moving the svq3 motion compensation stuff to dsputil (this also means that existing optimized halfpel code is used now ...)

Originally committed as revision 1885 to svn://svn.ffmpeg.org/ffmpeg/trunk

3 files changed, 274 insertions, 129 deletions

diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 05a310492a..1f9acf8cb7 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -466,6 +466,14 @@ CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c,
 #else // 64 bit variant
 
 #define PIXOP2(OPNAME, OP) \
+static void OPNAME ## _pixels2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        OP(*((uint16_t*)(block  )), LD16(pixels  ));\
+        pixels+=line_size;\
+        block +=line_size;\
+    }\
+}\
 static void OPNAME ## _pixels4_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
     int i;\
     for(i=0; i<h; i++){\
@@ -526,6 +534,17 @@ static inline void OPNAME ## _pixels4_l2(uint8_t *dst, const uint8_t *src1, cons
     }\
 }\
 \
+static inline void OPNAME ## _pixels2_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
+                                                int src_stride1, int src_stride2, int h){\
+    int i;\
+    for(i=0; i<h; i++){\
+        uint32_t a,b;\
+        a= LD16(&src1[i*src_stride1  ]);\
+        b= LD16(&src2[i*src_stride2  ]);\
+        OP(*((uint16_t*)&dst[i*dst_stride  ]), rnd_avg32(a, b));\
+    }\
+}\
+\
 static inline void OPNAME ## _pixels16_l2(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int dst_stride, \
                                                 int src_stride1, int src_stride2, int h){\
     OPNAME ## _pixels8_l2(dst  , src1  , src2  , dst_stride, src_stride1, src_stride2, h);\
@@ -589,6 +608,23 @@ static inline void OPNAME ## _pixels8_l4(uint8_t *dst, const uint8_t *src1, uint
         OP(*((uint32_t*)&dst[i*dst_stride+4]), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
     }\
 }\
+\
+static inline void OPNAME ## _pixels4_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    OPNAME ## _pixels4_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\
+}\
+\
+static inline void OPNAME ## _pixels4_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    OPNAME ## _pixels4_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\
+}\
+\
+static inline void OPNAME ## _pixels2_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    OPNAME ## _pixels2_l2(block, pixels, pixels+1, line_size, line_size, line_size, h);\
+}\
+\
+static inline void OPNAME ## _pixels2_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
+    OPNAME ## _pixels2_l2(block, pixels, pixels+line_size, line_size, line_size, line_size, h);\
+}\
+\
 static inline void OPNAME ## _no_rnd_pixels8_l4(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,\
                  int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\
     int i;\
@@ -635,6 +671,75 @@ static inline void OPNAME ## _no_rnd_pixels16_l4(uint8_t *dst, const uint8_t *sr
     OPNAME ## _no_rnd_pixels8_l4(dst+8, src1+8, src2+8, src3+8, src4+8, dst_stride, src_stride1, src_stride2, src_stride3, src_stride4, h);\
 }\
 \
+static inline void OPNAME ## _pixels2_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
+{\
+        int i, a0, b0, a1, b1;\
+        a0= pixels[0];\
+        b0= pixels[1] + 2;\
+        a0 += b0;\
+        b0 += pixels[2];\
+\
+        pixels+=line_size;\
+        for(i=0; i<h; i+=2){\
+            a1= pixels[0];\
+            b1= pixels[1];\
+            a1 += b1;\
+            b1 += pixels[2];\
+\
+            block[0]= (a1+a0)>>2; /* FIXME non put */\
+            block[1]= (b1+b0)>>2;\
+\
+            pixels+=line_size;\
+            block +=line_size;\
+\
+            a0= pixels[0];\
+            b0= pixels[1] + 2;\
+            a0 += b0;\
+            b0 += pixels[2];\
+\
+            block[0]= (a1+a0)>>2;\
+            block[1]= (b1+b0)>>2;\
+            pixels+=line_size;\
+            block +=line_size;\
+        }\
+}\
+\
+static inline void OPNAME ## _pixels4_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
+{\
+        int i;\
+        const uint32_t a= LD32(pixels  );\
+        const uint32_t b= LD32(pixels+1);\
+        uint32_t l0=  (a&0x03030303UL)\
+                    + (b&0x03030303UL)\
+                    + 0x02020202UL;\
+        uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\
+                   + ((b&0xFCFCFCFCUL)>>2);\
+        uint32_t l1,h1;\
+\
+        pixels+=line_size;\
+        for(i=0; i<h; i+=2){\
+            uint32_t a= LD32(pixels  );\
+            uint32_t b= LD32(pixels+1);\
+            l1=  (a&0x03030303UL)\
+               + (b&0x03030303UL);\
+            h1= ((a&0xFCFCFCFCUL)>>2)\
+              + ((b&0xFCFCFCFCUL)>>2);\
+            OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
+            pixels+=line_size;\
+            block +=line_size;\
+            a= LD32(pixels  );\
+            b= LD32(pixels+1);\
+            l0=  (a&0x03030303UL)\
+               + (b&0x03030303UL)\
+               + 0x02020202UL;\
+            h0= ((a&0xFCFCFCFCUL)>>2)\
