Merge horizontal and vertical hpel code
[x262.git] / common / x86 / mc-c.c
index a789f96..c32ec7d 100644 (file)
@@ -1,7 +1,7 @@
 /*****************************************************************************
  * mc-c.c: x86 motion compensation
  *****************************************************************************
- * Copyright (C) 2003-2011 x264 project
+ * Copyright (C) 2003-2014 x264 project
  *
  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
  *          Loren Merritt <lorenm@u.washington.edu>
 #include "mc.h"
 
 #define DECL_SUF( func, args )\
-    void func##_mmxext args;\
+    void func##_mmx2 args;\
     void func##_sse2 args;\
-    void func##_ssse3 args;
-
-DECL_SUF( x264_pixel_avg_16x16, ( pixel *, int, pixel *, int, pixel *, int, int ))
-DECL_SUF( x264_pixel_avg_16x8,  ( pixel *, int, pixel *, int, pixel *, int, int ))
-DECL_SUF( x264_pixel_avg_8x16,  ( pixel *, int, pixel *, int, pixel *, int, int ))
-DECL_SUF( x264_pixel_avg_8x8,   ( pixel *, int, pixel *, int, pixel *, int, int ))
-DECL_SUF( x264_pixel_avg_8x4,   ( pixel *, int, pixel *, int, pixel *, int, int ))
-DECL_SUF( x264_pixel_avg_4x8,   ( pixel *, int, pixel *, int, pixel *, int, int ))
-DECL_SUF( x264_pixel_avg_4x4,   ( pixel *, int, pixel *, int, pixel *, int, int ))
-DECL_SUF( x264_pixel_avg_4x2,   ( pixel *, int, pixel *, int, pixel *, int, int ))
+    void func##_ssse3 args;\
+    void func##_avx2 args;
+
+DECL_SUF( x264_pixel_avg_16x16, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
+DECL_SUF( x264_pixel_avg_16x8,  ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
+DECL_SUF( x264_pixel_avg_8x16,  ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
+DECL_SUF( x264_pixel_avg_8x8,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
+DECL_SUF( x264_pixel_avg_8x4,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
+DECL_SUF( x264_pixel_avg_4x16,  ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
+DECL_SUF( x264_pixel_avg_4x8,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
+DECL_SUF( x264_pixel_avg_4x4,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
+DECL_SUF( x264_pixel_avg_4x2,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
 
 #define MC_WEIGHT(w,type) \
-    void x264_mc_weight_w##w##_##type( pixel *,int, pixel *,int, const x264_weight_t *,int );
+    void x264_mc_weight_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int );
 
 #define MC_WEIGHT_OFFSET(w,type) \
-    void x264_mc_offsetadd_w##w##_##type( pixel *,int, pixel *,int, const x264_weight_t *,int ); \
-    void x264_mc_offsetsub_w##w##_##type( pixel *,int, pixel *,int, const x264_weight_t *,int ); \
+    void x264_mc_offsetadd_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ); \
+    void x264_mc_offsetsub_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ); \
     MC_WEIGHT(w,type)
 
-MC_WEIGHT_OFFSET( 4, mmxext )
-MC_WEIGHT_OFFSET( 8, mmxext )
-MC_WEIGHT_OFFSET( 12, mmxext )
-MC_WEIGHT_OFFSET( 16, mmxext )
-MC_WEIGHT_OFFSET( 20, mmxext )
+MC_WEIGHT_OFFSET( 4, mmx2 )
+MC_WEIGHT_OFFSET( 8, mmx2 )
+MC_WEIGHT_OFFSET( 12, mmx2 )
+MC_WEIGHT_OFFSET( 16, mmx2 )
+MC_WEIGHT_OFFSET( 20, mmx2 )
 MC_WEIGHT_OFFSET( 12, sse2 )
 MC_WEIGHT_OFFSET( 16, sse2 )
 MC_WEIGHT_OFFSET( 20, sse2 )
@@ -71,112 +73,140 @@ MC_WEIGHT( 8, ssse3 )
 MC_WEIGHT( 12, ssse3 )
 MC_WEIGHT( 16, ssse3 )
 MC_WEIGHT( 20, ssse3 )
-MC_WEIGHT( 4, avx )
-MC_WEIGHT( 8, avx )
-MC_WEIGHT( 12, avx )
-MC_WEIGHT( 16, avx )
-MC_WEIGHT( 20, avx )
+MC_WEIGHT( 8, avx2 )
+MC_WEIGHT( 16, avx2 )
+MC_WEIGHT( 20, avx2 )
 #undef MC_OFFSET
 #undef MC_WEIGHT
 
