]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/mpegvideo_enc.c
mpegvideo: Move QMAT_SHIFT* defines to the only place they are used
[ffmpeg] / libavcodec / mpegvideo_enc.c
index e84b0dae031db56df98e3c64b2e221c5e4381eac..2d0cd83a6d452f44ff5d255c5baba96803368607 100644 (file)
  * The simplest mpeg encoder (well, it was the simplest!).
  */
 
+#include <stdint.h>
+
+#include "libavutil/internal.h"
 #include "libavutil/intmath.h"
 #include "libavutil/mathematics.h"
+#include "libavutil/pixdesc.h"
 #include "libavutil/opt.h"
+#include "libavutil/timer.h"
 #include "avcodec.h"
-#include "dsputil.h"
+#include "dct.h"
+#include "idctdsp.h"
+#include "mpeg12.h"
 #include "mpegvideo.h"
-#include "mpegvideo_common.h"
+#include "h261.h"
 #include "h263.h"
+#include "mjpegenc_common.h"
+#include "mathops.h"
+#include "mpegutils.h"
 #include "mjpegenc.h"
 #include "msmpeg4.h"
+#include "pixblockdsp.h"
+#include "qpeldsp.h"
 #include "faandct.h"
 #include "thread.h"
 #include "aandcttab.h"
 #include "bytestream.h"
 #include <limits.h>
 
-//#undef NDEBUG
-//#include <assert.h>
+#define QUANT_BIAS_SHIFT 8
+
+#define QMAT_SHIFT_MMX 16
+#define QMAT_SHIFT 22
 
 static int encode_picture(MpegEncContext *s, int picture_number);
-static int dct_quantize_refine(MpegEncContext *s, DCTELEM *block, int16_t *weight, DCTELEM *orig, int n, int qscale);
+static int dct_quantize_refine(MpegEncContext *s, int16_t *block, int16_t *weight, int16_t *orig, int n, int qscale);
 static int sse_mb(MpegEncContext *s);
-static void denoise_dct_c(MpegEncContext *s, DCTELEM *block);
-static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
-
-/* enable all paranoid tests for rounding, overflows, etc... */
-//#define PARANOID
-
-//#define DEBUG
+static void denoise_dct_c(MpegEncContext *s, int16_t *block);
+static int dct_quantize_trellis_c(MpegEncContext *s, int16_t *block, int n, int qscale, int *overflow);
 
 static uint8_t default_mv_penalty[MAX_FCODE + 1][MAX_MV * 2 + 1];
 static uint8_t default_fcode_tab[MAX_MV * 2 + 1];
@@ -68,21 +77,22 @@ const AVOption ff_mpv_generic_options[] = {
     { NULL },
 };
 
-void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64],
+void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64],
                        uint16_t (*qmat16)[2][64],
                        const uint16_t *quant_matrix,
                        int bias, int qmin, int qmax, int intra)
 {
+    FDCTDSPContext *fdsp = &s->fdsp;
     int qscale;
     int shift = 0;
 
     for (qscale = qmin; qscale <= qmax; qscale++) {
         int i;
-        if (dsp->fdct == ff_jpeg_fdct_islow_8 ||
-            dsp->fdct == ff_jpeg_fdct_islow_10 ||
-            dsp->fdct == ff_faandct) {
+        if (fdsp->fdct == ff_jpeg_fdct_islow_8  ||
+            fdsp->fdct == ff_jpeg_fdct_islow_10 ||
+            fdsp->fdct == ff_faandct) {
             for (i = 0; i < 64; i++) {
-                const int j = dsp->idct_permutation[i];
+                const int j = s->idsp.idct_permutation[i];
                 /* 16 <= qscale * quant_matrix[i] <= 7905
                  * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
                  *             19952 <=              x  <= 249205026
@@ -92,9 +102,9 @@ void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64],
                 qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) /
                                         (qscale * quant_matrix[j]));
             }
-        } else if (dsp->fdct == ff_fdct_ifast) {
+        } else if (fdsp->fdct == ff_fdct_ifast) {
             for (i = 0; i < 64; i++) {
-                const int j = dsp->idct_permutation[i];
+                const int j = s->idsp.idct_permutation[i];
                 /* 16 <= qscale * quant_matrix[i] <= 7905
                  * Assume x = ff_aanscales[i] * qscale * quant_matrix[i]
                  *             19952 <=              x  <= 249205026
@@ -107,7 +117,7 @@ void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64],
             }
         } else {
             for (i = 0; i < 64; i++) {
-                const int j = dsp->idct_permutation[i];
+                const int j = s->idsp.idct_permutation[i];
                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
                  * Assume x = qscale * quant_matrix[i]
                  * So             16 <=              x  <= 7905
@@ -131,7 +141,7 @@ void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64],
 
         for (i = intra; i < 64; i++) {
             int64_t max = 8191;
-            if (dsp->fdct == ff_fdct_ifast) {
+            if (fdsp->fdct == ff_fdct_ifast) {
                 max = (8191LL * ff_aanscales[i]) >> 14;
             }
             while (((max * qmat[qscale][i]) >> shift) > INT_MAX) {
@@ -174,7 +184,7 @@ void ff_write_quant_matrix(PutBitContext *pb, uint16_t *matrix)
  */
 void ff_init_qscale_tab(MpegEncContext *s)
 {
-    int8_t * const qscale_table = s->current_picture.f.qscale_table;
+    int8_t * const qscale_table = s->current_picture.qscale_table;
     int i;
 
     for (i = 0; i < s->mb_num; i++) {
@@ -185,54 +195,6 @@ void ff_init_qscale_tab(MpegEncContext *s)
     }
 }
 
-static void copy_picture_attributes(MpegEncContext *s,
-                                    AVFrame *dst,
-                                    AVFrame *src)
-{
-    int i;
-
-    dst->pict_type              = src->pict_type;
-    dst->quality                = src->quality;
-    dst->coded_picture_number   = src->coded_picture_number;
-    dst->display_picture_number = src->display_picture_number;
-    //dst->reference              = src->reference;
-    dst->pts                    = src->pts;
-    dst->interlaced_frame       = src->interlaced_frame;
-    dst->top_field_first        = src->top_field_first;
-
-    if (s->avctx->me_threshold) {
-        if (!src->motion_val[0])
-            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.motion_val not set!\n");
-        if (!src->mb_type)
-            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.mb_type not set!\n");
-        if (!src->ref_index[0])
-            av_log(s->avctx, AV_LOG_ERROR, "AVFrame.ref_index not set!\n");
-        if (src->motion_subsample_log2 != dst->motion_subsample_log2)
-            av_log(s->avctx, AV_LOG_ERROR,
-                   "AVFrame.motion_subsample_log2 doesn't match! (%d!=%d)\n",
-                   src->motion_subsample_log2, dst->motion_subsample_log2);
-
-        memcpy(dst->mb_type, src->mb_type,
-               s->mb_stride * s->mb_height * sizeof(dst->mb_type[0]));
-
-        for (i = 0; i < 2; i++) {
-            int stride = ((16 * s->mb_width ) >>
-                          src->motion_subsample_log2) + 1;
-            int height = ((16 * s->mb_height) >> src->motion_subsample_log2);
-
-            if (src->motion_val[i] &&
-                src->motion_val[i] != dst->motion_val[i]) {
-                memcpy(dst->motion_val[i], src->motion_val[i],
-                       2 * stride * height * sizeof(int16_t));
-            }
-            if (src->ref_index[i] && src->ref_index[i] != dst->ref_index[i]) {
-                memcpy(dst->ref_index[i], src->ref_index[i],
-                       s->mb_stride * 4 * s->mb_height * sizeof(int8_t));
-            }
-        }
-    }
-}
-
 static void update_duplicate_context_after_me(MpegEncContext *dst,
                                               MpegEncContext *src)
 {
@@ -266,63 +228,62 @@ static void MPV_encode_defaults(MpegEncContext *s)
     }
     s->me.mv_penalty = default_mv_penalty;
     s->fcode_tab     = default_fcode_tab;
+
+    s->input_picture_number  = 0;
+    s->picture_in_gop_number = 0;
 }
 
 /* init video encoder */
 av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
 {
     MpegEncContext *s = avctx->priv_data;
-    int i;
-    int chroma_h_shift, chroma_v_shift;
+    int i, ret, format_supported;
 
     MPV_encode_defaults(s);
 
     switch (avctx->codec_id) {
     case AV_CODEC_ID_MPEG2VIDEO:
-        if (avctx->pix_fmt != PIX_FMT_YUV420P &&
-            avctx->pix_fmt != PIX_FMT_YUV422P) {
+        if (avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
+            avctx->pix_fmt != AV_PIX_FMT_YUV422P) {
             av_log(avctx, AV_LOG_ERROR,
                    "only YUV420 and YUV422 are supported\n");
             return -1;
         }
         break;
-    case AV_CODEC_ID_LJPEG:
-        if (avctx->pix_fmt != PIX_FMT_YUVJ420P &&
-            avctx->pix_fmt != PIX_FMT_YUVJ422P &&
-            avctx->pix_fmt != PIX_FMT_YUVJ444P &&
-            avctx->pix_fmt != PIX_FMT_BGRA     &&
-            ((avctx->pix_fmt != PIX_FMT_YUV420P &&
-              avctx->pix_fmt != PIX_FMT_YUV422P &&
-              avctx->pix_fmt != PIX_FMT_YUV444P) ||
-             avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL)) {
-            av_log(avctx, AV_LOG_ERROR, "colorspace not supported in LJPEG\n");
-            return -1;
-        }
-        break;
     case AV_CODEC_ID_MJPEG:
-        if (avctx->pix_fmt != PIX_FMT_YUVJ420P &&
-            avctx->pix_fmt != PIX_FMT_YUVJ422P &&
-            ((avctx->pix_fmt != PIX_FMT_YUV420P &&
-              avctx->pix_fmt != PIX_FMT_YUV422P) ||
-             avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL)) {
+        format_supported = 0;
+        /* JPEG color space */
+        if (avctx->pix_fmt == AV_PIX_FMT_YUVJ420P ||
+            avctx->pix_fmt == AV_PIX_FMT_YUVJ422P ||
+            (avctx->color_range == AVCOL_RANGE_JPEG &&
+             (avctx->pix_fmt == AV_PIX_FMT_YUV420P ||
+              avctx->pix_fmt == AV_PIX_FMT_YUV422P)))
+            format_supported = 1;
+        /* MPEG color space */
+        else if (avctx->strict_std_compliance <= FF_COMPLIANCE_UNOFFICIAL &&
+                 (avctx->pix_fmt == AV_PIX_FMT_YUV420P ||
+                  avctx->pix_fmt == AV_PIX_FMT_YUV422P))
+            format_supported = 1;
+
+        if (!format_supported) {
             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
             return -1;
         }
         break;
     default:
-        if (avctx->pix_fmt != PIX_FMT_YUV420P) {
+        if (avctx->pix_fmt != AV_PIX_FMT_YUV420P) {
             av_log(avctx, AV_LOG_ERROR, "only YUV420 is supported\n");
             return -1;
         }
     }
 
     switch (avctx->pix_fmt) {
-    case PIX_FMT_YUVJ422P:
-    case PIX_FMT_YUV422P:
+    case AV_PIX_FMT_YUVJ422P:
+    case AV_PIX_FMT_YUV422P:
         s->chroma_format = CHROMA_422;
         break;
-    case PIX_FMT_YUVJ420P:
-    case PIX_FMT_YUV420P:
+    case AV_PIX_FMT_YUVJ420P:
+    case AV_PIX_FMT_YUV420P:
     default:
         s->chroma_format = CHROMA_420;
         break;
@@ -341,14 +302,12 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     s->avctx        = avctx;
     s->flags        = avctx->flags;
     s->flags2       = avctx->flags2;
+    if (avctx->max_b_frames > MAX_B_FRAMES) {
+        av_log(avctx, AV_LOG_ERROR, "Too many B-frames requested, maximum "
+               "is %d.\n", MAX_B_FRAMES);
+    }
     s->max_b_frames = avctx->max_b_frames;
     s->codec_id     = avctx->codec->id;
-#if FF_API_MPV_GLOBAL_OPTS
-    if (avctx->luma_elim_threshold)
-        s->luma_elim_threshold   = avctx->luma_elim_threshold;
-    if (avctx->chroma_elim_threshold)
-        s->chroma_elim_threshold = avctx->chroma_elim_threshold;
-#endif
     s->strict_std_compliance = avctx->strict_std_compliance;
     s->quarter_sample     = (avctx->flags & CODEC_FLAG_QPEL) != 0;
     s->mpeg_quant         = avctx->mpeg_quant;
@@ -368,11 +327,6 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     /* Fixed QSCALE */
     s->fixed_qscale = !!(avctx->flags & CODEC_FLAG_QSCALE);
 
