Cacheline-split SSSE3 chroma MC

[x264] / common / x86 / mc-c.c

diff --git a/common/x86/mc-c.c b/common/x86/mc-c.c
index 2027d3ed2940af1ade57062a13ebf3d50ce0f08a..1241a2325ebcc4511c4e7785ebb799f2ad586594 100644 (file)
--- a/common/x86/mc-c.c
+++ b/common/x86/mc-c.c
@@ -59,6 +59,9 @@ extern void x264_mc_chroma_sse2( uint8_t *src, int i_src_stride,
 extern void x264_mc_chroma_ssse3( uint8_t *src, int i_src_stride,
                                   uint8_t *dst, int i_dst_stride,
                                   int dx, int dy, int i_width, int i_height );
+extern void x264_mc_chroma_ssse3_cache64( uint8_t *src, int i_src_stride,
+                                  uint8_t *dst, int i_dst_stride,
+                                  int dx, int dy, int i_width, int i_height );
 extern void x264_plane_copy_mmxext( uint8_t *, int, uint8_t *, int, int w, int h);
 extern void *x264_memcpy_aligned_mmx( void * dst, const void * src, size_t n );
 extern void *x264_memcpy_aligned_sse2( void * dst, const void * src, size_t n );
@@ -70,6 +73,7 @@ extern void x264_integral_init4v_mmx( uint16_t *sum8, uint16_t *sum4, int stride
 extern void x264_integral_init4v_sse2( uint16_t *sum8, uint16_t *sum4, int stride );
 extern void x264_integral_init8v_mmx( uint16_t *sum8, int stride );
 extern void x264_integral_init8v_sse2( uint16_t *sum8, int stride );
+extern void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, int stride );
 #define LOWRES(cpu) \
 extern void x264_frame_init_lowres_core_##cpu( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,\
                                                int src_stride, int dst_stride, int width, int height );
@@ -339,6 +343,11 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
     pf->hpel_filter = x264_hpel_filter_ssse3;
     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
     pf->mc_chroma = x264_mc_chroma_ssse3;
+    if( cpu&X264_CPU_CACHELINE_64 )
+        pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
+
+    if( cpu&X264_CPU_SHUFFLE_IS_FAST )
+        pf->integral_init4v = x264_integral_init4v_ssse3;
 
     if( !(cpu&X264_CPU_SSE4) )
         return;