Merge commit '368f50359eb328b0b9d67451f56fda20b3255f9a'
authorMichael Niedermayer <michaelni@gmx.at>
Fri, 30 May 2014 00:17:55 +0000 (02:17 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Fri, 30 May 2014 00:43:34 +0000 (02:43 +0200)
* commit '368f50359eb328b0b9d67451f56fda20b3255f9a':
  dsputil: Split off quarterpel bits into their own context

Conflicts:
configure
libavcodec/dsputil.c
libavcodec/h263dec.c
libavcodec/mpegvideo.c
libavcodec/mpegvideo_enc.c
libavcodec/vc1dec.c
libavcodec/vc1dsp.c
libavcodec/x86/dsputil_init.c
libavcodec/x86/qpeldsp.asm

Merged-by: Michael Niedermayer <michaelni@gmx.at>
32 files changed:
1  2 
configure
libavcodec/Makefile
libavcodec/cavs.c
libavcodec/cavsdsp.h
libavcodec/dsputil.c
libavcodec/dsputil.h
libavcodec/h263dec.c
libavcodec/h264.h
libavcodec/h264_mb.c
libavcodec/h264qpel.h
libavcodec/motion_est.c
libavcodec/mpeg4videodec.c
libavcodec/mpegvideo.c
libavcodec/mpegvideo.h
libavcodec/mpegvideo_enc.c
libavcodec/mpegvideo_motion.c
libavcodec/mss2.c
libavcodec/qpel_template.c
libavcodec/qpeldsp.c
libavcodec/qpeldsp.h
libavcodec/rv34.c
libavcodec/rv34dsp.h
libavcodec/snow.c
libavcodec/snow.h
libavcodec/snowenc.c
libavcodec/vc1dec.c
libavcodec/vc1dsp.c
libavcodec/wmv2dsp.h
libavcodec/x86/Makefile
libavcodec/x86/dsputil_init.c
libavcodec/x86/qpeldsp.asm
libavcodec/x86/qpeldsp_init.c

diff --cc configure
+++ b/configure
@@@ -1986,8 -1704,8 +1987,8 @@@ mdct_select="fft
  rdft_select="fft"
  mpegaudio_select="mpegaudiodsp"
  mpegaudiodsp_select="dct"
 -mpegvideo_select="dsputil hpeldsp videodsp"
 +mpegvideo_select="dsputil h264chroma hpeldsp videodsp"
- mpegvideoenc_select="dsputil mpegvideo"
+ mpegvideoenc_select="dsputil mpegvideo qpeldsp"
  
  # decoders / encoders
  aac_decoder_select="mdct sinewin"
@@@ -2050,11 -1763,11 +2051,11 @@@ flv_decoder_select="h263_decoder
  flv_encoder_select="h263_encoder"
  fourxm_decoder_select="dsputil"
  fraps_decoder_select="dsputil huffman"
 -g2m_decoder_deps="zlib"
 -g2m_decoder_select="dsputil"
 +g2m_decoder_select="dsputil zlib"
 +g729_decoder_select="dsputil"
  h261_decoder_select="error_resilience mpegvideo"
  h261_encoder_select="aandcttables mpegvideoenc"
- h263_decoder_select="error_resilience h263_parser h263dsp mpegvideo"
+ h263_decoder_select="error_resilience h263_parser h263dsp mpegvideo qpeldsp"
  h263_encoder_select="aandcttables h263dsp mpegvideoenc"
  h263i_decoder_select="h263_decoder"
  h263p_encoder_select="h263_encoder"
@@@ -2257,9 -1937,8 +2258,9 @@@ wmv3_vdpau_hwaccel_select="vc1_vdpau_hw
  
  # parsers
  h264_parser_select="h264_decoder"
- mpeg4video_parser_select="error_resilience h263dsp mpegvideo"
 +hevc_parser_select="hevc_decoder"
  mpegvideo_parser_select="error_resilience mpegvideo"
+ mpeg4video_parser_select="error_resilience h263dsp mpegvideo qpeldsp"
  vc1_parser_select="mpegvideo"
  
  # external libraries
Simple merge
Simple merge
Simple merge
Simple merge
Simple merge
@@@ -38,7 -36,7 +38,8 @@@
  #include "mpeg4video_parser.h"
  #include "mpegvideo.h"
  #include "msmpeg4.h"
+ #include "qpeldsp.h"
 +#include "vdpau_internal.h"
  #include "thread.h"
  
  av_cold int ff_h263_decode_init(AVCodecContext *avctx)
Simple merge
Simple merge
Simple merge
Simple merge
@@@ -2131,91 -2065,15 +2131,91 @@@ int ff_mpeg4_workaround_bugs(AVCodecCon
          ctx->divx_build   = -1;
      }
  
-     s->dsp.put_        ## postfix1 = ff_put_        ## postfix2;    \
-     s->dsp.put_no_rnd_ ## postfix1 = ff_put_no_rnd_ ## postfix2;    \
-     s->dsp.avg_        ## postfix1 = ff_avg_        ## postfix2;
 +    if (s->workaround_bugs & FF_BUG_AUTODETECT) {
 +        if (s->codec_tag == AV_RL32("XVIX"))
 +            s->workaround_bugs |= FF_BUG_XVID_ILACE;
 +
 +        if (s->codec_tag == AV_RL32("UMP4"))
 +            s->workaround_bugs |= FF_BUG_UMP4;
 +
 +        if (ctx->divx_version >= 500 && ctx->divx_build < 1814)
 +            s->workaround_bugs |= FF_BUG_QPEL_CHROMA;
 +
 +        if (ctx->divx_version > 502 && ctx->divx_build < 1814)
 +            s->workaround_bugs |= FF_BUG_QPEL_CHROMA2;
 +
 +        if (ctx->xvid_build <= 3U)
 +            s->padding_bug_score = 256 * 256 * 256 * 64;
 +
 +        if (ctx->xvid_build <= 1U)
 +            s->workaround_bugs |= FF_BUG_QPEL_CHROMA;
 +
 +        if (ctx->xvid_build <= 12U)
 +            s->workaround_bugs |= FF_BUG_EDGE;
 +
 +        if (ctx->xvid_build <= 32U)
 +            s->workaround_bugs |= FF_BUG_DC_CLIP;
 +
 +#define SET_QPEL_FUNC(postfix1, postfix2)                           \
++    s->qdsp.put_        ## postfix1 = ff_put_        ## postfix2;   \
++    s->qdsp.put_no_rnd_ ## postfix1 = ff_put_no_rnd_ ## postfix2;   \
++    s->qdsp.avg_        ## postfix1 = ff_avg_        ## postfix2;
 +
 +        if (ctx->lavc_build < 4653U)
 +            s->workaround_bugs |= FF_BUG_STD_QPEL;
 +
 +        if (ctx->lavc_build < 4655U)
 +            s->workaround_bugs |= FF_BUG_DIRECT_BLOCKSIZE;
 +
 +        if (ctx->lavc_build < 4670U)
 +            s->workaround_bugs |= FF_BUG_EDGE;
 +
 +        if (ctx->lavc_build <= 4712U)
 +            s->workaround_bugs |= FF_BUG_DC_CLIP;
 +
 +        if (ctx->divx_version >= 0)
 +            s->workaround_bugs |= FF_BUG_DIRECT_BLOCKSIZE;
 +        if (ctx->divx_version == 501 && ctx->divx_build == 20020416)
 +            s->padding_bug_score = 256 * 256 * 256 * 64;
 +
 +        if (ctx->divx_version < 500U)
 +            s->workaround_bugs |= FF_BUG_EDGE;
 +
 +        if (ctx->divx_version >= 0)
 +            s->workaround_bugs |= FF_BUG_HPEL_CHROMA;
 +    }
 +
 +    if (s->workaround_bugs & FF_BUG_STD_QPEL) {
 +        SET_QPEL_FUNC(qpel_pixels_tab[0][5], qpel16_mc11_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[0][7], qpel16_mc31_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[0][9], qpel16_mc12_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_old_c)
 +
 +        SET_QPEL_FUNC(qpel_pixels_tab[1][5], qpel8_mc11_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[1][7], qpel8_mc31_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[1][9], qpel8_mc12_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_old_c)
 +        SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_old_c)
 +    }
 +
 +    if (avctx->debug & FF_DEBUG_BUGS)
 +        av_log(s->avctx, AV_LOG_DEBUG,
 +               "bugs: %X lavc_build:%d xvid_build:%d divx_version:%d divx_build:%d %s\n",
 +               s->workaround_bugs, ctx->lavc_build, ctx->xvid_build,
 +               ctx->divx_version, ctx->divx_build, s->divx_packed ? "p" : "");
 +
  #if HAVE_MMX
 -    if (ctx->xvid_build >= 0                &&
 -        s->avctx->idct_algo == FF_IDCT_AUTO &&
 +    if (s->codec_id == AV_CODEC_ID_MPEG4 && ctx->xvid_build >= 0 &&
 +        avctx->idct_algo == FF_IDCT_AUTO &&
          (av_get_cpu_flags() & AV_CPU_FLAG_MMX)) {
 -        s->avctx->idct_algo = FF_IDCT_XVIDMMX;
 +        avctx->idct_algo = FF_IDCT_XVIDMMX;
          ff_dct_common_init(s);
 +        return 1;
      }
  #endif
 -
      return 0;
  }
  
@@@ -41,6 -40,8 +41,7 @@@
  #include "mpegvideo.h"
  #include "mjpegenc.h"
  #include "msmpeg4.h"
 -#include "xvmc_internal.h"
+ #include "qpeldsp.h"
  #include "thread.h"
  #include <limits.h>
  
@@@ -354,8 -348,8 +355,9 @@@ typedef struct MpegEncContext 
      int h263_long_vectors;      ///< use horrible h263v1 long vector mode
  
      DSPContext dsp;             ///< pointers for accelerated dsp functions
 +    H264ChromaContext h264chroma;
      HpelDSPContext hdsp;
+     QpelDSPContext qdsp;
      VideoDSPContext vdsp;
      H263DSPContext h263dsp;
      int f_code;                 ///< forward MV resolution
@@@ -803,6 -685,11 +804,8 @@@ av_cold int ff_MPV_encode_init(AVCodecC
      if (ff_MPV_common_init(s) < 0)
          return -1;
  
 -    if (ARCH_X86)
 -        ff_MPV_encode_init_x86(s);
 -
+     ff_qpeldsp_init(&s->qdsp);
      s->avctx->coded_frame = s->current_picture.f;
  