-void x264_mc_copy_w4_mmx( pixel *, int, pixel *, int, int );
-void x264_mc_copy_w8_mmx( pixel *, int, pixel *, int, int );
-void x264_mc_copy_w8_sse2( pixel *, int, pixel *, int, int );
-void x264_mc_copy_w8_aligned_sse2( pixel *, int, pixel *, int, int );
-void x264_mc_copy_w16_mmx( pixel *, int, pixel *, int, int );
-void x264_mc_copy_w16_sse2( pixel *, int, pixel *, int, int );
-void x264_mc_copy_w16_sse3( uint8_t *, int, uint8_t *, int, int );
-void x264_mc_copy_w16_aligned_sse2( pixel *, int, pixel *, int, int );
-void x264_prefetch_fenc_mmxext( uint8_t *, int, uint8_t *, int, int );
-void x264_prefetch_ref_mmxext( uint8_t *, int, int );
-void x264_plane_copy_core_mmxext( pixel *, int, pixel *, int, int w, int h);
-void x264_plane_copy_c( pixel *, int, pixel *, int, int w, int h );
-void x264_plane_copy_interleave_core_mmxext( pixel *dst, int i_dst,
-                                             pixel *srcu, int i_srcu,
-                                             pixel *srcv, int i_srcv, int w, int h );
-void x264_plane_copy_interleave_core_sse2( pixel *dst, int i_dst,
-                                           pixel *srcu, int i_srcu,
-                                           pixel *srcv, int i_srcv, int w, int h );
-void x264_plane_copy_interleave_core_avx( pixel *dst, int i_dst,
-                                           pixel *srcu, int i_srcu,
-                                           pixel *srcv, int i_srcv, int w, int h );
-void x264_plane_copy_interleave_c( pixel *dst, int i_dst,
-                                   pixel *srcu, int i_srcu,
-                                   pixel *srcv, int i_srcv, int w, int h );
-void x264_plane_copy_deinterleave_mmx( pixel *dstu, int i_dstu,
-                                       pixel *dstv, int i_dstv,
-                                       pixel *src, int i_src, int w, int h );
-void x264_plane_copy_deinterleave_sse2( pixel *dstu, int i_dstu,
-                                        pixel *dstv, int i_dstv,
-                                        pixel *src, int i_src, int w, int h );
-void x264_plane_copy_deinterleave_ssse3( uint8_t *dstu, int i_dstu,
-                                         uint8_t *dstv, int i_dstv,
-                                         uint8_t *src, int i_src, int w, int h );
-void x264_plane_copy_deinterleave_avx( uint16_t *dstu, int i_dstu,
-                                         uint16_t *dstv, int i_dstv,
-                                         uint16_t *src, int i_src, int w, int h );
-void x264_store_interleave_8x8x2_mmxext( pixel *dst, int i_dst, pixel *srcu, pixel *srcv );
-void x264_store_interleave_8x8x2_sse2( pixel *dst, int i_dst, pixel *srcu, pixel *srcv );
-void x264_store_interleave_8x8x2_avx( pixel *dst, int i_dst, pixel *srcu, pixel *srcv );
-void x264_load_deinterleave_8x8x2_fenc_mmx( pixel *dst, pixel *src, int i_src );
-void x264_load_deinterleave_8x8x2_fenc_sse2( pixel *dst, pixel *src, int i_src );
-void x264_load_deinterleave_8x8x2_fenc_ssse3( uint8_t *dst, uint8_t *src, int i_src );
-void x264_load_deinterleave_8x8x2_fenc_avx( uint16_t *dst, uint16_t *src, int i_src );
-void x264_load_deinterleave_8x8x2_fdec_mmx( pixel *dst, pixel *src, int i_src );
-void x264_load_deinterleave_8x8x2_fdec_sse2( pixel *dst, pixel *src, int i_src );
-void x264_load_deinterleave_8x8x2_fdec_ssse3( uint8_t *dst, uint8_t *src, int i_src );
-void x264_load_deinterleave_8x8x2_fdec_avx( uint16_t *dst, uint16_t *src, int i_src );
-void *x264_memcpy_aligned_mmx( void * dst, const void * src, size_t n );
-void *x264_memcpy_aligned_sse2( void * dst, const void * src, size_t n );
-void x264_memzero_aligned_mmx( void * dst, int n );
-void x264_memzero_aligned_sse2( void * dst, int n );
-void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, int stride );
-void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, int stride );
-void x264_integral_init8h_avx ( uint16_t *sum, uint8_t *pix, int stride );
-void x264_integral_init4v_mmx( uint16_t *sum8, uint16_t *sum4, int stride );
-void x264_integral_init4v_sse2( uint16_t *sum8, uint16_t *sum4, int stride );
-void x264_integral_init8v_mmx( uint16_t *sum8, int stride );
-void x264_integral_init8v_sse2( uint16_t *sum8, int stride );
-void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, int stride );
-void x264_mbtree_propagate_cost_sse2( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
+void x264_mc_copy_w4_mmx ( pixel *, intptr_t, pixel *, intptr_t, int );
+void x264_mc_copy_w8_mmx ( pixel *, intptr_t, pixel *, intptr_t, int );
+void x264_mc_copy_w8_sse ( pixel *, intptr_t, pixel *, intptr_t, int );
+void x264_mc_copy_w16_mmx( pixel *, intptr_t, pixel *, intptr_t, int );
+void x264_mc_copy_w16_sse( pixel *, intptr_t, pixel *, intptr_t, int );
+void x264_mc_copy_w16_aligned_sse( pixel *, intptr_t, pixel *, intptr_t, int );
+void x264_mc_copy_w16_avx( uint16_t *, intptr_t, uint16_t *, intptr_t, int );
+void x264_mc_copy_w16_aligned_avx( uint16_t *, intptr_t, uint16_t *, intptr_t, int );
+void x264_prefetch_fenc_420_mmx2( pixel *, intptr_t, pixel *, intptr_t, int );
+void x264_prefetch_fenc_422_mmx2( pixel *, intptr_t, pixel *, intptr_t, int );
+void x264_prefetch_ref_mmx2( pixel *, intptr_t, int );
+void x264_plane_copy_core_mmx2( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
+void x264_plane_copy_c( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
+void x264_plane_copy_interleave_core_mmx2( pixel *dst,  intptr_t i_dst,
+                                           pixel *srcu, intptr_t i_srcu,
+                                           pixel *srcv, intptr_t i_srcv, int w, int h );
+void x264_plane_copy_interleave_core_sse2( pixel *dst,  intptr_t i_dst,
+                                           pixel *srcu, intptr_t i_srcu,
+                                           pixel *srcv, intptr_t i_srcv, int w, int h );
+void x264_plane_copy_interleave_core_avx( pixel *dst,  intptr_t i_dst,
+                                          pixel *srcu, intptr_t i_srcu,
+                                          pixel *srcv, intptr_t i_srcv, int w, int h );
+void x264_plane_copy_interleave_c( pixel *dst,  intptr_t i_dst,
+                                   pixel *srcu, intptr_t i_srcu,
+                                   pixel *srcv, intptr_t i_srcv, int w, int h );
+void x264_plane_copy_deinterleave_mmx( pixel *dstu, intptr_t i_dstu,
+                                       pixel *dstv, intptr_t i_dstv,
+                                       pixel *src,  intptr_t i_src, int w, int h );
+void x264_plane_copy_deinterleave_sse2( pixel *dstu, intptr_t i_dstu,
+                                        pixel *dstv, intptr_t i_dstv,
+                                        pixel *src,  intptr_t i_src, int w, int h );
+void x264_plane_copy_deinterleave_ssse3( uint8_t *dstu, intptr_t i_dstu,
+                                         uint8_t *dstv, intptr_t i_dstv,
+                                         uint8_t *src,  intptr_t i_src, int w, int h );
+void x264_plane_copy_deinterleave_avx( uint16_t *dstu, intptr_t i_dstu,
+                                       uint16_t *dstv, intptr_t i_dstv,
+                                       uint16_t *src,  intptr_t i_src, int w, int h );
+void x264_plane_copy_deinterleave_rgb_sse2 ( pixel *dsta, intptr_t i_dsta,
+                                             pixel *dstb, intptr_t i_dstb,
+                                             pixel *dstc, intptr_t i_dstc,
+                                             pixel *src,  intptr_t i_src, int pw, int w, int h );
+void x264_plane_copy_deinterleave_rgb_ssse3( pixel *dsta, intptr_t i_dsta,
+                                             pixel *dstb, intptr_t i_dstb,
+                                             pixel *dstc, intptr_t i_dstc,
+                                             pixel *src,  intptr_t i_src, int pw, int w, int h );
+void x264_plane_copy_deinterleave_v210_ssse3( uint16_t *dstu, intptr_t i_dstu,
+                                              uint16_t *dstv, intptr_t i_dstv,
+                                              uint32_t *src,  intptr_t i_src, int w, int h );
+void x264_plane_copy_deinterleave_v210_avx  ( uint16_t *dstu, intptr_t i_dstu,
+                                              uint16_t *dstv, intptr_t i_dstv,
+                                              uint32_t *src,  intptr_t i_src, int w, int h );
+void x264_plane_copy_deinterleave_v210_avx2 ( uint16_t *dstu, intptr_t i_dstu,
+                                              uint16_t *dstv, intptr_t i_dstv,
+                                              uint32_t *src,  intptr_t i_src, int w, int h );
+void x264_store_interleave_chroma_mmx2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
+void x264_store_interleave_chroma_sse2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
+void x264_store_interleave_chroma_avx ( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
+void x264_load_deinterleave_chroma_fenc_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
+void x264_load_deinterleave_chroma_fenc_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
+void x264_load_deinterleave_chroma_fenc_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
+void x264_load_deinterleave_chroma_fenc_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
+void x264_load_deinterleave_chroma_fdec_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
+void x264_load_deinterleave_chroma_fdec_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
+void x264_load_deinterleave_chroma_fdec_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
+void x264_load_deinterleave_chroma_fdec_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
+void *x264_memcpy_aligned_mmx( void *dst, const void *src, size_t n );
+void *x264_memcpy_aligned_sse( void *dst, const void *src, size_t n );
+void x264_memzero_aligned_mmx( void *dst, size_t n );
+void x264_memzero_aligned_sse( void *dst, size_t n );
+void x264_memzero_aligned_avx( void *dst, size_t n );
+void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
+void x264_integral_init4h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
+void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
+void x264_integral_init8h_avx ( uint16_t *sum, uint8_t *pix, intptr_t stride );
+void x264_integral_init8h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
+void x264_integral_init4v_mmx  ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
+void x264_integral_init4v_sse2 ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
+void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
+void x264_integral_init4v_avx2( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
+void x264_integral_init8v_mmx ( uint16_t *sum8, intptr_t stride );
+void x264_integral_init8v_sse2( uint16_t *sum8, intptr_t stride );
+void x264_integral_init8v_avx2( uint16_t *sum8, intptr_t stride );
+void x264_mbtree_propagate_cost_sse2( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
+                                      uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
+void x264_mbtree_propagate_cost_avx ( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
+void x264_mbtree_propagate_cost_fma4( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
+                                      uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
+void x264_mbtree_propagate_cost_avx2_fma3( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
+                                           uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 
 #define MC_CHROMA(cpu)\
-void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, int i_dst,\
-                           pixel *src, int i_src,\
+void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, intptr_t i_dst, pixel *src, intptr_t i_src,\
                            int dx, int dy, int i_width, int i_height );
-MC_CHROMA(mmxext)
+MC_CHROMA(mmx2)
 MC_CHROMA(sse2)
-MC_CHROMA(sse2_misalign)
 MC_CHROMA(ssse3)
 MC_CHROMA(ssse3_cache64)
 MC_CHROMA(avx)
-MC_CHROMA(avx_cache64)
+MC_CHROMA(avx2)
 
