x86: Use SSE instead of SSE2 for copying data

Reduces code size because movaps/movups is one byte shorter than movdqa/movdqu.
Also merge MMX and SSE versions of memcpy_aligned into a single macro.

diff --git a/common/cpu.c b/common/cpu.c
index 8463ede3720a045cd7033a4f3d7998b2a6f8165e..c5e34a9fd649f1d2b5712de2af1a36e1df1b4d13 100644 (file)
--- a/common/cpu.c
+++ b/common/cpu.c
@@ -51,7 +51,7 @@ const x264_cpu_name_t x264_cpu_names[] =
 //  {"MMX",         X264_CPU_MMX}, // we don't support asm on mmx1 cpus anymore
     {"MMX2",        X264_CPU_MMX|X264_CPU_MMX2},
     {"MMXEXT",      X264_CPU_MMX|X264_CPU_MMX2},
-//  {"SSE",         X264_CPU_MMX|X264_CPU_MMX2|X264_CPU_SSE}, // there are no sse1 functions in x264
+    {"SSE",         X264_CPU_MMX|X264_CPU_MMX2|X264_CPU_SSE},
 #define SSE2 X264_CPU_MMX|X264_CPU_MMX2|X264_CPU_SSE|X264_CPU_SSE2
     {"SSE2Slow",    SSE2|X264_CPU_SSE2_IS_SLOW},
     {"SSE2",        SSE2},

diff --git a/common/pixel.c b/common/pixel.c
index be6e4b2bf9e77c1ee6904836710140773d8e6796..8b5b7e22925169494f2bfb4e6168c5a69202ef32 100644 (file)
--- a/common/pixel.c
+++ b/common/pixel.c
@@ -528,6 +528,7 @@ void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[36], int res[3] )
 INTRA_MBCMP_8x8( sad,, _c )
 INTRA_MBCMP_8x8(sa8d,, _c )
 #if HIGH_BIT_DEPTH && HAVE_MMX
+#define x264_predict_8x8_v_sse2 x264_predict_8x8_v_sse
 INTRA_MBCMP_8x8( sad, _mmx2,  _c )
 INTRA_MBCMP_8x8(sa8d, _sse2,  _sse2 )
 #endif
@@ -554,6 +555,9 @@ INTRA_MBCMP(satd, 16x16,  v, h, dc,  ,, _c )
 
 #if HAVE_MMX
 #if HIGH_BIT_DEPTH
+#define x264_predict_8x8c_v_sse2 x264_predict_8x8c_v_sse
+#define x264_predict_8x16c_v_sse2 x264_predict_8x16c_v_sse
+#define x264_predict_16x16_v_sse2 x264_predict_16x16_v_sse
 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _mmx2, _c )
 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _mmx2, _c )
 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _mmx2, _mmx2 )

diff --git a/common/x86/mc-a.asm b/common/x86/mc-a.asm
index 9b46680b45a9eeaf2b38e31208e2f18ab667ed71..bc9b011b035a03d4b66662194b89b0dc2c0dc13d 100644 (file)
--- a/common/x86/mc-a.asm
+++ b/common/x86/mc-a.asm
@@ -1233,10 +1233,10 @@ cglobal mc_copy_w%1, 5,7,8*(%%w/2)
 INIT_MMX mmx
 MC_COPY  8
 MC_COPY 16
-INIT_XMM sse2
+INIT_XMM sse
 MC_COPY  8
 MC_COPY 16
-INIT_XMM aligned, sse2
+INIT_XMM aligned, sse
 MC_COPY 16
 
