey,
+                      int w, int h, int stride, int color)
+{
     int x, y, fr, f;
 
-    sx= av_clip(sx, 0, w-1);
-    sy= av_clip(sy, 0, h-1);
-    ex= av_clip(ex, 0, w-1);
-    ey= av_clip(ey, 0, h-1);
+    sx = av_clip(sx, 0, w - 1);
+    sy = av_clip(sy, 0, h - 1);
+    ex = av_clip(ex, 0, w - 1);
+    ey = av_clip(ey, 0, h - 1);
 
-    buf[sy*stride + sx]+= color;
+    buf[sy * stride + sx] += color;
 
-    if(FFABS(ex - sx) > FFABS(ey - sy)){
-        if(sx > ex){
+    if (FFABS(ex - sx) > FFABS(ey - sy)) {
+        if (sx > ex) {
             FFSWAP(int, sx, ex);
             FFSWAP(int, sy, ey);
         }
-        buf+= sx + sy*stride;
-        ex-= sx;
-        f= ((ey-sy)<<16)/ex;
-        for(x= 0; x <= ex; x++){
-            y = (x*f)>>16;
-            fr= (x*f)&0xFFFF;
-            buf[ y   *stride + x]+= (color*(0x10000-fr))>>16;
-            buf[(y+1)*stride + x]+= (color*         fr )>>16;
+        buf += sx + sy * stride;
+        ex  -= sx;
+        f    = ((ey - sy) << 16) / ex;
+        for (x = 0; x <= ex; x++) {
+            y  = (x * f) >> 16;
+            fr = (x * f) & 0xFFFF;
+            buf[y * stride + x]       += (color * (0x10000 - fr)) >> 16;
+            buf[(y + 1) * stride + x] += (color *            fr ) >> 16;
         }
-    }else{
-        if(sy > ey){
+    } else {
+        if (sy > ey) {
             FFSWAP(int, sx, ex);
             FFSWAP(int, sy, ey);
         }
-        buf+= sx + sy*stride;
-        ey-= sy;
-        if(ey) f= ((ex-sx)<<16)/ey;
-        else   f= 0;
-        for(y= 0; y <= ey; y++){
-            x = (y*f)>>16;
-            fr= (y*f)&0xFFFF;
-            buf[y*stride + x  ]+= (color*(0x10000-fr))>>16;
-            buf[y*stride + x+1]+= (color*         fr )>>16;
+        buf += sx + sy * stride;
+        ey  -= sy;
+        if (ey)
+            f  = ((ex - sx) << 16) / ey;
+        else
+            f = 0;
+        for (y = 0; y = ey; y++) {
+            x  = (y * f) >> 16;
+            fr = (y * f) & 0xFFFF;
+            buf[y * stride + x]     += (color * (0x10000 - fr)) >> 16;
+            buf[y * stride + x + 1] += (color *            fr ) >> 16;
         }
     }
 }
@@ -1455,25 +1713,27 @@ static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h
  * @param stride stride/linesize of the image
  * @param color color of the arrow
  */
-static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
+static void draw_arrow(uint8_t *buf, int sx, int sy, int ex,
+                       int ey, int w, int h, int stride, int color)
+{
     int dx,dy;
 
-    sx= av_clip(sx, -100, w+100);
-    sy= av_clip(sy, -100, h+100);
-    ex= av_clip(ex, -100, w+100);
-    ey= av_clip(ey, -100, h+100);
+    sx = av_clip(sx, -100, w + 100);
+    sy = av_clip(sy, -100, h + 100);
+    ex = av_clip(ex, -100, w + 100);
+    ey = av_clip(ey, -100, h + 100);
 
-    dx= ex - sx;
-    dy= ey - sy;
+    dx = ex - sx;
+    dy = ey - sy;
 
-    if(dx*dx + dy*dy > 3*3){
-        int rx=  dx + dy;
-        int ry= -dx + dy;
-        int length= ff_sqrt((rx*rx + ry*ry)<<8);
+    if (dx * dx + dy * dy > 3 * 3) {
+        int rx =  dx + dy;
+        int ry = -dx + dy;
+        int length = ff_sqrt((rx * rx + ry * ry) << 8);
 
-        //FIXME subpixel accuracy
-        rx= ROUNDED_DIV(rx*3<<4, length);
-        ry= ROUNDED_DIV(ry*3<<4, length);
+        // FIXME subpixel accuracy
+        rx = ROUNDED_DIV(rx * 3 << 4, length);
+        ry = ROUNDED_DIV(ry * 3 << 4, length);
 
         draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
         draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
@@ -1482,595 +1742,322 @@ static void draw_arrow(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int
 }
 
 /**
- * Print debuging info for the given picture.
+ * Print debugging info for the given picture.
  */
-void ff_print_debug_info(MpegEncContext *s, AVFrame *pict){
-
-    if(s->avctx->hwaccel || !pict || !pict->mb_type) return;
+void ff_print_debug_info(MpegEncContext *s, AVFrame *pict)
+{
+    if (s->avctx->hwaccel || !pict || !pict->mb_type)
+        return;
 
-    if(s->avctx->debug&(FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)){
+    if (s->avctx->debug & (FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)) {
         int x,y;
 
