Support for lossless and inter H264 4:2:2.

4 files changed, 103 insertions, 17 deletions

diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index f61f524508..8d652f13ce 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -457,6 +457,7 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
     const int full_my= my>>2;
     const int pic_width  = 16*s->mb_width;
     const int pic_height = 16*s->mb_height >> MB_FIELD;
+    int ysh;
 
     if(mx&7) extra_width -= 3;
     if(my&7) extra_height -= 3;
@@ -465,7 +466,8 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
        || full_my < 0-extra_height
        || full_mx + 16/*FIXME*/ > pic_width + extra_width
        || full_my + 16/*FIXME*/ > pic_height + extra_height){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
+                                16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
             src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
         emu=1;
     }
@@ -502,25 +504,27 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
         return;
     }
 
-    if(MB_FIELD){
+    ysh = 3 - !!(CHROMA422);
+    if(!CHROMA422 && MB_FIELD){
         // chroma offset when predicting from a field of opposite parity
         my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
     }
-    src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> 3) * h->mb_uvlinesize;
-    src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> 3) * h->mb_uvlinesize;
+
+    src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
+    src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
 
     if(emu){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, (16>>!(CHROMA422)) + 1, (mx>>3), (my>>ysh), pic_width>>1, pic_height>>!(CHROMA422));
             src_cb= s->edge_emu_buffer;
     }
-    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
+    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height << !!(CHROMA422), mx&7, (my << !!(CHROMA422)) &7);
 
     if(emu){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, (16>>!(CHROMA422)) + 1, (mx>>3), (my>>ysh), pic_width>>1, pic_height>>!(CHROMA422));
             src_cr= s->edge_emu_buffer;
     }
-    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
+    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height << !!(CHROMA422), mx&7, (my << !!(CHROMA422)) &7);
 }
 