      if (s->msmpeg4_version) {
Simple merge
Simple merge
@@@ -1,13 -1,9 +1,9 @@@
  /*
-  * DSP utils
-  * Copyright (c) 2000, 2001 Fabrice Bellard
-  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
-  *
-  * gmc & q-pel & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at>
+  * quarterpel DSP function templates
   *
 - * 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.
index 0000000,7b51ce2..0beeab8
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,764 +1,810 @@@
 - * This file is part of Libav.
+ /*
+  * quarterpel DSP functions
+  *
 - * Libav is free software; you can redistribute it and/or
++ * This file is part of FFmpeg.
+  *
 - * Libav is distributed in the hope that it will be useful,
++ * 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.
+  *
 - * License along with Libav; if not, write to the Free Software
++ * 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 FFmpeg; if not, write to the Free Software
+  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+  */
+ /**
+  * @file
+  * quarterpel DSP functions
+  */
+ #include <stddef.h>
+ #include <stdint.h>
+ #include "config.h"
+ #include "libavutil/attributes.h"
+ #include "copy_block.h"
+ #include "qpeldsp.h"
++#include "diracdsp.h"
+ #define BIT_DEPTH 8
+ #include "hpel_template.c"
+ #include "tpel_template.c"
+ #include "qpel_template.c"
+ #define QPEL_MC(r, OPNAME, RND, OP)                                           \
+ static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src,       \
+                                             int dstStride, int srcStride,     \
+                                             int h)                            \
+ {                                                                             \
+     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;                           \
+     int i;                                                                    \
+                                                                               \
+     for (i = 0; i < h; i++) {                                                 \
+         OP(dst[0], (src[0] + src[1]) * 20 - (src[0] + src[2]) * 6 + (src[1] + src[3]) * 3 - (src[2] + src[4])); \
+         OP(dst[1], (src[1] + src[2]) * 20 - (src[0] + src[3]) * 6 + (src[0] + src[4]) * 3 - (src[1] + src[5])); \
+         OP(dst[2], (src[2] + src[3]) * 20 - (src[1] + src[4]) * 6 + (src[0] + src[5]) * 3 - (src[0] + src[6])); \
+         OP(dst[3], (src[3] + src[4]) * 20 - (src[2] + src[5]) * 6 + (src[1] + src[6]) * 3 - (src[0] + src[7])); \
+         OP(dst[4], (src[4] + src[5]) * 20 - (src[3] + src[6]) * 6 + (src[2] + src[7]) * 3 - (src[1] + src[8])); \
+         OP(dst[5], (src[5] + src[6]) * 20 - (src[4] + src[7]) * 6 + (src[3] + src[8]) * 3 - (src[2] + src[8])); \
+         OP(dst[6], (src[6] + src[7]) * 20 - (src[5] + src[8]) * 6 + (src[4] + src[8]) * 3 - (src[3] + src[7])); \
+         OP(dst[7], (src[7] + src[8]) * 20 - (src[6] + src[8]) * 6 + (src[5] + src[7]) * 3 - (src[4] + src[6])); \
+         dst += dstStride;                                                     \
+         src += srcStride;                                                     \
+     }                                                                         \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src,       \
+                                             int dstStride, int srcStride)     \
+ {                                                                             \
+     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;                           \
+     const int w = 8;                                                          \
+     int i;                                                                    \
+                                                                               \
+     for (i = 0; i < w; i++) {                                                 \
+         const int src0 = src[0 * srcStride];                                  \
+         const int src1 = src[1 * srcStride];                                  \
+         const int src2 = src[2 * srcStride];                                  \
+         const int src3 = src[3 * srcStride];                                  \
+         const int src4 = src[4 * srcStride];                                  \
+         const int src5 = src[5 * srcStride];                                  \
+         const int src6 = src[6 * srcStride];                                  \
+         const int src7 = src[7 * srcStride];                                  \
+         const int src8 = src[8 * srcStride];                                  \
+         OP(dst[0 * dstStride], (src0 + src1) * 20 - (src0 + src2) * 6 + (src1 + src3) * 3 - (src2 + src4)); \
+         OP(dst[1 * dstStride], (src1 + src2) * 20 - (src0 + src3) * 6 + (src0 + src4) * 3 - (src1 + src5)); \
+         OP(dst[2 * dstStride], (src2 + src3) * 20 - (src1 + src4) * 6 + (src0 + src5) * 3 - (src0 + src6)); \
+         OP(dst[3 * dstStride], (src3 + src4) * 20 - (src2 + src5) * 6 + (src1 + src6) * 3 - (src0 + src7)); \
+         OP(dst[4 * dstStride], (src4 + src5) * 20 - (src3 + src6) * 6 + (src2 + src7) * 3 - (src1 + src8)); \
+         OP(dst[5 * dstStride], (src5 + src6) * 20 - (src4 + src7) * 6 + (src3 + src8) * 3 - (src2 + src8)); \
+         OP(dst[6 * dstStride], (src6 + src7) * 20 - (src5 + src8) * 6 + (src4 + src8) * 3 - (src3 + src7)); \
+         OP(dst[7 * dstStride], (src7 + src8) * 20 - (src6 + src8) * 6 + (src5 + src7) * 3 - (src4 + src6)); \
+         dst++;                                                                \
+         src++;                                                                \
+     }                                                                         \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src,      \
+                                              int dstStride, int srcStride,    \
+                                              int h)                           \
+ {                                                                             \
+     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;                           \
+     int i;                                                                    \
+                                                                               \
+     for (i = 0; i < h; i++) {                                                 \
+         OP(dst[0],  (src[0]  + src[1])  * 20 - (src[0]  + src[2])  * 6 + (src[1]  + src[3])  * 3 - (src[2]  + src[4]));  \
+         OP(dst[1],  (src[1]  + src[2])  * 20 - (src[0]  + src[3])  * 6 + (src[0]  + src[4])  * 3 - (src[1]  + src[5]));  \
+         OP(dst[2],  (src[2]  + src[3])  * 20 - (src[1]  + src[4])  * 6 + (src[0]  + src[5])  * 3 - (src[0]  + src[6]));  \
+         OP(dst[3],  (src[3]  + src[4])  * 20 - (src[2]  + src[5])  * 6 + (src[1]  + src[6])  * 3 - (src[0]  + src[7]));  \
+         OP(dst[4],  (src[4]  + src[5])  * 20 - (src[3]  + src[6])  * 6 + (src[2]  + src[7])  * 3 - (src[1]  + src[8]));  \
+         OP(dst[5],  (src[5]  + src[6])  * 20 - (src[4]  + src[7])  * 6 + (src[3]  + src[8])  * 3 - (src[2]  + src[9]));  \
+         OP(dst[6],  (src[6]  + src[7])  * 20 - (src[5]  + src[8])  * 6 + (src[4]  + src[9])  * 3 - (src[3]  + src[10])); \
+         OP(dst[7],  (src[7]  + src[8])  * 20 - (src[6]  + src[9])  * 6 + (src[5]  + src[10]) * 3 - (src[4]  + src[11])); \
+         OP(dst[8],  (src[8]  + src[9])  * 20 - (src[7]  + src[10]) * 6 + (src[6]  + src[11]) * 3 - (src[5]  + src[12])); \
+         OP(dst[9],  (src[9]  + src[10]) * 20 - (src[8]  + src[11]) * 6 + (src[7]  + src[12]) * 3 - (src[6]  + src[13])); \
+         OP(dst[10], (src[10] + src[11]) * 20 - (src[9]  + src[12]) * 6 + (src[8]  + src[13]) * 3 - (src[7]  + src[14])); \
+         OP(dst[11], (src[11] + src[12]) * 20 - (src[10] + src[13]) * 6 + (src[9]  + src[14]) * 3 - (src[8]  + src[15])); \
+         OP(dst[12], (src[12] + src[13]) * 20 - (src[11] + src[14]) * 6 + (src[10] + src[15]) * 3 - (src[9]  + src[16])); \
+         OP(dst[13], (src[13] + src[14]) * 20 - (src[12] + src[15]) * 6 + (src[11] + src[16]) * 3 - (src[10] + src[16])); \
+         OP(dst[14], (src[14] + src[15]) * 20 - (src[13] + src[16]) * 6 + (src[12] + src[16]) * 3 - (src[11] + src[15])); \
+         OP(dst[15], (src[15] + src[16]) * 20 - (src[14] + src[16]) * 6 + (src[13] + src[15]) * 3 - (src[12] + src[14])); \
+         dst += dstStride;                                                     \
+         src += srcStride;                                                     \
+     }                                                                         \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src,      \
+                                              int dstStride, int srcStride)    \
+ {                                                                             \
+     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;                           \
+     const int w = 16;                                                         \
+     int i;                                                                    \
+                                                                               \
+     for (i = 0; i < w; i++) {                                                 \
+         const int src0  = src[0  * srcStride];                                \
+         const int src1  = src[1  * srcStride];                                \
+         const int src2  = src[2  * srcStride];                                \
+         const int src3  = src[3  * srcStride];                                \
+         const int src4  = src[4  * srcStride];                                \
+         const int src5  = src[5  * srcStride];                                \
+         const int src6  = src[6  * srcStride];                                \
+         const int src7  = src[7  * srcStride];                                \
+         const int src8  = src[8  * srcStride];                                \
+         const int src9  = src[9  * srcStride];                                \
+         const int src10 = src[10 * srcStride];                                \
+         const int src11 = src[11 * srcStride];                                \
+         const int src12 = src[12 * srcStride];                                \
+         const int src13 = src[13 * srcStride];                                \
+         const int src14 = src[14 * srcStride];                                \
+         const int src15 = src[15 * srcStride];                                \
+         const int src16 = src[16 * srcStride];                                \
+         OP(dst[0  * dstStride], (src0  + src1)  * 20 - (src0  + src2)  * 6 + (src1  + src3)  * 3 - (src2  + src4));  \
+         OP(dst[1  * dstStride], (src1  + src2)  * 20 - (src0  + src3)  * 6 + (src0  + src4)  * 3 - (src1  + src5));  \
+         OP(dst[2  * dstStride], (src2  + src3)  * 20 - (src1  + src4)  * 6 + (src0  + src5)  * 3 - (src0  + src6));  \
+         OP(dst[3  * dstStride], (src3  + src4)  * 20 - (src2  + src5)  * 6 + (src1  + src6)  * 3 - (src0  + src7));  \
+         OP(dst[4  * dstStride], (src4  + src5)  * 20 - (src3  + src6)  * 6 + (src2  + src7)  * 3 - (src1  + src8));  \
+         OP(dst[5  * dstStride], (src5  + src6)  * 20 - (src4  + src7)  * 6 + (src3  + src8)  * 3 - (src2  + src9));  \
+         OP(dst[6  * dstStride], (src6  + src7)  * 20 - (src5  + src8)  * 6 + (src4  + src9)  * 3 - (src3  + src10)); \
+         OP(dst[7  * dstStride], (src7  + src8)  * 20 - (src6  + src9)  * 6 + (src5  + src10) * 3 - (src4  + src11)); \
+         OP(dst[8  * dstStride], (src8  + src9)  * 20 - (src7  + src10) * 6 + (src6  + src11) * 3 - (src5  + src12)); \
+         OP(dst[9  * dstStride], (src9  + src10) * 20 - (src8  + src11) * 6 + (src7  + src12) * 3 - (src6  + src13)); \
+         OP(dst[10 * dstStride], (src10 + src11) * 20 - (src9  + src12) * 6 + (src8  + src13) * 3 - (src7  + src14)); \
+         OP(dst[11 * dstStride], (src11 + src12) * 20 - (src10 + src13) * 6 + (src9  + src14) * 3 - (src8  + src15)); \
+         OP(dst[12 * dstStride], (src12 + src13) * 20 - (src11 + src14) * 6 + (src10 + src15) * 3 - (src9  + src16)); \
+         OP(dst[13 * dstStride], (src13 + src14) * 20 - (src12 + src15) * 6 + (src11 + src16) * 3 - (src10 + src16)); \
+         OP(dst[14 * dstStride], (src14 + src15) * 20 - (src13 + src16) * 6 + (src12 + src16) * 3 - (src11 + src15)); \
+         OP(dst[15 * dstStride], (src15 + src16) * 20 - (src14 + src16) * 6 + (src13 + src15) * 3 - (src12 + src14)); \
+         dst++;                                                                \
+         src++;                                                                \
+     }                                                                         \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t half[64];                                                         \
+                                                                               \
+     put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);             \
+     OPNAME ## pixels8_l2_8(dst, src, half, stride, stride, 8, 8);             \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);             \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t half[64];                                                         \
+                                                                               \
+     put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);             \
+     OPNAME ## pixels8_l2_8(dst, src + 1, half, stride, stride, 8, 8);         \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t half[64];                                                         \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);                   \
+     OPNAME ## pixels8_l2_8(dst, full, half, stride, 16, 8, 8);                \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);                   \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t half[64];                                                         \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);                   \
+     OPNAME ## pixels8_l2_8(dst, full + 16, half, stride, 16, 8, 8);           \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src,            \
+                                        ptrdiff_t stride)                      \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfV[64];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);                  \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l4_8(dst, full, halfH, halfV, halfHV,                   \
+                            stride, 16, 8, 8, 8, 8);                           \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## pixels8_l2_8(halfH, halfH, full, 8, 8, 16, 9);              \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l2_8(dst, halfH, halfHV, stride, 8, 8, 8);              \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src,            \
+                                        ptrdiff_t stride)                      \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfV[64];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full + 1, 8, 16);              \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l4_8(dst, full + 1, halfH, halfV, halfHV,               \
+                            stride, 16, 8, 8, 8, 8);                           \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## pixels8_l2_8(halfH, halfH, full + 1, 8, 8, 16, 9);          \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l2_8(dst, halfH, halfHV, stride, 8, 8, 8);              \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src,            \
+                                        ptrdiff_t stride)                      \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfV[64];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);                  \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l4_8(dst, full + 16, halfH + 8, halfV, halfHV,          \
+                            stride, 16, 8, 8, 8, 8);                           \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## pixels8_l2_8(halfH, halfH, full, 8, 8, 16, 9);              \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l2_8(dst, halfH + 8, halfHV, stride, 8, 8, 8);          \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src,            \
+                                        ptrdiff_t stride)                      \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfV[64];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full + 1, 8, 16);              \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l4_8(dst, full + 17, halfH + 8, halfV, halfHV,          \
+                            stride, 16, 8, 8, 8, 8);                           \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## pixels8_l2_8(halfH, halfH, full + 1, 8, 8, 16, 9);          \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l2_8(dst, halfH + 8, halfHV, stride, 8, 8, 8);          \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);            \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l2_8(dst, halfH, halfHV, stride, 8, 8, 8);              \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);            \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l2_8(dst, halfH + 8, halfHV, stride, 8, 8, 8);          \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel8_mc12_old_c(uint8_t *dst, uint8_t *src,            \
+                                        ptrdiff_t stride)                      \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfV[64];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);                  \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l2_8(dst, halfV, halfHV, stride, 8, 8, 8);              \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc12_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## pixels8_l2_8(halfH, halfH, full, 8, 8, 16, 9);              \
+     OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);                   \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel8_mc32_old_c(uint8_t *dst, uint8_t *src,            \
+                                        ptrdiff_t stride)                      \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+     uint8_t halfV[64];                                                        \
+     uint8_t halfHV[64];                                                       \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full + 1, 8, 16);              \
+     put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);                 \
+     OPNAME ## pixels8_l2_8(dst, halfV, halfHV, stride, 8, 8, 8);              \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t full[16 * 9];                                                     \
+     uint8_t halfH[72];                                                        \
+                                                                               \
+     copy_block9(full, src, 16, stride, 9);                                    \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);               \
+     put ## RND ## pixels8_l2_8(halfH, halfH, full + 1, 8, 8, 16, 9);          \
+     OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);                   \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src,                \
+                                    ptrdiff_t stride)                          \
+ {                                                                             \
+     uint8_t halfH[72];                                                        \
+                                                                               \
+     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);            \
+     OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);                   \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t half[256];                                                        \
+                                                                               \
+     put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);          \
+     OPNAME ## pixels16_l2_8(dst, src, half, stride, stride, 16, 16);          \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);           \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t half[256];                                                        \
+                                                                               \
+     put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);          \
+     OPNAME ## pixels16_l2_8(dst, src + 1, half, stride, stride, 16, 16);      \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t half[256];                                                        \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);                 \
+     OPNAME ## pixels16_l2_8(dst, full, half, stride, 24, 16, 16);             \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);                  \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t half[256];                                                        \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);                 \
+     OPNAME ## pixels16_l2_8(dst, full + 24, half, stride, 24, 16, 16);        \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src,           \
+                                         ptrdiff_t stride)                     \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfV[256];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);                \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l4_8(dst, full, halfH, halfV, halfHV,                  \
+                             stride, 24, 16, 16, 16, 16);                      \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## pixels16_l2_8(halfH, halfH, full, 16, 16, 24, 17);          \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l2_8(dst, halfH, halfHV, stride, 16, 16, 16);          \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src,           \
+                                         ptrdiff_t stride)                     \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfV[256];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full + 1, 16, 24);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l4_8(dst, full + 1, halfH, halfV, halfHV,              \
+                             stride, 24, 16, 16, 16, 16);                      \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## pixels16_l2_8(halfH, halfH, full + 1, 16, 16, 24, 17);      \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l2_8(dst, halfH, halfHV, stride, 16, 16, 16);          \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src,           \
+                                         ptrdiff_t stride)                     \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfV[256];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);                \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l4_8(dst, full + 24, halfH + 16, halfV, halfHV,        \
+                             stride, 24, 16, 16, 16, 16);                      \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## pixels16_l2_8(halfH, halfH, full, 16, 16, 24, 17);          \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l2_8(dst, halfH + 16, halfHV, stride, 16, 16, 16);     \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src,           \
+                                         ptrdiff_t stride)                     \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfV[256];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full + 1, 16, 24);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l4_8(dst, full + 25, halfH + 16, halfV, halfHV,        \
+                             stride, 24, 16, 16, 16, 16);                      \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## pixels16_l2_8(halfH, halfH, full + 1, 16, 16, 24, 17);      \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l2_8(dst, halfH + 16, halfHV, stride, 16, 16, 16);     \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);         \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l2_8(dst, halfH, halfHV, stride, 16, 16, 16);          \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);         \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l2_8(dst, halfH + 16, halfHV, stride, 16, 16, 16);     \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src,           \
+                                         ptrdiff_t stride)                     \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfV[256];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);                \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l2_8(dst, halfV, halfHV, stride, 16, 16, 16);          \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## pixels16_l2_8(halfH, halfH, full, 16, 16, 24, 17);          \
+     OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);                 \
+ }                                                                             \
+                                                                               \
+ void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src,           \
+                                         ptrdiff_t stride)                     \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+     uint8_t halfV[256];                                                       \
+     uint8_t halfHV[256];                                                      \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full + 1, 16, 24);            \
+     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);              \
+     OPNAME ## pixels16_l2_8(dst, halfV, halfHV, stride, 16, 16, 16);          \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t full[24 * 17];                                                    \
+     uint8_t halfH[272];                                                       \
+                                                                               \
+     copy_block17(full, src, 24, stride, 17);                                  \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);            \
+     put ## RND ## pixels16_l2_8(halfH, halfH, full + 1, 16, 16, 24, 17);      \
+     OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);                 \
+ }                                                                             \
+                                                                               \
+ static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src,               \
+                                     ptrdiff_t stride)                         \
+ {                                                                             \
+     uint8_t halfH[272];                                                       \
+                                                                               \
+     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);         \
+     OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);                 \
+ }
+ #define op_avg(a, b)        a = (((a) + cm[((b) + 16) >> 5] + 1) >> 1)
+ #define op_put(a, b)        a = cm[((b) + 16) >> 5]
+ #define op_put_no_rnd(a, b) a = cm[((b) + 15) >> 5]
+ QPEL_MC(0, put_, _, op_put)
+ QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd)
+ QPEL_MC(0, avg_, _, op_avg)
+ #undef op_avg
+ #undef op_put
+ #undef op_put_no_rnd
+ void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
+ {
+     put_pixels8_8_c(dst, src, stride, 8);
+ }
+ void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
+ {
+     avg_pixels8_8_c(dst, src, stride, 8);
+ }
+ void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
+ {
+     put_pixels16_8_c(dst, src, stride, 16);
+ }
+ void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
+ {
+     avg_pixels16_8_c(dst, src, stride, 16);
+ }
+ #define put_qpel8_mc00_c         ff_put_pixels8x8_c
+ #define avg_qpel8_mc00_c         ff_avg_pixels8x8_c
+ #define put_qpel16_mc00_c        ff_put_pixels16x16_c
+ #define avg_qpel16_mc00_c        ff_avg_pixels16x16_c
+ #define put_no_rnd_qpel8_mc00_c  ff_put_pixels8x8_c
+ #define put_no_rnd_qpel16_mc00_c ff_put_pixels16x16_c
+ void ff_put_pixels8_l2_8(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
+                          int dst_stride, int src_stride1, int src_stride2,
+                          int h)
+ {
+     put_pixels8_l2_8(dst, src1, src2, dst_stride, src_stride1, src_stride2, h);
+ }
++#if CONFIG_DIRAC_DECODER
++#define DIRAC_MC(OPNAME)\
++void ff_ ## OPNAME ## _dirac_pixels8_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++     OPNAME ## _pixels8_8_c(dst, src[0], stride, h);\
++}\
++void ff_ ## OPNAME ## _dirac_pixels16_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++    OPNAME ## _pixels16_8_c(dst, src[0], stride, h);\
++}\
++void ff_ ## OPNAME ## _dirac_pixels32_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++    OPNAME ## _pixels16_8_c(dst   , src[0]   , stride, h);\
++    OPNAME ## _pixels16_8_c(dst+16, src[0]+16, stride, h);\
++}\
++void ff_ ## OPNAME ## _dirac_pixels8_l2_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++    OPNAME ## _pixels8_l2_8(dst, src[0], src[1], stride, stride, stride, h);\
++}\
++void ff_ ## OPNAME ## _dirac_pixels16_l2_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++    OPNAME ## _pixels16_l2_8(dst, src[0], src[1], stride, stride, stride, h);\
++}\
++void ff_ ## OPNAME ## _dirac_pixels32_l2_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++    OPNAME ## _pixels16_l2_8(dst   , src[0]   , src[1]   , stride, stride, stride, h);\
++    OPNAME ## _pixels16_l2_8(dst+16, src[0]+16, src[1]+16, stride, stride, stride, h);\
++}\
++void ff_ ## OPNAME ## _dirac_pixels8_l4_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++    OPNAME ## _pixels8_l4_8(dst, src[0], src[1], src[2], src[3], stride, stride, stride, stride, stride, h);\
++}\
++void ff_ ## OPNAME ## _dirac_pixels16_l4_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++    OPNAME ## _pixels16_l4_8(dst, src[0], src[1], src[2], src[3], stride, stride, stride, stride, stride, h);\
++}\
++void ff_ ## OPNAME ## _dirac_pixels32_l4_c(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
++{\
++    OPNAME ## _pixels16_l4_8(dst   , src[0]   , src[1]   , src[2]   , src[3]   , stride, stride, stride, stride, stride, h);\
++    OPNAME ## _pixels16_l4_8(dst+16, src[0]+16, src[1]+16, src[2]+16, src[3]+16, stride, stride, stride, stride, stride, h);\
++}
++DIRAC_MC(put)
++DIRAC_MC(avg)
++#endif
++
+ av_cold void ff_qpeldsp_init(QpelDSPContext *c)
+ {
+ #define dspfunc(PFX, IDX, NUM)                              \
+     c->PFX ## _pixels_tab[IDX][0]  = PFX ## NUM ## _mc00_c; \
+     c->PFX ## _pixels_tab[IDX][1]  = PFX ## NUM ## _mc10_c; \
+     c->PFX ## _pixels_tab[IDX][2]  = PFX ## NUM ## _mc20_c; \
+     c->PFX ## _pixels_tab[IDX][3]  = PFX ## NUM ## _mc30_c; \
+     c->PFX ## _pixels_tab[IDX][4]  = PFX ## NUM ## _mc01_c; \
+     c->PFX ## _pixels_tab[IDX][5]  = PFX ## NUM ## _mc11_c; \
+     c->PFX ## _pixels_tab[IDX][6]  = PFX ## NUM ## _mc21_c; \
+     c->PFX ## _pixels_tab[IDX][7]  = PFX ## NUM ## _mc31_c; \
+     c->PFX ## _pixels_tab[IDX][8]  = PFX ## NUM ## _mc02_c; \
+     c->PFX ## _pixels_tab[IDX][9]  = PFX ## NUM ## _mc12_c; \
+     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_c; \
+     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_c; \
+     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_c; \
+     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_c; \
+     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_c; \
+     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_c
+     dspfunc(put_qpel, 0, 16);
+     dspfunc(put_qpel, 1, 8);
+     dspfunc(put_no_rnd_qpel, 0, 16);
+     dspfunc(put_no_rnd_qpel, 1, 8);
+     dspfunc(avg_qpel, 0, 16);
+     dspfunc(avg_qpel, 1, 8);
+     if (ARCH_X86)
+         ff_qpeldsp_init_x86(c);
+ }
index 0000000,a750ff8..94caf6f
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,78 +1,78 @@@
 - * This file is part of Libav.
+ /*
+  * quarterpel DSP functions
+  *
 - * Libav is free software; you can redistribute it and/or
++ * This file is part of FFmpeg.
+  *
 - * Libav is distributed in the hope that it will be useful,
++ * 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.
+  *
 - * License along with Libav; if not, write to the Free Software
++ * 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 FFmpeg; if not, write to the Free Software
+  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+  */
+ /**
+  * @file
+  * quarterpel DSP functions
+  */
+ #ifndef AVCODEC_QPELDSP_H
+ #define AVCODEC_QPELDSP_H
+ #include <stddef.h>
+ #include <stdint.h>
+ void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+ void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+ void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+ void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+ void ff_put_pixels8_l2_8(uint8_t *dst, const uint8_t *src1, const uint8_t *src2,
+                          int dst_stride, int src_stride1, int src_stride2,
+                          int h);
+ #define DEF_OLD_QPEL(name)                                                     \
+     void ff_put_        ## name(uint8_t *dst /* align width (8 or 16) */,      \
+                                 uint8_t *src /* align 1 */, ptrdiff_t stride); \
+     void ff_put_no_rnd_ ## name(uint8_t *dst /* align width (8 or 16) */,      \
+                                 uint8_t *src /* align 1 */, ptrdiff_t stride); \
+     void ff_avg_        ## name(uint8_t *dst /* align width (8 or 16) */,      \
+                                 uint8_t *src /* align 1 */, ptrdiff_t stride);
+ DEF_OLD_QPEL(qpel16_mc11_old_c)
+ DEF_OLD_QPEL(qpel16_mc31_old_c)
+ DEF_OLD_QPEL(qpel16_mc12_old_c)
+ DEF_OLD_QPEL(qpel16_mc32_old_c)
+ DEF_OLD_QPEL(qpel16_mc13_old_c)
+ DEF_OLD_QPEL(qpel16_mc33_old_c)
+ DEF_OLD_QPEL(qpel8_mc11_old_c)
+ DEF_OLD_QPEL(qpel8_mc31_old_c)
+ DEF_OLD_QPEL(qpel8_mc12_old_c)
+ DEF_OLD_QPEL(qpel8_mc32_old_c)
+ DEF_OLD_QPEL(qpel8_mc13_old_c)
+ DEF_OLD_QPEL(qpel8_mc33_old_c)
+ typedef void (*qpel_mc_func)(uint8_t *dst /* align width (8 or 16) */,
+                              uint8_t *src /* align 1 */, ptrdiff_t stride);
+ /**
+  * quarterpel DSP context
+  */
+ typedef struct QpelDSPContext {
+     qpel_mc_func put_qpel_pixels_tab[2][16];
+     qpel_mc_func avg_qpel_pixels_tab[2][16];
+     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
+ } QpelDSPContext;
+ void ff_qpeldsp_init(QpelDSPContext *c);
+ void ff_qpeldsp_init_x86(QpelDSPContext *c);
+ #endif /* AVCODEC_QPELDSP_H */
Simple merge
Simple merge
index 3961e67,0000000..3a825bd
mode 100644,000000..100644
--- /dev/null
@@@ -1,709 -1,0 +1,709 @@@
-     s->dsp.put_qpel_pixels_tab       [0][dy+dx/4]=\
-     s->dsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\
 +/*
 + * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
 + *
 + * This file is part of FFmpeg.
 + *
 + * 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.
 + *
 + * 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 FFmpeg; if not, write to the Free Software
 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 + */
 +
 +#include "libavutil/intmath.h"
 +#include "libavutil/log.h"
 +#include "libavutil/opt.h"
 +#include "avcodec.h"
 +#include "dsputil.h"
 +#include "snow_dwt.h"
 +#include "internal.h"
 +#include "snow.h"
 +#include "snowdata.h"
 +
 +#include "rangecoder.h"
 +#include "mathops.h"
 +#include "h263.h"
 +
 +
 +void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
 +                              int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
 +    int y, x;
 +    IDWTELEM * dst;
 +    for(y=0; y<b_h; y++){
 +        //FIXME ugly misuse of obmc_stride
 +        const uint8_t *obmc1= obmc + y*obmc_stride;
 +        const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
 +        const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
 +        const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
 +        dst = slice_buffer_get_line(sb, src_y + y);
 +        for(x=0; x<b_w; x++){
 +            int v=   obmc1[x] * block[3][x + y*src_stride]
 +                    +obmc2[x] * block[2][x + y*src_stride]
 +                    +obmc3[x] * block[1][x + y*src_stride]
 +                    +obmc4[x] * block[0][x + y*src_stride];
 +
 +            v <<= 8 - LOG2_OBMC_MAX;
 +            if(FRAC_BITS != 8){
 +                v >>= 8 - FRAC_BITS;
 +            }
 +            if(add){
 +                v += dst[x + src_x];
 +                v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
 +                if(v&(~255)) v= ~(v>>31);
 +                dst8[x + y*src_stride] = v;
 +            }else{
 +                dst[x + src_x] -= v;
 +            }
 +        }
 +    }
 +}
 +
 +void ff_snow_reset_contexts(SnowContext *s){ //FIXME better initial contexts
 +    int plane_index, level, orientation;
 +
 +    for(plane_index=0; plane_index<3; plane_index++){
 +        for(level=0; level<MAX_DECOMPOSITIONS; level++){
 +            for(orientation=level ? 1:0; orientation<4; orientation++){
 +                memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state));
 +            }
 +        }
 +    }
 +    memset(s->header_state, MID_STATE, sizeof(s->header_state));
 +    memset(s->block_state, MID_STATE, sizeof(s->block_state));
 +}
 +
 +int ff_snow_alloc_blocks(SnowContext *s){
 +    int w= FF_CEIL_RSHIFT(s->avctx->width,  LOG2_MB_SIZE);
 +    int h= FF_CEIL_RSHIFT(s->avctx->height, LOG2_MB_SIZE);
 +
 +    s->b_width = w;
 +    s->b_height= h;
 +
 +    av_free(s->block);
 +    s->block= av_mallocz_array(w * h,  sizeof(BlockNode) << (s->block_max_depth*2));
 +    if (!s->block)
 +        return AVERROR(ENOMEM);
 +
 +    return 0;
 +}
 +
 +static av_cold void init_qexp(void){
 +    int i;
 +    double v=128;
 +
 +    for(i=0; i<QROOT; i++){
 +        ff_qexp[i]= lrintf(v);
 +        v *= pow(2, 1.0 / QROOT);
 +    }
 +}
 +static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, int stride, int b_w, int b_h, int dx, int dy){
 +    static const uint8_t weight[64]={
 +    8,7,6,5,4,3,2,1,
 +    7,7,0,0,0,0,0,1,
 +    6,0,6,0,0,0,2,0,
 +    5,0,0,5,0,3,0,0,
 +    4,0,0,0,4,0,0,0,
 +    3,0,0,5,0,3,0,0,
 +    2,0,6,0,0,0,2,0,
 +    1,7,0,0,0,0,0,1,
 +    };
 +
 +    static const uint8_t brane[256]={
 +    0x00,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x11,0x12,0x12,0x12,0x12,0x12,0x12,0x12,
 +    0x04,0x05,0xcc,0xcc,0xcc,0xcc,0xcc,0x41,0x15,0x16,0xcc,0xcc,0xcc,0xcc,0xcc,0x52,
 +    0x04,0xcc,0x05,0xcc,0xcc,0xcc,0x41,0xcc,0x15,0xcc,0x16,0xcc,0xcc,0xcc,0x52,0xcc,
 +    0x04,0xcc,0xcc,0x05,0xcc,0x41,0xcc,0xcc,0x15,0xcc,0xcc,0x16,0xcc,0x52,0xcc,0xcc,
 +    0x04,0xcc,0xcc,0xcc,0x41,0xcc,0xcc,0xcc,0x15,0xcc,0xcc,0xcc,0x16,0xcc,0xcc,0xcc,
 +    0x04,0xcc,0xcc,0x41,0xcc,0x05,0xcc,0xcc,0x15,0xcc,0xcc,0x52,0xcc,0x16,0xcc,0xcc,
 +    0x04,0xcc,0x41,0xcc,0xcc,0xcc,0x05,0xcc,0x15,0xcc,0x52,0xcc,0xcc,0xcc,0x16,0xcc,
 +    0x04,0x41,0xcc,0xcc,0xcc,0xcc,0xcc,0x05,0x15,0x52,0xcc,0xcc,0xcc,0xcc,0xcc,0x16,
 +    0x44,0x45,0x45,0x45,0x45,0x45,0x45,0x45,0x55,0x56,0x56,0x56,0x56,0x56,0x56,0x56,
 +    0x48,0x49,0xcc,0xcc,0xcc,0xcc,0xcc,0x85,0x59,0x5A,0xcc,0xcc,0xcc,0xcc,0xcc,0x96,
 +    0x48,0xcc,0x49,0xcc,0xcc,0xcc,0x85,0xcc,0x59,0xcc,0x5A,0xcc,0xcc,0xcc,0x96,0xcc,
 +    0x48,0xcc,0xcc,0x49,0xcc,0x85,0xcc,0xcc,0x59,0xcc,0xcc,0x5A,0xcc,0x96,0xcc,0xcc,
 +    0x48,0xcc,0xcc,0xcc,0x49,0xcc,0xcc,0xcc,0x59,0xcc,0xcc,0xcc,0x96,0xcc,0xcc,0xcc,
