;******************************************************************************
;* H.264 intra prediction asm optimizations
;* Copyright (c) 2010 Jason Garrett-Glaser
+;* Copyright (c) 2010 Holger Lubitz
+;* Copyright (c) 2010 Loren Merritt
+;* Copyright (c) 2010 Ronald S. Bultje
;*
-;* This file is part of FFmpeg.
+;* This file is part of Libav.
;*
-;* FFmpeg is free software; you can redistribute it and/or
+;* Libav is free software; you can redistribute it and/or
;* modify it under the terms of the GNU Lesser General Public
;* License as published by the Free Software Foundation; either
;* version 2.1 of the License, or (at your option) any later version.
;*
-;* FFmpeg is distributed in the hope that it will be useful,
+;* Libav is distributed in the hope that it will be useful,
;* but WITHOUT ANY WARRANTY; without even the implied warranty of
;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
;* Lesser General Public License for more details.
;*
;* You should have received a copy of the GNU Lesser General Public
-;* License along with FFmpeg; if not, write to the Free Software
-;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;* License along with Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
-%include "x86inc.asm"
+%include "libavutil/x86/x86util.asm"
SECTION_RODATA
tm_shuf: times 8 db 0x03, 0x80
+pw_ff00: times 8 dw 0xff00
plane_shuf: db -8, -7, -6, -5, -4, -3, -2, -1
db 1, 2, 3, 4, 5, 6, 7, 8
plane8_shuf: db -4, -3, -2, -1, 0, 0, 0, 0
cextern pb_1
cextern pb_3
+cextern pw_4
cextern pw_5
+cextern pw_8
cextern pw_16
cextern pw_17
cextern pw_32
;-----------------------------------------------------------------------------
-; void pred16x16_vertical(uint8_t *src, int stride)
+; void pred16x16_vertical_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-cglobal pred16x16_vertical_mmx, 2,3
+INIT_MMX mmx
+cglobal pred16x16_vertical_8, 2,3
sub r0, r1
mov r2, 8
movq mm0, [r0+0]
jg .loop
REP_RET
-cglobal pred16x16_vertical_sse, 2,3
+INIT_XMM sse
+cglobal pred16x16_vertical_8, 2,3
sub r0, r1
mov r2, 4
movaps xmm0, [r0]
REP_RET
;-----------------------------------------------------------------------------
-; void pred16x16_horizontal(uint8_t *src, int stride)
+; void pred16x16_horizontal_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-%macro PRED16x16_H 1
-cglobal pred16x16_horizontal_%1, 2,3
+%macro PRED16x16_H 0
+cglobal pred16x16_horizontal_8, 2,3
mov r2, 8
-%ifidn %1, ssse3
+%if cpuflag(ssse3)
mova m2, [pb_3]
%endif
.loop:
movd m0, [r0+r1*0-4]
movd m1, [r0+r1*1-4]
-%ifidn %1, ssse3
+%if cpuflag(ssse3)
pshufb m0, m2
pshufb m1, m2
%else
punpcklbw m0, m0
punpcklbw m1, m1
-%ifidn %1, mmxext
- pshufw m0, m0, 0xff
- pshufw m1, m1, 0xff
-%else
- punpckhwd m0, m0
- punpckhwd m1, m1
- punpckhdq m0, m0
- punpckhdq m1, m1
-%endif
+ SPLATW m0, m0, 3
+ SPLATW m1, m1, 3
mova [r0+r1*0+8], m0
mova [r0+r1*1+8], m1
%endif
REP_RET
%endmacro
-INIT_MMX
-PRED16x16_H mmx
-PRED16x16_H mmxext
-INIT_XMM
-PRED16x16_H ssse3
+INIT_MMX mmx
+PRED16x16_H
+INIT_MMX mmxext
+PRED16x16_H
+INIT_XMM ssse3
+PRED16x16_H
;-----------------------------------------------------------------------------
-; void pred16x16_dc(uint8_t *src, int stride)
+; void pred16x16_dc_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-%macro PRED16x16_DC 1
-cglobal pred16x16_dc_%1, 2,7
+%macro PRED16x16_DC 0
+cglobal pred16x16_dc_8, 2,7
mov r4, r0
sub r0, r1
pxor mm0, mm0
add r5d, r6d
lea r2d, [r2+r5+16]
shr r2d, 5
-%ifidn %1, mmxext
- movd m0, r2d
- punpcklbw m0, m0
- pshufw m0, m0, 0
-%elifidn %1, sse2
- movd m0, r2d
- punpcklbw m0, m0
- pshuflw m0, m0, 0
- punpcklqdq m0, m0
-%elifidn %1, ssse3
+%if cpuflag(ssse3)
pxor m1, m1
- movd m0, r2d
- pshufb m0, m1
%endif
+ SPLATB_REG m0, r2, m1
%if mmsize==8
mov r3d, 8
REP_RET
%endmacro
-INIT_MMX
-PRED16x16_DC mmxext
-INIT_XMM
-PRED16x16_DC sse2
-PRED16x16_DC ssse3
+INIT_MMX mmxext
+PRED16x16_DC
+INIT_XMM sse2
+PRED16x16_DC
+INIT_XMM ssse3
+PRED16x16_DC
;-----------------------------------------------------------------------------
-; void pred16x16_tm_vp8(uint8_t *src, int stride)
+; void pred16x16_tm_vp8_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-%macro PRED16x16_TM_MMX 1
-cglobal pred16x16_tm_vp8_%1, 2,5
+%macro PRED16x16_TM 0
+cglobal pred16x16_tm_vp8_8, 2,5
sub r0, r1
pxor mm7, mm7
movq mm0, [r0+0]
movzx r2d, byte [r0+r1-1]
sub r2d, r3d
movd mm4, r2d
-%ifidn %1, mmx
- punpcklwd mm4, mm4
- punpckldq mm4, mm4
-%else
- pshufw mm4, mm4, 0
-%endif
+ SPLATW mm4, mm4, 0
movq mm5, mm4
movq mm6, mm4
movq mm7, mm4
REP_RET
%endmacro
-PRED16x16_TM_MMX mmx
-PRED16x16_TM_MMX mmxext
+INIT_MMX mmx
+PRED16x16_TM
+INIT_MMX mmxext
+PRED16x16_TM
-cglobal pred16x16_tm_vp8_sse2, 2,6,6
+INIT_XMM sse2
+cglobal pred16x16_tm_vp8_8, 2,6,6
sub r0, r1
pxor xmm2, xmm2
movdqa xmm0, [r0]
REP_RET
;-----------------------------------------------------------------------------
-; void pred16x16_plane(uint8_t *src, int stride)
+; void pred16x16_plane_*_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-%macro H264_PRED16x16_PLANE 3
-cglobal pred16x16_plane_%3_%1, 2, 7, %2
+%macro H264_PRED16x16_PLANE 1
+cglobal pred16x16_plane_%1_8, 2,9,7
mov r2, r1 ; +stride
neg r1 ; -stride
paddw m0, m2
paddw m1, m3
%else ; mmsize == 16
-%ifidn %1, sse2
+%if cpuflag(ssse3)
+ movhps m0, [r0+r1 +8]
+ pmaddubsw m0, [plane_shuf] ; H coefficients
+%else ; sse2
pxor m2, m2
movh m1, [r0+r1 +8]
punpcklbw m0, m2
pmullw m0, [pw_m8tom1]
pmullw m1, [pw_1to8]
paddw m0, m1
-%else ; ssse3
- movhps m0, [r0+r1 +8]
- pmaddubsw m0, [plane_shuf] ; H coefficients
%endif
movhlps m1, m0
%endif
paddw m0, m1
-%ifidn %1, mmx
+%if cpuflag(mmxext)
+ PSHUFLW m1, m0, 0xE
+%elif cpuflag(mmx)
mova m1, m0
psrlq m1, 32
-%elifidn %1, mmx2
- pshufw m1, m0, 0xE
-%else ; mmsize == 16
- pshuflw m1, m0, 0xE
%endif
paddw m0, m1
-%ifidn %1, mmx
+%if cpuflag(mmxext)
+ PSHUFLW m1, m0, 0x1
+%elif cpuflag(mmx)