 

diff --git a/common/x86/mc-a2.asm b/common/x86/mc-a2.asm
index 21fc217c6513f4cfb455dda9c6ef71bc527901d8..f1c179a8cfaff9b3fbb3f75e39065ce566cd1169 100644 (file)
--- a/common/x86/mc-a2.asm
+++ b/common/x86/mc-a2.asm
@@ -1131,65 +1131,45 @@ PLANE_DEINTERLEAVE
 ;-----------------------------------------------------------------------------
 ; void *memcpy_aligned( void *dst, const void *src, size_t n );
 ;-----------------------------------------------------------------------------
-INIT_MMX
-cglobal memcpy_aligned_mmx, 3,3
-    test r2d, 16
-    jz .copy32start
-    movq mm0, [r1 + r2 - 16]
-    movq mm1, [r1 + r2 -  8]
-    movq [r0 + r2 - 16], mm0
-    movq [r0 + r2 -  8], mm1
-    sub  r2d, 16
-.copy32start
-    test r2d, r2d
-    jz .ret
-.copy32:
-    movq mm0, [r1 + r2 - 32]
-    movq mm1, [r1 + r2 - 24]
-    movq mm2, [r1 + r2 - 16]
-    movq mm3, [r1 + r2 -  8]
-    movq [r0 + r2 - 32], mm0
-    movq [r0 + r2 - 24], mm1
-    movq [r0 + r2 - 16], mm2
-    movq [r0 + r2 -  8], mm3
-    sub  r2d, 32
-    jg .copy32
-.ret
-    RET
-
-;-----------------------------------------------------------------------------
-; void *memcpy_aligned( void *dst, const void *src, size_t n );
-;-----------------------------------------------------------------------------
-cglobal memcpy_aligned_sse2, 3,3
+%macro MEMCPY 0
+cglobal memcpy_aligned, 3,3
+%if mmsize == 16
     test r2d, 16
-    jz .copy32
-    movdqa xmm0, [r1 + r2 - 16]
-    movdqa [r0 + r2 - 16], xmm0
+    jz .copy2
+    mova  m0, [r1+r2-16]
+    mova [r0+r2-16], m0
     sub  r2d, 16
-.copy32:
-    test r2d, 32
-    jz .copy64start
-    movdqa xmm0, [r1 + r2 - 32]
-    movdqa [r0 + r2 - 32], xmm0
-    movdqa xmm1, [r1 + r2 - 16]
-    movdqa [r0 + r2 - 16], xmm1
-    sub  r2d, 32
-.copy64start
+.copy2:
+%endif
+    test r2d, 2*mmsize
+    jz .copy4start
+    mova  m0, [r1+r2-1*mmsize]
+    mova  m1, [r1+r2-2*mmsize]
+    mova [r0+r2-1*mmsize], m0
+    mova [r0+r2-2*mmsize], m1
+    sub  r2d, 2*mmsize
+.copy4start:
     test r2d, r2d
     jz .ret
-.copy64:
-    movdqa xmm0, [r1 + r2 - 64]
-    movdqa [r0 + r2 - 64], xmm0
-    movdqa xmm1, [r1 + r2 - 48]
-    movdqa [r0 + r2 - 48], xmm1
-    movdqa xmm2, [r1 + r2 - 32]
-    movdqa [r0 + r2 - 32], xmm2
-    movdqa xmm3, [r1 + r2 - 16]
-    movdqa [r0 + r2 - 16], xmm3
-    sub  r2d, 64
-    jg .copy64
+.copy4:
+    mova  m0, [r1+r2-1*mmsize]
+    mova  m1, [r1+r2-2*mmsize]
+    mova  m2, [r1+r2-3*mmsize]
+    mova  m3, [r1+r2-4*mmsize]
+    mova [r0+r2-1*mmsize], m0
+    mova [r0+r2-2*mmsize], m1
+    mova [r0+r2-3*mmsize], m2
+    mova [r0+r2-4*mmsize], m3
+    sub  r2d, 4*mmsize
+    jg .copy4
 .ret:
     REP_RET
+%endmacro
+
+INIT_MMX mmx
+MEMCPY
+INIT_XMM sse
+MEMCPY
 
 ;-----------------------------------------------------------------------------
 ; void *memzero_aligned( void *dst, size_t n );
@@ -1198,7 +1178,11 @@ cglobal memcpy_aligned_sse2, 3,3
 cglobal memzero_aligned, 2,2
     add  r0, r1
     neg  r1
+%if mmsize == 8
     pxor m0, m0
+%else
+    xorps m0, m0
+%endif
 .loop:
 %assign i 0
 %rep 8
@@ -1212,7 +1196,7 @@ cglobal memzero_aligned, 2,2
 
