* all: fixed ss2 runtime selection.

git-svn-id: svn://svn.videolan.org/x264/trunk@11 df754926-b1dd-0310-bc7b-ec298dee348c

diff --git a/Jamfile b/Jamfile
index 3266e836ace35f901f847413b682615577a9be71..7084c5249d9ff967c0542a36f425f1f8f1f7a42e 100644 (file)
--- a/Jamfile
+++ b/Jamfile
@@ -35,7 +35,7 @@ SOURCES_ALTIVEC = core/ppc/mc.c core/ppc/pixel.c ;
 SOURCES_X264 = $(SOURCES_C) ;
 if $(OS) = LINUX
 {
-    DEFINES      += ARCH_X86 HAVE_MMXEXT HAVE_MALLOC_H ;
+    DEFINES      += ARCH_X86 HAVE_MMXEXT HAVE_SSE2 HAVE_MALLOC_H ;
     SOURCES_X264 += $(SOURCES_MMX) ;
     SOURCES_X264 += $(SOURCES_X86) ;
     ASFLAGS = -f elf ;

diff --git a/core/i386/mc-c.c b/core/i386/mc-c.c
index bab42647a612e9f1c7452f15121cd27ae3c15393..bec61db45738f99cd88197f7262730274cf62acc 100644 (file)
--- a/core/i386/mc-c.c
+++ b/core/i386/mc-c.c
@@ -181,107 +181,87 @@ static inline int x264_tapfilter1( uint8_t *pix )
     return pix[-2] - 5*pix[-1] + 20*(pix[0] + pix[1]) - 5*pix[ 2] + pix[ 3];
 }
 
-#if 0
-static inline void pixel_avg_w4( uint8_t *dst,  int i_dst_stride,
-                                 uint8_t *src1, int i_src1_stride,
-                                 uint8_t *src2, int i_src2_stride,
-                                 int i_height )
-{
-    int x, y;
-    for( y = 0; y < i_height; y++ )
-    {
-        for( x = 0; x < 4; x++ )
-        {
-            dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
-        }
-        dst  += i_dst_stride;
-        src1 += i_src1_stride;
-        src2 += i_src2_stride;
-    }
+typedef void (*pf_mc_t)(uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height );
+
+/* NASM functions */
+extern void x264_pixel_avg_w4_mmxext( uint8_t *,  int, uint8_t *, int, uint8_t *, int, int  );
+extern void x264_pixel_avg_w8_mmxext( uint8_t *,  int, uint8_t *, int, uint8_t *, int, int  );
+extern void x264_pixel_avg_w16_mmxext( uint8_t *,  int, uint8_t *, int, uint8_t *, int, int  );
+extern void x264_pixel_avg_w16_sse2( uint8_t *,  int, uint8_t *, int, uint8_t *, int, int  );
+
+/* Macro to define NxM functions */
+/* mc I+H */
+#define MC_IH( name, cpu, width, height, off )  \
+static void name##_w##width##_##cpu( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height ) \
+{                                                               \
+    DECLARE_ALIGNED( uint8_t, tmp[width*height], width );       \
+                                                                \
+    mc_hh_w##width( src, i_src_stride, tmp, width, i_height );  \
+    x264_pixel_avg_w##width##_##cpu( dst, i_dst_stride,         \
+                                     src+(off), i_src_stride,   \
+                                     tmp, width, i_height );    \
 }
 
-static inline void pixel_avg_w8( uint8_t *dst,  int i_dst_stride,
-                                 uint8_t *src1, int i_src1_stride,
-                                 uint8_t *src2, int i_src2_stride,
-                                 int i_height )
-{
-    int y;
-    for( y = 0; y < i_height; y++ )
-    {
-        asm volatile(
-            "movq (%1), %%mm0\n"
-            "movq (%2), %%mm1\n"
-            "pavgb %%mm1, %%mm0\n"
-            "movq %%mm0, (%0)\n"
-            : : "r"(dst), "r"(src1), "r"(src2)
-            );
-        dst  += i_dst_stride;
-        src1 += i_src1_stride;
-        src2 += i_src2_stride;
-    }
+/* mc I+V */
+#define MC_IV( name, cpu, width, height, off )  \
+static void name##_w##width##_##cpu( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height ) \
+{                                                               \
+    DECLARE_ALIGNED( uint8_t, tmp[width*height], width );       \
+                                                                \
+    mc_hv_w##width( src, i_src_stride, tmp, width, i_height );  \
+    x264_pixel_avg_w##width##_##cpu( dst, i_dst_stride,         \
+                                     src+(off), i_src_stride,   \
+                                     tmp, width, i_height );    \
 }
-static inline void pixel_avg_w16( uint8_t *dst,  int i_dst_stride,
-                                  uint8_t *src1, int i_src1_stride,
-                                  uint8_t *src2, int i_src2_stride,
-                                  int i_height )
-{
-    int y;
 