-#if FF_API_MPV_GLOBAL_OPTS
-    if (s->flags & CODEC_FLAG_QP_RD)
-        s->mpv_flags |= FF_MPV_FLAG_QP_RD;
-#endif
-
     s->adaptive_quant = (s->avctx->lumi_masking ||
                          s->avctx->dark_masking ||
                          s->avctx->temporal_cplx_masking ||
@@ -489,11 +443,6 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         return -1;
     }
 
-#if FF_API_MPV_GLOBAL_OPTS
-    if (s->flags & CODEC_FLAG_CBP_RD)
-        s->mpv_flags |= FF_MPV_FLAG_CBP_RD;
-#endif
-
     if ((s->mpv_flags & FF_MPV_FLAG_CBP_RD) && !avctx->trellis) {
         av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
         return -1;
@@ -560,11 +509,6 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     }
 
     i = (INT_MAX / 2 + 128) >> 8;
-    if (avctx->me_threshold >= i) {
-        av_log(avctx, AV_LOG_ERROR, "me_threshold too large, max is %d\n",
-               i - 1);
-        return -1;
-    }
     if (avctx->mb_threshold >= i) {
         av_log(avctx, AV_LOG_ERROR, "mb_threshold too large, max is %d\n",
                i - 1);
@@ -601,9 +545,6 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     if (avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
         s->inter_quant_bias = avctx->inter_quant_bias;
 
-    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift,
-                                  &chroma_v_shift);
-
     if (avctx->codec_id == AV_CODEC_ID_MPEG4 &&
         s->avctx->time_base.den > (1 << 16) - 1) {
         av_log(avctx, AV_LOG_ERROR,
@@ -615,15 +556,6 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     }
     s->time_increment_bits = av_log2(s->avctx->time_base.den - 1) + 1;
 
-#if FF_API_MPV_GLOBAL_OPTS
-    if (avctx->flags2 & CODEC_FLAG2_SKIP_RD)
-        s->mpv_flags |= FF_MPV_FLAG_SKIP_RD;
-    if (avctx->flags2 & CODEC_FLAG2_STRICT_GOP)
-        s->mpv_flags |= FF_MPV_FLAG_STRICT_GOP;
-    if (avctx->quantizer_noise_shaping)
-        s->quantizer_noise_shaping = avctx->quantizer_noise_shaping;
-#endif
-
     switch (avctx->codec->id) {
     case AV_CODEC_ID_MPEG1VIDEO:
         s->out_format = FMT_MPEG1;
@@ -636,24 +568,10 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         avctx->delay  = s->low_delay ? 0 : (s->max_b_frames + 1);
         s->rtp_mode   = 1;
         break;
-    case AV_CODEC_ID_LJPEG:
     case AV_CODEC_ID_MJPEG:
         s->out_format = FMT_MJPEG;
         s->intra_only = 1; /* force intra only for jpeg */
-        if (avctx->codec->id == AV_CODEC_ID_LJPEG &&
-            avctx->pix_fmt   == PIX_FMT_BGRA) {
-            s->mjpeg_vsample[0] = s->mjpeg_hsample[0] =
-            s->mjpeg_vsample[1] = s->mjpeg_hsample[1] =
-            s->mjpeg_vsample[2] = s->mjpeg_hsample[2] = 1;
-        } else {
-            s->mjpeg_vsample[0] = 2;
-            s->mjpeg_vsample[1] = 2 >> chroma_v_shift;
-            s->mjpeg_vsample[2] = 2 >> chroma_v_shift;
-            s->mjpeg_hsample[0] = 2;
-            s->mjpeg_hsample[1] = 2 >> chroma_h_shift;
-            s->mjpeg_hsample[2] = 2 >> chroma_h_shift;
-        }
-        if (!(CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) ||
+        if (!CONFIG_MJPEG_ENCODER ||
             ff_mjpeg_encode_init(s) < 0)
             return -1;
         avctx->delay = 0;
@@ -785,6 +703,40 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     if (ff_MPV_common_init(s) < 0)
         return -1;
 
+    if (ARCH_X86)
+        ff_MPV_encode_init_x86(s);
+
+    ff_fdctdsp_init(&s->fdsp, avctx);
+    ff_me_cmp_init(&s->mecc, avctx);
+    ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
+    ff_pixblockdsp_init(&s->pdsp, avctx);
+    ff_qpeldsp_init(&s->qdsp);
+
+    s->avctx->coded_frame = s->current_picture.f;
+
+    if (s->msmpeg4_version) {
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats,
+                          2 * 2 * (MAX_LEVEL + 1) *
+                          (MAX_RUN + 1) * 2 * sizeof(int), fail);
+    }
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, 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,   64 * 32 * sizeof(int), fail);
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail);
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail);
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture,
+                      MAX_PICTURE_COUNT * sizeof(Picture *), fail);
+    FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture,
+                      MAX_PICTURE_COUNT * sizeof(Picture *), fail);
+
+    if (s->avctx->noise_reduction) {
+        FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset,
+                          2 * 64 * sizeof(uint16_t), fail);
+    }
+
+    if (CONFIG_H263_ENCODER)
+        ff_h263dsp_init(&s->h263dsp);
     if (!s->dct_quantize)
         s->dct_quantize = ff_dct_quantize_c;
     if (!s->denoise_dct)
@@ -798,8 +750,8 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
 
     s->quant_precision = 5;
 
-    ff_set_cmp(&s->dsp, s->dsp.ildct_cmp, s->avctx->ildct_cmp);
-    ff_set_cmp(&s->dsp, s->dsp.frame_skip_cmp, s->avctx->frame_skip_cmp);
+    ff_set_cmp(&s->mecc, s->mecc.ildct_cmp,      s->avctx->ildct_cmp);
+    ff_set_cmp(&s->mecc, s->mecc.frame_skip_cmp, s->avctx->frame_skip_cmp);
 
     if (CONFIG_H261_ENCODER && s->out_format == FMT_H261)
         ff_h261_encode_init(s);
@@ -813,7 +765,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
 
     /* init q matrix */
     for (i = 0; i < 64; i++) {
-        int j = s->dsp.idct_permutation[i];
+        int j = s->idsp.idct_permutation[i];
         if (CONFIG_MPEG4_ENCODER && s->codec_id == AV_CODEC_ID_MPEG4 &&
             s->mpeg_quant) {
             s->intra_matrix[j] = ff_mpeg4_default_intra_matrix[i];
@@ -835,10 +787,10 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     /* precompute matrix */
     /* for mjpeg, we do include qscale in the matrix */
     if (s->out_format != FMT_MJPEG) {
-        ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
+        ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
                           s->intra_matrix, s->intra_quant_bias, avctx->qmin,
                           31, 1);
-        ff_convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16,
+        ff_convert_matrix(s, s->q_inter_matrix, s->q_inter_matrix16,
                           s->inter_matrix, s->inter_quant_bias, avctx->qmin,
                           31, 0);
     }
@@ -846,22 +798,80 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     if (ff_rate_control_init(s) < 0)
         return -1;
 
+#if FF_API_ERROR_RATE
+    FF_DISABLE_DEPRECATION_WARNINGS
+    if (avctx->error_rate)
+        s->error_rate = avctx->error_rate;
+    FF_ENABLE_DEPRECATION_WARNINGS;
+#endif
+
+#if FF_API_NORMALIZE_AQP
+    FF_DISABLE_DEPRECATION_WARNINGS
+    if (avctx->flags & CODEC_FLAG_NORMALIZE_AQP)
+        s->mpv_flags |= FF_MPV_FLAG_NAQ;
+    FF_ENABLE_DEPRECATION_WARNINGS;
+#endif
+
+#if FF_API_MV0
+    FF_DISABLE_DEPRECATION_WARNINGS
+    if (avctx->flags & CODEC_FLAG_MV0)
+        s->mpv_flags |= FF_MPV_FLAG_MV0;
+    FF_ENABLE_DEPRECATION_WARNINGS
+#endif
+
+    if (avctx->b_frame_strategy == 2) {
+        for (i = 0; i < s->max_b_frames + 2; i++) {
+            s->tmp_frames[i] = av_frame_alloc();
+            if (!s->tmp_frames[i])
+                return AVERROR(ENOMEM);
+
+            s->tmp_frames[i]->format = AV_PIX_FMT_YUV420P;
+            s->tmp_frames[i]->width  = s->width  >> avctx->brd_scale;
+            s->tmp_frames[i]->height = s->height >> avctx->brd_scale;
+
+            ret = av_frame_get_buffer(s->tmp_frames[i], 32);
+            if (ret < 0)
+                return ret;
+        }
+    }
+
     return 0;
+fail:
+    ff_MPV_encode_end(avctx);
+    return AVERROR_UNKNOWN;
 }
 
 av_cold int ff_MPV_encode_end(AVCodecContext *avctx)
 {
     MpegEncContext *s = avctx->priv_data;
+    int i;
 
     ff_rate_control_uninit(s);
 
     ff_MPV_common_end(s);
-    if ((CONFIG_MJPEG_ENCODER || CONFIG_LJPEG_ENCODER) &&
+    if (CONFIG_MJPEG_ENCODER &&
         s->out_format == FMT_MJPEG)
         ff_mjpeg_encode_close(s);
 
     av_freep(&avctx->extradata);
 
+    for (i = 0; i < FF_ARRAY_ELEMS(s->tmp_frames); i++)
+        av_frame_free(&s->tmp_frames[i]);
+
+    ff_free_picture_tables(&s->new_picture);
+    ff_mpeg_unref_picture(s, &s->new_picture);
+
+    av_freep(&s->avctx->stats_out);
+    av_freep(&s->ac_stats);
+
+    av_freep(&s->q_intra_matrix);
+    av_freep(&s->q_inter_matrix);
+    av_freep(&s->q_intra_matrix16);
+    av_freep(&s->q_inter_matrix16);
+    av_freep(&s->input_picture);
+    av_freep(&s->reordered_input_picture);
+    av_freep(&s->dct_offset);
+
     return 0;
 }
 
@@ -891,9 +901,9 @@ static int get_intra_count(MpegEncContext *s, uint8_t *src,
     for (y = 0; y < h; y += 16) {
         for (x = 0; x < w; x += 16) {
             int offset = x + y * stride;
-            int sad  = s->dsp.sad[0](NULL, src + offset, ref + offset, stride,
-                                     16);
-            int mean = (s->dsp.pix_sum(src + offset, stride) + 128) >> 8;
+            int sad  = s->mecc.sad[0](NULL, src + offset, ref + offset,
+                                      stride, 16);
+            int mean = (s->mpvencdsp.pix_sum(src + offset, stride) + 128) >> 8;
             int sae  = get_sae(src + offset, mean, stride);
 
             acc += sae + 500 < sad;
@@ -903,18 +913,18 @@ static int get_intra_count(MpegEncContext *s, uint8_t *src,
 }
 
 
-static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg)
+static int load_input_picture(MpegEncContext *s, const AVFrame *pic_arg)
 {
-    AVFrame *pic = NULL;
+    Picture *pic = NULL;
     int64_t pts;
-    int i;
+    int i, display_picture_number = 0, ret;
     const int encoding_delay = s->max_b_frames ? s->max_b_frames :
                                                  (s->low_delay ? 0 : 1);
     int direct = 1;
 
     if (pic_arg) {
         pts = pic_arg->pts;
-        pic_arg->display_picture_number = s->input_picture_number++;
+        display_picture_number = s->input_picture_number++;
 
         if (pts != AV_NOPTS_VALUE) {
             if (s->user_specified_pts != AV_NOPTS_VALUE) {
@@ -928,7 +938,7 @@ static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg)
                     return -1;
                 }
 
-                if (!s->low_delay && pic_arg->display_picture_number == 1)
+                if (!s->low_delay && display_picture_number == 1)
                     s->dts_delta = time - last;
             }
             s->user_specified_pts = pts;
@@ -940,88 +950,91 @@ static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg)
                        "Warning: AVFrame.pts=? trying to guess (%"PRId64")\n",
                        pts);
             } else {
-                pts = pic_arg->display_picture_number;
+                pts = display_picture_number;
             }
         }
     }
 
