Merge commit '3360ad995530ea6967b1e83981b4aa8240fbb0ed'
[ffmpeg.git] / libavcodec / mpegvideo_enc.c
index 4fbe10e..dcbdc03 100644 (file)
@@ -5,20 +5,20 @@
  *
  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
  *
- * This file is part of Libav.
+ * This file is part of FFmpeg.
  *
- * Libav is free software; you can redistribute it and/or
+ * FFmpeg is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
- * Libav is distributed in the hope that it will be useful,
+ * FFmpeg is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
+ * License along with FFmpeg; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
@@ -52,6 +52,7 @@
 #include "internal.h"
 #include "bytestream.h"
 #include <limits.h>
+#include "sp5x.h"
 
 static int encode_picture(MpegEncContext *s, int picture_number);
 static int dct_quantize_refine(MpegEncContext *s, int16_t *block, int16_t *weight, int16_t *orig, int n, int qscale);
@@ -101,8 +102,7 @@ void ff_convert_matrix(DSPContext *dsp, int (*qmat)[64],
                  *           3444240 >= (1 << 36) / (x) >= 275 */
 
                 qmat[qscale][i] = (int)((UINT64_C(1) << (QMAT_SHIFT + 14)) /
-                                        (ff_aanscales[i] * qscale *
-                                         quant_matrix[j]));
+                                        (ff_aanscales[i] * (int64_t)qscale * quant_matrix[j]));
             }
         } else {
             for (i = 0; i < 64; i++) {
@@ -222,6 +222,22 @@ static void MPV_encode_defaults(MpegEncContext *s)
     s->picture_in_gop_number = 0;
 }
 
+av_cold int ff_dct_encode_init(MpegEncContext *s) {
+    if (ARCH_X86)
+        ff_dct_encode_init_x86(s);
+
+    ff_h263dsp_init(&s->h263dsp);
+    if (!s->dct_quantize)
+        s->dct_quantize = ff_dct_quantize_c;
+    if (!s->denoise_dct)
+        s->denoise_dct  = denoise_dct_c;
+    s->fast_dct_quantize = s->dct_quantize;
+    if (s->avctx->trellis)
+        s->dct_quantize  = dct_quantize_trellis_c;
+
+    return 0;
+}
+
 /* init video encoder */
 av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
 {
@@ -244,7 +260,9 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         if (avctx->pix_fmt != AV_PIX_FMT_YUVJ420P &&
             avctx->pix_fmt != AV_PIX_FMT_YUVJ422P &&
             avctx->pix_fmt != AV_PIX_FMT_YUVJ444P &&
+            avctx->pix_fmt != AV_PIX_FMT_BGR0     &&
             avctx->pix_fmt != AV_PIX_FMT_BGRA     &&
+            avctx->pix_fmt != AV_PIX_FMT_BGR24    &&
             ((avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
               avctx->pix_fmt != AV_PIX_FMT_YUV422P &&
               avctx->pix_fmt != AV_PIX_FMT_YUV444P) ||
@@ -254,10 +272,13 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         }
         break;
     case AV_CODEC_ID_MJPEG:
+    case AV_CODEC_ID_AMV:
         if (avctx->pix_fmt != AV_PIX_FMT_YUVJ420P &&
             avctx->pix_fmt != AV_PIX_FMT_YUVJ422P &&
+            avctx->pix_fmt != AV_PIX_FMT_YUVJ444P &&
             ((avctx->pix_fmt != AV_PIX_FMT_YUV420P &&
-              avctx->pix_fmt != AV_PIX_FMT_YUV422P) ||
+              avctx->pix_fmt != AV_PIX_FMT_YUV422P &&
+              avctx->pix_fmt != AV_PIX_FMT_YUV444P) ||
              avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL)) {
             av_log(avctx, AV_LOG_ERROR, "colorspace not supported in jpeg\n");
             return -1;
@@ -271,6 +292,10 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     }
 
     switch (avctx->pix_fmt) {
+    case AV_PIX_FMT_YUVJ444P:
+    case AV_PIX_FMT_YUV444P:
+        s->chroma_format = CHROMA_444;
+        break;
     case AV_PIX_FMT_YUVJ422P:
     case AV_PIX_FMT_YUV422P:
         s->chroma_format = CHROMA_422;
@@ -287,8 +312,9 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     s->height   = avctx->height;
     if (avctx->gop_size > 600 &&
         avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) {
-        av_log(avctx, AV_LOG_ERROR,
-               "Warning keyframe interval too large! reducing it ...\n");
+        av_log(avctx, AV_LOG_WARNING,
+               "keyframe interval too large!, reducing it from %d to %d\n",
+               avctx->gop_size, 600);
         avctx->gop_size = 600;
     }
     s->gop_size     = avctx->gop_size;
@@ -298,6 +324,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     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);
+        avctx->max_b_frames = MAX_B_FRAMES;
     }
     s->max_b_frames = avctx->max_b_frames;
     s->codec_id     = avctx->codec->id;
@@ -332,10 +359,35 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     s->loop_filter      = !!(s->flags & CODEC_FLAG_LOOP_FILTER);
 