mova m1, m0
psrlq m1, 16
-%elifidn %1, mmx2
- pshufw m1, m0, 0x1
-%else
- pshuflw m1, m0, 0x1
%endif
paddw m0, m1 ; sum of H coefficients
-%ifidn %3, h264
- pmullw m0, [pw_5]
- paddw m0, [pw_32]
- psraw m0, 6
-%elifidn %3, rv40
- pmullw m0, [pw_5]
- psraw m0, 6
-%elifidn %3, svq3
- movd r3d, m0
- movsx r3, r3w
- test r3, r3
- lea r4, [r3+3]
- cmovs r3, r4
- sar r3, 2 ; H/4
- lea r3, [r3*5] ; 5*(H/4)
- test r3, r3
- lea r4, [r3+15]
- cmovs r3, r4
- sar r3, 4 ; (5*(H/4))/16
- movd m0, r3d
-%endif
-
lea r4, [r0+r2*8-1]
lea r3, [r0+r2*4-1]
add r4, r2
-%ifdef ARCH_X86_64
-%define e_reg r11
+%if ARCH_X86_64
+%define e_reg r8
%else
%define e_reg r0
%endif
lea r5, [r5+r6*4]
movzx e_reg, byte [r3 ]
-%ifdef ARCH_X86_64
- movzx r10, byte [r4+r2 ]
- sub r10, e_reg
+%if ARCH_X86_64
+ movzx r7, byte [r4+r2 ]
+ sub r7, e_reg
%else
movzx r6, byte [r4+r2 ]
sub r6, e_reg
movzx r4, byte [e_reg+r2 ]
movzx r6, byte [r3 ]
sub r6, r4
-%ifdef ARCH_X86_64
- lea r6, [r10+r6*2]
+%if ARCH_X86_64
+ lea r6, [r7+r6*2]
lea r5, [r5+r6*2]
add r5, r6
%else
%endif
movzx r4, byte [e_reg ]
-%ifdef ARCH_X86_64
- movzx r10, byte [r3 +r2 ]
- sub r10, r4
- sub r5, r10
+%if ARCH_X86_64
+ movzx r7, byte [r3 +r2 ]
+ sub r7, r4
+ sub r5, r7
%else
movzx r6, byte [r3 +r2 ]
sub r6, r4
movzx r4, byte [e_reg+r1 ]
movzx r6, byte [r3 +r2*2]
sub r6, r4
-%ifdef ARCH_X86_64
- add r6, r10
+%if ARCH_X86_64
+ add r6, r7
%endif
lea r5, [r5+r6*8]
lea r5, [r5+r6*4]
add r5, r6 ; sum of V coefficients
-%ifndef ARCH_X86_64
+%if ARCH_X86_64 == 0
mov r0, r0m
%endif
-%ifidn %3, h264
+%ifidn %1, h264
lea r5, [r5*5+32]
sar r5, 6
-%elifidn %3, rv40
+%elifidn %1, rv40
lea r5, [r5*5]
sar r5, 6
-%elifidn %3, svq3
+%elifidn %1, svq3
test r5, r5
lea r6, [r5+3]
cmovs r5, r6
movzx r3, byte [r3+r2*2 ]
lea r3, [r3+r4+1]
shl r3, 4
+
movd r1d, m0
movsx r1d, r1w
+%ifnidn %1, svq3
+%ifidn %1, h264
+ lea r1d, [r1d*5+32]
+%else ; rv40
+ lea r1d, [r1d*5]
+%endif
+ sar r1d, 6
+%else ; svq3
+ test r1d, r1d
+ lea r4d, [r1d+3]
+ cmovs r1d, r4d
+ sar r1d, 2 ; H/4
+ lea r1d, [r1d*5] ; 5*(H/4)
+ test r1d, r1d
+ lea r4d, [r1d+15]
+ cmovs r1d, r4d
+ sar r1d, 4 ; (5*(H/4))/16
+%endif
+ movd m0, r1d
+
add r1d, r5d
add r3d, r1d
shl r1d, 3
movd m1, r5d
movd m3, r3d
-%ifidn %1, mmx
- punpcklwd m0, m0
- punpcklwd m1, m1
- punpcklwd m3, m3
- punpckldq m0, m0
- punpckldq m1, m1
- punpckldq m3, m3
-%elifidn %1, mmx2
- pshufw m0, m0, 0x0
- pshufw m1, m1, 0x0
- pshufw m3, m3, 0x0
-%else
- pshuflw m0, m0, 0x0
- pshuflw m1, m1, 0x0
- pshuflw m3, m3, 0x0
- punpcklqdq m0, m0 ; splat H (words)
- punpcklqdq m1, m1 ; splat V (words)
- punpcklqdq m3, m3 ; splat a (words)
-%endif
-%ifidn %3, svq3
+ SPLATW m0, m0, 0 ; H
+ SPLATW m1, m1, 0 ; V
+ SPLATW m3, m3, 0 ; a
+%ifidn %1, svq3
SWAP 0, 1
%endif
mova m2, m0
%endif
mov r4, 8
-.loop
+.loop:
mova m3, m0 ; b[0..7]
mova m4, m2 ; b[8..15]
psraw m3, 5
REP_RET
%endmacro
-INIT_MMX
-H264_PRED16x16_PLANE mmx, 0, h264
-H264_PRED16x16_PLANE mmx, 0, rv40
-H264_PRED16x16_PLANE mmx, 0, svq3
-H264_PRED16x16_PLANE mmx2, 0, h264
-H264_PRED16x16_PLANE mmx2, 0, rv40
-H264_PRED16x16_PLANE mmx2, 0, svq3
-INIT_XMM
-H264_PRED16x16_PLANE sse2, 8, h264
-H264_PRED16x16_PLANE sse2, 8, rv40
-H264_PRED16x16_PLANE sse2, 8, svq3
-H264_PRED16x16_PLANE ssse3, 8, h264
-H264_PRED16x16_PLANE ssse3, 8, rv40
-H264_PRED16x16_PLANE ssse3, 8, svq3
+INIT_MMX mmx
+H264_PRED16x16_PLANE h264
+H264_PRED16x16_PLANE rv40
+H264_PRED16x16_PLANE svq3
+INIT_MMX mmxext
+H264_PRED16x16_PLANE h264
+H264_PRED16x16_PLANE rv40
+H264_PRED16x16_PLANE svq3
+INIT_XMM sse2
+H264_PRED16x16_PLANE h264
+H264_PRED16x16_PLANE rv40
+H264_PRED16x16_PLANE svq3
+INIT_XMM ssse3
+H264_PRED16x16_PLANE h264
+H264_PRED16x16_PLANE rv40
+H264_PRED16x16_PLANE svq3
;-----------------------------------------------------------------------------
-; void pred8x8_plane(uint8_t *src, int stride)
+; void pred8x8_plane_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-%macro H264_PRED8x8_PLANE 2
-cglobal pred8x8_plane_%1, 2, 7, %2
+%macro H264_PRED8x8_PLANE 0
+cglobal pred8x8_plane_8, 2,9,7
mov r2, r1 ; +stride
neg r1 ; -stride
pmullw m0, [pw_m4to4]
pmullw m1, [pw_m4to4+8]
%else ; mmsize == 16
-%ifidn %1, sse2
+%if cpuflag(ssse3)
+ movhps m0, [r0+r1 +4] ; this reads 4 bytes more than necessary
+ pmaddubsw m0, [plane8_shuf] ; H coefficients
+%else ; sse2
pxor m2, m2
movd m1, [r0+r1 +4]
punpckldq m0, m1
punpcklbw m0, m2
pmullw m0, [pw_m4to4]
-%else ; ssse3
- movhps m0, [r0+r1 +4] ; this reads 4 bytes more than necessary
- pmaddubsw m0, [plane8_shuf] ; H coefficients
%endif
movhlps m1, m0
%endif
paddw m0, m1
-%ifnidn %1, ssse3
-%ifidn %1, mmx
+%if notcpuflag(ssse3)
+%if cpuflag(mmxext)
+ PSHUFLW m1, m0, 0xE
+%elif cpuflag(mmx)
mova m1, m0
psrlq m1, 32
-%elifidn %1, mmx2
- pshufw m1, m0, 0xE
-%else ; mmsize == 16
- pshuflw m1, m0, 0xE
%endif
paddw m0, m1
%endif ; !ssse3
-%ifidn %1, mmx
+%if cpuflag(mmxext)
+ PSHUFLW m1, m0, 0x1
+%elif cpuflag(mmx)
mova m1, m0
psrlq m1, 16
-%elifidn %1, mmx2
- pshufw m1, m0, 0x1
-%else
- pshuflw m1, m0, 0x1
%endif
paddw m0, m1 ; sum of H coefficients
- pmullw m0, [pw_17]
- paddw m0, [pw_16]
- psraw m0, 5
-
lea r4, [r0+r2*4-1]
lea r3, [r0 -1]
add r4, r2
-%ifdef ARCH_X86_64
-%define e_reg r11
+%if ARCH_X86_64
+%define e_reg r8
%else
%define e_reg r0
%endif
sub r5, e_reg
movzx e_reg, byte [r3 ]
-%ifdef ARCH_X86_64
- movzx r10, byte [r4+r2 ]
- sub r10, e_reg
- sub r5, r10
+%if ARCH_X86_64
+ movzx r7, byte [r4+r2 ]
+ sub r7, e_reg
+ sub r5, r7
%else
movzx r6, byte [r4+r2 ]
sub r6, e_reg
movzx e_reg, byte [r3+r1 ]
movzx r6, byte [r4+r2*2 ]
sub r6, e_reg
-%ifdef ARCH_X86_64
- add r6, r10
+%if ARCH_X86_64
+ add r6, r7
%endif
lea r5, [r5+r6*4]
lea r5, [r5+r6*8]
sar r5, 5
-%ifndef ARCH_X86_64
+%if ARCH_X86_64 == 0
mov r0, r0m
%endif
shl r3, 4
movd r1d, m0
movsx r1d, r1w
+ imul r1d, 17
+ add r1d, 16
+ sar r1d, 5
+ movd m0, r1d
add r1d, r5d
sub r3d, r1d