 +    0x48,0xcc,0xcc,0x85,0xcc,0x49,0xcc,0xcc,0x59,0xcc,0xcc,0x96,0xcc,0x5A,0xcc,0xcc,
 +    0x48,0xcc,0x85,0xcc,0xcc,0xcc,0x49,0xcc,0x59,0xcc,0x96,0xcc,0xcc,0xcc,0x5A,0xcc,
 +    0x48,0x85,0xcc,0xcc,0xcc,0xcc,0xcc,0x49,0x59,0x96,0xcc,0xcc,0xcc,0xcc,0xcc,0x5A,
 +    };
 +
 +    static const uint8_t needs[16]={
 +    0,1,0,0,
 +    2,4,2,0,
 +    0,1,0,0,
 +    15
 +    };
 +
 +    int x, y, b, r, l;
 +    int16_t tmpIt   [64*(32+HTAPS_MAX)];
 +    uint8_t tmp2t[3][64*(32+HTAPS_MAX)];
 +    int16_t *tmpI= tmpIt;
 +    uint8_t *tmp2= tmp2t[0];
 +    const uint8_t *hpel[11];
 +    av_assert2(dx<16 && dy<16);
 +    r= brane[dx + 16*dy]&15;
 +    l= brane[dx + 16*dy]>>4;
 +
 +    b= needs[l] | needs[r];
 +    if(p && !p->diag_mc)
 +        b= 15;
 +
 +    if(b&5){
 +        for(y=0; y < b_h+HTAPS_MAX-1; y++){
 +            for(x=0; x < b_w; x++){
 +                int a_1=src[x + HTAPS_MAX/2-4];
 +                int a0= src[x + HTAPS_MAX/2-3];
 +                int a1= src[x + HTAPS_MAX/2-2];
 +                int a2= src[x + HTAPS_MAX/2-1];
 +                int a3= src[x + HTAPS_MAX/2+0];
 +                int a4= src[x + HTAPS_MAX/2+1];
 +                int a5= src[x + HTAPS_MAX/2+2];
 +                int a6= src[x + HTAPS_MAX/2+3];
 +                int am=0;
 +                if(!p || p->fast_mc){
 +                    am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
 +                    tmpI[x]= am;
 +                    am= (am+16)>>5;
 +                }else{
 +                    am= p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6);
 +                    tmpI[x]= am;
 +                    am= (am+32)>>6;
 +                }
 +
 +                if(am&(~255)) am= ~(am>>31);
 +                tmp2[x]= am;
 +            }
 +            tmpI+= 64;
 +            tmp2+= 64;
 +            src += stride;
 +        }
 +        src -= stride*y;
 +    }
 +    src += HTAPS_MAX/2 - 1;
 +    tmp2= tmp2t[1];
 +
 +    if(b&2){
 +        for(y=0; y < b_h; y++){
 +            for(x=0; x < b_w+1; x++){
 +                int a_1=src[x + (HTAPS_MAX/2-4)*stride];
 +                int a0= src[x + (HTAPS_MAX/2-3)*stride];
 +                int a1= src[x + (HTAPS_MAX/2-2)*stride];
 +                int a2= src[x + (HTAPS_MAX/2-1)*stride];
 +                int a3= src[x + (HTAPS_MAX/2+0)*stride];
 +                int a4= src[x + (HTAPS_MAX/2+1)*stride];
 +                int a5= src[x + (HTAPS_MAX/2+2)*stride];
 +                int a6= src[x + (HTAPS_MAX/2+3)*stride];
 +                int am=0;
 +                if(!p || p->fast_mc)
 +                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
 +                else
 +                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;
 +
 +                if(am&(~255)) am= ~(am>>31);
 +                tmp2[x]= am;
 +            }
 +            src += stride;
 +            tmp2+= 64;
 +        }
 +        src -= stride*y;
 +    }
 +    src += stride*(HTAPS_MAX/2 - 1);
 +    tmp2= tmp2t[2];
 +    tmpI= tmpIt;
 +    if(b&4){
 +        for(y=0; y < b_h; y++){
 +            for(x=0; x < b_w; x++){
 +                int a_1=tmpI[x + (HTAPS_MAX/2-4)*64];
 +                int a0= tmpI[x + (HTAPS_MAX/2-3)*64];
 +                int a1= tmpI[x + (HTAPS_MAX/2-2)*64];
 +                int a2= tmpI[x + (HTAPS_MAX/2-1)*64];
 +                int a3= tmpI[x + (HTAPS_MAX/2+0)*64];
 +                int a4= tmpI[x + (HTAPS_MAX/2+1)*64];
 +                int a5= tmpI[x + (HTAPS_MAX/2+2)*64];
 +                int a6= tmpI[x + (HTAPS_MAX/2+3)*64];
 +                int am=0;
 +                if(!p || p->fast_mc)
 +                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
 +                else
 +                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;
 +                if(am&(~255)) am= ~(am>>31);
 +                tmp2[x]= am;
 +            }
 +            tmpI+= 64;
 +            tmp2+= 64;
 +        }
 +    }
 +
 +    hpel[ 0]= src;
 +    hpel[ 1]= tmp2t[0] + 64*(HTAPS_MAX/2-1);
 +    hpel[ 2]= src + 1;
 +
 +    hpel[ 4]= tmp2t[1];
 +    hpel[ 5]= tmp2t[2];
 +    hpel[ 6]= tmp2t[1] + 1;
 +
 +    hpel[ 8]= src + stride;
 +    hpel[ 9]= hpel[1] + 64;
 +    hpel[10]= hpel[8] + 1;
 +
 +#define MC_STRIDE(x) (needs[x] ? 64 : stride)
 +
 +    if(b==15){
 +        int dxy = dx / 8 + dy / 8 * 4;
 +        const uint8_t *src1 = hpel[dxy    ];
 +        const uint8_t *src2 = hpel[dxy + 1];
 +        const uint8_t *src3 = hpel[dxy + 4];
 +        const uint8_t *src4 = hpel[dxy + 5];
 +        int stride1 = MC_STRIDE(dxy);
 +        int stride2 = MC_STRIDE(dxy + 1);
 +        int stride3 = MC_STRIDE(dxy + 4);
 +        int stride4 = MC_STRIDE(dxy + 5);
 +        dx&=7;
 +        dy&=7;
 +        for(y=0; y < b_h; y++){
 +            for(x=0; x < b_w; x++){
 +                dst[x]= ((8-dx)*(8-dy)*src1[x] + dx*(8-dy)*src2[x]+
 +                         (8-dx)*   dy *src3[x] + dx*   dy *src4[x]+32)>>6;
 +            }
 +            src1+=stride1;
 +            src2+=stride2;
 +            src3+=stride3;
 +            src4+=stride4;
 +            dst +=stride;
 +        }
 +    }else{
 +        const uint8_t *src1= hpel[l];
 +        const uint8_t *src2= hpel[r];
 +        int stride1 = MC_STRIDE(l);
 +        int stride2 = MC_STRIDE(r);
 +        int a= weight[((dx&7) + (8*(dy&7)))];
 +        int b= 8-a;
 +        for(y=0; y < b_h; y++){
 +            for(x=0; x < b_w; x++){
 +                dst[x]= (a*src1[x] + b*src2[x] + 4)>>3;
 +            }
 +            src1+=stride1;
 +            src2+=stride2;
 +            dst +=stride;
 +        }
 +    }
 +}
 +
 +void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){
 +    if(block->type & BLOCK_INTRA){
 +        int x, y;
 +        const unsigned color  = block->color[plane_index];
 +        const unsigned color4 = color*0x01010101;
 +        if(b_w==32){
 +            for(y=0; y < b_h; y++){
 +                *(uint32_t*)&dst[0 + y*stride]= color4;
 +                *(uint32_t*)&dst[4 + y*stride]= color4;
 +                *(uint32_t*)&dst[8 + y*stride]= color4;
 +                *(uint32_t*)&dst[12+ y*stride]= color4;
 +                *(uint32_t*)&dst[16+ y*stride]= color4;
 +                *(uint32_t*)&dst[20+ y*stride]= color4;
 +                *(uint32_t*)&dst[24+ y*stride]= color4;
 +                *(uint32_t*)&dst[28+ y*stride]= color4;
 +            }
 +        }else if(b_w==16){
 +            for(y=0; y < b_h; y++){
 +                *(uint32_t*)&dst[0 + y*stride]= color4;
 +                *(uint32_t*)&dst[4 + y*stride]= color4;
 +                *(uint32_t*)&dst[8 + y*stride]= color4;
 +                *(uint32_t*)&dst[12+ y*stride]= color4;
 +            }
 +        }else if(b_w==8){
 +            for(y=0; y < b_h; y++){
 +                *(uint32_t*)&dst[0 + y*stride]= color4;
 +                *(uint32_t*)&dst[4 + y*stride]= color4;
 +            }
 +        }else if(b_w==4){
 +            for(y=0; y < b_h; y++){
 +                *(uint32_t*)&dst[0 + y*stride]= color4;
 +            }
 +        }else{
 +            for(y=0; y < b_h; y++){
 +                for(x=0; x < b_w; x++){
 +                    dst[x + y*stride]= color;
 +                }
 +            }
 +        }
 +    }else{
 +        uint8_t *src= s->last_picture[block->ref]->data[plane_index];
 +        const int scale= plane_index ?  (2*s->mv_scale)>>s->chroma_h_shift : 2*s->mv_scale;
 +        int mx= block->mx*scale;
 +        int my= block->my*scale;
 +        const int dx= mx&15;
 +        const int dy= my&15;
 +        const int tab_index= 3 - (b_w>>2) + (b_w>>4);
 +        sx += (mx>>4) - (HTAPS_MAX/2-1);
 +        sy += (my>>4) - (HTAPS_MAX/2-1);
 +        src += sx + sy*stride;
 +        if(   (unsigned)sx >= FFMAX(w - b_w - (HTAPS_MAX-2), 0)
 +           || (unsigned)sy >= FFMAX(h - b_h - (HTAPS_MAX-2), 0)){
 +            s->vdsp.emulated_edge_mc(tmp + MB_SIZE, src,
 +                                     stride, stride,
 +                                     b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1,
 +                                     sx, sy, w, h);
 +            src= tmp + MB_SIZE;
 +        }
 +
 +        av_assert2(s->chroma_h_shift == s->chroma_v_shift); // only one mv_scale
 +
 +        av_assert2(b_w>1 && b_h>1);
 +        av_assert2((tab_index>=0 && tab_index<4) || b_w==32);
 +        if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1)) || !s->plane[plane_index].fast_mc )
 +            mc_block(&s->plane[plane_index], dst, src, stride, b_w, b_h, dx, dy);
 +        else if(b_w==32){
 +            int y;
 +            for(y=0; y<b_h; y+=16){
 +                s->h264qpel.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 3 + (y+3)*stride,stride);
 +                s->h264qpel.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 19 + (y+3)*stride,stride);
 +            }
 +        }else if(b_w==b_h)
 +            s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst,src + 3 + 3*stride,stride);
 +        else if(b_w==2*b_h){
 +            s->h264qpel.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst    ,src + 3       + 3*stride,stride);
 +            s->h264qpel.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 3 + b_h + 3*stride,stride);
 +        }else{
 +            av_assert2(2*b_w==b_h);
 +            s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst           ,src + 3 + 3*stride           ,stride);
 +            s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst+b_w*stride,src + 3 + 3*stride+b_w*stride,stride);
 +        }
 +    }
 +}
 +
 +#define mca(dx,dy,b_w)\
 +static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, ptrdiff_t stride, int h){\
 +    av_assert2(h==b_w);\
 +    mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, stride, b_w, b_w, dx, dy);\
 +}
 +
 +mca( 0, 0,16)
 +mca( 8, 0,16)
 +mca( 0, 8,16)
 +mca( 8, 8,16)
 +mca( 0, 0,8)
 +mca( 8, 0,8)
 +mca( 0, 8,8)
 +mca( 8, 8,8)
 +
 +av_cold int ff_snow_common_init(AVCodecContext *avctx){
 +    SnowContext *s = avctx->priv_data;
 +    int width, height;
 +    int i, j;
 +
 +    s->avctx= avctx;
 +    s->max_ref_frames=1; //just make sure it's not an invalid value in case of no initial keyframe
 +
 +    ff_dsputil_init(&s->dsp, avctx);
 +    ff_hpeldsp_init(&s->hdsp, avctx->flags);
 +    ff_videodsp_init(&s->vdsp, 8);
 +    ff_dwt_init(&s->dwt);
 +    ff_h264qpel_init(&s->h264qpel, 8);
 +
 +#define mcf(dx,dy)\
-     s->dsp.put_qpel_pixels_tab       [1][dy+dx/4]=\
-     s->dsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\
++    s->qdsp.put_qpel_pixels_tab       [0][dy+dx/4]=\
++    s->qdsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\
 +        s->h264qpel.put_h264_qpel_pixels_tab[0][dy+dx/4];\
++    s->qdsp.put_qpel_pixels_tab       [1][dy+dx/4]=\
++    s->qdsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\
 +        s->h264qpel.put_h264_qpel_pixels_tab[1][dy+dx/4];
 +
 +    mcf( 0, 0)
 +    mcf( 4, 0)
 +    mcf( 8, 0)
 +    mcf(12, 0)
 +    mcf( 0, 4)
 +    mcf( 4, 4)
 +    mcf( 8, 4)
 +    mcf(12, 4)
 +    mcf( 0, 8)
 +    mcf( 4, 8)
 +    mcf( 8, 8)
 +    mcf(12, 8)
 +    mcf( 0,12)
 +    mcf( 4,12)
 +    mcf( 8,12)
 +    mcf(12,12)
 +
 +#define mcfh(dx,dy)\
 +    s->hdsp.put_pixels_tab       [0][dy/4+dx/8]=\
 +    s->hdsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\
 +        mc_block_hpel ## dx ## dy ## 16;\
 +    s->hdsp.put_pixels_tab       [1][dy/4+dx/8]=\
 +    s->hdsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\
 +        mc_block_hpel ## dx ## dy ## 8;
 +
 +    mcfh(0, 0)
 +    mcfh(8, 0)
 +    mcfh(0, 8)
 +    mcfh(8, 8)
 +
 +    init_qexp();
 +
 +//    dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);
 +
 +    width= s->avctx->width;
 +    height= s->avctx->height;
 +
 +    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->spatial_idwt_buffer, width, height * sizeof(IDWTELEM), fail);
 +    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->spatial_dwt_buffer,  width, height * sizeof(DWTELEM),  fail); //FIXME this does not belong here
 +    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->temp_dwt_buffer,     width, sizeof(DWTELEM),  fail);
 +    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->temp_idwt_buffer,    width, sizeof(IDWTELEM), fail);
 +    FF_ALLOC_ARRAY_OR_GOTO(avctx,  s->run_buffer,          ((width + 1) >> 1), ((height + 1) >> 1) * sizeof(*s->run_buffer), fail);
 +
 +    for(i=0; i<MAX_REF_FRAMES; i++) {
 +        for(j=0; j<MAX_REF_FRAMES; j++)
 +            ff_scale_mv_ref[i][j] = 256*(i+1)/(j+1);
 +        s->last_picture[i] = av_frame_alloc();
 +        if (!s->last_picture[i])
 +            goto fail;
 +    }
 +
 +    s->mconly_picture = av_frame_alloc();
 +    s->current_picture = av_frame_alloc();
 +    if (!s->mconly_picture || !s->current_picture)
 +        goto fail;
 +
 +    return 0;
 +fail:
 +    return AVERROR(ENOMEM);
 +}
 +
 +int ff_snow_common_init_after_header(AVCodecContext *avctx) {
 +    SnowContext *s = avctx->priv_data;
 +    int plane_index, level, orientation;
 +    int ret, emu_buf_size;
 +
 +    if(!s->scratchbuf) {
 +        if ((ret = ff_get_buffer(s->avctx, s->mconly_picture,
 +                                 AV_GET_BUFFER_FLAG_REF)) < 0)
 +            return ret;
 +        FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->scratchbuf, FFMAX(s->mconly_picture->linesize[0], 2*avctx->width+256), 7*MB_SIZE, fail);
 +        emu_buf_size = FFMAX(s->mconly_picture->linesize[0], 2*avctx->width+256) * (2 * MB_SIZE + HTAPS_MAX - 1);
 +        FF_ALLOC_OR_GOTO(avctx, s->emu_edge_buffer, emu_buf_size, fail);
 +    }
 +
 +    if(s->mconly_picture->format != avctx->pix_fmt) {
 +        av_log(avctx, AV_LOG_ERROR, "pixel format changed\n");
 +        return AVERROR_INVALIDDATA;
 +    }
 +
 +    for(plane_index=0; plane_index < s->nb_planes; plane_index++){
 +        int w= s->avctx->width;
 +        int h= s->avctx->height;
 +
 +        if(plane_index){
 +            w>>= s->chroma_h_shift;
 +            h>>= s->chroma_v_shift;
 +        }
 +        s->plane[plane_index].width = w;
 +        s->plane[plane_index].height= h;
 +
 +        for(level=s->spatial_decomposition_count-1; level>=0; level--){
 +            for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +                SubBand *b= &s->plane[plane_index].band[level][orientation];
 +
 +                b->buf= s->spatial_dwt_buffer;
 +                b->level= level;
 +                b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level);
 +                b->width = (w + !(orientation&1))>>1;
 +                b->height= (h + !(orientation>1))>>1;
 +
 +                b->stride_line = 1 << (s->spatial_decomposition_count - level);
 +                b->buf_x_offset = 0;
 +                b->buf_y_offset = 0;
 +
 +                if(orientation&1){
 +                    b->buf += (w+1)>>1;
 +                    b->buf_x_offset = (w+1)>>1;
 +                }
 +                if(orientation>1){
 +                    b->buf += b->stride>>1;
 +                    b->buf_y_offset = b->stride_line >> 1;
 +                }
 +                b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer);
 +
 +                if(level)
 +                    b->parent= &s->plane[plane_index].band[level-1][orientation];
 +                //FIXME avoid this realloc
 +                av_freep(&b->x_coeff);
 +                b->x_coeff=av_mallocz_array(((b->width+1) * b->height+1), sizeof(x_and_coeff));
 +                if (!b->x_coeff)
 +                    goto fail;
 +            }
 +            w= (w+1)>>1;
 +            h= (h+1)>>1;
 +        }
 +    }
 +
 +    return 0;
 +fail:
 +    return AVERROR(ENOMEM);
 +}
 +
 +#define USE_HALFPEL_PLANE 0
 +
 +static int halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *frame){
 +    int p,x,y;
 +
 +    for(p=0; p < s->nb_planes; p++){
 +        int is_chroma= !!p;
 +        int w= is_chroma ? s->avctx->width >>s->chroma_h_shift : s->avctx->width;
 +        int h= is_chroma ? s->avctx->height>>s->chroma_v_shift : s->avctx->height;
 +        int ls= frame->linesize[p];
 +        uint8_t *src= frame->data[p];
 +
 +        halfpel[1][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
 +        halfpel[2][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
 +        halfpel[3][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
 +        if (!halfpel[1][p] || !halfpel[2][p] || !halfpel[3][p])
 +            return AVERROR(ENOMEM);
 +
 +        halfpel[0][p]= src;
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int i= y*ls + x;
 +
 +                halfpel[1][p][i]= (20*(src[i] + src[i+1]) - 5*(src[i-1] + src[i+2]) + (src[i-2] + src[i+3]) + 16 )>>5;
 +            }
 +        }
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int i= y*ls + x;
 +
 +                halfpel[2][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
 +            }
 +        }
 +        src= halfpel[1][p];
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int i= y*ls + x;
 +
 +                halfpel[3][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
 +            }
 +        }
 +
 +//FIXME border!
 +    }
 +    return 0;
 +}
 +
 +void ff_snow_release_buffer(AVCodecContext *avctx)
 +{
 +    SnowContext *s = avctx->priv_data;
 +    int i;
 +
 +    if(s->last_picture[s->max_ref_frames-1]->data[0]){
 +        av_frame_unref(s->last_picture[s->max_ref_frames-1]);
 +        for(i=0; i<9; i++)
 +            if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3])
 +                av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture->linesize[i%3]));
 +    }
 +}
 +
 +int ff_snow_frame_start(SnowContext *s){
 +   AVFrame *tmp;
 +   int i, ret;
 +   int w= s->avctx->width; //FIXME round up to x16 ?
 +   int h= s->avctx->height;
 +
 +    if (s->current_picture->data[0] && !(s->avctx->flags&CODEC_FLAG_EMU_EDGE)) {
 +        s->dsp.draw_edges(s->current_picture->data[0],
 +                          s->current_picture->linesize[0], w   , h   ,
 +                          EDGE_WIDTH  , EDGE_WIDTH  , EDGE_TOP | EDGE_BOTTOM);
 +        if (s->current_picture->data[2]) {
 +            s->dsp.draw_edges(s->current_picture->data[1],
 +                            s->current_picture->linesize[1], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
 +                            EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
 +            s->dsp.draw_edges(s->current_picture->data[2],
 +                            s->current_picture->linesize[2], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
 +                            EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
 +        }
 +    }
 +
 +    ff_snow_release_buffer(s->avctx);
 +
 +    tmp= s->last_picture[s->max_ref_frames-1];
 +    for(i=s->max_ref_frames-1; i>0; i--)
 +        s->last_picture[i] = s->last_picture[i-1];
 +    memmove(s->halfpel_plane+1, s->halfpel_plane, (s->max_ref_frames-1)*sizeof(void*)*4*4);
 +    if(USE_HALFPEL_PLANE && s->current_picture->data[0]) {
 +        if((ret = halfpel_interpol(s, s->halfpel_plane[0], s->current_picture)) < 0)
 +            return ret;
 +    }
 +    s->last_picture[0] = s->current_picture;
 +    s->current_picture = tmp;
 +
 +    if(s->keyframe){
 +        s->ref_frames= 0;
 +    }else{
 +        int i;
 +        for(i=0; i<s->max_ref_frames && s->last_picture[i]->data[0]; i++)
 +            if(i && s->last_picture[i-1]->key_frame)
 +                break;
 +        s->ref_frames= i;
 +        if(s->ref_frames==0){
 +            av_log(s->avctx,AV_LOG_ERROR, "No reference frames\n");
 +            return -1;
 +        }
 +    }
 +
 +    if ((ret = ff_get_buffer(s->avctx, s->current_picture, AV_GET_BUFFER_FLAG_REF)) < 0)
 +        return ret;
 +
 +    s->current_picture->key_frame= s->keyframe;
 +
 +    return 0;
 +}
 +
 +av_cold void ff_snow_common_end(SnowContext *s)
 +{
 +    int plane_index, level, orientation, i;
 +
 +    av_freep(&s->spatial_dwt_buffer);
 +    av_freep(&s->temp_dwt_buffer);
 +    av_freep(&s->spatial_idwt_buffer);
 +    av_freep(&s->temp_idwt_buffer);
 +    av_freep(&s->run_buffer);
 +
 +    s->m.me.temp= NULL;
 +    av_freep(&s->m.me.scratchpad);
 +    av_freep(&s->m.me.map);
 +    av_freep(&s->m.me.score_map);
 +    av_freep(&s->m.obmc_scratchpad);
 +
 +    av_freep(&s->block);
 +    av_freep(&s->scratchbuf);
 +    av_freep(&s->emu_edge_buffer);
 +
 +    for(i=0; i<MAX_REF_FRAMES; i++){
 +        av_freep(&s->ref_mvs[i]);
 +        av_freep(&s->ref_scores[i]);
 +        if(s->last_picture[i]->data[0]) {
 +            av_assert0(s->last_picture[i]->data[0] != s->current_picture->data[0]);
 +        }
 +        av_frame_free(&s->last_picture[i]);
 +    }
 +
 +    for(plane_index=0; plane_index < s->nb_planes; plane_index++){
 +        for(level=s->spatial_decomposition_count-1; level>=0; level--){
 +            for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +                SubBand *b= &s->plane[plane_index].band[level][orientation];
 +
 +                av_freep(&b->x_coeff);
 +            }
 +        }
 +    }
 +    av_frame_free(&s->mconly_picture);
 +    av_frame_free(&s->current_picture);
 +}
index 60b0062,0000000..d0e5c5b
mode 100644,000000..100644
--- /dev/null
@@@ -1,706 -1,0 +1,708 @@@
 +/*
 + * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
 + * Copyright (C) 2006 Robert Edele <yartrebo@earthlink.net>
 + *
 + * This file is part of FFmpeg.
 + *
 + * 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.
 + *
 + * 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 FFmpeg; if not, write to the Free Software
 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 + */
 +
 +#ifndef AVCODEC_SNOW_H
 +#define AVCODEC_SNOW_H
 +
 +#include "dsputil.h"
 +#include "hpeldsp.h"
++#include "qpeldsp.h"
 +#include "snow_dwt.h"
 +
 +#include "rangecoder.h"
 +#include "mathops.h"
 +#include "mpegvideo.h"
 +#include "h264qpel.h"
 +
 +#define MID_STATE 128
 +
 +#define MAX_PLANES 4
 +#define QSHIFT 5
 +#define QROOT (1<<QSHIFT)
 +#define LOSSLESS_QLOG -128
 +#define FRAC_BITS 4
 +#define MAX_REF_FRAMES 8
 +
 +#define LOG2_OBMC_MAX 8
 +#define OBMC_MAX (1<<(LOG2_OBMC_MAX))
 +typedef struct BlockNode{
 +    int16_t mx;
 +    int16_t my;
 +    uint8_t ref;
 +    uint8_t color[3];
 +    uint8_t type;
 +//#define TYPE_SPLIT    1
 +#define BLOCK_INTRA   1
 +#define BLOCK_OPT     2
 +//#define TYPE_NOCOLOR  4
 +    uint8_t level; //FIXME merge into type?
 +}BlockNode;
 +
 +static const BlockNode null_block= { //FIXME add border maybe
 +    .color= {128,128,128},
 +    .mx= 0,
 +    .my= 0,
 +    .ref= 0,
 +    .type= 0,
 +    .level= 0,
 +};
 +
 +#define LOG2_MB_SIZE 4
 +#define MB_SIZE (1<<LOG2_MB_SIZE)
 +#define ENCODER_EXTRA_BITS 4
 +#define HTAPS_MAX 8
 +
 +typedef struct x_and_coeff{
 +    int16_t x;
 +    uint16_t coeff;
 +} x_and_coeff;
 +
 +typedef struct SubBand{
 +    int level;
 +    int stride;
 +    int width;
 +    int height;
 +    int qlog;        ///< log(qscale)/log[2^(1/6)]
 +    DWTELEM *buf;
 +    IDWTELEM *ibuf;
 +    int buf_x_offset;
 +    int buf_y_offset;
 +    int stride_line; ///< Stride measured in lines, not pixels.
 +    x_and_coeff * x_coeff;
 +    struct SubBand *parent;
 +    uint8_t state[/*7*2*/ 7 + 512][32];
 +}SubBand;
 +
 +typedef struct Plane{
 +    int width;
 +    int height;
 +    SubBand band[MAX_DECOMPOSITIONS][4];
 +
 +    int htaps;
 +    int8_t hcoeff[HTAPS_MAX/2];
 +    int diag_mc;
 +    int fast_mc;
 +
 +    int last_htaps;
 +    int8_t last_hcoeff[HTAPS_MAX/2];
 +    int last_diag_mc;
 +}Plane;
 +
 +typedef struct SnowContext{
 +    AVClass *class;
 +    AVCodecContext *avctx;
 +    RangeCoder c;
 +    DSPContext dsp;
 +    HpelDSPContext hdsp;
++    QpelDSPContext qdsp;
 +    VideoDSPContext vdsp;
 +    H264QpelContext h264qpel;
 +    SnowDWTContext dwt;
 +    AVFrame *new_picture;
 +    AVFrame *input_picture;              ///< new_picture with the internal linesizes
 +    AVFrame *current_picture;
 +    AVFrame *last_picture[MAX_REF_FRAMES];
 +    uint8_t *halfpel_plane[MAX_REF_FRAMES][4][4];
 +    AVFrame *mconly_picture;
 +//     uint8_t q_context[16];
 +    uint8_t header_state[32];
 +    uint8_t block_state[128 + 32*128];
 +    int keyframe;
 +    int always_reset;
 +    int version;
 +    int spatial_decomposition_type;
 +    int last_spatial_decomposition_type;
 +    int temporal_decomposition_type;
 +    int spatial_decomposition_count;
 +    int last_spatial_decomposition_count;
 +    int temporal_decomposition_count;
 +    int max_ref_frames;
 +    int ref_frames;
 +    int16_t (*ref_mvs[MAX_REF_FRAMES])[2];
 +    uint32_t *ref_scores[MAX_REF_FRAMES];
 +    DWTELEM *spatial_dwt_buffer;
 +    DWTELEM *temp_dwt_buffer;
 +    IDWTELEM *spatial_idwt_buffer;
 +    IDWTELEM *temp_idwt_buffer;
 +    int *run_buffer;
 +    int colorspace_type;
 +    int chroma_h_shift;
 +    int chroma_v_shift;
 +    int spatial_scalability;
 +    int qlog;
 +    int last_qlog;
 +    int lambda;
 +    int lambda2;
 +    int pass1_rc;
 +    int mv_scale;
 +    int last_mv_scale;
 +    int qbias;
 +    int last_qbias;
 +#define QBIAS_SHIFT 3
 +    int b_width;
 +    int b_height;
 +    int block_max_depth;
 +    int last_block_max_depth;
 +    int nb_planes;
 +    Plane plane[MAX_PLANES];
 +    BlockNode *block;
 +#define ME_CACHE_SIZE 1024
 +    unsigned me_cache[ME_CACHE_SIZE];
 +    unsigned me_cache_generation;
 +    slice_buffer sb;
 +    int memc_only;
 +    int no_bitstream;
 +
 +    MpegEncContext m; // needed for motion estimation, should not be used for anything else, the idea is to eventually make the motion estimation independent of MpegEncContext, so this will be removed then (FIXME/XXX)
 +
 +    uint8_t *scratchbuf;
 +    uint8_t *emu_edge_buffer;
 +}SnowContext;
 +
 +/* Tables */
 +extern const uint8_t * const ff_obmc_tab[4];
 +extern uint8_t ff_qexp[QROOT];
 +extern int ff_scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES];
 +
 +/* C bits used by mmx/sse2/altivec */
 +
 +static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){
 +    (*i) = (width) - 2;
 +
 +    if (width & 1){
 +        low[(*i)+1] = low[((*i)+1)>>1];
 +        (*i)--;
 +    }
 +}
 +
 +static av_always_inline void snow_interleave_line_footer(int * i, IDWTELEM * low, IDWTELEM * high){
 +    for (; (*i)>=0; (*i)-=2){
 +        low[(*i)+1] = high[(*i)>>1];
 +        low[*i] = low[(*i)>>1];
 +    }
 +}
 +
 +static av_always_inline void snow_horizontal_compose_lift_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w, int lift_high, int mul, int add, int shift){
 +    for(; i<w; i++){
 +        dst[i] = src[i] - ((mul * (ref[i] + ref[i + 1]) + add) >> shift);
 +    }
 +
 +    if((width^lift_high)&1){
 +        dst[w] = src[w] - ((mul * 2 * ref[w] + add) >> shift);
 +    }
 +}
 +
 +static av_always_inline void snow_horizontal_compose_liftS_lead_out(int i, IDWTELEM * dst, IDWTELEM * src, IDWTELEM * ref, int width, int w){
 +        for(; i<w; i++){
 +            dst[i] = src[i] + ((ref[i] + ref[(i+1)]+W_BO + 4 * src[i]) >> W_BS);
 +        }
 +
 +        if(width&1){
 +            dst[w] = src[w] + ((2 * ref[w] + W_BO + 4 * src[w]) >> W_BS);
 +        }
 +}
 +
 +/* common code */
 +
 +int ff_snow_common_init(AVCodecContext *avctx);
 +int ff_snow_common_init_after_header(AVCodecContext *avctx);
 +void ff_snow_common_end(SnowContext *s);
 +void ff_snow_release_buffer(AVCodecContext *avctx);
 +void ff_snow_reset_contexts(SnowContext *s);
 +int ff_snow_alloc_blocks(SnowContext *s);
 +int ff_snow_frame_start(SnowContext *s);
 +void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride,
 +                     int sx, int sy, int b_w, int b_h, BlockNode *block,
 +                     int plane_index, int w, int h);
 +/* common inline functions */
 +//XXX doublecheck all of them should stay inlined
 +
 +static inline void snow_set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
 +    const int w= s->b_width << s->block_max_depth;
 +    const int rem_depth= s->block_max_depth - level;
 +    const int index= (x + y*w) << rem_depth;
 +    const int block_w= 1<<rem_depth;
 +    BlockNode block;
 +    int i,j;
 +
 +    block.color[0]= l;
 +    block.color[1]= cb;
 +    block.color[2]= cr;
 +    block.mx= mx;
 +    block.my= my;
 +    block.ref= ref;
 +    block.type= type;
 +    block.level= level;
 +
 +    for(j=0; j<block_w; j++){
 +        for(i=0; i<block_w; i++){
 +            s->block[index + i + j*w]= block;
 +        }
 +    }
 +}
 +
 +static inline void pred_mv(SnowContext *s, int *mx, int *my, int ref,
 +                           const BlockNode *left, const BlockNode *top, const BlockNode *tr){
 +    if(s->ref_frames == 1){
 +        *mx = mid_pred(left->mx, top->mx, tr->mx);
 +        *my = mid_pred(left->my, top->my, tr->my);
 +    }else{
 +        const int *scale = ff_scale_mv_ref[ref];
 +        *mx = mid_pred((left->mx * scale[left->ref] + 128) >>8,
 +                       (top ->mx * scale[top ->ref] + 128) >>8,
 +                       (tr  ->mx * scale[tr  ->ref] + 128) >>8);
 +        *my = mid_pred((left->my * scale[left->ref] + 128) >>8,
 +                       (top ->my * scale[top ->ref] + 128) >>8,
 +                       (tr  ->my * scale[tr  ->ref] + 128) >>8);
 +    }
 +}
 +
 +static av_always_inline int same_block(BlockNode *a, BlockNode *b){
 +    if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){
 +        return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2]));
 +    }else{
 +        return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA));
 +    }
 +}
 +
 +//FIXME name cleanup (b_w, block_w, b_width stuff)
 +//XXX should we really inline it?
 +static av_always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, IDWTELEM *dst, uint8_t *dst8, const uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int offset_dst, int plane_index){
 +    const int b_width = s->b_width  << s->block_max_depth;
 +    const int b_height= s->b_height << s->block_max_depth;
 +    const int b_stride= b_width;
 +    BlockNode *lt= &s->block[b_x + b_y*b_stride];
 +    BlockNode *rt= lt+1;
 +    BlockNode *lb= lt+b_stride;
 +    BlockNode *rb= lb+1;
 +    uint8_t *block[4];
 +    int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride;
 +    uint8_t *tmp = s->scratchbuf;
 +    uint8_t *ptmp;
 +    int x,y;
 +
 +    if(b_x<0){
 +        lt= rt;
 +        lb= rb;
 +    }else if(b_x + 1 >= b_width){
 +        rt= lt;
 +        rb= lb;
 +    }
 +    if(b_y<0){
 +        lt= lb;
 +        rt= rb;
 +    }else if(b_y + 1 >= b_height){
 +        lb= lt;
 +        rb= rt;
 +    }
 +
 +    if(src_x<0){ //FIXME merge with prev & always round internal width up to *16
 +        obmc -= src_x;
 +        b_w += src_x;
 +        if(!sliced && !offset_dst)
 +            dst -= src_x;
 +        src_x=0;
 +    }
 +    if(src_x + b_w > w){
 +        b_w = w - src_x;
 +    }
 +    if(src_y<0){
 +        obmc -= src_y*obmc_stride;
 +        b_h += src_y;
 +        if(!sliced && !offset_dst)
 +            dst -= src_y*dst_stride;
 +        src_y=0;
 +    }
 +    if(src_y + b_h> h){
 +        b_h = h - src_y;
 +    }
 +
 +    if(b_w<=0 || b_h<=0) return;
 +
 +    av_assert2(src_stride > 2*MB_SIZE + 5);
 +
 +    if(!sliced && offset_dst)
 +        dst += src_x + src_y*dst_stride;
 +    dst8+= src_x + src_y*src_stride;
 +//    src += src_x + src_y*src_stride;
 +
 +    ptmp= tmp + 3*tmp_step;
 +    block[0]= ptmp;
 +    ptmp+=tmp_step;
 +    ff_snow_pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h);
 +
 +    if(same_block(lt, rt)){
 +        block[1]= block[0];
 +    }else{
 +        block[1]= ptmp;
 +        ptmp+=tmp_step;
 +        ff_snow_pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h);
 +    }
 +
 +    if(same_block(lt, lb)){
 +        block[2]= block[0];
 +    }else if(same_block(rt, lb)){
 +        block[2]= block[1];
 +    }else{
 +        block[2]= ptmp;
 +        ptmp+=tmp_step;
 +        ff_snow_pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h);
 +    }
 +
 +    if(same_block(lt, rb) ){
 +        block[3]= block[0];
 +    }else if(same_block(rt, rb)){
 +        block[3]= block[1];
 +    }else if(same_block(lb, rb)){
 +        block[3]= block[2];
 +    }else{
 +        block[3]= ptmp;
 +        ff_snow_pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h);
 +    }
 +    if(sliced){
 +        s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
 +    }else{
 +        for(y=0; y<b_h; y++){
 +            //FIXME ugly misuse of obmc_stride
 +            const uint8_t *obmc1= obmc + y*obmc_stride;
 +            const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
 +            const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
 +            const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
 +            for(x=0; x<b_w; x++){
 +                int v=   obmc1[x] * block[3][x + y*src_stride]
 +                        +obmc2[x] * block[2][x + y*src_stride]
 +                        +obmc3[x] * block[1][x + y*src_stride]
 +                        +obmc4[x] * block[0][x + y*src_stride];
 +
 +                v <<= 8 - LOG2_OBMC_MAX;
 +                if(FRAC_BITS != 8){
 +                    v >>= 8 - FRAC_BITS;
 +                }
 +                if(add){
 +                    v += dst[x + y*dst_stride];
 +                    v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
 +                    if(v&(~255)) v= ~(v>>31);
 +                    dst8[x + y*src_stride] = v;
 +                }else{
 +                    dst[x + y*dst_stride] -= v;
 +                }
 +            }
 +        }
 +    }
 +}
 +
 +static av_always_inline void predict_slice(SnowContext *s, IDWTELEM *buf, int plane_index, int add, int mb_y){
 +    Plane *p= &s->plane[plane_index];
 +    const int mb_w= s->b_width  << s->block_max_depth;
 +    const int mb_h= s->b_height << s->block_max_depth;
 +    int x, y, mb_x;
 +    int block_size = MB_SIZE >> s->block_max_depth;
 +    int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
 +    int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
 +    const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
 +    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
 +    int ref_stride= s->current_picture->linesize[plane_index];
 +    uint8_t *dst8= s->current_picture->data[plane_index];
 +    int w= p->width;
 +    int h= p->height;
 +    av_assert2(s->chroma_h_shift == s->chroma_v_shift); // obmc params assume squares
 +    if(s->keyframe || (s->avctx->debug&512)){
 +        if(mb_y==mb_h)
 +            return;
 +
 +        if(add){
 +            for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){
 +                for(x=0; x<w; x++){
 +                    int v= buf[x + y*w] + (128<<FRAC_BITS) + (1<<(FRAC_BITS-1));
 +                    v >>= FRAC_BITS;
 +                    if(v&(~255)) v= ~(v>>31);
 +                    dst8[x + y*ref_stride]= v;
 +                }
 +            }
 +        }else{
 +            for(y=block_h*mb_y; y<FFMIN(h,block_h*(mb_y+1)); y++){
 +                for(x=0; x<w; x++){
 +                    buf[x + y*w]-= 128<<FRAC_BITS;
 +                }
 +            }
 +        }
 +
 +        return;
 +    }
 +
 +    for(mb_x=0; mb_x<=mb_w; mb_x++){
 +        add_yblock(s, 0, NULL, buf, dst8, obmc,
 +                   block_w*mb_x - block_w/2,
 +                   block_h*mb_y - block_h/2,
 +                   block_w, block_h,
 +                   w, h,
 +                   w, ref_stride, obmc_stride,
 +                   mb_x - 1, mb_y - 1,
 +                   add, 1, plane_index);
 +    }
 +}
 +
 +static av_always_inline void predict_plane(SnowContext *s, IDWTELEM *buf, int plane_index, int add){
 +    const int mb_h= s->b_height << s->block_max_depth;
 +    int mb_y;
 +    for(mb_y=0; mb_y<=mb_h; mb_y++)
 +        predict_slice(s, buf, plane_index, add, mb_y);
 +}
 +
 +static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){
 +    const int w= s->b_width << s->block_max_depth;
 +    const int rem_depth= s->block_max_depth - level;
 +    const int index= (x + y*w) << rem_depth;
 +    const int block_w= 1<<rem_depth;
 +    const int block_h= 1<<rem_depth; //FIXME "w!=h"
 +    BlockNode block;
 +    int i,j;
 +
 +    block.color[0]= l;
 +    block.color[1]= cb;
 +    block.color[2]= cr;
 +    block.mx= mx;
 +    block.my= my;
 +    block.ref= ref;
 +    block.type= type;
 +    block.level= level;
 +
 +    for(j=0; j<block_h; j++){
 +        for(i=0; i<block_w; i++){
 +            s->block[index + i + j*w]= block;
 +        }
 +    }
 +}
 +
 +static inline void init_ref(MotionEstContext *c, uint8_t *src[3], uint8_t *ref[3], uint8_t *ref2[3], int x, int y, int ref_index){
 +    SnowContext *s = c->avctx->priv_data;
 +    const int offset[3]= {
 +          y*c->  stride + x,
 +        ((y*c->uvstride + x)>>s->chroma_h_shift),
 +        ((y*c->uvstride + x)>>s->chroma_h_shift),
 +    };
 +    int i;
 +    for(i=0; i<3; i++){
 +        c->src[0][i]= src [i];
 +        c->ref[0][i]= ref [i] + offset[i];
 +    }
 +    av_assert2(!ref_index);
 +}
 +
 +
 +/* bitstream functions */
 +
 +extern const int8_t ff_quant3bA[256];
 +
 +#define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
 +
 +static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
 +    int i;
 +
 +    if(v){
 +        const int a= FFABS(v);
 +        const int e= av_log2(a);
 +        const int el= FFMIN(e, 10);
 +        put_rac(c, state+0, 0);
 +
 +        for(i=0; i<el; i++){
 +            put_rac(c, state+1+i, 1);  //1..10
 +        }
 +        for(; i<e; i++){
 +            put_rac(c, state+1+9, 1);  //1..10
 +        }
 +        put_rac(c, state+1+FFMIN(i,9), 0);
 +
 +        for(i=e-1; i>=el; i--){
 +            put_rac(c, state+22+9, (a>>i)&1); //22..31
 +        }
 +        for(; i>=0; i--){
 +            put_rac(c, state+22+i, (a>>i)&1); //22..31
 +        }
 +
 +        if(is_signed)
 +            put_rac(c, state+11 + el, v < 0); //11..21
 +    }else{
 +        put_rac(c, state+0, 1);
 +    }
 +}
 +
 +static inline int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){
 +    if(get_rac(c, state+0))
 +        return 0;
 +    else{
 +        int i, e, a;
 +        e= 0;
 +        while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10
 +            e++;
 +        }
 +
 +        a= 1;
 +        for(i=e-1; i>=0; i--){
 +            a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31
 +        }
 +
 +        e= -(is_signed && get_rac(c, state+11 + FFMIN(e,10))); //11..21
 +        return (a^e)-e;
 +    }
 +}
 +
 +static inline void put_symbol2(RangeCoder *c, uint8_t *state, int v, int log2){
 +    int i;
 +    int r= log2>=0 ? 1<<log2 : 1;
 +
 +    av_assert2(v>=0);
 +    av_assert2(log2>=-4);
 +
 +    while(v >= r){
 +        put_rac(c, state+4+log2, 1);
 +        v -= r;
 +        log2++;
 +        if(log2>0) r+=r;
 +    }
 +    put_rac(c, state+4+log2, 0);
 +
 +    for(i=log2-1; i>=0; i--){
 +        put_rac(c, state+31-i, (v>>i)&1);
 +    }
 +}
 +
 +static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
 +    int i;
 +    int r= log2>=0 ? 1<<log2 : 1;
 +    int v=0;
 +
 +    av_assert2(log2>=-4);
 +
 +    while(log2<28 && get_rac(c, state+4+log2)){
 +        v+= r;
 +        log2++;
 +        if(log2>0) r+=r;
 +    }
 +
 +    for(i=log2-1; i>=0; i--){
 +        v+= get_rac(c, state+31-i)<<i;
 +    }
 +
 +    return v;
 +}
 +
 +static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){
 +    const int w= b->width;
 +    const int h= b->height;
 +    int x,y;
 +
 +    int run, runs;
 +    x_and_coeff *xc= b->x_coeff;
 +    x_and_coeff *prev_xc= NULL;
 +    x_and_coeff *prev2_xc= xc;
 +    x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