     if (avctx->rc_max_rate && !avctx->rc_buffer_size) {
-        av_log(avctx, AV_LOG_ERROR,
-               "a vbv buffer size is needed, "
-               "for encoding with a maximum bitrate\n");
-        return -1;
+        switch(avctx->codec_id) {
+        case AV_CODEC_ID_MPEG1VIDEO:
+        case AV_CODEC_ID_MPEG2VIDEO:
+            avctx->rc_buffer_size = FFMAX(avctx->rc_max_rate, 15000000) * 112L / 15000000 * 16384;
+            break;
+        case AV_CODEC_ID_MPEG4:
+        case AV_CODEC_ID_MSMPEG4V1:
+        case AV_CODEC_ID_MSMPEG4V2:
+        case AV_CODEC_ID_MSMPEG4V3:
+            if       (avctx->rc_max_rate >= 15000000) {
+                avctx->rc_buffer_size = 320 + (avctx->rc_max_rate - 15000000L) * (760-320) / (38400000 - 15000000);
+            } else if(avctx->rc_max_rate >=  2000000) {
+                avctx->rc_buffer_size =  80 + (avctx->rc_max_rate -  2000000L) * (320- 80) / (15000000 -  2000000);
+            } else if(avctx->rc_max_rate >=   384000) {
+                avctx->rc_buffer_size =  40 + (avctx->rc_max_rate -   384000L) * ( 80- 40) / ( 2000000 -   384000);
+            } else
+                avctx->rc_buffer_size = 40;
+            avctx->rc_buffer_size *= 16384;
+            break;
+        }
+        if (avctx->rc_buffer_size) {
+            av_log(avctx, AV_LOG_INFO, "Automatically choosing VBV buffer size of %d kbyte\n", avctx->rc_buffer_size/8192);
+        }
+    }
+
+    if ((!avctx->rc_max_rate) != (!avctx->rc_buffer_size)) {
+        av_log(avctx, AV_LOG_ERROR, "Either both buffer size and max rate or neither must be specified\n");
+        if (avctx->rc_max_rate && !avctx->rc_buffer_size)
+            return -1;
     }
 
     if (avctx->rc_min_rate && avctx->rc_max_rate != avctx->rc_min_rate) {
@@ -349,7 +401,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     }
 
     if (avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate) {
-        av_log(avctx, AV_LOG_INFO, "bitrate above max bitrate\n");
+        av_log(avctx, AV_LOG_ERROR, "bitrate above max bitrate\n");
         return -1;
     }
 
@@ -411,18 +463,74 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
         return -1;
     }
+    if (s->max_b_frames < 0) {
+        av_log(avctx, AV_LOG_ERROR,
+               "max b frames must be 0 or positive for mpegvideo based encoders\n");
+        return -1;
+    }
 
     if ((s->codec_id == AV_CODEC_ID_MPEG4 ||
          s->codec_id == AV_CODEC_ID_H263  ||
          s->codec_id == AV_CODEC_ID_H263P) &&
         (avctx->sample_aspect_ratio.num > 255 ||
          avctx->sample_aspect_ratio.den > 255)) {
-        av_log(avctx, AV_LOG_ERROR,
-               "Invalid pixel aspect ratio %i/%i, limit is 255/255\n",
+        av_log(avctx, AV_LOG_WARNING,
+               "Invalid pixel aspect ratio %i/%i, limit is 255/255 reducing\n",
                avctx->sample_aspect_ratio.num, avctx->sample_aspect_ratio.den);
+        av_reduce(&avctx->sample_aspect_ratio.num, &avctx->sample_aspect_ratio.den,
+                   avctx->sample_aspect_ratio.num,  avctx->sample_aspect_ratio.den, 255);
+    }
+
+    if ((s->codec_id == AV_CODEC_ID_H263  ||
+         s->codec_id == AV_CODEC_ID_H263P) &&
+        (avctx->width  > 2048 ||
+         avctx->height > 1152 )) {
+        av_log(avctx, AV_LOG_ERROR, "H.263 does not support resolutions above 2048x1152\n");
+        return -1;
+    }
+    if ((s->codec_id == AV_CODEC_ID_H263  ||
+         s->codec_id == AV_CODEC_ID_H263P) &&
+        ((avctx->width &3) ||
+         (avctx->height&3) )) {
+        av_log(avctx, AV_LOG_ERROR, "w/h must be a multiple of 4\n");
+        return -1;
+    }
+
+    if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO &&
+        (avctx->width  > 4095 ||
+         avctx->height > 4095 )) {
+        av_log(avctx, AV_LOG_ERROR, "MPEG-1 does not support resolutions above 4095x4095\n");
+        return -1;
+    }
+
+    if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO &&
+        (avctx->width  > 16383 ||
+         avctx->height > 16383 )) {
+        av_log(avctx, AV_LOG_ERROR, "MPEG-2 does not support resolutions above 16383x16383\n");
         return -1;
     }
 
+    if (s->codec_id == AV_CODEC_ID_RV10 &&
+        (avctx->width &15 ||
+         avctx->height&15 )) {
+        av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 16\n");
+        return AVERROR(EINVAL);
+    }
+
+    if (s->codec_id == AV_CODEC_ID_RV20 &&
+        (avctx->width &3 ||
+         avctx->height&3 )) {
+        av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 4\n");
+        return AVERROR(EINVAL);
+    }
+
+    if ((s->codec_id == AV_CODEC_ID_WMV1 ||
+         s->codec_id == AV_CODEC_ID_WMV2) &&
+         avctx->width & 1) {
+         av_log(avctx, AV_LOG_ERROR, "width must be multiple of 2\n");
+         return -1;
+    }
+
     if ((s->flags & (CODEC_FLAG_INTERLACED_DCT | CODEC_FLAG_INTERLACED_ME)) &&
         s->codec_id != AV_CODEC_ID_MPEG4 && s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
         av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
@@ -430,7 +538,8 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     }
 