add r1d, r1d
movd m1, r5d
movd m3, r3d
-%ifidn %1, mmx
- punpcklwd m0, m0
- punpcklwd m1, m1
- punpcklwd m3, m3
- punpckldq m0, m0
- punpckldq m1, m1
- punpckldq m3, m3
-%elifidn %1, mmx2
- pshufw m0, m0, 0x0
- pshufw m1, m1, 0x0
- pshufw m3, m3, 0x0
-%else
- pshuflw m0, m0, 0x0
- pshuflw m1, m1, 0x0
- pshuflw m3, m3, 0x0
- punpcklqdq m0, m0 ; splat H (words)
- punpcklqdq m1, m1 ; splat V (words)
- punpcklqdq m3, m3 ; splat a (words)
-%endif
+ SPLATW m0, m0, 0 ; H
+ SPLATW m1, m1, 0 ; V
+ SPLATW m3, m3, 0 ; a
%if mmsize == 8
mova m2, m0
%endif
mov r4, 4
ALIGN 16
-.loop
+.loop:
%if mmsize == 16
mova m3, m0 ; b[0..7]
paddw m0, m1
REP_RET
%endmacro
-INIT_MMX
-H264_PRED8x8_PLANE mmx, 0
-H264_PRED8x8_PLANE mmx2, 0
-INIT_XMM
-H264_PRED8x8_PLANE sse2, 8
-H264_PRED8x8_PLANE ssse3, 8
+INIT_MMX mmx
+H264_PRED8x8_PLANE
+INIT_MMX mmxext
+H264_PRED8x8_PLANE
+INIT_XMM sse2
+H264_PRED8x8_PLANE
+INIT_XMM ssse3
+H264_PRED8x8_PLANE
;-----------------------------------------------------------------------------
-; void pred8x8_vertical(uint8_t *src, int stride)
+; void pred8x8_vertical_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-cglobal pred8x8_vertical_mmx, 2,2
+INIT_MMX mmx
+cglobal pred8x8_vertical_8, 2,2
sub r0, r1
movq mm0, [r0]
%rep 3
RET
;-----------------------------------------------------------------------------
-; void pred8x8_horizontal(uint8_t *src, int stride)
+; void pred8x8_horizontal_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-%macro PRED8x8_H 1
-cglobal pred8x8_horizontal_%1, 2,3
+%macro PRED8x8_H 0
+cglobal pred8x8_horizontal_8, 2,3
mov r2, 4
-%ifidn %1, ssse3
+%if cpuflag(ssse3)
mova m2, [pb_3]
%endif
.loop:
- movd m0, [r0+r1*0-4]
- movd m1, [r0+r1*1-4]
-%ifidn %1, ssse3
- pshufb m0, m2
- pshufb m1, m2
-%else
- punpcklbw m0, m0
- punpcklbw m1, m1
-%ifidn %1, mmxext
- pshufw m0, m0, 0xff
- pshufw m1, m1, 0xff
-%else
- punpckhwd m0, m0
- punpckhwd m1, m1
- punpckhdq m0, m0
- punpckhdq m1, m1
-%endif
-%endif
+ SPLATB_LOAD m0, r0+r1*0-1, m2
+ SPLATB_LOAD m1, r0+r1*1-1, m2
mova [r0+r1*0], m0
mova [r0+r1*1], m1
lea r0, [r0+r1*2]
REP_RET
%endmacro
-INIT_MMX
-PRED8x8_H mmx
-PRED8x8_H mmxext
-PRED8x8_H ssse3
+INIT_MMX mmx
+PRED8x8_H
+INIT_MMX mmxext
+PRED8x8_H
+INIT_MMX ssse3
+PRED8x8_H
;-----------------------------------------------------------------------------
-; void pred8x8_dc_rv40(uint8_t *src, int stride)
+; void pred8x8_top_dc_8_mmxext(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+INIT_MMX mmxext
+cglobal pred8x8_top_dc_8, 2,5
+ sub r0, r1
+ movq mm0, [r0]
+ pxor mm1, mm1
+ pxor mm2, mm2
+ lea r2, [r0+r1*2]
+ punpckhbw mm1, mm0
+ punpcklbw mm0, mm2
+ psadbw mm1, mm2 ; s1
+ lea r3, [r2+r1*2]
+ psadbw mm0, mm2 ; s0
+ psrlw mm1, 1
+ psrlw mm0, 1
+ pavgw mm1, mm2
+ lea r4, [r3+r1*2]
+ pavgw mm0, mm2
+ pshufw mm1, mm1, 0
+ pshufw mm0, mm0, 0 ; dc0 (w)
+ packuswb mm0, mm1 ; dc0,dc1 (b)
+ movq [r0+r1*1], mm0
+ movq [r0+r1*2], mm0
+ lea r0, [r3+r1*2]
+ movq [r2+r1*1], mm0
+ movq [r2+r1*2], mm0
+ movq [r3+r1*1], mm0
+ movq [r3+r1*2], mm0
+ movq [r0+r1*1], mm0
+ movq [r0+r1*2], mm0
+ RET
+
+;-----------------------------------------------------------------------------
+; void pred8x8_dc_8_mmxext(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-cglobal pred8x8_dc_rv40_mmxext, 2,7
+INIT_MMX mmxext
+cglobal pred8x8_dc_8, 2,5
+ sub r0, r1
+ pxor m7, m7
+ movd m0, [r0+0]
+ movd m1, [r0+4]
+ psadbw m0, m7 ; s0
+ mov r4, r0
+ psadbw m1, m7 ; s1
+
+ movzx r2d, byte [r0+r1*1-1]
+ movzx r3d, byte [r0+r1*2-1]
+ lea r0, [r0+r1*2]
+ add r2d, r3d
+ movzx r3d, byte [r0+r1*1-1]
+ add r2d, r3d
+ movzx r3d, byte [r0+r1*2-1]
+ add r2d, r3d
+ lea r0, [r0+r1*2]
+ movd m2, r2d ; s2
+ movzx r2d, byte [r0+r1*1-1]
+ movzx r3d, byte [r0+r1*2-1]
+ lea r0, [r0+r1*2]
+ add r2d, r3d
+ movzx r3d, byte [r0+r1*1-1]
+ add r2d, r3d
+ movzx r3d, byte [r0+r1*2-1]
+ add r2d, r3d
+ movd m3, r2d ; s3
+
+ punpcklwd m0, m1
+ mov r0, r4
+ punpcklwd m2, m3
+ punpckldq m0, m2 ; s0, s1, s2, s3
+ pshufw m3, m0, 11110110b ; s2, s1, s3, s3
+ lea r2, [r0+r1*2]
+ pshufw m0, m0, 01110100b ; s0, s1, s3, s1
+ paddw m0, m3
+ lea r3, [r2+r1*2]
+ psrlw m0, 2
+ pavgw m0, m7 ; s0+s2, s1, s3, s1+s3
+ lea r4, [r3+r1*2]
+ packuswb m0, m0
+ punpcklbw m0, m0
+ movq m1, m0
+ punpcklbw m0, m0
+ punpckhbw m1, m1
+ movq [r0+r1*1], m0
+ movq [r0+r1*2], m0
+ movq [r2+r1*1], m0
+ movq [r2+r1*2], m0
+ movq [r3+r1*1], m1
+ movq [r3+r1*2], m1
+ movq [r4+r1*1], m1
+ movq [r4+r1*2], m1
+ RET
+
+;-----------------------------------------------------------------------------
+; void pred8x8_dc_rv40_8(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred8x8_dc_rv40_8, 2,7
mov r4, r0
sub r0, r1
pxor mm0, mm0
REP_RET
;-----------------------------------------------------------------------------
-; void pred8x8_tm_vp8(uint8_t *src, int stride)
+; void pred8x8_tm_vp8_8(uint8_t *src, int stride)
;-----------------------------------------------------------------------------
-%macro PRED8x8_TM_MMX 1
-cglobal pred8x8_tm_vp8_%1, 2,6
+%macro PRED8x8_TM 0
+cglobal pred8x8_tm_vp8_8, 2,6
sub r0, r1
pxor mm7, mm7
movq mm0, [r0]
sub r3d, r4d
movd mm2, r2d
movd mm4, r3d
-%ifidn %1, mmx
- punpcklwd mm2, mm2
- punpcklwd mm4, mm4
- punpckldq mm2, mm2
- punpckldq mm4, mm4
-%else
- pshufw mm2, mm2, 0
- pshufw mm4, mm4, 0
-%endif
+ SPLATW mm2, mm2, 0
+ SPLATW mm4, mm4, 0
movq mm3, mm2
movq mm5, mm4
paddw mm2, mm0
REP_RET
%endmacro
-PRED8x8_TM_MMX mmx
-PRED8x8_TM_MMX mmxext
+INIT_MMX mmx
+PRED8x8_TM
+INIT_MMX mmxext
+PRED8x8_TM
-cglobal pred8x8_tm_vp8_sse2, 2,6,4
+INIT_XMM sse2
+cglobal pred8x8_tm_vp8_8, 2,6,4
sub r0, r1
pxor xmm1, xmm1
movq xmm0, [r0]
jg .loop
REP_RET
-cglobal pred8x8_tm_vp8_ssse3, 2,3,6
+INIT_XMM ssse3
+cglobal pred8x8_tm_vp8_8, 2,3,6
sub r0, r1
movdqa xmm4, [tm_shuf]
pxor xmm1, xmm1
jg .loop
REP_RET
+; dest, left, right, src, tmp
+; output: %1 = (t[n-1] + t[n]*2 + t[n+1] + 2) >> 2