 INIT_MMX mmx
 MEMZERO
-INIT_XMM sse2
+INIT_XMM sse
 MEMZERO
 
 

diff --git a/common/x86/mc-c.c b/common/x86/mc-c.c
index c43a0841f55cf5e7d3706faff3a2a7c1ab65c5a9..b0a31e1537d831b2fabf45da433d7e84ae56ac8b 100644 (file)
--- a/common/x86/mc-c.c
+++ b/common/x86/mc-c.c
@@ -75,12 +75,12 @@ MC_WEIGHT( 20, ssse3 )
 #undef MC_OFFSET
 #undef MC_WEIGHT
 
-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_sse2 ( 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_sse2( pixel *, intptr_t, pixel *, intptr_t, int );
-void x264_mc_copy_w16_aligned_sse2( pixel *, intptr_t, pixel *, intptr_t, int );
+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_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 );
@@ -121,10 +121,10 @@ void x264_load_deinterleave_chroma_fdec_mmx ( pixel *dst, pixel *src, intptr_t i
 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_sse2( void *dst, const void *src, size_t n );
-void x264_memzero_aligned_mmx ( void *dst, size_t n );
-void x264_memzero_aligned_sse2( void *dst, size_t n );
+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_integral_init4h_sse4( 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 );
@@ -228,9 +228,9 @@ static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, intptr_t, pixel *,
 
 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
 #if HIGH_BIT_DEPTH
-MC_COPY_WTAB(sse2,mmx,sse2,sse2)
+MC_COPY_WTAB(sse,mmx,sse,sse)
 #else
-MC_COPY_WTAB(sse2,mmx,mmx,sse2)
+MC_COPY_WTAB(sse,mmx,mmx,sse)
 #endif
 
 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
@@ -357,14 +357,14 @@ static void mc_luma_##name( pixel *dst,    intptr_t i_dst_stride,\
 }
 
 MC_LUMA(mmx2,mmx2,mmx)
-MC_LUMA(sse2,sse2,sse2)
+MC_LUMA(sse2,sse2,sse)
 #if !HIGH_BIT_DEPTH
 #if ARCH_X86
 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)
 #endif // !HIGH_BIT_DEPTH
 
 #define GET_REF(name)\
@@ -545,6 +545,12 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 
     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) )
@@ -569,8 +575,6 @@ 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;
@@ -591,7 +595,7 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
     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) )
@@ -640,8 +644,6 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
     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;
@@ -657,7 +659,7 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
         pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
     }
 
-    pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse2;
+    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;

diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
index d376b3df65b325f315a3fd81bd9b2aa804a8b001..6b3faf7a5f4db0bfa48bf3464a70af008cae6386 100644 (file)
--- a/common/x86/predict-a.asm
+++ b/common/x86/predict-a.asm
@@ -760,7 +760,7 @@ cglobal predict_8x8_v, 2,2
 %endmacro
 
 %if HIGH_BIT_DEPTH
-INIT_XMM sse2
+INIT_XMM sse
 PREDICT_8x8_V
 %else
 INIT_MMX mmx2
@@ -1630,7 +1630,7 @@ cglobal predict_8x8c_v, 1,1
 %endmacro
 
 %if HIGH_BIT_DEPTH
-INIT_XMM sse2
+INIT_XMM sse
 PREDICT_8x8C_V
 %else
 INIT_MMX mmx
@@ -1664,7 +1664,7 @@ cglobal predict_8x16c_v, 1,1
 %endmacro
 