 +    x_and_coeff *prev_parent_xc= parent_xc;
 +
 +    runs= get_symbol2(&s->c, b->state[30], 0);
 +    if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
 +    else           run= INT_MAX;
 +
 +    for(y=0; y<h; y++){
 +        int v=0;
 +        int lt=0, t=0, rt=0;
 +
 +        if(y && prev_xc->x == 0){
 +            rt= prev_xc->coeff;
 +        }
 +        for(x=0; x<w; x++){
 +            int p=0;
 +            const int l= v;
 +
 +            lt= t; t= rt;
 +
 +            if(y){
 +                if(prev_xc->x <= x)
 +                    prev_xc++;
 +                if(prev_xc->x == x + 1)
 +                    rt= prev_xc->coeff;
 +                else
 +                    rt=0;
 +            }
 +            if(parent_xc){
 +                if(x>>1 > parent_xc->x){
 +                    parent_xc++;
 +                }
 +                if(x>>1 == parent_xc->x){
 +                    p= parent_xc->coeff;
 +                }
 +            }
 +            if(/*ll|*/l|lt|t|rt|p){
 +                int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
 +
 +                v=get_rac(&s->c, &b->state[0][context]);
 +                if(v){
 +                    v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
 +                    v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l&0xFF] + 3*ff_quant3bA[t&0xFF]]);
 +
 +                    xc->x=x;
 +                    (xc++)->coeff= v;
 +                }
 +            }else{
 +                if(!run){
 +                    if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
 +                    else           run= INT_MAX;
 +                    v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
 +                    v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
 +
 +                    xc->x=x;
 +                    (xc++)->coeff= v;
 +                }else{
 +                    int max_run;
 +                    run--;
 +                    v=0;
 +                    av_assert2(run >= 0);
 +                    if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
 +                    else  max_run= FFMIN(run, w-x-1);
 +                    if(parent_xc)
 +                        max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
 +                    av_assert2(max_run >= 0 && max_run <= run);
 +
 +                    x+= max_run;
 +                    run-= max_run;
 +                }
 +            }
 +        }
 +        (xc++)->x= w+1; //end marker
 +        prev_xc= prev2_xc;
 +        prev2_xc= xc;
 +
 +        if(parent_xc){
 +            if(y&1){
 +                while(parent_xc->x != parent->width+1)
 +                    parent_xc++;
 +                parent_xc++;
 +                prev_parent_xc= parent_xc;
 +            }else{
 +                parent_xc= prev_parent_xc;
 +            }
 +        }
 +    }
 +
 +    (xc++)->x= w+1; //end marker
 +}
 +
 +#endif /* AVCODEC_SNOW_H */
index 5067a36,0000000..167b205
mode 100644,000000..100644
--- /dev/null
@@@ -1,2005 -1,0 +1,2006 @@@
 +/*
 + * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
 + *
 + * This file is part of FFmpeg.
 + *
 + * 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.
 + *
 + * 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 FFmpeg; if not, write to the Free Software
 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 + */
 +
 +#include "libavutil/intmath.h"
 +#include "libavutil/log.h"
 +#include "libavutil/opt.h"
 +#include "avcodec.h"
 +#include "dsputil.h"
 +#include "internal.h"
 +#include "snow_dwt.h"
 +#include "snow.h"
 +
 +#include "rangecoder.h"
 +#include "mathops.h"
 +
 +#include "mpegvideo.h"
 +#include "h263.h"
 +
 +static av_cold int encode_init(AVCodecContext *avctx)
 +{
 +    SnowContext *s = avctx->priv_data;
 +    int plane_index, ret;
 +
 +    if(avctx->prediction_method == DWT_97
 +       && (avctx->flags & CODEC_FLAG_QSCALE)
 +       && avctx->global_quality == 0){
 +        av_log(avctx, AV_LOG_ERROR, "The 9/7 wavelet is incompatible with lossless mode.\n");
 +        return -1;
 +    }
 +
 +    s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type
 +
 +    s->mv_scale       = (avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
 +    s->block_max_depth= (avctx->flags & CODEC_FLAG_4MV ) ? 1 : 0;
 +
 +    for(plane_index=0; plane_index<3; plane_index++){
 +        s->plane[plane_index].diag_mc= 1;
 +        s->plane[plane_index].htaps= 6;
 +        s->plane[plane_index].hcoeff[0]=  40;
 +        s->plane[plane_index].hcoeff[1]= -10;
 +        s->plane[plane_index].hcoeff[2]=   2;
 +        s->plane[plane_index].fast_mc= 1;
 +    }
 +
 +    if ((ret = ff_snow_common_init(avctx)) < 0) {
 +        ff_snow_common_end(avctx->priv_data);
 +        return ret;
 +    }
 +    ff_snow_alloc_blocks(s);
 +
 +    s->version=0;
 +
 +    s->m.avctx   = avctx;
 +    s->m.flags   = avctx->flags;
 +    s->m.bit_rate= avctx->bit_rate;
 +
 +    s->m.me.temp      =
 +    s->m.me.scratchpad= av_mallocz_array((avctx->width+64), 2*16*2*sizeof(uint8_t));
 +    s->m.me.map       = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
 +    s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
 +    s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t));
 +    if (!s->m.me.scratchpad || !s->m.me.map || !s->m.me.score_map || !s->m.obmc_scratchpad)
 +        return AVERROR(ENOMEM);
 +
 +    ff_h263_encode_init(&s->m); //mv_penalty
 +
 +    s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1);
 +
 +    if(avctx->flags&CODEC_FLAG_PASS1){
 +        if(!avctx->stats_out)
 +            avctx->stats_out = av_mallocz(256);
 +
 +        if (!avctx->stats_out)
 +            return AVERROR(ENOMEM);
 +    }
 +    if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){
 +        if(ff_rate_control_init(&s->m) < 0)
 +            return -1;
 +    }
 +    s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2));
 +
 +    switch(avctx->pix_fmt){
 +    case AV_PIX_FMT_YUV444P:
 +//    case AV_PIX_FMT_YUV422P:
 +    case AV_PIX_FMT_YUV420P:
 +//    case AV_PIX_FMT_YUV411P:
 +    case AV_PIX_FMT_YUV410P:
 +        s->nb_planes = 3;
 +        s->colorspace_type= 0;
 +        break;
 +    case AV_PIX_FMT_GRAY8:
 +        s->nb_planes = 1;
 +        s->colorspace_type = 1;
 +        break;
 +/*    case AV_PIX_FMT_RGB32:
 +        s->colorspace= 1;
 +        break;*/
 +    default:
 +        av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
 +        return -1;
 +    }
 +    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
 +
 +    ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp);
 +    ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
 +
 +    s->input_picture = av_frame_alloc();
 +    if (!s->input_picture)
 +        return AVERROR(ENOMEM);
 +    if ((ret = ff_get_buffer(s->avctx, s->input_picture, AV_GET_BUFFER_FLAG_REF)) < 0)
 +        return ret;
 +
 +    if(s->avctx->me_method == ME_ITER){
 +        int i;
 +        int size= s->b_width * s->b_height << 2*s->block_max_depth;
 +        for(i=0; i<s->max_ref_frames; i++){
 +            s->ref_mvs[i]= av_mallocz_array(size, sizeof(int16_t[2]));
 +            s->ref_scores[i]= av_mallocz_array(size, sizeof(uint32_t));
 +            if (!s->ref_mvs[i] || !s->ref_scores[i])
 +                return AVERROR(ENOMEM);
 +        }
 +    }
 +
 +    return 0;
 +}
 +
 +//near copy & paste from dsputil, FIXME
 +static int pix_sum(uint8_t * pix, int line_size, int w, int h)
 +{
 +    int s, i, j;
 +
 +    s = 0;
 +    for (i = 0; i < h; i++) {
 +        for (j = 0; j < w; j++) {
 +            s += pix[0];
 +            pix ++;
 +        }
 +        pix += line_size - w;
 +    }
 +    return s;
 +}
 +
 +//near copy & paste from dsputil, FIXME
 +static int pix_norm1(uint8_t * pix, int line_size, int w)
 +{
 +    int s, i, j;
 +    uint32_t *sq = ff_square_tab + 256;
 +
 +    s = 0;
 +    for (i = 0; i < w; i++) {
 +        for (j = 0; j < w; j ++) {
 +            s += sq[pix[0]];
 +            pix ++;
 +        }
 +        pix += line_size - w;
 +    }
 +    return s;
 +}
 +
 +static inline int get_penalty_factor(int lambda, int lambda2, int type){
 +    switch(type&0xFF){
 +    default:
 +    case FF_CMP_SAD:
 +        return lambda>>FF_LAMBDA_SHIFT;
 +    case FF_CMP_DCT:
 +        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 +    case FF_CMP_W53:
 +        return (4*lambda)>>(FF_LAMBDA_SHIFT);
 +    case FF_CMP_W97:
 +        return (2*lambda)>>(FF_LAMBDA_SHIFT);
 +    case FF_CMP_SATD:
 +    case FF_CMP_DCT264:
 +        return (2*lambda)>>FF_LAMBDA_SHIFT;
 +    case FF_CMP_RD:
 +    case FF_CMP_PSNR:
 +    case FF_CMP_SSE:
 +    case FF_CMP_NSSE:
 +        return lambda2>>FF_LAMBDA_SHIFT;
 +    case FF_CMP_BIT:
 +        return 1;
 +    }
 +}
 +
 +//FIXME copy&paste
 +#define P_LEFT P[1]
 +#define P_TOP P[2]
 +#define P_TOPRIGHT P[3]
 +#define P_MEDIAN P[4]
 +#define P_MV1 P[9]
 +#define FLAG_QPEL   1 //must be 1
 +
 +static int encode_q_branch(SnowContext *s, int level, int x, int y){
 +    uint8_t p_buffer[1024];
 +    uint8_t i_buffer[1024];
 +    uint8_t p_state[sizeof(s->block_state)];
 +    uint8_t i_state[sizeof(s->block_state)];
 +    RangeCoder pc, ic;
 +    uint8_t *pbbak= s->c.bytestream;
 +    uint8_t *pbbak_start= s->c.bytestream_start;
 +    int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
 +    const int w= s->b_width  << s->block_max_depth;
 +    const int h= s->b_height << s->block_max_depth;
 +    const int rem_depth= s->block_max_depth - level;
 +    const int index= (x + y*w) << rem_depth;
 +    const int block_w= 1<<(LOG2_MB_SIZE - level);
 +    int trx= (x+1)<<rem_depth;
 +    int try= (y+1)<<rem_depth;
 +    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
 +    const BlockNode *top   = y ? &s->block[index-w] : &null_block;
 +    const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
 +    const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
 +    const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
 +    const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
 +    int pl = left->color[0];
 +    int pcb= left->color[1];
 +    int pcr= left->color[2];
 +    int pmx, pmy;
 +    int mx=0, my=0;
 +    int l,cr,cb;
 +    const int stride= s->current_picture->linesize[0];
 +    const int uvstride= s->current_picture->linesize[1];
 +    uint8_t *current_data[3]= { s->input_picture->data[0] + (x + y*  stride)*block_w,
 +                                s->input_picture->data[1] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift),
 +                                s->input_picture->data[2] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift)};
 +    int P[10][2];
 +    int16_t last_mv[3][2];
 +    int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
 +    const int shift= 1+qpel;
 +    MotionEstContext *c= &s->m.me;
 +    int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
 +    int mx_context= av_log2(2*FFABS(left->mx - top->mx));
 +    int my_context= av_log2(2*FFABS(left->my - top->my));
 +    int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
 +    int ref, best_ref, ref_score, ref_mx, ref_my;
 +
 +    av_assert0(sizeof(s->block_state) >= 256);
 +    if(s->keyframe){
 +        set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
 +        return 0;
 +    }
 +
 +//    clip predictors / edge ?
 +
 +    P_LEFT[0]= left->mx;
 +    P_LEFT[1]= left->my;
 +    P_TOP [0]= top->mx;
 +    P_TOP [1]= top->my;
 +    P_TOPRIGHT[0]= tr->mx;
 +    P_TOPRIGHT[1]= tr->my;
 +
 +    last_mv[0][0]= s->block[index].mx;
 +    last_mv[0][1]= s->block[index].my;
 +    last_mv[1][0]= right->mx;
 +    last_mv[1][1]= right->my;
 +    last_mv[2][0]= bottom->mx;
 +    last_mv[2][1]= bottom->my;
 +
 +    s->m.mb_stride=2;
 +    s->m.mb_x=
 +    s->m.mb_y= 0;
 +    c->skip= 0;
 +
 +    av_assert1(c->  stride ==   stride);
 +    av_assert1(c->uvstride == uvstride);
 +
 +    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
 +    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
 +    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
 +    c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;
 +
 +    c->xmin = - x*block_w - 16+3;
 +    c->ymin = - y*block_w - 16+3;
 +    c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
 +    c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
 +
 +    if(P_LEFT[0]     > (c->xmax<<shift)) P_LEFT[0]    = (c->xmax<<shift);
 +    if(P_LEFT[1]     > (c->ymax<<shift)) P_LEFT[1]    = (c->ymax<<shift);
 +    if(P_TOP[0]      > (c->xmax<<shift)) P_TOP[0]     = (c->xmax<<shift);
 +    if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
 +    if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
 +    if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
 +    if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
 +
 +    P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
 +    P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
 +
 +    if (!y) {
 +        c->pred_x= P_LEFT[0];
 +        c->pred_y= P_LEFT[1];
 +    } else {
 +        c->pred_x = P_MEDIAN[0];
 +        c->pred_y = P_MEDIAN[1];
 +    }
 +
 +    score= INT_MAX;
 +    best_ref= 0;
 +    for(ref=0; ref<s->ref_frames; ref++){
 +        init_ref(c, current_data, s->last_picture[ref]->data, NULL, block_w*x, block_w*y, 0);
 +
 +        ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv,
 +                                         (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);
 +
 +        av_assert2(ref_mx >= c->xmin);
 +        av_assert2(ref_mx <= c->xmax);
 +        av_assert2(ref_my >= c->ymin);
 +        av_assert2(ref_my <= c->ymax);
 +
 +        ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
 +        ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
 +        ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
 +        if(s->ref_mvs[ref]){
 +            s->ref_mvs[ref][index][0]= ref_mx;
 +            s->ref_mvs[ref][index][1]= ref_my;
 +            s->ref_scores[ref][index]= ref_score;
 +        }
 +        if(score > ref_score){
 +            score= ref_score;
 +            best_ref= ref;
 +            mx= ref_mx;
 +            my= ref_my;
 +        }
 +    }
 +    //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2
 +
 +  //  subpel search
 +    base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
 +    pc= s->c;
 +    pc.bytestream_start=
 +    pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
 +    memcpy(p_state, s->block_state, sizeof(s->block_state));
 +
 +    if(level!=s->block_max_depth)
 +        put_rac(&pc, &p_state[4 + s_context], 1);
 +    put_rac(&pc, &p_state[1 + left->type + top->type], 0);
 +    if(s->ref_frames > 1)
 +        put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0);
 +    pred_mv(s, &pmx, &pmy, best_ref, left, top, tr);
 +    put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
 +    put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1);
 +    p_len= pc.bytestream - pc.bytestream_start;
 +    score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;
 +
 +    block_s= block_w*block_w;
 +    sum = pix_sum(current_data[0], stride, block_w, block_w);
 +    l= (sum + block_s/2)/block_s;
 +    iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;
 +
 +    if (s->nb_planes > 2) {
 +        block_s= block_w*block_w>>(s->chroma_h_shift + s->chroma_v_shift);
 +        sum = pix_sum(current_data[1], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
 +        cb= (sum + block_s/2)/block_s;
 +    //    iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
 +        sum = pix_sum(current_data[2], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
 +        cr= (sum + block_s/2)/block_s;
 +    //    iscore += pix_norm1(&current_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
 +    }else
 +        cb = cr = 0;
 +
 +    ic= s->c;
 +    ic.bytestream_start=
 +    ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
 +    memcpy(i_state, s->block_state, sizeof(s->block_state));
 +    if(level!=s->block_max_depth)
 +        put_rac(&ic, &i_state[4 + s_context], 1);
 +    put_rac(&ic, &i_state[1 + left->type + top->type], 1);
 +    put_symbol(&ic, &i_state[32],  l-pl , 1);
 +    if (s->nb_planes > 2) {
 +        put_symbol(&ic, &i_state[64], cb-pcb, 1);
 +        put_symbol(&ic, &i_state[96], cr-pcr, 1);
 +    }
 +    i_len= ic.bytestream - ic.bytestream_start;
 +    iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;
 +
 +//    assert(score==256*256*256*64-1);
 +    av_assert1(iscore < 255*255*256 + s->lambda2*10);
 +    av_assert1(iscore >= 0);
 +    av_assert1(l>=0 && l<=255);
 +    av_assert1(pl>=0 && pl<=255);
 +
 +    if(level==0){
 +        int varc= iscore >> 8;
 +        int vard= score >> 8;
 +        if (vard <= 64 || vard < varc)
 +            c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
 +        else
 +            c->scene_change_score+= s->m.qscale;
 +    }
 +
 +    if(level!=s->block_max_depth){
 +        put_rac(&s->c, &s->block_state[4 + s_context], 0);
 +        score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
 +        score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
 +        score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
 +        score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
 +        score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead
 +
 +        if(score2 < score && score2 < iscore)
 +            return score2;
 +    }
 +
 +    if(iscore < score){
 +        pred_mv(s, &pmx, &pmy, 0, left, top, tr);
 +        memcpy(pbbak, i_buffer, i_len);
 +        s->c= ic;
 +        s->c.bytestream_start= pbbak_start;
 +        s->c.bytestream= pbbak + i_len;
 +        set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA);
 +        memcpy(s->block_state, i_state, sizeof(s->block_state));
 +        return iscore;
 +    }else{
 +        memcpy(pbbak, p_buffer, p_len);
 +        s->c= pc;
 +        s->c.bytestream_start= pbbak_start;
 +        s->c.bytestream= pbbak + p_len;
 +        set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0);
 +        memcpy(s->block_state, p_state, sizeof(s->block_state));
 +        return score;
 +    }
 +}
 +
 +static void encode_q_branch2(SnowContext *s, int level, int x, int y){
 +    const int w= s->b_width  << s->block_max_depth;
 +    const int rem_depth= s->block_max_depth - level;
 +    const int index= (x + y*w) << rem_depth;
 +    int trx= (x+1)<<rem_depth;
 +    BlockNode *b= &s->block[index];
 +    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
 +    const BlockNode *top   = y ? &s->block[index-w] : &null_block;
 +    const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
 +    const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
 +    int pl = left->color[0];
 +    int pcb= left->color[1];
 +    int pcr= left->color[2];
 +    int pmx, pmy;
 +    int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
 +    int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref;
 +    int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref;
 +    int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
 +
 +    if(s->keyframe){
 +        set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
 +        return;
 +    }
 +
 +    if(level!=s->block_max_depth){
 +        if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){
 +            put_rac(&s->c, &s->block_state[4 + s_context], 1);
 +        }else{
 +            put_rac(&s->c, &s->block_state[4 + s_context], 0);
 +            encode_q_branch2(s, level+1, 2*x+0, 2*y+0);
 +            encode_q_branch2(s, level+1, 2*x+1, 2*y+0);
 +            encode_q_branch2(s, level+1, 2*x+0, 2*y+1);
 +            encode_q_branch2(s, level+1, 2*x+1, 2*y+1);
 +            return;
 +        }
 +    }
 +    if(b->type & BLOCK_INTRA){
 +        pred_mv(s, &pmx, &pmy, 0, left, top, tr);
 +        put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1);
 +        put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1);
 +        if (s->nb_planes > 2) {
 +            put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1);
 +            put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1);
 +        }
 +        set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA);
 +    }else{
 +        pred_mv(s, &pmx, &pmy, b->ref, left, top, tr);
 +        put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0);
 +        if(s->ref_frames > 1)
 +            put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0);
 +        put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1);
 +        put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1);
 +        set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0);
 +    }
 +}
 +
 +static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
 +    int i, x2, y2;
 +    Plane *p= &s->plane[plane_index];
 +    const int block_size = MB_SIZE >> s->block_max_depth;
 +    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
 +    const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
 +    const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
 +    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
 +    const int ref_stride= s->current_picture->linesize[plane_index];
 +    uint8_t *src= s-> input_picture->data[plane_index];
 +    IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned
 +    const int b_stride = s->b_width << s->block_max_depth;
 +    const int w= p->width;
 +    const int h= p->height;
 +    int index= mb_x + mb_y*b_stride;
 +    BlockNode *b= &s->block[index];
 +    BlockNode backup= *b;
 +    int ab=0;
 +    int aa=0;
 +
 +    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc stuff above
 +
 +    b->type|= BLOCK_INTRA;
 +    b->color[plane_index]= 0;
 +    memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM));
 +
 +    for(i=0; i<4; i++){
 +        int mb_x2= mb_x + (i &1) - 1;
 +        int mb_y2= mb_y + (i>>1) - 1;
 +        int x= block_w*mb_x2 + block_w/2;
 +        int y= block_h*mb_y2 + block_h/2;
 +
 +        add_yblock(s, 0, NULL, dst + (i&1)*block_w + (i>>1)*obmc_stride*block_h, NULL, obmc,
 +                    x, y, block_w, block_h, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index);
 +
 +        for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_h); y2++){
 +            for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){
 +                int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_h*mb_y - block_h/2))*obmc_stride;
 +                int obmc_v= obmc[index];
 +                int d;
 +                if(y<0) obmc_v += obmc[index + block_h*obmc_stride];
 +                if(x<0) obmc_v += obmc[index + block_w];
 +                if(y+block_h>h) obmc_v += obmc[index - block_h*obmc_stride];
 +                if(x+block_w>w) obmc_v += obmc[index - block_w];
 +                //FIXME precalculate this or simplify it somehow else
 +
 +                d = -dst[index] + (1<<(FRAC_BITS-1));
 +                dst[index] = d;
 +                ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v;
 +                aa += obmc_v * obmc_v; //FIXME precalculate this
 +            }
 +        }
 +    }
 +    *b= backup;
 +
 +    return av_clip( ROUNDED_DIV(ab<<LOG2_OBMC_MAX, aa), 0, 255); //FIXME we should not need clipping
 +}
 +
 +static inline int get_block_bits(SnowContext *s, int x, int y, int w){
 +    const int b_stride = s->b_width << s->block_max_depth;
 +    const int b_height = s->b_height<< s->block_max_depth;
 +    int index= x + y*b_stride;
 +    const BlockNode *b     = &s->block[index];
 +    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
 +    const BlockNode *top   = y ? &s->block[index-b_stride] : &null_block;
 +    const BlockNode *tl    = y && x ? &s->block[index-b_stride-1] : left;
 +    const BlockNode *tr    = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl;
 +    int dmx, dmy;
 +//  int mx_context= av_log2(2*FFABS(left->mx - top->mx));
 +//  int my_context= av_log2(2*FFABS(left->my - top->my));
 +
 +    if(x<0 || x>=b_stride || y>=b_height)
 +        return 0;
 +/*
 +1            0      0
 +01X          1-2    1
 +001XX        3-6    2-3
 +0001XXX      7-14   4-7
 +00001XXXX   15-30   8-15
 +*/
 +//FIXME try accurate rate
 +//FIXME intra and inter predictors if surrounding blocks are not the same type
 +    if(b->type & BLOCK_INTRA){
 +        return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0]))
 +                   + av_log2(2*FFABS(left->color[1] - b->color[1]))
 +                   + av_log2(2*FFABS(left->color[2] - b->color[2])));
 +    }else{
 +        pred_mv(s, &dmx, &dmy, b->ref, left, top, tr);
 +        dmx-= b->mx;
 +        dmy-= b->my;
 +        return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda
 +                    + av_log2(2*FFABS(dmy))
 +                    + av_log2(2*b->ref));
 +    }
 +}
 +
 +static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, uint8_t (*obmc_edged)[MB_SIZE * 2]){
 +    Plane *p= &s->plane[plane_index];
 +    const int block_size = MB_SIZE >> s->block_max_depth;
 +    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
 +    const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
 +    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
 +    const int ref_stride= s->current_picture->linesize[plane_index];
 +    uint8_t *dst= s->current_picture->data[plane_index];
 +    uint8_t *src= s->  input_picture->data[plane_index];
 +    IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
 +    uint8_t *cur = s->scratchbuf;
 +    uint8_t *tmp = s->emu_edge_buffer;
 +    const int b_stride = s->b_width << s->block_max_depth;
 +    const int b_height = s->b_height<< s->block_max_depth;
 +    const int w= p->width;
 +    const int h= p->height;
 +    int distortion;
 +    int rate= 0;
 +    const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
 +    int sx= block_w*mb_x - block_w/2;
 +    int sy= block_h*mb_y - block_h/2;
 +    int x0= FFMAX(0,-sx);
 +    int y0= FFMAX(0,-sy);
 +    int x1= FFMIN(block_w*2, w-sx);
 +    int y1= FFMIN(block_h*2, h-sy);
 +    int i,x,y;
 +
 +    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below chckinhg only block_w
 +
 +    ff_snow_pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_h*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h);
 +
 +    for(y=y0; y<y1; y++){
 +        const uint8_t *obmc1= obmc_edged[y];
 +        const IDWTELEM *pred1 = pred + y*obmc_stride;
 +        uint8_t *cur1 = cur + y*ref_stride;
 +        uint8_t *dst1 = dst + sx + (sy+y)*ref_stride;
 +        for(x=x0; x<x1; x++){
 +#if FRAC_BITS >= LOG2_OBMC_MAX
 +            int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX);
 +#else
 +            int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS);
 +#endif
 +            v = (v + pred1[x]) >> FRAC_BITS;
 +            if(v&(~255)) v= ~(v>>31);
 +            dst1[x] = v;
 +        }
 +    }
 +
 +    /* copy the regions where obmc[] = (uint8_t)256 */
 +    if(LOG2_OBMC_MAX == 8
 +        && (mb_x == 0 || mb_x == b_stride-1)
 +        && (mb_y == 0 || mb_y == b_height-1)){
 +        if(mb_x == 0)
 +            x1 = block_w;
 +        else
 +            x0 = block_w;
 +        if(mb_y == 0)
 +            y1 = block_h;
 +        else
 +            y0 = block_h;
 +        for(y=y0; y<y1; y++)
 +            memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0);
 +    }
 +
 +    if(block_w==16){
 +        /* FIXME rearrange dsputil to fit 32x32 cmp functions */
 +        /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */
 +        /* FIXME cmps overlap but do not cover the wavelet's whole support.
 +         * So improving the score of one block is not strictly guaranteed
 +         * to improve the score of the whole frame, thus iterative motion
 +         * estimation does not always converge. */
 +        if(s->avctx->me_cmp == FF_CMP_W97)
 +            distortion = ff_w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
 +        else if(s->avctx->me_cmp == FF_CMP_W53)
 +            distortion = ff_w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
 +        else{
 +            distortion = 0;
 +            for(i=0; i<4; i++){
 +                int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride;
 +                distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16);
 +            }
 +        }
 +    }else{
 +        av_assert2(block_w==8);
 +        distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2);
 +    }
 +
 +    if(plane_index==0){
 +        for(i=0; i<4; i++){
 +/* ..RRr
 + * .RXx.
 + * rxx..
 + */
 +            rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1);
 +        }
 +        if(mb_x == b_stride-2)
 +            rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1);
 +    }
 +    return distortion + rate*penalty_factor;
 +}
 +
 +static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
 +    int i, y2;
 +    Plane *p= &s->plane[plane_index];
 +    const int block_size = MB_SIZE >> s->block_max_depth;
 +    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
 +    const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
 +    const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
 +    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
 +    const int ref_stride= s->current_picture->linesize[plane_index];
 +    uint8_t *dst= s->current_picture->data[plane_index];
 +    uint8_t *src= s-> input_picture->data[plane_index];
 +    //FIXME zero_dst is const but add_yblock changes dst if add is 0 (this is never the case for dst=zero_dst
 +    // const has only been removed from zero_dst to suppress a warning
 +    static IDWTELEM zero_dst[4096]; //FIXME
 +    const int b_stride = s->b_width << s->block_max_depth;
 +    const int w= p->width;
 +    const int h= p->height;
 +    int distortion= 0;
 +    int rate= 0;
 +    const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
 +
 +    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below
 +
 +    for(i=0; i<9; i++){
 +        int mb_x2= mb_x + (i%3) - 1;
 +        int mb_y2= mb_y + (i/3) - 1;
 +        int x= block_w*mb_x2 + block_w/2;
 +        int y= block_h*mb_y2 + block_h/2;
 +
 +        add_yblock(s, 0, NULL, zero_dst, dst, obmc,
 +                   x, y, block_w, block_h, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index);
 +
 +        //FIXME find a cleaner/simpler way to skip the outside stuff
 +        for(y2= y; y2<0; y2++)
 +            memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
 +        for(y2= h; y2<y+block_h; y2++)
 +            memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
 +        if(x<0){
 +            for(y2= y; y2<y+block_h; y2++)
 +                memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x);
 +        }
 +        if(x+block_w > w){
 +            for(y2= y; y2<y+block_h; y2++)
 +                memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w);
 +        }
 +
 +        av_assert1(block_w== 8 || block_w==16);
 +        distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_h);
 +    }
 +
 +    if(plane_index==0){
 +        BlockNode *b= &s->block[mb_x+mb_y*b_stride];
 +        int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1);
 +
 +/* ..RRRr
 + * .RXXx.
 + * .RXXx.
 + * rxxx.
 + */
 +        if(merged)
 +            rate = get_block_bits(s, mb_x, mb_y, 2);
 +        for(i=merged?4:0; i<9; i++){
 +            static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}};
 +            rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1);
 +        }
 +    }
 +    return distortion + rate*penalty_factor;
 +}
 +
 +static int encode_subband_c0run(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
 +    const int w= b->width;
 +    const int h= b->height;
 +    int x, y;
 +
 +    if(1){
 +        int run=0;
 +        int *runs = s->run_buffer;
 +        int run_index=0;
 +        int max_index;
 +
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int v, p=0;
 +                int /*ll=0, */l=0, lt=0, t=0, rt=0;
 +                v= src[x + y*stride];
 +
 +                if(y){
 +                    t= src[x + (y-1)*stride];
 +                    if(x){
 +                        lt= src[x - 1 + (y-1)*stride];
 +                    }
 +                    if(x + 1 < w){
 +                        rt= src[x + 1 + (y-1)*stride];
 +                    }
 +                }
 +                if(x){
 +                    l= src[x - 1 + y*stride];
 +                    /*if(x > 1){
 +                        if(orientation==1) ll= src[y + (x-2)*stride];
 +                        else               ll= src[x - 2 + y*stride];
 +                    }*/
 +                }
 +                if(parent){
 +                    int px= x>>1;
 +                    int py= y>>1;
 +                    if(px<b->parent->width && py<b->parent->height)
 +                        p= parent[px + py*2*stride];
 +                }
 +                if(!(/*ll|*/l|lt|t|rt|p)){
 +                    if(v){
 +                        runs[run_index++]= run;
 +                        run=0;
 +                    }else{
 +                        run++;
 +                    }
 +                }
 +            }
 +        }
 +        max_index= run_index;
 +        runs[run_index++]= run;
 +        run_index=0;
 +        run= runs[run_index++];
 +
 +        put_symbol2(&s->c, b->state[30], max_index, 0);
 +        if(run_index <= max_index)
 +            put_symbol2(&s->c, b->state[1], run, 3);
 +
 +        for(y=0; y<h; y++){
 +            if(s->c.bytestream_end - s->c.bytestream < w*40){
 +                av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
 +                return -1;
 +            }
 +            for(x=0; x<w; x++){
 +                int v, p=0;
 +                int /*ll=0, */l=0, lt=0, t=0, rt=0;
 +                v= src[x + y*stride];
 +
 +                if(y){
 +                    t= src[x + (y-1)*stride];
 +                    if(x){
 +                        lt= src[x - 1 + (y-1)*stride];
 +                    }
 +                    if(x + 1 < w){
 +                        rt= src[x + 1 + (y-1)*stride];
 +                    }
 +                }
 +                if(x){
 +                    l= src[x - 1 + y*stride];
 +                    /*if(x > 1){
 +                        if(orientation==1) ll= src[y + (x-2)*stride];
 +                        else               ll= src[x - 2 + y*stride];
 +                    }*/
 +                }
 +                if(parent){
 +                    int px= x>>1;
 +                    int py= y>>1;
 +                    if(px<b->parent->width && py<b->parent->height)
 +                        p= parent[px + py*2*stride];
 +                }
 +                if(/*ll|*/l|lt|t|rt|p){
 +                    int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
 +
 +                    put_rac(&s->c, &b->state[0][context], !!v);
 +                }else{
 +                    if(!run){
 +                        run= runs[run_index++];
 +
 +                        if(run_index <= max_index)
 +                            put_symbol2(&s->c, b->state[1], run, 3);
 +                        av_assert2(v);
 +                    }else{
 +                        run--;
 +                        av_assert2(!v);
 +                    }
 +                }
 +                if(v){
 +                    int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
 +                    int l2= 2*FFABS(l) + (l<0);
 +                    int t2= 2*FFABS(t) + (t<0);
 +
 +                    put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4);
 +                    put_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l2&0xFF] + 3*ff_quant3bA[t2&0xFF]], v<0);
 +                }
 +            }
 +        }
 +    }
 +    return 0;
 +}
 +
 +static int encode_subband(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
 +//    encode_subband_qtree(s, b, src, parent, stride, orientation);
 +//    encode_subband_z0run(s, b, src, parent, stride, orientation);
 +    return encode_subband_c0run(s, b, src, parent, stride, orientation);
 +//    encode_subband_dzr(s, b, src, parent, stride, orientation);
 +}
 +
 +static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, uint8_t (*obmc_edged)[MB_SIZE * 2], int *best_rd){
 +    const int b_stride= s->b_width << s->block_max_depth;
 +    BlockNode *block= &s->block[mb_x + mb_y * b_stride];
 +    BlockNode backup= *block;
 +    unsigned value;
 +    int rd, index;
 +
 +    av_assert2(mb_x>=0 && mb_y>=0);
 +    av_assert2(mb_x<b_stride);
 +
 +    if(intra){
 +        block->color[0] = p[0];
 +        block->color[1] = p[1];
 +        block->color[2] = p[2];
 +        block->type |= BLOCK_INTRA;
 +    }else{
 +        index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1);
 +        value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12);
 +        if(s->me_cache[index] == value)
 +            return 0;
 +        s->me_cache[index]= value;
 +
 +        block->mx= p[0];
 +        block->my= p[1];
 +        block->type &= ~BLOCK_INTRA;
 +    }
 +
 +    rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged);
 +
 +//FIXME chroma
 +    if(rd < *best_rd){
 +        *best_rd= rd;
 +        return 1;
 +    }else{
 +        *block= backup;
 +        return 0;
 +    }
 +}
 +
 +/* special case for int[2] args we discard afterwards,
 + * fixes compilation problem with gcc 2.95 */
 +static av_always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, uint8_t (*obmc_edged)[MB_SIZE * 2], int *best_rd){
 +    int p[2] = {p0, p1};
 +    return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd);
 +}
 +
 +static av_always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){
 +    const int b_stride= s->b_width << s->block_max_depth;
 +    BlockNode *block= &s->block[mb_x + mb_y * b_stride];
 +    BlockNode backup[4];
 +    unsigned value;
 +    int rd, index;
 +
 +    /* We don't initialize backup[] during variable declaration, because
 +     * that fails to compile on MSVC: "cannot convert from 'BlockNode' to
 +     * 'int16_t'". */
 +    backup[0] = block[0];
 +    backup[1] = block[1];
 +    backup[2] = block[b_stride];
 +    backup[3] = block[b_stride + 1];
 +
 +    av_assert2(mb_x>=0 && mb_y>=0);
 +    av_assert2(mb_x<b_stride);
 +    av_assert2(((mb_x|mb_y)&1) == 0);
 +
 +    index= (p0 + 31*p1) & (ME_CACHE_SIZE-1);
 +    value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12);
 +    if(s->me_cache[index] == value)
 +        return 0;
 +    s->me_cache[index]= value;
 +
 +    block->mx= p0;
 +    block->my= p1;
 +    block->ref= ref;
 +    block->type &= ~BLOCK_INTRA;
 +    block[1]= block[b_stride]= block[b_stride+1]= *block;
 +
 +    rd= get_4block_rd(s, mb_x, mb_y, 0);
 +
 +//FIXME chroma
 +    if(rd < *best_rd){
 +        *best_rd= rd;
 +        return 1;
 +    }else{
 +        block[0]= backup[0];
 +        block[1]= backup[1];
 +        block[b_stride]= backup[2];
 +        block[b_stride+1]= backup[3];
 +        return 0;
 +    }
 +}
 +
 +static void iterative_me(SnowContext *s){
 +    int pass, mb_x, mb_y;
 +    const int b_width = s->b_width  << s->block_max_depth;
 +    const int b_height= s->b_height << s->block_max_depth;
 +    const int b_stride= b_width;
 +    int color[3];
 +
 +    {
 +        RangeCoder r = s->c;
 +        uint8_t state[sizeof(s->block_state)];
 +        memcpy(state, s->block_state, sizeof(s->block_state));
 +        for(mb_y= 0; mb_y<s->b_height; mb_y++)
 +            for(mb_x= 0; mb_x<s->b_width; mb_x++)
 +                encode_q_branch(s, 0, mb_x, mb_y);
 +        s->c = r;
 +        memcpy(s->block_state, state, sizeof(s->block_state));
 +    }
 +