+%macro PRED4x4_LOWPASS 5
+ mova %5, %2
+ pavgb %2, %3
+ pxor %3, %5
+ mova %1, %4
+ pand %3, [pb_1]
+ psubusb %2, %3
+ pavgb %1, %2
+%endmacro
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_top_dc_8(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+%macro PRED8x8L_TOP_DC 0
+cglobal pred8x8l_top_dc_8, 4,4
+ sub r0, r3
+ pxor mm7, mm7
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1 ; top_left
+ jz .fix_lt_2
+ test r2, r2 ; top_right
+ jz .fix_tr_1
+ jmp .body
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2 ; top_right
+ jnz .body
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+.body:
+ PRED4x4_LOWPASS mm0, mm2, mm1, mm3, mm5
+ psadbw mm7, mm0
+ paddw mm7, [pw_4]
+ psrlw mm7, 3
+ pshufw mm7, mm7, 0
+ packuswb mm7, mm7
+%rep 3
+ movq [r0+r3*1], mm7
+ movq [r0+r3*2], mm7
+ lea r0, [r0+r3*2]
+%endrep
+ movq [r0+r3*1], mm7
+ movq [r0+r3*2], mm7
+ RET
+%endmacro
+
+INIT_MMX mmxext
+PRED8x8L_TOP_DC
+INIT_MMX ssse3
+PRED8x8L_TOP_DC
+
+;-----------------------------------------------------------------------------
+;void pred8x8l_dc_8(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED8x8L_DC 0
+cglobal pred8x8l_dc_8, 4,5
+ sub r0, r3
+ lea r4, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ punpckhbw mm0, [r0+r3*0-8]
+ movq mm1, [r4+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r4, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r4]
+ mov r0, r4
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ test r1, r1
+ jnz .do_left
+.fix_lt_1:
+ movq mm5, mm3
+ pxor mm5, mm4
+ psrlq mm5, 56
+ psllq mm5, 48
+ pxor mm1, mm5
+ jmp .do_left
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2
+ jnz .body
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .body
+.do_left:
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq mm4, mm0
+ movq mm7, mm2
+ PRED4x4_LOWPASS mm1, mm3, mm0, mm4, mm5
+ psllq mm1, 56
+ PALIGNR mm7, mm1, 7, mm3
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1
+ jz .fix_lt_2
+ test r2, r2
+ jz .fix_tr_1
+.body:
+ lea r1, [r0+r3*2]
+ PRED4x4_LOWPASS mm6, mm2, mm1, mm3, mm5
+ pxor mm0, mm0
+ pxor mm1, mm1
+ lea r2, [r1+r3*2]
+ psadbw mm0, mm7
+ psadbw mm1, mm6
+ paddw mm0, [pw_8]
+ paddw mm0, mm1
+ lea r4, [r2+r3*2]
+ psrlw mm0, 4
+ pshufw mm0, mm0, 0
+ packuswb mm0, mm0
+ movq [r0+r3*1], mm0
+ movq [r0+r3*2], mm0
+ movq [r1+r3*1], mm0
+ movq [r1+r3*2], mm0
+ movq [r2+r3*1], mm0
+ movq [r2+r3*2], mm0
+ movq [r4+r3*1], mm0
+ movq [r4+r3*2], mm0
+ RET
+%endmacro
+
+INIT_MMX mmxext
+PRED8x8L_DC
+INIT_MMX ssse3
+PRED8x8L_DC
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_horizontal_8(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED8x8L_HORIZONTAL 0
+cglobal pred8x8l_horizontal_8, 4,4
+ sub r0, r3
+ lea r2, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ test r1, r1
+ lea r1, [r0+r3]
+ cmovnz r1, r0
+ punpckhbw mm0, [r1+r3*0-8]
+ movq mm1, [r2+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r2, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r1+r3*0-8]
+ mov r0, r2
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq mm4, mm0
+ movq mm7, mm2
+ PRED4x4_LOWPASS mm1, mm3, mm0, mm4, mm5
+ psllq mm1, 56
+ PALIGNR mm7, mm1, 7, mm3
+ movq mm3, mm7
+ lea r1, [r0+r3*2]
+ movq mm7, mm3
+ punpckhbw mm3, mm3
+ punpcklbw mm7, mm7
+ pshufw mm0, mm3, 0xff
+ pshufw mm1, mm3, 0xaa
+ lea r2, [r1+r3*2]
+ pshufw mm2, mm3, 0x55
+ pshufw mm3, mm3, 0x00
+ pshufw mm4, mm7, 0xff
+ pshufw mm5, mm7, 0xaa
+ pshufw mm6, mm7, 0x55
+ pshufw mm7, mm7, 0x00
+ movq [r0+r3*1], mm0
+ movq [r0+r3*2], mm1
+ movq [r1+r3*1], mm2
+ movq [r1+r3*2], mm3
+ movq [r2+r3*1], mm4
+ movq [r2+r3*2], mm5
+ lea r0, [r2+r3*2]
+ movq [r0+r3*1], mm6
+ movq [r0+r3*2], mm7
+ RET
+%endmacro
+
+INIT_MMX mmxext
+PRED8x8L_HORIZONTAL
+INIT_MMX ssse3
+PRED8x8L_HORIZONTAL
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_vertical_8(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED8x8L_VERTICAL 0
+cglobal pred8x8l_vertical_8, 4,4
+ sub r0, r3
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1 ; top_left
+ jz .fix_lt_2
+ test r2, r2 ; top_right
+ jz .fix_tr_1
+ jmp .body
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2 ; top_right
+ jnz .body
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+.body:
+ PRED4x4_LOWPASS mm0, mm2, mm1, mm3, mm5
+%rep 3
+ movq [r0+r3*1], mm0
+ movq [r0+r3*2], mm0
+ lea r0, [r0+r3*2]
+%endrep
+ movq [r0+r3*1], mm0
+ movq [r0+r3*2], mm0
+ RET
+%endmacro
+
+INIT_MMX mmxext
+PRED8x8L_VERTICAL
+INIT_MMX ssse3
+PRED8x8L_VERTICAL
+
;-----------------------------------------------------------------------------
-; void pred4x4_dc_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+;void pred8x8l_down_left_8(uint8_t *src, int has_topleft, int has_topright, int stride)
;-----------------------------------------------------------------------------
-cglobal pred4x4_dc_mmxext, 3,5
+INIT_MMX mmxext
+cglobal pred8x8l_down_left_8, 4,5
+ sub r0, r3
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1
+ jz .fix_lt_2
+ test r2, r2
+ jz .fix_tr_1
+ jmp .do_top
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.fix_tr_2:
+ punpckhbw mm3, mm3
+ pshufw mm1, mm3, 0xFF
+ jmp .do_topright
+.do_top:
+ PRED4x4_LOWPASS mm4, mm2, mm1, mm3, mm5
+ movq mm7, mm4
+ test r2, r2
+ jz .fix_tr_2
+ movq mm0, [r0+8]
+ movq mm5, mm0
+ movq mm2, mm0
+ movq mm4, mm0
+ psrlq mm5, 56
+ PALIGNR mm2, mm3, 7, mm3
+ PALIGNR mm5, mm4, 1, mm4
+ PRED4x4_LOWPASS mm1, mm2, mm5, mm0, mm4
+.do_topright:
+ lea r1, [r0+r3*2]
+ movq mm6, mm1
+ psrlq mm1, 56
+ movq mm4, mm1
+ lea r2, [r1+r3*2]
+ movq mm2, mm6
+ PALIGNR mm2, mm7, 1, mm0
+ movq mm3, mm6
+ PALIGNR mm3, mm7, 7, mm0
+ PALIGNR mm4, mm6, 1, mm0
+ movq mm5, mm7
+ movq mm1, mm7
+ movq mm7, mm6
+ lea r4, [r2+r3*2]
+ psllq mm1, 8
+ PRED4x4_LOWPASS mm0, mm1, mm2, mm5, mm6
+ PRED4x4_LOWPASS mm1, mm3, mm4, mm7, mm6
+ movq [r4+r3*2], mm1
+ movq mm2, mm0
+ psllq mm1, 8
+ psrlq mm2, 56