-  if (pic_arg) {
-    if (encoding_delay && !(s->flags & CODEC_FLAG_INPUT_PRESERVED))
-        direct = 0;
-    if (pic_arg->linesize[0] != s->linesize)
-        direct = 0;
-    if (pic_arg->linesize[1] != s->uvlinesize)
-        direct = 0;
-    if (pic_arg->linesize[2] != s->uvlinesize)
-        direct = 0;
-
-    //av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0],
-    //       pic_arg->linesize[1], s->linesize, s->uvlinesize);
-
-    if (direct) {
-        i = ff_find_unused_picture(s, 1);
-        if (i < 0)
-            return i;
+    if (pic_arg) {
+        if (!pic_arg->buf[0]);
+            direct = 0;
+        if (pic_arg->linesize[0] != s->linesize)
+            direct = 0;
+        if (pic_arg->linesize[1] != s->uvlinesize)
+            direct = 0;
+        if (pic_arg->linesize[2] != s->uvlinesize)
+            direct = 0;
+
+        av_dlog(s->avctx, "%d %d %td %td\n", pic_arg->linesize[0],
+                pic_arg->linesize[1], s->linesize, s->uvlinesize);
+
+        if (direct) {
+            i = ff_find_unused_picture(s, 1);
+            if (i < 0)
+                return i;
 
-        pic = &s->picture[i].f;
-        pic->reference = 3;
+            pic = &s->picture[i];
+            pic->reference = 3;
 
-        for (i = 0; i < 4; i++) {
-            pic->data[i]     = pic_arg->data[i];
-            pic->linesize[i] = pic_arg->linesize[i];
-        }
-        if (ff_alloc_picture(s, (Picture *) pic, 1) < 0) {
-            return -1;
-        }
-    } else {
-        i = ff_find_unused_picture(s, 0);
-        if (i < 0)
-            return i;
+            if ((ret = av_frame_ref(pic->f, pic_arg)) < 0)
+                return ret;
+            if (ff_alloc_picture(s, pic, 1) < 0) {
+                return -1;
+            }
+        } else {
+            i = ff_find_unused_picture(s, 0);
+            if (i < 0)
+                return i;
 
-        pic = &s->picture[i].f;
-        pic->reference = 3;
+            pic = &s->picture[i];
+            pic->reference = 3;
 
-        if (ff_alloc_picture(s, (Picture *) pic, 0) < 0) {
-            return -1;
-        }
+            if (ff_alloc_picture(s, pic, 0) < 0) {
+                return -1;
+            }
 
-        if (pic->data[0] + INPLACE_OFFSET == pic_arg->data[0] &&
-            pic->data[1] + INPLACE_OFFSET == pic_arg->data[1] &&
-            pic->data[2] + INPLACE_OFFSET == pic_arg->data[2]) {
-            // empty
-        } else {
-            int h_chroma_shift, v_chroma_shift;
-            avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift,
-                                          &v_chroma_shift);
-
-            for (i = 0; i < 3; i++) {
-                int src_stride = pic_arg->linesize[i];
-                int dst_stride = i ? s->uvlinesize : s->linesize;
-                int h_shift = i ? h_chroma_shift : 0;
-                int v_shift = i ? v_chroma_shift : 0;
-                int w = s->width  >> h_shift;
-                int h = s->height >> v_shift;
-                uint8_t *src = pic_arg->data[i];
-                uint8_t *dst = pic->data[i];
-
-                if (!s->avctx->rc_buffer_size)
-                    dst += INPLACE_OFFSET;
-
-                if (src_stride == dst_stride)
-                    memcpy(dst, src, src_stride * h);
-                else {
-                    while (h--) {
-                        memcpy(dst, src, w);
-                        dst += dst_stride;
-                        src += src_stride;
+            if (pic->f->data[0] + INPLACE_OFFSET == pic_arg->data[0] &&
+                pic->f->data[1] + INPLACE_OFFSET == pic_arg->data[1] &&
+                pic->f->data[2] + INPLACE_OFFSET == pic_arg->data[2]) {
+                // empty
+            } else {
+                int h_chroma_shift, v_chroma_shift;
+                av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
+                                                 &h_chroma_shift,
+                                                 &v_chroma_shift);
+
+                for (i = 0; i < 3; i++) {
+                    int src_stride = pic_arg->linesize[i];
+                    int dst_stride = i ? s->uvlinesize : s->linesize;
+                    int h_shift = i ? h_chroma_shift : 0;
+                    int v_shift = i ? v_chroma_shift : 0;
+                    int w = s->width  >> h_shift;
+                    int h = s->height >> v_shift;
+                    uint8_t *src = pic_arg->data[i];
+                    uint8_t *dst = pic->f->data[i];
+
+                    if (!s->avctx->rc_buffer_size)
+                        dst += INPLACE_OFFSET;
+
+                    if (src_stride == dst_stride)
+                        memcpy(dst, src, src_stride * h);
+                    else {
+                        while (h--) {
+                            memcpy(dst, src, w);
+                            dst += dst_stride;
+                            src += src_stride;
+                        }
                     }
                 }
             }
         }
+        ret = av_frame_copy_props(pic->f, pic_arg);
+        if (ret < 0)
+            return ret;
+
+        pic->f->display_picture_number = display_picture_number;
+        pic->f->pts = pts; // we set this here to avoid modifiying pic_arg
     }
-    copy_picture_attributes(s, pic, pic_arg);
-    pic->pts = pts; // we set this here to avoid modifiying pic_arg
-  }
 
     /* shift buffer entries */
     for (i = 1; i < MAX_PICTURE_COUNT /*s->encoding_delay + 1*/; i++)
@@ -1039,14 +1052,14 @@ static int skip_check(MpegEncContext *s, Picture *p, Picture *ref)
     int64_t score64 = 0;
 
     for (plane = 0; plane < 3; plane++) {
-        const int stride = p->f.linesize[plane];
+        const int stride = p->f->linesize[plane];
         const int bw = plane ? 1 : 2;
         for (y = 0; y < s->mb_height * bw; y++) {
             for (x = 0; x < s->mb_width * bw; x++) {
-                int off = p->f.type == FF_BUFFER_TYPE_SHARED ? 0 : 16;
-                uint8_t *dptr = p->f.data[plane] + 8 * (x + y * stride) + off;
-                uint8_t *rptr = ref->f.data[plane] + 8 * (x + y * stride);
-                int v   = s->dsp.frame_skip_cmp[1](s, dptr, rptr, stride, 8);
+                int off = p->shared ? 0 : 16;
+                uint8_t *dptr = p->f->data[plane] + 8 * (x + y * stride) + off;
+                uint8_t *rptr = ref->f->data[plane] + 8 * (x + y * stride);
+                int v = s->mecc.frame_skip_cmp[1](s, dptr, rptr, stride, 8);
 
                 switch (s->avctx->frame_skip_exp) {
                 case 0: score    =  FFMAX(score, v);          break;
@@ -1069,15 +1082,27 @@ static int skip_check(MpegEncContext *s, Picture *p, Picture *ref)
     return 0;
 }
 
+static int encode_frame(AVCodecContext *c, AVFrame *frame)
+{
+    AVPacket pkt = { 0 };
+    int ret, got_output;
+
+    av_init_packet(&pkt);
+    ret = avcodec_encode_video2(c, &pkt, frame, &got_output);
+    if (ret < 0)
+        return ret;
+
+    ret = pkt.size;
+    av_free_packet(&pkt);
+    return ret;
+}
+
 static int estimate_best_b_count(MpegEncContext *s)
 {
     AVCodec *codec    = avcodec_find_encoder(s->avctx->codec_id);
     AVCodecContext *c = avcodec_alloc_context3(NULL);
-    AVFrame input[FF_MAX_B_FRAMES + 2];
     const int scale = s->avctx->brd_scale;
     int i, j, out_size, p_lambda, b_lambda, lambda2;
-    int outbuf_size  = s->width * s->height; // FIXME
-    uint8_t *outbuf  = av_malloc(outbuf_size);
     int64_t best_rd  = INT64_MAX;
     int best_b_count = -1;
 
@@ -1095,14 +1120,13 @@ static int estimate_best_b_count(MpegEncContext *s)
 
     c->width        = s->width  >> scale;
     c->height       = s->height >> scale;
-    c->flags        = CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR |
-                      CODEC_FLAG_INPUT_PRESERVED /*| CODEC_FLAG_EMU_EDGE*/;
+    c->flags        = CODEC_FLAG_QSCALE | CODEC_FLAG_PSNR;
     c->flags       |= s->avctx->flags & CODEC_FLAG_QPEL;
     c->mb_decision  = s->avctx->mb_decision;
     c->me_cmp       = s->avctx->me_cmp;
     c->mb_cmp       = s->avctx->mb_cmp;
     c->me_sub_cmp   = s->avctx->me_sub_cmp;
-    c->pix_fmt      = PIX_FMT_YUV420P;
+    c->pix_fmt      = AV_PIX_FMT_YUV420P;
     c->time_base    = s->avctx->time_base;
     c->max_b_frames = s->max_b_frames;
 
@@ -1110,37 +1134,33 @@ static int estimate_best_b_count(MpegEncContext *s)
         return -1;
 
     for (i = 0; i < s->max_b_frames + 2; i++) {
-        int ysize = c->width * c->height;
-        int csize = (c->width / 2) * (c->height / 2);
         Picture pre_input, *pre_input_ptr = i ? s->input_picture[i - 1] :
                                                 s->next_picture_ptr;
 
-        avcodec_get_frame_defaults(&input[i]);
-        input[i].data[0]     = av_malloc(ysize + 2 * csize);
-        input[i].data[1]     = input[i].data[0] + ysize;
-        input[i].data[2]     = input[i].data[1] + csize;
-        input[i].linesize[0] = c->width;
-        input[i].linesize[1] =
-        input[i].linesize[2] = c->width / 2;
-
         if (pre_input_ptr && (!i || s->input_picture[i - 1])) {
             pre_input = *pre_input_ptr;
 
-            if (pre_input.f.type != FF_BUFFER_TYPE_SHARED && i) {
-                pre_input.f.data[0] += INPLACE_OFFSET;
-                pre_input.f.data[1] += INPLACE_OFFSET;
-                pre_input.f.data[2] += INPLACE_OFFSET;
+            if (!pre_input.shared && i) {
+                pre_input.f->data[0] += INPLACE_OFFSET;
+                pre_input.f->data[1] += INPLACE_OFFSET;
+                pre_input.f->data[2] += INPLACE_OFFSET;
             }
 
-            s->dsp.shrink[scale](input[i].data[0], input[i].linesize[0],
-                                 pre_input.f.data[0], pre_input.f.linesize[0],
-                                 c->width,      c->height);
-            s->dsp.shrink[scale](input[i].data[1], input[i].linesize[1],
-                                 pre_input.f.data[1], pre_input.f.linesize[1],
-                                 c->width >> 1, c->height >> 1);
-            s->dsp.shrink[scale](input[i].data[2], input[i].linesize[2],
-                                 pre_input.f.data[2], pre_input.f.linesize[2],
-                                 c->width >> 1, c->height >> 1);
+            s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[0],
+                                       s->tmp_frames[i]->linesize[0],
+                                       pre_input.f->data[0],
+                                       pre_input.f->linesize[0],
+                                       c->width, c->height);
+            s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[1],
+                                       s->tmp_frames[i]->linesize[1],
+                                       pre_input.f->data[1],
+                                       pre_input.f->linesize[1],
+                                       c->width >> 1, c->height >> 1);
+            s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[2],
+                                       s->tmp_frames[i]->linesize[2],
+                                       pre_input.f->data[2],
+                                       pre_input.f->linesize[2],
+                                       c->width >> 1, c->height >> 1);
         }
     }
 
@@ -1152,26 +1172,28 @@ static int estimate_best_b_count(MpegEncContext *s)
 
         c->error[0] = c->error[1] = c->error[2] = 0;
 
-        input[0].pict_type = AV_PICTURE_TYPE_I;
-        input[0].quality   = 1 * FF_QP2LAMBDA;
-        out_size           = avcodec_encode_video(c, outbuf,
-                                                  outbuf_size, &input[0]);
+        s->tmp_frames[0]->pict_type = AV_PICTURE_TYPE_I;
+        s->tmp_frames[0]->quality   = 1 * FF_QP2LAMBDA;
+
+        out_size = encode_frame(c, s->tmp_frames[0]);
+
         //rd += (out_size * lambda2) >> FF_LAMBDA_SHIFT;
 
         for (i = 0; i < s->max_b_frames + 1; i++) {
             int is_p = i % (j + 1) == j || i == s->max_b_frames;
 
-            input[i + 1].pict_type = is_p ?
+            s->tmp_frames[i + 1]->pict_type = is_p ?
                                      AV_PICTURE_TYPE_P : AV_PICTURE_TYPE_B;
-            input[i + 1].quality   = is_p ? p_lambda : b_lambda;
-            out_size = avcodec_encode_video(c, outbuf, outbuf_size,
-                                            &input[i + 1]);
+            s->tmp_frames[i + 1]->quality   = is_p ? p_lambda : b_lambda;
+
+            out_size = encode_frame(c, s->tmp_frames[i + 1]);
+
             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
         }
 