+              + ((b&0xFCFCFCFCUL)>>2);\
+            OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\
+            pixels+=line_size;\
+            block +=line_size;\
+        }\
+}\
+\
 static inline void OPNAME ## _pixels8_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
 {\
     int j;\
@@ -819,6 +924,125 @@ static void gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
         oy += dyy;
     }
 }
+
+static inline void put_tpel_pixels_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    switch(width){
+    case 2: put_pixels2_c (dst, src, stride, height); break;
+    case 4: put_pixels4_c (dst, src, stride, height); break;
+    case 8: put_pixels8_c (dst, src, stride, height); break;
+    case 16:put_pixels16_c(dst, src, stride, height); break;
+    }
+}
+
+static inline void put_tpel_pixels_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    int i,j;
+    for (i=0; i < height; i++) {
+      for (j=0; j < width; j++) {
+	dst[j] = (683*(2*src[j] + src[j+1] + 1)) >> 11;
+      }
+      src += stride;
+      dst += stride;
+    }
+}
+
+static inline void put_tpel_pixels_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    int i,j;
+    for (i=0; i < height; i++) {
+      for (j=0; j < width; j++) {
+	dst[j] = (683*(src[j] + 2*src[j+1] + 1)) >> 11;
+      }
+      src += stride;
+      dst += stride;
+    }
+}
+    
+static inline void put_tpel_pixels_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    int i,j;
+    for (i=0; i < height; i++) {
+      for (j=0; j < width; j++) {
+	dst[j] = (683*(2*src[j] + src[j+stride] + 1)) >> 11;
+      }
+      src += stride;
+      dst += stride;
+    }
+}
+    
+static inline void put_tpel_pixels_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    int i,j;
+    for (i=0; i < height; i++) {
+      for (j=0; j < width; j++) {
+	dst[j] = (2731*(4*src[j] + 3*src[j+1] + 3*src[j+stride] + 2*src[j+stride+1] + 6)) >> 15;
+      }
+      src += stride;
+      dst += stride;
+    }
+}
+
+static inline void put_tpel_pixels_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    int i,j;
+    for (i=0; i < height; i++) {
+      for (j=0; j < width; j++) {
+	dst[j] = (2731*(3*src[j] + 4*src[j+1] + 2*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;
+      }
+      src += stride;
+      dst += stride;
+    }
+}
+
+static inline void put_tpel_pixels_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    int i,j;
+    for (i=0; i < height; i++) {
+      for (j=0; j < width; j++) {
+	dst[j] = (683*(src[j] + 2*src[j+stride] + 1)) >> 11;
+      }
+      src += stride;
+      dst += stride;
+    }
+}
+
+static inline void put_tpel_pixels_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    int i,j;
+    for (i=0; i < height; i++) {
+      for (j=0; j < width; j++) {
+	dst[j] = (2731*(3*src[j] + 2*src[j+1] + 4*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;
+      }
+      src += stride;
+      dst += stride;
+    }
+}
+
+static inline void put_tpel_pixels_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int width, int height){
+    int i,j;
+    for (i=0; i < height; i++) {
+      for (j=0; j < width; j++) {
+	dst[j] = (2731*(2*src[j] + 3*src[j+1] + 3*src[j+stride] + 4*src[j+stride+1] + 6)) >> 15;
+      }
+      src += stride;
+      dst += stride;
+    }
+}
+#if 0
+#define TPEL_WIDTH(width)\
+static void put_tpel_pixels ## width ## _mc00_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc00_c(dst, src, stride, width, height);}\
+static void put_tpel_pixels ## width ## _mc10_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc10_c(dst, src, stride, width, height);}\
+static void put_tpel_pixels ## width ## _mc20_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc20_c(dst, src, stride, width, height);}\
+static void put_tpel_pixels ## width ## _mc01_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc01_c(dst, src, stride, width, height);}\
+static void put_tpel_pixels ## width ## _mc11_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc11_c(dst, src, stride, width, height);}\
+static void put_tpel_pixels ## width ## _mc21_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc21_c(dst, src, stride, width, height);}\
+static void put_tpel_pixels ## width ## _mc02_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc02_c(dst, src, stride, width, height);}\
+static void put_tpel_pixels ## width ## _mc12_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc12_c(dst, src, stride, width, height);}\
+static void put_tpel_pixels ## width ## _mc22_c(uint8_t *dst, const uint8_t *src, int stride, int height){\
+    void put_tpel_pixels_mc22_c(dst, src, stride, width, height);}
+#endif
+
 #define H264_CHROMA_MC(OPNAME, OP)\
 static void OPNAME ## h264_chroma_mc2_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\
     const int A=(8-x)*(8-y);\
@@ -2561,6 +2785,8 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
     dspfunc(put_no_rnd, 0, 16);
     dspfunc(put, 1, 8);
     dspfunc(put_no_rnd, 1, 8);
+    dspfunc(put, 2, 4);
+    dspfunc(put, 3, 2);
 