         av_log(s->avctx,AV_LOG_DEBUG,"New frame, type: ");
         switch (pict->pict_type) {
-            case AV_PICTURE_TYPE_I: av_log(s->avctx,AV_LOG_DEBUG,"I\n"); break;
-            case AV_PICTURE_TYPE_P: av_log(s->avctx,AV_LOG_DEBUG,"P\n"); break;
-            case AV_PICTURE_TYPE_B: av_log(s->avctx,AV_LOG_DEBUG,"B\n"); break;
-            case AV_PICTURE_TYPE_S: av_log(s->avctx,AV_LOG_DEBUG,"S\n"); break;
-            case AV_PICTURE_TYPE_SI: av_log(s->avctx,AV_LOG_DEBUG,"SI\n"); break;
-            case AV_PICTURE_TYPE_SP: av_log(s->avctx,AV_LOG_DEBUG,"SP\n"); break;
+        case AV_PICTURE_TYPE_I:
+            av_log(s->avctx,AV_LOG_DEBUG,"I\n");
+            break;
+        case AV_PICTURE_TYPE_P:
+            av_log(s->avctx,AV_LOG_DEBUG,"P\n");
+            break;
+        case AV_PICTURE_TYPE_B:
+            av_log(s->avctx,AV_LOG_DEBUG,"B\n");
+            break;
+        case AV_PICTURE_TYPE_S:
+            av_log(s->avctx,AV_LOG_DEBUG,"S\n");
+            break;
+        case AV_PICTURE_TYPE_SI:
+            av_log(s->avctx,AV_LOG_DEBUG,"SI\n");
+            break;
+        case AV_PICTURE_TYPE_SP:
+            av_log(s->avctx,AV_LOG_DEBUG,"SP\n");
+            break;
         }
-        for(y=0; y<s->mb_height; y++){
-            for(x=0; x<s->mb_width; x++){
-                if(s->avctx->debug&FF_DEBUG_SKIP){
-                    int count= s->mbskip_table[x + y*s->mb_stride];
-                    if(count>9) count=9;
+        for (y = 0; y < s->mb_height; y++) {
+            for (x = 0; x < s->mb_width; x++) {
+                if (s->avctx->debug & FF_DEBUG_SKIP) {
+                    int count = s->mbskip_table[x + y * s->mb_stride];
+                    if (count > 9)
+                        count = 9;
                     av_log(s->avctx, AV_LOG_DEBUG, "%1d", count);
                 }
-                if(s->avctx->debug&FF_DEBUG_QP){
-                    av_log(s->avctx, AV_LOG_DEBUG, "%2d", pict->qscale_table[x + y*s->mb_stride]);
+                if (s->avctx->debug & FF_DEBUG_QP) {
+                    av_log(s->avctx, AV_LOG_DEBUG, "%2d",
+                           pict->qscale_table[x + y * s->mb_stride]);
                 }
-                if(s->avctx->debug&FF_DEBUG_MB_TYPE){
-                    int mb_type= pict->mb_type[x + y*s->mb_stride];
-                    //Type & MV direction
-                    if(IS_PCM(mb_type))
+                if (s->avctx->debug & FF_DEBUG_MB_TYPE) {
+                    int mb_type = pict->mb_type[x + y * s->mb_stride];
+                    // Type & MV direction
+                    if (IS_PCM(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "P");
-                    else if(IS_INTRA(mb_type) && IS_ACPRED(mb_type))
+                    else if (IS_INTRA(mb_type) && IS_ACPRED(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "A");
-                    else if(IS_INTRA4x4(mb_type))
+                    else if (IS_INTRA4x4(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "i");
-                    else if(IS_INTRA16x16(mb_type))
+                    else if (IS_INTRA16x16(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "I");
-                    else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type))
+                    else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "d");
-                    else if(IS_DIRECT(mb_type))
+                    else if (IS_DIRECT(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "D");
-                    else if(IS_GMC(mb_type) && IS_SKIP(mb_type))
+                    else if (IS_GMC(mb_type) && IS_SKIP(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "g");
-                    else if(IS_GMC(mb_type))
+                    else if (IS_GMC(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "G");
-                    else if(IS_SKIP(mb_type))
+                    else if (IS_SKIP(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "S");
-                    else if(!USES_LIST(mb_type, 1))
+                    else if (!USES_LIST(mb_type, 1))
                         av_log(s->avctx, AV_LOG_DEBUG, ">");
-                    else if(!USES_LIST(mb_type, 0))
+                    else if (!USES_LIST(mb_type, 0))
                         av_log(s->avctx, AV_LOG_DEBUG, "<");
-                    else{
+                    else {
                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
                         av_log(s->avctx, AV_LOG_DEBUG, "X");
                     }
 
-                    //segmentation
-                    if(IS_8X8(mb_type))
+                    // segmentation
+                    if (IS_8X8(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "+");
-                    else if(IS_16X8(mb_type))
+                    else if (IS_16X8(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "-");
-                    else if(IS_8X16(mb_type))
+                    else if (IS_8X16(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "|");
-                    else if(IS_INTRA(mb_type) || IS_16X16(mb_type))
+                    else if (IS_INTRA(mb_type) || IS_16X16(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, " ");
                     else
                         av_log(s->avctx, AV_LOG_DEBUG, "?");
 
 
-                    if(IS_INTERLACED(mb_type))
+                    if (IS_INTERLACED(mb_type))
                         av_log(s->avctx, AV_LOG_DEBUG, "=");
                     else
                         av_log(s->avctx, AV_LOG_DEBUG, " ");
                 }
-//                av_log(s->avctx, AV_LOG_DEBUG, " ");
             }
             av_log(s->avctx, AV_LOG_DEBUG, "\n");
         }
     }
 
     if ((s->avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
-        s->avctx->debug_mv) {
-        const int shift= 1 + s->quarter_sample;
+        (s->avctx->debug_mv)) {
+        const int shift = 1 + s->quarter_sample;
         int mb_y;
         uint8_t *ptr;
         int i;
         int h_chroma_shift, v_chroma_shift, block_height;
-        const int width = s->avctx->width;
-        const int height= s->avctx->height;
-        const int mv_sample_log2= 4 - pict->motion_subsample_log2;
-        const int mv_stride= (s->mb_width << mv_sample_log2) + (s->codec_id == CODEC_ID_H264 ? 0 : 1);
-        s->low_delay=0; //needed to see the vectors without trashing the buffers
-
-        avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
-        for(i=0; i<3; i++){
-            memcpy(s->visualization_buffer[i], pict->data[i], (i==0) ? pict->linesize[i]*height:pict->linesize[i]*height >> v_chroma_shift);
-            pict->data[i]= s->visualization_buffer[i];
+        const int width          = s->avctx->width;
+        const int height         = s->avctx->height;
+        const int mv_sample_log2 = 4 - pict->motion_subsample_log2;
+        const int mv_stride      = (s->mb_width << mv_sample_log2) +
+                                   (s->codec_id == AV_CODEC_ID_H264 ? 0 : 1);
+        s->low_delay = 0; // needed to see the vectors without trashing the buffers
+
+        av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
+                                         &h_chroma_shift, &v_chroma_shift);
+        for (i = 0; i < 3; i++) {
+            memcpy(s->visualization_buffer[i], pict->data[i],
+                   (i == 0) ? pict->linesize[i] * height:
+                              pict->linesize[i] * height >> v_chroma_shift);
+            pict->data[i] = s->visualization_buffer[i];
         }
-        pict->type= FF_BUFFER_TYPE_COPY;
-        ptr= pict->data[0];
-        block_height = 16>>v_chroma_shift;
+        pict->type   = FF_BUFFER_TYPE_COPY;
+        ptr          = pict->data[0];
+        block_height = 16 >> v_chroma_shift;
 