-    for( y = 0; y < i_height; y++ )
-    {
-        asm volatile(
-            "movq (%1), %%mm0\n"
-            "movq 8(%1), %%mm2\n"
-            "movq (%2), %%mm1\n"
-            "movq 8(%2), %%mm3\n"
-
-            "pavgb %%mm1, %%mm0\n"
-            "movq %%mm0, (%0)\n"
-            "pavgb %%mm3, %%mm2\n"
-            "movq %%mm2, 8(%0)\n"
-            : : "r"(dst), "r"(src1), "r"(src2)
-            );
-        dst  += i_dst_stride;
-        src1 += i_src1_stride;
-        src2 += i_src2_stride;
-    }
+/* mc H+V */
+#define MC_HV( name, cpu, width, height, off1, off2 ) \
+static void name##_w##width##_##cpu( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height ) \
+{                                                               \
+    DECLARE_ALIGNED( uint8_t, tmp1[width*height], width );      \
+    DECLARE_ALIGNED( uint8_t, tmp2[width*height], width );      \
+                                                                \
+    mc_hv_w##width( src+(off1), i_src_stride, tmp1, width, i_height );  \
+    mc_hh_w##width( src+(off2), i_src_stride, tmp2, width, i_height );  \
+    x264_pixel_avg_w##width##_##cpu( dst, i_dst_stride,         \
+                                     tmp1, width, tmp2, width,  \
+                                     i_height );                \
+}
+
+/* mc C+H */
+#define MC_CH( name, cpu, width, height, off ) \
+static void name##_w##width##_##cpu( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height ) \
+{                                                               \
+    DECLARE_ALIGNED( uint8_t, tmp1[width*height], width );      \
+    DECLARE_ALIGNED( uint8_t, tmp2[width*height], width );      \
+                                                                \
+    mc_hc_w##width( src,       i_src_stride, tmp1, width, i_height );  \
+    mc_hh_w##width( src+(off), i_src_stride, tmp2, width, i_height );  \
+    x264_pixel_avg_w##width##_##cpu( dst, i_dst_stride,         \
+                                     tmp1, width, tmp2, width,  \
+                                     i_height );                \
+}
+
+/* mc C+V */
+#define MC_CV( name, cpu, width, height, off ) \
+static void name##_w##width##_##cpu( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height ) \
+{                                                               \
+    DECLARE_ALIGNED( uint8_t, tmp1[width*height], width );      \
+    DECLARE_ALIGNED( uint8_t, tmp2[width*height], width );      \
+                                                                \
+    mc_hc_w##width( src,       i_src_stride, tmp1, width, i_height );  \
+    mc_hv_w##width( src+(off), i_src_stride, tmp2, width, i_height );  \
+    x264_pixel_avg_w##width##_##cpu( dst, i_dst_stride,         \
+                                     tmp1, width, tmp2, width,  \
+                                     i_height );                \
 }
-#else
-extern void pixel_avg_w4( uint8_t *dst,  int i_dst_stride,
-                          uint8_t *src1, int i_src1_stride,
-                          uint8_t *src2, int i_src2_stride,
-                          int i_height );
-extern void pixel_avg_w8( uint8_t *dst,  int i_dst_stride,
-                          uint8_t *src1, int i_src1_stride,
-                          uint8_t *src2, int i_src2_stride,
-                          int i_height );
-extern void pixel_avg_w16( uint8_t *dst,  int i_dst_stride,
-                           uint8_t *src1, int i_src1_stride,
-                           uint8_t *src2, int i_src2_stride,
-                           int i_height );
-#endif
 
-typedef void (*pf_mc_t)(uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height );
 
 /*****************************************************************************
  * MC with width == 4 (height <= 8)
  *****************************************************************************/
-#if 0
-static void mc_copy_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
-{
-    int y;
 
-    for( y = 0; y < i_height; y++ )
-    {
-        memcpy( dst, src, 4 );
-
-        src += i_src_stride;
-        dst += i_dst_stride;
-    }
-}
-#else
-extern void mc_copy_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height );
-#endif
+extern void x264_mc_copy_w4_mmxext( uint8_t *, int, uint8_t *, int, int );
 
 static inline void mc_hh_w4( uint8_t *src, int i_src, uint8_t *dst, int i_dst, int i_height )
 {
@@ -384,7 +364,24 @@ static inline void mc_hc_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i
     }
 }
 
