Merge commit 'a1bcc76e6036e78f25cbb7323c145056cfca9d93'
authorMichael Niedermayer <michaelni@gmx.at>
Fri, 24 Aug 2012 12:22:38 +0000 (14:22 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Fri, 24 Aug 2012 12:30:40 +0000 (14:30 +0200)
* commit 'a1bcc76e6036e78f25cbb7323c145056cfca9d93': (21 commits)
  cmdutils: fix a memleak when specifying an option twice.
  x86: mpegvideo: more sensible names for optimization file and init function
  x86: mpegvideoenc: Split optimizations off into a separate file
  dnxhdenc: x86: more sensible names for optimization file and init function
  svq1/svq3: Move common code out of SVQ1 decoder-specific file
  dirac: add Comments and references to the standard
  lavr: x86: optimized 6-channel flt to fltp conversion
  lavr: x86: optimized 2-channel flt to fltp conversion
  lavr: x86: optimized 6-channel flt to s16p conversion
  lavr: x86: optimized 2-channel flt to s16p conversion
  lavr: x86: optimized 6-channel s16 to fltp conversion
  lavr: x86: optimized 2-channel s16 to fltp conversion
  lavr: x86: optimized 6-channel s16 to s16p conversion
  lavr: x86: optimized 2-channel s16 to s16p conversion
  lavr: x86: optimized 2-channel fltp to flt conversion
  lavr: x86: optimized 6-channel fltp to s16 conversion
  lavr: x86: optimized 2-channel fltp to s16 conversion
  lavr: x86: optimized 6-channel s16p to flt conversion
  lavr: x86: optimized 2-channel s16p to flt conversion
  lavr: x86: optimized 6-channel s16p to s16 conversion
  ...

Conflicts:
libavcodec/dirac.c
libavcodec/mpegvideo.h
libavcodec/x86/Makefile

Merged-by: Michael Niedermayer <michaelni@gmx.at>
16 files changed:
1  2 
cmdutils.c
libavcodec/Makefile
libavcodec/dirac.c
libavcodec/dnxhdenc.c
libavcodec/dnxhdenc.h
libavcodec/mpegvideo.c
libavcodec/mpegvideo.h
libavcodec/mpegvideo_enc.c
libavcodec/svq13.c
libavcodec/svq1dec.c
libavcodec/x86/Makefile
libavcodec/x86/dnxhdenc.c
libavcodec/x86/mpegvideo.c
libavcodec/x86/mpegvideoenc.c
libavcodec/x86/mpegvideoenc_template.c
libavutil/x86/x86util.asm

diff --cc cmdutils.c
@@@ -301,6 -271,7 +301,7 @@@ int parse_option(void *optctx, const ch
      if (po->flags & OPT_STRING) {
          char *str;
          str = av_strdup(arg);
 -        av_freep(dst);
++//         av_freep(dst);
          *(char **)dst = str;
      } else if (po->flags & OPT_BOOL) {
          *(int *)dst = bool_val;
@@@ -394,25 -347,17 +394,24 @@@ OBJS-$(CONFIG_SNOW_DECODER)            
  OBJS-$(CONFIG_SNOW_ENCODER)            += snowenc.o snow.o              \
                                            h263.o ituh263enc.o
  OBJS-$(CONFIG_SOL_DPCM_DECODER)        += dpcm.o
 +OBJS-$(CONFIG_SONIC_DECODER)           += sonic.o
 +OBJS-$(CONFIG_SONIC_ENCODER)           += sonic.o
 +OBJS-$(CONFIG_SONIC_LS_ENCODER)        += sonic.o
  OBJS-$(CONFIG_SP5X_DECODER)            += sp5xdec.o mjpegdec.o mjpeg.o
  OBJS-$(CONFIG_SRT_DECODER)             += srtdec.o ass.o
 +OBJS-$(CONFIG_SRT_ENCODER)             += srtenc.o ass_split.o
 +OBJS-$(CONFIG_SUBRIP_DECODER)          += srtdec.o ass.o
 +OBJS-$(CONFIG_SUBRIP_ENCODER)          += srtenc.o ass_split.o
 +OBJS-$(CONFIG_SUBVIEWER_DECODER)       += subviewerdec.o ass.o
  OBJS-$(CONFIG_SUNRAST_DECODER)         += sunrast.o
  OBJS-$(CONFIG_SUNRAST_ENCODER)         += sunrastenc.o
- OBJS-$(CONFIG_SVQ1_DECODER)            += svq1dec.o svq1.o h263.o
+ OBJS-$(CONFIG_SVQ1_DECODER)            += svq1dec.o svq1.o svq13.o h263.o
  OBJS-$(CONFIG_SVQ1_ENCODER)            += svq1enc.o svq1.o    \
                                            h263.o ituh263enc.o
- OBJS-$(CONFIG_SVQ3_DECODER)            += h264.o svq3.o                       \
+ OBJS-$(CONFIG_SVQ3_DECODER)            += svq3.o svq13.o h263.o h264.o        \
                                            h264_loopfilter.o h264_direct.o     \
                                            h264_sei.o h264_ps.o h264_refs.o    \
-                                           h264_cavlc.o h264_cabac.o cabac.o   \
-                                           svq1dec.o svq1.o h263.o
+                                           h264_cavlc.o h264_cabac.o cabac.o
  OBJS-$(CONFIG_TARGA_DECODER)           += targa.o
  OBJS-$(CONFIG_TARGA_ENCODER)           += targaenc.o rle.o
  OBJS-$(CONFIG_THEORA_DECODER)          += xiph.o
@@@ -138,47 -129,57 +128,57 @@@ static int parse_source_parameters(AVCo
      }
  
      /* [DIRAC_STD] 10.3.3 Chroma Sampling Format.
-        chroma_sampling_format(video_params) */
-     if (get_bits1(gb))                                  /* [DIRAC_STD] custom_chroma_format_flag */
-         source->chroma_format = svq3_get_ue_golomb(gb); /*[DIRAC_STD] CHROMA_FORMAT_INDEX        */
+      *  chroma_sampling_format(video_params) */
+     /* [DIRAC_STD] custom_chroma_format_flag */
+     if (get_bits1(gb))
+         /* [DIRAC_STD] CHROMA_FORMAT_INDEX */
+         source->chroma_format = svq3_get_ue_golomb(gb);
 -    if (source->chroma_format > 2) {
 +    if (source->chroma_format > 2U) {
          av_log(avctx, AV_LOG_ERROR, "Unknown chroma format %d\n",
                 source->chroma_format);
          return AVERROR_INVALIDDATA;
      }
  
      /* [DIRAC_STD] 10.3.4 Scan Format. scan_format(video_params) */
-     if (get_bits1(gb))                               /* [DIRAC_STD] custom_scan_format_flag */
-         source->interlaced = svq3_get_ue_golomb(gb); /* [DIRAC_STD] SOURCE_SAMPLING         */
+     /* [DIRAC_STD] custom_scan_format_flag */
+     if (get_bits1(gb))
+         /* [DIRAC_STD] SOURCE_SAMPLING */
+         source->interlaced = svq3_get_ue_golomb(gb);
 -    if (source->interlaced > 1)
 +    if (source->interlaced > 1U)
          return AVERROR_INVALIDDATA;
  
      /* [DIRAC_STD] 10.3.5 Frame Rate. frame_rate(video_params) */
      if (get_bits1(gb)) { /* [DIRAC_STD] custom_frame_rate_flag */
          source->frame_rate_index = svq3_get_ue_golomb(gb);
  
 -        if (source->frame_rate_index > 10)
 +        if (source->frame_rate_index > 10U)
              return AVERROR_INVALIDDATA;
  
-         if (!source->frame_rate_index){
-             frame_rate.num = svq3_get_ue_golomb(gb); /* [DIRAC_STD] FRAME_RATE_NUMER */
-             frame_rate.den = svq3_get_ue_golomb(gb); /* [DIRAC_STD] FRAME_RATE_DENOM */
+         if (!source->frame_rate_index) {
+             /* [DIRAC_STD] FRAME_RATE_NUMER */
+             frame_rate.num = svq3_get_ue_golomb(gb);
+             /* [DIRAC_STD] FRAME_RATE_DENOM */
+             frame_rate.den = svq3_get_ue_golomb(gb);
          }
      }
-     if (source->frame_rate_index > 0) { /* [DIRAC_STD] preset_frame_rate(video_params,index) */
+     /* [DIRAC_STD] preset_frame_rate(video_params, index) */
+     if (source->frame_rate_index > 0) {
          if (source->frame_rate_index <= 8)
 -            frame_rate = avpriv_frame_rate_tab[source->frame_rate_index];
 +            frame_rate = avpriv_frame_rate_tab[source->frame_rate_index];  /* [DIRAC_STD] Table 10.3 values 1-8  */
          else
-             frame_rate = dirac_frame_rate[source->frame_rate_index-9]; /* [DIRAC_STD] Table 10.3 values 9-10 */
+              /* [DIRAC_STD] Table 10.3 values 9-10 */
+             frame_rate = dirac_frame_rate[source->frame_rate_index-9];
      }
      av_reduce(&avctx->time_base.num, &avctx->time_base.den,
                frame_rate.den, frame_rate.num, 1<<30);
  