-        for(mb_y=0; mb_y<s->mb_height; mb_y++){
+        for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
             int mb_x;
-            for(mb_x=0; mb_x<s->mb_width; mb_x++){
-                const int mb_index= mb_x + mb_y*s->mb_stride;
-                if (s->avctx->debug_mv && pict->motion_val) {
-                  int type;
-                  for(type=0; type<3; type++){
-                    int direction = 0;
-                    switch (type) {
-                      case 0: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_P_FOR)) || (pict->pict_type!=AV_PICTURE_TYPE_P))
+            for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
+                const int mb_index = mb_x + mb_y * s->mb_stride;
+                if ((s->avctx->debug_mv) && pict->motion_val) {
+                    int type;
+                    for (type = 0; type < 3; type++) {
+                        int direction = 0;
+                        switch (type) {
+                        case 0:
+                            if ((!(s->avctx->debug_mv & FF_DEBUG_VIS_MV_P_FOR)) ||
+                                (pict->pict_type!= AV_PICTURE_TYPE_P))
                                 continue;
-                              direction = 0;
-                              break;
-                      case 1: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_FOR)) || (pict->pict_type!=AV_PICTURE_TYPE_B))
+                            direction = 0;
+                            break;
+                        case 1:
+                            if ((!(s->avctx->debug_mv & FF_DEBUG_VIS_MV_B_FOR)) ||
+                                (pict->pict_type!= AV_PICTURE_TYPE_B))
                                 continue;
-                              direction = 0;
-                              break;
-                      case 2: if ((!(s->avctx->debug_mv&FF_DEBUG_VIS_MV_B_BACK)) || (pict->pict_type!=AV_PICTURE_TYPE_B))
+                            direction = 0;
+                            break;
+                        case 2:
+                            if ((!(s->avctx->debug_mv & FF_DEBUG_VIS_MV_B_BACK)) ||
+                                (pict->pict_type!= AV_PICTURE_TYPE_B))
                                 continue;
-                              direction = 1;
-                              break;
-                    }
-                    if(!USES_LIST(pict->mb_type[mb_index], direction))
-                        continue;
-
-                    if(IS_8X8(pict->mb_type[mb_index])){
-                      int i;
-                      for(i=0; i<4; i++){
-                        int sx= mb_x*16 + 4 + 8*(i&1);
-                        int sy= mb_y*16 + 4 + 8*(i>>1);
-                        int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
-                        int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
-                        int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
-                        draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
-                      }
-                    }else if(IS_16X8(pict->mb_type[mb_index])){
-                      int i;
-                      for(i=0; i<2; i++){
-                        int sx=mb_x*16 + 8;
-                        int sy=mb_y*16 + 4 + 8*i;
-                        int xy= (mb_x*2 + (mb_y*2 + i)*mv_stride) << (mv_sample_log2-1);
-                        int mx=(pict->motion_val[direction][xy][0]>>shift);
-                        int my=(pict->motion_val[direction][xy][1]>>shift);
-
-                        if(IS_INTERLACED(pict->mb_type[mb_index]))
-                            my*=2;
-
-                        draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
-                      }
-                    }else if(IS_8X16(pict->mb_type[mb_index])){
-                      int i;
-                      for(i=0; i<2; i++){
-                        int sx=mb_x*16 + 4 + 8*i;
-                        int sy=mb_y*16 + 8;
-                        int xy= (mb_x*2 + i + mb_y*2*mv_stride) << (mv_sample_log2-1);
-                        int mx=(pict->motion_val[direction][xy][0]>>shift);
-                        int my=(pict->motion_val[direction][xy][1]>>shift);
-
-                        if(IS_INTERLACED(pict->mb_type[mb_index]))
-                            my*=2;
-
-                        draw_arrow(ptr, sx, sy, mx+sx, my+sy, width, height, s->linesize, 100);
-                      }
-                    }else{
-                      int sx= mb_x*16 + 8;
-                      int sy= mb_y*16 + 8;
-                      int xy= (mb_x + mb_y*mv_stride) << mv_sample_log2;
-                      int mx= (pict->motion_val[direction][xy][0]>>shift) + sx;
-                      int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
-                      draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
+                            direction = 1;
+                            break;
+                        }
+                        if (!USES_LIST(pict->mb_type[mb_index], direction))
+                            continue;
+
+                        if (IS_8X8(pict->mb_type[mb_index])) {
+                            int i;
+                            for (i = 0; i < 4; i++) {
+                                int sx = mb_x * 16 + 4 + 8 * (i & 1);
+                                int sy = mb_y * 16 + 4 + 8 * (i >> 1);
+                                int xy = (mb_x * 2 + (i & 1) +
+                                          (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
+                                int mx = (pict->motion_val[direction][xy][0] >> shift) + sx;
+                                int my = (pict->motion_val[direction][xy][1] >> shift) + sy;
+                                draw_arrow(ptr, sx, sy, mx, my, width,
+                                           height, s->linesize, 100);
+                            }
+                        } else if (IS_16X8(pict->mb_type[mb_index])) {
+                            int i;
+                            for (i = 0; i < 2; i++) {
+                                int sx = mb_x * 16 + 8;
+                                int sy = mb_y * 16 + 4 + 8 * i;
+                                int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
+                                int mx = (pict->motion_val[direction][xy][0] >> shift);
+                                int my = (pict->motion_val[direction][xy][1] >> shift);
+
+                                if (IS_INTERLACED(pict->mb_type[mb_index]))
+                                    my *= 2;
+
+                            draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
+                                       height, s->linesize, 100);
+                            }
+                        } else if (IS_8X16(pict->mb_type[mb_index])) {
+                            int i;
+                            for (i = 0; i < 2; i++) {
+                                int sx = mb_x * 16 + 4 + 8 * i;
+                                int sy = mb_y * 16 + 8;
+                                int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
+                                int mx = pict->motion_val[direction][xy][0] >> shift;
+                                int my = pict->motion_val[direction][xy][1] >> shift;
+
+                                if (IS_INTERLACED(pict->mb_type[mb_index]))
+                                    my *= 2;
+
+                                draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
+                                           height, s->linesize, 100);
+                            }
+                        } else {
+                              int sx = mb_x * 16 + 8;
+                              int sy = mb_y * 16 + 8;
+                              int xy = (mb_x + mb_y * mv_stride) << mv_sample_log2;
+                              int mx = pict->motion_val[direction][xy][0] >> shift + sx;
+                              int my = pict->motion_val[direction][xy][1] >> shift + sy;
+                              draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
+                        }
                     }
-                  }
                 }
-                if((s->avctx->debug&FF_DEBUG_VIS_QP) && pict->motion_val){
-                    uint64_t c= (pict->qscale_table[mb_index]*128/31) * 0x0101010101010101ULL;
+                if ((s->avctx->debug & FF_DEBUG_VIS_QP) && pict->motion_val) {
+                    uint64_t c = (pict->qscale_table[mb_index] * 128 / 31) *
+                                 0x0101010101010101ULL;
                     int y;
-                    for(y=0; y<block_height; y++){
-                        *(uint64_t*)(pict->data[1] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[1])= c;
-                        *(uint64_t*)(pict->data[2] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[2])= c;
+                    for (y = 0; y < block_height; y++) {
+                        *(uint64_t *)(pict->data[1] + 8 * mb_x +
+                                      (block_height * mb_y + y) *
+                                      pict->linesize[1]) = c;
+                        *(uint64_t *)(pict->data[2] + 8 * mb_x +
+                                      (block_height * mb_y + y) *
+                                      pict->linesize[2]) = c;
                     }
                 }
-                if((s->avctx->debug&FF_DEBUG_VIS_MB_TYPE) && pict->motion_val){
-                    int mb_type= pict->mb_type[mb_index];
+                if ((s->avctx->debug & FF_DEBUG_VIS_MB_TYPE) &&
+                    pict->motion_val) {
+                    int mb_type = pict->mb_type[mb_index];
                     uint64_t u,v;
                     int y;
-#define COLOR(theta, r)\
-u= (int)(128 + r*cos(theta*3.141592/180));\
-v= (int)(128 + r*sin(theta*3.141592/180));
-
-
-                    u=v=128;
-                    if(IS_PCM(mb_type)){
-                        COLOR(120,48)
-                    }else if((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) || IS_INTRA16x16(mb_type)){
-                        COLOR(30,48)
-                    }else if(IS_INTRA4x4(mb_type)){
-                        COLOR(90,48)
-                    }else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type)){
-//                        COLOR(120,48)
-                    }else if(IS_DIRECT(mb_type)){
-                        COLOR(150,48)
-                    }else if(IS_GMC(mb_type) && IS_SKIP(mb_type)){
-                        COLOR(170,48)
-                    }else if(IS_GMC(mb_type)){
-                        COLOR(190,48)
-                    }else if(IS_SKIP(mb_type)){
-//                        COLOR(180,48)
-                    }else if(!USES_LIST(mb_type, 1)){
-                        COLOR(240,48)
-                    }else if(!USES_LIST(mb_type, 0)){
-                        COLOR(0,48)
-                    }else{
+#define COLOR(theta, r) \
+    u = (int)(128 + r * cos(theta * 3.141592 / 180)); \
+    v = (int)(128 + r * sin(theta * 3.141592 / 180));
+
+
+                    u = v = 128;
+                    if (IS_PCM(mb_type)) {
+                        COLOR(120, 48)
+                    } else if ((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) ||
+                               IS_INTRA16x16(mb_type)) {
+                        COLOR(30, 48)
+                    } else if (IS_INTRA4x4(mb_type)) {
+                        COLOR(90, 48)
+                    } else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type)) {
+                        // COLOR(120, 48)
+                    } else if (IS_DIRECT(mb_type)) {
+                        COLOR(150, 48)
+                    } else if (IS_GMC(mb_type) && IS_SKIP(mb_type)) {
+                        COLOR(170, 48)
+                    } else if (IS_GMC(mb_type)) {
+                        COLOR(190, 48)
+                    } else if (IS_SKIP(mb_type)) {
+                        // COLOR(180, 48)
+                    } else if (!USES_LIST(mb_type, 1)) {
+                        COLOR(240, 48)
+                    } else if (!USES_LIST(mb_type, 0)) {
+                        COLOR(0, 48)
+                    } else {
                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
                         COLOR(300,48)
                     }
 