     // FIXME mpeg2 uses that too
-    if (s->mpeg_quant && s->codec_id != AV_CODEC_ID_MPEG4) {
+    if (s->mpeg_quant && (   s->codec_id != AV_CODEC_ID_MPEG4
+                          && s->codec_id != AV_CODEC_ID_MPEG2VIDEO)) {
         av_log(avctx, AV_LOG_ERROR,
                "mpeg2 style quantization not supported by codec\n");
         return -1;
@@ -480,6 +589,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         s->codec_id != AV_CODEC_ID_MPEG4      &&
         s->codec_id != AV_CODEC_ID_MPEG1VIDEO &&
         s->codec_id != AV_CODEC_ID_MPEG2VIDEO &&
+        s->codec_id != AV_CODEC_ID_MJPEG      &&
         (s->codec_id != AV_CODEC_ID_H263P)) {
         av_log(avctx, AV_LOG_ERROR,
                "multi threaded encoding not supported by codec\n");
@@ -488,14 +598,17 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
 
     if (s->avctx->thread_count < 1) {
         av_log(avctx, AV_LOG_ERROR,
-               "automatic thread number detection not supported by codec,"
+               "automatic thread number detection not supported by codec, "
                "patch welcome\n");
         return -1;
     }
 
-    if (s->avctx->thread_count > 1)
+    if (s->avctx->slices > 1 || s->avctx->thread_count > 1)
         s->rtp_mode = 1;
 
+    if (s->avctx->thread_count > 1 && s->codec_id == AV_CODEC_ID_H263P)
+        s->h263_slice_structured = 1;
+
     if (!avctx->time_base.den || !avctx->time_base.num) {
         av_log(avctx, AV_LOG_ERROR, "framerate not set\n");
         return -1;
@@ -522,8 +635,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         //return -1;
     }
 
-    if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
-        s->codec_id == AV_CODEC_ID_MPEG2VIDEO || s->codec_id == AV_CODEC_ID_MJPEG) {
+    if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG1VIDEO || s->codec_id == AV_CODEC_ID_MPEG2VIDEO || s->codec_id == AV_CODEC_ID_MJPEG || s->codec_id==AV_CODEC_ID_AMV) {
         // (a + x * 3 / 8) / x
         s->intra_quant_bias = 3 << (QUANT_BIAS_SHIFT - 3);
         s->inter_quant_bias = 0;
@@ -533,13 +645,19 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         s->inter_quant_bias = -(1 << (QUANT_BIAS_SHIFT - 2));
     }
 
+    if (avctx->qmin > avctx->qmax || avctx->qmin <= 0) {
+        av_log(avctx, AV_LOG_ERROR, "qmin and or qmax are invalid, they must be 0 < min <= max\n");
+        return AVERROR(EINVAL);
+    }
+
     if (avctx->intra_quant_bias != FF_DEFAULT_QUANT_BIAS)
         s->intra_quant_bias = avctx->intra_quant_bias;
     if (avctx->inter_quant_bias != FF_DEFAULT_QUANT_BIAS)
         s->inter_quant_bias = avctx->inter_quant_bias;
 
-    av_pix_fmt_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift,
-                                     &chroma_v_shift);
+    av_log(avctx, AV_LOG_DEBUG, "intra_quant_bias = %d inter_quant_bias = %d\n",s->intra_quant_bias,s->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) {
@@ -566,13 +684,19 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         break;
     case AV_CODEC_ID_LJPEG:
     case AV_CODEC_ID_MJPEG:
+    case AV_CODEC_ID_AMV:
         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   == AV_PIX_FMT_BGRA) {
+            (avctx->pix_fmt == AV_PIX_FMT_BGR0
+             || s->avctx->pix_fmt == AV_PIX_FMT_BGRA
+             || s->avctx->pix_fmt == AV_PIX_FMT_BGR24)) {
             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 if (avctx->pix_fmt == AV_PIX_FMT_YUV444P || avctx->pix_fmt == AV_PIX_FMT_YUVJ444P) {
+            s->mjpeg_vsample[0] = s->mjpeg_vsample[1] = s->mjpeg_vsample[2] = 2;
+            s->mjpeg_hsample[0] = s->mjpeg_hsample[1] = s->mjpeg_hsample[2] = 1;
         } else {
             s->mjpeg_vsample[0] = 2;
             s->mjpeg_vsample[1] = 2 >> chroma_v_shift;
@@ -603,13 +727,13 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
         break;
     case AV_CODEC_ID_H263:
         if (!CONFIG_H263_ENCODER)
-        return -1;
+            return -1;
         if (ff_match_2uint16(ff_h263_format, FF_ARRAY_ELEMS(ff_h263_format),
                              s->width, s->height) == 8) {
-            av_log(avctx, AV_LOG_INFO,
+            av_log(avctx, AV_LOG_ERROR,
                    "The specified picture size of %dx%d is not valid for "
                    "the H.263 codec.\nValid sizes are 128x96, 176x144, "
-                   "352x288, 704x576, and 1408x1152."
+                   "352x288, 704x576, and 1408x1152. "
                    "Try H.263+.\n", s->width, s->height);
             return -1;
         }
@@ -713,9 +837,6 @@ 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);
-
     s->avctx->coded_frame = &s->current_picture.f;
 
     if (s->msmpeg4_version) {
@@ -726,8 +847,10 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     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_chroma_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_chroma_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);
@@ -739,14 +862,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
                           2 * 64 * sizeof(uint16_t), fail);
     }
 