+ psllq mm0, 8
+ por mm1, mm2
+ movq [r4+r3*1], mm1
+ movq mm2, mm0
+ psllq mm1, 8
+ psrlq mm2, 56
+ psllq mm0, 8
+ por mm1, mm2
+ movq [r2+r3*2], mm1
+ movq mm2, mm0
+ psllq mm1, 8
+ psrlq mm2, 56
+ psllq mm0, 8
+ por mm1, mm2
+ movq [r2+r3*1], mm1
+ movq mm2, mm0
+ psllq mm1, 8
+ psrlq mm2, 56
+ psllq mm0, 8
+ por mm1, mm2
+ movq [r1+r3*2], mm1
+ movq mm2, mm0
+ psllq mm1, 8
+ psrlq mm2, 56
+ psllq mm0, 8
+ por mm1, mm2
+ movq [r1+r3*1], mm1
+ movq mm2, mm0
+ psllq mm1, 8
+ psrlq mm2, 56
+ psllq mm0, 8
+ por mm1, mm2
+ movq [r0+r3*2], mm1
+ psllq mm1, 8
+ psrlq mm0, 56
+ por mm1, mm0
+ movq [r0+r3*1], mm1
+ RET
+
+%macro PRED8x8L_DOWN_LEFT 0
+cglobal pred8x8l_down_left_8, 4,4
+ sub r0, r3
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1 ; top_left
+ jz .fix_lt_2
+ test r2, r2 ; top_right
+ jz .fix_tr_1
+ jmp .do_top
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2 ; top_right
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.fix_tr_2:
+ punpckhbw mm3, mm3
+ pshufw mm1, mm3, 0xFF
+ jmp .do_topright
+.do_top:
+ PRED4x4_LOWPASS mm4, mm2, mm1, mm3, mm5
+ movq2dq xmm3, mm4
+ test r2, r2 ; top_right
+ jz .fix_tr_2
+ movq mm0, [r0+8]
+ movq mm5, mm0
+ movq mm2, mm0
+ movq mm4, mm0
+ psrlq mm5, 56
+ PALIGNR mm2, mm3, 7, mm3
+ PALIGNR mm5, mm4, 1, mm4
+ PRED4x4_LOWPASS mm1, mm2, mm5, mm0, mm4
+.do_topright:
+ movq2dq xmm4, mm1
+ psrlq mm1, 56
+ movq2dq xmm5, mm1
+ lea r1, [r0+r3*2]
+ pslldq xmm4, 8
+ por xmm3, xmm4
+ movdqa xmm2, xmm3
+ psrldq xmm2, 1
+ pslldq xmm5, 15
+ por xmm2, xmm5
+ lea r2, [r1+r3*2]
+ movdqa xmm1, xmm3
+ pslldq xmm1, 1
+INIT_XMM cpuname
+ PRED4x4_LOWPASS xmm0, xmm1, xmm2, xmm3, xmm4
+ psrldq xmm0, 1
+ movq [r0+r3*1], xmm0
+ psrldq xmm0, 1
+ movq [r0+r3*2], xmm0
+ psrldq xmm0, 1
+ lea r0, [r2+r3*2]
+ movq [r1+r3*1], xmm0
+ psrldq xmm0, 1
+ movq [r1+r3*2], xmm0
+ psrldq xmm0, 1
+ movq [r2+r3*1], xmm0
+ psrldq xmm0, 1
+ movq [r2+r3*2], xmm0
+ psrldq xmm0, 1
+ movq [r0+r3*1], xmm0
+ psrldq xmm0, 1
+ movq [r0+r3*2], xmm0
+ RET
+%endmacro
+
+INIT_MMX sse2
+PRED8x8L_DOWN_LEFT
+INIT_MMX ssse3
+PRED8x8L_DOWN_LEFT
+
+;-----------------------------------------------------------------------------
+;void pred8x8l_down_right_8_mmxext(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred8x8l_down_right_8, 4,5
+ sub r0, r3
+ lea r4, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ punpckhbw mm0, [r0+r3*0-8]
+ movq mm1, [r4+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r4, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r4]
+ mov r0, r4
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ test r1, r1 ; top_left
+ jz .fix_lt_1
+.do_left:
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq mm4, mm0
+ movq mm7, mm2
+ movq mm6, mm2
+ PRED4x4_LOWPASS mm1, mm3, mm0, mm4, mm5
+ psllq mm1, 56
+ PALIGNR mm7, mm1, 7, mm3
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1 ; top_left
+ jz .fix_lt_2
+ test r2, r2 ; top_right
+ jz .fix_tr_1
+.do_top:
+ PRED4x4_LOWPASS mm4, mm2, mm1, mm3, mm5
+ movq mm5, mm4
+ jmp .body
+.fix_lt_1:
+ movq mm5, mm3
+ pxor mm5, mm4
+ psrlq mm5, 56
+ psllq mm5, 48
+ pxor mm1, mm5
+ jmp .do_left
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2 ; top_right
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.body:
+ lea r1, [r0+r3*2]
+ movq mm1, mm7
+ movq mm7, mm5
+ movq mm5, mm6
+ movq mm2, mm7
+ lea r2, [r1+r3*2]
+ PALIGNR mm2, mm6, 1, mm0
+ movq mm3, mm7
+ PALIGNR mm3, mm6, 7, mm0
+ movq mm4, mm7
+ lea r4, [r2+r3*2]
+ psrlq mm4, 8
+ PRED4x4_LOWPASS mm0, mm1, mm2, mm5, mm6
+ PRED4x4_LOWPASS mm1, mm3, mm4, mm7, mm6
+ movq [r4+r3*2], mm0
+ movq mm2, mm1
+ psrlq mm0, 8
+ psllq mm2, 56
+ psrlq mm1, 8
+ por mm0, mm2
+ movq [r4+r3*1], mm0
+ movq mm2, mm1
+ psrlq mm0, 8
+ psllq mm2, 56
+ psrlq mm1, 8
+ por mm0, mm2
+ movq [r2+r3*2], mm0
+ movq mm2, mm1
+ psrlq mm0, 8
+ psllq mm2, 56
+ psrlq mm1, 8
+ por mm0, mm2
+ movq [r2+r3*1], mm0
+ movq mm2, mm1
+ psrlq mm0, 8
+ psllq mm2, 56
+ psrlq mm1, 8
+ por mm0, mm2
+ movq [r1+r3*2], mm0
+ movq mm2, mm1
+ psrlq mm0, 8
+ psllq mm2, 56
+ psrlq mm1, 8
+ por mm0, mm2
+ movq [r1+r3*1], mm0
+ movq mm2, mm1
+ psrlq mm0, 8
+ psllq mm2, 56
+ psrlq mm1, 8
+ por mm0, mm2
+ movq [r0+r3*2], mm0
+ psrlq mm0, 8
+ psllq mm1, 56
+ por mm0, mm1
+ movq [r0+r3*1], mm0
+ RET
+
+%macro PRED8x8L_DOWN_RIGHT 0
+cglobal pred8x8l_down_right_8, 4,5
+ sub r0, r3
+ lea r4, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ punpckhbw mm0, [r0+r3*0-8]
+ movq mm1, [r4+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r4, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r4]
+ mov r0, r4
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ test r1, r1
+ jz .fix_lt_1
+ jmp .do_left
+.fix_lt_1:
+ movq mm5, mm3
+ pxor mm5, mm4
+ psrlq mm5, 56
+ psllq mm5, 48
+ pxor mm1, mm5
+ jmp .do_left
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.do_left:
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq mm4, mm0
+ movq mm7, mm2
+ movq2dq xmm3, mm2
+ PRED4x4_LOWPASS mm1, mm3, mm0, mm4, mm5
+ psllq mm1, 56
+ PALIGNR mm7, mm1, 7, mm3
+ movq2dq xmm1, mm7
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1
+ jz .fix_lt_2
+ test r2, r2
+ jz .fix_tr_1
+.do_top:
+ PRED4x4_LOWPASS mm4, mm2, mm1, mm3, mm5
+ movq2dq xmm4, mm4
+ lea r1, [r0+r3*2]
+ movdqa xmm0, xmm3
+ pslldq xmm4, 8
+ por xmm3, xmm4
+ lea r2, [r1+r3*2]
+ pslldq xmm4, 1
+ por xmm1, xmm4
+ psrldq xmm0, 7
+ pslldq xmm0, 15
+ psrldq xmm0, 7
+ por xmm1, xmm0
+ lea r0, [r2+r3*2]
+ movdqa xmm2, xmm3
+ psrldq xmm2, 1
+INIT_XMM cpuname
+ PRED4x4_LOWPASS xmm0, xmm1, xmm2, xmm3, xmm4
+ movdqa xmm1, xmm0
+ psrldq xmm1, 1
+ movq [r0+r3*2], xmm0
+ movq [r0+r3*1], xmm1
+ psrldq xmm0, 2
+ psrldq xmm1, 2
+ movq [r2+r3*2], xmm0
+ movq [r2+r3*1], xmm1
+ psrldq xmm0, 2
+ psrldq xmm1, 2
+ movq [r1+r3*2], xmm0
+ movq [r1+r3*1], xmm1
+ psrldq xmm0, 2
+ psrldq xmm1, 2
+ movq [r4+r3*2], xmm0
+ movq [r4+r3*1], xmm1
+ RET