     dspfunc(avg, 0, 16);
     dspfunc(avg_no_rnd, 0, 16);
@@ -2568,6 +2794,16 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
     dspfunc(avg_no_rnd, 1, 8);
 #undef dspfunc
 
+    c->put_tpel_pixels_tab[ 0] = put_tpel_pixels_mc00_c;
+    c->put_tpel_pixels_tab[ 1] = put_tpel_pixels_mc10_c;
+    c->put_tpel_pixels_tab[ 2] = put_tpel_pixels_mc20_c;
+    c->put_tpel_pixels_tab[ 4] = put_tpel_pixels_mc01_c;
+    c->put_tpel_pixels_tab[ 5] = put_tpel_pixels_mc11_c;
+    c->put_tpel_pixels_tab[ 6] = put_tpel_pixels_mc21_c;
+    c->put_tpel_pixels_tab[ 8] = put_tpel_pixels_mc02_c;
+    c->put_tpel_pixels_tab[ 9] = put_tpel_pixels_mc12_c;
+    c->put_tpel_pixels_tab[10] = put_tpel_pixels_mc22_c;
+
 #define dspfunc(PFX, IDX, NUM) \
     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_c; \
     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_c; \
@@ -2621,7 +2857,7 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
     c->put_mspel_pixels_tab[5]= put_mspel8_mc12_c;
     c->put_mspel_pixels_tab[6]= put_mspel8_mc22_c;
     c->put_mspel_pixels_tab[7]= put_mspel8_mc32_c;
-    
+        
     c->hadamard8_diff[0]= hadamard8_diff16_c;
     c->hadamard8_diff[1]= hadamard8_diff_c;
     c->hadamard8_abs = hadamard8_abs_c;
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 0060d77dce..d43175e25b 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -77,6 +77,7 @@ void clear_blocks_c(DCTELEM *blocks);
 /* add and put pixel (decoding) */
 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
+typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h);
 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
 
@@ -146,18 +147,18 @@ typedef struct DSPContext {
     me_cmp_func me_sub_cmp[11];
     me_cmp_func mb_cmp[11];
 