-    ff_h263dsp_init(&s->h263dsp);
-    if (!s->dct_quantize)
-        s->dct_quantize = ff_dct_quantize_c;
-    if (!s->denoise_dct)
-        s->denoise_dct  = denoise_dct_c;
-    s->fast_dct_quantize = s->dct_quantize;
-    if (avctx->trellis)
-        s->dct_quantize  = dct_quantize_trellis_c;
+    ff_dct_encode_init(s);
 
     if ((CONFIG_H263P_ENCODER || CONFIG_RV20_ENCODER) && s->modified_quant)
         s->chroma_qscale_table = ff_h263_chroma_qscale_table;
@@ -853,6 +969,10 @@ av_cold int ff_MPV_encode_end(AVCodecContext *avctx)
     av_freep(&s->avctx->stats_out);
     av_freep(&s->ac_stats);
 
+    if(s->q_chroma_intra_matrix   != s->q_intra_matrix  ) av_freep(&s->q_chroma_intra_matrix);
+    if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
+    s->q_chroma_intra_matrix=   NULL;
+    s->q_chroma_intra_matrix16= NULL;
     av_freep(&s->q_intra_matrix);
     av_freep(&s->q_inter_matrix);
     av_freep(&s->q_intra_matrix16);
@@ -917,18 +1037,17 @@ static int load_input_picture(MpegEncContext *s, const AVFrame *pic_arg)
 
         if (pts != AV_NOPTS_VALUE) {
             if (s->user_specified_pts != AV_NOPTS_VALUE) {
-                int64_t time = pts;
                 int64_t last = s->user_specified_pts;
 
-                if (time <= last) {
+                if (pts <= last) {
                     av_log(s->avctx, AV_LOG_ERROR,
-                           "Error, Invalid timestamp=%"PRId64", "
-                           "last=%"PRId64"\n", pts, s->user_specified_pts);
-                    return -1;
+                           "Invalid pts (%"PRId64") <= last (%"PRId64")\n",
+                           pts, last);
+                    return AVERROR(EINVAL);
                 }
 
                 if (!s->low_delay && display_picture_number == 1)
-                    s->dts_delta = time - last;
+                    s->dts_delta = pts - last;
             }
             s->user_specified_pts = pts;
         } else {
@@ -945,7 +1064,7 @@ static int load_input_picture(MpegEncContext *s, const AVFrame *pic_arg)
     }
 
     if (pic_arg) {
-        if (!pic_arg->buf[0]);
+        if (!pic_arg->buf[0])
             direct = 0;
         if (pic_arg->linesize[0] != s->linesize)
             direct = 0;
@@ -1002,18 +1121,31 @@ static int load_input_picture(MpegEncContext *s, const AVFrame *pic_arg)
                     uint8_t *src = pic_arg->data[i];
                     uint8_t *dst = pic->f.data[i];
 
+                    if (s->codec_id == AV_CODEC_ID_AMV && !(s->avctx->flags & CODEC_FLAG_EMU_EDGE)) {
+                        h = ((s->height + 15)/16*16) >> v_shift;
+                    }
+
                     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;
+                        int h2 = h;
+                        uint8_t *dst2 = dst;
+                        while (h2--) {
+                            memcpy(dst2, src, w);
+                            dst2 += dst_stride;
                             src += src_stride;
                         }
                     }
+                    if ((s->width & 15) || (s->height & 15)) {
+                        s->dsp.draw_edges(dst, dst_stride,
+                                          w, h,
+                                          16>>h_shift,
+                                          16>>v_shift,
+                                          EDGE_BOTTOM);
+                    }
                 }
             }
         }
@@ -1095,7 +1227,7 @@ static int estimate_best_b_count(MpegEncContext *s)
     int64_t best_rd  = INT64_MAX;
     int best_b_count = -1;
 
-    assert(scale >= 0 && scale <= 3);
+    av_assert0(scale >= 0 && scale <= 3);
 
     //emms_c();
     //s->next_picture_ptr->quality;
@@ -1383,15 +1515,15 @@ static void frame_end(MpegEncContext *s)
         const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
         int hshift = desc->log2_chroma_w;
         int vshift = desc->log2_chroma_h;
-        s->dsp.draw_edges(s->current_picture.f.data[0], s->linesize,
+        s->dsp.draw_edges(s->current_picture.f.data[0], s->current_picture.f.linesize[0],
                           s->h_edge_pos, s->v_edge_pos,
                           EDGE_WIDTH, EDGE_WIDTH,
                           EDGE_TOP | EDGE_BOTTOM);
-        s->dsp.draw_edges(s->current_picture.f.data[1], s->uvlinesize,
+        s->dsp.draw_edges(s->current_picture.f.data[1], s->current_picture.f.linesize[1],
                           s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
                           EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
                           EDGE_TOP | EDGE_BOTTOM);
-        s->dsp.draw_edges(s->current_picture.f.data[2], s->uvlinesize,
+        s->dsp.draw_edges(s->current_picture.f.data[2], s->current_picture.f.linesize[2],
                           s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
                           EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
                           EDGE_TOP | EDGE_BOTTOM);
@@ -1404,20 +1536,12 @@ static void frame_end(MpegEncContext *s)
     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;
 
 }
 
 int ff_MPV_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
-                          const AVFrame *pic_arg, int *got_packet)
+                          AVFrame *pic_arg, int *got_packet)
 {
     MpegEncContext *s = avctx->priv_data;
     int i, stuffing_count, ret;
@@ -1434,8 +1558,7 @@ int ff_MPV_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
 