 +    for(pass=0; pass<25; pass++){
 +        int change= 0;
 +
 +        for(mb_y= 0; mb_y<b_height; mb_y++){
 +            for(mb_x= 0; mb_x<b_width; mb_x++){
 +                int dia_change, i, j, ref;
 +                int best_rd= INT_MAX, ref_rd;
 +                BlockNode backup, ref_b;
 +                const int index= mb_x + mb_y * b_stride;
 +                BlockNode *block= &s->block[index];
 +                BlockNode *tb =                   mb_y            ? &s->block[index-b_stride  ] : NULL;
 +                BlockNode *lb = mb_x                              ? &s->block[index         -1] : NULL;
 +                BlockNode *rb = mb_x+1<b_width                    ? &s->block[index         +1] : NULL;
 +                BlockNode *bb =                   mb_y+1<b_height ? &s->block[index+b_stride  ] : NULL;
 +                BlockNode *tlb= mb_x           && mb_y            ? &s->block[index-b_stride-1] : NULL;
 +                BlockNode *trb= mb_x+1<b_width && mb_y            ? &s->block[index-b_stride+1] : NULL;
 +                BlockNode *blb= mb_x           && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL;
 +                BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL;
 +                const int b_w= (MB_SIZE >> s->block_max_depth);
 +                uint8_t obmc_edged[MB_SIZE * 2][MB_SIZE * 2];
 +
 +                if(pass && (block->type & BLOCK_OPT))
 +                    continue;
 +                block->type |= BLOCK_OPT;
 +
 +                backup= *block;
 +
 +                if(!s->me_cache_generation)
 +                    memset(s->me_cache, 0, sizeof(s->me_cache));
 +                s->me_cache_generation += 1<<22;
 +
 +                //FIXME precalculate
 +                {
 +                    int x, y;
 +                    for (y = 0; y < b_w * 2; y++)
 +                        memcpy(obmc_edged[y], ff_obmc_tab[s->block_max_depth] + y * b_w * 2, b_w * 2);
 +                    if(mb_x==0)
 +                        for(y=0; y<b_w*2; y++)
 +                            memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w);
 +                    if(mb_x==b_stride-1)
 +                        for(y=0; y<b_w*2; y++)
 +                            memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w);
 +                    if(mb_y==0){
 +                        for(x=0; x<b_w*2; x++)
 +                            obmc_edged[0][x] += obmc_edged[b_w-1][x];
 +                        for(y=1; y<b_w; y++)
 +                            memcpy(obmc_edged[y], obmc_edged[0], b_w*2);
 +                    }
 +                    if(mb_y==b_height-1){
 +                        for(x=0; x<b_w*2; x++)
 +                            obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x];
 +                        for(y=b_w; y<b_w*2-1; y++)
 +                            memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2);
 +                    }
 +                }
 +
 +                //skip stuff outside the picture
 +                if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1){
 +                    uint8_t *src= s->  input_picture->data[0];
 +                    uint8_t *dst= s->current_picture->data[0];
 +                    const int stride= s->current_picture->linesize[0];
 +                    const int block_w= MB_SIZE >> s->block_max_depth;
 +                    const int block_h= MB_SIZE >> s->block_max_depth;
 +                    const int sx= block_w*mb_x - block_w/2;
 +                    const int sy= block_h*mb_y - block_h/2;
 +                    const int w= s->plane[0].width;
 +                    const int h= s->plane[0].height;
 +                    int y;
 +
 +                    for(y=sy; y<0; y++)
 +                        memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
 +                    for(y=h; y<sy+block_h*2; y++)
 +                        memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
 +                    if(sx<0){
 +                        for(y=sy; y<sy+block_h*2; y++)
 +                            memcpy(dst + sx + y*stride, src + sx + y*stride, -sx);
 +                    }
 +                    if(sx+block_w*2 > w){
 +                        for(y=sy; y<sy+block_h*2; y++)
 +                            memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w);
 +                    }
 +                }
 +
 +                // intra(black) = neighbors' contribution to the current block
 +                for(i=0; i < s->nb_planes; i++)
 +                    color[i]= get_dc(s, mb_x, mb_y, i);
 +
 +                // get previous score (cannot be cached due to OBMC)
 +                if(pass > 0 && (block->type&BLOCK_INTRA)){
 +                    int color0[3]= {block->color[0], block->color[1], block->color[2]};
 +                    check_block(s, mb_x, mb_y, color0, 1, obmc_edged, &best_rd);
 +                }else
 +                    check_block_inter(s, mb_x, mb_y, block->mx, block->my, obmc_edged, &best_rd);
 +
 +                ref_b= *block;
 +                ref_rd= best_rd;
 +                for(ref=0; ref < s->ref_frames; ref++){
 +                    int16_t (*mvr)[2]= &s->ref_mvs[ref][index];
 +                    if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold
 +                        continue;
 +                    block->ref= ref;
 +                    best_rd= INT_MAX;
 +
 +                    check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], obmc_edged, &best_rd);
 +                    check_block_inter(s, mb_x, mb_y, 0, 0, obmc_edged, &best_rd);
 +                    if(tb)
 +                        check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], obmc_edged, &best_rd);
 +                    if(lb)
 +                        check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], obmc_edged, &best_rd);
 +                    if(rb)
 +                        check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], obmc_edged, &best_rd);
 +                    if(bb)
 +                        check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], obmc_edged, &best_rd);
 +
 +                    /* fullpel ME */
 +                    //FIXME avoid subpel interpolation / round to nearest integer
 +                    do{
 +                        dia_change=0;
 +                        for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){
 +                            for(j=0; j<i; j++){
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), obmc_edged, &best_rd);
 +                            }
 +                        }
 +                    }while(dia_change);
 +                    /* subpel ME */
 +                    do{
 +                        static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},};
 +                        dia_change=0;
 +                        for(i=0; i<8; i++)
 +                            dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], obmc_edged, &best_rd);
 +                    }while(dia_change);
 +                    //FIXME or try the standard 2 pass qpel or similar
 +
 +                    mvr[0][0]= block->mx;
 +                    mvr[0][1]= block->my;
 +                    if(ref_rd > best_rd){
 +                        ref_rd= best_rd;
 +                        ref_b= *block;
 +                    }
 +                }
 +                best_rd= ref_rd;
 +                *block= ref_b;
 +                check_block(s, mb_x, mb_y, color, 1, obmc_edged, &best_rd);
 +                //FIXME RD style color selection
 +                if(!same_block(block, &backup)){
 +                    if(tb ) tb ->type &= ~BLOCK_OPT;
 +                    if(lb ) lb ->type &= ~BLOCK_OPT;
 +                    if(rb ) rb ->type &= ~BLOCK_OPT;
 +                    if(bb ) bb ->type &= ~BLOCK_OPT;
 +                    if(tlb) tlb->type &= ~BLOCK_OPT;
 +                    if(trb) trb->type &= ~BLOCK_OPT;
 +                    if(blb) blb->type &= ~BLOCK_OPT;
 +                    if(brb) brb->type &= ~BLOCK_OPT;
 +                    change ++;
 +                }
 +            }
 +        }
 +        av_log(s->avctx, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change);
 +        if(!change)
 +            break;
 +    }
 +
 +    if(s->block_max_depth == 1){
 +        int change= 0;
 +        for(mb_y= 0; mb_y<b_height; mb_y+=2){
 +            for(mb_x= 0; mb_x<b_width; mb_x+=2){
 +                int i;
 +                int best_rd, init_rd;
 +                const int index= mb_x + mb_y * b_stride;
 +                BlockNode *b[4];
 +
 +                b[0]= &s->block[index];
 +                b[1]= b[0]+1;
 +                b[2]= b[0]+b_stride;
 +                b[3]= b[2]+1;
 +                if(same_block(b[0], b[1]) &&
 +                   same_block(b[0], b[2]) &&
 +                   same_block(b[0], b[3]))
 +                    continue;
 +
 +                if(!s->me_cache_generation)
 +                    memset(s->me_cache, 0, sizeof(s->me_cache));
 +                s->me_cache_generation += 1<<22;
 +
 +                init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0);
 +
 +                //FIXME more multiref search?
 +                check_4block_inter(s, mb_x, mb_y,
 +                                   (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2,
 +                                   (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd);
 +
 +                for(i=0; i<4; i++)
 +                    if(!(b[i]->type&BLOCK_INTRA))
 +                        check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd);
 +
 +                if(init_rd != best_rd)
 +                    change++;
 +            }
 +        }
 +        av_log(s->avctx, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4);
 +    }
 +}
 +
 +static void encode_blocks(SnowContext *s, int search){
 +    int x, y;
 +    int w= s->b_width;
 +    int h= s->b_height;
 +
 +    if(s->avctx->me_method == ME_ITER && !s->keyframe && search)
 +        iterative_me(s);
 +
 +    for(y=0; y<h; y++){
 +        if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit
 +            av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
 +            return;
 +        }
 +        for(x=0; x<w; x++){
 +            if(s->avctx->me_method == ME_ITER || !search)
 +                encode_q_branch2(s, 0, x, y);
 +            else
 +                encode_q_branch (s, 0, x, y);
 +        }
 +    }
 +}
 +
 +static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){
 +    const int w= b->width;
 +    const int h= b->height;
 +    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
 +    const int qmul= ff_qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);
 +    int x,y, thres1, thres2;
 +
 +    if(s->qlog == LOSSLESS_QLOG){
 +        for(y=0; y<h; y++)
 +            for(x=0; x<w; x++)
 +                dst[x + y*stride]= src[x + y*stride];
 +        return;
 +    }
 +
 +    bias= bias ? 0 : (3*qmul)>>3;
 +    thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
 +    thres2= 2*thres1;
 +
 +    if(!bias){
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int i= src[x + y*stride];
 +
 +                if((unsigned)(i+thres1) > thres2){
 +                    if(i>=0){
 +                        i<<= QEXPSHIFT;
 +                        i/= qmul; //FIXME optimize
 +                        dst[x + y*stride]=  i;
 +                    }else{
 +                        i= -i;
 +                        i<<= QEXPSHIFT;
 +                        i/= qmul; //FIXME optimize
 +                        dst[x + y*stride]= -i;
 +                    }
 +                }else
 +                    dst[x + y*stride]= 0;
 +            }
 +        }
 +    }else{
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int i= src[x + y*stride];
 +
 +                if((unsigned)(i+thres1) > thres2){
 +                    if(i>=0){
 +                        i<<= QEXPSHIFT;
 +                        i= (i + bias) / qmul; //FIXME optimize
 +                        dst[x + y*stride]=  i;
 +                    }else{
 +                        i= -i;
 +                        i<<= QEXPSHIFT;
 +                        i= (i + bias) / qmul; //FIXME optimize
 +                        dst[x + y*stride]= -i;
 +                    }
 +                }else
 +                    dst[x + y*stride]= 0;
 +            }
 +        }
 +    }
 +}
 +
 +static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){
 +    const int w= b->width;
 +    const int h= b->height;
 +    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
 +    const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
 +    const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
 +    int x,y;
 +
 +    if(s->qlog == LOSSLESS_QLOG) return;
 +
 +    for(y=0; y<h; y++){
 +        for(x=0; x<w; x++){
 +            int i= src[x + y*stride];
 +            if(i<0){
 +                src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
 +            }else if(i>0){
 +                src[x + y*stride]=  (( i*qmul + qadd)>>(QEXPSHIFT));
 +            }
 +        }
 +    }
 +}
 +
 +static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
 +    const int w= b->width;
 +    const int h= b->height;
 +    int x,y;
 +
 +    for(y=h-1; y>=0; y--){
 +        for(x=w-1; x>=0; x--){
 +            int i= x + y*stride;
 +
 +            if(x){
 +                if(use_median){
 +                    if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
 +                    else  src[i] -= src[i - 1];
 +                }else{
 +                    if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
 +                    else  src[i] -= src[i - 1];
 +                }
 +            }else{
 +                if(y) src[i] -= src[i - stride];
 +            }
 +        }
 +    }
 +}
 +
 +static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
 +    const int w= b->width;
 +    const int h= b->height;
 +    int x,y;
 +
 +    for(y=0; y<h; y++){
 +        for(x=0; x<w; x++){
 +            int i= x + y*stride;
 +
 +            if(x){
 +                if(use_median){
 +                    if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
 +                    else  src[i] += src[i - 1];
 +                }else{
 +                    if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
 +                    else  src[i] += src[i - 1];
 +                }
 +            }else{
 +                if(y) src[i] += src[i - stride];
 +            }
 +        }
 +    }
 +}
 +
 +static void encode_qlogs(SnowContext *s){
 +    int plane_index, level, orientation;
 +
 +    for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
 +        for(level=0; level<s->spatial_decomposition_count; level++){
 +            for(orientation=level ? 1:0; orientation<4; orientation++){
 +                if(orientation==2) continue;
 +                put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1);
 +            }
 +        }
 +    }
 +}
 +
 +static void encode_header(SnowContext *s){
 +    int plane_index, i;
 +    uint8_t kstate[32];
 +
 +    memset(kstate, MID_STATE, sizeof(kstate));
 +
 +    put_rac(&s->c, kstate, s->keyframe);
 +    if(s->keyframe || s->always_reset){
 +        ff_snow_reset_contexts(s);
 +        s->last_spatial_decomposition_type=
 +        s->last_qlog=
 +        s->last_qbias=
 +        s->last_mv_scale=
 +        s->last_block_max_depth= 0;
 +        for(plane_index=0; plane_index<2; plane_index++){
 +            Plane *p= &s->plane[plane_index];
 +            p->last_htaps=0;
 +            p->last_diag_mc=0;
 +            memset(p->last_hcoeff, 0, sizeof(p->last_hcoeff));
 +        }
 +    }
 +    if(s->keyframe){
 +        put_symbol(&s->c, s->header_state, s->version, 0);
 +        put_rac(&s->c, s->header_state, s->always_reset);
 +        put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0);
 +        put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0);
 +        put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
 +        put_symbol(&s->c, s->header_state, s->colorspace_type, 0);
 +        if (s->nb_planes > 2) {
 +            put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0);
 +            put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0);
 +        }
 +        put_rac(&s->c, s->header_state, s->spatial_scalability);
 +//        put_rac(&s->c, s->header_state, s->rate_scalability);
 +        put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0);
 +
 +        encode_qlogs(s);
 +    }
 +
 +    if(!s->keyframe){
 +        int update_mc=0;
 +        for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
 +            Plane *p= &s->plane[plane_index];
 +            update_mc |= p->last_htaps   != p->htaps;
 +            update_mc |= p->last_diag_mc != p->diag_mc;
 +            update_mc |= !!memcmp(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
 +        }
 +        put_rac(&s->c, s->header_state, update_mc);
 +        if(update_mc){
 +            for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
 +                Plane *p= &s->plane[plane_index];
 +                put_rac(&s->c, s->header_state, p->diag_mc);
 +                put_symbol(&s->c, s->header_state, p->htaps/2-1, 0);
 +                for(i= p->htaps/2; i; i--)
 +                    put_symbol(&s->c, s->header_state, FFABS(p->hcoeff[i]), 0);
 +            }
 +        }
 +        if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
 +            put_rac(&s->c, s->header_state, 1);
 +            put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
 +            encode_qlogs(s);
 +        }else
 +            put_rac(&s->c, s->header_state, 0);
 +    }
 +
 +    put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1);
 +    put_symbol(&s->c, s->header_state, s->qlog            - s->last_qlog    , 1);
 +    put_symbol(&s->c, s->header_state, s->mv_scale        - s->last_mv_scale, 1);
 +    put_symbol(&s->c, s->header_state, s->qbias           - s->last_qbias   , 1);
 +    put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1);
 +
 +}
 +
 +static void update_last_header_values(SnowContext *s){
 +    int plane_index;
 +
 +    if(!s->keyframe){
 +        for(plane_index=0; plane_index<2; plane_index++){
 +            Plane *p= &s->plane[plane_index];
 +            p->last_diag_mc= p->diag_mc;
 +            p->last_htaps  = p->htaps;
 +            memcpy(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
 +        }
 +    }
 +
 +    s->last_spatial_decomposition_type  = s->spatial_decomposition_type;
 +    s->last_qlog                        = s->qlog;
 +    s->last_qbias                       = s->qbias;
 +    s->last_mv_scale                    = s->mv_scale;
 +    s->last_block_max_depth             = s->block_max_depth;
 +    s->last_spatial_decomposition_count = s->spatial_decomposition_count;
 +}
 +
 +static int qscale2qlog(int qscale){
 +    return rint(QROOT*log2(qscale / (float)FF_QP2LAMBDA))
 +           + 61*QROOT/8; ///< 64 > 60
 +}
 +
 +static int ratecontrol_1pass(SnowContext *s, AVFrame *pict)
 +{
 +    /* Estimate the frame's complexity as a sum of weighted dwt coefficients.
 +     * FIXME we know exact mv bits at this point,
 +     * but ratecontrol isn't set up to include them. */
 +    uint32_t coef_sum= 0;
 +    int level, orientation, delta_qlog;
 +
 +    for(level=0; level<s->spatial_decomposition_count; level++){
 +        for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +            SubBand *b= &s->plane[0].band[level][orientation];
 +            IDWTELEM *buf= b->ibuf;
 +            const int w= b->width;
 +            const int h= b->height;
 +            const int stride= b->stride;
 +            const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16);
 +            const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
 +            const int qdiv= (1<<16)/qmul;
 +            int x, y;
 +            //FIXME this is ugly
 +            for(y=0; y<h; y++)
 +                for(x=0; x<w; x++)
 +                    buf[x+y*stride]= b->buf[x+y*stride];
 +            if(orientation==0)
 +                decorrelate(s, b, buf, stride, 1, 0);
 +            for(y=0; y<h; y++)
 +                for(x=0; x<w; x++)
 +                    coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16;
 +        }
 +    }
 +
 +    /* ugly, ratecontrol just takes a sqrt again */
 +    av_assert0(coef_sum < INT_MAX);
 +    coef_sum = (uint64_t)coef_sum * coef_sum >> 16;
 +
 +    if(pict->pict_type == AV_PICTURE_TYPE_I){
 +        s->m.current_picture.mb_var_sum= coef_sum;
 +        s->m.current_picture.mc_mb_var_sum= 0;
 +    }else{
 +        s->m.current_picture.mc_mb_var_sum= coef_sum;
 +        s->m.current_picture.mb_var_sum= 0;
 +    }
 +
 +    pict->quality= ff_rate_estimate_qscale(&s->m, 1);
 +    if (pict->quality < 0)
 +        return INT_MIN;
 +    s->lambda= pict->quality * 3/2;
 +    delta_qlog= qscale2qlog(pict->quality) - s->qlog;
 +    s->qlog+= delta_qlog;
 +    return delta_qlog;
 +}
 +
 +static void calculate_visual_weight(SnowContext *s, Plane *p){
 +    int width = p->width;
 +    int height= p->height;
 +    int level, orientation, x, y;
 +
 +    for(level=0; level<s->spatial_decomposition_count; level++){
 +        for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +            SubBand *b= &p->band[level][orientation];
 +            IDWTELEM *ibuf= b->ibuf;
 +            int64_t error=0;
 +
 +            memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height);
 +            ibuf[b->width/2 + b->height/2*b->stride]= 256*16;
 +            ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
 +            for(y=0; y<height; y++){
 +                for(x=0; x<width; x++){
 +                    int64_t d= s->spatial_idwt_buffer[x + y*width]*16;
 +                    error += d*d;
 +                }
 +            }
 +
 +            b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5);
 +        }
 +    }
 +}
 +
 +static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
 +                        const AVFrame *pict, int *got_packet)
 +{
 +    SnowContext *s = avctx->priv_data;
 +    RangeCoder * const c= &s->c;
 +    AVFrame *pic = pict;
 +    const int width= s->avctx->width;
 +    const int height= s->avctx->height;
 +    int level, orientation, plane_index, i, y, ret;
 +    uint8_t rc_header_bak[sizeof(s->header_state)];
 +    uint8_t rc_block_bak[sizeof(s->block_state)];
 +
 +    if ((ret = ff_alloc_packet2(avctx, pkt, s->b_width*s->b_height*MB_SIZE*MB_SIZE*3 + FF_MIN_BUFFER_SIZE)) < 0)
 +        return ret;
 +
 +    ff_init_range_encoder(c, pkt->data, pkt->size);
 +    ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
 +
 +    for(i=0; i < s->nb_planes; i++){
 +        int hshift= i ? s->chroma_h_shift : 0;
 +        int vshift= i ? s->chroma_v_shift : 0;
 +        for(y=0; y<(height>>vshift); y++)
 +            memcpy(&s->input_picture->data[i][y * s->input_picture->linesize[i]],
 +                   &pict->data[i][y * pict->linesize[i]],
 +                   width>>hshift);
 +        s->dsp.draw_edges(s->input_picture->data[i], s->input_picture->linesize[i],
 +                            width >> hshift, height >> vshift,
 +                            EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
 +                            EDGE_TOP | EDGE_BOTTOM);
 +
 +    }
 +    emms_c();
 +    s->new_picture = pict;
 +
 +    s->m.picture_number= avctx->frame_number;
 +    if(avctx->flags&CODEC_FLAG_PASS2){
 +        s->m.pict_type = pic->pict_type = s->m.rc_context.entry[avctx->frame_number].new_pict_type;
 +        s->keyframe = pic->pict_type == AV_PICTURE_TYPE_I;
 +        if(!(avctx->flags&CODEC_FLAG_QSCALE)) {
 +            pic->quality = ff_rate_estimate_qscale(&s->m, 0);
 +            if (pic->quality < 0)
 +                return -1;
 +        }
 +    }else{
 +        s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
 +        s->m.pict_type = pic->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
 +    }
 +
 +    if(s->pass1_rc && avctx->frame_number == 0)
 +        pic->quality = 2*FF_QP2LAMBDA;
 +    if (pic->quality) {
 +        s->qlog   = qscale2qlog(pic->quality);
 +        s->lambda = pic->quality * 3/2;
 +    }
 +    if (s->qlog < 0 || (!pic->quality && (avctx->flags & CODEC_FLAG_QSCALE))) {
 +        s->qlog= LOSSLESS_QLOG;
 +        s->lambda = 0;
 +    }//else keep previous frame's qlog until after motion estimation
 +
 +    ff_snow_frame_start(s);
 +    avctx->coded_frame= s->current_picture;
 +
 +    s->m.current_picture_ptr= &s->m.current_picture;
 +    s->m.current_picture.f = s->current_picture;
 +    s->m.current_picture.f->pts = pict->pts;
 +    if(pic->pict_type == AV_PICTURE_TYPE_P){
 +        int block_width = (width +15)>>4;
 +        int block_height= (height+15)>>4;
 +        int stride= s->current_picture->linesize[0];
 +
 +        av_assert0(s->current_picture->data[0]);
 +        av_assert0(s->last_picture[0]->data[0]);
 +
 +        s->m.avctx= s->avctx;
 +        s->m.   last_picture.f = s->last_picture[0];
 +        s->m.    new_picture.f = s->input_picture;
 +        s->m.   last_picture_ptr= &s->m.   last_picture;
 +        s->m.linesize = stride;
 +        s->m.uvlinesize= s->current_picture->linesize[1];
 +        s->m.width = width;
 +        s->m.height= height;
 +        s->m.mb_width = block_width;
 +        s->m.mb_height= block_height;
 +        s->m.mb_stride=   s->m.mb_width+1;
 +        s->m.b8_stride= 2*s->m.mb_width+1;
 +        s->m.f_code=1;
 +        s->m.pict_type = pic->pict_type;
 +        s->m.me_method= s->avctx->me_method;
 +        s->m.me.scene_change_score=0;
 +        s->m.flags= s->avctx->flags;
 +        s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0;
 +        s->m.out_format= FMT_H263;
 +        s->m.unrestricted_mv= 1;
 +
 +        s->m.lambda = s->lambda;
 +        s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 +        s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 +
 +        s->m.dsp= s->dsp; //move
++        s->m.qdsp= s->qdsp; //move
 +        s->m.hdsp = s->hdsp;
 +        ff_init_me(&s->m);
 +        s->hdsp = s->m.hdsp;
 +        s->dsp= s->m.dsp;
 +    }
 +
 +    if(s->pass1_rc){
 +        memcpy(rc_header_bak, s->header_state, sizeof(s->header_state));
 +        memcpy(rc_block_bak, s->block_state, sizeof(s->block_state));
 +    }
 +
 +redo_frame:
 +
 +    s->spatial_decomposition_count= 5;
 +
 +    while(   !(width >>(s->chroma_h_shift + s->spatial_decomposition_count))
 +          || !(height>>(s->chroma_v_shift + s->spatial_decomposition_count)))
 +        s->spatial_decomposition_count--;
 +
 +    if (s->spatial_decomposition_count <= 0) {
 +        av_log(avctx, AV_LOG_ERROR, "Resolution too low\n");
 +        return AVERROR(EINVAL);
 +    }
 +
 +    s->m.pict_type = pic->pict_type;
 +    s->qbias = pic->pict_type == AV_PICTURE_TYPE_P ? 2 : 0;
 +
 +    ff_snow_common_init_after_header(avctx);
 +
 +    if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
 +        for(plane_index=0; plane_index < s->nb_planes; plane_index++){
 +            calculate_visual_weight(s, &s->plane[plane_index]);
 +        }
 +    }
 +
 +    encode_header(s);
 +    s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start);
 +    encode_blocks(s, 1);
 +    s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits;
 +
 +    for(plane_index=0; plane_index < s->nb_planes; plane_index++){
 +        Plane *p= &s->plane[plane_index];
 +        int w= p->width;
 +        int h= p->height;
 +        int x, y;
 +//        int bits= put_bits_count(&s->c.pb);
 +
 +        if (!s->memc_only) {
 +            //FIXME optimize
 +            if(pict->data[plane_index]) //FIXME gray hack
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->spatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
 +                    }
 +                }
 +            predict_plane(s, s->spatial_idwt_buffer, plane_index, 0);
 +
 +            if(   plane_index==0
 +               && pic->pict_type == AV_PICTURE_TYPE_P
 +               && !(avctx->flags&CODEC_FLAG_PASS2)
 +               && s->m.me.scene_change_score > s->avctx->scenechange_threshold){
 +                ff_init_range_encoder(c, pkt->data, pkt->size);
 +                ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
 +                pic->pict_type= AV_PICTURE_TYPE_I;
 +                s->keyframe=1;
 +                s->current_picture->key_frame=1;
 +                goto redo_frame;
 +            }
 +
 +            if(s->qlog == LOSSLESS_QLOG){
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
 +                    }
 +                }
 +            }else{
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->spatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<<ENCODER_EXTRA_BITS;
 +                    }
 +                }
 +            }
 +
 +            ff_spatial_dwt(s->spatial_dwt_buffer, s->temp_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
 +
 +            if(s->pass1_rc && plane_index==0){
 +                int delta_qlog = ratecontrol_1pass(s, pic);
 +                if (delta_qlog <= INT_MIN)
 +                    return -1;
 +                if(delta_qlog){
 +                    //reordering qlog in the bitstream would eliminate this reset
 +                    ff_init_range_encoder(c, pkt->data, pkt->size);
 +                    memcpy(s->header_state, rc_header_bak, sizeof(s->header_state));
 +                    memcpy(s->block_state, rc_block_bak, sizeof(s->block_state));
 +                    encode_header(s);
 +                    encode_blocks(s, 0);
 +                }
 +            }
 +
 +            for(level=0; level<s->spatial_decomposition_count; level++){
 +                for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +                    SubBand *b= &p->band[level][orientation];
 +
 +                    quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias);
 +                    if(orientation==0)
 +                        decorrelate(s, b, b->ibuf, b->stride, pic->pict_type == AV_PICTURE_TYPE_P, 0);
 +                    if (!s->no_bitstream)
 +                    encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation);
 +                    av_assert0(b->parent==NULL || b->parent->stride == b->stride*2);
 +                    if(orientation==0)
 +                        correlate(s, b, b->ibuf, b->stride, 1, 0);
 +                }
 +            }
 +
 +            for(level=0; level<s->spatial_decomposition_count; level++){
 +                for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +                    SubBand *b= &p->band[level][orientation];
 +
 +                    dequantize(s, b, b->ibuf, b->stride);
 +                }
 +            }
 +
 +            ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
 +            if(s->qlog == LOSSLESS_QLOG){
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS;
 +                    }
 +                }
 +            }
 +            predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
 +        }else{
 +            //ME/MC only
 +            if(pic->pict_type == AV_PICTURE_TYPE_I){
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->current_picture->data[plane_index][y*s->current_picture->linesize[plane_index] + x]=
 +                            pict->data[plane_index][y*pict->linesize[plane_index] + x];
 +                    }
 +                }
 +            }else{
 +                memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h);
 +                predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
 +            }
 +        }
 +        if(s->avctx->flags&CODEC_FLAG_PSNR){
 +            int64_t error= 0;
 +
 +            if(pict->data[plane_index]) //FIXME gray hack
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        int d= s->current_picture->data[plane_index][y*s->current_picture->linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x];
 +                        error += d*d;
 +                    }
 +                }
 +            s->avctx->error[plane_index] += error;
 +            s->current_picture->error[plane_index] = error;
 +        }
 +
 +    }
 +
 +    update_last_header_values(s);
 +
 +    ff_snow_release_buffer(avctx);
 +
 +    s->current_picture->coded_picture_number = avctx->frame_number;
 +    s->current_picture->pict_type = pict->pict_type;
 +    s->current_picture->quality = pict->quality;
 +    s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start);
 +    s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits;
 +    s->m.current_picture.f->display_picture_number =
 +    s->m.current_picture.f->coded_picture_number   = avctx->frame_number;
 +    s->m.current_picture.f->quality                = pic->quality;
 +    s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
 +    if(s->pass1_rc)
 +        if (ff_rate_estimate_qscale(&s->m, 0) < 0)
 +            return -1;
 +    if(avctx->flags&CODEC_FLAG_PASS1)
 +        ff_write_pass1_stats(&s->m);
 +    s->m.last_pict_type = s->m.pict_type;
 +    avctx->frame_bits = s->m.frame_bits;
 +    avctx->mv_bits = s->m.mv_bits;
 +    avctx->misc_bits = s->m.misc_bits;
 +    avctx->p_tex_bits = s->m.p_tex_bits;
 +
 +    emms_c();
 +
 +    pkt->size = ff_rac_terminate(c);
 +    if (avctx->coded_frame->key_frame)
 +        pkt->flags |= AV_PKT_FLAG_KEY;
 +    *got_packet = 1;
 +
 +    return 0;
 +}
 +
 +static av_cold int encode_end(AVCodecContext *avctx)
 +{
 +    SnowContext *s = avctx->priv_data;
 +
 +    ff_snow_common_end(s);
 +    ff_rate_control_uninit(&s->m);
 +    av_frame_free(&s->input_picture);
 +    av_free(avctx->stats_out);
 +
 +    return 0;
 +}
 +
 +#define OFFSET(x) offsetof(SnowContext, x)
 +#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
 +static const AVOption options[] = {
 +    { "memc_only",      "Only do ME/MC (I frames -> ref, P frame -> ME+MC).",   OFFSET(memc_only), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
 +    { "no_bitstream",   "Skip final bitstream writeout.",                    OFFSET(no_bitstream), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
 +    { NULL },
 +};
 +
 +static const AVClass snowenc_class = {
 +    .class_name = "snow encoder",
 +    .item_name  = av_default_item_name,
 +    .option     = options,
 +    .version    = LIBAVUTIL_VERSION_INT,
 +};
 +
 +AVCodec ff_snow_encoder = {
 +    .name           = "snow",
 +    .long_name      = NULL_IF_CONFIG_SMALL("Snow"),
 +    .type           = AVMEDIA_TYPE_VIDEO,
 +    .id             = AV_CODEC_ID_SNOW,
 +    .priv_data_size = sizeof(SnowContext),
 +    .init           = encode_init,
 +    .encode2        = encode_frame,
 +    .close          = encode_end,
 +    .pix_fmts       = (const enum AVPixelFormat[]){
 +        AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV444P,
 +        AV_PIX_FMT_GRAY8,
 +        AV_PIX_FMT_NONE
 +    },
 +    .priv_class     = &snowenc_class,
 +};
 +
 +
 +#ifdef TEST
 +#undef malloc
 +#undef free
 +#undef printf
 +
 +#include "libavutil/lfg.h"
 +#include "libavutil/mathematics.h"
 +
 +int main(void){
 +#define width  256
 +#define height 256
 +    int buffer[2][width*height];
 +    SnowContext s;
 +    int i;
 +    AVLFG prng;
 +    s.spatial_decomposition_count=6;
 +    s.spatial_decomposition_type=1;
 +
 +    s.temp_dwt_buffer  = av_mallocz(width * sizeof(DWTELEM));
 +    s.temp_idwt_buffer = av_mallocz(width * sizeof(IDWTELEM));
 +
 +    av_lfg_init(&prng, 1);
 +
 +    printf("testing 5/3 DWT\n");
 +    for(i=0; i<width*height; i++)
 +        buffer[0][i] = buffer[1][i] = av_lfg_get(&prng) % 54321 - 12345;
 +
 +    ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +    ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +
 +    for(i=0; i<width*height; i++)
 +        if(buffer[0][i]!= buffer[1][i]) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]);
 +
 +    printf("testing 9/7 DWT\n");
 +    s.spatial_decomposition_type=0;
 +    for(i=0; i<width*height; i++)
 +        buffer[0][i] = buffer[1][i] = av_lfg_get(&prng) % 54321 - 12345;
 +
 +    ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +    ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +
 +    for(i=0; i<width*height; i++)
 +        if(FFABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]);
 +
 +    {
 +    int level, orientation, x, y;
 +    int64_t errors[8][4];
 +    int64_t g=0;
 +
 +        memset(errors, 0, sizeof(errors));
 +        s.spatial_decomposition_count=3;
 +        s.spatial_decomposition_type=0;
 +        for(level=0; level<s.spatial_decomposition_count; level++){
 +            for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +                int w= width  >> (s.spatial_decomposition_count-level);
 +                int h= height >> (s.spatial_decomposition_count-level);
 +                int stride= width  << (s.spatial_decomposition_count-level);
 +                DWTELEM *buf= buffer[0];
 +                int64_t error=0;
 +
 +                if(orientation&1) buf+=w;
 +                if(orientation>1) buf+=stride>>1;
 +
 +                memset(buffer[0], 0, sizeof(int)*width*height);
 +                buf[w/2 + h/2*stride]= 256*256;
 +                ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +                for(y=0; y<height; y++){
 +                    for(x=0; x<width; x++){
 +                        int64_t d= buffer[0][x + y*width];
 +                        error += d*d;
 +                        if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9 && level==2) printf("%8"PRId64" ", d);
 +                    }
 +                    if(FFABS(height/2-y)<9 && level==2) printf("\n");
 +                }
 +                error= (int)(sqrt(error)+0.5);
 +                errors[level][orientation]= error;
 +                if(g) g=av_gcd(g, error);
 +                else g= error;
 +            }
 +        }
 +        printf("static int const visual_weight[][4]={\n");
 +        for(level=0; level<s.spatial_decomposition_count; level++){
 +            printf("  {");
 +            for(orientation=0; orientation<4; orientation++){
 +                printf("%8"PRId64",", errors[level][orientation]/g);
 +            }
 +            printf("},\n");
 +        }
 +        printf("};\n");
 +        {
 +            int level=2;
 +            int w= width  >> (s.spatial_decomposition_count-level);
 +            //int h= height >> (s.spatial_decomposition_count-level);
 +            int stride= width  << (s.spatial_decomposition_count-level);
 +            DWTELEM *buf= buffer[0];
 +            int64_t error=0;
 +
 +            buf+=w;
 +            buf+=stride>>1;
 +
 +            memset(buffer[0], 0, sizeof(int)*width*height);
 +            for(y=0; y<height; y++){
 +                for(x=0; x<width; x++){
 +                    int tab[4]={0,2,3,1};
 +                    buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)];
 +                }
 +            }
 +            ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +            for(y=0; y<height; y++){
 +                for(x=0; x<width; x++){
 +                    int64_t d= buffer[0][x + y*width];
 +                    error += d*d;
 +                    if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9) printf("%8"PRId64" ", d);
 +                }
 +                if(FFABS(height/2-y)<9) printf("\n");
 +            }
 +        }
 +
 +    }
 +    return 0;
 +}
 +#endif /* TEST */
@@@ -5611,19 -5601,11 +5612,20 @@@ static av_cold int vc1_decode_init(AVCo
          avctx->pix_fmt = AV_PIX_FMT_GRAY8;
      v->s.avctx = avctx;
  
