;*****************************************************************************
;* mc-a.asm: x86 motion compensation
;*****************************************************************************
-;* Copyright (C) 2003-2011 x264 project
+;* Copyright (C) 2003-2012 x264 project
;*
;* Authors: Loren Merritt <lorenm@u.washington.edu>
;* Fiona Glaser <fiona@x264.com>
; implicit weighted biprediction
;=============================================================================
; assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64
-%ifdef WIN64
+%if WIN64
DECLARE_REG_TMP 0,1,2,3,4,5,4,5
%macro AVG_START 0-1 0
- PROLOGUE 5,7,%1
- movsxd r5, dword r5m
+ PROLOGUE 6,7,%1
%endmacro
-%elifdef UNIX64
+%elif UNIX64
DECLARE_REG_TMP 0,1,2,3,4,5,7,8
%macro AVG_START 0-1 0
PROLOGUE 6,9,%1
REP_RET
%endmacro
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
%macro BIWEIGHT_MMX 2
movh m0, %1
SPLATW m3, m3 ; weight_dst,src
%endmacro
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
%macro BIWEIGHT_ROW 4
BIWEIGHT [%2], [%3]
%if %4==mmsize/4
%endif ;HIGH_BIT_DEPTH
;-----------------------------------------------------------------------------
-; int pixel_avg_weight_w16( pixel *dst, int, pixel *src1, int, pixel *src2, int, int i_weight )
+; int pixel_avg_weight_w16( pixel *dst, intptr_t, pixel *src1, intptr_t, pixel *src2, intptr_t, int i_weight )
;-----------------------------------------------------------------------------
%macro AVG_WEIGHT 1-2 0
cglobal pixel_avg_weight_w%1
BIWEIGHT_START
AVG_START %2
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
mova m7, [pw_pixel_max]
%endif
.height_loop:
BIWEIGHT [t2], [t4]
SWAP 0, 6
BIWEIGHT [t2+SIZEOF_PIXEL*t3], [t4+SIZEOF_PIXEL*t5]
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
packssdw m6, m0
CLIPW m6, m5, m7
%else ;!HIGH_BIT_DEPTH
AVG_WEIGHT 4
AVG_WEIGHT 8
AVG_WEIGHT 16
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
INIT_XMM sse2
AVG_WEIGHT 4, 8
AVG_WEIGHT 8, 8
; P frame explicit weighted prediction
;=============================================================================
-%ifdef HIGH_BIT_DEPTH
-%macro WEIGHT_START 1 ; (width)
+%if HIGH_BIT_DEPTH
+; width
+%macro WEIGHT_START 1
mova m0, [r4+ 0] ; 1<<denom
mova m3, [r4+16]
movd m2, [r4+32] ; denom
paddw m2, [sq_1] ; denom+1
%endmacro
-%macro WEIGHT 2 ; (src1, src2)
+; src1, src2
+%macro WEIGHT 2
movh m5, [%1]
movh m6, [%2]
punpcklwd m5, m0
packssdw m5, m6
%endmacro
-%macro WEIGHT_TWO_ROW 3 ; (src, dst, width)
+; src, dst, width
+%macro WEIGHT_TWO_ROW 3
%assign x 0
%rep (%3+mmsize/2-1)/(mmsize/2)
%if %3-x/2 <= 4 && mmsize == 16
%macro WEIGHT_START 1
mova m3, [r4]
- mova m6, [r4+16]
+ mova m4, [r4+16]
+%if notcpuflag(ssse3)
movd m5, [r4+32]
- pxor m2, m2
-%if (%1 == 20 || %1 == 12) && mmsize == 16
- movdq2q mm3, xmm3
- movdq2q mm4, xmm4
- movdq2q mm5, xmm5
- movdq2q mm6, xmm6
- pxor mm2, mm2
%endif
-%endmacro
-
-%macro WEIGHT_START_SSSE3 1
- mova m3, [r4]
- mova m4, [r4+16]
pxor m2, m2
-%if %1 == 20 || %1 == 12
- movdq2q mm3, xmm3
- movdq2q mm4, xmm4
- pxor mm2, mm2
-%endif
%endmacro
-;; macro to weight mmsize bytes taking half from %1 and half from %2
-%macro WEIGHT 2 ; (src1,src2)
- movh m0, [%1]
- movh m1, [%2]
- punpcklbw m0, m2 ;setup
- punpcklbw m1, m2 ;setup
- pmullw m0, m3 ;scale
- pmullw m1, m3 ;scale
- paddsw m0, m6 ;1<<(denom-1)+(offset<<denom)
- paddsw m1, m6 ;1<<(denom-1)+(offset<<denom)