     /* output? */
     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)
+        if ((ret = ff_alloc_packet2(avctx, pkt, s->mb_width*s->mb_height*(MAX_MB_BYTES+100)+10000)) < 0)
             return ret;
         if (s->mb_info) {
             s->mb_info_ptr = av_packet_new_side_data(pkt,
@@ -1456,7 +1579,8 @@ int ff_MPV_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
 
         s->pict_type = s->new_picture.f.pict_type;
         //emms_c();
-        ff_MPV_frame_start(s, avctx);
+        if (ff_MPV_frame_start(s, avctx) < 0)
+            return -1;
 vbv_retry:
         if (encode_picture(s, s->picture_number) < 0)
             return -1;
@@ -1530,6 +1654,7 @@ vbv_retry:
         s->frame_bits  = put_bits_count(&s->pb);
 
         stuffing_count = ff_vbv_update(s, s->frame_bits);
+        s->stuffing_bits = 8*stuffing_count;
         if (stuffing_count) {
             if (s->pb.buf_end - s->pb.buf - (put_bits_count(&s->pb) >> 3) <
                     stuffing_count + 50) {
@@ -1584,7 +1709,7 @@ vbv_retry:
 
             vbv_delay = FFMAX(vbv_delay, min_delay);
 
-            assert(vbv_delay < 0xFFFF);
+            av_assert0(vbv_delay < 0xFFFF);
 
             s->vbv_delay_ptr[0] &= 0xF8;
             s->vbv_delay_ptr[0] |= vbv_delay >> 13;
@@ -1597,12 +1722,12 @@ vbv_retry:
         avctx->frame_bits  = s->frame_bits;
 
         pkt->pts = s->current_picture.f.pts;
-        if (!s->low_delay) {
+        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)
@@ -1612,6 +1737,13 @@ vbv_retry:
     } else {
         s->frame_bits = 0;
     }
+
+    /* 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]);
+    }
+
     assert((s->frame_bits & 7) == 0);
 
     pkt->size = s->frame_bits / 8;
@@ -1736,15 +1868,17 @@ static void get_visual_weight(int16_t *weight, uint8_t *ptr, int stride)
 static av_always_inline void encode_mb_internal(MpegEncContext *s,
                                                 int motion_x, int motion_y,
                                                 int mb_block_height,
+                                                int mb_block_width,
                                                 int mb_block_count)
 {
-    int16_t weight[8][64];
-    int16_t orig[8][64];
+    int16_t weight[12][64];
+    int16_t orig[12][64];
     const int mb_x = s->mb_x;
     const int mb_y = s->mb_y;
     int i;
-    int skip_dct[8];
+    int skip_dct[12];
     int dct_offset = s->linesize * 8; // default for progressive frames
+    int uv_dct_offset = s->uvlinesize * 8;
     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
     ptrdiff_t wrap_y, wrap_c;
 
@@ -1786,12 +1920,14 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
     ptr_y  = s->new_picture.f.data[0] +
              (mb_y * 16 * wrap_y)              + mb_x * 16;
     ptr_cb = s->new_picture.f.data[1] +
-             (mb_y * mb_block_height * wrap_c) + mb_x * 8;
+             (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
     ptr_cr = s->new_picture.f.data[2] +
-             (mb_y * mb_block_height * wrap_c) + mb_x * 8;
+             (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
 
-    if (mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) {
+    if((mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) && s->codec_id != AV_CODEC_ID_AMV){
         uint8_t *ebuf = s->edge_emu_buffer + 32;
+        int cw = (s->width  + s->chroma_x_shift) >> s->chroma_x_shift;
+        int ch = (s->height + s->chroma_y_shift) >> s->chroma_y_shift;
         s->vdsp.emulated_edge_mc(ebuf, ptr_y,
                                  wrap_y, wrap_y,
                                  16, 16, mb_x * 16, mb_y * 16,
@@ -1799,14 +1935,16 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
         ptr_y = ebuf;
         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);
+                                 mb_block_width, mb_block_height,
+                                 mb_x * mb_block_width, mb_y * mb_block_height,
+                                 cw, ch);
         ptr_cb = ebuf + 18 * wrap_y;
-        s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y + 8, ptr_cr,
+        s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y + 16, 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;
+                                 mb_block_width, mb_block_height,
+                                 mb_x * mb_block_width, mb_y * mb_block_height,
+                                 cw, ch);
+        ptr_cr = ebuf + 18 * wrap_y + 16;
     }
 
     if (s->mb_intra) {
@@ -1828,8 +1966,10 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
                     s->interlaced_dct = 1;
 
                     dct_offset = wrap_y;
+                    uv_dct_offset = wrap_c;
                     wrap_y <<= 1;
-                    if (s->chroma_format == CHROMA_422)
+                    if (s->chroma_format == CHROMA_422 ||
+                        s->chroma_format == CHROMA_444)
                         wrap_c <<= 1;
                 }
             }
@@ -1846,11 +1986,16 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
         } else {
             s->dsp.get_pixels(s->block[4], ptr_cb, wrap_c);
             s->dsp.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);
+            if (!s->chroma_y_shift && s->chroma_x_shift) { /* 422 */
+                s->dsp.get_pixels(s->block[6], ptr_cb + uv_dct_offset, wrap_c);
+                s->dsp.get_pixels(s->block[7], ptr_cr + uv_dct_offset, wrap_c);
+            } else if (!s->chroma_y_shift && !s->chroma_x_shift) { /* 444 */
+                s->dsp.get_pixels(s->block[6], ptr_cb + 8, wrap_c);
+                s->dsp.get_pixels(s->block[7], ptr_cr + 8, wrap_c);
+                s->dsp.get_pixels(s->block[8], ptr_cb + uv_dct_offset, wrap_c);
+                s->dsp.get_pixels(s->block[9], ptr_cr + uv_dct_offset, wrap_c);
+                s->dsp.get_pixels(s->block[10], ptr_cb + uv_dct_offset + 8, wrap_c);
+                s->dsp.get_pixels(s->block[11], ptr_cr + uv_dct_offset + 8, wrap_c);
             }
         }
     } else {
@@ -1909,6 +2054,7 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
                     s->interlaced_dct = 1;
 