-     /* [DIRAC_STD] 10.3.6 Pixel Aspect Ratio. pixel_aspect_ratio(video_params) */
+     /* [DIRAC_STD] 10.3.6 Pixel Aspect Ratio.
+      * pixel_aspect_ratio(video_params) */
      if (get_bits1(gb)) { /* [DIRAC_STD] custom_pixel_aspect_ratio_flag */
-         source->aspect_ratio_index = svq3_get_ue_golomb(gb); /* [DIRAC_STD] index */
+          /* [DIRAC_STD] index */
+         source->aspect_ratio_index = svq3_get_ue_golomb(gb);
  
 -        if (source->aspect_ratio_index > 6)
 +        if (source->aspect_ratio_index > 6U)
              return AVERROR_INVALIDDATA;
  
          if (!source->aspect_ratio_index) {
              avctx->sample_aspect_ratio.den = svq3_get_ue_golomb(gb);
          }
      }
-     if (source->aspect_ratio_index > 0) /* [DIRAC_STD] Take value from Table 10.4 Available preset pixel aspect ratio values */
+     /* [DIRAC_STD] Take value from Table 10.4 Available preset pixel
+      *  aspect ratio values */
+     if (source->aspect_ratio_index > 0)
          avctx->sample_aspect_ratio =
 -                dirac_preset_aspect_ratios[source->aspect_ratio_index-1];
 +            dirac_preset_aspect_ratios[source->aspect_ratio_index-1];
  
      /* [DIRAC_STD] 10.3.7 Clean area. clean_area(video_params) */
      if (get_bits1(gb)) { /* [DIRAC_STD] custom_clean_area_flag */
-         source->clean_width        = svq3_get_ue_golomb(gb); /* [DIRAC_STD] CLEAN_WIDTH        */
-         source->clean_height       = svq3_get_ue_golomb(gb); /* [DIRAC_STD] CLEAN_HEIGHT       */
-         source->clean_left_offset  = svq3_get_ue_golomb(gb); /* [DIRAC_STD] CLEAN_LEFT_OFFSET  */
-         source->clean_right_offset = svq3_get_ue_golomb(gb); /* [DIRAC_STD] CLEAN_RIGHT_OFFSET */
+         /* [DIRAC_STD] CLEAN_WIDTH */
+         source->clean_width        = svq3_get_ue_golomb(gb);
+          /* [DIRAC_STD] CLEAN_HEIGHT */
+         source->clean_height       = svq3_get_ue_golomb(gb);
+          /* [DIRAC_STD] CLEAN_LEFT_OFFSET */
+         source->clean_left_offset  = svq3_get_ue_golomb(gb);
+         /* [DIRAC_STD] CLEAN_RIGHT_OFFSET */
+         source->clean_right_offset = svq3_get_ue_golomb(gb);
      }
  
-     /*[DIRAC_STD] 10.3.8 Signal range. signal_range(video_params)
-       WARNING: Some adaptation seemed to be done using the AVCOL_RANGE_MPEG/JPEG values */
-     if (get_bits1(gb)) {                                    /*[DIRAC_STD] custom_signal_range_flag */
-         source->pixel_range_index = svq3_get_ue_golomb(gb); /*[DIRAC_STD] index */
+     /* [DIRAC_STD] 10.3.8 Signal range. signal_range(video_params)
+      * WARNING: Some adaptation seems to be done using the
+      * AVCOL_RANGE_MPEG/JPEG values */
+     if (get_bits1(gb)) { /* [DIRAC_STD] custom_signal_range_flag */
+         /* [DIRAC_STD] index */
+         source->pixel_range_index = svq3_get_ue_golomb(gb);
  
 -        if (source->pixel_range_index > 4)
 +        if (source->pixel_range_index > 4U)
              return AVERROR_INVALIDDATA;
  
 -        // This assumes either fullrange or MPEG levels only
 +        /* This assumes either fullrange or MPEG levels only */
          if (!source->pixel_range_index) {
              luma_offset = svq3_get_ue_golomb(gb);
              luma_depth  = av_log2(svq3_get_ue_golomb(gb))+1;
  
      /* [DIRAC_STD] 10.3.9 Colour specification. colour_spec(video_params) */
      if (get_bits1(gb)) { /* [DIRAC_STD] custom_colour_spec_flag */
-         idx = source->color_spec_index = svq3_get_ue_golomb(gb); /* [DIRAC_STD] index */
+          /* [DIRAC_STD] index */
+         idx = source->color_spec_index = svq3_get_ue_golomb(gb);
  
 -        if (source->color_spec_index > 4)
 +        if (source->color_spec_index > 4U)
              return AVERROR_INVALIDDATA;
  
          avctx->color_primaries = dirac_color_presets[idx].color_primaries;
          avctx->color_trc       = dirac_color_presets[idx].color_trc;
  
          if (!source->color_spec_index) {
-             /* [DIRAC_STD] 10.3.9.1 Color primaries */
+             /* [DIRAC_STD] 10.3.9.1 Colour primaries */
              if (get_bits1(gb)) {
                  idx = svq3_get_ue_golomb(gb);
 -                if (idx < 3)
 +                if (idx < 3U)
                      avctx->color_primaries = dirac_primaries[idx];
              }
-             /* [DIRAC_STD] 10.3.9.2 Color matrix */
+             /* [DIRAC_STD] 10.3.9.2 Colour matrix */
              if (get_bits1(gb)) {
                  idx = svq3_get_ue_golomb(gb);
                  if (!idx)
@@@ -292,13 -300,14 +299,14 @@@ int avpriv_dirac_parse_sequence_header(
      else if (version_major > 2)
          av_log(avctx, AV_LOG_WARNING, "Stream may have unhandled features\n");
  
 -    if (video_format > 20)
 +    if (video_format > 20U)
          return AVERROR_INVALIDDATA;
  
 -    // Fill in defaults for the source parameters.
 +    /* Fill in defaults for the source parameters. */
      *source = dirac_source_parameters_defaults[video_format];
  
-     // Override the defaults.
+     /* [DIRAC_STD] 10.3 Source Parameters
+      * Override the defaults. */
      if (ret = parse_source_parameters(avctx, gb, source))
          return ret;
  