         /* get the delayed frames */
         while (out_size) {
-            out_size = avcodec_encode_video(c, outbuf, outbuf_size, NULL);
+            out_size = encode_frame(c, NULL);
             rd += (out_size * lambda2) >> (FF_LAMBDA_SHIFT - 3);
         }
 
@@ -1183,20 +1205,15 @@ static int estimate_best_b_count(MpegEncContext *s)
         }
     }
 
-    av_freep(&outbuf);
     avcodec_close(c);
     av_freep(&c);
 
-    for (i = 0; i < s->max_b_frames + 2; i++) {
-        av_freep(&input[i].data[0]);
-    }
-
     return best_b_count;
 }
 
 static int select_input_picture(MpegEncContext *s)
 {
-    int i;
+    int i, ret;
 
     for (i = 1; i < MAX_PICTURE_COUNT; i++)
         s->reordered_input_picture[i - 1] = s->reordered_input_picture[i];
@@ -1207,8 +1224,8 @@ static int select_input_picture(MpegEncContext *s)
         if (/*s->picture_in_gop_number >= s->gop_size ||*/
             s->next_picture_ptr == NULL || s->intra_only) {
             s->reordered_input_picture[0] = s->input_picture[0];
-            s->reordered_input_picture[0]->f.pict_type = AV_PICTURE_TYPE_I;
-            s->reordered_input_picture[0]->f.coded_picture_number =
+            s->reordered_input_picture[0]->f->pict_type = AV_PICTURE_TYPE_I;
+            s->reordered_input_picture[0]->f->coded_picture_number =
                 s->coded_picture_number++;
         } else {
             int b_frames;
@@ -1217,21 +1234,7 @@ static int select_input_picture(MpegEncContext *s)
                 if (s->picture_in_gop_number < s->gop_size &&
                     skip_check(s, s->input_picture[0], s->next_picture_ptr)) {
                     // FIXME check that te gop check above is +-1 correct
-                    //av_log(NULL, AV_LOG_DEBUG, "skip %p %"PRId64"\n",
-                    //       s->input_picture[0]->f.data[0],
-                    //       s->input_picture[0]->pts);
-
-                    if (s->input_picture[0]->f.type == FF_BUFFER_TYPE_SHARED) {
-                        for (i = 0; i < 4; i++)
-                            s->input_picture[0]->f.data[i] = NULL;
-                        s->input_picture[0]->f.type = 0;
-                    } else {
-                        assert(s->input_picture[0]->f.type == FF_BUFFER_TYPE_USER ||
-                               s->input_picture[0]->f.type == FF_BUFFER_TYPE_INTERNAL);
-
-                        s->avctx->release_buffer(s->avctx,
-                                                 &s->input_picture[0]->f);
-                    }
+                    av_frame_unref(s->input_picture[0]->f);
 
                     emms_c();
                     ff_vbv_update(s, 0);
@@ -1242,7 +1245,7 @@ static int select_input_picture(MpegEncContext *s)
 
             if (s->flags & CODEC_FLAG_PASS2) {
                 for (i = 0; i < s->max_b_frames + 1; i++) {
-                    int pict_num = s->input_picture[0]->f.display_picture_number + i;
+                    int pict_num = s->input_picture[0]->f->display_picture_number + i;
 
                     if (pict_num >= s->rc_context.num_entries)
                         break;
@@ -1251,7 +1254,7 @@ static int select_input_picture(MpegEncContext *s)
                         break;
                     }
 
-                    s->input_picture[i]->f.pict_type =
+                    s->input_picture[i]->f->pict_type =
                         s->rc_context.entry[pict_num].new_pict_type;
                 }
             }
@@ -1266,8 +1269,8 @@ static int select_input_picture(MpegEncContext *s)
                         s->input_picture[i]->b_frame_score == 0) {
                         s->input_picture[i]->b_frame_score =
                             get_intra_count(s,
-                                            s->input_picture[i    ]->f.data[0],
-                                            s->input_picture[i - 1]->f.data[0],
+                                            s->input_picture[i    ]->f->data[0],
+                                            s->input_picture[i - 1]->f->data[0],
                                             s->linesize) + 1;
                     }
                 }
@@ -1292,16 +1295,13 @@ static int select_input_picture(MpegEncContext *s)
             }
 
             emms_c();
-            //static int b_count = 0;
-            //b_count += b_frames;
-            //av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
 
             for (i = b_frames - 1; i >= 0; i--) {
-                int type = s->input_picture[i]->f.pict_type;
+                int type = s->input_picture[i]->f->pict_type;
                 if (type && type != AV_PICTURE_TYPE_B)
                     b_frames = i;
             }
-            if (s->input_picture[b_frames]->f.pict_type == AV_PICTURE_TYPE_B &&
+            if (s->input_picture[b_frames]->f->pict_type == AV_PICTURE_TYPE_B &&
                 b_frames == s->max_b_frames) {
                 av_log(s->avctx, AV_LOG_ERROR,
                        "warning, too many b frames in a row\n");
@@ -1314,38 +1314,39 @@ static int select_input_picture(MpegEncContext *s)
                 } else {
                     if (s->flags & CODEC_FLAG_CLOSED_GOP)
                         b_frames = 0;
-                    s->input_picture[b_frames]->f.pict_type = AV_PICTURE_TYPE_I;
+                    s->input_picture[b_frames]->f->pict_type = AV_PICTURE_TYPE_I;
                 }
             }
 
             if ((s->flags & CODEC_FLAG_CLOSED_GOP) && b_frames &&
-                s->input_picture[b_frames]->f.pict_type == AV_PICTURE_TYPE_I)
+                s->input_picture[b_frames]->f->pict_type == AV_PICTURE_TYPE_I)
                 b_frames--;
 
             s->reordered_input_picture[0] = s->input_picture[b_frames];
-            if (s->reordered_input_picture[0]->f.pict_type != AV_PICTURE_TYPE_I)
-                s->reordered_input_picture[0]->f.pict_type = AV_PICTURE_TYPE_P;
-            s->reordered_input_picture[0]->f.coded_picture_number =
+            if (s->reordered_input_picture[0]->f->pict_type != AV_PICTURE_TYPE_I)
+                s->reordered_input_picture[0]->f->pict_type = AV_PICTURE_TYPE_P;
+            s->reordered_input_picture[0]->f->coded_picture_number =
                 s->coded_picture_number++;
             for (i = 0; i < b_frames; i++) {
                 s->reordered_input_picture[i + 1] = s->input_picture[i];
-                s->reordered_input_picture[i + 1]->f.pict_type =
+                s->reordered_input_picture[i + 1]->f->pict_type =
                     AV_PICTURE_TYPE_B;
-                s->reordered_input_picture[i + 1]->f.coded_picture_number =
+                s->reordered_input_picture[i + 1]->f->coded_picture_number =
                     s->coded_picture_number++;
             }
         }
     }
 no_output_pic:
     if (s->reordered_input_picture[0]) {
-        s->reordered_input_picture[0]->f.reference =
-           s->reordered_input_picture[0]->f.pict_type !=
+        s->reordered_input_picture[0]->reference =
+           s->reordered_input_picture[0]->f->pict_type !=
                AV_PICTURE_TYPE_B ? 3 : 0;
 
-        ff_copy_picture(&s->new_picture, s->reordered_input_picture[0]);
+        ff_mpeg_unref_picture(s, &s->new_picture);
+        if ((ret = ff_mpeg_ref_picture(s, &s->new_picture, s->reordered_input_picture[0])))
+            return ret;
 
-        if (s->reordered_input_picture[0]->f.type == FF_BUFFER_TYPE_SHARED ||
-            s->avctx->rc_buffer_size) {
+        if (s->reordered_input_picture[0]->shared || s->avctx->rc_buffer_size) {
             // input is a shared pix, so we can't modifiy it -> alloc a new
             // one & ensure that the shared one is reuseable
 
@@ -1355,43 +1356,176 @@ no_output_pic:
                 return i;
             pic = &s->picture[i];
 
-            pic->f.reference = s->reordered_input_picture[0]->f.reference;
+            pic->reference = s->reordered_input_picture[0]->reference;
             if (ff_alloc_picture(s, pic, 0) < 0) {
                 return -1;
             }
 
-            /* mark us unused / free shared pic */
-            if (s->reordered_input_picture[0]->f.type == FF_BUFFER_TYPE_INTERNAL)
-                s->avctx->release_buffer(s->avctx,
-                                         &s->reordered_input_picture[0]->f);
-            for (i = 0; i < 4; i++)
-                s->reordered_input_picture[0]->f.data[i] = NULL;
-            s->reordered_input_picture[0]->f.type = 0;
+            ret = av_frame_copy_props(pic->f, s->reordered_input_picture[0]->f);
+            if (ret < 0)
+                return ret;
 
-            copy_picture_attributes(s, &pic->f,
-                                    &s->reordered_input_picture[0]->f);
+            /* mark us unused / free shared pic */
+            av_frame_unref(s->reordered_input_picture[0]->f);
+            s->reordered_input_picture[0]->shared = 0;
 
             s->current_picture_ptr = pic;
         } else {
             // input is not a shared pix -> reuse buffer for current_pix
-
-            assert(s->reordered_input_picture[0]->f.type ==
-                       FF_BUFFER_TYPE_USER ||
-                   s->reordered_input_picture[0]->f.type ==
-                       FF_BUFFER_TYPE_INTERNAL);
-
             s->current_picture_ptr = s->reordered_input_picture[0];
             for (i = 0; i < 4; i++) {
-                s->new_picture.f.data[i] += INPLACE_OFFSET;
+                s->new_picture.f->data[i] += INPLACE_OFFSET;
+            }
+        }
+        ff_mpeg_unref_picture(s, &s->current_picture);
+        if ((ret = ff_mpeg_ref_picture(s, &s->current_picture,
+                                       s->current_picture_ptr)) < 0)
+            return ret;
+
+        s->picture_number = s->new_picture.f->display_picture_number;
+    } else {
+        ff_mpeg_unref_picture(s, &s->new_picture);
+    }
+    return 0;
+}
+
+static void frame_end(MpegEncContext *s)
+{
+    int i;
+
+    if (s->unrestricted_mv &&
+        s->current_picture.reference &&
+        !s->intra_only) {
+        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->mpvencdsp.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->mpvencdsp.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->mpvencdsp.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;
+
+    if (s->encoding) {
+        /* release non-reference frames */
+        for (i = 0; i < MAX_PICTURE_COUNT; i++) {
+            if (!s->picture[i].reference)
+                ff_mpeg_unref_picture(s, &s->picture[i]);
+        }
+    }
+
+    s->avctx->coded_frame = s->current_picture_ptr->f;
+
+}
+
+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;
             }
+            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);
         }
-        ff_copy_picture(&s->current_picture, s->current_picture_ptr);
+    }
+}
+
+static int frame_start(MpegEncContext *s)
+{
+    int ret;
+
+    /* 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->buf[0]) {
+        ff_mpeg_unref_picture(s, s->last_picture_ptr);
+    }
+
+    s->current_picture_ptr->f->pict_type = s->pict_type;
+    s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
 
-        s->picture_number = s->new_picture.f.display_picture_number;
-        //printf("dpn:%d\n", s->picture_number);
+    ff_mpeg_unref_picture(s, &s->current_picture);
+    if ((ret = ff_mpeg_ref_picture(s, &s->current_picture,
+                                   s->current_picture_ptr)) < 0)
+        return ret;
+
+    if (s->pict_type != AV_PICTURE_TYPE_B) {
+        s->last_picture_ptr = s->next_picture_ptr;
+        if (!s->droppable)
+            s->next_picture_ptr = s->current_picture_ptr;
+    }
+
+    if (s->last_picture_ptr) {
+        ff_mpeg_unref_picture(s, &s->last_picture);
+        if (s->last_picture_ptr->f->buf[0] &&
+            (ret = ff_mpeg_ref_picture(s, &s->last_picture,
+                                       s->last_picture_ptr)) < 0)
+            return ret;
+    }
+    if (s->next_picture_ptr) {
+        ff_mpeg_unref_picture(s, &s->next_picture);
+        if (s->next_picture_ptr->f->buf[0] &&
+            (ret = ff_mpeg_ref_picture(s, &s->next_picture,
+                                       s->next_picture_ptr)) < 0)
+            return ret;
+    }
+
+    if (s->picture_structure!= PICT_FRAME) {
+        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];
+            }
+            s->current_picture.f->linesize[i] *= 2;
+            s->last_picture.f->linesize[i]    *= 2;
+            s->next_picture.f->linesize[i]    *= 2;
+        }
+    }
+
+    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) {
+        s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
+        s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
     } else {
-        memset(&s->new_picture, 0, sizeof(Picture));
+        s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
+        s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
     }
+
+    if (s->dct_error_sum) {
+        assert(s->avctx->noise_reduction && s->encoding);
+        update_noise_reduction(s);
+    }
+
     return 0;
 }
 