 %if HIGH_BIT_DEPTH
-INIT_XMM sse2
+INIT_XMM sse
 PREDICT_8x16C_V
 %else
 INIT_MMX mmx
@@ -1997,8 +1997,8 @@ cglobal predict_16x16_v_mmx2, 1,2
     mova        m3, [r0 - FDEC_STRIDEB+24]
     STORE16x16  m0, m1, m2, m3
     RET
-INIT_XMM
-cglobal predict_16x16_v_sse2, 2,2
+INIT_XMM sse
+cglobal predict_16x16_v, 1,2
     mova      m0, [r0 - FDEC_STRIDEB+ 0]
     mova      m1, [r0 - FDEC_STRIDEB+16]
     STORE16x16_SSE2 m0, m1
@@ -2010,10 +2010,10 @@ cglobal predict_16x16_v_mmx2, 1,2
     movq        m1, [r0 - FDEC_STRIDE + 8]
     STORE16x16  m0, m1
     RET
-INIT_XMM
-cglobal predict_16x16_v_sse2, 1,1
-    movdqa      xmm0, [r0 - FDEC_STRIDE]
-    STORE16x16_SSE2 xmm0
+INIT_XMM sse
+cglobal predict_16x16_v, 1,1
+    mova        m0, [r0 - FDEC_STRIDE]
+    STORE16x16_SSE2 m0
     RET
 %endif
 

diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
index 50b7173f516257fc139b0496df5ad6e49492a21d..395b8237f50b541c581f267460d010da536e6be3 100644 (file)
--- a/common/x86/predict-c.c
+++ b/common/x86/predict-c.c
@@ -329,12 +329,14 @@ void x264_predict_16x16_init_mmx( int cpu, x264_predict_t pf[7] )
     pf[I_PRED_16x16_V]       = x264_predict_16x16_v_mmx2;
     pf[I_PRED_16x16_H]       = x264_predict_16x16_h_mmx2;
 #if HIGH_BIT_DEPTH
+    if( !(cpu&X264_CPU_SSE) )
+        return;
+    pf[I_PRED_16x16_V]       = x264_predict_16x16_v_sse;
     if( !(cpu&X264_CPU_SSE2) )
         return;
     pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_sse2;
     pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_sse2;
     pf[I_PRED_16x16_DC_LEFT] = x264_predict_16x16_dc_left_sse2;
-    pf[I_PRED_16x16_V]       = x264_predict_16x16_v_sse2;
     pf[I_PRED_16x16_H]       = x264_predict_16x16_h_sse2;
 #if HAVE_X86_INLINE_ASM
     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_sse2;
@@ -343,10 +345,12 @@ void x264_predict_16x16_init_mmx( int cpu, x264_predict_t pf[7] )
 #if !ARCH_X86_64
     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_mmx2;
 #endif
+    if( !(cpu&X264_CPU_SSE) )
+        return;
+    pf[I_PRED_16x16_V]       = x264_predict_16x16_v_sse;
     if( !(cpu&X264_CPU_SSE2) )
         return;
     pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_sse2;
-    pf[I_PRED_16x16_V]       = x264_predict_16x16_v_sse2;
     if( cpu&X264_CPU_SSE2_IS_SLOW )
         return;
     pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_sse2;
@@ -374,9 +378,11 @@ void x264_predict_8x8c_init_mmx( int cpu, x264_predict_t pf[7] )
         return;
     pf[I_PRED_CHROMA_DC]      = x264_predict_8x8c_dc_mmx2;
     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_mmx2;
+    if( !(cpu&X264_CPU_SSE) )
+        return;
+    pf[I_PRED_CHROMA_V]       = x264_predict_8x8c_v_sse;
     if( !(cpu&X264_CPU_SSE2) )
         return;
-    pf[I_PRED_CHROMA_V]       = x264_predict_8x8c_v_sse2;
     pf[I_PRED_CHROMA_DC]      = x264_predict_8x8c_dc_sse2;
     pf[I_PRED_CHROMA_DC_TOP]  = x264_predict_8x8c_dc_top_sse2;
     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_sse2;
@@ -423,9 +429,11 @@ void x264_predict_8x16c_init_mmx( int cpu, x264_predict_t pf[7] )
         return;
     pf[I_PRED_CHROMA_DC]      = x264_predict_8x16c_dc_mmx2;
     pf[I_PRED_CHROMA_H]       = x264_predict_8x16c_h_mmx2;
+    if( !(cpu&X264_CPU_SSE) )
+        return;
+    pf[I_PRED_CHROMA_V]       = x264_predict_8x16c_v_sse;
     if( !(cpu&X264_CPU_SSE2) )
         return;
-    pf[I_PRED_CHROMA_V]       = x264_predict_8x16c_v_sse2;
     pf[I_PRED_CHROMA_DC_TOP]  = x264_predict_8x16c_dc_top_sse2;
     pf[I_PRED_CHROMA_DC]      = x264_predict_8x16c_dc_sse2;
     pf[I_PRED_CHROMA_H]       = x264_predict_8x16c_h_sse2;
@@ -460,9 +468,11 @@ void x264_predict_8x8_init_mmx( int cpu, x264_predict8x8_t pf[12], x264_predict_
     if( !(cpu&X264_CPU_MMX2) )
         return;
 #if HIGH_BIT_DEPTH
+    if( !(cpu&X264_CPU_SSE) )
+        return;
+    pf[I_PRED_8x8_V]      = x264_predict_8x8_v_sse;
     if( !(cpu&X264_CPU_SSE2) )
         return;
-    pf[I_PRED_8x8_V]      = x264_predict_8x8_v_sse2;
     pf[I_PRED_8x8_H]      = x264_predict_8x8_h_sse2;
     pf[I_PRED_8x8_DC]     = x264_predict_8x8_dc_sse2;
     pf[I_PRED_8x8_DC_TOP] = x264_predict_8x8_dc_top_sse2;