Simple merge
Simple merge
Simple merge
@@@ -764,8 -755,9 +764,9 @@@ void ff_MPV_frame_end(MpegEncContext *s
  int ff_MPV_encode_init(AVCodecContext *avctx);
  int ff_MPV_encode_end(AVCodecContext *avctx);
  int ff_MPV_encode_picture(AVCodecContext *avctx, AVPacket *pkt,
 -                          const AVFrame *frame, int *got_packet);
 +                          AVFrame *frame, int *got_packet);
- void ff_MPV_common_init_mmx(MpegEncContext *s);
+ void ff_MPV_encode_init_x86(MpegEncContext *s);
+ void ff_MPV_common_init_x86(MpegEncContext *s);
  void ff_MPV_common_init_axp(MpegEncContext *s);
  void ff_MPV_common_init_mmi(MpegEncContext *s);
  void ff_MPV_common_init_arm(MpegEncContext *s);
Simple merge
index 0000000,5d9d498..65a7902
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,68 +1,68 @@@
 - * This file is part of Libav.
+ /*
+  * SVQ1/SVQ3 decoder common code
+  *
 - * 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
+  */
+ #include <stdint.h>
+ #include "svq1.h"
+ static const uint16_t checksum_table[256] = {
+   0x0000, 0x1021, 0x2042, 0x3063, 0x4084, 0x50A5, 0x60C6, 0x70E7,
+   0x8108, 0x9129, 0xA14A, 0xB16B, 0xC18C, 0xD1AD, 0xE1CE, 0xF1EF,
+   0x1231, 0x0210, 0x3273, 0x2252, 0x52B5, 0x4294, 0x72F7, 0x62D6,
+   0x9339, 0x8318, 0xB37B, 0xA35A, 0xD3BD, 0xC39C, 0xF3FF, 0xE3DE,
+   0x2462, 0x3443, 0x0420, 0x1401, 0x64E6, 0x74C7, 0x44A4, 0x5485,
+   0xA56A, 0xB54B, 0x8528, 0x9509, 0xE5EE, 0xF5CF, 0xC5AC, 0xD58D,
+   0x3653, 0x2672, 0x1611, 0x0630, 0x76D7, 0x66F6, 0x5695, 0x46B4,
+   0xB75B, 0xA77A, 0x9719, 0x8738, 0xF7DF, 0xE7FE, 0xD79D, 0xC7BC,
+   0x48C4, 0x58E5, 0x6886, 0x78A7, 0x0840, 0x1861, 0x2802, 0x3823,
+   0xC9CC, 0xD9ED, 0xE98E, 0xF9AF, 0x8948, 0x9969, 0xA90A, 0xB92B,
+   0x5AF5, 0x4AD4, 0x7AB7, 0x6A96, 0x1A71, 0x0A50, 0x3A33, 0x2A12,
+   0xDBFD, 0xCBDC, 0xFBBF, 0xEB9E, 0x9B79, 0x8B58, 0xBB3B, 0xAB1A,
+   0x6CA6, 0x7C87, 0x4CE4, 0x5CC5, 0x2C22, 0x3C03, 0x0C60, 0x1C41,
+   0xEDAE, 0xFD8F, 0xCDEC, 0xDDCD, 0xAD2A, 0xBD0B, 0x8D68, 0x9D49,
+   0x7E97, 0x6EB6, 0x5ED5, 0x4EF4, 0x3E13, 0x2E32, 0x1E51, 0x0E70,
+   0xFF9F, 0xEFBE, 0xDFDD, 0xCFFC, 0xBF1B, 0xAF3A, 0x9F59, 0x8F78,
+   0x9188, 0x81A9, 0xB1CA, 0xA1EB, 0xD10C, 0xC12D, 0xF14E, 0xE16F,
+   0x1080, 0x00A1, 0x30C2, 0x20E3, 0x5004, 0x4025, 0x7046, 0x6067,
+   0x83B9, 0x9398, 0xA3FB, 0xB3DA, 0xC33D, 0xD31C, 0xE37F, 0xF35E,
+   0x02B1, 0x1290, 0x22F3, 0x32D2, 0x4235, 0x5214, 0x6277, 0x7256,
+   0xB5EA, 0xA5CB, 0x95A8, 0x8589, 0xF56E, 0xE54F, 0xD52C, 0xC50D,
+   0x34E2, 0x24C3, 0x14A0, 0x0481, 0x7466, 0x6447, 0x5424, 0x4405,
+   0xA7DB, 0xB7FA, 0x8799, 0x97B8, 0xE75F, 0xF77E, 0xC71D, 0xD73C,
+   0x26D3, 0x36F2, 0x0691, 0x16B0, 0x6657, 0x7676, 0x4615, 0x5634,
+   0xD94C, 0xC96D, 0xF90E, 0xE92F, 0x99C8, 0x89E9, 0xB98A, 0xA9AB,
+   0x5844, 0x4865, 0x7806, 0x6827, 0x18C0, 0x08E1, 0x3882, 0x28A3,
+   0xCB7D, 0xDB5C, 0xEB3F, 0xFB1E, 0x8BF9, 0x9BD8, 0xABBB, 0xBB9A,
+   0x4A75, 0x5A54, 0x6A37, 0x7A16, 0x0AF1, 0x1AD0, 0x2AB3, 0x3A92,
+   0xFD2E, 0xED0F, 0xDD6C, 0xCD4D, 0xBDAA, 0xAD8B, 0x9DE8, 0x8DC9,
+   0x7C26, 0x6C07, 0x5C64, 0x4C45, 0x3CA2, 0x2C83, 0x1CE0, 0x0CC1,
+   0xEF1F, 0xFF3E, 0xCF5D, 0xDF7C, 0xAF9B, 0xBFBA, 0x8FD9, 0x9FF8,
+   0x6E17, 0x7E36, 0x4E55, 0x5E74, 0x2E93, 0x3EB2, 0x0ED1, 0x1EF0
+ };
+ uint16_t ff_svq1_packet_checksum (const uint8_t *data, const int length, int value) {
+   int i;
+   for (i=0; i < length; i++) {
+     value = checksum_table[data[i] ^ (value >> 8)] ^ ((value & 0xFF) << 8);
+   }
+   return value;
+ }
Simple merge
@@@ -14,20 -14,18 +14,21 @@@ MMX-OBJS                               
  MMX-OBJS-$(CONFIG_AAC_DECODER)         += x86/sbrdsp_init.o
  MMX-OBJS-$(CONFIG_AC3DSP)              += x86/ac3dsp_init.o
  MMX-OBJS-$(CONFIG_CAVS_DECODER)        += x86/cavsdsp_mmx.o
- MMX-OBJS-$(CONFIG_DNXHD_ENCODER)       += x86/dnxhd_mmx.o
+ MMX-OBJS-$(CONFIG_DNXHD_ENCODER)       += x86/dnxhdenc.o
 -MMX-OBJS-$(CONFIG_DWT)                 += x86/snowdsp_mmx.o
 +MMX-OBJS-$(CONFIG_DWT)                 += x86/snowdsp_mmx.o \
 +                                          x86/dwt.o
  MMX-OBJS-$(CONFIG_ENCODERS)            += x86/dsputilenc_mmx.o
  MMX-OBJS-$(CONFIG_FFT)                 += x86/fft_init.o
 +MMX-OBJS-$(CONFIG_GPL)                 += x86/idct_mmx.o
  MMX-OBJS-$(CONFIG_H264DSP)             += x86/h264dsp_init.o
  MMX-OBJS-$(CONFIG_H264PRED)            += x86/h264_intrapred_init.o
  MMX-OBJS-$(CONFIG_LPC)                 += x86/lpc_mmx.o
  MMX-OBJS-$(CONFIG_MPEGAUDIODSP)        += x86/mpegaudiodec_mmx.o
- MMX-OBJS-$(CONFIG_MPEGVIDEO)           += x86/mpegvideo_mmx.o
+ MMX-OBJS-$(CONFIG_MPEGVIDEO)           += x86/mpegvideo.o
+ MMX-OBJS-$(CONFIG_MPEGVIDEOENC)        += x86/mpegvideoenc.o
  MMX-OBJS-$(CONFIG_PNG_DECODER)         += x86/pngdsp_init.o
  MMX-OBJS-$(CONFIG_PRORES_DECODER)      += x86/proresdsp_init.o
 +MMX-OBJS-$(CONFIG_PRORES_LGPL_DECODER) += x86/proresdsp_init.o
  MMX-OBJS-$(CONFIG_RV30_DECODER)        += x86/rv34dsp_init.o
  MMX-OBJS-$(CONFIG_RV40_DECODER)        += x86/rv34dsp_init.o            \
                                            x86/rv40dsp_init.o
index 0000000,c344afe..8b0c2ad
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,65 +1,65 @@@
 - * This file is part of Libav.
+ /*
+  * VC3/DNxHD SIMD functions
+  * Copyright (c) 2007 Baptiste Coudurier <baptiste dot coudurier at smartjog dot com>
+  *
+  * VC-3 encoder funded by the British Broadcasting Corporation
+  *
 - * 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
+  */
+ #include "libavutil/x86/asm.h"
+ #include "libavcodec/dnxhdenc.h"
+ #if HAVE_INLINE_ASM
+ static void get_pixels_8x4_sym_sse2(DCTELEM *block, const uint8_t *pixels, int line_size)
+ {
+     __asm__ volatile(
+         "pxor %%xmm5,      %%xmm5       \n\t"
+         "movq (%0),        %%xmm0       \n\t"
+         "add  %2,          %0           \n\t"
+         "movq (%0),        %%xmm1       \n\t"
+         "movq (%0, %2),    %%xmm2       \n\t"
+         "movq (%0, %2,2),  %%xmm3       \n\t"
+         "punpcklbw %%xmm5, %%xmm0       \n\t"
+         "punpcklbw %%xmm5, %%xmm1       \n\t"
+         "punpcklbw %%xmm5, %%xmm2       \n\t"
+         "punpcklbw %%xmm5, %%xmm3       \n\t"
+         "movdqa %%xmm0,      (%1)       \n\t"
+         "movdqa %%xmm1,    16(%1)       \n\t"
+         "movdqa %%xmm2,    32(%1)       \n\t"
+         "movdqa %%xmm3,    48(%1)       \n\t"
+         "movdqa %%xmm3 ,   64(%1)       \n\t"
+         "movdqa %%xmm2 ,   80(%1)       \n\t"
+         "movdqa %%xmm1 ,   96(%1)       \n\t"
+         "movdqa %%xmm0,   112(%1)       \n\t"
+         : "+r" (pixels)
+         : "r" (block), "r" ((x86_reg)line_size)
+     );
+ }
+ #endif /* HAVE_INLINE_ASM */
+ void ff_dnxhdenc_init_x86(DNXHDEncContext *ctx)
+ {
+ #if HAVE_INLINE_ASM
+     if (av_get_cpu_flags() & AV_CPU_FLAG_SSE2) {
+         if (ctx->cid_table->bit_depth == 8)
+             ctx->get_pixels_8x4_sym = get_pixels_8x4_sym_sse2;
+     }
+ #endif /* HAVE_INLINE_ASM */
+ }
index 0000000,077031a..2d475ab
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,608 +1,608 @@@
 - * This file is part of Libav.
+ /*
+  * Optimized for ia32 CPUs by Nick Kurshev <nickols_k@mail.ru>
+  * h263, mpeg1, mpeg2 dequantizer & draw_edges by Michael Niedermayer <michaelni@gmx.at>
+  *
 - * 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
+  */
+ #include "libavutil/cpu.h"
+ #include "libavutil/x86/asm.h"
+ #include "libavcodec/avcodec.h"
+ #include "libavcodec/dsputil.h"
+ #include "libavcodec/mpegvideo.h"
+ #include "dsputil_mmx.h"
+ #if HAVE_INLINE_ASM
+ static void dct_unquantize_h263_intra_mmx(MpegEncContext *s,
+                                   DCTELEM *block, int n, int qscale)
+ {
+     x86_reg level, qmul, qadd, nCoeffs;
+     qmul = qscale << 1;
+     assert(s->block_last_index[n]>=0 || s->h263_aic);
+     if (!s->h263_aic) {
+         if (n < 4)
+             level = block[0] * s->y_dc_scale;
+         else
+             level = block[0] * s->c_dc_scale;
+         qadd = (qscale - 1) | 1;
+     }else{
+         qadd = 0;
+         level= block[0];
+     }
+     if(s->ac_pred)
+         nCoeffs=63;
+     else
+         nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+ //printf("%d %d  ", qmul, qadd);
+ __asm__ volatile(
+                 "movd %1, %%mm6                 \n\t" //qmul
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "movd %2, %%mm5                 \n\t" //qadd
+                 "pxor %%mm7, %%mm7              \n\t"
+                 "packssdw %%mm5, %%mm5          \n\t"
+                 "packssdw %%mm5, %%mm5          \n\t"
+                 "psubw %%mm5, %%mm7             \n\t"
+                 "pxor %%mm4, %%mm4              \n\t"
+                 ".p2align 4                     \n\t"
+                 "1:                             \n\t"
+                 "movq (%0, %3), %%mm0           \n\t"