-                    u*= 0x0101010101010101ULL;
-                    v*= 0x0101010101010101ULL;
-                    for(y=0; y<block_height; y++){
-                        *(uint64_t*)(pict->data[1] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[1])= u;
-                        *(uint64_t*)(pict->data[2] + 8*mb_x + (block_height*mb_y + y)*pict->linesize[2])= v;
+                    u *= 0x0101010101010101ULL;
+                    v *= 0x0101010101010101ULL;
+                    for (y = 0; y < block_height; y++) {
+                        *(uint64_t *)(pict->data[1] + 8 * mb_x +
+                                      (block_height * mb_y + y) * pict->linesize[1]) = u;
+                        *(uint64_t *)(pict->data[2] + 8 * mb_x +
+                                      (block_height * mb_y + y) * pict->linesize[2]) = v;
                     }
 
-                    //segmentation
-                    if(IS_8X8(mb_type) || IS_16X8(mb_type)){
-                        *(uint64_t*)(pict->data[0] + 16*mb_x + 0 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
-                        *(uint64_t*)(pict->data[0] + 16*mb_x + 8 + (16*mb_y + 8)*pict->linesize[0])^= 0x8080808080808080ULL;
+                    // segmentation
+                    if (IS_8X8(mb_type) || IS_16X8(mb_type)) {
+                        *(uint64_t *)(pict->data[0] + 16 * mb_x + 0 +
+                                      (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
+                        *(uint64_t *)(pict->data[0] + 16 * mb_x + 8 +
+                                      (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
                     }
-                    if(IS_8X8(mb_type) || IS_8X16(mb_type)){
-                        for(y=0; y<16; y++)
-                            pict->data[0][16*mb_x + 8 + (16*mb_y + y)*pict->linesize[0]]^= 0x80;
+                    if (IS_8X8(mb_type) || IS_8X16(mb_type)) {
+                        for (y = 0; y < 16; y++)
+                            pict->data[0][16 * mb_x + 8 + (16 * mb_y + y) *
+                                          pict->linesize[0]] ^= 0x80;
                     }
-                    if(IS_8X8(mb_type) && mv_sample_log2 >= 2){
-                        int dm= 1 << (mv_sample_log2-2);
-                        for(i=0; i<4; i++){
-                            int sx= mb_x*16 + 8*(i&1);
-                            int sy= mb_y*16 + 8*(i>>1);
-                            int xy= (mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*mv_stride) << (mv_sample_log2-1);
-                            //FIXME bidir
-                            int32_t *mv = (int32_t*)&pict->motion_val[0][xy];
-                            if(mv[0] != mv[dm] || mv[dm*mv_stride] != mv[dm*(mv_stride+1)])
-                                for(y=0; y<8; y++)
-                                    pict->data[0][sx + 4 + (sy + y)*pict->linesize[0]]^= 0x80;
-                            if(mv[0] != mv[dm*mv_stride] || mv[dm] != mv[dm*(mv_stride+1)])
-                                *(uint64_t*)(pict->data[0] + sx + (sy + 4)*pict->linesize[0])^= 0x8080808080808080ULL;
+                    if (IS_8X8(mb_type) && mv_sample_log2 >= 2) {
+                        int dm = 1 << (mv_sample_log2 - 2);
+                        for (i = 0; i < 4; i++) {
+                            int sx = mb_x * 16 + 8 * (i & 1);
+                            int sy = mb_y * 16 + 8 * (i >> 1);
+                            int xy = (mb_x * 2 + (i & 1) +
+                                     (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
+                            // FIXME bidir
+                            int32_t *mv = (int32_t *) &pict->motion_val[0][xy];
+                            if (mv[0] != mv[dm] ||
+                                mv[dm * mv_stride] != mv[dm * (mv_stride + 1)])
+                                for (y = 0; y < 8; y++)
+                                    pict->data[0][sx + 4 + (sy + y) * pict->linesize[0]] ^= 0x80;
+                            if (mv[0] != mv[dm * mv_stride] || mv[dm] != mv[dm * (mv_stride + 1)])
+                                *(uint64_t *)(pict->data[0] + sx + (sy + 4) *
+                                              pict->linesize[0]) ^= 0x8080808080808080ULL;
                         }
                     }
 