-/* mc I+H */
+MC_IH( mc_xy10, mmxext, 4, 8, 0 )
+MC_IH( mc_xy30, mmxext, 4, 8, 1 )
+
+MC_IV( mc_xy01, mmxext, 4, 8, 0 )
+MC_IV( mc_xy03, mmxext, 4, 8, i_src_stride )
+
+MC_HV( mc_xy11, mmxext, 4, 8, 0, 0 )
+MC_HV( mc_xy31, mmxext, 4, 8, 1, 0 )
+MC_HV( mc_xy13, mmxext, 4, 8, 0, i_src_stride )
+MC_HV( mc_xy33, mmxext, 4, 8, 1, i_src_stride )
+
+MC_CH( mc_xy21, mmxext, 4, 8, 0 )
+MC_CH( mc_xy23, mmxext, 4, 8, i_src_stride )
+
+MC_CV( mc_xy12, mmxext, 4, 8, 0 )
+MC_CV( mc_xy32, mmxext, 4, 8, 1 )
+
+#if 0
 static void mc_xy10_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
     uint8_t tmp[4*8];
@@ -397,7 +394,7 @@ static void mc_xy30_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_
     mc_hh_w4( src, i_src_stride, tmp, 4, i_height );
     pixel_avg_w4( dst, i_dst_stride, src+1, i_src_stride, tmp, 4, i_height );
 }
-/* mc I+V */
+
 static void mc_xy01_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
     uint8_t tmp[4*8];
@@ -410,7 +407,7 @@ static void mc_xy03_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_
     mc_hv_w4( src, i_src_stride, tmp, 4, i_height );
     pixel_avg_w4( dst, i_dst_stride, src+i_src_stride, i_src_stride, tmp, 4, i_height );
 }
-/* H+V */
+
 static void mc_xy11_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
     uint8_t tmp1[4*8];
@@ -447,6 +444,7 @@ static void mc_xy33_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_
     mc_hh_w4( src+i_src_stride, i_src_stride, tmp2, 4, i_height );
     pixel_avg_w4( dst, i_dst_stride, tmp1, 4, tmp2, 4, i_height );
 }
+
 static void mc_xy21_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
     uint8_t tmp1[4*8];
@@ -456,54 +454,40 @@ static void mc_xy21_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_
     mc_hh_w4( src, i_src_stride, tmp2, 4, i_height );
     pixel_avg_w4( dst, i_dst_stride, tmp1, 4, tmp2, 4, i_height );
 }
-static void mc_xy12_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
+static void mc_xy23_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
     uint8_t tmp1[4*8];
     uint8_t tmp2[4*8];
 
-    mc_hc_w4( src, i_src_stride, tmp1, 4, i_height );
-    mc_hv_w4( src, i_src_stride, tmp2, 4, i_height );
+    mc_hc_w4( src,              i_src_stride, tmp1, 4, i_height );
+    mc_hh_w4( src+i_src_stride, i_src_stride, tmp2, 4, i_height );
     pixel_avg_w4( dst, i_dst_stride, tmp1, 4, tmp2, 4, i_height );
 }
-static void mc_xy32_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
+
+static void mc_xy12_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
     uint8_t tmp1[4*8];
     uint8_t tmp2[4*8];
 
-    mc_hc_w4( src,   i_src_stride, tmp1, 4, i_height );
-    mc_hv_w4( src+1, i_src_stride, tmp2, 4, i_height );
+    mc_hc_w4( src, i_src_stride, tmp1, 4, i_height );
+    mc_hv_w4( src, i_src_stride, tmp2, 4, i_height );
     pixel_avg_w4( dst, i_dst_stride, tmp1, 4, tmp2, 4, i_height );
 }
-static void mc_xy23_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
+static void mc_xy32_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
     uint8_t tmp1[4*8];
     uint8_t tmp2[4*8];
 
-    mc_hc_w4( src,              i_src_stride, tmp1, 4, i_height );
-    mc_hh_w4( src+i_src_stride, i_src_stride, tmp2, 4, i_height );
+    mc_hc_w4( src,   i_src_stride, tmp1, 4, i_height );
+    mc_hv_w4( src+1, i_src_stride, tmp2, 4, i_height );
     pixel_avg_w4( dst, i_dst_stride, tmp1, 4, tmp2, 4, i_height );
 }
-
+#endif
 
 /*****************************************************************************
  * MC with width == 8 (height <= 16)
  *****************************************************************************/
-#if 0
-static void mc_copy_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
-{
-    int y;
-
-    for( y = 0; y < i_height; y++ )
-    {
-        memcpy( dst, src, 8 );
-
-        src += i_src_stride;
-        dst += i_dst_stride;
-    }
-}
-#else
-extern void mc_copy_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height );
-#endif
+extern void x264_mc_copy_w8_mmxext( uint8_t *, int, uint8_t *, int, int );
 
 static inline void mc_hh_w8( uint8_t *src, int i_src, uint8_t *dst, int i_dst, int i_height )
 {
@@ -670,6 +654,24 @@ static inline void mc_hc_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i
     }
 }
 