+                 "movq 8(%0, %3), %%mm1          \n\t"
+                 "pmullw %%mm6, %%mm0            \n\t"
+                 "pmullw %%mm6, %%mm1            \n\t"
+                 "movq (%0, %3), %%mm2           \n\t"
+                 "movq 8(%0, %3), %%mm3          \n\t"
+                 "pcmpgtw %%mm4, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                 "pcmpgtw %%mm4, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "paddw %%mm7, %%mm0             \n\t"
+                 "paddw %%mm7, %%mm1             \n\t"
+                 "pxor %%mm0, %%mm2              \n\t"
+                 "pxor %%mm1, %%mm3              \n\t"
+                 "pcmpeqw %%mm7, %%mm0           \n\t" // block[i] == 0 ? -1 : 0
+                 "pcmpeqw %%mm7, %%mm1           \n\t" // block[i] == 0 ? -1 : 0
+                 "pandn %%mm2, %%mm0             \n\t"
+                 "pandn %%mm3, %%mm1             \n\t"
+                 "movq %%mm0, (%0, %3)           \n\t"
+                 "movq %%mm1, 8(%0, %3)          \n\t"
+                 "add $16, %3                    \n\t"
+                 "jng 1b                         \n\t"
+                 ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
+                 : "memory"
+         );
+         block[0]= level;
+ }
+ static void dct_unquantize_h263_inter_mmx(MpegEncContext *s,
+                                   DCTELEM *block, int n, int qscale)
+ {
+     x86_reg qmul, qadd, nCoeffs;
+     qmul = qscale << 1;
+     qadd = (qscale - 1) | 1;
+     assert(s->block_last_index[n]>=0 || s->h263_aic);
+     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+ //printf("%d %d  ", qmul, qadd);
+ __asm__ volatile(
+                 "movd %1, %%mm6                 \n\t" //qmul
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "movd %2, %%mm5                 \n\t" //qadd
+                 "pxor %%mm7, %%mm7              \n\t"
+                 "packssdw %%mm5, %%mm5          \n\t"
+                 "packssdw %%mm5, %%mm5          \n\t"
+                 "psubw %%mm5, %%mm7             \n\t"
+                 "pxor %%mm4, %%mm4              \n\t"
+                 ".p2align 4                     \n\t"
+                 "1:                             \n\t"
+                 "movq (%0, %3), %%mm0           \n\t"
+                 "movq 8(%0, %3), %%mm1          \n\t"
+                 "pmullw %%mm6, %%mm0            \n\t"
+                 "pmullw %%mm6, %%mm1            \n\t"
+                 "movq (%0, %3), %%mm2           \n\t"
+                 "movq 8(%0, %3), %%mm3          \n\t"
+                 "pcmpgtw %%mm4, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                 "pcmpgtw %%mm4, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "paddw %%mm7, %%mm0             \n\t"
+                 "paddw %%mm7, %%mm1             \n\t"
+                 "pxor %%mm0, %%mm2              \n\t"
+                 "pxor %%mm1, %%mm3              \n\t"
+                 "pcmpeqw %%mm7, %%mm0           \n\t" // block[i] == 0 ? -1 : 0
+                 "pcmpeqw %%mm7, %%mm1           \n\t" // block[i] == 0 ? -1 : 0
+                 "pandn %%mm2, %%mm0             \n\t"
+                 "pandn %%mm3, %%mm1             \n\t"
+                 "movq %%mm0, (%0, %3)           \n\t"
+                 "movq %%mm1, 8(%0, %3)          \n\t"
+                 "add $16, %3                    \n\t"
+                 "jng 1b                         \n\t"
+                 ::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
+                 : "memory"
+         );
+ }
+ /*
+   NK:
+   Note: looking at PARANOID:
+   "enable all paranoid tests for rounding, overflows, etc..."
+ #ifdef PARANOID
+                 if (level < -2048 || level > 2047)
+                     fprintf(stderr, "unquant error %d %d\n", i, level);
+ #endif
+   We can suppose that result of two multiplications can't be greater than 0xFFFF
+   i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
+   a complex multiplication.
+ =====================================================
+  Full formula for multiplication of 2 integer numbers
+  which are represent as high:low words:
+  input: value1 = high1:low1
+         value2 = high2:low2
+  output: value3 = value1*value2
+  value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
+  this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
+  but this algorithm will compute only 0x66cb0ce4
+  this limited by 16-bit size of operands
+  ---------------------------------
+  tlow1 = high1*low2
+  tlow2 = high2*low1
+  tlow1 = tlow1 + tlow2
+  high3:low3 = low1*low2
+  high3 += tlow1
+ */
+ static void dct_unquantize_mpeg1_intra_mmx(MpegEncContext *s,
+                                      DCTELEM *block, int n, int qscale)
+ {
+     x86_reg nCoeffs;
+     const uint16_t *quant_matrix;
+     int block0;
+     assert(s->block_last_index[n]>=0);
+     nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
+     if (n < 4)
+         block0 = block[0] * s->y_dc_scale;
+     else
+         block0 = block[0] * s->c_dc_scale;
+     /* XXX: only mpeg1 */
+     quant_matrix = s->intra_matrix;
+ __asm__ volatile(
+                 "pcmpeqw %%mm7, %%mm7           \n\t"
+                 "psrlw $15, %%mm7               \n\t"
+                 "movd %2, %%mm6                 \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "mov %3, %%"REG_a"              \n\t"
+                 ".p2align 4                     \n\t"
+                 "1:                             \n\t"
+                 "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                 "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
+                 "movq (%1, %%"REG_a"), %%mm4    \n\t"
+                 "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
+                 "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
+                 "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
+                 "pxor %%mm2, %%mm2              \n\t"
+                 "pxor %%mm3, %%mm3              \n\t"
+                 "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                 "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
+                 "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
+                 "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*q
+                 "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*q
+                 "pxor %%mm4, %%mm4              \n\t"
+                 "pxor %%mm5, %%mm5              \n\t" // FIXME slow
+                 "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
+                 "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
+                 "psraw $3, %%mm0                \n\t"
+                 "psraw $3, %%mm1                \n\t"
+                 "psubw %%mm7, %%mm0             \n\t"
+                 "psubw %%mm7, %%mm1             \n\t"
+                 "por %%mm7, %%mm0               \n\t"
+                 "por %%mm7, %%mm1               \n\t"
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "psubw %%mm2, %%mm0             \n\t"
+                 "psubw %%mm3, %%mm1             \n\t"
+                 "pandn %%mm0, %%mm4             \n\t"
+                 "pandn %%mm1, %%mm5             \n\t"
+                 "movq %%mm4, (%0, %%"REG_a")    \n\t"
+                 "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
+                 "add $16, %%"REG_a"             \n\t"
+                 "js 1b                          \n\t"
+                 ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
+                 : "%"REG_a, "memory"
+         );
+     block[0]= block0;
+ }
+ static void dct_unquantize_mpeg1_inter_mmx(MpegEncContext *s,
+                                      DCTELEM *block, int n, int qscale)
+ {
+     x86_reg nCoeffs;
+     const uint16_t *quant_matrix;
+     assert(s->block_last_index[n]>=0);
+     nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
+         quant_matrix = s->inter_matrix;
+ __asm__ volatile(
+                 "pcmpeqw %%mm7, %%mm7           \n\t"
+                 "psrlw $15, %%mm7               \n\t"
+                 "movd %2, %%mm6                 \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "mov %3, %%"REG_a"              \n\t"
+                 ".p2align 4                     \n\t"
+                 "1:                             \n\t"
+                 "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                 "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
+                 "movq (%1, %%"REG_a"), %%mm4    \n\t"
+                 "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
+                 "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