-                    if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264){
+                    if (IS_INTERLACED(mb_type) &&
+                        s->codec_id == AV_CODEC_ID_H264) {
                         // hmm
                     }
                 }
-                s->mbskip_table[mb_index]=0;
+                s->mbskip_table[mb_index] = 0;
             }
         }
     }
 }
 
-static inline int hpel_motion_lowres(MpegEncContext *s,
-                                  uint8_t *dest, uint8_t *src,
-                                  int field_based, int field_select,
-                                  int src_x, int src_y,
-                                  int width, int height, int stride,
-                                  int h_edge_pos, int v_edge_pos,
-                                  int w, int h, h264_chroma_mc_func *pix_op,
-                                  int motion_x, int motion_y)
-{
-    const int lowres= s->avctx->lowres;
-    const int op_index= FFMIN(lowres, 2);
-    const int s_mask= (2<<lowres)-1;
-    int emu=0;
-    int sx, sy;
-
-    if(s->quarter_sample){
-        motion_x/=2;
-        motion_y/=2;
-    }
-
-    sx= motion_x & s_mask;
-    sy= motion_y & s_mask;
-    src_x += motion_x >> (lowres+1);
-    src_y += motion_y >> (lowres+1);
-
-    src += src_y * stride + src_x;
-
-    if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - w
-       || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w+1, (h+1)<<field_based,
-                            src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
-        src= s->edge_emu_buffer;
-        emu=1;
-    }
-
-    sx= (sx << 2) >> lowres;
-    sy= (sy << 2) >> lowres;
-    if(field_select)
-        src += s->linesize;
-    pix_op[op_index](dest, src, stride, h, sx, sy);
-    return emu;
-}
-
-/* apply one mpeg motion vector to the three components */
-static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
-                               uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                               int field_based, int bottom_field, int field_select,
-                               uint8_t **ref_picture, h264_chroma_mc_func *pix_op,
-                               int motion_x, int motion_y, int h, int mb_y)
-{
-    uint8_t *ptr_y, *ptr_cb, *ptr_cr;
-    int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, uvlinesize, linesize, sx, sy, uvsx, uvsy;
-    const int lowres= s->avctx->lowres;
-    const int op_index= FFMIN(lowres, 2);
-    const int block_s= 8>>lowres;
-    const int s_mask= (2<<lowres)-1;
-    const int h_edge_pos = s->h_edge_pos >> lowres;
-    const int v_edge_pos = s->v_edge_pos >> lowres;
-    linesize   = s->current_picture.f.linesize[0] << field_based;
-    uvlinesize = s->current_picture.f.linesize[1] << field_based;
-
-    if(s->quarter_sample){ //FIXME obviously not perfect but qpel will not work in lowres anyway
-        motion_x/=2;
-        motion_y/=2;
-    }
-
-    if(field_based){
-        motion_y += (bottom_field - field_select)*((1<<lowres)-1);
-    }
-
-    sx= motion_x & s_mask;
-    sy= motion_y & s_mask;
-    src_x = s->mb_x*2*block_s               + (motion_x >> (lowres+1));
-    src_y =(   mb_y*2*block_s>>field_based) + (motion_y >> (lowres+1));
-
-    if (s->out_format == FMT_H263) {
-        uvsx = ((motion_x>>1) & s_mask) | (sx&1);
-        uvsy = ((motion_y>>1) & s_mask) | (sy&1);
-        uvsrc_x = src_x>>1;
-        uvsrc_y = src_y>>1;
-    }else if(s->out_format == FMT_H261){//even chroma mv's are full pel in H261
-        mx = motion_x / 4;
-        my = motion_y / 4;
-        uvsx = (2*mx) & s_mask;
-        uvsy = (2*my) & s_mask;
-        uvsrc_x = s->mb_x*block_s               + (mx >> lowres);
-        uvsrc_y =    mb_y*block_s               + (my >> lowres);
-    } else {
-        mx = motion_x / 2;
-        my = motion_y / 2;
-        uvsx = mx & s_mask;
-        uvsy = my & s_mask;
-        uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
-        uvsrc_y =(   mb_y*block_s>>field_based) + (my >> (lowres+1));
-    }
-
-    ptr_y  = ref_picture[0] + src_y * linesize + src_x;
-    ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
-    ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
-
-    if(   (unsigned)src_x > h_edge_pos                 - (!!sx) - 2*block_s
-       || (unsigned)src_y >(v_edge_pos >> field_based) - (!!sy) - h){
-            s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, 17, 17+field_based,
-                             src_x, src_y<<field_based, h_edge_pos, v_edge_pos);
-            ptr_y = s->edge_emu_buffer;
-            if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-                uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize;
-                s->dsp.emulated_edge_mc(uvbuf  , ptr_cb, s->uvlinesize, 9, 9+field_based,
-                                 uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
-                s->dsp.emulated_edge_mc(uvbuf+16, ptr_cr, s->uvlinesize, 9, 9+field_based,
-                                 uvsrc_x, uvsrc_y<<field_based, h_edge_pos>>1, v_edge_pos>>1);
-                ptr_cb= uvbuf;
-                ptr_cr= uvbuf+16;
-            }
-    }
-
-    if(bottom_field){ //FIXME use this for field pix too instead of the obnoxious hack which changes picture.f.data
-        dest_y += s->linesize;
-        dest_cb+= s->uvlinesize;
-        dest_cr+= s->uvlinesize;
-    }
-
-    if(field_select){
-        ptr_y += s->linesize;
-        ptr_cb+= s->uvlinesize;
-        ptr_cr+= s->uvlinesize;
-    }
-
-    sx= (sx << 2) >> lowres;
-    sy= (sy << 2) >> lowres;
-    pix_op[lowres-1](dest_y, ptr_y, linesize, h, sx, sy);
-
-    if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-        uvsx= (uvsx << 2) >> lowres;
-        uvsy= (uvsy << 2) >> lowres;
-        pix_op[op_index](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
-        pix_op[op_index](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
-    }
-    //FIXME h261 lowres loop filter
-}
-
-static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
-                                     uint8_t *dest_cb, uint8_t *dest_cr,
-                                     uint8_t **ref_picture,
-                                     h264_chroma_mc_func *pix_op,
-                                     int mx, int my){
-    const int lowres= s->avctx->lowres;
-    const int op_index= FFMIN(lowres, 2);
-    const int block_s= 8>>lowres;
-    const int s_mask= (2<<lowres)-1;
-    const int h_edge_pos = s->h_edge_pos >> (lowres+1);
-    const int v_edge_pos = s->v_edge_pos >> (lowres+1);
-    int emu=0, src_x, src_y, offset, sx, sy;