-    /* maybe create an array for 16/8 functions */
+    /* maybe create an array for 16/8/4/2 functions */
     /**
      * Halfpel motion compensation with rounding (a+b+1)>>1.
-     * this is an array[2][4] of motion compensation funcions for 2 
-     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
+     * this is an array[4][4] of motion compensation funcions for 4 
+     * horizontal blocksizes (2,4,8,16) and the 4 halfpel positions<br>
      * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
      * @param block destination where the result is stored
      * @param pixels source
      * @param line_size number of bytes in a horizontal line of block
      * @param h height
      */
-    op_pixels_func put_pixels_tab[2][4];
+    op_pixels_func put_pixels_tab[4][4];
 
     /**
      * Halfpel motion compensation with rounding (a+b+1)>>1.
@@ -194,6 +195,18 @@ typedef struct DSPContext {
      * @param h height
      */
     op_pixels_func avg_no_rnd_pixels_tab[2][4];
+    
+    /**
+     * Thirdpel motion compensation with rounding (a+b+1)>>1.
+     * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
+     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
+     * @param block destination where the result is stored
+     * @param pixels source
+     * @param line_size number of bytes in a horizontal line of block
+     * @param h height
+     */
+    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
+    
     qpel_mc_func put_qpel_pixels_tab[2][16];
     qpel_mc_func avg_qpel_pixels_tab[2][16];
     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
@@ -380,7 +393,9 @@ void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 
 struct unaligned_64 { uint64_t l; } __attribute__((packed));
 struct unaligned_32 { uint32_t l; } __attribute__((packed));
+struct unaligned_16 { uint16_t l; } __attribute__((packed));
 
+#define LD16(a) (((const struct unaligned_16 *) (a))->l)
 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
 
@@ -388,6 +403,7 @@ struct unaligned_32 { uint32_t l; } __attribute__((packed));
 
 #else /* __GNUC__ */
 
+#define LD16(a) (*((uint16_t*)(a)))
 #define LD32(a) (*((uint32_t*)(a)))
 #define LD64(a) (*((uint64_t*)(a)))
 
diff --git a/libavcodec/svq3.c b/libavcodec/svq3.c
index 6ff1c0b28e..4e79a22281 100644
--- a/libavcodec/svq3.c
+++ b/libavcodec/svq3.c
@@ -262,125 +262,11 @@ static inline int svq3_decode_block (GetBitContext *gb, DCTELEM *block,
   return 0;
 }
 
-static void sixpel_mc_put (MpegEncContext *s,
-			   uint8_t *src, uint8_t *dst, int stride,
-			   int dxy, int width, int height) {
-  int i, j;
-
-  switch (dxy) {
-  case 6*0+0:
-    for (i=0; i < height; i++) {
-      memcpy (dst, src, width);
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*0+2:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (683*(2*src[j] + src[j+1] + 1)) >> 11;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*0+3:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (src[j] + src[j+1] + 1) >> 1;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*0+4:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (683*(src[j] + 2*src[j+1] + 1)) >> 11;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*2+0:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (683*(2*src[j] + src[j+stride] + 1)) >> 11;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*2+2:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (2731*(4*src[j] + 3*src[j+1] + 3*src[j+stride] + 2*src[j+stride+1] + 6)) >> 15;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*2+4:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (2731*(3*src[j] + 4*src[j+1] + 2*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*3+0:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (src[j] + src[j+stride]+1) >> 1;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*3+3:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (src[j] + src[j+1] + src[j+stride] + src[j+stride+1] + 2) >> 2;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*4+0:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (683*(src[j] + 2*src[j+stride] + 1)) >> 11;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*4+2:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (2731*(3*src[j] + 2*src[j+1] + 4*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  case 6*4+4:
-    for (i=0; i < height; i++) {
-      for (j=0; j < width; j++) {
-	dst[j] = (2731*(2*src[j] + 3*src[j+1] + 3*src[j+stride] + 4*src[j+stride+1] + 6)) >> 15;
-      }
-      src += stride;
-      dst += stride;
-    }
-    break;
-  }
-}
-
 static inline void svq3_mc_dir_part (MpegEncContext *s, int x, int y,
-				     int width, int height, int mx, int my, int dxy) {
+				     int width, int height, int mx, int my, int dxy, int thirdpel) {
   uint8_t *src, *dest;
   int i, emu = 0;
+  int blocksize= 2 - (width>>3); //16->0, 8->1, 4->2
 