 #define LOWRES(cpu)\
 void x264_frame_init_lowres_core_##cpu( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,\
-                                        int src_stride, int dst_stride, int width, int height );
-LOWRES(mmxext)
-LOWRES(cache32_mmxext)
+                                        intptr_t src_stride, intptr_t dst_stride, int width, int height );
+LOWRES(mmx2)
+LOWRES(cache32_mmx2)
 LOWRES(sse2)
 LOWRES(ssse3)
+LOWRES(avx)
+LOWRES(xop)
+LOWRES(avx2)
 
 #define PIXEL_AVG_W(width,cpu)\
-void x264_pixel_avg2_w##width##_##cpu( pixel *, int, pixel *, int, pixel *, int );
+void x264_pixel_avg2_w##width##_##cpu( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t );
 /* This declares some functions that don't exist, but that isn't a problem. */
 #define PIXEL_AVG_WALL(cpu)\
 PIXEL_AVG_W(4,cpu); PIXEL_AVG_W(8,cpu); PIXEL_AVG_W(10,cpu); PIXEL_AVG_W(12,cpu); PIXEL_AVG_W(16,cpu); PIXEL_AVG_W(18,cpu); PIXEL_AVG_W(20,cpu);
 
-PIXEL_AVG_WALL(mmxext)
-PIXEL_AVG_WALL(cache32_mmxext)
-PIXEL_AVG_WALL(cache64_mmxext)
+PIXEL_AVG_WALL(mmx2)
+PIXEL_AVG_WALL(cache32_mmx2)
+PIXEL_AVG_WALL(cache64_mmx2)
 PIXEL_AVG_WALL(cache64_sse2)
 PIXEL_AVG_WALL(sse2)
-PIXEL_AVG_WALL(sse2_misalign)
 PIXEL_AVG_WALL(cache64_ssse3)
+PIXEL_AVG_WALL(avx2)
 
 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
-static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, int, pixel *, int, pixel *, int ) =\
+static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t ) =\
 {\
     NULL,\
     x264_pixel_avg2_w4_##name1,\
@@ -188,10 +218,12 @@ static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, int, pixel *, int
 
 #if HIGH_BIT_DEPTH
 /* we can replace w12/w20 with w10/w18 as only 9/17 pixels in fact are important */
-#define x264_pixel_avg2_w12_mmxext       x264_pixel_avg2_w10_mmxext
-#define x264_pixel_avg2_w20_mmxext       x264_pixel_avg2_w18_mmxext
+#define x264_pixel_avg2_w12_mmx2       x264_pixel_avg2_w10_mmx2
+#define x264_pixel_avg2_w20_mmx2       x264_pixel_avg2_w18_mmx2
 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w10_sse2
 #define x264_pixel_avg2_w20_sse2         x264_pixel_avg2_w18_sse2
+#define x264_pixel_avg2_w12_avx2         x264_pixel_avg2_w16_avx2
+#define x264_pixel_avg2_w20_avx2         x264_pixel_avg2_w18_avx2
 #else
 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
 #define x264_pixel_avg2_w12_cache64_ssse3 x264_pixel_avg2_w16_cache64_ssse3
@@ -200,22 +232,24 @@ static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, int, pixel *, int
 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w16_sse2
 #endif // HIGH_BIT_DEPTH
 