+MC_IH( mc_xy10, mmxext, 8, 16, 0 )
+MC_IH( mc_xy30, mmxext, 8, 16, 1 )
+
+MC_IV( mc_xy01, mmxext, 8, 16, 0 )
+MC_IV( mc_xy03, mmxext, 8, 16, i_src_stride )
+
+MC_HV( mc_xy11, mmxext, 8, 16, 0, 0 )
+MC_HV( mc_xy31, mmxext, 8, 16, 1, 0 )
+MC_HV( mc_xy13, mmxext, 8, 16, 0, i_src_stride )
+MC_HV( mc_xy33, mmxext, 8, 16, 1, i_src_stride )
+
+MC_CH( mc_xy21, mmxext, 8, 16, 0 )
+MC_CH( mc_xy23, mmxext, 8, 16, i_src_stride )
+
+MC_CV( mc_xy12, mmxext, 8, 16, 0 )
+MC_CV( mc_xy32, mmxext, 8, 16, 1 )
+
+#if 0
 /* mc I+H */
 static void mc_xy10_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
@@ -769,27 +771,15 @@ static void mc_xy23_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_
     mc_hh_w8( src+i_src_stride, i_src_stride, tmp2, 8, i_height );
     pixel_avg_w8( dst, i_dst_stride, tmp1, 8, tmp2, 8, i_height );
 }
-
+#endif
 
 /*****************************************************************************
  * MC with width == 16 (height <= 16)
  *****************************************************************************/
-#if 0
-static void mc_copy_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
-{
-    int y;
 
-    for( y = 0; y < i_height; y++ )
-    {
-        memcpy( dst, src, 16 );
+extern void x264_mc_copy_w16_mmxext( uint8_t *, int, uint8_t *, int, int );
+extern void x264_mc_copy_w16_sse2( uint8_t *, int, uint8_t *, int, int );
 
-        src += i_src_stride;
-        dst += i_dst_stride;
-    }
-}
-#else
-extern void mc_copy_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height );
-#endif
 static inline void mc_hh_w16( uint8_t *src, int i_src, uint8_t *dst, int i_dst, int i_height )
 {
     mc_hh_w4( &src[ 0], i_src, &dst[ 0], i_dst, i_height );
@@ -809,6 +799,44 @@ static inline void mc_hc_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int
     mc_hc_w8( &src[8], i_src_stride, &dst[8], i_dst_stride, i_height );
 }
 
+/* MMX avg/copy */
+MC_IH( mc_xy10, mmxext, 16, 16, 0 )
+MC_IH( mc_xy30, mmxext, 16, 16, 1 )
+
+MC_IV( mc_xy01, mmxext, 16, 16, 0 )
+MC_IV( mc_xy03, mmxext, 16, 16, i_src_stride )
+
+MC_HV( mc_xy11, mmxext, 16, 16, 0, 0 )
+MC_HV( mc_xy31, mmxext, 16, 16, 1, 0 )
+MC_HV( mc_xy13, mmxext, 16, 16, 0, i_src_stride )
+MC_HV( mc_xy33, mmxext, 16, 16, 1, i_src_stride )
+
+MC_CH( mc_xy21, mmxext, 16, 16, 0 )
+MC_CH( mc_xy23, mmxext, 16, 16, i_src_stride )
+
+MC_CV( mc_xy12, mmxext, 16, 16, 0 )
+MC_CV( mc_xy32, mmxext, 16, 16, 1 )
+
+/* SSE2 avg/copy */
+MC_IH( mc_xy10, sse2, 16, 16, 0 )
+MC_IH( mc_xy30, sse2, 16, 16, 1 )
+
+MC_IV( mc_xy01, sse2, 16, 16, 0 )
+MC_IV( mc_xy03, sse2, 16, 16, i_src_stride )
+
+MC_HV( mc_xy11, sse2, 16, 16, 0, 0 )
+MC_HV( mc_xy31, sse2, 16, 16, 1, 0 )
+MC_HV( mc_xy13, sse2, 16, 16, 0, i_src_stride )
+MC_HV( mc_xy33, sse2, 16, 16, 1, i_src_stride )
+
+MC_CH( mc_xy21, sse2, 16, 16, 0 )
+MC_CH( mc_xy23, sse2, 16, 16, i_src_stride )
+
+MC_CV( mc_xy12, sse2, 16, 16, 0 )
+MC_CV( mc_xy32, sse2, 16, 16, 1 )
+
+
+#if 0
 /* mc I+H */
 static void mc_xy10_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
 {
@@ -908,55 +936,92 @@ static void mc_xy23_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst
     mc_hh_w16( src+i_src_stride, i_src_stride, tmp2, 16, i_height );
     pixel_avg_w16( dst, i_dst_stride, tmp1, 16, tmp2, 16, i_height );
 }
+#endif
 