 -    if (ff_vc1_init_common(v) < 0)
 -        return -1;
 +    if ((ret = ff_vc1_init_common(v)) < 0)
 +        return ret;
 +    // ensure static VLC tables are initialized
 +    if ((ret = ff_msmpeg4_decode_init(avctx)) < 0)
 +        return ret;
 +    if ((ret = ff_vc1_decode_init_alloc_tables(v)) < 0)
 +        return ret;
 +    // Hack to ensure the above functions will be called
 +    // again once we know all necessary settings.
 +    // That this is necessary might indicate a bug.
 +    ff_vc1_decode_end(avctx);
 +
      ff_h264chroma_init(&v->h264chroma, 8);
 -    ff_vc1dsp_init(&v->vc1dsp);
+     ff_qpeldsp_init(&s->qdsp);
  
      if (avctx->codec_id == AV_CODEC_ID_WMV3 || avctx->codec_id == AV_CODEC_ID_WMV3IMAGE) {
          int count = 0;
@@@ -6032,54 -5973,16 +6034,54 @@@ static int vc1_decode_frame(AVCodecCont
          s->current_picture_ptr->f->repeat_pict = v->rptfrm * 2;
      }
  
-     s->me.qpel_put = s->dsp.put_qpel_pixels_tab;
-     s->me.qpel_avg = s->dsp.avg_qpel_pixels_tab;
+     s->me.qpel_put = s->qdsp.put_qpel_pixels_tab;
+     s->me.qpel_avg = s->qdsp.avg_qpel_pixels_tab;
  