- psraw m0, m5 ;denom
- psraw m1, m5 ;denom
+; src1, src2, dst1, dst2
+%macro WEIGHT_ROWx2 4
+ movh m0, [%1 ]
+ movh m1, [%1+mmsize/2]
+ movh m6, [%2 ]
+ movh m7, [%2+mmsize/2]
+ punpcklbw m0, m2
+ punpcklbw m1, m2
+ punpcklbw m6, m2
+ punpcklbw m7, m2
+%if cpuflag(ssse3)
+ psllw m0, 7
+ psllw m1, 7
+ psllw m6, 7
+ psllw m7, 7
+ pmulhrsw m0, m3
+ pmulhrsw m1, m3
+ pmulhrsw m6, m3
+ pmulhrsw m7, m3
+ paddw m0, m4
+ paddw m1, m4
+ paddw m6, m4
+ paddw m7, m4
+%else
+ pmullw m0, m3
+ pmullw m1, m3
+ pmullw m6, m3
+ pmullw m7, m3
+ paddsw m0, m4 ;1<<(denom-1)+(offset<<denom)
+ paddsw m1, m4
+ paddsw m6, m4
+ paddsw m7, m4
+ psraw m0, m5
+ psraw m1, m5
+ psraw m6, m5
+ psraw m7, m5
+%endif
+ packuswb m0, m1
+ packuswb m6, m7
+ mova [%3], m0
+ mova [%4], m6
%endmacro
-%macro WEIGHT_SSSE3 2
+; src1, src2, dst1, dst2, width
+%macro WEIGHT_COL 5
movh m0, [%1]
movh m1, [%2]
punpcklbw m0, m2
punpcklbw m1, m2
+%if cpuflag(ssse3)
psllw m0, 7
psllw m1, 7
pmulhrsw m0, m3
pmulhrsw m1, m3
paddw m0, m4
paddw m1, m4
-%endmacro
-
-%macro WEIGHT_SAVE_ROW 3 ;(src,dst,width)
-%if %3 == 16
- mova [%2], %1
-%elif %3 == 8
- movq [%2], %1
%else
- movd [%2], %1 ; width 2 can write garbage for last 2 bytes
-%endif
-%endmacro
-
-%macro WEIGHT_ROW 3 ; (src,dst,width)
- ;; load weights
- WEIGHT %1, (%1+(mmsize/2))
- packuswb m0, m1 ;put bytes into m0
- WEIGHT_SAVE_ROW m0, %2, %3
-%endmacro
-
-%macro WEIGHT_SAVE_COL 2 ;(dst,size)
-%if %2 == 8
- packuswb m0, m1
- movq [%1], m0
- movhps [%1+r1], m0
-%else
- packuswb m0, m0
- packuswb m1, m1
- movd [%1], m0 ; width 2 can write garbage for last 2 bytes
- movd [%1+r1], m1
-%endif
-%endmacro
-
-%macro WEIGHT_COL 3 ; (src,dst,width)
-%if %3 <= 4 && mmsize == 16
- INIT_MMX
- ;; load weights
- WEIGHT %1, (%1+r3)
- WEIGHT_SAVE_COL %2, %3
- INIT_XMM
+ pmullw m0, m3
+ pmullw m1, m3
+ paddsw m0, m4 ;1<<(denom-1)+(offset<<denom)
+ paddsw m1, m4
+ psraw m0, m5
+ psraw m1, m5
+%endif
+%if %5 == 8
+ packuswb m0, m1
+ movh [%3], m0
+ movhps [%4], m0
%else
- WEIGHT %1, (%1+r3)
- WEIGHT_SAVE_COL %2, %3
+ packuswb m0, m0
+ packuswb m1, m1
+ movd [%3], m0 ; width 2 can write garbage for the last 2 bytes
+ movd [%4], m1
%endif
-
%endmacro
-%macro WEIGHT_TWO_ROW 3 ; (src,dst,width)
+; src, dst, width
+%macro WEIGHT_TWO_ROW 3
%assign x 0
%rep %3
%if (%3-x) >= mmsize
- WEIGHT_ROW (%1+x), (%2+x), mmsize ; weight 1 mmsize
- WEIGHT_ROW (%1+r3+x), (%2+r1+x), mmsize ; weight 1 mmsize
+ WEIGHT_ROWx2 %1+x, %1+r3+x, %2+x, %2+r1+x
%assign x (x+mmsize)
%else
- WEIGHT_COL (%1+x),(%2+x),(%3-x)
+ WEIGHT_COL %1+x, %1+r3+x, %2+x, %2+r1+x, %3-x
%exitrep
%endif
%if x >= %3
%endif ; HIGH_BIT_DEPTH
;-----------------------------------------------------------------------------
-;void mc_weight_wX( pixel *dst, int i_dst_stride, pixel *src, int i_src_stride, weight_t *weight, int h )
+;void mc_weight_wX( pixel *dst, intptr_t i_dst_stride, pixel *src, intptr_t i_src_stride, weight_t *weight, int h )
;-----------------------------------------------------------------------------
-%ifdef ARCH_X86_64
-%define NUMREGS 6
-%define LOAD_HEIGHT
-%define HEIGHT_REG r5d
-%define TMP_REG r6d
-%else
-%define NUMREGS 5
-%define TMP_REG r5d
-%define LOAD_HEIGHT mov r4d, r5m
-%define HEIGHT_REG r4d
-%endif
-
-%assign XMMREGS 7
-%ifdef HIGH_BIT_DEPTH
-%assign NUMREGS NUMREGS+1
-%assign XMMREGS 8
-%endif
-
%macro WEIGHTER 1
- cglobal mc_weight_w%1, NUMREGS, NUMREGS, XMMREGS
