Add weighted motion compensation for RV40 B-frames

Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

4 files changed, 96 insertions, 11 deletions

diff --git a/libavcodec/rv34.c b/libavcodec/rv34.c
index 58e4552bd8..cdc559fa07 100644
--- a/libavcodec/rv34.c
+++ b/libavcodec/rv34.c
@@ -717,7 +717,7 @@ static const int chroma_coeffs[3] = { 0, 3, 5 };
 static inline void rv34_mc(RV34DecContext *r, const int block_type,
                           const int xoff, const int yoff, int mv_off,
                           const int width, const int height, int dir,
-                          const int thirdpel,
+                          const int thirdpel, int weighted,
                           qpel_mc_func (*qpel_mc)[16],
                           h264_chroma_mc_func (*chroma_mc))
 {
@@ -781,9 +781,15 @@ static inline void rv34_mc(RV34DecContext *r, const int block_type,
         srcU = uvbuf;
         srcV = uvbuf + 16;
     }
-    Y = s->dest[0] + xoff      + yoff     *s->linesize;
-    U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
-    V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
+    if(!weighted){
+        Y = s->dest[0] + xoff      + yoff     *s->linesize;
+        U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
+        V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
+    }else{
+        Y = r->tmp_b_block_y [dir]     +  xoff     +  yoff    *s->linesize;
+        U = r->tmp_b_block_uv[dir*2]   + (xoff>>1) + (yoff>>1)*s->uvlinesize;
+        V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
+    }
 
     if(block_type == RV34_MB_P_16x8){
         qpel_mc[1][dxy](Y, srcY, s->linesize);
@@ -804,33 +810,70 @@ static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
                         const int xoff, const int yoff, int mv_off,
                         const int width, const int height, int dir)
 {
-    rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30,
+    rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
             r->rdsp.put_pixels_tab,
             r->rdsp.put_chroma_pixels_tab);
 }
 
+static void rv4_weight(RV34DecContext *r)
+{
+    r->rdsp.rv40_weight_pixels_tab[0](r->s.dest[0],
+                                      r->tmp_b_block_y[0],
+                                      r->tmp_b_block_y[1],
+                                      r->weight1,
+                                      r->weight2,
+                                      r->s.linesize);
+    r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[1],
+                                      r->tmp_b_block_uv[0],
+                                      r->tmp_b_block_uv[2],
+                                      r->weight1,
+                                      r->weight2,
+                                      r->s.uvlinesize);
+    r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[2],
+                                      r->tmp_b_block_uv[1],
+                                      r->tmp_b_block_uv[3],
+                                      r->weight1,
+                                      r->weight2,
+                                      r->s.uvlinesize);
+}
+
 static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
 {
-    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30,
+    int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
+
+    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
             r->rdsp.put_pixels_tab,
             r->rdsp.put_chroma_pixels_tab);
-    rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30,
-            r->rdsp.avg_pixels_tab,
-            r->rdsp.avg_chroma_pixels_tab);
+    if(!weighted){
+        rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
+                r->rdsp.avg_pixels_tab,
+                r->rdsp.avg_chroma_pixels_tab);
+    }else{
+        rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
+                r->rdsp.put_pixels_tab,
+                r->rdsp.put_chroma_pixels_tab);
+        rv4_weight(r);
+    }
 }
 
 static void rv34_mc_2mv_skip(RV34DecContext *r)
 {
     int i, j;
+    int weighted = !r->rv30 && r->weight1 != 8192;
+
     for(j = 0; j < 2; j++)
         for(i = 0; i < 2; i++){
              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
+                     weighted,
                      r->rdsp.put_pixels_tab,
                      r->rdsp.put_chroma_pixels_tab);
              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
-                     r->rdsp.avg_pixels_tab,
-                     r->rdsp.avg_chroma_pixels_tab);
+                     weighted,
+                     weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
+                     weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
         }
+    if(weighted)
+        rv4_weight(r);
 }
 