-    uint8_t *ptr;
-
-    if(s->quarter_sample){
-        mx/=2;
-        my/=2;
-    }
-
-    /* In case of 8X8, we construct a single chroma motion vector
-       with a special rounding */
-    mx= ff_h263_round_chroma(mx);
-    my= ff_h263_round_chroma(my);
-
-    sx= mx & s_mask;
-    sy= my & s_mask;
-    src_x = s->mb_x*block_s + (mx >> (lowres+1));
-    src_y = s->mb_y*block_s + (my >> (lowres+1));
-
-    offset = src_y * s->uvlinesize + src_x;
-    ptr = ref_picture[1] + offset;
-    if(s->flags&CODEC_FLAG_EMU_EDGE){
-        if(   (unsigned)src_x > h_edge_pos - (!!sx) - block_s
-           || (unsigned)src_y > v_edge_pos - (!!sy) - block_s){
-            s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
-            ptr= s->edge_emu_buffer;
-            emu=1;
-        }
-    }
-    sx= (sx << 2) >> lowres;
-    sy= (sy << 2) >> lowres;
-    pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
-
-    ptr = ref_picture[2] + offset;
-    if(emu){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos);
-        ptr= s->edge_emu_buffer;
-    }
-    pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
-}
-
-/**
- * motion compensation of a single macroblock
- * @param s context
- * @param dest_y luma destination pointer
- * @param dest_cb chroma cb/u destination pointer
- * @param dest_cr chroma cr/v destination pointer
- * @param dir direction (0->forward, 1->backward)
- * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
- * @param pix_op halfpel motion compensation function (average or put normally)
- * the motion vectors are taken from s->mv and the MV type from s->mv_type
- */
-static inline void MPV_motion_lowres(MpegEncContext *s,
-                              uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                              int dir, uint8_t **ref_picture,
-                              h264_chroma_mc_func *pix_op)
-{
-    int mx, my;
-    int mb_x, mb_y, i;
-    const int lowres= s->avctx->lowres;
-    const int block_s= 8>>lowres;
-
-    mb_x = s->mb_x;
-    mb_y = s->mb_y;
-
-    switch(s->mv_type) {
-    case MV_TYPE_16X16:
-        mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
-                    0, 0, 0,
-                    ref_picture, pix_op,
-                    s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s, mb_y);
-        break;
-    case MV_TYPE_8X8:
-        mx = 0;
-        my = 0;
-            for(i=0;i<4;i++) {
-                hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) * s->linesize)*block_s,
-                            ref_picture[0], 0, 0,
-                            (2*mb_x + (i & 1))*block_s, (2*mb_y + (i >>1))*block_s,
-                            s->width, s->height, s->linesize,
-                            s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
-                            block_s, block_s, pix_op,
-                            s->mv[dir][i][0], s->mv[dir][i][1]);
-
-                mx += s->mv[dir][i][0];
-                my += s->mv[dir][i][1];
-            }
-
-        if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY))
-            chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture, pix_op, mx, my);
-        break;
-    case MV_TYPE_FIELD:
-        if (s->picture_structure == PICT_FRAME) {
-            /* top field */
-            mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
-                        1, 0, s->field_select[dir][0],
-                        ref_picture, pix_op,
-                        s->mv[dir][0][0], s->mv[dir][0][1], block_s, mb_y);
-            /* bottom field */
-            mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
-                        1, 1, s->field_select[dir][1],
-                        ref_picture, pix_op,
-                        s->mv[dir][1][0], s->mv[dir][1][1], block_s, mb_y);
-        } else {
-            if(s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field){
-                ref_picture = s->current_picture_ptr->f.data;
-            }
-
-            mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
-                        0, 0, s->field_select[dir][0],
-                        ref_picture, pix_op,
-                        s->mv[dir][0][0], s->mv[dir][0][1], 2*block_s, mb_y>>1);
-        }
-        break;
-    case MV_TYPE_16X8:
-        for(i=0; i<2; i++){
-            uint8_t ** ref2picture;
-
-            if(s->picture_structure == s->field_select[dir][i] + 1 || s->pict_type == AV_PICTURE_TYPE_B || s->first_field){
-                ref2picture= ref_picture;
-            }else{
-                ref2picture = s->current_picture_ptr->f.data;
-            }
-
-            mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
-                        0, 0, s->field_select[dir][i],
-                        ref2picture, pix_op,
-                        s->mv[dir][i][0], s->mv[dir][i][1] + 2*block_s*i, block_s, mb_y>>1);
-
-            dest_y += 2*block_s*s->linesize;
-            dest_cb+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
-            dest_cr+= (2*block_s>>s->chroma_y_shift)*s->uvlinesize;
-        }
-        break;
-    case MV_TYPE_DMV:
-        if(s->picture_structure == PICT_FRAME){
-            for(i=0; i<2; i++){
-                int j;
-                for(j=0; j<2; j++){
-                    mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
-                                1, j, j^i,
-                                ref_picture, pix_op,
-                                s->mv[dir][2*i + j][0], s->mv[dir][2*i + j][1], block_s, mb_y);
-                }
-                pix_op = s->dsp.avg_h264_chroma_pixels_tab;
-            }
-        }else{
-            for(i=0; i<2; i++){
-                mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
-                            0, 0, s->picture_structure != i+1,
-                            ref_picture, pix_op,
-                            s->mv[dir][2*i][0],s->mv[dir][2*i][1],2*block_s, mb_y>>1);
-
-                // after put we make avg of the same block
-                pix_op = s->dsp.avg_h264_chroma_pixels_tab;
-
-                //opposite parity is always in the same frame if this is second field
-                if(!s->first_field){
-                    ref_picture = s->current_picture_ptr->f.data;
-                }
-            }
-        }
-    break;
-    default: assert(0);
-    }
-}
-
 /**
  * find the lowest MB row referenced in the MVs
  */
-int MPV_lowest_referenced_row(MpegEncContext *s, int dir)
+int ff_MPV_lowest_referenced_row(MpegEncContext *s, int dir)
 {
     int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
     int my, off, i, mvs;
 