@@ -1412,7 +1546,7 @@ int ff_MPV_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
     }
 
     /* output? */
-    if (s->new_picture.f.data[0]) {
+    if (s->new_picture.f->data[0]) {
         if (!pkt->data &&
             (ret = ff_alloc_packet(pkt, s->mb_width*s->mb_height*MAX_MB_BYTES)) < 0)
             return ret;
@@ -1433,11 +1567,11 @@ int ff_MPV_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
             init_put_bits(&s->thread_context[i]->pb, start, end - start);
         }
 
-        s->pict_type = s->new_picture.f.pict_type;
+        s->pict_type = s->new_picture.f->pict_type;
         //emms_c();
-        //printf("qs:%f %f %d\n", s->new_picture.quality,
-        //       s->current_picture.quality, s->qscale);
-        ff_MPV_frame_start(s, avctx);
+        ret = frame_start(s);
+        if (ret < 0)
+            return ret;
 vbv_retry:
         if (encode_picture(s, s->picture_number) < 0)
             return -1;
@@ -1452,10 +1586,10 @@ vbv_retry:
         avctx->p_count     = s->mb_num - s->i_count - s->skip_count;
         avctx->skip_count  = s->skip_count;
 
-        ff_MPV_frame_end(s);
+        frame_end(s);
 
         if (CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG)
-            ff_mjpeg_encode_picture_trailer(s);
+            ff_mjpeg_encode_picture_trailer(&s->pb, s->header_bits);
 
         if (avctx->rc_buffer_size) {
             RateControlContext *rcc = &s->rc_context;
@@ -1473,7 +1607,7 @@ vbv_retry:
                                   s->lambda_table[i] * (s->qscale + 1) /
                                   s->qscale);
                 }
-                s->mb_skipped = 0;        // done in MPV_frame_start()
+                s->mb_skipped = 0;        // done in frame_start()
                 // done in encode_picture() so we must undo it
                 if (s->pict_type == AV_PICTURE_TYPE_P) {
                     if (s->flipflop_rounding          ||
@@ -1485,7 +1619,6 @@ vbv_retry:
                     s->time_base       = s->last_time_base;
                     s->last_non_b_time = s->time - s->pp_time;
                 }
-                //av_log(NULL, AV_LOG_ERROR, "R:%d ", s->next_lambda);
                 for (i = 0; i < context_count; i++) {
                     PutBitContext *pb = &s->thread_context[i]->pb;
                     init_put_bits(pb, pb->buf, pb->buf_end - pb->buf);
@@ -1500,8 +1633,8 @@ vbv_retry:
             ff_write_pass1_stats(s);
 
         for (i = 0; i < 4; i++) {
-            s->current_picture_ptr->f.error[i] = s->current_picture.f.error[i];
-            avctx->error[i] += s->current_picture_ptr->f.error[i];
+            s->current_picture_ptr->f->error[i] = s->current_picture.f->error[i];
+            avctx->error[i] += s->current_picture_ptr->f->error[i];
         }
 
         if (s->flags & CODEC_FLAG_PASS1)
@@ -1578,21 +1711,20 @@ vbv_retry:
         s->total_bits     += s->frame_bits;
         avctx->frame_bits  = s->frame_bits;
 
-        pkt->pts = s->current_picture.f.pts;
-        if (!s->low_delay) {
-            if (!s->current_picture.f.coded_picture_number)
+        pkt->pts = s->current_picture.f->pts;
+        if (!s->low_delay && s->pict_type != AV_PICTURE_TYPE_B) {
+            if (!s->current_picture.f->coded_picture_number)
                 pkt->dts = pkt->pts - s->dts_delta;
             else
                 pkt->dts = s->reordered_pts;
-            s->reordered_pts = s->input_picture[0]->f.pts;
+            s->reordered_pts = pkt->pts;
         } else
             pkt->dts = pkt->pts;
-        if (s->current_picture.f.key_frame)
+        if (s->current_picture.f->key_frame)
             pkt->flags |= AV_PKT_FLAG_KEY;
         if (s->mb_info)
             av_packet_shrink_side_data(pkt, AV_PKT_DATA_H263_MB_INFO, s->mb_info_size);
     } else {
-        assert((put_bits_ptr(&s->pb) == s->pb.buf));
         s->frame_bits = 0;
     }
     assert((s->frame_bits & 7) == 0);
@@ -1618,7 +1750,7 @@ static inline void dct_single_coeff_elimination(MpegEncContext *s,
     int score = 0;
     int run = 0;
     int i;
-    DCTELEM *block = s->block[n];
+    int16_t *block = s->block[n];
     const int last_index = s->block_last_index[n];
     int skip_dc;
 
@@ -1658,7 +1790,7 @@ static inline void dct_single_coeff_elimination(MpegEncContext *s,
         s->block_last_index[n] = -1;
 }
 
-static inline void clip_coeffs(MpegEncContext *s, DCTELEM *block,
+static inline void clip_coeffs(MpegEncContext *s, int16_t *block,
                                int last_index)
 {
     int i;
@@ -1722,14 +1854,14 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
                                                 int mb_block_count)
 {
     int16_t weight[8][64];
-    DCTELEM orig[8][64];
+    int16_t orig[8][64];
     const int mb_x = s->mb_x;
     const int mb_y = s->mb_y;
     int i;
     int skip_dct[8];
     int dct_offset = s->linesize * 8; // default for progressive frames
     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
-    int wrap_y, wrap_c;
+    ptrdiff_t wrap_y, wrap_c;
 
     for (i = 0; i < mb_block_count; i++)
         skip_dct[i] = s->skipdct;
@@ -1742,7 +1874,7 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
         update_qscale(s);
 
         if (!(s->mpv_flags & FF_MPV_FLAG_QP_RD)) {
-            s->qscale = s->current_picture_ptr->f.qscale_table[mb_xy];
+            s->qscale = s->current_picture_ptr->qscale_table[mb_xy];
             s->dquant = s->qscale - last_qp;
 
             if (s->out_format == FMT_H263) {
@@ -1766,25 +1898,29 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
 
     wrap_y = s->linesize;
     wrap_c = s->uvlinesize;
-    ptr_y  = s->new_picture.f.data[0] +
+    ptr_y  = s->new_picture.f->data[0] +
              (mb_y * 16 * wrap_y)              + mb_x * 16;
-    ptr_cb = s->new_picture.f.data[1] +
+    ptr_cb = s->new_picture.f->data[1] +
              (mb_y * mb_block_height * wrap_c) + mb_x * 8;
-    ptr_cr = s->new_picture.f.data[2] +
+    ptr_cr = s->new_picture.f->data[2] +
              (mb_y * mb_block_height * wrap_c) + mb_x * 8;
 
     if (mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) {
         uint8_t *ebuf = s->edge_emu_buffer + 32;
-        s->dsp.emulated_edge_mc(ebuf, ptr_y, wrap_y, 16, 16, mb_x * 16,
-                                mb_y * 16, s->width, s->height);
+        s->vdsp.emulated_edge_mc(ebuf, ptr_y,
+                                 wrap_y, wrap_y,
+                                 16, 16, mb_x * 16, mb_y * 16,
+                                 s->width, s->height);
         ptr_y = ebuf;
-        s->dsp.emulated_edge_mc(ebuf + 18 * wrap_y, ptr_cb, wrap_c, 8,
-                                mb_block_height, mb_x * 8, mb_y * 8,
-                                s->width >> 1, s->height >> 1);
+        s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y, ptr_cb,
+                                 wrap_c, wrap_c,
+                                 8, mb_block_height, mb_x * 8, mb_y * 8,
+                                 s->width >> 1, s->height >> 1);
         ptr_cb = ebuf + 18 * wrap_y;
-        s->dsp.emulated_edge_mc(ebuf + 18 * wrap_y + 8, ptr_cr, wrap_c, 8,
-                                mb_block_height, mb_x * 8, mb_y * 8,
-                                s->width >> 1, s->height >> 1);
+        s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y + 8, ptr_cr,
+                                 wrap_c, wrap_c,
+                                 8, mb_block_height, mb_x * 8, mb_y * 8,
+                                 s->width >> 1, s->height >> 1);
         ptr_cr = ebuf + 18 * wrap_y + 8;
     }
 
@@ -1793,16 +1929,15 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
             int progressive_score, interlaced_score;
 
             s->interlaced_dct = 0;
-            progressive_score = s->dsp.ildct_cmp[4](s, ptr_y,
-                                                    NULL, wrap_y, 8) +
-                                s->dsp.ildct_cmp[4](s, ptr_y + wrap_y * 8,
-                                                    NULL, wrap_y, 8) - 400;
+            progressive_score = s->mecc.ildct_cmp[4](s, ptr_y, NULL, wrap_y, 8) +
+                                s->mecc.ildct_cmp[4](s, ptr_y + wrap_y * 8,
+                                                     NULL, wrap_y, 8) - 400;
 
             if (progressive_score > 0) {
-                interlaced_score = s->dsp.ildct_cmp[4](s, ptr_y,
-                                                       NULL, wrap_y * 2, 8) +
-                                   s->dsp.ildct_cmp[4](s, ptr_y + wrap_y,
-                                                       NULL, wrap_y * 2, 8);
+                interlaced_score = s->mecc.ildct_cmp[4](s, ptr_y,
+                                                        NULL, wrap_y * 2, 8) +
+                                   s->mecc.ildct_cmp[4](s, ptr_y + wrap_y,
+                                                        NULL, wrap_y * 2, 8);
                 if (progressive_score > interlaced_score) {
                     s->interlaced_dct = 1;
 
@@ -1814,22 +1949,22 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
             }
         }
 
-        s->dsp.get_pixels(s->block[0], ptr_y                  , wrap_y);
-        s->dsp.get_pixels(s->block[1], ptr_y              + 8 , wrap_y);
-        s->dsp.get_pixels(s->block[2], ptr_y + dct_offset     , wrap_y);
-        s->dsp.get_pixels(s->block[3], ptr_y + dct_offset + 8 , wrap_y);
+        s->pdsp.get_pixels(s->block[0], ptr_y,                  wrap_y);
+        s->pdsp.get_pixels(s->block[1], ptr_y + 8,              wrap_y);
+        s->pdsp.get_pixels(s->block[2], ptr_y + dct_offset,     wrap_y);
+        s->pdsp.get_pixels(s->block[3], ptr_y + dct_offset + 8, wrap_y);
 
         if (s->flags & CODEC_FLAG_GRAY) {
             skip_dct[4] = 1;
             skip_dct[5] = 1;
         } else {
-            s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
-            s->dsp.get_pixels(s->block[5], ptr_cr, wrap_c);
+            s->pdsp.get_pixels(s->block[4], ptr_cb, wrap_c);
+            s->pdsp.get_pixels(s->block[5], ptr_cr, wrap_c);
             if (!s->chroma_y_shift) { /* 422 */
-                s->dsp.get_pixels(s->block[6],
-                                  ptr_cb + (dct_offset >> 1), wrap_c);
-                s->dsp.get_pixels(s->block[7],
-                                  ptr_cr + (dct_offset >> 1), wrap_c);
+                s->pdsp.get_pixels(s->block[6],
+                                   ptr_cb + (dct_offset >> 1), wrap_c);
+                s->pdsp.get_pixels(s->block[7],
+                                   ptr_cr + (dct_offset >> 1), wrap_c);
             }
         }
     } else {
@@ -1842,45 +1977,44 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
         dest_cr = s->dest[2];
 
         if ((!s->no_rounding) || s->pict_type == AV_PICTURE_TYPE_B) {
-            op_pix  = s->dsp.put_pixels_tab;
-            op_qpix = s->dsp.put_qpel_pixels_tab;
+            op_pix  = s->hdsp.put_pixels_tab;
+            op_qpix = s->qdsp.put_qpel_pixels_tab;
         } else {
-            op_pix  = s->dsp.put_no_rnd_pixels_tab;
-            op_qpix = s->dsp.put_no_rnd_qpel_pixels_tab;
+            op_pix  = s->hdsp.put_no_rnd_pixels_tab;
+            op_qpix = s->qdsp.put_no_rnd_qpel_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->dsp.avg_qpel_pixels_tab;
+            ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 0,
+                          s->last_picture.f->data,
+                          op_pix, op_qpix);
+            op_pix  = s->hdsp.avg_pixels_tab;
+            op_qpix = s->qdsp.avg_qpel_pixels_tab;
         }
         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);
+            ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 1,
+                          s->next_picture.f->data,
+                          op_pix, op_qpix);
         }
 
         if (s->flags & CODEC_FLAG_INTERLACED_DCT) {
             int progressive_score, interlaced_score;
 
             s->interlaced_dct = 0;
-            progressive_score = s->dsp.ildct_cmp[0](s, dest_y,
-                                                    ptr_y,              wrap_y,
-                                                    8) +
-                                s->dsp.ildct_cmp[0](s, dest_y + wrap_y * 8,
-                                                    ptr_y + wrap_y * 8, wrap_y,
-                                                    8) - 400;
+            progressive_score = s->mecc.ildct_cmp[0](s, dest_y, ptr_y, wrap_y, 8) +
+                                s->mecc.ildct_cmp[0](s, dest_y + wrap_y * 8,
+                                                     ptr_y + wrap_y * 8,
+                                                     wrap_y, 8) - 400;
 
             if (s->avctx->ildct_cmp == FF_CMP_VSSE)
                 progressive_score -= 400;
 
             if (progressive_score > 0) {
-                interlaced_score = s->dsp.ildct_cmp[0](s, dest_y,
-                                                       ptr_y,
-                                                       wrap_y * 2, 8) +
-                                   s->dsp.ildct_cmp[0](s, dest_y + wrap_y,
-                                                       ptr_y + wrap_y,
-                                                       wrap_y * 2, 8);
+                interlaced_score = s->mecc.ildct_cmp[0](s, dest_y, ptr_y,
+                                                        wrap_y * 2, 8) +
+                                   s->mecc.ildct_cmp[0](s, dest_y + wrap_y,
+                                                        ptr_y + wrap_y,
+                                                        wrap_y * 2, 8);
 
                 if (progressive_score > interlaced_score) {
                     s->interlaced_dct = 1;
@@ -1893,57 +2027,52 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
             }
         }
 