-static void motion_compensation_luma( uint8_t *src, int i_src_stride,
-                                      uint8_t *dst, int i_dst_stride,
-                                      int mvx,int mvy,
-                                      int i_width, int i_height )
+#define MOTION_COMPENSATION_LUMA \
+    src += (mvy >> 2) * i_src_stride + (mvx >> 2);  \
+    if( i_width == 4 )                              \
+    {                                               \
+        pf_mc[0][mvy&0x03][mvx&0x03]( src, i_src_stride, dst, i_dst_stride, i_height ); \
+    }                                               \
+    else if( i_width == 8 )                         \
+    {                                               \
+        pf_mc[1][mvy&0x03][mvx&0x03]( src, i_src_stride, dst, i_dst_stride, i_height ); \
+    }                                               \
+    else if( i_width == 16 )                        \
+    {                                               \
+        pf_mc[2][mvy&0x03][mvx&0x03]( src, i_src_stride, dst, i_dst_stride, i_height ); \
+    }                                               \
+    else                                            \
+    {                                               \
+        fprintf( stderr, "Error: motion_compensation_luma called with invalid width" ); \
+    }
+
+static void motion_compensation_luma_mmxext( uint8_t *src, int i_src_stride,
+                                             uint8_t *dst, int i_dst_stride,
+                                             int mvx,int mvy,
+                                             int i_width, int i_height )
 {
     static const pf_mc_t pf_mc[3][4][4] =    /*XXX [dqy][dqx] */
     {
         {
-            { mc_copy_w4,  mc_xy10_w4,    mc_hh_w4,      mc_xy30_w4 },
-            { mc_xy01_w4,  mc_xy11_w4,    mc_xy21_w4,    mc_xy31_w4 },
-            { mc_hv_w4,    mc_xy12_w4,    mc_hc_w4,      mc_xy32_w4 },
-            { mc_xy03_w4,  mc_xy13_w4,    mc_xy23_w4,    mc_xy33_w4 },
+            { x264_mc_copy_w4_mmxext,   mc_xy10_w4_mmxext,    mc_hh_w4,             mc_xy30_w4_mmxext },
+            { mc_xy01_w4_mmxext,        mc_xy11_w4_mmxext,    mc_xy21_w4_mmxext,    mc_xy31_w4_mmxext },
+            { mc_hv_w4,                 mc_xy12_w4_mmxext,    mc_hc_w4,             mc_xy32_w4_mmxext },
+            { mc_xy03_w4_mmxext,        mc_xy13_w4_mmxext,    mc_xy23_w4_mmxext,    mc_xy33_w4_mmxext },
         },
         {
-            { mc_copy_w8,  mc_xy10_w8,    mc_hh_w8,      mc_xy30_w8 },
-            { mc_xy01_w8,  mc_xy11_w8,    mc_xy21_w8,    mc_xy31_w8 },
-            { mc_hv_w8,    mc_xy12_w8,    mc_hc_w8,      mc_xy32_w8 },
-            { mc_xy03_w8,  mc_xy13_w8,    mc_xy23_w8,    mc_xy33_w8 },
+            { x264_mc_copy_w8_mmxext,   mc_xy10_w8_mmxext,    mc_hh_w8,             mc_xy30_w8_mmxext },
+            { mc_xy01_w8_mmxext,        mc_xy11_w8_mmxext,    mc_xy21_w8_mmxext,    mc_xy31_w8_mmxext },
+            { mc_hv_w8,                 mc_xy12_w8_mmxext,    mc_hc_w8,             mc_xy32_w8_mmxext },
+            { mc_xy03_w8_mmxext,        mc_xy13_w8_mmxext,    mc_xy23_w8_mmxext,    mc_xy33_w8_mmxext },
         },
         {
-            { mc_copy_w16,  mc_xy10_w16,    mc_hh_w16,      mc_xy30_w16 },
-            { mc_xy01_w16,  mc_xy11_w16,    mc_xy21_w16,    mc_xy31_w16 },
-            { mc_hv_w16,    mc_xy12_w16,    mc_hc_w16,      mc_xy32_w16 },
-            { mc_xy03_w16,  mc_xy13_w16,    mc_xy23_w16,    mc_xy33_w16 },
+            { x264_mc_copy_w16_mmxext,   mc_xy10_w16_mmxext,    mc_hh_w16,             mc_xy30_w16_mmxext },
+            { mc_xy01_w16_mmxext,        mc_xy11_w16_mmxext,    mc_xy21_w16_mmxext,    mc_xy31_w16_mmxext },
+            { mc_hv_w16,                 mc_xy12_w16_mmxext,    mc_hc_w16,             mc_xy32_w16_mmxext },
+            { mc_xy03_w16_mmxext,        mc_xy13_w16_mmxext,    mc_xy23_w16_mmxext,    mc_xy33_w16_mmxext },
         }
     };
 