-PIXEL_AVG_WTAB(mmxext, mmxext, mmxext, mmxext, mmxext, mmxext)
+PIXEL_AVG_WTAB(mmx2, mmx2, mmx2, mmx2, mmx2, mmx2)
 #if HIGH_BIT_DEPTH
-PIXEL_AVG_WTAB(sse2, mmxext, sse2, sse2, sse2, sse2)
+PIXEL_AVG_WTAB(sse2, mmx2, sse2, sse2, sse2, sse2)
+PIXEL_AVG_WTAB(avx2, mmx2, sse2, avx2, avx2, avx2)
 #else // !HIGH_BIT_DEPTH
 #if ARCH_X86
-PIXEL_AVG_WTAB(cache32_mmxext, mmxext, cache32_mmxext, cache32_mmxext, cache32_mmxext, cache32_mmxext)
-PIXEL_AVG_WTAB(cache64_mmxext, mmxext, cache64_mmxext, cache64_mmxext, cache64_mmxext, cache64_mmxext)
+PIXEL_AVG_WTAB(cache32_mmx2, mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2)
+PIXEL_AVG_WTAB(cache64_mmx2, mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2)
 #endif
-PIXEL_AVG_WTAB(sse2, mmxext, mmxext, sse2, sse2, sse2)
-PIXEL_AVG_WTAB(sse2_misalign, mmxext, mmxext, sse2, sse2, sse2_misalign)
-PIXEL_AVG_WTAB(cache64_sse2, mmxext, cache64_mmxext, cache64_sse2, cache64_sse2, cache64_sse2)
-PIXEL_AVG_WTAB(cache64_ssse3, mmxext, cache64_mmxext, cache64_ssse3, cache64_ssse3, cache64_sse2)
+PIXEL_AVG_WTAB(sse2, mmx2, mmx2, sse2, sse2, sse2)
+PIXEL_AVG_WTAB(cache64_sse2, mmx2, cache64_mmx2, cache64_sse2, cache64_sse2, cache64_sse2)
+PIXEL_AVG_WTAB(cache64_ssse3, mmx2, cache64_mmx2, cache64_ssse3, cache64_ssse3, cache64_sse2)
+PIXEL_AVG_WTAB(cache64_ssse3_atom, mmx2, mmx2, cache64_ssse3, cache64_ssse3, sse2)
+PIXEL_AVG_WTAB(avx2, mmx2, mmx2, sse2, sse2, avx2)
 #endif // HIGH_BIT_DEPTH
 
 #define MC_COPY_WTAB(instr, name1, name2, name3)\
-static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, int, pixel *, int, int ) =\
+static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, intptr_t, pixel *, intptr_t, int ) =\
 {\
     NULL,\
     x264_mc_copy_w4_##name1,\
@@ -225,10 +259,15 @@ static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, int, pixel *, int,
 };
 
 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
-MC_COPY_WTAB(sse2,mmx,mmx,sse2)
+#if HIGH_BIT_DEPTH
+MC_COPY_WTAB(sse,mmx,sse,sse)
+MC_COPY_WTAB(avx,mmx,sse,avx)
+#else
+MC_COPY_WTAB(sse,mmx,mmx,sse)
+#endif
 
 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
-    static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, int, pixel *, int, const x264_weight_t *, int ) =\
+    static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ) =\
 {\
     x264_mc_##function##_w4_##name1,\
     x264_mc_##function##_w4_##name1,\
@@ -239,14 +278,14 @@ MC_COPY_WTAB(sse2,mmx,mmx,sse2)
 };
 
 #if HIGH_BIT_DEPTH
-MC_WEIGHT_WTAB(weight,mmxext,mmxext,mmxext,12)
-MC_WEIGHT_WTAB(offsetadd,mmxext,mmxext,mmxext,12)
-MC_WEIGHT_WTAB(offsetsub,mmxext,mmxext,mmxext,12)
-MC_WEIGHT_WTAB(weight,sse2,mmxext,sse2,12)
-MC_WEIGHT_WTAB(offsetadd,sse2,mmxext,sse2,16)
-MC_WEIGHT_WTAB(offsetsub,sse2,mmxext,sse2,16)
-
-static void x264_weight_cache_mmxext( x264_t *h, x264_weight_t *w )
+MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
+MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
+MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
+MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,12)
+MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,sse2,16)
+MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,sse2,16)
+
+static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 {
     if( w->i_scale == 1<<w->i_denom )
     {
@@ -269,15 +308,16 @@ static void x264_weight_cache_mmxext( x264_t *h, x264_weight_t *w )
     }
 }
 #else
-MC_WEIGHT_WTAB(weight,mmxext,mmxext,mmxext,12)
-MC_WEIGHT_WTAB(offsetadd,mmxext,mmxext,mmxext,12)
-MC_WEIGHT_WTAB(offsetsub,mmxext,mmxext,mmxext,12)
-MC_WEIGHT_WTAB(weight,sse2,mmxext,sse2,16)
-MC_WEIGHT_WTAB(offsetadd,sse2,mmxext,mmxext,16)
-MC_WEIGHT_WTAB(offsetsub,sse2,mmxext,mmxext,16)
+MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
+MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
+MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
+MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,16)
+MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,mmx2,16)
+MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,mmx2,16)
 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
+MC_WEIGHT_WTAB(weight,avx2,ssse3,avx2,16)
 
-static void x264_weight_cache_mmxext( x264_t *h, x264_weight_t *w )
+static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 {
     int i;
     int16_t den1;
@@ -327,10 +367,10 @@ static const uint8_t hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
 static const uint8_t hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
 
 #define MC_LUMA(name,instr1,instr2)\
-static void mc_luma_##name( pixel *dst,    int i_dst_stride,\
-                  pixel *src[4], int i_src_stride,\
-                  int mvx, int mvy,\
-                  int i_width, int i_height, const x264_weight_t *weight )\
+static void mc_luma_##name( pixel *dst,    intptr_t i_dst_stride,\
+                            pixel *src[4], intptr_t i_src_stride,\
+                            int mvx, int mvy,\
+                            int i_width, int i_height, const x264_weight_t *weight )\
 {\
     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
@@ -350,22 +390,25 @@ static void mc_luma_##name( pixel *dst,    int i_dst_stride,\
         x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
 }
 
-MC_LUMA(mmxext,mmxext,mmx)
-MC_LUMA(sse2,sse2,sse2)
-#if !HIGH_BIT_DEPTH
+MC_LUMA(mmx2,mmx2,mmx)
+MC_LUMA(sse2,sse2,sse)
+#if HIGH_BIT_DEPTH
+MC_LUMA(avx2,avx2,avx)
+#else
 #if ARCH_X86
-MC_LUMA(cache32_mmxext,cache32_mmxext,mmx)
-MC_LUMA(cache64_mmxext,cache64_mmxext,mmx)
+MC_LUMA(cache32_mmx2,cache32_mmx2,mmx)
+MC_LUMA(cache64_mmx2,cache64_mmx2,mmx)
 #endif
-MC_LUMA(cache64_sse2,cache64_sse2,sse2)
-MC_LUMA(cache64_ssse3,cache64_ssse3,sse2)
+MC_LUMA(cache64_sse2,cache64_sse2,sse)
+MC_LUMA(cache64_ssse3,cache64_ssse3,sse)
+MC_LUMA(cache64_ssse3_atom,cache64_ssse3_atom,sse)
 #endif // !HIGH_BIT_DEPTH
 
