Merge remote-tracking branch 'qatar/master'

* qatar/master: (35 commits)
  flvdec: Do not call parse_keyframes_index with a NULL stream
  libspeexdec: include system headers before local headers
  libspeexdec: return meaningful error codes
  libspeexdec: cosmetics: reindent
  libspeexdec: decode one frame at a time.
  swscale: fix signed shift overflows in ff_yuv2rgb_c_init_tables()
  Move timefilter code from lavf to lavd.
  mov: add support for hdvd and pgapmetadata atoms
  mov: rename function _stik, some indentation cosmetics
  mov: rename function _int8 to remove ambiguity, some indentation cosmetics
  mov: parse the gnre atom
  mp3on4: check for allocation failures in decode_init_mp3on4()
  mp3on4: create a separate flush function for MP3onMP4.
  mp3on4: ensure that the frame channel count does not exceed the codec channel count.
  mp3on4: set channel layout
  mp3on4: fix the output channel order
  mp3on4: allocate temp buffer with av_malloc() instead of on the stack.
  mp3on4: copy MPADSPContext from first context to all contexts.
  fmtconvert: port float_to_int16_interleave() 2-channel x86 inline asm to yasm
  fmtconvert: port int32_to_float_fmul_scalar() x86 inline asm to yasm
  ...

Conflicts:
	libavcodec/arm/h264dsp_init_arm.c
	libavcodec/h264.c
	libavcodec/h264.h
	libavcodec/h264_cabac.c
	libavcodec/h264_cavlc.c
	libavcodec/h264_ps.c
	libavcodec/h264dsp_template.c
	libavcodec/h264idct_template.c
	libavcodec/h264pred.c
	libavcodec/h264pred_template.c
	libavcodec/x86/h264dsp_mmx.c
	libavdevice/Makefile
	libavdevice/jack_audio.c
	libavformat/Makefile
	libavformat/flvdec.c
	libavformat/flvenc.c
	libavutil/pixfmt.h
	libswscale/utils.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

1 files changed, 38 insertions, 0 deletions

diff --git a/libavcodec/h264pred_template.c b/libavcodec/h264pred_template.c
index 8021895378..074fad50ca 100644
--- a/libavcodec/h264pred_template.c
+++ b/libavcodec/h264pred_template.c
@@ -663,23 +663,45 @@ static void FUNC(pred8x8_mad_cow_dc_l0t)(uint8_t *src, int stride){
     FUNCC(pred4x4_dc)(src, NULL, stride);
 }
 
+static void FUNC(pred8x16_mad_cow_dc_l0t)(uint8_t *src, int stride){
+    FUNCC(pred8x16_top_dc)(src, stride);
+    FUNCC(pred4x4_dc)(src, NULL, stride);
+}
+
 static void FUNC(pred8x8_mad_cow_dc_0lt)(uint8_t *src, int stride){
     FUNCC(pred8x8_dc)(src, stride);
     FUNCC(pred4x4_top_dc)(src, NULL, stride);
 }
 
+static void FUNC(pred8x16_mad_cow_dc_0lt)(uint8_t *src, int stride){
+    FUNCC(pred8x16_dc)(src, stride);
+    FUNCC(pred4x4_top_dc)(src, NULL, stride);
+}
+
 static void FUNC(pred8x8_mad_cow_dc_l00)(uint8_t *src, int stride){
     FUNCC(pred8x8_left_dc)(src, stride);
     FUNCC(pred4x4_128_dc)(src + 4*stride                  , NULL, stride);
     FUNCC(pred4x4_128_dc)(src + 4*stride + 4*sizeof(pixel), NULL, stride);
 }
 
+static void FUNC(pred8x16_mad_cow_dc_l00)(uint8_t *src, int stride){
+    FUNCC(pred8x16_left_dc)(src, stride);
+    FUNCC(pred4x4_128_dc)(src + 4*stride                  , NULL, stride);
+    FUNCC(pred4x4_128_dc)(src + 4*stride + 4*sizeof(pixel), NULL, stride);
+}
+
 static void FUNC(pred8x8_mad_cow_dc_0l0)(uint8_t *src, int stride){
     FUNCC(pred8x8_left_dc)(src, stride);
     FUNCC(pred4x4_128_dc)(src                  , NULL, stride);
     FUNCC(pred4x4_128_dc)(src + 4*sizeof(pixel), NULL, stride);
 }
 
+static void FUNC(pred8x16_mad_cow_dc_0l0)(uint8_t *src, int stride){
+    FUNCC(pred8x16_left_dc)(src, stride);
+    FUNCC(pred4x4_128_dc)(src                  , NULL, stride);
+    FUNCC(pred4x4_128_dc)(src + 4*sizeof(pixel), NULL, stride);
+}
+
 static void FUNCC(pred8x8_plane)(uint8_t *_src, int _stride){
   int j, k;
   int a;
@@ -1126,8 +1148,24 @@ static void FUNCC(pred8x8_vertical_add)(uint8_t *pix, const int *block_offset, c
         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
 }
 
+static void FUNCC(pred8x16_vertical_add)(uint8_t *pix, const int *block_offset, const DCTELEM *block, int stride){
+    int i;
+    for(i=0; i<4; i++)
+        FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
+    for(i=4; i<8; i++)
+        FUNCC(pred4x4_vertical_add)(pix + block_offset[i+4], block + i*16*sizeof(pixel), stride);
+}
+
 static void FUNCC(pred8x8_horizontal_add)(uint8_t *pix, const int *block_offset, const DCTELEM *block, int stride){
     int i;
     for(i=0; i<4; i++)
         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
 }
+
+static void FUNCC(pred8x16_horizontal_add)(uint8_t *pix, const int *block_offset, const DCTELEM *block, int stride){
+    int i;
+    for(i=0; i<4; i++)
+        FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
+    for(i=4; i<8; i++)
+        FUNCC(pred4x4_horizontal_add)(pix + block_offset[i+4], block + i*16*sizeof(pixel), stride);
+}