-    src += (mvy >> 2) * i_src_stride + (mvx >> 2);
-    if( i_width == 4 )
-    {
-        pf_mc[0][mvy&0x03][mvx&0x03]( src, i_src_stride, dst, i_dst_stride, i_height );
-    }
-    else if( i_width == 8 )
-    {
-        pf_mc[1][mvy&0x03][mvx&0x03]( src, i_src_stride, dst, i_dst_stride, i_height );
-    }
-    else if( i_width == 16 )
-    {
-        pf_mc[2][mvy&0x03][mvx&0x03]( src, i_src_stride, dst, i_dst_stride, i_height );
-    }
-    else
+    MOTION_COMPENSATION_LUMA
+}
+
+static void motion_compensation_luma_sse2( uint8_t *src, int i_src_stride,
+                                           uint8_t *dst, int i_dst_stride,
+                                           int mvx,int mvy,
+                                           int i_width, int i_height )
+{
+    static const pf_mc_t pf_mc[3][4][4] =    /*XXX [dqy][dqx] */
     {
-        fprintf( stderr, "Error: motion_compensation_luma called with invalid width" );
-    }
+        {
+            { x264_mc_copy_w4_mmxext,   mc_xy10_w4_mmxext,    mc_hh_w4,             mc_xy30_w4_mmxext },
+            { mc_xy01_w4_mmxext,        mc_xy11_w4_mmxext,    mc_xy21_w4_mmxext,    mc_xy31_w4_mmxext },
+            { mc_hv_w4,                 mc_xy12_w4_mmxext,    mc_hc_w4,             mc_xy32_w4_mmxext },
+            { mc_xy03_w4_mmxext,        mc_xy13_w4_mmxext,    mc_xy23_w4_mmxext,    mc_xy33_w4_mmxext },
+        },
+        {
+            { x264_mc_copy_w8_mmxext,   mc_xy10_w8_mmxext,    mc_hh_w8,             mc_xy30_w8_mmxext },
+            { mc_xy01_w8_mmxext,        mc_xy11_w8_mmxext,    mc_xy21_w8_mmxext,    mc_xy31_w8_mmxext },
+            { mc_hv_w8,                 mc_xy12_w8_mmxext,    mc_hc_w8,             mc_xy32_w8_mmxext },
+            { mc_xy03_w8_mmxext,        mc_xy13_w8_mmxext,    mc_xy23_w8_mmxext,    mc_xy33_w8_mmxext },
+        },
+        {
+            { x264_mc_copy_w16_sse2,   mc_xy10_w16_sse2,    mc_hh_w16,             mc_xy30_w16_sse2 },
+            { mc_xy01_w16_sse2,        mc_xy11_w16_sse2,    mc_xy21_w16_sse2,    mc_xy31_w16_sse2 },
+            { mc_hv_w16,                 mc_xy12_w16_sse2,    mc_hc_w16,             mc_xy32_w16_sse2 },
+            { mc_xy03_w16_sse2,        mc_xy13_w16_sse2,    mc_xy23_w16_sse2,    mc_xy33_w16_sse2 },
+        }
+    };
+    MOTION_COMPENSATION_LUMA
 }
 
 void x264_mc_mmxext_init( x264_mc_function_t pf[2] )
 {
-    pf[MC_LUMA]   = motion_compensation_luma;
+    pf[MC_LUMA]   = motion_compensation_luma_mmxext;
+}
+void x264_mc_sse2_init( x264_mc_function_t pf[2] )
+{
+    pf[MC_LUMA]   = motion_compensation_luma_sse2;
 }
 

diff --git a/core/i386/mc.asm b/core/i386/mc.asm
index a932e159cbaa0fbb3bd98eb856db64548bdb3996..9ee4191a47e311b497c8f7f13267b9bba974c8b8 100644 (file)
--- a/core/i386/mc.asm
+++ b/core/i386/mc.asm
@@ -67,16 +67,25 @@ ALIGN 16
 