-        s->dsp.diff_pixels(s->block[0], ptr_y, dest_y, wrap_y);
-        s->dsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
-        s->dsp.diff_pixels(s->block[2], ptr_y + dct_offset,
-                           dest_y + dct_offset, wrap_y);
-        s->dsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8,
-                           dest_y + dct_offset + 8, wrap_y);
+        s->pdsp.diff_pixels(s->block[0], ptr_y, dest_y, wrap_y);
+        s->pdsp.diff_pixels(s->block[1], ptr_y + 8, dest_y + 8, wrap_y);
+        s->pdsp.diff_pixels(s->block[2], ptr_y + dct_offset,
+                            dest_y + dct_offset, wrap_y);
+        s->pdsp.diff_pixels(s->block[3], ptr_y + dct_offset + 8,
+                            dest_y + dct_offset + 8, wrap_y);
 
         if (s->flags & CODEC_FLAG_GRAY) {
             skip_dct[4] = 1;
             skip_dct[5] = 1;
         } else {
-            s->dsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
-            s->dsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
+            s->pdsp.diff_pixels(s->block[4], ptr_cb, dest_cb, wrap_c);
+            s->pdsp.diff_pixels(s->block[5], ptr_cr, dest_cr, wrap_c);
             if (!s->chroma_y_shift) { /* 422 */
-                s->dsp.diff_pixels(s->block[6], ptr_cb + (dct_offset >> 1),
-                                   dest_cb + (dct_offset >> 1), wrap_c);
-                s->dsp.diff_pixels(s->block[7], ptr_cr + (dct_offset >> 1),
-                                   dest_cr + (dct_offset >> 1), wrap_c);
+                s->pdsp.diff_pixels(s->block[6], ptr_cb + (dct_offset >> 1),
+                                    dest_cb + (dct_offset >> 1), wrap_c);
+                s->pdsp.diff_pixels(s->block[7], ptr_cr + (dct_offset >> 1),
+                                    dest_cr + (dct_offset >> 1), wrap_c);
             }
         }
         /* pre quantization */
         if (s->current_picture.mc_mb_var[s->mb_stride * mb_y + mb_x] <
                 2 * s->qscale * s->qscale) {
             // FIXME optimize
-            if (s->dsp.sad[1](NULL, ptr_y , dest_y,
-                              wrap_y, 8) < 20 * s->qscale)
+            if (s->mecc.sad[1](NULL, ptr_y, dest_y, wrap_y, 8) < 20 * s->qscale)
                 skip_dct[0] = 1;
-            if (s->dsp.sad[1](NULL, ptr_y + 8,
-                              dest_y + 8, wrap_y, 8) < 20 * s->qscale)
+            if (s->mecc.sad[1](NULL, ptr_y + 8, dest_y + 8, wrap_y, 8) < 20 * s->qscale)
                 skip_dct[1] = 1;
-            if (s->dsp.sad[1](NULL, ptr_y + dct_offset,
-                              dest_y + dct_offset, wrap_y, 8) < 20 * s->qscale)
+            if (s->mecc.sad[1](NULL, ptr_y + dct_offset, dest_y + dct_offset,
+                               wrap_y, 8) < 20 * s->qscale)
                 skip_dct[2] = 1;
-            if (s->dsp.sad[1](NULL, ptr_y + dct_offset + 8,
-                              dest_y + dct_offset + 8,
-                              wrap_y, 8) < 20 * s->qscale)
+            if (s->mecc.sad[1](NULL, ptr_y + dct_offset + 8, dest_y + dct_offset + 8,
+                               wrap_y, 8) < 20 * s->qscale)
                 skip_dct[3] = 1;
-            if (s->dsp.sad[1](NULL, ptr_cb, dest_cb,
-                              wrap_c, 8) < 20 * s->qscale)
+            if (s->mecc.sad[1](NULL, ptr_cb, dest_cb, wrap_c, 8) < 20 * s->qscale)
                 skip_dct[4] = 1;
-            if (s->dsp.sad[1](NULL, ptr_cr, dest_cr,
-                              wrap_c, 8) < 20 * s->qscale)
+            if (s->mecc.sad[1](NULL, ptr_cr, dest_cr, wrap_c, 8) < 20 * s->qscale)
                 skip_dct[5] = 1;
             if (!s->chroma_y_shift) { /* 422 */
-                if (s->dsp.sad[1](NULL, ptr_cb + (dct_offset >> 1),
-                                  dest_cb + (dct_offset >> 1),
-                                  wrap_c, 8) < 20 * s->qscale)
+                if (s->mecc.sad[1](NULL, ptr_cb + (dct_offset >> 1),
+                                   dest_cb + (dct_offset >> 1),
+                                   wrap_c, 8) < 20 * s->qscale)
                     skip_dct[6] = 1;
-                if (s->dsp.sad[1](NULL, ptr_cr + (dct_offset >> 1),
-                                  dest_cr + (dct_offset >> 1),
-                                  wrap_c, 8) < 20 * s->qscale)
+                if (s->mecc.sad[1](NULL, ptr_cr + (dct_offset >> 1),
+                                   dest_cr + (dct_offset >> 1),
+                                   wrap_c, 8) < 20 * s->qscale)
                     skip_dct[7] = 1;
             }
         }
@@ -1970,7 +2099,7 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
                 get_visual_weight(weight[7], ptr_cr + (dct_offset >> 1),
                                   wrap_c);
         }
-        memcpy(orig[0], s->block[0], sizeof(DCTELEM) * 64 * mb_block_count);
+        memcpy(orig[0], s->block[0], sizeof(int16_t) * 64 * mb_block_count);
     }
 
     /* DCT & quantize */
@@ -2203,14 +2332,14 @@ static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegE
 }
 
 static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
-    uint32_t *sq = ff_squareTbl + 256;
+    uint32_t *sq = ff_square_tab + 256;
     int acc=0;
     int x,y;
 
     if(w==16 && h==16)
-        return s->dsp.sse[0](NULL, src1, src2, stride, 16);
+        return s->mecc.sse[0](NULL, src1, src2, stride, 16);
     else if(w==8 && h==8)
-        return s->dsp.sse[1](NULL, src1, src2, stride, 8);
+        return s->mecc.sse[1](NULL, src1, src2, stride, 8);
 
     for(y=0; y<h; y++){
         for(x=0; x<w; x++){
@@ -2232,18 +2361,18 @@ static int sse_mb(MpegEncContext *s){
 
     if(w==16 && h==16)
       if(s->avctx->mb_cmp == FF_CMP_NSSE){
-        return  s->dsp.nsse[0](s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
-               +s->dsp.nsse[1](s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
-               +s->dsp.nsse[1](s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
+        return s->mecc.nsse[0](s, s->new_picture.f->data[0] + s->mb_x * 16 + s->mb_y * s->linesize   * 16, s->dest[0], s->linesize,   16) +
+               s->mecc.nsse[1](s, s->new_picture.f->data[1] + s->mb_x *  8 + s->mb_y * s->uvlinesize *  8, s->dest[1], s->uvlinesize,  8) +
+               s->mecc.nsse[1](s, s->new_picture.f->data[2] + s->mb_x *  8 + s->mb_y * s->uvlinesize *  8, s->dest[2], s->uvlinesize,  8);
       }else{
-        return  s->dsp.sse[0](NULL, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize, 16)
-               +s->dsp.sse[1](NULL, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize, 8)
-               +s->dsp.sse[1](NULL, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize, 8);
+        return s->mecc.sse[0](NULL, s->new_picture.f->data[0] + s->mb_x * 16 + s->mb_y * s->linesize   * 16, s->dest[0], s->linesize,   16) +
+               s->mecc.sse[1](NULL, s->new_picture.f->data[1] + s->mb_x *  8 + s->mb_y * s->uvlinesize *  8, s->dest[1], s->uvlinesize,  8) +
+               s->mecc.sse[1](NULL, s->new_picture.f->data[2] + s->mb_x *  8 + s->mb_y * s->uvlinesize *  8, s->dest[2], s->uvlinesize,  8);
       }
     else
-        return  sse(s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
-               +sse(s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
-               +sse(s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
+        return  sse(s, s->new_picture.f->data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
+               +sse(s, s->new_picture.f->data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
+               +sse(s, s->new_picture.f->data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
 }
 
 static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
@@ -2268,8 +2397,6 @@ static int pre_estimate_motion_thread(AVCodecContext *c, void *arg){
 static int estimate_motion_thread(AVCodecContext *c, void *arg){
     MpegEncContext *s= *(void**)arg;
 
-    ff_check_alignment();
-
     s->me.dia_size= s->avctx->dia_size;
     s->first_slice_line=1;
     for(s->mb_y= s->start_mb_y; s->mb_y < s->end_mb_y; s->mb_y++) {
@@ -2296,17 +2423,16 @@ static int mb_var_thread(AVCodecContext *c, void *arg){
     MpegEncContext *s= *(void**)arg;
     int mb_x, mb_y;
 
-    ff_check_alignment();
-
     for(mb_y=s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
             int xx = mb_x * 16;
             int yy = mb_y * 16;
-            uint8_t *pix = s->new_picture.f.data[0] + (yy * s->linesize) + xx;
+            uint8_t *pix = s->new_picture.f->data[0] + (yy * s->linesize) + xx;
             int varc;
-            int sum = s->dsp.pix_sum(pix, s->linesize);
+            int sum = s->mpvencdsp.pix_sum(pix, s->linesize);
 
-            varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500 + 128)>>8;
+            varc = (s->mpvencdsp.pix_norm1(pix, s->linesize) -
+                    (((unsigned) sum * sum) >> 8) + 500 + 128) >> 8;
 
             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
@@ -2387,9 +2513,6 @@ static int encode_thread(AVCodecContext *c, void *arg){
     uint8_t bit_buf2[2][MAX_MB_BYTES];
     uint8_t bit_buf_tex[2][MAX_MB_BYTES];
     PutBitContext pb[2], pb2[2], tex_pb[2];
-//printf("%d->%d\n", s->resync_mb_y, s->end_mb_y);
-
-    ff_check_alignment();
 
     for(i=0; i<2; i++){
         init_put_bits(&pb    [i], bit_buf    [i], MAX_MB_BYTES);
@@ -2412,7 +2535,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
         /* note: quant matrix value (8) is implied here */
         s->last_dc[i] = 128 << s->intra_dc_precision;
 
-        s->current_picture.f.error[i] = 0;
+        s->current_picture.f->error[i] = 0;
     }
     s->mb_skip_run = 0;
     memset(s->last_mv, 0, sizeof(s->last_mv));
@@ -2437,7 +2560,6 @@ static int encode_thread(AVCodecContext *c, void *arg){
     s->first_slice_line = 1;
     s->ptr_lastgob = s->pb.buf;
     for(mb_y= s->start_mb_y; mb_y < s->end_mb_y; mb_y++) {
-//    printf("row %d at %X\n", s->mb_y, (int)s);
         s->mb_x=0;
         s->mb_y= mb_y;
 