-    if (s->picture_structure != PICT_FRAME) goto unhandled;
+    if (s->picture_structure != PICT_FRAME || s->mcsel)
+        goto unhandled;
 
     switch (s->mv_type) {
         case MV_TYPE_16X16:
@@ -2101,7 +2088,7 @@ unhandled:
 
 /* put block[] to dest[] */
 static inline void put_dct(MpegEncContext *s,
-                           DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
+                           int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
 {
     s->dct_unquantize_intra(s, block, i, qscale);
     s->dsp.idct_put (dest, line_size, block);
@@ -2109,7 +2096,7 @@ static inline void put_dct(MpegEncContext *s,
 
 /* add block[] to dest[] */
 static inline void add_dct(MpegEncContext *s,
-                           DCTELEM *block, int i, uint8_t *dest, int line_size)
+                           int16_t *block, int i, uint8_t *dest, int line_size)
 {
     if (s->block_last_index[i] >= 0) {
         s->dsp.idct_add (dest, line_size, block);
@@ -2117,7 +2104,7 @@ static inline void add_dct(MpegEncContext *s,
 }
 
 static inline void add_dequant_dct(MpegEncContext *s,
-                           DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
+                           int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
 {
     if (s->block_last_index[i] >= 0) {
         s->dct_unquantize_inter(s, block, i, qscale);
@@ -2170,8 +2157,8 @@ void ff_clean_intra_table_entries(MpegEncContext *s)
    s->interlaced_dct : true if interlaced dct used (mpeg2)
  */
 static av_always_inline
-void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
-                            int lowres_flag, int is_mpeg12)
+void MPV_decode_mb_internal(MpegEncContext *s, int16_t block[12][64],
+                            int is_mpeg12)
 {
     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
     if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
@@ -2182,7 +2169,7 @@ void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
        /* save DCT coefficients */
        int i,j;
-       DCTELEM *dct = &s->current_picture.f.dct_coeff[mb_xy * 64 * 6];
+       int16_t *dct = &s->current_picture.f.dct_coeff[mb_xy * 64 * 6];
        av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
        for(i=0; i<6; i++){
            for(j=0; j<64; j++){
@@ -2216,38 +2203,27 @@ void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
         qpel_mc_func (*op_qpix)[16];
         const int linesize   = s->current_picture.f.linesize[0]; //not s->linesize as this would be wrong for field pics
         const int uvlinesize = s->current_picture.f.linesize[1];
-        const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
-        const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
+        const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band;
+        const int block_size = 8;
 
         /* avoid copy if macroblock skipped in last frame too */
         /* skip only during decoding as we might trash the buffers during encoding a bit */
         if(!s->encoding){
             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
-            const int age = s->current_picture.f.age;
-
-            assert(age);
 
             if (s->mb_skipped) {
                 s->mb_skipped= 0;
                 assert(s->pict_type!=AV_PICTURE_TYPE_I);
-
-                (*mbskip_ptr) ++; /* indicate that this time we skipped it */
-                if(*mbskip_ptr >99) *mbskip_ptr= 99;
-
-                /* if previous was skipped too, then nothing to do !  */
-                if (*mbskip_ptr >= age && s->current_picture.f.reference){
-                    return;
-                }
+                *mbskip_ptr = 1;
             } else if(!s->current_picture.f.reference) {
-                (*mbskip_ptr) ++; /* increase counter so the age can be compared cleanly */
-                if(*mbskip_ptr >99) *mbskip_ptr= 99;
+                *mbskip_ptr = 1;
             } else{
                 *mbskip_ptr = 0; /* not skipped */
             }
         }
 
         dct_linesize = linesize << s->interlaced_dct;
-        dct_offset =(s->interlaced_dct)? linesize : linesize*block_size;
+        dct_offset   = s->interlaced_dct ? linesize : linesize * block_size;
 
         if(readable){
             dest_y=  s->dest[0];
@@ -2266,38 +2242,30 @@ void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
 
                 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
                     if (s->mv_dir & MV_DIR_FORWARD) {
-                        ff_thread_await_progress((AVFrame*)s->last_picture_ptr, MPV_lowest_referenced_row(s, 0), 0);
+                        ff_thread_await_progress(&s->last_picture_ptr->f,
+                                                 ff_MPV_lowest_referenced_row(s, 0),
+                                                 0);
                     }
                     if (s->mv_dir & MV_DIR_BACKWARD) {
-                        ff_thread_await_progress((AVFrame*)s->next_picture_ptr, MPV_lowest_referenced_row(s, 1), 0);
+                        ff_thread_await_progress(&s->next_picture_ptr->f,
+                                                 ff_MPV_lowest_referenced_row(s, 1),
+                                                 0);
                     }
                 }
 
-                if(lowres_flag){
-                    h264_chroma_mc_func *op_pix = s->dsp.put_h264_chroma_pixels_tab;
-
-                    if (s->mv_dir & MV_DIR_FORWARD) {
-                        MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix);
-                        op_pix = s->dsp.avg_h264_chroma_pixels_tab;
-                    }
-                    if (s->mv_dir & MV_DIR_BACKWARD) {
-                        MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix);
-                    }
+                op_qpix= s->me.qpel_put;
+                if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
+                    op_pix = s->dsp.put_pixels_tab;
                 }else{
-                    op_qpix= s->me.qpel_put;
-                    if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
-                        op_pix = s->dsp.put_pixels_tab;
-                    }else{
-                        op_pix = s->dsp.put_no_rnd_pixels_tab;
-                    }
-                    if (s->mv_dir & MV_DIR_FORWARD) {
-                        MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix, op_qpix);
-                        op_pix = s->dsp.avg_pixels_tab;
-                        op_qpix= s->me.qpel_avg;
-                    }
-                    if (s->mv_dir & MV_DIR_BACKWARD) {
-                        MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix, op_qpix);
-                    }
+                    op_pix = s->dsp.put_no_rnd_pixels_tab;
+                }
+                if (s->mv_dir & MV_DIR_FORWARD) {
+                    ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix, op_qpix);
+                    op_pix = s->dsp.avg_pixels_tab;
+                    op_qpix= s->me.qpel_avg;
+                }
+                if (s->mv_dir & MV_DIR_BACKWARD) {
+                    ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix, op_qpix);
                 }
             }
 
@@ -2310,8 +2278,8 @@ void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
             }
 