 /** number of motion vectors in each macroblock type */
@@ -1265,6 +1308,16 @@ static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int
         if(MPV_frame_start(s, s->avctx) < 0)
             return -1;
         ff_er_frame_start(s);
+        if (!r->tmp_b_block_base || s->width != r->si.width || s->height != r->si.height) {
+            int i;
+
+            r->tmp_b_block_base = av_realloc(r->tmp_b_block_base, s->linesize * 48);
+            for (i = 0; i < 2; i++)
+                r->tmp_b_block_y[i] = r->tmp_b_block_base + i * 16 * s->linesize;
+            for (i = 0; i < 4; i++)
+                r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
+                                       + (i >> 1) * 8 * s->uvlinesize + (i & 1) * 16;
+        }
         r->cur_pts = r->si.pts;
         if(s->pict_type != AV_PICTURE_TYPE_B){
             r->last_pts = r->next_pts;
@@ -1500,6 +1553,7 @@ av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
 
     av_freep(&r->intra_types_hist);
     r->intra_types = NULL;
+    av_freep(&r->tmp_b_block_base);
     av_freep(&r->mb_type);
     av_freep(&r->cbp_luma);
     av_freep(&r->cbp_chroma);
diff --git a/libavcodec/rv34.h b/libavcodec/rv34.h
index ef19813cf8..12607fb806 100644
--- a/libavcodec/rv34.h
+++ b/libavcodec/rv34.h
@@ -116,6 +116,11 @@ typedef struct RV34DecContext{
     /** 8x8 block available flags (for MV prediction) */
     DECLARE_ALIGNED(8, uint32_t, avail_cache)[3*4];
 
+    /** temporary blocks for RV4 weighted MC */
+    uint8_t *tmp_b_block_y[2];
+    uint8_t *tmp_b_block_uv[4];
+    uint8_t *tmp_b_block_base;
+
     int (*parse_slice_header)(struct RV34DecContext *r, GetBitContext *gb, SliceInfo *si);
     int (*decode_mb_info)(struct RV34DecContext *r);
     int (*decode_intra_types)(struct RV34DecContext *r, GetBitContext *gb, int8_t *dst);
diff --git a/libavcodec/rv34dsp.h b/libavcodec/rv34dsp.h
index 771a6c0f08..e1def7dc26 100644
--- a/libavcodec/rv34dsp.h
+++ b/libavcodec/rv34dsp.h
@@ -29,11 +29,17 @@
 
 #include "dsputil.h"
 
+typedef void (*rv40_weight_func)(uint8_t *dst/*align width (8 or 16)*/,
+                                 uint8_t *src1/*align width (8 or 16)*/,
+                                 uint8_t *src2/*align width (8 or 16)*/,
+                                 int w1, int w2, int stride);
+
 typedef struct RV34DSPContext {
     qpel_mc_func put_pixels_tab[4][16];
     qpel_mc_func avg_pixels_tab[4][16];
     h264_chroma_mc_func put_chroma_pixels_tab[3];
     h264_chroma_mc_func avg_chroma_pixels_tab[3];
+    rv40_weight_func rv40_weight_pixels_tab[2];
 } RV34DSPContext;
 
 void ff_rv30dsp_init(RV34DSPContext *c, DSPContext* dsp);
diff --git a/libavcodec/rv40dsp.c b/libavcodec/rv40dsp.c
index 132f063dfa..ca620ab7d7 100644
--- a/libavcodec/rv40dsp.c
+++ b/libavcodec/rv40dsp.c
@@ -285,6 +285,23 @@ static void OPNAME ## rv40_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*a
 RV40_CHROMA_MC(put_, op_put)
 RV40_CHROMA_MC(avg_, op_avg)
 
+#define RV40_WEIGHT_FUNC(size) \
+static void rv40_weight_func_ ## size (uint8_t *dst, uint8_t *src1, uint8_t *src2, int w1, int w2, int stride)\
+{\
+    int i, j;\
+\
+    for (j = 0; j < size; j++) {\
+        for (i = 0; i < size; i++)\
+            dst[i] = (((w2 * src1[i]) >> 9) + ((w1 * src2[i]) >> 9) + 0x10) >> 5;\
+        src1 += stride;\
+        src2 += stride;\
+        dst  += stride;\
+    }\
+}
+
+RV40_WEIGHT_FUNC(16)
+RV40_WEIGHT_FUNC(8)
+
 av_cold void ff_rv40dsp_init(RV34DSPContext *c, DSPContext* dsp) {
     c->put_pixels_tab[0][ 0] = dsp->put_h264_qpel_pixels_tab[0][0];
     c->put_pixels_tab[0][ 1] = put_rv40_qpel16_mc10_c;
@@ -356,6 +373,9 @@ av_cold void ff_rv40dsp_init(RV34DSPContext *c, DSPContext* dsp) {
     c->avg_chroma_pixels_tab[0] = avg_rv40_chroma_mc8_c;
     c->avg_chroma_pixels_tab[1] = avg_rv40_chroma_mc4_c;
 
+    c->rv40_weight_pixels_tab[0] = rv40_weight_func_16;
+    c->rv40_weight_pixels_tab[1] = rv40_weight_func_8;
+
     if (HAVE_MMX)
         ff_rv40dsp_init_x86(c, dsp);
 }