+                 "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
+                 "pxor %%mm2, %%mm2              \n\t"
+                 "pxor %%mm3, %%mm3              \n\t"
+                 "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                 "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
+                 "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
+                 "paddw %%mm0, %%mm0             \n\t" // abs(block[i])*2
+                 "paddw %%mm1, %%mm1             \n\t" // abs(block[i])*2
+                 "paddw %%mm7, %%mm0             \n\t" // abs(block[i])*2 + 1
+                 "paddw %%mm7, %%mm1             \n\t" // abs(block[i])*2 + 1
+                 "pmullw %%mm4, %%mm0            \n\t" // (abs(block[i])*2 + 1)*q
+                 "pmullw %%mm5, %%mm1            \n\t" // (abs(block[i])*2 + 1)*q
+                 "pxor %%mm4, %%mm4              \n\t"
+                 "pxor %%mm5, %%mm5              \n\t" // FIXME slow
+                 "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
+                 "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
+                 "psraw $4, %%mm0                \n\t"
+                 "psraw $4, %%mm1                \n\t"
+                 "psubw %%mm7, %%mm0             \n\t"
+                 "psubw %%mm7, %%mm1             \n\t"
+                 "por %%mm7, %%mm0               \n\t"
+                 "por %%mm7, %%mm1               \n\t"
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "psubw %%mm2, %%mm0             \n\t"
+                 "psubw %%mm3, %%mm1             \n\t"
+                 "pandn %%mm0, %%mm4             \n\t"
+                 "pandn %%mm1, %%mm5             \n\t"
+                 "movq %%mm4, (%0, %%"REG_a")    \n\t"
+                 "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
+                 "add $16, %%"REG_a"             \n\t"
+                 "js 1b                          \n\t"
+                 ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
+                 : "%"REG_a, "memory"
+         );
+ }
+ static void dct_unquantize_mpeg2_intra_mmx(MpegEncContext *s,
+                                      DCTELEM *block, int n, int qscale)
+ {
+     x86_reg nCoeffs;
+     const uint16_t *quant_matrix;
+     int block0;
+     assert(s->block_last_index[n]>=0);
+     if(s->alternate_scan) nCoeffs= 63; //FIXME
+     else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
+     if (n < 4)
+         block0 = block[0] * s->y_dc_scale;
+     else
+         block0 = block[0] * s->c_dc_scale;
+     quant_matrix = s->intra_matrix;
+ __asm__ volatile(
+                 "pcmpeqw %%mm7, %%mm7           \n\t"
+                 "psrlw $15, %%mm7               \n\t"
+                 "movd %2, %%mm6                 \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "mov %3, %%"REG_a"              \n\t"
+                 ".p2align 4                     \n\t"
+                 "1:                             \n\t"
+                 "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                 "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
+                 "movq (%1, %%"REG_a"), %%mm4    \n\t"
+                 "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
+                 "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
+                 "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
+                 "pxor %%mm2, %%mm2              \n\t"
+                 "pxor %%mm3, %%mm3              \n\t"
+                 "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                 "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
+                 "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
+                 "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*q
+                 "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*q
+                 "pxor %%mm4, %%mm4              \n\t"
+                 "pxor %%mm5, %%mm5              \n\t" // FIXME slow
+                 "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
+                 "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
+                 "psraw $3, %%mm0                \n\t"
+                 "psraw $3, %%mm1                \n\t"
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "psubw %%mm2, %%mm0             \n\t"
+                 "psubw %%mm3, %%mm1             \n\t"
+                 "pandn %%mm0, %%mm4             \n\t"
+                 "pandn %%mm1, %%mm5             \n\t"
+                 "movq %%mm4, (%0, %%"REG_a")    \n\t"
+                 "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
+                 "add $16, %%"REG_a"             \n\t"
+                 "jng 1b                         \n\t"
+                 ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
+                 : "%"REG_a, "memory"
+         );
+     block[0]= block0;
+         //Note, we do not do mismatch control for intra as errors cannot accumulate
+ }
+ static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s,
+                                      DCTELEM *block, int n, int qscale)
+ {
+     x86_reg nCoeffs;
+     const uint16_t *quant_matrix;
+     assert(s->block_last_index[n]>=0);
+     if(s->alternate_scan) nCoeffs= 63; //FIXME
+     else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
+         quant_matrix = s->inter_matrix;
+ __asm__ volatile(
+                 "pcmpeqw %%mm7, %%mm7           \n\t"
+                 "psrlq $48, %%mm7               \n\t"
+                 "movd %2, %%mm6                 \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "packssdw %%mm6, %%mm6          \n\t"
+                 "mov %3, %%"REG_a"              \n\t"
+                 ".p2align 4                     \n\t"
+                 "1:                             \n\t"
+                 "movq (%0, %%"REG_a"), %%mm0    \n\t"
+                 "movq 8(%0, %%"REG_a"), %%mm1   \n\t"
+                 "movq (%1, %%"REG_a"), %%mm4    \n\t"
+                 "movq 8(%1, %%"REG_a"), %%mm5   \n\t"
+                 "pmullw %%mm6, %%mm4            \n\t" // q=qscale*quant_matrix[i]
+                 "pmullw %%mm6, %%mm5            \n\t" // q=qscale*quant_matrix[i]
+                 "pxor %%mm2, %%mm2              \n\t"
+                 "pxor %%mm3, %%mm3              \n\t"
+                 "pcmpgtw %%mm0, %%mm2           \n\t" // block[i] < 0 ? -1 : 0
+                 "pcmpgtw %%mm1, %%mm3           \n\t" // block[i] < 0 ? -1 : 0
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "psubw %%mm2, %%mm0             \n\t" // abs(block[i])
+                 "psubw %%mm3, %%mm1             \n\t" // abs(block[i])
+                 "paddw %%mm0, %%mm0             \n\t" // abs(block[i])*2
+                 "paddw %%mm1, %%mm1             \n\t" // abs(block[i])*2
+                 "pmullw %%mm4, %%mm0            \n\t" // abs(block[i])*2*q
+                 "pmullw %%mm5, %%mm1            \n\t" // abs(block[i])*2*q
+                 "paddw %%mm4, %%mm0             \n\t" // (abs(block[i])*2 + 1)*q
+                 "paddw %%mm5, %%mm1             \n\t" // (abs(block[i])*2 + 1)*q
+                 "pxor %%mm4, %%mm4              \n\t"
+                 "pxor %%mm5, %%mm5              \n\t" // FIXME slow
+                 "pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
+                 "pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
+                 "psrlw $4, %%mm0                \n\t"
+                 "psrlw $4, %%mm1                \n\t"
+                 "pxor %%mm2, %%mm0              \n\t"
+                 "pxor %%mm3, %%mm1              \n\t"
+                 "psubw %%mm2, %%mm0             \n\t"
+                 "psubw %%mm3, %%mm1             \n\t"
+                 "pandn %%mm0, %%mm4             \n\t"
+                 "pandn %%mm1, %%mm5             \n\t"
+                 "pxor %%mm4, %%mm7              \n\t"
+                 "pxor %%mm5, %%mm7              \n\t"
+                 "movq %%mm4, (%0, %%"REG_a")    \n\t"
+                 "movq %%mm5, 8(%0, %%"REG_a")   \n\t"
+                 "add $16, %%"REG_a"             \n\t"
+                 "jng 1b                         \n\t"
+                 "movd 124(%0, %3), %%mm0        \n\t"
+                 "movq %%mm7, %%mm6              \n\t"
+                 "psrlq $32, %%mm7               \n\t"
+                 "pxor %%mm6, %%mm7              \n\t"
+                 "movq %%mm7, %%mm6              \n\t"
+                 "psrlq $16, %%mm7               \n\t"
+                 "pxor %%mm6, %%mm7              \n\t"
+                 "pslld $31, %%mm7               \n\t"
+                 "psrlq $15, %%mm7               \n\t"
+                 "pxor %%mm7, %%mm0              \n\t"