@@ -2508,9 +2630,9 @@ static int encode_thread(AVCodecContext *c, void *arg){
                     assert((put_bits_count(&s->pb)&7) == 0);
                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
 
-                    if(s->avctx->error_rate && s->resync_mb_x + s->resync_mb_y > 0){
+                    if (s->error_rate && s->resync_mb_x + s->resync_mb_y > 0) {
                         int r= put_bits_count(&s->pb)/8 + s->picture_number + 16 + s->mb_x + s->mb_y;
-                        int d= 100 / s->avctx->error_rate;
+                        int d = 100 / s->error_rate;
                         if(r % d == 0){
                             current_packet_size=0;
                             s->pb.buf_ptr= s->ptr_lastgob;
@@ -2616,8 +2738,8 @@ static int encode_thread(AVCodecContext *c, void *arg){
                     s->mv_type = MV_TYPE_8X8;
                     s->mb_intra= 0;
                     for(i=0; i<4; i++){
-                        s->mv[0][i][0] = s->current_picture.f.motion_val[0][s->block_index[i]][0];
-                        s->mv[0][i][1] = s->current_picture.f.motion_val[0][s->block_index[i]][1];
+                        s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
+                        s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
                     }
                     encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER4V, pb, pb2, tex_pb,
                                  &dmin, &next_block, 0, 0);
@@ -2709,7 +2831,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
                     if(best_s.mv_type==MV_TYPE_16X16){ //FIXME move 4mv after QPRD
                         const int last_qp= backup_s.qscale;
                         int qpi, qp, dc[6];
-                        DCTELEM ac[6][16];
+                        int16_t ac[6][16];
                         const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
                         static const int dquant_tab[4]={-1,1,-2,2};
 
@@ -2734,7 +2856,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
                             if(s->mb_intra && s->dc_val[0]){
                                 for(i=0; i<6; i++){
                                     dc[i]= s->dc_val[0][ s->block_index[i] ];
-                                    memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
+                                    memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(int16_t)*16);
                                 }
                             }
 
@@ -2744,7 +2866,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
                                 if(s->mb_intra && s->dc_val[0]){
                                     for(i=0; i<6; i++){
                                         s->dc_val[0][ s->block_index[i] ]= dc[i];
-                                        memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
+                                        memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(int16_t)*16);
                                     }
                                 }
                             }
@@ -2803,7 +2925,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
                     }
                 }
 
-                s->current_picture.f.qscale_table[xy] = best_s.qscale;
+                s->current_picture.qscale_table[xy] = best_s.qscale;
 
                 copy_context_after_encode(s, &best_s, -1);
 
@@ -2830,9 +2952,9 @@ static int encode_thread(AVCodecContext *c, void *arg){
                     ff_h263_update_motion_val(s);
 
                 if(next_block==0){ //FIXME 16 vs linesize16
-                    s->dsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
-                    s->dsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
-                    s->dsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
+                    s->hdsp.put_pixels_tab[0][0](s->dest[0], s->rd_scratchpad                     , s->linesize  ,16);
+                    s->hdsp.put_pixels_tab[1][0](s->dest[1], s->rd_scratchpad + 16*s->linesize    , s->uvlinesize, 8);
+                    s->hdsp.put_pixels_tab[1][0](s->dest[2], s->rd_scratchpad + 16*s->linesize + 8, s->uvlinesize, 8);
                 }
 
                 if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
@@ -2870,8 +2992,8 @@ static int encode_thread(AVCodecContext *c, void *arg){
                     s->mv_type = MV_TYPE_8X8;
                     s->mb_intra= 0;
                     for(i=0; i<4; i++){
-                        s->mv[0][i][0] = s->current_picture.f.motion_val[0][s->block_index[i]][0];
-                        s->mv[0][i][1] = s->current_picture.f.motion_val[0][s->block_index[i]][1];
+                        s->mv[0][i][0] = s->current_picture.motion_val[0][s->block_index[i]][0];
+                        s->mv[0][i][1] = s->current_picture.motion_val[0][s->block_index[i]][1];
                     }
                     break;
                 case CANDIDATE_MB_TYPE_DIRECT:
@@ -2909,7 +3031,6 @@ static int encode_thread(AVCodecContext *c, void *arg){
                     s->mb_intra= 0;
                     motion_x= s->mv[0][0][0] = s->b_forw_mv_table[xy][0];
                     motion_y= s->mv[0][0][1] = s->b_forw_mv_table[xy][1];
-//                    printf(" %d %d ", motion_x, motion_y);
                     break;
                 case CANDIDATE_MB_TYPE_FORWARD_I:
                     s->mv_dir = MV_DIR_FORWARD;
@@ -2972,21 +3093,22 @@ static int encode_thread(AVCodecContext *c, void *arg){
                 if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
 
-                s->current_picture.f.error[0] += sse(
-                    s, s->new_picture.f.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
+                s->current_picture.f->error[0] += sse(
+                    s, s->new_picture.f->data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
                     s->dest[0], w, h, s->linesize);
-                s->current_picture.f.error[1] += sse(
-                    s, s->new_picture.f.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
+                s->current_picture.f->error[1] += sse(
+                    s, s->new_picture.f->data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
                     s->dest[1], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
-                s->current_picture.f.error[2] += sse(
-                    s, s->new_picture.f.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
+                s->current_picture.f->error[2] += sse(
+                    s, s->new_picture.f->data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*chr_h,
                     s->dest[2], w>>1, h>>s->chroma_y_shift, s->uvlinesize);
             }
             if(s->loop_filter){
                 if(CONFIG_H263_ENCODER && s->out_format == FMT_H263)
                     ff_h263_loop_filter(s);
             }
-//printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, put_bits_count(&s->pb));
+            av_dlog(s->avctx, "MB %d %d bits\n",
+                    s->mb_x + s->mb_y * s->mb_stride, put_bits_count(&s->pb));
         }
     }
 
@@ -3028,11 +3150,11 @@ static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src)
     MERGE(b_count);
     MERGE(skip_count);
     MERGE(misc_bits);
-    MERGE(error_count);
+    MERGE(er.error_count);
     MERGE(padding_bug_score);
-    MERGE(current_picture.f.error[0]);
-    MERGE(current_picture.f.error[1]);
-    MERGE(current_picture.f.error[2]);
+    MERGE(current_picture.f->error[0]);
+    MERGE(current_picture.f->error[1]);
+    MERGE(current_picture.f->error[2]);
 
     if(dst->avctx->noise_reduction){
         for(i=0; i<64; i++){
@@ -3049,13 +3171,13 @@ static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src)
 
 static int estimate_qp(MpegEncContext *s, int dry_run){
     if (s->next_lambda){
-        s->current_picture_ptr->f.quality =
-        s->current_picture.f.quality = s->next_lambda;
+        s->current_picture_ptr->f->quality =
+        s->current_picture.f->quality = s->next_lambda;
         if(!dry_run) s->next_lambda= 0;
     } else if (!s->fixed_qscale) {
-        s->current_picture_ptr->f.quality =
-        s->current_picture.f.quality = ff_rate_estimate_qscale(s, dry_run);
-        if (s->current_picture.f.quality < 0)
+        s->current_picture_ptr->f->quality =
+        s->current_picture.f->quality = ff_rate_estimate_qscale(s, dry_run);
+        if (s->current_picture.f->quality < 0)
             return -1;
     }
 
@@ -3078,16 +3200,15 @@ static int estimate_qp(MpegEncContext *s, int dry_run){
         s->lambda= s->lambda_table[0];
         //FIXME broken
     }else
-        s->lambda = s->current_picture.f.quality;
-//printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
+        s->lambda = s->current_picture.f->quality;
     update_qscale(s);
     return 0;
 }
 
 /* must be called before writing the header */
 static void set_frame_distances(MpegEncContext * s){
-    assert(s->current_picture_ptr->f.pts != AV_NOPTS_VALUE);
-    s->time = s->current_picture_ptr->f.pts * s->avctx->time_base.num;
+    assert(s->current_picture_ptr->f->pts != AV_NOPTS_VALUE);
+    s->time = s->current_picture_ptr->f->pts * s->avctx->time_base.num;
 
     if(s->pict_type==AV_PICTURE_TYPE_B){
         s->pb_time= s->pp_time - (s->last_non_b_time - s->time);
@@ -3101,7 +3222,7 @@ static void set_frame_distances(MpegEncContext * s){
 
 static int encode_picture(MpegEncContext *s, int picture_number)
 {
-    int i;
+    int i, ret;
     int bits;
     int context_count = s->slice_context_count;
 
@@ -3144,7 +3265,9 @@ static int encode_picture(MpegEncContext *s, int picture_number)
 
     s->mb_intra=0; //for the rate distortion & bit compare functions
     for(i=1; i<context_count; i++){
-        ff_update_duplicate_context(s->thread_context[i], s);
+        ret = ff_update_duplicate_context(s->thread_context[i], s);
+        if (ret < 0)
+            return ret;
     }
 
     if(ff_init_me(s)<0)
@@ -3154,7 +3277,7 @@ static int encode_picture(MpegEncContext *s, int picture_number)
     if(s->pict_type != AV_PICTURE_TYPE_I){
         s->lambda = (s->lambda * s->avctx->me_penalty_compensation + 128)>>8;
         s->lambda2= (s->lambda2* (int64_t)s->avctx->me_penalty_compensation + 128)>>8;
-        if(s->pict_type != AV_PICTURE_TYPE_B && s->avctx->me_threshold==0){
+        if (s->pict_type != AV_PICTURE_TYPE_B) {
             if((s->avctx->pre_me && s->last_non_b_pict_type==AV_PICTURE_TYPE_I) || s->avctx->pre_me==2){
                 s->avctx->execute(s->avctx, pre_estimate_motion_thread, &s->thread_context[0], NULL, context_count, sizeof(void*));
             }
@@ -3182,7 +3305,8 @@ static int encode_picture(MpegEncContext *s, int picture_number)
         s->pict_type= AV_PICTURE_TYPE_I;
         for(i=0; i<s->mb_stride*s->mb_height; i++)
             s->mb_type[i]= CANDIDATE_MB_TYPE_INTRA;
-//printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
+        av_dlog(s, "Scene change detected, encoding as I Frame %d %d\n",
+                s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
     }
 
     if(!s->umvplus){
@@ -3248,32 +3372,33 @@ static int encode_picture(MpegEncContext *s, int picture_number)
     if (s->out_format == FMT_MJPEG) {
         /* for mjpeg, we do include qscale in the matrix */
         for(i=1;i<64;i++){
-            int j= s->dsp.idct_permutation[i];
+            int j = s->idsp.idct_permutation[i];
 
             s->intra_matrix[j] = av_clip_uint8((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
         }
         s->y_dc_scale_table=
         s->c_dc_scale_table= ff_mpeg2_dc_scale_table[s->intra_dc_precision];
         s->intra_matrix[0] = ff_mpeg2_dc_scale_table[s->intra_dc_precision][8];
-        ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
+        ff_convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16,
                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
         s->qscale= 8;
     }
 
     //FIXME var duplication
-    s->current_picture_ptr->f.key_frame =
-    s->current_picture.f.key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
-    s->current_picture_ptr->f.pict_type =
-    s->current_picture.f.pict_type = s->pict_type;
+    s->current_picture_ptr->f->key_frame =
+    s->current_picture.f->key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
+    s->current_picture_ptr->f->pict_type =
+    s->current_picture.f->pict_type = s->pict_type;
 
-    if (s->current_picture.f.key_frame)
+    if (s->current_picture.f->key_frame)
         s->picture_in_gop_number=0;
 
     s->last_bits= put_bits_count(&s->pb);
     switch(s->out_format) {
     case FMT_MJPEG:
         if (CONFIG_MJPEG_ENCODER)
-            ff_mjpeg_encode_picture_header(s);
+            ff_mjpeg_encode_picture_header(s->avctx, &s->pb, &s->intra_scantable,
+                                           s->intra_matrix);
         break;
     case FMT_H261:
         if (CONFIG_H261_ENCODER)
@@ -3299,8 +3424,6 @@ static int encode_picture(MpegEncContext *s, int picture_number)
         if (CONFIG_MPEG1VIDEO_ENCODER || CONFIG_MPEG2VIDEO_ENCODER)
             ff_mpeg1_encode_picture_header(s, picture_number);
         break;
-    case FMT_H264:
-        break;
     default:
         assert(0);
     }
@@ -3318,7 +3441,7 @@ static int encode_picture(MpegEncContext *s, int picture_number)
     return 0;
 }
 