 #define GET_REF(name)\
-static pixel *get_ref_##name( pixel *dst,   int *i_dst_stride,\
-                         pixel *src[4], int i_src_stride,\
-                         int mvx, int mvy,\
-                         int i_width, int i_height, const x264_weight_t *weight )\
+static pixel *get_ref_##name( pixel *dst,   intptr_t *i_dst_stride,\
+                              pixel *src[4], intptr_t i_src_stride,\
+                              int mvx, int mvy,\
+                              int i_width, int i_height, const x264_weight_t *weight )\
 {\
     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
@@ -392,26 +435,68 @@ static pixel *get_ref_##name( pixel *dst,   int *i_dst_stride,\
     }\
 }
 
-GET_REF(mmxext)
+GET_REF(mmx2)
 GET_REF(sse2)
+GET_REF(avx2)
 #if !HIGH_BIT_DEPTH
 #if ARCH_X86
-GET_REF(cache32_mmxext)
-GET_REF(cache64_mmxext)
+GET_REF(cache32_mmx2)
+GET_REF(cache64_mmx2)
 #endif
-GET_REF(sse2_misalign)
 GET_REF(cache64_sse2)
 GET_REF(cache64_ssse3)
+GET_REF(cache64_ssse3_atom)
 #endif // !HIGH_BIT_DEPTH
 
+#define GET_REF_MPEG2(name)\
+static pixel *get_ref_mpeg2_##name( pixel *dst,   intptr_t *i_dst_stride,\
+                                    pixel *src[4], intptr_t i_src_stride,\
+                                    int mvx, int mvy,\
+                                    int i_width, int i_height, const x264_weight_t *weight )\
+{\
+    mvx >>= 1;\
+    mvy >>= 1;\
+    pixel *src1 = src[0] + (mvy>>1)*i_src_stride + (mvx>>1);\
+    pixel *dst_bak = dst;\
+    if( !((mvx|mvy)&1) )\
+    {\
+        *i_dst_stride = i_src_stride;\
+        return src1;\
+    }\
+    else if( (mvx&mvy)&1 ) /* centre hpel positions */\
+    {\
+        pixel *srcp = src1 + i_src_stride;\
+        for( int y = 0; y < i_height; y++ )\
+        {\
+            for( int x = 0; x < i_width; x++ )\
+                dst[x] = ( src1[x] + src1[x+1] + srcp[x] + srcp[x+1] + 2 ) >> 2;\
+            dst  += *i_dst_stride;\
+            src1  = srcp;\
+            srcp += i_src_stride;\
+        }\
+    }\
+    else /* horizontal/vertical hpel positions */\
+    {\
+        pixel *src2 = src1 + (mvy&1)*i_src_stride + (mvx&1);\
+        x264_pixel_avg_wtab_##name[i_width>>2](\
+                dst, *i_dst_stride, src1, i_src_stride,\
+                src2, i_height );\
+    }\
+    return dst_bak;\
+}
+
+GET_REF_MPEG2(mmx2)
+GET_REF_MPEG2(sse2)
+GET_REF_MPEG2(avx2)
+
 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
-void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, int stride, int width);\
-void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, int width );\
-void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, int width );\
+void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, intptr_t stride, intptr_t width);\
+void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, intptr_t width );\
+void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, intptr_t width );\
 static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,\
-                             int stride, int width, int height, int16_t *buf )\
+                                    intptr_t stride, int width, int height, int16_t *buf )\
 {\
-    int realign = (intptr_t)src & (align-1);\
+    intptr_t realign = (intptr_t)src & (align-1);\
     src -= realign;\
     dstv -= realign;\
     dstc -= realign;\
@@ -419,8 +504,8 @@ static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel
     width += realign;\
     while( height-- )\
     {\
-        x264_hpel_filter_v_##cpuv( dstv, src, buf+8, stride, width );\
-        x264_hpel_filter_c_##cpuc( dstc, buf+8, width );\
+        x264_hpel_filter_v_##cpuv( dstv, src, buf+16, stride, width );\
+        x264_hpel_filter_c_##cpuc( dstc, buf+16, width );\
         x264_hpel_filter_h_##cpuh( dsth, src, width );\
         dsth += stride;\
         dstv += stride;\
@@ -430,44 +515,45 @@ static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel
     x264_sfence();\
 }
 
-HPEL(8, mmxext, mmxext, mmxext, mmxext)
+HPEL(8, mmx2, mmx2, mmx2, mmx2)
 #if HIGH_BIT_DEPTH
 HPEL(16, sse2, sse2, sse2, sse2)
 #else // !HIGH_BIT_DEPTH
-HPEL(16, sse2_amd, mmxext, mmxext, sse2)
+HPEL(16, sse2_amd, mmx2, mmx2, sse2)
 #if ARCH_X86_64
-void x264_hpel_filter_sse2( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height, int16_t *buf );
-void x264_hpel_filter_ssse3( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height, int16_t *buf );
-void x264_hpel_filter_avx( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height, int16_t *buf );
+void x264_hpel_filter_sse2 ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
+void x264_hpel_filter_ssse3( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
+void x264_hpel_filter_avx  ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
+void x264_hpel_filter_avx2 ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
 #else
 HPEL(16, sse2, sse2, sse2, sse2)
 HPEL(16, ssse3, ssse3, ssse3, ssse3)
-HPEL(16, avx, avx, avx, ssse3)
+HPEL(16, avx, avx, avx, avx)
+HPEL(32, avx2, avx2, avx2, avx2)
 #endif
-HPEL(16, sse2_misalign, sse2, sse2_misalign, sse2)
 #endif // HIGH_BIT_DEPTH
 
-static void x264_plane_copy_mmxext( pixel *dst, int i_dst, pixel *src, int i_src, int w, int h )
+static void x264_plane_copy_mmx2( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )
 {
     int c_w = 16/sizeof(pixel) - 1;
     if( w < 256 ) { // tiny resolutions don't want non-temporal hints. dunno the exact threshold.
         x264_plane_copy_c( dst, i_dst, src, i_src, w, h );
     } else if( !(w&c_w) ) {
-        x264_plane_copy_core_mmxext( dst, i_dst, src, i_src, w, h );
+        x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, w, h );
     } else if( i_src > 0 ) {
         // have to use plain memcpy on the last line (in memory order) to avoid overreading src
-        x264_plane_copy_core_mmxext( dst, i_dst, src, i_src, (w+c_w)&~c_w, h-1 );
+        x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, (w+c_w)&~c_w, h-1 );
         memcpy( dst+i_dst*(h-1), src+i_src*(h-1), w*sizeof(pixel) );
     } else {
         memcpy( dst, src, w*sizeof(pixel) );
-        x264_plane_copy_core_mmxext( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h-1 );
+        x264_plane_copy_core_mmx2( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h-1 );
     }
 }
 