diff --git a/common/x86/predict.h b/common/x86/predict.h
index dac352edf6c97625e72473709db0a32bc4a56247..ca2c20fad09b4ffe555afac93549458b7acb5597 100644 (file)
--- a/common/x86/predict.h
+++ b/common/x86/predict.h
@@ -34,7 +34,7 @@ void x264_predict_4x4_init_mmx   ( int cpu, x264_predict_t pf[12] );
 void x264_predict_8x8_init_mmx   ( int cpu, x264_predict8x8_t pf[12], x264_predict_8x8_filter_t *predict_8x8_filter );
 
 void x264_predict_16x16_v_mmx2( pixel *src );
-void x264_predict_16x16_v_sse2( pixel *src );
+void x264_predict_16x16_v_sse ( pixel *src );
 void x264_predict_16x16_h_mmx2( pixel *src );
 void x264_predict_16x16_h_sse2( uint16_t *src );
 void x264_predict_16x16_h_ssse3( uint8_t *src );
@@ -55,7 +55,7 @@ void x264_predict_8x16c_dc_sse2( uint16_t *src );
 void x264_predict_8x16c_dc_top_mmx2( uint8_t *src );
 void x264_predict_8x16c_dc_top_sse2( uint16_t *src );
 void x264_predict_8x16c_v_mmx( uint8_t *src );
-void x264_predict_8x16c_v_sse2( uint16_t *src );
+void x264_predict_8x16c_v_sse( uint16_t *src );
 void x264_predict_8x16c_h_mmx2( pixel *src );
 void x264_predict_8x16c_h_sse2( pixel *src );
 void x264_predict_8x16c_h_ssse3( uint8_t *src );
@@ -70,12 +70,12 @@ void x264_predict_8x8c_dc_sse2( uint16_t *src );
 void x264_predict_8x8c_dc_top_mmx2( uint8_t *src );
 void x264_predict_8x8c_dc_top_sse2( uint16_t *src );
 void x264_predict_8x8c_v_mmx( pixel *src );
-void x264_predict_8x8c_v_sse2( uint16_t *src );
+void x264_predict_8x8c_v_sse( uint16_t *src );
 void x264_predict_8x8c_h_mmx2( pixel *src );
 void x264_predict_8x8c_h_sse2( pixel *src );
 void x264_predict_8x8c_h_ssse3( uint8_t *src );
 void x264_predict_8x8_v_mmx2( uint8_t *src, uint8_t edge[36] );
-void x264_predict_8x8_v_sse2( uint16_t *src, uint16_t edge[36] );
+void x264_predict_8x8_v_sse ( uint16_t *src, uint16_t edge[36] );
 void x264_predict_8x8_h_mmx2( uint8_t *src, uint8_t edge[36] );
 void x264_predict_8x8_h_sse2( uint16_t *src, uint16_t edge[36] );
 void x264_predict_8x8_hd_mmx2( uint8_t *src, uint8_t edge[36] );