+                 "movd %%mm0, 124(%0, %3)        \n\t"
+                 ::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "r" (-2*nCoeffs)
+                 : "%"REG_a, "memory"
+         );
+ }
+ static void  denoise_dct_mmx(MpegEncContext *s, DCTELEM *block){
+     const int intra= s->mb_intra;
+     int *sum= s->dct_error_sum[intra];
+     uint16_t *offset= s->dct_offset[intra];
+     s->dct_count[intra]++;
+     __asm__ volatile(
+         "pxor %%mm7, %%mm7                      \n\t"
+         "1:                                     \n\t"
+         "pxor %%mm0, %%mm0                      \n\t"
+         "pxor %%mm1, %%mm1                      \n\t"
+         "movq (%0), %%mm2                       \n\t"
+         "movq 8(%0), %%mm3                      \n\t"
+         "pcmpgtw %%mm2, %%mm0                   \n\t"
+         "pcmpgtw %%mm3, %%mm1                   \n\t"
+         "pxor %%mm0, %%mm2                      \n\t"
+         "pxor %%mm1, %%mm3                      \n\t"
+         "psubw %%mm0, %%mm2                     \n\t"
+         "psubw %%mm1, %%mm3                     \n\t"
+         "movq %%mm2, %%mm4                      \n\t"
+         "movq %%mm3, %%mm5                      \n\t"
+         "psubusw (%2), %%mm2                    \n\t"
+         "psubusw 8(%2), %%mm3                   \n\t"
+         "pxor %%mm0, %%mm2                      \n\t"
+         "pxor %%mm1, %%mm3                      \n\t"
+         "psubw %%mm0, %%mm2                     \n\t"
+         "psubw %%mm1, %%mm3                     \n\t"
+         "movq %%mm2, (%0)                       \n\t"
+         "movq %%mm3, 8(%0)                      \n\t"
+         "movq %%mm4, %%mm2                      \n\t"
+         "movq %%mm5, %%mm3                      \n\t"
+         "punpcklwd %%mm7, %%mm4                 \n\t"
+         "punpckhwd %%mm7, %%mm2                 \n\t"
+         "punpcklwd %%mm7, %%mm5                 \n\t"
+         "punpckhwd %%mm7, %%mm3                 \n\t"
+         "paddd (%1), %%mm4                      \n\t"
+         "paddd 8(%1), %%mm2                     \n\t"
+         "paddd 16(%1), %%mm5                    \n\t"
+         "paddd 24(%1), %%mm3                    \n\t"
+         "movq %%mm4, (%1)                       \n\t"
+         "movq %%mm2, 8(%1)                      \n\t"
+         "movq %%mm5, 16(%1)                     \n\t"
+         "movq %%mm3, 24(%1)                     \n\t"
+         "add $16, %0                            \n\t"
+         "add $32, %1                            \n\t"
+         "add $16, %2                            \n\t"
+         "cmp %3, %0                             \n\t"
+             " jb 1b                             \n\t"
+         : "+r" (block), "+r" (sum), "+r" (offset)
+         : "r"(block+64)
+     );
+ }
+ static void  denoise_dct_sse2(MpegEncContext *s, DCTELEM *block){
+     const int intra= s->mb_intra;
+     int *sum= s->dct_error_sum[intra];
+     uint16_t *offset= s->dct_offset[intra];
+     s->dct_count[intra]++;
+     __asm__ volatile(
+         "pxor %%xmm7, %%xmm7                    \n\t"
+         "1:                                     \n\t"
+         "pxor %%xmm0, %%xmm0                    \n\t"
+         "pxor %%xmm1, %%xmm1                    \n\t"
+         "movdqa (%0), %%xmm2                    \n\t"
+         "movdqa 16(%0), %%xmm3                  \n\t"
+         "pcmpgtw %%xmm2, %%xmm0                 \n\t"
+         "pcmpgtw %%xmm3, %%xmm1                 \n\t"
+         "pxor %%xmm0, %%xmm2                    \n\t"
+         "pxor %%xmm1, %%xmm3                    \n\t"
+         "psubw %%xmm0, %%xmm2                   \n\t"
+         "psubw %%xmm1, %%xmm3                   \n\t"
+         "movdqa %%xmm2, %%xmm4                  \n\t"
+         "movdqa %%xmm3, %%xmm5                  \n\t"
+         "psubusw (%2), %%xmm2                   \n\t"
+         "psubusw 16(%2), %%xmm3                 \n\t"
+         "pxor %%xmm0, %%xmm2                    \n\t"
+         "pxor %%xmm1, %%xmm3                    \n\t"
+         "psubw %%xmm0, %%xmm2                   \n\t"
+         "psubw %%xmm1, %%xmm3                   \n\t"
+         "movdqa %%xmm2, (%0)                    \n\t"
+         "movdqa %%xmm3, 16(%0)                  \n\t"
+         "movdqa %%xmm4, %%xmm6                  \n\t"
+         "movdqa %%xmm5, %%xmm0                  \n\t"
+         "punpcklwd %%xmm7, %%xmm4               \n\t"
+         "punpckhwd %%xmm7, %%xmm6               \n\t"
+         "punpcklwd %%xmm7, %%xmm5               \n\t"
+         "punpckhwd %%xmm7, %%xmm0               \n\t"
+         "paddd (%1), %%xmm4                     \n\t"
+         "paddd 16(%1), %%xmm6                   \n\t"
+         "paddd 32(%1), %%xmm5                   \n\t"
+         "paddd 48(%1), %%xmm0                   \n\t"
+         "movdqa %%xmm4, (%1)                    \n\t"
+         "movdqa %%xmm6, 16(%1)                  \n\t"
+         "movdqa %%xmm5, 32(%1)                  \n\t"
+         "movdqa %%xmm0, 48(%1)                  \n\t"
+         "add $32, %0                            \n\t"
+         "add $64, %1                            \n\t"
+         "add $32, %2                            \n\t"
+         "cmp %3, %0                             \n\t"
+             " jb 1b                             \n\t"
+         : "+r" (block), "+r" (sum), "+r" (offset)
+         : "r"(block+64)
+           XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
+                             "%xmm4", "%xmm5", "%xmm6", "%xmm7")
+     );
+ }
+ #endif /* HAVE_INLINE_ASM */
+ void ff_MPV_common_init_x86(MpegEncContext *s)
+ {
+ #if HAVE_INLINE_ASM
+     int mm_flags = av_get_cpu_flags();
+     if (mm_flags & AV_CPU_FLAG_MMX) {
+         s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_mmx;
+         s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_mmx;
+         s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_mmx;
+         s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_mmx;
+         if(!(s->flags & CODEC_FLAG_BITEXACT))
+             s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx;
+         s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx;
+         if (mm_flags & AV_CPU_FLAG_SSE2) {
+             s->denoise_dct= denoise_dct_sse2;
+         } else {
+                 s->denoise_dct= denoise_dct_mmx;
+         }
+     }
+ #endif /* HAVE_INLINE_ASM */
+ }
index 0000000,5949a63..2d190f9
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,96 +1,96 @@@
 - * This file is part of Libav.
+ /*
+  * The simplest mpeg encoder (well, it was the simplest!)
+  * Copyright (c) 2000,2001 Fabrice Bellard
+  *
 - * 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
+  */
+ #include "libavutil/cpu.h"
+ #include "libavutil/x86/asm.h"
+ #include "libavcodec/avcodec.h"
+ #include "libavcodec/dsputil.h"
+ #include "libavcodec/mpegvideo.h"
+ #include "dsputil_mmx.h"
+ #if HAVE_INLINE_ASM
+ extern uint16_t ff_inv_zigzag_direct16[64];
+ #if HAVE_SSSE3
+ #define HAVE_SSSE3_BAK
+ #endif
+ #undef HAVE_SSSE3
+ #define HAVE_SSSE3 0
+ #undef HAVE_SSE2
+ #undef HAVE_MMXEXT
+ #define HAVE_SSE2 0
+ #define HAVE_MMXEXT 0
+ #define RENAME(a) a ## _MMX
+ #define RENAMEl(a) a ## _mmx
+ #include "mpegvideoenc_template.c"
+ #undef HAVE_MMXEXT
+ #define HAVE_MMXEXT 1
+ #undef RENAME
+ #undef RENAMEl
+ #define RENAME(a) a ## _MMX2
+ #define RENAMEl(a) a ## _mmx2
+ #include "mpegvideoenc_template.c"
+ #undef HAVE_SSE2
+ #define HAVE_SSE2 1
+ #undef RENAME
+ #undef RENAMEl
+ #define RENAME(a) a ## _SSE2
+ #define RENAMEl(a) a ## _sse2
+ #include "mpegvideoenc_template.c"
+ #ifdef HAVE_SSSE3_BAK
+ #undef HAVE_SSSE3
+ #define HAVE_SSSE3 1
+ #undef RENAME
+ #undef RENAMEl
+ #define RENAME(a) a ## _SSSE3
+ #define RENAMEl(a) a ## _sse2
+ #include "mpegvideoenc_template.c"
+ #endif
+ #endif /* HAVE_INLINE_ASM */
+ void ff_MPV_encode_init_x86(MpegEncContext *s)
+ {
+ #if HAVE_INLINE_ASM
+     int mm_flags = av_get_cpu_flags();
+     const int dct_algo = s->avctx->dct_algo;
+     if (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX) {
+ #if HAVE_SSSE3
+         if (mm_flags & AV_CPU_FLAG_SSSE3) {
+             s->dct_quantize = dct_quantize_SSSE3;
+         } else
+ #endif
+         if (mm_flags & AV_CPU_FLAG_SSE2) {
+             s->dct_quantize = dct_quantize_SSE2;
+         } else if (mm_flags & AV_CPU_FLAG_MMXEXT) {
+             s->dct_quantize = dct_quantize_MMX2;
+         } else {
+             s->dct_quantize = dct_quantize_MMX;
+         }
+     }
+ #endif /* HAVE_INLINE_ASM */
+ }
index 0000000,195298d..d538a39
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,361 +1,364 @@@
 - * This file is part of Libav.
+ /*
+  * MPEG video MMX templates
+  *