 #define PLANE_INTERLEAVE(cpu) \
-static void x264_plane_copy_interleave_##cpu( pixel *dst, int i_dst,\
-                                              pixel *srcu, int i_srcu,\
-                                              pixel *srcv, int i_srcv, int w, int h )\
+static void x264_plane_copy_interleave_##cpu( pixel *dst,  intptr_t i_dst,\
+                                              pixel *srcu, intptr_t i_srcu,\
+                                              pixel *srcv, intptr_t i_srcv, int w, int h )\
 {\
     if( !(w&15) ) {\
         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
@@ -482,19 +568,126 @@ static void x264_plane_copy_interleave_##cpu( pixel *dst, int i_dst,\
     }\
 }
 
-PLANE_INTERLEAVE(mmxext)
+PLANE_INTERLEAVE(mmx2)
 PLANE_INTERLEAVE(sse2)
 #if HIGH_BIT_DEPTH
 PLANE_INTERLEAVE(avx)
 #endif
 
+#if HAVE_X86_INLINE_ASM
+#define CLIP_ADD(s,x)\
+do\
+{\
+    int temp;\
+    asm("movd       %0, %%xmm0     \n"\
+        "movd       %2, %%xmm1     \n"\
+        "paddsw %%xmm1, %%xmm0     \n"\
+        "movd   %%xmm0, %1         \n"\
+        :"+m"(s), "=&r"(temp)\
+        :"m"(x)\
+    );\
+    s = temp;\
+} while(0)
+
+#define CLIP_ADD2(s,x)\
+do\
+{\
+    asm("movd       %0, %%xmm0     \n"\
+        "movd       %1, %%xmm1     \n"\
+        "paddsw %%xmm1, %%xmm0     \n"\
+        "movd   %%xmm0, %0         \n"\
+        :"+m"(M32(s))\
+        :"m"(M32(x))\
+    );\
+} while(0)
+#else
+#define CLIP_ADD(s,x) (s) = X264_MIN((s)+(x),(1<<15)-1)
+#define CLIP_ADD2(s,x)\
+do\
+{\
+    CLIP_ADD((s)[0], (x)[0]);\
+    CLIP_ADD((s)[1], (x)[1]);\
+} while(0)
+#endif
+
+#define PROPAGATE_LIST(cpu)\
+void x264_mbtree_propagate_list_internal_##cpu( int16_t (*mvs)[2], int16_t *propagate_amount,\
+                                                uint16_t *lowres_costs, int16_t *output,\
+                                                int bipred_weight, int mb_y, int len );\
+\
+static void x264_mbtree_propagate_list_##cpu( x264_t *h, uint16_t *ref_costs, int16_t (*mvs)[2],\
+                                              int16_t *propagate_amount, uint16_t *lowres_costs,\
+                                              int bipred_weight, int mb_y, int len, int list )\
+{\
+    int16_t *current = h->scratch_buffer2;\
+\
+    x264_mbtree_propagate_list_internal_##cpu( mvs, propagate_amount, lowres_costs,\
+                                               current, bipred_weight, mb_y, len );\
+\
+    unsigned stride = h->mb.i_mb_stride;\
+    unsigned width = h->mb.i_mb_width;\
+    unsigned height = h->mb.i_mb_height;\
+\
+    for( unsigned i = 0; i < len; current += 32 )\
+    {\
+        int end = X264_MIN( i+8, len );\
+        for( ; i < end; i++, current += 2 )\
+        {\
+            if( !(lowres_costs[i] & (1 << (list+LOWRES_COST_SHIFT))) )\
+                continue;\
+\
+            unsigned mbx = current[0];\
+            unsigned mby = current[1];\
+            unsigned idx0 = mbx + mby * stride;\
+            unsigned idx2 = idx0 + stride;\
+\
+            /* Shortcut for the simple/common case of zero MV */\
+            if( !M32( mvs[i] ) )\
+            {\
+                CLIP_ADD( ref_costs[idx0], current[16] );\
+                continue;\
+            }\
+\
+            if( mbx < width-1 && mby < height-1 )\
+            {\
+                CLIP_ADD2( ref_costs+idx0, current+16 );\
+                CLIP_ADD2( ref_costs+idx2, current+32 );\
+            }\
+            else\
+            {\
+                /* Note: this takes advantage of unsigned representation to\
+                 * catch negative mbx/mby. */\
+                if( mby < height )\
+                {\
+                    if( mbx < width )\
+                        CLIP_ADD( ref_costs[idx0+0], current[16] );\
+                    if( mbx+1 < width )\
+                        CLIP_ADD( ref_costs[idx0+1], current[17] );\
+                }\
+                if( mby+1 < height )\
+                {\
+                    if( mbx < width )\
+                        CLIP_ADD( ref_costs[idx2+0], current[32] );\
+                    if( mbx+1 < width )\
+                        CLIP_ADD( ref_costs[idx2+1], current[33] );\
+                }\
+            }\
+        }\
+    }\
+}
+
+PROPAGATE_LIST(ssse3)
+PROPAGATE_LIST(avx)
+#undef CLIP_ADD
+#undef CLIP_ADD2
+
 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 {
     if( !(cpu&X264_CPU_MMX) )
         return;
 
-    pf->load_deinterleave_8x8x2_fenc = x264_load_deinterleave_8x8x2_fenc_mmx;
-    pf->load_deinterleave_8x8x2_fdec = x264_load_deinterleave_8x8x2_fdec_mmx;
+    pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_mmx;
+    pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_mmx;
 
     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
 
@@ -507,37 +700,48 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
     pf->integral_init4v = x264_integral_init4v_mmx;
     pf->integral_init8v = x264_integral_init8v_mmx;
 
-    if( !(cpu&X264_CPU_MMXEXT) )
+    if( !(cpu&X264_CPU_MMX2) )
         return;
 