 SECTION .text
 
-cglobal pixel_avg_w4
+cglobal x264_pixel_avg_w4_mmxext
+cglobal x264_pixel_avg_w8_mmxext
+cglobal x264_pixel_avg_w16_mmxext
+cglobal x264_pixel_avg_w16_sse2
+
+cglobal x264_mc_copy_w4_mmxext
+cglobal x264_mc_copy_w8_mmxext
+cglobal x264_mc_copy_w16_mmxext
+cglobal x264_mc_copy_w16_sse2
+
 
 ALIGN 16
 ;-----------------------------------------------------------------------------
-; void pixel_avg_w4( uint8_t *dst,  int i_dst_stride,
-;                    uint8_t *src1, int i_src1_stride,
-;                    uint8_t *src2, int i_src2_stride,
-;                    int i_height );
+; void x264_pixel_avg_w4_mmxext( uint8_t *dst,  int i_dst_stride,
+;                                uint8_t *src1, int i_src1_stride,
+;                                uint8_t *src2, int i_src2_stride,
+;                                int i_height );
 ;-----------------------------------------------------------------------------
-pixel_avg_w4:
+x264_pixel_avg_w4_mmxext:
     push        ebp
     push        ebx
     push        esi
@@ -111,16 +120,15 @@ ALIGN 4
     ret
 
                           
-cglobal pixel_avg_w8
 
 ALIGN 16
 ;-----------------------------------------------------------------------------
-; void pixel_avg_w8( uint8_t *dst,  int i_dst_stride,
-;                    uint8_t *src1, int i_src1_stride,
-;                    uint8_t *src2, int i_src2_stride,
-;                    int i_height );
+; void x264_pixel_avg_w8_mmxext( uint8_t *dst,  int i_dst_stride,
+;                                uint8_t *src1, int i_src1_stride,
+;                                uint8_t *src2, int i_src2_stride,
+;                                int i_height );
 ;-----------------------------------------------------------------------------
-pixel_avg_w8:
+x264_pixel_avg_w8_mmxext:
     push        ebp
     push        ebx
     push        esi
@@ -151,16 +159,15 @@ ALIGN 4
     ret
 
 
-cglobal pixel_avg_w16
 
 ALIGN 16
 ;-----------------------------------------------------------------------------
-; void pixel_avg_w16( uint8_t *dst,  int i_dst_stride,
-;                     uint8_t *src1, int i_src1_stride,
-;                     uint8_t *src2, int i_src2_stride,
-;                     int i_height );
+; void x264_pixel_avg_w16_mmxext( uint8_t *dst,  int i_dst_stride,
+;                                 uint8_t *src1, int i_src1_stride,
+;                                 uint8_t *src2, int i_src2_stride,
+;                                 int i_height );
 ;-----------------------------------------------------------------------------
-pixel_avg_w16:
+x264_pixel_avg_w16_mmxext:
     push        ebp
     push        ebx
     push        esi
@@ -175,18 +182,50 @@ pixel_avg_w16:
     mov         ebp, [esp+44]       ; i_height
 ALIGN 4
 .height_loop    
-%ifndef HAVE_SSE2
     movq        mm0, [ebx  ]
     movq        mm1, [ebx+8]
     pavgb       mm0, [ecx  ]
     pavgb       mm1, [ecx+8]
     movq        [edi  ], mm0
     movq        [edi+8], mm1
-%else
+    dec         ebp
+    lea         ebx, [ebx+eax]
+    lea         ecx, [ecx+edx]
+    lea         edi, [edi+esi]
+    jne         .height_loop
+
+    pop         edi
+    pop         esi
+    pop         ebx
+    pop         ebp
+    ret
+
+ALIGN 16
+;-----------------------------------------------------------------------------
+; void x264_pixel_avg_w16_sse2( uint8_t *dst,  int i_dst_stride,
+;                               uint8_t *src1, int i_src1_stride,
+;                               uint8_t *src2, int i_src2_stride,
+;                               int i_height );
+;-----------------------------------------------------------------------------
+x264_pixel_avg_w16_sse2:
+    push        ebp
+    push        ebx
+    push        esi
+    push        edi
+
+    mov         edi, [esp+20]       ; dst
+    mov         ebx, [esp+28]       ; src1
+    mov         ecx, [esp+36]       ; src2
+    mov         esi, [esp+24]       ; i_dst_stride
+    mov         eax, [esp+32]       ; i_src1_stride
+    mov         edx, [esp+40]       ; i_src2_stride
+    mov         ebp, [esp+44]       ; i_height
+ALIGN 4
+.height_loop    
     movdqu      xmm0, [ebx]
     pavgb       xmm0, [ecx]
     movdqu      [edi], xmm0
-%endif
+
     dec         ebp
     lea         ebx, [ebx+eax]
     lea         ecx, [ecx+edx]
@@ -200,13 +239,13 @@ ALIGN 4
     ret
 