+  * Copyright (c) 2002 Michael Niedermayer <michaelni@gmx.at>
+  *
 - * 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
 -        if (n < 4)
++ * License along with FFmpeg; if not, write to the Free Software
+  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+  */
+ #undef MMREG_WIDTH
+ #undef MM
+ #undef MOVQ
+ #undef SPREADW
+ #undef PMAXW
+ #undef PMAX
+ #undef SAVE_SIGN
+ #undef RESTORE_SIGN
+ #if HAVE_SSE2
+ #define MMREG_WIDTH "16"
+ #define MM "%%xmm"
+ #define MOVQ "movdqa"
+ #define SPREADW(a) \
+             "pshuflw $0, "a", "a"       \n\t"\
+             "punpcklwd "a", "a"         \n\t"
+ #define PMAXW(a,b) "pmaxsw "a", "b"     \n\t"
+ #define PMAX(a,b) \
+             "movhlps "a", "b"           \n\t"\
+             PMAXW(b, a)\
+             "pshuflw $0x0E, "a", "b"    \n\t"\
+             PMAXW(b, a)\
+             "pshuflw $0x01, "a", "b"    \n\t"\
+             PMAXW(b, a)
+ #else
+ #define MMREG_WIDTH "8"
+ #define MM "%%mm"
+ #define MOVQ "movq"
+ #if HAVE_MMXEXT
+ #define SPREADW(a) "pshufw $0, "a", "a" \n\t"
+ #define PMAXW(a,b) "pmaxsw "a", "b"     \n\t"
+ #define PMAX(a,b) \
+             "pshufw $0x0E, "a", "b"     \n\t"\
+             PMAXW(b, a)\
+             "pshufw $0x01, "a", "b"     \n\t"\
+             PMAXW(b, a)
+ #else
+ #define SPREADW(a) \
+             "punpcklwd "a", "a"         \n\t"\
+             "punpcklwd "a", "a"         \n\t"
+ #define PMAXW(a,b) \
+             "psubusw "a", "b"           \n\t"\
+             "paddw "a", "b"             \n\t"
+ #define PMAX(a,b)  \
+             "movq "a", "b"              \n\t"\
+             "psrlq $32, "a"             \n\t"\
+             PMAXW(b, a)\
+             "movq "a", "b"              \n\t"\
+             "psrlq $16, "a"             \n\t"\
+             PMAXW(b, a)
+ #endif
+ #endif
+ #if HAVE_SSSE3
+ #define SAVE_SIGN(a,b) \
+             "movdqa "b", "a"            \n\t"\
+             "pabsw  "b", "b"            \n\t"
+ #define RESTORE_SIGN(a,b) \
+             "psignw "a", "b"            \n\t"
+ #else
+ #define SAVE_SIGN(a,b) \
+             "pxor "a", "a"              \n\t"\
+             "pcmpgtw "b", "a"           \n\t" /* block[i] <= 0 ? 0xFF : 0x00 */\
+             "pxor "a", "b"              \n\t"\
+             "psubw "a", "b"             \n\t" /* ABS(block[i]) */
+ #define RESTORE_SIGN(a,b) \
+             "pxor "a", "b"              \n\t"\
+             "psubw "a", "b"             \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
+ #endif
+ static int RENAME(dct_quantize)(MpegEncContext *s,
+                             DCTELEM *block, int n,
+                             int qscale, int *overflow)
+ {
+     x86_reg last_non_zero_p1;
+     int level=0, q; //=0 is because gcc says uninitialized ...
+     const uint16_t *qmat, *bias;
+     LOCAL_ALIGNED_16(int16_t, temp_block, [64]);
+     assert((7&(int)(&temp_block[0])) == 0); //did gcc align it correctly?
+     //s->fdct (block);
+     RENAMEl(ff_fdct) (block); //cannot be anything else ...
+     if(s->dct_error_sum)
+         s->denoise_dct(s, block);
+     if (s->mb_intra) {
+         int dummy;
 -        else
++        if (n < 4){
+             q = s->y_dc_scale;
 -        bias = s->q_intra_matrix16[qscale][1];
 -        qmat = s->q_intra_matrix16[qscale][0];
++            bias = s->q_intra_matrix16[qscale][1];
++            qmat = s->q_intra_matrix16[qscale][0];
++        }else{
+             q = s->c_dc_scale;
++            bias = s->q_chroma_intra_matrix16[qscale][1];
++            qmat = s->q_chroma_intra_matrix16[qscale][0];
++        }
+         /* note: block[0] is assumed to be positive */
+         if (!s->h263_aic) {
+         __asm__ volatile (
+                 "mul %%ecx                \n\t"
+                 : "=d" (level), "=a"(dummy)
+                 : "a" ((block[0]>>2) + q), "c" (ff_inverse[q<<1])
+         );
+         } else
+             /* For AIC we skip quant/dequant of INTRADC */
+             level = (block[0] + 4)>>3;
+         block[0]=0; //avoid fake overflow
+ //        temp_block[0] = (block[0] + (q >> 1)) / q;
+         last_non_zero_p1 = 1;
+     } else {
+         last_non_zero_p1 = 0;
+         bias = s->q_inter_matrix16[qscale][1];
+         qmat = s->q_inter_matrix16[qscale][0];
+     }
+     if((s->out_format == FMT_H263 || s->out_format == FMT_H261) && s->mpeg_quant==0){
+         __asm__ volatile(
+             "movd %%"REG_a", "MM"3              \n\t" // last_non_zero_p1
+             SPREADW(MM"3")
+             "pxor "MM"7, "MM"7                  \n\t" // 0
+             "pxor "MM"4, "MM"4                  \n\t" // 0
+             MOVQ" (%2), "MM"5                   \n\t" // qmat[0]
+             "pxor "MM"6, "MM"6                  \n\t"
+             "psubw (%3), "MM"6                  \n\t" // -bias[0]
+             "mov $-128, %%"REG_a"               \n\t"
+             ".p2align 4                         \n\t"
+             "1:                                 \n\t"
+             MOVQ" (%1, %%"REG_a"), "MM"0        \n\t" // block[i]
+             SAVE_SIGN(MM"1", MM"0")                   // ABS(block[i])
+             "psubusw "MM"6, "MM"0               \n\t" // ABS(block[i]) + bias[0]
+             "pmulhw "MM"5, "MM"0                \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16
+             "por "MM"0, "MM"4                   \n\t"
+             RESTORE_SIGN(MM"1", MM"0")                // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
+             MOVQ" "MM"0, (%5, %%"REG_a")        \n\t"
+             "pcmpeqw "MM"7, "MM"0               \n\t" // out==0 ? 0xFF : 0x00
+             MOVQ" (%4, %%"REG_a"), "MM"1        \n\t"
+             MOVQ" "MM"7, (%1, %%"REG_a")        \n\t" // 0
+             "pandn "MM"1, "MM"0                 \n\t"
+             PMAXW(MM"0", MM"3")
+             "add $"MMREG_WIDTH", %%"REG_a"      \n\t"
+             " js 1b                             \n\t"
+             PMAX(MM"3", MM"0")
+             "movd "MM"3, %%"REG_a"              \n\t"
+             "movzb %%al, %%"REG_a"              \n\t" // last_non_zero_p1
+             : "+a" (last_non_zero_p1)
+             : "r" (block+64), "r" (qmat), "r" (bias),
+               "r" (ff_inv_zigzag_direct16+64), "r" (temp_block+64)
+               XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
+                                 "%xmm4", "%xmm5", "%xmm6", "%xmm7")
+         );
+     }else{ // FMT_H263
+         __asm__ volatile(
+             "movd %%"REG_a", "MM"3              \n\t" // last_non_zero_p1
+             SPREADW(MM"3")
+             "pxor "MM"7, "MM"7                  \n\t" // 0
+             "pxor "MM"4, "MM"4                  \n\t" // 0
+             "mov $-128, %%"REG_a"               \n\t"
+             ".p2align 4                         \n\t"
+             "1:                                 \n\t"
+             MOVQ" (%1, %%"REG_a"), "MM"0        \n\t" // block[i]
+             SAVE_SIGN(MM"1", MM"0")                   // ABS(block[i])
+             MOVQ" (%3, %%"REG_a"), "MM"6        \n\t" // bias[0]
+             "paddusw "MM"6, "MM"0               \n\t" // ABS(block[i]) + bias[0]
+             MOVQ" (%2, %%"REG_a"), "MM"5        \n\t" // qmat[i]
+             "pmulhw "MM"5, "MM"0                \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16
+             "por "MM"0, "MM"4                   \n\t"
+             RESTORE_SIGN(MM"1", MM"0")                // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
+             MOVQ" "MM"0, (%5, %%"REG_a")        \n\t"
+             "pcmpeqw "MM"7, "MM"0               \n\t" // out==0 ? 0xFF : 0x00
+             MOVQ" (%4, %%"REG_a"), "MM"1        \n\t"
+             MOVQ" "MM"7, (%1, %%"REG_a")        \n\t" // 0
+             "pandn "MM"1, "MM"0                 \n\t"
+             PMAXW(MM"0", MM"3")
+             "add $"MMREG_WIDTH", %%"REG_a"      \n\t"
+             " js 1b                             \n\t"
+             PMAX(MM"3", MM"0")
+             "movd "MM"3, %%"REG_a"              \n\t"
+             "movzb %%al, %%"REG_a"              \n\t" // last_non_zero_p1
+             : "+a" (last_non_zero_p1)
+             : "r" (block+64), "r" (qmat+64), "r" (bias+64),
+               "r" (ff_inv_zigzag_direct16+64), "r" (temp_block+64)
+               XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
+                                 "%xmm4", "%xmm5", "%xmm6", "%xmm7")
+         );
+     }
+     __asm__ volatile(
+         "movd %1, "MM"1                     \n\t" // max_qcoeff
+         SPREADW(MM"1")
+         "psubusw "MM"1, "MM"4               \n\t"
+         "packuswb "MM"4, "MM"4              \n\t"
+ #if HAVE_SSE2
+         "packuswb "MM"4, "MM"4              \n\t"
+ #endif
+         "movd "MM"4, %0                     \n\t" // *overflow
+         : "=g" (*overflow)