                     dct_offset = wrap_y;
+                    uv_dct_offset = wrap_c;
                     wrap_y <<= 1;
                     if (s->chroma_format == CHROMA_422)
                         wrap_c <<= 1;
@@ -1930,10 +2076,10 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
             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);
             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->dsp.diff_pixels(s->block[6], ptr_cb + uv_dct_offset,
+                                   dest_cb + uv_dct_offset, wrap_c);
+                s->dsp.diff_pixels(s->block[7], ptr_cr + uv_dct_offset,
+                                   dest_cr + uv_dct_offset, wrap_c);
             }
         }
         /* pre quantization */
@@ -1960,12 +2106,12 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
                               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),
+                if (s->dsp.sad[1](NULL, ptr_cb + uv_dct_offset,
+                                  dest_cb + uv_dct_offset,
                                   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),
+                if (s->dsp.sad[1](NULL, ptr_cr + uv_dct_offset,
+                                  dest_cr + uv_dct_offset,
                                   wrap_c, 8) < 20 * s->qscale)
                     skip_dct[7] = 1;
             }
@@ -1987,17 +2133,17 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
             get_visual_weight(weight[5], ptr_cr                , wrap_c);
         if (!s->chroma_y_shift) { /* 422 */
             if (!skip_dct[6])
-                get_visual_weight(weight[6], ptr_cb + (dct_offset >> 1),
+                get_visual_weight(weight[6], ptr_cb + uv_dct_offset,
                                   wrap_c);
             if (!skip_dct[7])
-                get_visual_weight(weight[7], ptr_cr + (dct_offset >> 1),
+                get_visual_weight(weight[7], ptr_cr + uv_dct_offset,
                                   wrap_c);
         }
         memcpy(orig[0], s->block[0], sizeof(int16_t) * 64 * mb_block_count);
     }
 
     /* DCT & quantize */
-    assert(s->out_format != FMT_MJPEG || s->qscale == 8);
+    av_assert2(s->out_format != FMT_MJPEG || s->qscale == 8);
     {
         for (i = 0; i < mb_block_count; i++) {
             if (!skip_dct[i]) {
@@ -2043,6 +2189,12 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
         s->block_last_index[5] = 0;
         s->block[4][0] =
         s->block[5][0] = (1024 + s->c_dc_scale / 2) / s->c_dc_scale;
+        if (!s->chroma_y_shift) { /* 422 / 444 */
+            for (i=6; i<12; i++) {
+                s->block_last_index[i] = 0;
+                s->block[i][0] = s->block[4][0];
+            }
+        }
     }
 
     // non c quantize code returns incorrect block_last_index FIXME
@@ -2093,18 +2245,20 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s,
             ff_h263_encode_mb(s, s->block, motion_x, motion_y);
         break;
     case AV_CODEC_ID_MJPEG:
+    case AV_CODEC_ID_AMV:
         if (CONFIG_MJPEG_ENCODER)
             ff_mjpeg_encode_mb(s, s->block);
         break;
     default:
-        assert(0);
+        av_assert1(0);
     }
 }
 
 static av_always_inline void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
 {
-    if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 6);
-    else                                encode_mb_internal(s, motion_x, motion_y, 16, 8);
+    if (s->chroma_format == CHROMA_420) encode_mb_internal(s, motion_x, motion_y,  8, 8, 6);
+    else if (s->chroma_format == CHROMA_422) encode_mb_internal(s, motion_x, motion_y, 16, 8, 8);
+    else encode_mb_internal(s, motion_x, motion_y, 16, 16, 12);
 }
 
 static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext *s, int type){
@@ -2241,7 +2395,7 @@ static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, in
         }
     }
 
-    assert(acc>=0);
+    av_assert2(acc>=0);
 
     return acc;
 }
@@ -2347,7 +2501,7 @@ static void write_slice_end(MpegEncContext *s){
 
         ff_mpeg4_stuffing(&s->pb);
     }else if(CONFIG_MJPEG_ENCODER && s->out_format == FMT_MJPEG){
-        ff_mjpeg_encode_stuffing(&s->pb);
+        ff_mjpeg_encode_stuffing(s);
     }
 
     avpriv_align_put_bits(&s->pb);
@@ -2436,6 +2590,11 @@ static int encode_thread(AVCodecContext *c, void *arg){
 
         s->current_picture.f.error[i] = 0;
     }
+    if(s->codec_id==AV_CODEC_ID_AMV){
+        s->last_dc[0] = 128*8/13;
+        s->last_dc[1] = 128*8/14;
+        s->last_dc[2] = 128*8/14;
+    }
     s->mb_skip_run = 0;
     memset(s->last_mv, 0, sizeof(s->last_mv));
 