+cglobal mc_weight_w%1, 6,6,8
FIX_STRIDES r1, r3
WEIGHT_START %1
- LOAD_HEIGHT
.loop:
WEIGHT_TWO_ROW r2, r0, %1
lea r0, [r0+r1*2]
lea r2, [r2+r3*2]
- sub HEIGHT_REG, 2
+ sub r5d, 2
jg .loop
REP_RET
%endmacro
WEIGHTER 8
WEIGHTER 16
WEIGHTER 20
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
WEIGHTER 12
-INIT_XMM avx
-WEIGHTER 8
-WEIGHTER 12
-WEIGHTER 16
-WEIGHTER 20
%else
-%define WEIGHT WEIGHT_SSSE3
-%define WEIGHT_START WEIGHT_START_SSSE3
INIT_MMX ssse3
WEIGHTER 4
INIT_XMM ssse3
WEIGHTER 8
WEIGHTER 16
WEIGHTER 20
-INIT_XMM avx
-WEIGHTER 8
-WEIGHTER 16
-WEIGHTER 20
%endif
%macro OFFSET_OP 7
mov%6 m0, [%1]
mov%6 m1, [%2]
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
p%5usw m0, m2
p%5usw m1, m2
%ifidn %5,add
OFFSET_OP (%1+x), (%1+x+r3), (%2+x), (%2+x+r1), %4, u, a
%assign x (x+mmsize)
%else
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
OFFSET_OP (%1+x), (%1+x+r3), (%2+x), (%2+x+r1), %4, h, h
%else
OFFSET_OP (%1+x), (%1+x+r3), (%2+x), (%2+x+r1), %4, d, d
%endmacro
;-----------------------------------------------------------------------------
-;void mc_offset_wX( pixel *src, int i_src_stride, pixel *dst, int i_dst_stride, weight_t *w, int h )
+;void mc_offset_wX( pixel *src, intptr_t i_src_stride, pixel *dst, intptr_t i_dst_stride, weight_t *w, int h )
;-----------------------------------------------------------------------------
%macro OFFSET 2
- cglobal mc_offset%2_w%1, NUMREGS, NUMREGS
+cglobal mc_offset%2_w%1, 6,6
FIX_STRIDES r1, r3
mova m2, [r4]
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
%ifidn %2,add
mova m3, [pw_pixel_max]
%endif
%endif
- LOAD_HEIGHT
.loop:
OFFSET_TWO_ROW r2, r0, %1, %2
lea r0, [r0+r1*2]
lea r2, [r2+r3*2]
- sub HEIGHT_REG, 2
+ sub r5d, 2
jg .loop
REP_RET
%endmacro
OFFSETPN 12
OFFSETPN 16
OFFSETPN 20
-INIT_XMM avx
-OFFSETPN 12
-OFFSETPN 16
-OFFSETPN 20
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
INIT_XMM sse2
OFFSETPN 8
-INIT_XMM avx
-OFFSETPN 8
%endif
-%undef LOAD_HEIGHT
-%undef HEIGHT_REG
-%undef NUMREGS
-
;=============================================================================
;=============================================================================
;-----------------------------------------------------------------------------
-; void pixel_avg_4x4( pixel *dst, int dst_stride,
-; pixel *src1, int src1_stride, pixel *src2, int src2_stride, int weight );
+; void pixel_avg_4x4( pixel *dst, intptr_t dst_stride, pixel *src1, intptr_t src1_stride,
+; pixel *src2, intptr_t src2_stride, int weight );
;-----------------------------------------------------------------------------
%macro AVGH 2
cglobal pixel_avg_%1x%2
%endmacro
;-----------------------------------------------------------------------------
-; void pixel_avg_w4( pixel *dst, int dst_stride,
-; pixel *src1, int src1_stride, pixel *src2, int src2_stride,
-; int height, int weight );
+; void pixel_avg_w4( pixel *dst, intptr_t dst_stride, pixel *src1, intptr_t src1_stride,
+; pixel *src2, intptr_t src2_stride, int height, int weight );
;-----------------------------------------------------------------------------
%macro AVG_FUNC 3
%rep (%1*SIZEOF_PIXEL+mmsize-1)/mmsize
%2 m0, [t2+x]
%2 m1, [t2+x+SIZEOF_PIXEL*t3]
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
pavgw m0, [t4+x]
pavgw m1, [t4+x+SIZEOF_PIXEL*t5]
%else ;!HIGH_BIT_DEPTH
AVG_END
%endmacro