-    pf->plane_copy = x264_plane_copy_mmxext;
-    pf->plane_copy_interleave = x264_plane_copy_interleave_mmxext;
-    pf->store_interleave_8x8x2 = x264_store_interleave_8x8x2_mmxext;
-
-    pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmxext;
-    pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmxext;
-    pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmxext;
-    pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmxext;
-    pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmxext;
-    pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmxext;
-    pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmxext;
-    pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmxext;
-
-    pf->mc_luma = mc_luma_mmxext;
-    pf->get_ref = get_ref_mmxext;
-    pf->mc_chroma = x264_mc_chroma_mmxext;
-    pf->hpel_filter = x264_hpel_filter_mmxext;
-    pf->weight = x264_mc_weight_wtab_mmxext;
-    pf->weight_cache = x264_weight_cache_mmxext;
-    pf->offsetadd = x264_mc_offsetadd_wtab_mmxext;
-    pf->offsetsub = x264_mc_offsetsub_wtab_mmxext;
-
-    pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmxext;
+    pf->prefetch_fenc_420 = x264_prefetch_fenc_420_mmx2;
+    pf->prefetch_fenc_422 = x264_prefetch_fenc_422_mmx2;
+    pf->prefetch_ref  = x264_prefetch_ref_mmx2;
+
+    pf->plane_copy = x264_plane_copy_mmx2;
+    pf->plane_copy_interleave = x264_plane_copy_interleave_mmx2;
+    pf->store_interleave_chroma = x264_store_interleave_chroma_mmx2;
+
+    pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmx2;
+    pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmx2;
+    pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmx2;
+    pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmx2;
+    pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmx2;
+    pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_mmx2;
+    pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmx2;
+    pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmx2;
+    pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmx2;
+
+    pf->mc_luma = mc_luma_mmx2;
+    pf->get_ref = get_ref_mmx2;
+    pf->mc_chroma = x264_mc_chroma_mmx2;
+    pf->hpel_filter = x264_hpel_filter_mmx2;
+    pf->weight = x264_mc_weight_wtab_mmx2;
+    pf->weight_cache = x264_weight_cache_mmx2;
+    pf->offsetadd = x264_mc_offsetadd_wtab_mmx2;
+    pf->offsetsub = x264_mc_offsetsub_wtab_mmx2;
+
+    pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmx2;
+
+    if( cpu&X264_CPU_SSE )
+    {
+        pf->memcpy_aligned  = x264_memcpy_aligned_sse;
+        pf->memzero_aligned = x264_memzero_aligned_sse;
+    }
 
 #if HIGH_BIT_DEPTH
 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
     if( cpu&(X264_CPU_CACHELINE_32|X264_CPU_CACHELINE_64) )
-        pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
+        pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 #endif
 
     if( !(cpu&X264_CPU_SSE2) )
@@ -545,8 +749,8 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 
     pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 
-    pf->load_deinterleave_8x8x2_fenc = x264_load_deinterleave_8x8x2_fenc_sse2;
-    pf->load_deinterleave_8x8x2_fdec = x264_load_deinterleave_8x8x2_fdec_sse2;
+    pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
+    pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 
     pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
@@ -558,12 +762,10 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
         pf->hpel_filter = x264_hpel_filter_sse2;
     }
 
-    pf->memcpy_aligned  = x264_memcpy_aligned_sse2;
-    pf->memzero_aligned = x264_memzero_aligned_sse2;
     pf->integral_init4v = x264_integral_init4v_sse2;
     pf->integral_init8v = x264_integral_init8v_sse2;
     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
-    pf->store_interleave_8x8x2 = x264_store_interleave_8x8x2_sse2;
+    pf->store_interleave_chroma = x264_store_interleave_chroma_sse2;
     pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
     pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 
@@ -575,11 +777,12 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_sse2;
     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_sse2;
     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_sse2;
+    pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_sse2;
     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_sse2;
     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_sse2;
     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_sse2;
 
-    pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse2;
+    pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
     pf->weight = x264_mc_weight_wtab_sse2;
 
     if( !(cpu&X264_CPU_STACK_MOD4) )
@@ -589,92 +792,95 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
         return;
 
     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
+    pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_ssse3;
+    pf->mbtree_propagate_list = x264_mbtree_propagate_list_ssse3;
 
-    if( (cpu&X264_CPU_SHUFFLE_IS_FAST) && !(cpu&X264_CPU_SLOW_ATOM) )
+    if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
         pf->integral_init4v = x264_integral_init4v_ssse3;
 
     if( !(cpu&X264_CPU_AVX) )
         return;
 
-    pf->load_deinterleave_8x8x2_fenc = x264_load_deinterleave_8x8x2_fenc_avx;
-    pf->load_deinterleave_8x8x2_fdec = x264_load_deinterleave_8x8x2_fdec_avx;
+    pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
+    pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_avx;
+    pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_avx;
     pf->plane_copy_interleave        = x264_plane_copy_interleave_avx;
     pf->plane_copy_deinterleave      = x264_plane_copy_deinterleave_avx;
-    pf->store_interleave_8x8x2       = x264_store_interleave_8x8x2_avx;
+    pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx;
+    pf->store_interleave_chroma      = x264_store_interleave_chroma_avx;
+    pf->copy[PIXEL_16x16]            = x264_mc_copy_w16_aligned_avx;
 
     if( !(cpu&X264_CPU_STACK_MOD4) )
         pf->mc_chroma = x264_mc_chroma_avx;
+
+    if( cpu&X264_CPU_XOP )
+        pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
+
+    if( cpu&X264_CPU_AVX2 )
+    {
+        pf->mc_luma = mc_luma_avx2;
+        pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx2;
+    }
 #else // !HIGH_BIT_DEPTH
-    pf->prefetch_fenc = x264_prefetch_fenc_mmxext;
-    pf->prefetch_ref  = x264_prefetch_ref_mmxext;
 
 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
     if( cpu&X264_CPU_CACHELINE_32 )
     {
-        pf->mc_luma = mc_luma_cache32_mmxext;
-        pf->get_ref = get_ref_cache32_mmxext;
-        pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
+        pf->mc_luma = mc_luma_cache32_mmx2;
+        pf->get_ref = get_ref_cache32_mmx2;
+        pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
     }
     else if( cpu&X264_CPU_CACHELINE_64 )
     {
-        pf->mc_luma = mc_luma_cache64_mmxext;
-        pf->get_ref = get_ref_cache64_mmxext;
-        pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmxext;
+        pf->mc_luma = mc_luma_cache64_mmx2;
+        pf->get_ref = get_ref_cache64_mmx2;
+        pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
     }
 #endif
 
     if( !(cpu&X264_CPU_SSE2) )
         return;
 
-    pf->memcpy_aligned = x264_memcpy_aligned_sse2;
-    pf->memzero_aligned = x264_memzero_aligned_sse2;
     pf->integral_init4v = x264_integral_init4v_sse2;
     pf->integral_init8v = x264_integral_init8v_sse2;
     pf->hpel_filter = x264_hpel_filter_sse2_amd;
     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
+    pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_sse2;
 