 -    if (avctx->hwaccel) {
 -        if (avctx->hwaccel->start_frame(avctx, buf, buf_size) < 0)
 -            goto err;
 -        if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
 -            goto err;
 -        if (avctx->hwaccel->end_frame(avctx) < 0)
 -            goto err;
 +    if ((CONFIG_VC1_VDPAU_DECODER)
 +        &&s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU) {
 +        if (v->field_mode && buf_start_second_field) {
 +            ff_vdpau_vc1_decode_picture(s, buf_start, buf_start_second_field - buf_start);
 +            ff_vdpau_vc1_decode_picture(s, buf_start_second_field, (buf + buf_size) - buf_start_second_field);
 +        } else {
 +            ff_vdpau_vc1_decode_picture(s, buf_start, (buf + buf_size) - buf_start);
 +        }
 +    } else if (avctx->hwaccel) {
 +        if (v->field_mode && buf_start_second_field) {
 +            // decode first field
 +            s->picture_structure = PICT_BOTTOM_FIELD - v->tff;
 +            if (avctx->hwaccel->start_frame(avctx, buf_start, buf_start_second_field - buf_start) < 0)
 +                goto err;
 +            if (avctx->hwaccel->decode_slice(avctx, buf_start, buf_start_second_field - buf_start) < 0)
 +                goto err;
 +            if (avctx->hwaccel->end_frame(avctx) < 0)
 +                goto err;
 +
 +            // decode second field
 +            s->gb = slices[n_slices1 + 1].gb;
 +            s->picture_structure = PICT_TOP_FIELD + v->tff;
 +            v->second_field = 1;
 +            v->pic_header_flag = 0;
 +            if (ff_vc1_parse_frame_header_adv(v, &s->gb) < 0) {
 +                av_log(avctx, AV_LOG_ERROR, "parsing header for second field failed");
 +                goto err;
 +            }
 +            v->s.current_picture_ptr->f->pict_type = v->s.pict_type;
 +
 +            if (avctx->hwaccel->start_frame(avctx, buf_start_second_field, (buf + buf_size) - buf_start_second_field) < 0)
 +                goto err;
 +            if (avctx->hwaccel->decode_slice(avctx, buf_start_second_field, (buf + buf_size) - buf_start_second_field) < 0)
 +                goto err;
 +            if (avctx->hwaccel->end_frame(avctx) < 0)
 +                goto err;
 +        } else {
 +            s->picture_structure = PICT_FRAME;
 +            if (avctx->hwaccel->start_frame(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
 +                goto err;
 +            if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
 +                goto err;
 +            if (avctx->hwaccel->end_frame(avctx) < 0)
 +                goto err;
 +        }
      } else {
          int header_ret = 0;
  