+         : "g" (s->max_qcoeff)
+     );
+     if(s->mb_intra) block[0]= level;
+     else            block[0]= temp_block[0];
+     if(s->dsp.idct_permutation_type == FF_SIMPLE_IDCT_PERM){
+         if(last_non_zero_p1 <= 1) goto end;
+         block[0x08] = temp_block[0x01]; block[0x10] = temp_block[0x08];
+         block[0x20] = temp_block[0x10];
+         if(last_non_zero_p1 <= 4) goto end;
+         block[0x18] = temp_block[0x09]; block[0x04] = temp_block[0x02];
+         block[0x09] = temp_block[0x03];
+         if(last_non_zero_p1 <= 7) goto end;
+         block[0x14] = temp_block[0x0A]; block[0x28] = temp_block[0x11];
+         block[0x12] = temp_block[0x18]; block[0x02] = temp_block[0x20];
+         if(last_non_zero_p1 <= 11) goto end;
+         block[0x1A] = temp_block[0x19]; block[0x24] = temp_block[0x12];
+         block[0x19] = temp_block[0x0B]; block[0x01] = temp_block[0x04];
+         block[0x0C] = temp_block[0x05];
+         if(last_non_zero_p1 <= 16) goto end;
+         block[0x11] = temp_block[0x0C]; block[0x29] = temp_block[0x13];
+         block[0x16] = temp_block[0x1A]; block[0x0A] = temp_block[0x21];
+         block[0x30] = temp_block[0x28]; block[0x22] = temp_block[0x30];
+         block[0x38] = temp_block[0x29]; block[0x06] = temp_block[0x22];
+         if(last_non_zero_p1 <= 24) goto end;
+         block[0x1B] = temp_block[0x1B]; block[0x21] = temp_block[0x14];
+         block[0x1C] = temp_block[0x0D]; block[0x05] = temp_block[0x06];
+         block[0x0D] = temp_block[0x07]; block[0x15] = temp_block[0x0E];
+         block[0x2C] = temp_block[0x15]; block[0x13] = temp_block[0x1C];
+         if(last_non_zero_p1 <= 32) goto end;
+         block[0x0B] = temp_block[0x23]; block[0x34] = temp_block[0x2A];
+         block[0x2A] = temp_block[0x31]; block[0x32] = temp_block[0x38];
+         block[0x3A] = temp_block[0x39]; block[0x26] = temp_block[0x32];
+         block[0x39] = temp_block[0x2B]; block[0x03] = temp_block[0x24];
+         if(last_non_zero_p1 <= 40) goto end;
+         block[0x1E] = temp_block[0x1D]; block[0x25] = temp_block[0x16];
+         block[0x1D] = temp_block[0x0F]; block[0x2D] = temp_block[0x17];
+         block[0x17] = temp_block[0x1E]; block[0x0E] = temp_block[0x25];
+         block[0x31] = temp_block[0x2C]; block[0x2B] = temp_block[0x33];
+         if(last_non_zero_p1 <= 48) goto end;
+         block[0x36] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B];
+         block[0x23] = temp_block[0x34]; block[0x3C] = temp_block[0x2D];
+         block[0x07] = temp_block[0x26]; block[0x1F] = temp_block[0x1F];
+         block[0x0F] = temp_block[0x27]; block[0x35] = temp_block[0x2E];
+         if(last_non_zero_p1 <= 56) goto end;
+         block[0x2E] = temp_block[0x35]; block[0x33] = temp_block[0x3C];
+         block[0x3E] = temp_block[0x3D]; block[0x27] = temp_block[0x36];
+         block[0x3D] = temp_block[0x2F]; block[0x2F] = temp_block[0x37];
+         block[0x37] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
+     }else if(s->dsp.idct_permutation_type == FF_LIBMPEG2_IDCT_PERM){
+         if(last_non_zero_p1 <= 1) goto end;
+         block[0x04] = temp_block[0x01];
+         block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10];
+         if(last_non_zero_p1 <= 4) goto end;
+         block[0x0C] = temp_block[0x09]; block[0x01] = temp_block[0x02];
+         block[0x05] = temp_block[0x03];
+         if(last_non_zero_p1 <= 7) goto end;
+         block[0x09] = temp_block[0x0A]; block[0x14] = temp_block[0x11];
+         block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20];
+         if(last_non_zero_p1 <= 11) goto end;
+         block[0x1C] = temp_block[0x19];
+         block[0x11] = temp_block[0x12]; block[0x0D] = temp_block[0x0B];
+         block[0x02] = temp_block[0x04]; block[0x06] = temp_block[0x05];
+         if(last_non_zero_p1 <= 16) goto end;
+         block[0x0A] = temp_block[0x0C]; block[0x15] = temp_block[0x13];
+         block[0x19] = temp_block[0x1A]; block[0x24] = temp_block[0x21];
+         block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30];
+         block[0x2C] = temp_block[0x29]; block[0x21] = temp_block[0x22];
+         if(last_non_zero_p1 <= 24) goto end;
+         block[0x1D] = temp_block[0x1B]; block[0x12] = temp_block[0x14];
+         block[0x0E] = temp_block[0x0D]; block[0x03] = temp_block[0x06];
+         block[0x07] = temp_block[0x07]; block[0x0B] = temp_block[0x0E];
+         block[0x16] = temp_block[0x15]; block[0x1A] = temp_block[0x1C];
+         if(last_non_zero_p1 <= 32) goto end;
+         block[0x25] = temp_block[0x23]; block[0x29] = temp_block[0x2A];
+         block[0x34] = temp_block[0x31]; block[0x38] = temp_block[0x38];
+         block[0x3C] = temp_block[0x39]; block[0x31] = temp_block[0x32];
+         block[0x2D] = temp_block[0x2B]; block[0x22] = temp_block[0x24];
+         if(last_non_zero_p1 <= 40) goto end;
+         block[0x1E] = temp_block[0x1D]; block[0x13] = temp_block[0x16];
+         block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17];
+         block[0x1B] = temp_block[0x1E]; block[0x26] = temp_block[0x25];
+         block[0x2A] = temp_block[0x2C]; block[0x35] = temp_block[0x33];
+         if(last_non_zero_p1 <= 48) goto end;
+         block[0x39] = temp_block[0x3A]; block[0x3D] = temp_block[0x3B];
+         block[0x32] = temp_block[0x34]; block[0x2E] = temp_block[0x2D];
+             block[0x23] = temp_block[0x26]; block[0x1F] = temp_block[0x1F];
+         block[0x27] = temp_block[0x27]; block[0x2B] = temp_block[0x2E];
+         if(last_non_zero_p1 <= 56) goto end;
+         block[0x36] = temp_block[0x35]; block[0x3A] = temp_block[0x3C];
+         block[0x3E] = temp_block[0x3D]; block[0x33] = temp_block[0x36];
+         block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37];
+         block[0x3B] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
+     }else{
+         if(last_non_zero_p1 <= 1) goto end;
+         block[0x01] = temp_block[0x01];
+         block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10];
+         if(last_non_zero_p1 <= 4) goto end;
+         block[0x09] = temp_block[0x09]; block[0x02] = temp_block[0x02];
+         block[0x03] = temp_block[0x03];
+         if(last_non_zero_p1 <= 7) goto end;
+         block[0x0A] = temp_block[0x0A]; block[0x11] = temp_block[0x11];
+         block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20];
+         if(last_non_zero_p1 <= 11) goto end;
+         block[0x19] = temp_block[0x19];
+         block[0x12] = temp_block[0x12]; block[0x0B] = temp_block[0x0B];
+         block[0x04] = temp_block[0x04]; block[0x05] = temp_block[0x05];
+         if(last_non_zero_p1 <= 16) goto end;
+         block[0x0C] = temp_block[0x0C]; block[0x13] = temp_block[0x13];
+         block[0x1A] = temp_block[0x1A]; block[0x21] = temp_block[0x21];
+         block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30];
+         block[0x29] = temp_block[0x29]; block[0x22] = temp_block[0x22];
+         if(last_non_zero_p1 <= 24) goto end;
+         block[0x1B] = temp_block[0x1B]; block[0x14] = temp_block[0x14];
+         block[0x0D] = temp_block[0x0D]; block[0x06] = temp_block[0x06];
+         block[0x07] = temp_block[0x07]; block[0x0E] = temp_block[0x0E];
+         block[0x15] = temp_block[0x15]; block[0x1C] = temp_block[0x1C];
+         if(last_non_zero_p1 <= 32) goto end;
+         block[0x23] = temp_block[0x23]; block[0x2A] = temp_block[0x2A];
+         block[0x31] = temp_block[0x31]; block[0x38] = temp_block[0x38];
+         block[0x39] = temp_block[0x39]; block[0x32] = temp_block[0x32];
+         block[0x2B] = temp_block[0x2B]; block[0x24] = temp_block[0x24];
+         if(last_non_zero_p1 <= 40) goto end;
+         block[0x1D] = temp_block[0x1D]; block[0x16] = temp_block[0x16];
+         block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17];
+         block[0x1E] = temp_block[0x1E]; block[0x25] = temp_block[0x25];
+         block[0x2C] = temp_block[0x2C]; block[0x33] = temp_block[0x33];
+         if(last_non_zero_p1 <= 48) goto end;
+         block[0x3A] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B];
+         block[0x34] = temp_block[0x34]; block[0x2D] = temp_block[0x2D];
+         block[0x26] = temp_block[0x26]; block[0x1F] = temp_block[0x1F];
+         block[0x27] = temp_block[0x27]; block[0x2E] = temp_block[0x2E];
+         if(last_non_zero_p1 <= 56) goto end;
+         block[0x35] = temp_block[0x35]; block[0x3C] = temp_block[0x3C];
+         block[0x3D] = temp_block[0x3D]; block[0x36] = temp_block[0x36];
+         block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37];
+         block[0x3E] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
+     }
+     end:
+     return last_non_zero_p1 - 1;
+ }
Simple merge