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
INIT_MMX mmx2
AVG_FUNC 4, movq, movq
; pixel avg2
;=============================================================================
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
;-----------------------------------------------------------------------------
-; void pixel_avg2_wN( uint16_t *dst, int dst_stride,
-; uint16_t *src1, int src_stride,
+; void pixel_avg2_wN( uint16_t *dst, intptr_t dst_stride,
+; uint16_t *src1, intptr_t src_stride,
; uint16_t *src2, int height );
;-----------------------------------------------------------------------------
%macro AVG2_W_ONE 1
REP_RET
%endif ; HIGH_BIT_DEPTH
-%ifndef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH == 0
;-----------------------------------------------------------------------------
-; void pixel_avg2_w4( uint8_t *dst, int dst_stride,
-; uint8_t *src1, int src_stride,
+; void pixel_avg2_w4( uint8_t *dst, intptr_t dst_stride,
+; uint8_t *src1, intptr_t src_stride,
; uint8_t *src2, int height );
;-----------------------------------------------------------------------------
%macro AVG2_W8 2
%endif
%if 0 ; or %1==8 - but the extra branch seems too expensive
ja cachesplit
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
test r4b, 1
%else
test byte r4m, 1
INIT_MMX
AVG_CACHELINE_CHECK 8, 64, mmx2
AVG_CACHELINE_CHECK 12, 64, mmx2
-%ifndef ARCH_X86_64
+%if ARCH_X86_64 == 0
AVG_CACHELINE_CHECK 16, 64, mmx2
AVG_CACHELINE_CHECK 20, 64, mmx2
AVG_CACHELINE_CHECK 8, 32, mmx2
%else
lea r6, [avg_w16_addr + r6]
%endif
-%ifdef UNIX64
- jmp r6
-%else
- call r6
- RET
-%endif
+ TAIL_CALL r6, 1
%assign j 0
%assign k 1
%endmacro
;-----------------------------------------------------------------------------
-; void mc_copy_w4( uint8_t *dst, int i_dst_stride,
-; uint8_t *src, int i_src_stride, int i_height )
+; void mc_copy_w4( uint8_t *dst, intptr_t i_dst_stride,
+; uint8_t *src, intptr_t i_src_stride, int i_height )
;-----------------------------------------------------------------------------
INIT_MMX
cglobal mc_copy_w4_mmx, 4,6
lea r5, [r3*3]
lea r4, [r1*3]
je .end
-%ifndef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH == 0
%define mova movd
%define movu movd
%endif
; FIXME doesn't cover all pixels in high depth and/or 4:4:4
;-----------------------------------------------------------------------------
-; void prefetch_fenc( pixel *pix_y, int stride_y,
-; pixel *pix_uv, int stride_uv, int mb_x )
+; void prefetch_fenc( pixel *pix_y, intptr_t stride_y,
+; pixel *pix_uv, intptr_t stride_uv, int mb_x )
;-----------------------------------------------------------------------------
%macro PREFETCH_FENC 1
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
cglobal prefetch_fenc_%1, 5,5
- FIX_STRIDES r1d, r3d
+ FIX_STRIDES r1, r3
and r4d, 3
mov eax, r4d
imul r4d, r1d
PREFETCH_FENC 422
;-----------------------------------------------------------------------------
-; void prefetch_ref( pixel *pix, int stride, int parity )
+; void prefetch_ref( pixel *pix, intptr_t stride, int parity )
;-----------------------------------------------------------------------------
INIT_MMX mmx2
cglobal prefetch_ref, 3,3
- FIX_STRIDES r1d
+ FIX_STRIDES r1
dec r2d
and r2d, r1d
lea r0, [r0+r2*8+64*SIZEOF_PIXEL]
; chroma MC
;=============================================================================
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