   *
   */
  
 +#include "libavutil/avassert.h"
  #include "libavutil/common.h"
- #include "dsputil.h"
 +#include "libavutil/intreadwrite.h"
  #include "h264chroma.h"
+ #include "qpeldsp.h"
 +#include "rnd_avg.h"
  #include "vc1dsp.h"
 +#include "startcode.h"
  
  /* Apply overlap transform to horizontal edge */
  static void vc1_v_overlap_c(uint8_t *src, int stride)
Simple merge
@@@ -68,16 -61,10 +69,13 @@@ YASM-OBJS                              
  
  YASM-OBJS-$(CONFIG_AC3DSP)             += x86/ac3dsp.o
  YASM-OBJS-$(CONFIG_DCT)                += x86/dct32.o
 +YASM-OBJS-$(CONFIG_DIRAC_DECODER)      += x86/diracdsp_mmx.o x86/diracdsp_yasm.o\
 +                                          x86/dwt_yasm.o
  YASM-OBJS-$(CONFIG_DNXHD_ENCODER)      += x86/dnxhdenc.o
- YASM-OBJS-$(CONFIG_DSPUTIL)            += x86/dsputil.o                 \
-                                           x86/fpel.o                    \
-                                           x86/mpeg4qpel.o               \
-                                           x86/qpel.o
+ YASM-OBJS-$(CONFIG_DSPUTIL)            += x86/dsputil.o
  YASM-OBJS-$(CONFIG_ENCODERS)           += x86/dsputilenc.o
  YASM-OBJS-$(CONFIG_FFT)                += x86/fft.o
 +YASM-OBJS-$(CONFIG_FLAC_DECODER)       += x86/flacdsp.o
  YASM-OBJS-$(CONFIG_H263DSP)            += x86/h263_loopfilter.o
  YASM-OBJS-$(CONFIG_H264CHROMA)         += x86/h264_chromamc.o           \
                                            x86/h264_chromamc_10bit.o
@@@ -98,8 -83,10 +96,11 @@@ YASM-OBJS-$(CONFIG_HEVC_DECODER)       
  YASM-OBJS-$(CONFIG_HPELDSP)            += x86/fpel.o                    \
                                            x86/hpeldsp.o
  YASM-OBJS-$(CONFIG_HUFFYUVDSP)         += x86/huffyuvdsp.o
 +YASM-OBJS-$(CONFIG_LLVIDDSP)           += x86/lossless_videodsp.o
  YASM-OBJS-$(CONFIG_MPEGAUDIODSP)       += x86/imdct36.o
+ YASM-OBJS-$(CONFIG_QPELDSP)            += x86/qpeldsp.o                 \
+                                           x86/fpel.o                    \
+                                           x86/qpel.o
  YASM-OBJS-$(CONFIG_VIDEODSP)           += x86/videodsp.o
  YASM-OBJS-$(CONFIG_VP3DSP)             += x86/vp3dsp.o
  
  #include "libavcodec/simple_idct.h"
  #include "libavcodec/version.h"
  #include "dsputil_x86.h"
- #include "fpel.h"
  #include "idct_xvid.h"
  
- void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                               int dstStride, int src1Stride, int h);
- void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1,
-                                      uint8_t *src2, int dstStride,
-                                      int src1Stride, int h);
- void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                               int dstStride, int src1Stride, int h);
- void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                                int dstStride, int src1Stride, int h);
- void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                                int dstStride, int src1Stride, int h);
- void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
-                                       int dstStride, int src1Stride, int h);
- void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                           int dstStride, int srcStride, int h);
- void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                           int dstStride, int srcStride, int h);
- void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                                  int dstStride, int srcStride,
-                                                  int h);
- void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                          int dstStride, int srcStride, int h);
- void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                          int dstStride, int srcStride, int h);
- void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                                 int dstStride, int srcStride,
-                                                 int h);
- void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                           int dstStride, int srcStride);
- void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                           int dstStride, int srcStride);
- void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                                  int dstStride, int srcStride);
- void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                          int dstStride, int srcStride);
- void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                          int dstStride, int srcStride);
- void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
-                                                 int dstStride, int srcStride);
- #define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmx
- #define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmx
 +
  int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2,
                                        int order);
  int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2,
Simple merge
index 0000000,435b765..8f296a1
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,501 +1,501 @@@
 - * This file is part of Libav.
+ /*
+  * quarterpel DSP functions
+  *
 - * Libav is free software; you can redistribute it and/or
++ * This file is part of FFmpeg.
+  *
 - * Libav is distributed in the hope that it will be useful,
++ * 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.
+  *
 - * License along with Libav; if not, write to the Free Software
++ * 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
 -#define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmxext
 -#define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmxext
++ * License along with FFmpeg; if not, write to the Free Software
+  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+  */
+ #include <stddef.h>
+ #include <stdint.h>
+ #include "config.h"
+ #include "libavutil/attributes.h"
+ #include "libavutil/cpu.h"
+ #include "libavutil/x86/cpu.h"
+ #include "libavcodec/pixels.h"
+ #include "libavcodec/qpeldsp.h"
+ #include "fpel.h"
+ void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                               int dstStride, int src1Stride, int h);
+ void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1,
+                                      uint8_t *src2, int dstStride,
+                                      int src1Stride, int h);
+ void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                               int dstStride, int src1Stride, int h);
+ void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                                int dstStride, int src1Stride, int h);
+ void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                                int dstStride, int src1Stride, int h);
+ void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
+                                       int dstStride, int src1Stride, int h);
+ void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                           int dstStride, int srcStride, int h);
+ void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                           int dstStride, int srcStride, int h);
+ void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                                  int dstStride, int srcStride,
+                                                  int h);
+ void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                          int dstStride, int srcStride, int h);
+ void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                          int dstStride, int srcStride, int h);
+ void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                                 int dstStride, int srcStride,
+                                                 int h);
+ void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                           int dstStride, int srcStride);
+ void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                           int dstStride, int srcStride);
+ void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                                  int dstStride, int srcStride);
+ void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                          int dstStride, int srcStride);
+ void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                          int dstStride, int srcStride);
+ void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
+                                                 int dstStride, int srcStride);
 -CALL_2X_PIXELS(ff_avg_pixels16_mmxext, ff_avg_pixels8_mmxext, 8)
 -CALL_2X_PIXELS(ff_put_pixels16_mmxext, ff_put_pixels8_mmxext, 8)
++#define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmx
++#define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmx
+ #if HAVE_YASM
++#define ff_put_pixels16_mmxext ff_put_pixels16_mmx
++#define ff_put_pixels8_mmxext  ff_put_pixels8_mmx
+ #define QPEL_OP(OPNAME, RND, MMX)                                       \
+ static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);              \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t temp[8];                                                   \
+     uint8_t *const half = (uint8_t *) temp;                             \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
+                                                    stride, 8);          \
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
+                                         stride, stride, 8);             \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride,    \
+                                                    stride, 8);          \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t temp[8];                                                   \
+     uint8_t *const half = (uint8_t *) temp;                             \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
+                                                    stride, 8);          \
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride,     \
+                                         stride, 8);                     \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t temp[8];                                                   \
+     uint8_t *const half = (uint8_t *) temp;                             \
+     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
+                                                    8, stride);          \
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
+                                         stride, stride, 8);             \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src,            \
+                                                    stride, stride);     \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t temp[8];                                                   \
+     uint8_t *const half = (uint8_t *) temp;                             \
+     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
+                                                    8, stride);          \
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\
+                                         stride, 8);                     \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[8 + 9];                                               \
+     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
+                                         stride, 9);                     \
+     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
+                                         stride, 8, 8);                  \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[8 + 9];                                               \
+     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
+                                         stride, 9);                     \
+     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
+                                         stride, 8, 8);                  \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[8 + 9];                                               \
+     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
+                                         stride, 9);                     \
+     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
+                                         stride, 8, 8);                  \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[8 + 9];                                               \
+     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
+                                         stride, 9);                     \
+     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
+                                         stride, 8, 8);                  \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[8 + 9];                                               \
+     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
+                                         stride, 8, 8);                  \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[8 + 9];                                               \
+     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
+                                         stride, 8, 8);                  \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[8 + 9];                                               \
+     uint8_t *const halfH = (uint8_t *) half;                            \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH,              \
+                                         8, stride, 9);                  \
+     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
+                                                    stride, 8);          \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[8 + 9];                                               \
+     uint8_t *const halfH = (uint8_t *) half;                            \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
+                                         stride, 9);                     \
+     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
+                                                    stride, 8);          \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src,    \
+                                          ptrdiff_t stride)              \
+ {                                                                       \
+     uint64_t half[9];                                                   \
+     uint8_t *const halfH = (uint8_t *) half;                            \
+     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
+                                                    stride, 9);          \
+     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
+                                                    stride, 8);          \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc00_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);            \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t temp[32];                                                  \
+     uint8_t *const half = (uint8_t *) temp;                             \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
+                                                     stride, 16);        \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
+                                          stride, 16);                   \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src,           \
+                                                     stride, stride, 16);\
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t temp[32];                                                  \
+     uint8_t *const half = (uint8_t*) temp;                              \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
+                                                     stride, 16);        \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half,            \
+                                          stride, stride, 16);           \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t temp[32];                                                  \
+     uint8_t *const half = (uint8_t *) temp;                             \
+     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
+                                                     stride);            \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
+                                          stride, 16);                   \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src,           \
+                                                     stride, stride);    \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t temp[32];                                                  \
+     uint8_t *const half = (uint8_t *) temp;                             \
+     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
+                                                     stride);            \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half,         \
+                                          stride, stride, 16);           \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[16 * 2 + 17 * 2];                                     \
+     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
+                                          stride, 17);                   \
+     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                     16, 16);            \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
+                                          stride, 16, 16);               \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[16 * 2 + 17 * 2];                                     \
+     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
+                                          stride, 17);                   \
+     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                     16, 16);            \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
+                                          stride, 16, 16);               \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[16 * 2 + 17 * 2];                                     \
+     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
+                                          stride, 17);                   \
+     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                     16, 16);            \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
+                                          stride, 16, 16);               \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[16 * 2 + 17 * 2];                                     \
+     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
+                                          stride, 17);                   \
+     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                     16, 16);            \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
+                                          stride, 16, 16);               \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[16 * 2 + 17 * 2];                                     \
+     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                     16, 16);            \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
+                                          stride, 16, 16);               \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[16 * 2 + 17 * 2];                                     \
+     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
+     uint8_t *const halfHV = (uint8_t *) half;                           \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
+                                                     16, 16);            \
+     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
+                                          stride, 16, 16);               \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[17 * 2];                                              \
+     uint8_t *const halfH = (uint8_t *) half;                            \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
+                                          stride, 17);                   \
+     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
+                                                     stride, 16);        \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[17 * 2];                                              \
+     uint8_t *const halfH = (uint8_t *) half;                            \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
+                                          stride, 17);                   \
+     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
+                                                     stride, 16);        \
+ }                                                                       \
+                                                                         \
+ static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src,   \
+                                           ptrdiff_t stride)             \
+ {                                                                       \
+     uint64_t half[17 * 2];                                              \
+     uint8_t *const halfH = (uint8_t *) half;                            \
+     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
+                                                     stride, 17);        \
+     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
+                                                     stride, 16);        \
+ }
+ QPEL_OP(put_,        _,        mmxext)
+ QPEL_OP(avg_,        _,        mmxext)
+ QPEL_OP(put_no_rnd_, _no_rnd_, mmxext)
+ #endif /* HAVE_YASM */
+ #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
+ do {                                                                         \
+     c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
+     c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
+ } while (0)
+ av_cold void ff_qpeldsp_init_x86(QpelDSPContext *c)
+ {
+     int cpu_flags = av_get_cpu_flags();
+     if (X86_MMXEXT(cpu_flags)) {
+ #if HAVE_MMXEXT_EXTERNAL
+         SET_QPEL_FUNCS(avg_qpel,        0, 16, mmxext, );
+         SET_QPEL_FUNCS(avg_qpel,        1,  8, mmxext, );
+         SET_QPEL_FUNCS(put_qpel,        0, 16, mmxext, );
+         SET_QPEL_FUNCS(put_qpel,        1,  8, mmxext, );
+         SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, );
+         SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, mmxext, );
+ #endif /* HAVE_MMXEXT_EXTERNAL */
+     }
+ }