@@ -2479,7 +2638,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
             if(s->data_partitioning){
                 if(   s->pb2   .buf_end - s->pb2   .buf - (put_bits_count(&s->    pb2)>>3) < MAX_MB_BYTES
                    || s->tex_pb.buf_end - s->tex_pb.buf - (put_bits_count(&s->tex_pb )>>3) < MAX_MB_BYTES){
-                    av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
+                    av_log(s->avctx, AV_LOG_ERROR, "encoded partitioned frame too large\n");
                     return -1;
                 }
             }
@@ -2515,6 +2674,9 @@ static int encode_thread(AVCodecContext *c, void *arg){
                 case AV_CODEC_ID_MPEG1VIDEO:
                     if(s->mb_skip_run) is_gob_start=0;
                     break;
+                case AV_CODEC_ID_MJPEG:
+                    if(s->mb_x==0 && s->mb_y!=0) is_gob_start=1;
+                    break;
                 }
 
                 if(is_gob_start){
@@ -2526,7 +2688,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
                         }
                     }
 
-                    assert((put_bits_count(&s->pb)&7) == 0);
+                    av_assert2((put_bits_count(&s->pb)&7) == 0);
                     current_packet_size= put_bits_ptr(&s->pb) - s->ptr_lastgob;
 
                     if (s->error_rate && s->resync_mb_x + s->resync_mb_y > 0) {
@@ -2733,8 +2895,9 @@ static int encode_thread(AVCodecContext *c, void *arg){
                         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};
+                        int storecoefs = s->mb_intra && s->dc_val[0];
 
-                        assert(backup_s.dquant == 0);
+                        av_assert2(backup_s.dquant == 0);
 
                         //FIXME intra
                         s->mv_dir= best_s.mv_dir;
@@ -2752,7 +2915,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
                             if(qp < s->avctx->qmin || qp > s->avctx->qmax)
                                 continue;
                             backup_s.dquant= dquant;
-                            if(s->mb_intra && s->dc_val[0]){
+                            if(storecoefs){
                                 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(int16_t)*16);
@@ -2762,7 +2925,7 @@ static int encode_thread(AVCodecContext *c, void *arg){
                             encode_mb_hq(s, &backup_s, &best_s, CANDIDATE_MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb,
                                          &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
                             if(best_s.qscale != qp){
-                                if(s->mb_intra && s->dc_val[0]){
+                                if(storecoefs){
                                     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(int16_t)*16);
@@ -3162,6 +3325,13 @@ static int encode_picture(MpegEncContext *s, int picture_number)
         update_qscale(s);
     }
 
+    if(s->codec_id != AV_CODEC_ID_AMV){
+        if(s->q_chroma_intra_matrix   != s->q_intra_matrix  ) av_freep(&s->q_chroma_intra_matrix);
+        if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
+        s->q_chroma_intra_matrix   = s->q_intra_matrix;
+        s->q_chroma_intra_matrix16 = s->q_intra_matrix16;
+    }
+
     s->mb_intra=0; //for the rate distortion & bit compare functions
     for(i=1; i<context_count; i++){
         ret = ff_update_duplicate_context(s->thread_context[i], s);
@@ -3204,6 +3374,8 @@ 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;
+        if(s->msmpeg4_version >= 3)
+            s->no_rounding=1;
         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);
     }
@@ -3282,6 +3454,25 @@ static int encode_picture(MpegEncContext *s, int picture_number)
                        s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
         s->qscale= 8;
     }
+    if(s->codec_id == AV_CODEC_ID_AMV){
+        static const uint8_t y[32]={13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13};
+        static const uint8_t c[32]={14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14};
+        for(i=1;i<64;i++){
+            int j= s->dsp.idct_permutation[ff_zigzag_direct[i]];
+
+            s->intra_matrix[j] = sp5x_quant_table[5*2+0][i];
+            s->chroma_intra_matrix[j] = sp5x_quant_table[5*2+1][i];
+        }
+        s->y_dc_scale_table= y;
+        s->c_dc_scale_table= c;
+        s->intra_matrix[0] = 13;
+        s->chroma_intra_matrix[0] = 14;
+        ff_convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16,
+                       s->intra_matrix, s->intra_quant_bias, 8, 8, 1);
+        ff_convert_matrix(&s->dsp, s->q_chroma_intra_matrix, s->q_chroma_intra_matrix16,
+                       s->chroma_intra_matrix, s->intra_quant_bias, 8, 8, 1);
+        s->qscale= 8;
+    }
 
     //FIXME var duplication
     s->current_picture_ptr->f.key_frame =
@@ -3292,6 +3483,7 @@ static int encode_picture(MpegEncContext *s, int picture_number)
     if (s->current_picture.f.key_frame)
         s->picture_in_gop_number=0;
 
+    s->mb_x = s->mb_y = 0;
     s->last_bits= put_bits_count(&s->pb);
     switch(s->out_format) {
     case FMT_MJPEG:
@@ -3324,7 +3516,7 @@ static int encode_picture(MpegEncContext *s, int picture_number)
             ff_mpeg1_encode_picture_header(s, picture_number);
         break;
     default:
-        assert(0);
+        av_assert0(0);
     }
     bits= put_bits_count(&s->pb);
     s->header_bits= bits - s->last_bits;
@@ -3415,7 +3607,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
         block[0] = (block[0] + (q >> 1)) / q;
         start_i = 1;
         last_non_zero = 0;
-        qmat = s->q_intra_matrix[qscale];
+        qmat = n < 4 ? s->q_intra_matrix[qscale] : s->q_chroma_intra_matrix[qscale];
         if(s->mpeg_quant || s->out_format == FMT_MPEG1)
             bias= 1<<(QMAT_SHIFT-1);
         length     = s->intra_ac_vlc_length;
@@ -3461,7 +3653,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
 //                coeff[2][k]= -level+2;
             }
             coeff_count[i]= FFMIN(level, 2);
-            assert(coeff_count[i]);
+            av_assert2(coeff_count[i]);
             max |=level;
         }else{
             coeff[0][i]= (level>>31)|1;
@@ -3495,9 +3687,9 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
             const int alevel= FFABS(level);
             int unquant_coeff;
 