-static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
+static void denoise_dct_c(MpegEncContext *s, int16_t *block){
     const int intra= s->mb_intra;
     int i;
 
@@ -3343,7 +3466,7 @@ static void denoise_dct_c(MpegEncContext *s, DCTELEM *block){
 }
 
 static int dct_quantize_trellis_c(MpegEncContext *s,
-                                  DCTELEM *block, int n,
+                                  int16_t *block, int n,
                                   int qscale, int *overflow){
     const int *qmat;
     const uint8_t *scantable= s->intra_scantable.scantable;
@@ -3368,7 +3491,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
     uint8_t * last_length;
     const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
 
-    s->dsp.fdct (block);
+    s->fdsp.fdct(block);
 
     if(s->dct_error_sum)
         s->denoise_dct(s, block);
@@ -3450,7 +3573,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
     *overflow= s->max_qcoeff < max; //overflow might have happened
 
     if(last_non_zero < start_i){
-        memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
+        memset(block + start_i, 0, (64-start_i)*sizeof(int16_t));
         return last_non_zero;
     }
 
@@ -3463,7 +3586,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
         int dct_coeff= FFABS(block[ scantable[i] ]);
         int best_score=256*256*256*120;
 
-        if (s->dsp.fdct == ff_fdct_ifast)
+        if (s->fdsp.fdct == ff_fdct_ifast)
             dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12;
         zero_distortion= dct_coeff*dct_coeff;
 
@@ -3478,7 +3601,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
             if(s->out_format == FMT_H263){
                 unquant_coeff= alevel*qmul + qadd;
             }else{ //MPEG1
-                j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
+                j = s->idsp.idct_permutation[scantable[i]]; // FIXME: optimize
                 if(s->mb_intra){
                         unquant_coeff = (int)(  alevel  * qscale * s->intra_matrix[j]) >> 3;
                         unquant_coeff =   (unquant_coeff - 1) | 1;
@@ -3582,7 +3705,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
 
     dc= FFABS(block[0]);
     last_non_zero= last_i - 1;
-    memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
+    memset(block + start_i, 0, (64-start_i)*sizeof(int16_t));
 
     if(last_non_zero < start_i)
         return last_non_zero;
@@ -3657,10 +3780,10 @@ static void build_basis(uint8_t *perm){
 }
 
 static int dct_quantize_refine(MpegEncContext *s, //FIXME breaks denoise?
-                        DCTELEM *block, int16_t *weight, DCTELEM *orig,
+                        int16_t *block, int16_t *weight, int16_t *orig,
                         int n, int qscale){
     int16_t rem[64];
-    LOCAL_ALIGNED_16(DCTELEM, d1, [64]);
+    LOCAL_ALIGNED_16(int16_t, d1, [64]);
     const uint8_t *scantable= s->intra_scantable.scantable;
     const uint8_t *perm_scantable= s->intra_scantable.permutated;
 //    unsigned int threshold1, threshold2;
@@ -3684,7 +3807,7 @@ static int messed_sign=0;
 #endif
 
     if(basis[0][0] == 0)
-        build_basis(s->dsp.idct_permutation);
+        build_basis(s->idsp.idct_permutation);
 
     qmul= qscale*2;
     qadd= (qscale-1)|1;
@@ -3759,7 +3882,7 @@ STOP_TIMER("memset rem[]")}
             run_tab[rle_index++]=run;
             run=0;
 
-            s->dsp.add_8x8basis(rem, basis[j], coeff);
+            s->mpvencdsp.add_8x8basis(rem, basis[j], coeff);
         }else{
             run++;
         }
@@ -3773,7 +3896,7 @@ STOP_TIMER("init rem[]")
 {START_TIMER
 #endif
     for(;;){
-        int best_score=s->dsp.try_8x8basis(rem, weight, basis[0], 0);
+        int best_score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0], 0);
         int best_coeff=0;
         int best_change=0;
         int run2, best_unquant_change=0, analyze_gradient;
@@ -3795,7 +3918,7 @@ STOP_TIMER("init rem[]")
 STOP_TIMER("rem*w*w")}
 {START_TIMER
 #endif
-            s->dsp.fdct(d1);
+            s->fdsp.fdct(d1);
 #ifdef REFINE_STATS
 STOP_TIMER("dct")}
 #endif
@@ -3817,7 +3940,8 @@ STOP_TIMER("dct")}
                 if(new_coeff >= 2048 || new_coeff < 0)
                     continue;
 
-                score= s->dsp.try_8x8basis(rem, weight, basis[0], new_coeff - old_coeff);
+                score = s->mpvencdsp.try_8x8basis(rem, weight, basis[0],
+                                                  new_coeff - old_coeff);
                 if(score<best_score){
                     best_score= score;
                     best_coeff= 0;
@@ -3940,7 +4064,8 @@ STOP_TIMER("dct")}
                 unquant_change= new_coeff - old_coeff;
                 assert((score < 100*lambda && score > -100*lambda) || lambda==0);
 
-                score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
+                score += s->mpvencdsp.try_8x8basis(rem, weight, basis[j],
+                                                   unquant_change);
                 if(score<best_score){
                     best_score= score;
                     best_coeff= i;
@@ -4014,7 +4139,7 @@ if(256*256*256*64 % count == 0){
                  }
             }
 
-            s->dsp.add_8x8basis(rem, basis[j], best_unquant_change);
+            s->mpvencdsp.add_8x8basis(rem, basis[j], best_unquant_change);
         }else{
             break;
         }
@@ -4030,7 +4155,7 @@ STOP_TIMER("iterative search")
 }
 
 int ff_dct_quantize_c(MpegEncContext *s,
-                        DCTELEM *block, int n,
+                        int16_t *block, int n,
                         int qscale, int *overflow)
 {
     int i, j, level, last_non_zero, q, start_i;
@@ -4040,7 +4165,7 @@ int ff_dct_quantize_c(MpegEncContext *s,
     int max=0;
     unsigned int threshold1, threshold2;
 
-    s->dsp.fdct (block);
+    s->fdsp.fdct(block);
 
     if(s->dct_error_sum)
         s->denoise_dct(s, block);
@@ -4103,8 +4228,9 @@ int ff_dct_quantize_c(MpegEncContext *s,
     *overflow= s->max_qcoeff < max; //overflow might have happened
 
     /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */
-    if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)
-        ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero);
+    if (s->idsp.perm_type != FF_IDCT_PERM_NONE)
+        ff_block_permute(block, s->idsp.idct_permutation,
+                         scantable, last_non_zero);
 
     return last_non_zero;
 }
@@ -4112,9 +4238,9 @@ int ff_dct_quantize_c(MpegEncContext *s,
 #define OFFSET(x) offsetof(MpegEncContext, x)
 #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
 static const AVOption h263_options[] = {
-    { "obmc",         "use overlapped block motion compensation.", OFFSET(obmc), AV_OPT_TYPE_INT, { 0 }, 0, 1, VE },
-    { "structured_slices","Write slice start position at every GOB header instead of just GOB number.", OFFSET(h263_slice_structured), AV_OPT_TYPE_INT, { 0 }, 0, 1, VE},
-    { "mb_info",      "emit macroblock info for RFC 2190 packetization, the parameter value is the maximum payload size", OFFSET(mb_info), AV_OPT_TYPE_INT, { 0 }, 0, INT_MAX, VE },
+    { "obmc",         "use overlapped block motion compensation.", OFFSET(obmc), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
+    { "structured_slices","Write slice start position at every GOB header instead of just GOB number.", OFFSET(h263_slice_structured), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE},
+    { "mb_info",      "emit macroblock info for RFC 2190 packetization, the parameter value is the maximum payload size", OFFSET(mb_info), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, VE },
     FF_MPV_COMMON_OPTS
     { NULL },
 };
@@ -4128,22 +4254,22 @@ static const AVClass h263_class = {
 
 AVCodec ff_h263_encoder = {
     .name           = "h263",
+    .long_name      = NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_H263,
     .priv_data_size = sizeof(MpegEncContext),
     .init           = ff_MPV_encode_init,
     .encode2        = ff_MPV_encode_picture,
     .close          = ff_MPV_encode_end,
-    .pix_fmts= (const enum PixelFormat[]){PIX_FMT_YUV420P, PIX_FMT_NONE},
-    .long_name= NULL_IF_CONFIG_SMALL("H.263 / H.263-1996"),
+    .pix_fmts= (const enum AVPixelFormat[]){AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE},
     .priv_class     = &h263_class,
 };
 
 static const AVOption h263p_options[] = {
-    { "umv",        "Use unlimited motion vectors.",    OFFSET(umvplus), AV_OPT_TYPE_INT, { 0 }, 0, 1, VE },
-    { "aiv",        "Use alternative inter VLC.",       OFFSET(alt_inter_vlc), AV_OPT_TYPE_INT, { 0 }, 0, 1, VE },
-    { "obmc",       "use overlapped block motion compensation.", OFFSET(obmc), AV_OPT_TYPE_INT, { 0 }, 0, 1, VE },
-    { "structured_slices", "Write slice start position at every GOB header instead of just GOB number.", OFFSET(h263_slice_structured), AV_OPT_TYPE_INT, { 0 }, 0, 1, VE},
+    { "umv",        "Use unlimited motion vectors.",    OFFSET(umvplus), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
+    { "aiv",        "Use alternative inter VLC.",       OFFSET(alt_inter_vlc), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
+    { "obmc",       "use overlapped block motion compensation.", OFFSET(obmc), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
+    { "structured_slices", "Write slice start position at every GOB header instead of just GOB number.", OFFSET(h263_slice_structured), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE},
     FF_MPV_COMMON_OPTS
     { NULL },
 };
@@ -4156,6 +4282,7 @@ static const AVClass h263p_class = {
 
 AVCodec ff_h263p_encoder = {
     .name           = "h263p",
+    .long_name      = NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_H263P,
     .priv_data_size = sizeof(MpegEncContext),
@@ -4163,8 +4290,7 @@ AVCodec ff_h263p_encoder = {
     .encode2        = ff_MPV_encode_picture,
     .close          = ff_MPV_encode_end,
     .capabilities   = CODEC_CAP_SLICE_THREADS,
-    .pix_fmts       = (const enum PixelFormat[]){ PIX_FMT_YUV420P, PIX_FMT_NONE },
-    .long_name      = NULL_IF_CONFIG_SMALL("H.263+ / H.263-1998 / H.263 version 2"),
+    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
     .priv_class     = &h263p_class,
 };
 
@@ -4172,14 +4298,14 @@ FF_MPV_GENERIC_CLASS(msmpeg4v2)
 
 AVCodec ff_msmpeg4v2_encoder = {
     .name           = "msmpeg4v2",
+    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_MSMPEG4V2,
     .priv_data_size = sizeof(MpegEncContext),
     .init           = ff_MPV_encode_init,
     .encode2        = ff_MPV_encode_picture,
     .close          = ff_MPV_encode_end,
-    .pix_fmts       = (const enum PixelFormat[]){ PIX_FMT_YUV420P, PIX_FMT_NONE },
-    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 2"),
+    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
     .priv_class     = &msmpeg4v2_class,
 };
 
@@ -4187,14 +4313,14 @@ FF_MPV_GENERIC_CLASS(msmpeg4v3)
 
 AVCodec ff_msmpeg4v3_encoder = {
     .name           = "msmpeg4",
+    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_MSMPEG4V3,
     .priv_data_size = sizeof(MpegEncContext),
     .init           = ff_MPV_encode_init,
     .encode2        = ff_MPV_encode_picture,
     .close          = ff_MPV_encode_end,
-    .pix_fmts       = (const enum PixelFormat[]){ PIX_FMT_YUV420P, PIX_FMT_NONE },
-    .long_name      = NULL_IF_CONFIG_SMALL("MPEG-4 part 2 Microsoft variant version 3"),
+    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
     .priv_class     = &msmpeg4v3_class,
 };
 
@@ -4202,13 +4328,13 @@ FF_MPV_GENERIC_CLASS(wmv1)
 
 AVCodec ff_wmv1_encoder = {
     .name           = "wmv1",
+    .long_name      = NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
     .type           = AVMEDIA_TYPE_VIDEO,
     .id             = AV_CODEC_ID_WMV1,
     .priv_data_size = sizeof(MpegEncContext),
     .init           = ff_MPV_encode_init,
     .encode2        = ff_MPV_encode_picture,
     .close          = ff_MPV_encode_end,
-    .pix_fmts       = (const enum PixelFormat[]){ PIX_FMT_YUV420P, PIX_FMT_NONE },
-    .long_name      = NULL_IF_CONFIG_SMALL("Windows Media Video 7"),
+    .pix_fmts       = (const enum AVPixelFormat[]){ AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE },
     .priv_class     = &wmv1_class,
 };