+%endmacro
+
+INIT_MMX sse2
+PRED8x8L_DOWN_RIGHT
+INIT_MMX ssse3
+PRED8x8L_DOWN_RIGHT
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_vertical_right_8(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred8x8l_vertical_right_8, 4,5
+ sub r0, r3
+ lea r4, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ punpckhbw mm0, [r0+r3*0-8]
+ movq mm1, [r4+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r4, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r4]
+ mov r0, r4
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ test r1, r1
+ jz .fix_lt_1
+ jmp .do_left
+.fix_lt_1:
+ movq mm5, mm3
+ pxor mm5, mm4
+ psrlq mm5, 56
+ psllq mm5, 48
+ pxor mm1, mm5
+ jmp .do_left
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.do_left:
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq mm7, mm2
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1
+ jz .fix_lt_2
+ test r2, r2
+ jz .fix_tr_1
+.do_top:
+ PRED4x4_LOWPASS mm6, mm2, mm1, mm3, mm5
+ lea r1, [r0+r3*2]
+ movq mm2, mm6
+ movq mm3, mm6
+ PALIGNR mm3, mm7, 7, mm0
+ PALIGNR mm6, mm7, 6, mm1
+ movq mm4, mm3
+ pavgb mm3, mm2
+ lea r2, [r1+r3*2]
+ PRED4x4_LOWPASS mm0, mm6, mm2, mm4, mm5
+ movq [r0+r3*1], mm3
+ movq [r0+r3*2], mm0
+ movq mm5, mm0
+ movq mm6, mm3
+ movq mm1, mm7
+ movq mm2, mm1
+ psllq mm2, 8
+ movq mm3, mm1
+ psllq mm3, 16
+ lea r4, [r2+r3*2]
+ PRED4x4_LOWPASS mm0, mm1, mm3, mm2, mm4
+ PALIGNR mm6, mm0, 7, mm2
+ movq [r1+r3*1], mm6
+ psllq mm0, 8
+ PALIGNR mm5, mm0, 7, mm1
+ movq [r1+r3*2], mm5
+ psllq mm0, 8
+ PALIGNR mm6, mm0, 7, mm2
+ movq [r2+r3*1], mm6
+ psllq mm0, 8
+ PALIGNR mm5, mm0, 7, mm1
+ movq [r2+r3*2], mm5
+ psllq mm0, 8
+ PALIGNR mm6, mm0, 7, mm2
+ movq [r4+r3*1], mm6
+ psllq mm0, 8
+ PALIGNR mm5, mm0, 7, mm1
+ movq [r4+r3*2], mm5
+ RET
+
+%macro PRED8x8L_VERTICAL_RIGHT 0
+cglobal pred8x8l_vertical_right_8, 4,5,7
+ ; manually spill XMM registers for Win64 because
+ ; the code here is initialized with INIT_MMX
+ WIN64_SPILL_XMM 7
+ sub r0, r3
+ lea r4, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ punpckhbw mm0, [r0+r3*0-8]
+ movq mm1, [r4+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r4, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r4]
+ mov r0, r4
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ test r1, r1
+ jnz .do_left
+.fix_lt_1:
+ movq mm5, mm3
+ pxor mm5, mm4
+ psrlq mm5, 56
+ psllq mm5, 48
+ pxor mm1, mm5
+ jmp .do_left
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.do_left:
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq2dq xmm0, mm2
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1
+ jz .fix_lt_2
+ test r2, r2
+ jz .fix_tr_1
+.do_top:
+ PRED4x4_LOWPASS mm6, mm2, mm1, mm3, mm5
+ lea r1, [r0+r3*2]
+ movq2dq xmm4, mm6
+ pslldq xmm4, 8
+ por xmm0, xmm4
+ movdqa xmm6, [pw_ff00]
+ movdqa xmm1, xmm0
+ lea r2, [r1+r3*2]
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm0
+ pslldq xmm0, 1
+ pslldq xmm1, 2
+ pavgb xmm2, xmm0
+INIT_XMM cpuname
+ PRED4x4_LOWPASS xmm4, xmm3, xmm1, xmm0, xmm5
+ pandn xmm6, xmm4
+ movdqa xmm5, xmm4
+ psrlw xmm4, 8
+ packuswb xmm6, xmm4
+ movhlps xmm4, xmm6
+ movhps [r0+r3*2], xmm5
+ movhps [r0+r3*1], xmm2
+ psrldq xmm5, 4
+ movss xmm5, xmm6
+ psrldq xmm2, 4
+ movss xmm2, xmm4
+ lea r0, [r2+r3*2]
+ psrldq xmm5, 1
+ psrldq xmm2, 1
+ movq [r0+r3*2], xmm5
+ movq [r0+r3*1], xmm2
+ psrldq xmm5, 1
+ psrldq xmm2, 1
+ movq [r2+r3*2], xmm5
+ movq [r2+r3*1], xmm2
+ psrldq xmm5, 1
+ psrldq xmm2, 1
+ movq [r1+r3*2], xmm5
+ movq [r1+r3*1], xmm2
+ RET
+%endmacro
+
+INIT_MMX sse2
+PRED8x8L_VERTICAL_RIGHT
+INIT_MMX ssse3
+PRED8x8L_VERTICAL_RIGHT
+
+;-----------------------------------------------------------------------------
+;void pred8x8l_vertical_left_8(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED8x8L_VERTICAL_LEFT 0
+cglobal pred8x8l_vertical_left_8, 4,4
+ sub r0, r3
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1
+ jz .fix_lt_2
+ test r2, r2
+ jz .fix_tr_1
+ jmp .do_top
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.fix_tr_2:
+ punpckhbw mm3, mm3
+ pshufw mm1, mm3, 0xFF
+ jmp .do_topright
+.do_top:
+ PRED4x4_LOWPASS mm4, mm2, mm1, mm3, mm5
+ movq2dq xmm4, mm4
+ test r2, r2
+ jz .fix_tr_2
+ movq mm0, [r0+8]
+ movq mm5, mm0
+ movq mm2, mm0
+ movq mm4, mm0
+ psrlq mm5, 56
+ PALIGNR mm2, mm3, 7, mm3
+ PALIGNR mm5, mm4, 1, mm4
+ PRED4x4_LOWPASS mm1, mm2, mm5, mm0, mm4
+.do_topright:
+ movq2dq xmm3, mm1
+ lea r1, [r0+r3*2]
+ pslldq xmm3, 8
+ por xmm4, xmm3
+ movdqa xmm2, xmm4
+ movdqa xmm1, xmm4
+ movdqa xmm3, xmm4
+ psrldq xmm2, 1
+ pslldq xmm1, 1
+ pavgb xmm3, xmm2
+ lea r2, [r1+r3*2]
+INIT_XMM cpuname
+ PRED4x4_LOWPASS xmm0, xmm1, xmm2, xmm4, xmm5
+ psrldq xmm0, 1
+ movq [r0+r3*1], xmm3
+ movq [r0+r3*2], xmm0
+ lea r0, [r2+r3*2]
+ psrldq xmm3, 1
+ psrldq xmm0, 1
+ movq [r1+r3*1], xmm3
+ movq [r1+r3*2], xmm0
+ psrldq xmm3, 1
+ psrldq xmm0, 1
+ movq [r2+r3*1], xmm3
+ movq [r2+r3*2], xmm0
+ psrldq xmm3, 1
+ psrldq xmm0, 1
+ movq [r0+r3*1], xmm3
+ movq [r0+r3*2], xmm0
+ RET
+%endmacro
+
+INIT_MMX sse2
+PRED8x8L_VERTICAL_LEFT
+INIT_MMX ssse3
+PRED8x8L_VERTICAL_LEFT
+
+;-----------------------------------------------------------------------------
+; void pred8x8l_horizontal_up_8(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED8x8L_HORIZONTAL_UP 0
+cglobal pred8x8l_horizontal_up_8, 4,4
+ sub r0, r3
+ lea r2, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ test r1, r1
+ lea r1, [r0+r3]
+ cmovnz r1, r0
+ punpckhbw mm0, [r1+r3*0-8]
+ movq mm1, [r2+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r2, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r1+r3*0-8]
+ mov r0, r2
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq mm4, mm0
+ movq mm7, mm2
+ PRED4x4_LOWPASS mm1, mm3, mm0, mm4, mm5
+ psllq mm1, 56
+ PALIGNR mm7, mm1, 7, mm3
+ lea r1, [r0+r3*2]
+ pshufw mm0, mm7, 00011011b ; l6 l7 l4 l5 l2 l3 l0 l1
+ psllq mm7, 56 ; l7 .. .. .. .. .. .. ..