             /* add dct residue */
-            if(s->encoding || !(   s->msmpeg4_version || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO
-                                || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){
+            if(s->encoding || !(   s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
+                                || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
@@ -2330,7 +2298,7 @@ void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
                     }
                 }
-            } else if(is_mpeg12 || (s->codec_id != CODEC_ID_WMV2)){
+            } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
@@ -2343,7 +2311,7 @@ void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
                     }else{
                         //chroma422
                         dct_linesize = uvlinesize << s->interlaced_dct;
-                        dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
+                        dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize * 8;
 
                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
@@ -2363,7 +2331,7 @@ void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
             }
         } else {
             /* dct only in intra block */
-            if(s->encoding || !(s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO)){
+            if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
@@ -2395,7 +2363,7 @@ void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
                     }else{
 
                         dct_linesize = uvlinesize << s->interlaced_dct;
-                        dct_offset =(s->interlaced_dct)? uvlinesize : uvlinesize*8;
+                        dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize * 8;
 
                         s->dsp.idct_put(dest_cb,              dct_linesize, block[4]);
                         s->dsp.idct_put(dest_cr,              dct_linesize, block[5]);
@@ -2420,15 +2388,13 @@ skip_idct:
     }
 }
 
-void MPV_decode_mb(MpegEncContext *s, DCTELEM block[12][64]){
+void ff_MPV_decode_mb(MpegEncContext *s, int16_t block[12][64]){
 #if !CONFIG_SMALL
     if(s->out_format == FMT_MPEG1) {
-        if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 1);
-        else                 MPV_decode_mb_internal(s, block, 0, 1);
+        MPV_decode_mb_internal(s, block, 1);
     } else
 #endif
-    if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 0);
-    else                  MPV_decode_mb_internal(s, block, 0, 0);
+        MPV_decode_mb_internal(s, block, 0);
 }
 
 /**
@@ -2447,9 +2413,10 @@ void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
        && s->current_picture.f.reference
        && !s->intra_only
        && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
+        const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
         int sides = 0, edge_h;
-        int hshift = av_pix_fmt_descriptors[s->avctx->pix_fmt].log2_chroma_w;
-        int vshift = av_pix_fmt_descriptors[s->avctx->pix_fmt].log2_chroma_h;
+        int hshift = desc->log2_chroma_w;
+        int vshift = desc->log2_chroma_h;
         if (y==0) sides |= EDGE_TOP;
         if (y + h >= s->v_edge_pos) sides |= EDGE_BOTTOM;
 
@@ -2476,9 +2443,9 @@ void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
         int i;
 
         if(s->pict_type==AV_PICTURE_TYPE_B || s->low_delay || (s->avctx->slice_flags&SLICE_FLAG_CODED_ORDER))
-            src= (AVFrame*)s->current_picture_ptr;
+            src = &s->current_picture_ptr->f;
         else if(s->last_picture_ptr)
-            src= (AVFrame*)s->last_picture_ptr;
+            src = &s->last_picture_ptr->f;
         else
             return;
 
@@ -2503,7 +2470,7 @@ void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
     const int linesize   = s->current_picture.f.linesize[0]; //not s->linesize as this would be wrong for field pics
     const int uvlinesize = s->current_picture.f.linesize[1];
-    const int mb_size= 4 - s->avctx->lowres;
+    const int mb_size= 4;
 
     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
@@ -2532,6 +2499,35 @@ void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
     }
 }
 
+/**
+ * Permute an 8x8 block.
+ * @param block the block which will be permuted according to the given permutation vector
+ * @param permutation the permutation vector
+ * @param last the last non zero coefficient in scantable order, used to speed the permutation up
+ * @param scantable the used scantable, this is only used to speed the permutation up, the block is not
+ *                  (inverse) permutated to scantable order!
+ */
+void ff_block_permute(int16_t *block, uint8_t *permutation, const uint8_t *scantable, int last)
+{
+    int i;
+    int16_t temp[64];
+
+    if(last<=0) return;
+    //if(permutation[1]==1) return; //FIXME it is ok but not clean and might fail for some permutations
+
+    for(i=0; i<=last; i++){
+        const int j= scantable[i];
+        temp[j]= block[j];
+        block[j]=0;
+    }
+
+    for(i=0; i<=last; i++){
+        const int j= scantable[i];
+        const int perm_j= permutation[j];
+        block[perm_j]= temp[j];
+    }
+}
+
 void ff_mpeg_flush(AVCodecContext *avctx){
     int i;
     MpegEncContext *s = avctx->priv_data;
@@ -2560,7 +2556,7 @@ void ff_mpeg_flush(AVCodecContext *avctx){
 }
 
 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale)
+                                   int16_t *block, int n, int qscale)
 {
     int i, level, nCoeffs;
     const uint16_t *quant_matrix;
@@ -2592,7 +2588,7 @@ static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
 }
 
 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale)
+                                   int16_t *block, int n, int qscale)
 {
     int i, level, nCoeffs;
     const uint16_t *quant_matrix;
@@ -2621,7 +2617,7 @@ static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
 }
 
 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale)
+                                   int16_t *block, int n, int qscale)
 {
     int i, level, nCoeffs;
     const uint16_t *quant_matrix;
@@ -2651,7 +2647,7 @@ static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
 }
 
 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale)
+                                   int16_t *block, int n, int qscale)
 {
     int i, level, nCoeffs;
     const uint16_t *quant_matrix;
@@ -2684,7 +2680,7 @@ static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
 }
 
 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
-                                   DCTELEM *block, int n, int qscale)
+                                   int16_t *block, int n, int qscale)
 {
     int i, level, nCoeffs;
     const uint16_t *quant_matrix;
@@ -2715,7 +2711,7 @@ static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
 }
 
 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
-                                  DCTELEM *block, int n, int qscale)
+                                  int16_t *block, int n, int qscale)
 {
     int i, level, qmul, qadd;
     int nCoeffs;
@@ -2752,7 +2748,7 @@ static void dct_unquantize_h263_intra_c(MpegEncContext *s,
 }
 
 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
-                                  DCTELEM *block, int n, int qscale)
+                                  int16_t *block, int n, int qscale)
 {
     int i, level, qmul, qadd;
     int nCoeffs;
@@ -2794,8 +2790,8 @@ void ff_set_qscale(MpegEncContext * s, int qscale)
     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
 }
 
-void MPV_report_decode_progress(MpegEncContext *s)
+void ff_MPV_report_decode_progress(MpegEncContext *s)
 {
     if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->error_occurred)
-        ff_thread_report_progress((AVFrame*)s->current_picture_ptr, s->mb_y, 0);
+        ff_thread_report_progress(&s->current_picture_ptr->f, s->mb_y, 0);
 }