diff --git a/common/x86/x86inc.asm b/common/x86/x86inc.asm
index 4fe33b3b17022aa9c9dc3fbf8d51a81d20360b4b..2e68578472187f5fcf2105d997fdf4135bfb5bd3 100644 (file)
--- a/common/x86/x86inc.asm
+++ b/common/x86/x86inc.asm
@@ -429,7 +429,7 @@ DECLARE_REG 14, R15, 120
     %assign %%i xmm_regs_used
     %rep (xmm_regs_used-6)
         %assign %%i %%i-1
-        movdqa [rsp + (%%i-6)*16 + stack_size + (~stack_offset&8)], xmm %+ %%i
+        movaps [rsp + (%%i-6)*16 + stack_size + (~stack_offset&8)], xmm %+ %%i
     %endrep
 %endmacro
 
@@ -447,7 +447,7 @@ DECLARE_REG 14, R15, 120
         %assign %%i xmm_regs_used
         %rep (xmm_regs_used-6)
             %assign %%i %%i-1
-            movdqa xmm %+ %%i, [%1 + (%%i-6)*16+stack_size+(~stack_offset&8)]
+            movaps xmm %+ %%i, [%1 + (%%i-6)*16+stack_size+(~stack_offset&8)]
         %endrep
         %if stack_size_padded == 0
             add %1, (xmm_regs_used-6)*16+16

diff --git a/tools/checkasm.c b/tools/checkasm.c
index 2086641234710309d2e364086b0545ffae851ed2..c1e8188ee38fe05ddcac05d5cb77841708b0c527 100644 (file)
--- a/tools/checkasm.c
+++ b/tools/checkasm.c
@@ -176,6 +176,7 @@ static void print_bench(void)
                     /* print sse2slow only if there's also a sse2fast version of the same func */
                     b->cpu&X264_CPU_SSE2_IS_SLOW && j<MAX_CPUS-1 && b[1].cpu&X264_CPU_SSE2_IS_FAST && !(b[1].cpu&X264_CPU_SSE3) ? "sse2slow" :
                     b->cpu&X264_CPU_SSE2 ? "sse2" :
+                    b->cpu&X264_CPU_SSE ? "sse" :
                     b->cpu&X264_CPU_MMX ? "mmx" :
                     b->cpu&X264_CPU_ALTIVEC ? "altivec" :
                     b->cpu&X264_CPU_NEON ? "neon" :
@@ -2394,9 +2395,11 @@ static int check_all_flags( void )
         ret |= add_flags( &cpu0, &cpu1, X264_CPU_SLOW_CTZ, "MMX SlowCTZ" );
         cpu1 &= ~X264_CPU_SLOW_CTZ;
     }
+    if( x264_cpu_detect() & X264_CPU_SSE )
+        ret |= add_flags( &cpu0, &cpu1, X264_CPU_SSE, "SSE" );
     if( x264_cpu_detect() & X264_CPU_SSE2 )
     {
-        ret |= add_flags( &cpu0, &cpu1, X264_CPU_SSE | X264_CPU_SSE2 | X264_CPU_SSE2_IS_SLOW, "SSE2Slow" );
+        ret |= add_flags( &cpu0, &cpu1, X264_CPU_SSE2 | X264_CPU_SSE2_IS_SLOW, "SSE2Slow" );
         ret |= add_flags( &cpu0, &cpu1, X264_CPU_SSE2_IS_FAST, "SSE2Fast" );
         ret |= add_flags( &cpu0, &cpu1, X264_CPU_CACHELINE_64, "SSE2Fast Cache64" );
         cpu1 &= ~X264_CPU_CACHELINE_64;