+ movq mm2, mm0
+ psllw mm0, 8
+ psrlw mm2, 8
+ por mm2, mm0 ; l7 l6 l5 l4 l3 l2 l1 l0
+ movq mm3, mm2
+ movq mm4, mm2
+ movq mm5, mm2
+ psrlq mm2, 8
+ psrlq mm3, 16
+ lea r2, [r1+r3*2]
+ por mm2, mm7 ; l7 l7 l6 l5 l4 l3 l2 l1
+ punpckhbw mm7, mm7
+ por mm3, mm7 ; l7 l7 l7 l6 l5 l4 l3 l2
+ pavgb mm4, mm2
+ PRED4x4_LOWPASS mm1, mm3, mm5, mm2, mm6
+ movq mm5, mm4
+ punpcklbw mm4, mm1 ; p4 p3 p2 p1
+ punpckhbw mm5, mm1 ; p8 p7 p6 p5
+ movq mm6, mm5
+ movq mm7, mm5
+ movq mm0, mm5
+ PALIGNR mm5, mm4, 2, mm1
+ pshufw mm1, mm6, 11111001b
+ PALIGNR mm6, mm4, 4, mm2
+ pshufw mm2, mm7, 11111110b
+ PALIGNR mm7, mm4, 6, mm3
+ pshufw mm3, mm0, 11111111b
+ movq [r0+r3*1], mm4
+ movq [r0+r3*2], mm5
+ lea r0, [r2+r3*2]
+ movq [r1+r3*1], mm6
+ movq [r1+r3*2], mm7
+ movq [r2+r3*1], mm0
+ movq [r2+r3*2], mm1
+ movq [r0+r3*1], mm2
+ movq [r0+r3*2], mm3
+ RET
+%endmacro
+
+INIT_MMX mmxext
+PRED8x8L_HORIZONTAL_UP
+INIT_MMX ssse3
+PRED8x8L_HORIZONTAL_UP
+
+;-----------------------------------------------------------------------------
+;void pred8x8l_horizontal_down_8(uint8_t *src, int has_topleft, int has_topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred8x8l_horizontal_down_8, 4,5
+ sub r0, r3
+ lea r4, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ punpckhbw mm0, [r0+r3*0-8]
+ movq mm1, [r4+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r4, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r4]
+ mov r0, r4
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ test r1, r1
+ jnz .do_left
+.fix_lt_1:
+ movq mm5, mm3
+ pxor mm5, mm4
+ psrlq mm5, 56
+ psllq mm5, 48
+ pxor mm1, mm5
+ jmp .do_left
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.do_left:
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq mm4, mm0
+ movq mm7, mm2
+ movq mm6, mm2
+ PRED4x4_LOWPASS mm1, mm3, mm0, mm4, mm5
+ psllq mm1, 56
+ PALIGNR mm7, mm1, 7, mm3
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1
+ jz .fix_lt_2
+ test r2, r2
+ jz .fix_tr_1
+.do_top:
+ PRED4x4_LOWPASS mm4, mm2, mm1, mm3, mm5
+ movq mm5, mm4
+ lea r1, [r0+r3*2]
+ psllq mm7, 56
+ movq mm2, mm5
+ movq mm3, mm6
+ movq mm4, mm2
+ PALIGNR mm2, mm6, 7, mm5
+ PALIGNR mm6, mm7, 7, mm0
+ lea r2, [r1+r3*2]
+ PALIGNR mm4, mm3, 1, mm7
+ movq mm5, mm3
+ pavgb mm3, mm6
+ PRED4x4_LOWPASS mm0, mm4, mm6, mm5, mm7
+ movq mm4, mm2
+ movq mm1, mm2
+ lea r4, [r2+r3*2]
+ psrlq mm4, 16
+ psrlq mm1, 8
+ PRED4x4_LOWPASS mm6, mm4, mm2, mm1, mm5
+ movq mm7, mm3
+ punpcklbw mm3, mm0
+ punpckhbw mm7, mm0
+ movq mm1, mm7
+ movq mm0, mm7
+ movq mm4, mm7
+ movq [r4+r3*2], mm3
+ PALIGNR mm7, mm3, 2, mm5
+ movq [r4+r3*1], mm7
+ PALIGNR mm1, mm3, 4, mm5
+ movq [r2+r3*2], mm1
+ PALIGNR mm0, mm3, 6, mm3
+ movq [r2+r3*1], mm0
+ movq mm2, mm6
+ movq mm3, mm6
+ movq [r1+r3*2], mm4
+ PALIGNR mm6, mm4, 2, mm5
+ movq [r1+r3*1], mm6
+ PALIGNR mm2, mm4, 4, mm5
+ movq [r0+r3*2], mm2
+ PALIGNR mm3, mm4, 6, mm4
+ movq [r0+r3*1], mm3
+ RET
+
+%macro PRED8x8L_HORIZONTAL_DOWN 0
+cglobal pred8x8l_horizontal_down_8, 4,5
+ sub r0, r3
+ lea r4, [r0+r3*2]
+ movq mm0, [r0+r3*1-8]
+ punpckhbw mm0, [r0+r3*0-8]
+ movq mm1, [r4+r3*1-8]
+ punpckhbw mm1, [r0+r3*2-8]
+ mov r4, r0
+ punpckhwd mm1, mm0
+ lea r0, [r0+r3*4]
+ movq mm2, [r0+r3*1-8]
+ punpckhbw mm2, [r0+r3*0-8]
+ lea r0, [r0+r3*2]
+ movq mm3, [r0+r3*1-8]
+ punpckhbw mm3, [r0+r3*0-8]
+ punpckhwd mm3, mm2
+ punpckhdq mm3, mm1
+ lea r0, [r0+r3*2]
+ movq mm0, [r0+r3*0-8]
+ movq mm1, [r4]
+ mov r0, r4
+ movq mm4, mm3
+ movq mm2, mm3
+ PALIGNR mm4, mm0, 7, mm0
+ PALIGNR mm1, mm2, 1, mm2
+ test r1, r1
+ jnz .do_left
+.fix_lt_1:
+ movq mm5, mm3
+ pxor mm5, mm4
+ psrlq mm5, 56
+ psllq mm5, 48
+ pxor mm1, mm5
+ jmp .do_left
+.fix_lt_2:
+ movq mm5, mm3
+ pxor mm5, mm2
+ psllq mm5, 56
+ psrlq mm5, 56
+ pxor mm2, mm5
+ test r2, r2
+ jnz .do_top
+.fix_tr_1:
+ movq mm5, mm3
+ pxor mm5, mm1
+ psrlq mm5, 56
+ psllq mm5, 56
+ pxor mm1, mm5
+ jmp .do_top
+.fix_tr_2:
+ punpckhbw mm3, mm3
+ pshufw mm1, mm3, 0xFF
+ jmp .do_topright
+.do_left:
+ movq mm0, mm4
+ PRED4x4_LOWPASS mm2, mm1, mm4, mm3, mm5
+ movq2dq xmm0, mm2
+ pslldq xmm0, 8
+ movq mm4, mm0
+ PRED4x4_LOWPASS mm1, mm3, mm0, mm4, mm5
+ movq2dq xmm2, mm1
+ pslldq xmm2, 15
+ psrldq xmm2, 8
+ por xmm0, xmm2
+ movq mm0, [r0-8]
+ movq mm3, [r0]
+ movq mm1, [r0+8]
+ movq mm2, mm3
+ movq mm4, mm3
+ PALIGNR mm2, mm0, 7, mm0
+ PALIGNR mm1, mm4, 1, mm4
+ test r1, r1
+ jz .fix_lt_2
+ test r2, r2
+ jz .fix_tr_1
+.do_top:
+ PRED4x4_LOWPASS mm4, mm2, mm1, mm3, mm5
+ movq2dq xmm1, mm4
+ test r2, r2
+ jz .fix_tr_2
+ movq mm0, [r0+8]
+ movq mm5, mm0
+ movq mm2, mm0
+ movq mm4, mm0
+ psrlq mm5, 56
+ PALIGNR mm2, mm3, 7, mm3
+ PALIGNR mm5, mm4, 1, mm4
+ PRED4x4_LOWPASS mm1, mm2, mm5, mm0, mm4
+.do_topright:
+ movq2dq xmm5, mm1
+ pslldq xmm5, 8
+ por xmm1, xmm5
+INIT_XMM cpuname
+ lea r2, [r4+r3*2]
+ movdqa xmm2, xmm1
+ movdqa xmm3, xmm1