-            assert(level);
+            av_assert2(level);
 
-            if(s->out_format == FMT_H263){
+            if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
                 unquant_coeff= alevel*qmul + qadd;
             }else{ //MPEG1
                 j= s->dsp.idct_permutation[ scantable[i] ]; //FIXME optimize
@@ -3526,7 +3718,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
                     }
                 }
 
-                if(s->out_format == FMT_H263){
+                if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
                     for(j=survivor_count-1; j>=0; j--){
                         int run= i - survivor[j];
                         int score= distortion + last_length[UNI_AC_ENC_INDEX(run, level)]*lambda;
@@ -3552,7 +3744,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
                     }
                 }
 
-                if(s->out_format == FMT_H263){
+                if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
                   for(j=survivor_count-1; j>=0; j--){
                         int run= i - survivor[j];
                         int score= distortion + score_tab[i-run];
@@ -3585,7 +3777,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
         survivor[ survivor_count++ ]= i+1;
     }
 
-    if(s->out_format != FMT_H263){
+    if(s->out_format != FMT_H263 && s->out_format != FMT_H261){
         last_score= 256*256*256*120;
         for(i= survivor[0]; i<=last_non_zero + 1; i++){
             int score= score_tab[i];
@@ -3618,7 +3810,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
             int alevel= FFABS(level);
             int unquant_coeff, score, distortion;
 
-            if(s->out_format == FMT_H263){
+            if(s->out_format == FMT_H263 || s->out_format == FMT_H261){
                     unquant_coeff= (alevel*qmul + qadd)>>3;
             }else{ //MPEG1
                     unquant_coeff = (((  alevel  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
@@ -3644,7 +3836,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
     }
 
     i= last_i;
-    assert(last_level);
+    av_assert2(last_level);
 
     block[ perm_scantable[last_non_zero] ]= last_level;
     i -= last_run + 1;
@@ -3760,8 +3952,8 @@ STOP_TIMER("memset rem[]")}
         weight[i] = w;
 //        w=weight[i] = (63*qns + (w/2)) / w;
 
-        assert(w>0);
-        assert(w<(1<<6));
+        av_assert2(w>0);
+        av_assert2(w<(1<<6));
         sum += w*w;
     }
     lambda= sum*(uint64_t)s->lambda2 >> (FF_LAMBDA_SHIFT - 6 + 6 + 6 + 6);
@@ -3827,7 +4019,7 @@ STOP_TIMER("dct")}
             const int level= block[0];
             int change, old_coeff;
 
-            assert(s->mb_intra);
+            av_assert2(s->mb_intra);
 
             old_coeff= q*level;
 
@@ -3870,7 +4062,7 @@ STOP_TIMER("dct")}
             }else{
                 old_coeff=0;
                 run2--;
-                assert(run2>=0 || i >= last_non_zero );
+                av_assert2(run2>=0 || i >= last_non_zero );
             }
 
             for(change=-1; change<=1; change+=2){
@@ -3898,7 +4090,7 @@ STOP_TIMER("dct")}
                                          - last_length[UNI_AC_ENC_INDEX(run, level+64)];
                         }
                     }else{
-                        assert(FFABS(new_level)==1);
+                        av_assert2(FFABS(new_level)==1);
 
                         if(analyze_gradient){
                             int g= d1[ scantable[i] ];
@@ -3931,7 +4123,7 @@ STOP_TIMER("dct")}
                     }
                 }else{
                     new_coeff=0;
-                    assert(FFABS(level)==1);
+                    av_assert2(FFABS(level)==1);
 
                     if(i < last_non_zero){
                         int next_i= i + run2 + 1;
@@ -3960,7 +4152,7 @@ STOP_TIMER("dct")}
                 score *= lambda;
 
                 unquant_change= new_coeff - old_coeff;
-                assert((score < 100*lambda && score > -100*lambda) || lambda==0);
+                av_assert2((score < 100*lambda && score > -100*lambda) || lambda==0);
 
                 score+= s->dsp.try_8x8basis(rem, weight, basis[j], unquant_change);
                 if(score<best_score){
@@ -3991,7 +4183,7 @@ STOP_TIMER("iterative step")}
 
             if(best_coeff > last_non_zero){
                 last_non_zero= best_coeff;
-                assert(block[j]);
+                av_assert2(block[j]);
 #ifdef REFINE_STATS
 after_last++;
 #endif
@@ -4019,7 +4211,7 @@ if(block[j]){
 #ifdef REFINE_STATS
 count++;
 if(256*256*256*64 % count == 0){
-    printf("after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number);
+    av_log(s->avctx, AV_LOG_DEBUG, "after_last:%d to_zero:%d from_zero:%d raise:%d lower:%d sign:%d xyp:%d/%d/%d\n", after_last, to_zero, from_zero, raise, lower, messed_sign, s->mb_x, s->mb_y, s->picture_number);
 }
 #endif
             run=0;
@@ -4082,7 +4274,7 @@ int ff_dct_quantize_c(MpegEncContext *s,
         block[0] = (block[0] + (q >> 1)) / q;
         start_i = 1;
         last_non_zero = 0;
-        qmat = s->q_intra_matrix[qscale];
+        qmat = n < 4 ? s->q_intra_matrix[qscale] : s->q_chroma_intra_matrix[qscale];
         bias= s->intra_quant_bias<<(QMAT_SHIFT - QUANT_BIAS_SHIFT);
     } else {
         start_i = 0;