   mx += x;
   my += y;
@@ -405,13 +291,17 @@ static inline void svq3_mc_dir_part (MpegEncContext *s, int x, int y,
 			 mx, my, s->width, s->height);
     src = s->edge_emu_buffer;
   }
-  sixpel_mc_put (s, src, dest, s->linesize, dxy, width, height);
+  if(thirdpel)
+    s->dsp.put_tpel_pixels_tab[dxy](dest, src, s->linesize, width, height);
+  else
+    s->dsp.put_pixels_tab[blocksize][dxy](dest, src, s->linesize, height);
 
   if (!(s->flags & CODEC_FLAG_GRAY)) {
     mx	   = (mx + (mx < (int) x)) >> 1;
     my	   = (my + (my < (int) y)) >> 1;
     width  = (width  >> 1);
     height = (height >> 1);
+    blocksize++;
 
     for (i=1; i < 3; i++) {
       dest = s->current_picture.data[i] + (x >> 1) + (y >> 1)*s->uvlinesize;
@@ -422,7 +312,10 @@ static inline void svq3_mc_dir_part (MpegEncContext *s, int x, int y,
 			     mx, my, (s->width >> 1), (s->height >> 1));
 	src = s->edge_emu_buffer;
       }
-      sixpel_mc_put (s, src, dest, s->uvlinesize, dxy, width, height);
+      if(thirdpel)
+        s->dsp.put_tpel_pixels_tab[dxy](dest, src, s->uvlinesize, width, height);
+      else
+        s->dsp.put_pixels_tab[blocksize][dxy](dest, src, s->uvlinesize, height);
     }
   }
 }
@@ -441,7 +334,7 @@ static int svq3_decode_mb (H264Context *h, unsigned int mb_type) {
   h->topright_samples_available	= 0xFFFF;
 
   if (mb_type == 0) {		/* SKIP */
-    svq3_mc_dir_part (s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0, 0);
+    svq3_mc_dir_part (s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0, 0, 0);
 
     cbp = 0;
     mb_type = MB_TYPE_SKIP;
@@ -521,17 +414,17 @@ static int svq3_decode_mb (H264Context *h, unsigned int mb_type) {
           my = ((my + 1)>>1) + dy;
           fx= ((unsigned)(mx + 0x3000))/3 - 0x1000;
           fy= ((unsigned)(my + 0x3000))/3 - 0x1000;
-          dxy= 2*(mx - 3*fx) + 2*6*(my - 3*fy);
+          dxy= (mx - 3*fx) + 4*(my - 3*fy);
 
-          svq3_mc_dir_part (s, x, y, part_width, part_height, fx, fy, dxy);
+          svq3_mc_dir_part (s, x, y, part_width, part_height, fx, fy, dxy, 1);
           mx += mx;
           my += my;
 	} else if (mode == HALFPEL_MODE) {
 	  mx = ((unsigned)(mx + 1 + 0x3000))/3 + dx - 0x1000;
 	  my = ((unsigned)(my + 1 + 0x3000))/3 + dy - 0x1000;
-          dxy= 3*(mx&1) + 6*3*(my&1);
+          dxy= (mx&1) + 2*(my&1);
 
-          svq3_mc_dir_part (s, x, y, part_width, part_height, mx>>1, my>>1, dxy);
+          svq3_mc_dir_part (s, x, y, part_width, part_height, mx>>1, my>>1, dxy, 0);
           mx *= 3;
           my *= 3;
 	} else {
@@ -539,7 +432,7 @@ static int svq3_decode_mb (H264Context *h, unsigned int mb_type) {
 	  mx = ((unsigned)(mx + 3 + 0x6000))/6 + dx - 0x1000;
 	  my = ((unsigned)(my + 3 + 0x6000))/6 + dy - 0x1000;
 
-	  svq3_mc_dir_part (s, x, y, part_width, part_height, mx, my, 0);
+	  svq3_mc_dir_part (s, x, y, part_width, part_height, mx, my, 0, 0);
           mx *= 6;
           my *= 6;
 	}