-    if( cpu&X264_CPU_SSE2_IS_SLOW )
-        return;
-
-    pf->weight = x264_mc_weight_wtab_sse2;
-    if( !(cpu&X264_CPU_SLOW_ATOM) )
-    {
-        pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
-        pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
-    }
-
-    pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse2;
-    pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
-    pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
-    pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
-    pf->avg[PIXEL_8x8]  = x264_pixel_avg_8x8_sse2;
-    pf->avg[PIXEL_8x4]  = x264_pixel_avg_8x4_sse2;
-    pf->hpel_filter = x264_hpel_filter_sse2;
-    if( cpu&X264_CPU_SSE_MISALIGN )
-        pf->hpel_filter = x264_hpel_filter_sse2_misalign;
-    pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
-    if( !(cpu&X264_CPU_STACK_MOD4) )
-        pf->mc_chroma = x264_mc_chroma_sse2;
-
-    if( cpu&X264_CPU_SSE2_IS_FAST )
+    if( !(cpu&X264_CPU_SSE2_IS_SLOW) )
     {
-        pf->store_interleave_8x8x2  = x264_store_interleave_8x8x2_sse2; // FIXME sse2fast? sse2medium?
-        pf->load_deinterleave_8x8x2_fenc = x264_load_deinterleave_8x8x2_fenc_sse2;
-        pf->load_deinterleave_8x8x2_fdec = x264_load_deinterleave_8x8x2_fdec_sse2;
-        pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
-        pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
-        pf->mc_luma = mc_luma_sse2;
-        pf->get_ref = get_ref_sse2;
-        if( cpu&X264_CPU_CACHELINE_64 )
+        pf->weight = x264_mc_weight_wtab_sse2;
+        if( !(cpu&X264_CPU_SLOW_ATOM) )
         {
-            pf->mc_luma = mc_luma_cache64_sse2;
-            pf->get_ref = get_ref_cache64_sse2;
+            pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
+            pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
         }
-        if( cpu&X264_CPU_SSE_MISALIGN )
+
+        pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
+        pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
+        pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
+        pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
+        pf->avg[PIXEL_8x8]  = x264_pixel_avg_8x8_sse2;
+        pf->avg[PIXEL_8x4]  = x264_pixel_avg_8x4_sse2;
+        pf->hpel_filter = x264_hpel_filter_sse2;
+        pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
+        if( !(cpu&X264_CPU_STACK_MOD4) )
+            pf->mc_chroma = x264_mc_chroma_sse2;
+
+        if( cpu&X264_CPU_SSE2_IS_FAST )
         {
-            pf->get_ref = get_ref_sse2_misalign;
-            if( !(cpu&X264_CPU_STACK_MOD4) )
-                pf->mc_chroma = x264_mc_chroma_sse2_misalign;
+            pf->store_interleave_chroma = x264_store_interleave_chroma_sse2; // FIXME sse2fast? sse2medium?
+            pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
+            pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
+            pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
+            pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
+            pf->mc_luma = mc_luma_sse2;
+            pf->get_ref = get_ref_sse2;
+            if( cpu&X264_CPU_CACHELINE_64 )
+            {
+                pf->mc_luma = mc_luma_cache64_sse2;
+                pf->get_ref = get_ref_cache64_sse2;
+            }
         }
     }
 
@@ -686,16 +892,28 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_ssse3;
     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_ssse3;
     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_ssse3;
+    pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_ssse3;
     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_ssse3;
     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_ssse3;
     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_ssse3;
+    pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_ssse3;
+    pf->mbtree_propagate_list = x264_mbtree_propagate_list_ssse3;
 
-    pf->load_deinterleave_8x8x2_fenc = x264_load_deinterleave_8x8x2_fenc_ssse3;
-    pf->load_deinterleave_8x8x2_fdec = x264_load_deinterleave_8x8x2_fdec_ssse3;
-    pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
+    if( !(cpu&X264_CPU_SLOW_PSHUFB) )
+    {
+        pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_ssse3;
+        pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_ssse3;
+        pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
+    }
 
-    pf->hpel_filter = x264_hpel_filter_ssse3;
-    pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
+    if( !(cpu&X264_CPU_SLOW_PALIGNR) )
+    {
+#if ARCH_X86_64
+        if( !(cpu&X264_CPU_SLOW_ATOM) ) /* The 64-bit version is slower, but the 32-bit version is faster? */
+#endif
+            pf->hpel_filter = x264_hpel_filter_ssse3;
+        pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
+    }
     if( !(cpu&X264_CPU_STACK_MOD4) )
         pf->mc_chroma = x264_mc_chroma_ssse3;
 
@@ -705,13 +923,17 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
             pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
         pf->mc_luma = mc_luma_cache64_ssse3;
         pf->get_ref = get_ref_cache64_ssse3;
-
-        /* ssse3 weight is slower on Nehalem, so only assign here. */
-        pf->weight_cache = x264_weight_cache_ssse3;
-        pf->weight = x264_mc_weight_wtab_ssse3;
+        if( cpu&X264_CPU_SLOW_ATOM )
+        {
+            pf->mc_luma = mc_luma_cache64_ssse3_atom;
+            pf->get_ref = get_ref_cache64_ssse3_atom;
+        }
     }
 
-    if( (cpu&X264_CPU_SHUFFLE_IS_FAST) && !(cpu&X264_CPU_SLOW_ATOM) )
+    pf->weight_cache = x264_weight_cache_ssse3;
+    pf->weight = x264_mc_weight_wtab_ssse3;
+
+    if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
         pf->integral_init4v = x264_integral_init4v_ssse3;
 
     if( !(cpu&X264_CPU_SSE4) )
@@ -723,9 +945,62 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
     if( !(cpu&X264_CPU_AVX) )
         return;
 
+    pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
     pf->integral_init8h = x264_integral_init8h_avx;
     pf->hpel_filter = x264_hpel_filter_avx;
+
     if( !(cpu&X264_CPU_STACK_MOD4) )
         pf->mc_chroma = x264_mc_chroma_avx;
+
+    if( cpu&X264_CPU_XOP )
+        pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
+
+    if( cpu&X264_CPU_AVX2 )
+    {
+        pf->hpel_filter = x264_hpel_filter_avx2;
+        pf->mc_chroma = x264_mc_chroma_avx2;
+        pf->weight = x264_mc_weight_wtab_avx2;
+        pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_avx2;
+        pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_avx2;
+        pf->integral_init8v = x264_integral_init8v_avx2;
+        pf->integral_init4v = x264_integral_init4v_avx2;
+        pf->integral_init8h = x264_integral_init8h_avx2;
+        pf->integral_init4h = x264_integral_init4h_avx2;
+        pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx2;
+    }
 #endif // HIGH_BIT_DEPTH
+
+    if( !(cpu&X264_CPU_AVX) )
+        return;
+    pf->memzero_aligned = x264_memzero_aligned_avx;
+    pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
+    pf->mbtree_propagate_list = x264_mbtree_propagate_list_avx;
+
+    if( cpu&X264_CPU_FMA4 )
+        pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_fma4;
+
+    if( !(cpu&X264_CPU_AVX2) )
+        return;
+    pf->get_ref = get_ref_avx2;
+
+    if( cpu&X264_CPU_FMA3 )
+        pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx2_fma3;
+}
+
+void x264_mc_init_mmx_mpeg2( int cpu, x264_mc_functions_t *pf )
+{
+    if( !(cpu&X264_CPU_MMX) )
+        return;
+
+    pf->get_ref = get_ref_mpeg2_mmx2;
+
+    if( !(cpu&X264_CPU_SSE2) )
+        return;
+
+    pf->get_ref = get_ref_mpeg2_sse2;
+
+    if( !(cpu&X264_CPU_AVX2) )
+        return;
+
+    pf->get_ref = get_ref_mpeg2_avx2;
 }