 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
@@ -537,6 +541,9 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
     if(chroma444){
         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
+    } else if (CHROMA422) {
+        dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+        dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
     }else{
         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
@@ -577,6 +584,9 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
         chroma_weight_op = luma_weight_op;
         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
+    } else if (CHROMA422) {
+        dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+        dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
     }else{
         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
@@ -606,6 +616,14 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
             luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
+            if (CHROMA422) {
+                chroma_weight_avg(dest_cb + chroma_height * h->mb_uvlinesize,
+                                  tmp_cb + chroma_height * h->mb_uvlinesize,
+                                  h->mb_uvlinesize, 5, weight0, weight1, 0);
+                chroma_weight_avg(dest_cr + chroma_height * h->mb_uvlinesize,
+                                  tmp_cr + chroma_height * h->mb_uvlinesize,
+                                  h->mb_uvlinesize, 5, weight0, weight1, 0);
+            }
         }else{
             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
@@ -616,6 +634,18 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
+            if (CHROMA422) {
+                chroma_weight_avg(dest_cb + chroma_height * h->mb_uvlinesize,
+                                  tmp_cb + chroma_height * h->mb_uvlinesize,
+                                  h->mb_uvlinesize, h->chroma_log2_weight_denom,
+                                  h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
+                                  h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
+                chroma_weight_avg(dest_cr + chroma_height * h->mb_uvlinesize,
+                                  tmp_cr + chroma_height * h->mb_uvlinesize,
+                                  h->mb_uvlinesize, h->chroma_log2_weight_denom,
+                                  h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
+                                  h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
+            }
         }
     }else{
         int list = list1 ? 1 : 0;
@@ -632,6 +662,14 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
             chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
+            if (CHROMA422) {
+                chroma_weight_op(dest_cb + chroma_height * h->mb_uvlinesize,
+                                 h->mb_uvlinesize, h->chroma_log2_weight_denom,
+                                 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
+                chroma_weight_op(dest_cr + chroma_height * h->mb_uvlinesize,
+                                 h->mb_uvlinesize, h->chroma_log2_weight_denom,
+                                 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
+            }
         }
     }
 }
@@ -1851,13 +1889,13 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
             }
             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
                 if (!h->sps.chroma_format_idc) {
-                    for (i = 0; i < 8; i++) {
+                    for (i = 0; i < block_h; i++) {
                         uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
                         for (j = 0; j < 8; j++) {
                             tmp_cb[j] = 1 << (bit_depth - 1);
                         }
                     }
-                    for (i = 0; i < 8; i++) {
+                    for (i = 0; i < block_h; i++) {
                         uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
                         for (j = 0; j < 8; j++) {
                             tmp_cr[j] = 1 << (bit_depth - 1);
@@ -1882,7 +1920,7 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
             }
             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
                 if (!h->sps.chroma_format_idc) {
-                    for (i = 0; i < 8; i++) {
+                    for (i = 0; i < block_h; i++) {
                         memset(dest_cb + i*uvlinesize, 128, 8);
                         memset(dest_cr + i*uvlinesize, 128, 8);
                     }
@@ -1931,6 +1969,12 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
                             if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
                                 idct_add   (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
                         }
+                        if (CHROMA422) {
+                            for(i=j*16+4; i<j*16+8; i++){
+                                if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
+                                    idct_add   (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
+                            }
+                        }
                     }
                 }
             }else{
diff --git a/libavcodec/h264pred.c b/libavcodec/h264pred.c
index 17199d01e6..37a4cf1486 100644
--- a/libavcodec/h264pred.c
+++ b/libavcodec/h264pred.c
@@ -462,10 +462,10 @@ void ff_h264_pred_init(H264PredContext *h, int codec_id, const int bit_depth, co
             h->pred8x8[DC_PRED8x8     ]= FUNCC(pred8x16_dc                    , depth);\
             h->pred8x8[LEFT_DC_PRED8x8]= FUNCC(pred8x16_left_dc               , depth);\
             h->pred8x8[TOP_DC_PRED8x8 ]= FUNCC(pred8x16_top_dc                , depth);\
-            h->pred8x8[ALZHEIMER_DC_L0T_PRED8x8 ]= FUNC(pred8x8_mad_cow_dc_l0t, depth);\
-            h->pred8x8[ALZHEIMER_DC_0LT_PRED8x8 ]= FUNC(pred8x8_mad_cow_dc_0lt, depth);\
-            h->pred8x8[ALZHEIMER_DC_L00_PRED8x8 ]= FUNC(pred8x8_mad_cow_dc_l00, depth);\
-            h->pred8x8[ALZHEIMER_DC_0L0_PRED8x8 ]= FUNC(pred8x8_mad_cow_dc_0l0, depth);\
+            h->pred8x8[ALZHEIMER_DC_L0T_PRED8x8 ]= FUNC(pred8x16_mad_cow_dc_l0t, depth);\
+            h->pred8x8[ALZHEIMER_DC_0LT_PRED8x8 ]= FUNC(pred8x16_mad_cow_dc_0lt, depth);\
+            h->pred8x8[ALZHEIMER_DC_L00_PRED8x8 ]= FUNC(pred8x16_mad_cow_dc_l00, depth);\
+            h->pred8x8[ALZHEIMER_DC_0L0_PRED8x8 ]= FUNC(pred8x16_mad_cow_dc_0l0, depth);\
         }\
     }else{\
         h->pred8x8[DC_PRED8x8     ]= FUNCD(pred8x8_dc_rv40);\
@@ -510,8 +510,13 @@ void ff_h264_pred_init(H264PredContext *h, int codec_id, const int bit_depth, co
     h->pred4x4_add  [ HOR_PRED   ]= FUNCC(pred4x4_horizontal_add          , depth);\
     h->pred8x8l_add [VERT_PRED   ]= FUNCC(pred8x8l_vertical_add           , depth);\
     h->pred8x8l_add [ HOR_PRED   ]= FUNCC(pred8x8l_horizontal_add         , depth);\
+    if (chroma_format_idc == 1) {\
     h->pred8x8_add  [VERT_PRED8x8]= FUNCC(pred8x8_vertical_add            , depth);\
     h->pred8x8_add  [ HOR_PRED8x8]= FUNCC(pred8x8_horizontal_add          , depth);\
+    } else {\
+        h->pred8x8_add  [VERT_PRED8x8]= FUNCC(pred8x16_vertical_add            , depth);\
+        h->pred8x8_add  [ HOR_PRED8x8]= FUNCC(pred8x16_horizontal_add          , depth);\
+    }\
     h->pred16x16_add[VERT_PRED8x8]= FUNCC(pred16x16_vertical_add          , depth);\
     h->pred16x16_add[ HOR_PRED8x8]= FUNCC(pred16x16_horizontal_add        , depth);\
 
diff --git a/libavcodec/h264pred_template.c b/libavcodec/h264pred_template.c
index d4f654e18c..318b56196d 100644
--- a/libavcodec/h264pred_template.c
+++ b/libavcodec/h264pred_template.c
@@ -657,29 +657,50 @@ static void FUNCC(pred8x16_dc)(uint8_t *_src, int stride){
     }
 }
 
-//the following 4 function should not be optimized!
 static void FUNC(pred8x8_mad_cow_dc_l0t)(uint8_t *src, int stride){
     FUNCC(pred8x8_top_dc)(src, 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 +1147,24 @@ 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);
+}
diff --git a/libavcodec/x86/h264dsp_mmx.c b/libavcodec/x86/h264dsp_mmx.c
index 910ad8401f..06ee7cad43 100644
--- a/libavcodec/x86/h264dsp_mmx.c
+++ b/libavcodec/x86/h264dsp_mmx.c
@@ -354,7 +354,7 @@ void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth, const int chrom
 {
     int mm_flags = av_get_cpu_flags();
 
-    if (mm_flags & AV_CPU_FLAG_MMX2) {
+    if (chroma_format_idc == 1 && mm_flags & AV_CPU_FLAG_MMX2) {
         c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;
     }