 
-cglobal mc_copy_w4
 
 ALIGN 16
 ;-----------------------------------------------------------------------------
-;   void mc_copy_w4( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
+;  void x264_mc_copy_w4_mmxext( uint8_t *src, int i_src_stride,
+;                               uint8_t *dst, int i_dst_stride, int i_height )
 ;-----------------------------------------------------------------------------
-mc_copy_w4:
+x264_mc_copy_w4_mmxext:
     push    ebx
     push    esi
     push    edi
@@ -237,9 +276,10 @@ cglobal mc_copy_w8
 
 ALIGN 16
 ;-----------------------------------------------------------------------------
-;   void mc_copy_w8( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
+;   void x264_mc_copy_w8_mmxext( uint8_t *src, int i_src_stride,
+;                                uint8_t *dst, int i_dst_stride, int i_height )
 ;-----------------------------------------------------------------------------
-mc_copy_w8:
+x264_mc_copy_w8_mmxext:
     push    ebx
     push    esi
     push    edi
@@ -276,9 +316,10 @@ cglobal mc_copy_w16
 
 ALIGN 16
 ;-----------------------------------------------------------------------------
-;   void mc_copy_w16( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
+;   void x264_mc_copy_w16_mmxext( uint8_t *src, int i_src_stride,
+;                                 uint8_t *dst, int i_dst_stride, int i_height )
 ;-----------------------------------------------------------------------------
-mc_copy_w16:
+x264_mc_copy_w16_mmxext:
     push    ebx
     push    esi
     push    edi
@@ -288,9 +329,9 @@ mc_copy_w16:
     mov     ebx, [esp+20]       ; i_src_stride
     mov     edx, [esp+28]       ; i_dst_stride
     mov     ecx, [esp+32]       ; i_height
+
 ALIGN 4
 .height_loop
-%ifndef HAVE_SSE2
     movq    mm0, [esi]
     movq    mm1, [esi+8]
     movq    [edi], mm0
@@ -313,7 +354,30 @@ ALIGN 4
     lea     edi, [edi+edx*2]
     sub     ecx, byte 4
     jnz     .height_loop
-%else
+    
+    pop     edi
+    pop     esi
+    pop     ebx
+    ret
+
+
+ALIGN 16
+;-----------------------------------------------------------------------------
+;   void x264_mc_copy_w16_sse2( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_height )
+;-----------------------------------------------------------------------------
+x264_mc_copy_w16_sse2:
+    push    ebx
+    push    esi
+    push    edi
+
+    mov     esi, [esp+16]       ; src
+    mov     edi, [esp+24]       ; dst
+    mov     ebx, [esp+20]       ; i_src_stride
+    mov     edx, [esp+28]       ; i_dst_stride
+    mov     ecx, [esp+32]       ; i_height
+
+ALIGN 4
+.height_loop
     movdqu  xmm0, [esi]
     movdqu  xmm1, [esi+ebx]
     movdqu  [edi], xmm0
@@ -323,7 +387,6 @@ ALIGN 4
     lea     esi, [esi+ebx*2]
     lea     edi, [edi+edx*2]
     jnz     .height_loop
-%endif
     
     pop     edi
     pop     esi

diff --git a/core/i386/mc.h b/core/i386/mc.h
index c3e906fc668248d271fd54e17a7f3bdbca9014d9..8cfc0a4ff652134a2a1873bc624a01a2d9ce2fa6 100644 (file)
--- a/core/i386/mc.h
+++ b/core/i386/mc.h
@@ -25,5 +25,6 @@
 #define _I386_MC_H 1
 
 void x264_mc_mmxext_init( x264_mc_function_t pf[2] );
+void x264_mc_sse2_init( x264_mc_function_t pf[2] );
 
 #endif

diff --git a/core/mc.c b/core/mc.c
index e7ff7541ff816197e8304e13352cda110490bcbd..2c3fd792654e821f4e2fc1e88f161493371f7155 100644 (file)
--- a/core/mc.c
+++ b/core/mc.c
@@ -306,15 +306,16 @@ void x264_mc_init( int cpu, x264_mc_function_t pf[2] )
 
 #ifdef HAVE_MMXEXT
     if( cpu&X264_CPU_MMXEXT )
-    {
         x264_mc_mmxext_init( pf );
-    }
 #endif
+#ifdef HAVE_SSE2
+    if( cpu&X264_CPU_SSE2 )
+        x264_mc_sse2_init( pf );
+#endif
+
 #ifdef HAVE_ALTIVEC
     if( cpu&X264_CPU_ALTIVEC )
-    {
         x264_mc_altivec_init( pf );
-    }
 #endif
 }