DECLARE_REG_TMP 6,7,8
%else
DECLARE_REG_TMP 0,1,2
%endif
%macro MC_CHROMA_START 1
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
PROLOGUE 0,9,%1
%else
PROLOGUE 0,6,%1
add r3, t0 ; src += (dx>>3) + (dy>>3) * src_stride
%endmacro
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
%macro UNPACK_UNALIGNED 4
movu %1, [%4+0]
movu %2, [%4+4]
%endif ; HIGH_BIT_DEPTH
;-----------------------------------------------------------------------------
-; void mc_chroma( uint8_t *dstu, uint8_t *dstv, int dst_stride,
-; uint8_t *src, int src_stride,
+; void mc_chroma( uint8_t *dstu, uint8_t *dstv, intptr_t dst_stride,
+; uint8_t *src, intptr_t src_stride,
; int dx, int dy,
; int width, int height )
;-----------------------------------------------------------------------------
MC_CHROMA_START 0
FIX_STRIDES r4
and r5d, 7
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
jz .mc1dy
%endif
and t2d, 7
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
jz .mc1dx
%endif
shl r5d, 16
pshufw m5, m5, q1111
jge .width4
%else
-%ifdef WIN64
+%if WIN64
cmp dword r7m, 4 ; flags were clobbered by WIN64_SPILL_XMM
%endif
pshufd m7, m5, q1111
pshufd m5, m5, q1111
jg .width8
%endif
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
add r2, r2
UNPACK_UNALIGNED m0, m1, m2, r3
%else
SWAP 3, 0
ALIGN 4
.loop2:
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
UNPACK_UNALIGNED m0, m1, m2, r3+r4
pmullw m3, m6
%else ; !HIGH_BIT_DEPTH
pmullw m0, m5
paddw m0, m2
psrlw m0, 6
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
movh [r0], m0
%if mmsize == 8
psrlq m0, 32
%if mmsize==8
.width4:
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
mov t0, r0
mov t1, r1
mov t2, r3
%endif
%else
.width8:
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
%define multy0 m8
SWAP 8, 5
%else
%endif
FIX_STRIDES r2
.loopx:
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
UNPACK_UNALIGNED m0, m2, m4, r3
UNPACK_UNALIGNED m1, m3, m5, r3+mmsize
%else
add r3, r4
ALIGN 4
.loop4:
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
UNPACK_UNALIGNED m0, m1, m2, r3
pmaddwd m0, m7
pmaddwd m1, m7
paddw m1, m3
psrlw m0, 6
psrlw m1, 6
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
movh [r0], m0
movh [r0+mmsize/2], m1
%if mmsize==8
jg .width8
REP_RET
.width8:
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
lea r3, [t2+8*SIZEOF_PIXEL]
lea r0, [t0+4*SIZEOF_PIXEL]
lea r1, [t1+4*SIZEOF_PIXEL]
jmp .loopx
%endif
-%ifdef ARCH_X86_64 ; too many regs for x86_32
+%if ARCH_X86_64 ; too many regs for x86_32
RESET_MM_PERMUTATION
-%ifdef WIN64
+%if WIN64
%if xmm_regs_used > 6
%assign stack_offset stack_offset-(xmm_regs_used-6)*16-16
%assign xmm_regs_used 6
movd m5, r5d
mov r6d, 2*SIZEOF_PIXEL
.mc1d:
-%ifdef HIGH_BIT_DEPTH
-%if mmsize == 16
+%if HIGH_BIT_DEPTH && mmsize == 16
WIN64_SPILL_XMM 8
-%endif
%endif
mova m4, [pw_8]
SPLATW m5, m5
shr r5d, 1
%endif
.loop1d_w4:
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
%if mmsize == 8
movq m0, [r3+0]
movq m2, [r3+8]
paddw m2, m3
psrlw m0, 3
psrlw m2, 3
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
%if mmsize == 8
xchg r4, r8
xchg r2, r7
pshufb m0, m5
movu m1, [r3+8]
pshufb m1, m5
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
SWAP 8, 6
%define mult1 m8
%else
REP_RET
%endmacro
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
INIT_MMX mmx2
MC_CHROMA
INIT_XMM sse2
projects / x264 / blobdiff