+ PALIGNR xmm1, xmm0, 7, xmm4
+ PALIGNR xmm2, xmm0, 9, xmm5
+ lea r1, [r2+r3*2]
+ PALIGNR xmm3, xmm0, 8, xmm0
+ movdqa xmm4, xmm1
+ pavgb xmm4, xmm3
+ lea r0, [r1+r3*2]
+ PRED4x4_LOWPASS xmm0, xmm1, xmm2, xmm3, xmm5
+ punpcklbw xmm4, xmm0
+ movhlps xmm0, xmm4
+ movq [r0+r3*2], xmm4
+ movq [r2+r3*2], xmm0
+ psrldq xmm4, 2
+ psrldq xmm0, 2
+ movq [r0+r3*1], xmm4
+ movq [r2+r3*1], xmm0
+ psrldq xmm4, 2
+ psrldq xmm0, 2
+ movq [r1+r3*2], xmm4
+ movq [r4+r3*2], xmm0
+ psrldq xmm4, 2
+ psrldq xmm0, 2
+ movq [r1+r3*1], xmm4
+ movq [r4+r3*1], xmm0
+ RET
+%endmacro
+
+INIT_MMX sse2
+PRED8x8L_HORIZONTAL_DOWN
+INIT_MMX ssse3
+PRED8x8L_HORIZONTAL_DOWN
+
+;-----------------------------------------------------------------------------
+; void pred4x4_dc_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred4x4_dc_8, 3,5
pxor mm7, mm7
mov r4, r0
sub r0, r2
RET
;-----------------------------------------------------------------------------
-; void pred4x4_tm_vp8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+; void pred4x4_tm_vp8_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
;-----------------------------------------------------------------------------
-%macro PRED4x4_TM_MMX 1
-cglobal pred4x4_tm_vp8_%1, 3,6
+%macro PRED4x4_TM 0
+cglobal pred4x4_tm_vp8_8, 3,6
sub r0, r2
pxor mm7, mm7
movd mm0, [r0]
sub r3d, r4d
movd mm2, r1d
movd mm4, r3d
-%ifidn %1, mmx
+%if cpuflag(mmxext)
+ pshufw mm2, mm2, 0
+ pshufw mm4, mm4, 0
+%else
punpcklwd mm2, mm2
punpcklwd mm4, mm4
punpckldq mm2, mm2
punpckldq mm4, mm4
-%else
- pshufw mm2, mm2, 0
- pshufw mm4, mm4, 0
%endif
paddw mm2, mm0
paddw mm4, mm0
REP_RET
%endmacro
-PRED4x4_TM_MMX mmx
-PRED4x4_TM_MMX mmxext
+INIT_MMX mmx
+PRED4x4_TM
+INIT_MMX mmxext
+PRED4x4_TM
-cglobal pred4x4_tm_vp8_ssse3, 3,3
+INIT_XMM ssse3
+cglobal pred4x4_tm_vp8_8, 3,3
sub r0, r2
movq mm6, [tm_shuf]
pxor mm1, mm1
movd [r1+r2*2], mm5
RET
-; dest, left, right, src, tmp
-; output: %1 = (t[n-1] + t[n]*2 + t[n+1] + 2) >> 2
-%macro PRED4x4_LOWPASS 5
- mova %5, %2
- pavgb %2, %3
- pxor %3, %5
- mova %1, %4
- pand %3, [pb_1]
- psubusb %2, %3
- pavgb %1, %2
-%endmacro
-
;-----------------------------------------------------------------------------
-; void pred4x4_vertical_vp8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+; void pred4x4_vertical_vp8_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
;-----------------------------------------------------------------------------
-INIT_MMX
-cglobal pred4x4_vertical_vp8_mmxext, 3,3
+INIT_MMX mmxext
+cglobal pred4x4_vertical_vp8_8, 3,3
sub r0, r2
movd m1, [r0-1]
movd m0, [r0]
RET
;-----------------------------------------------------------------------------
-; void pred4x4_down_left_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+; void pred4x4_down_left_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
;-----------------------------------------------------------------------------
-%ifdef CONFIG_GPL
-INIT_MMX
-cglobal pred4x4_down_left_mmxext, 3,3
+INIT_MMX mmxext
+cglobal pred4x4_down_left_8, 3,3
sub r0, r2
movq m1, [r0]
punpckldq m1, [r1]
movq m2, m1
movq m3, m1
- movq m4, m1
psllq m1, 8
pxor m2, m1
psrlq m2, 8
- pxor m3, m2
- PRED4x4_LOWPASS m0, m1, m3, m4, m5
+ pxor m2, m3
+ PRED4x4_LOWPASS m0, m1, m2, m3, m4
lea r1, [r0+r2*2]
psrlq m0, 8
movd [r0+r2*1], m0
psrlq m0, 8
movd [r1+r2*2], m0
RET
-%endif
+
+;-----------------------------------------------------------------------------
+; void pred4x4_vertical_left_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred4x4_vertical_left_8, 3,3
+ sub r0, r2
+ movq m1, [r0]
+ punpckldq m1, [r1]
+ movq m3, m1
+ movq m2, m1
+ psrlq m3, 8
+ psrlq m2, 16
+ movq m4, m3
+ pavgb m4, m1
+ PRED4x4_LOWPASS m0, m1, m2, m3, m5
+ lea r1, [r0+r2*2]
+ movh [r0+r2*1], m4
+ movh [r0+r2*2], m0
+ psrlq m4, 8
+ psrlq m0, 8
+ movh [r1+r2*1], m4
+ movh [r1+r2*2], m0
+ RET
+
+;-----------------------------------------------------------------------------
+; void pred4x4_horizontal_up_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred4x4_horizontal_up_8, 3,3
+ sub r0, r2
+ lea r1, [r0+r2*2]
+ movd m0, [r0+r2*1-4]
+ punpcklbw m0, [r0+r2*2-4]
+ movd m1, [r1+r2*1-4]
+ punpcklbw m1, [r1+r2*2-4]
+ punpckhwd m0, m1
+ movq m1, m0
+ punpckhbw m1, m1
+ pshufw m1, m1, 0xFF
+ punpckhdq m0, m1
+ movq m2, m0
+ movq m3, m0
+ movq m7, m0
+ psrlq m2, 16
+ psrlq m3, 8
+ pavgb m7, m3
+ PRED4x4_LOWPASS m4, m0, m2, m3, m5
+ punpcklbw m7, m4
+ movd [r0+r2*1], m7
+ psrlq m7, 16
+ movd [r0+r2*2], m7
+ psrlq m7, 16
+ movd [r1+r2*1], m7
+ movd [r1+r2*2], m1
+ RET
+
+;-----------------------------------------------------------------------------
+; void pred4x4_horizontal_down_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred4x4_horizontal_down_8, 3,3
+ sub r0, r2
+ lea r1, [r0+r2*2]
+ movh m0, [r0-4] ; lt ..
+ punpckldq m0, [r0] ; t3 t2 t1 t0 lt .. .. ..
+ psllq m0, 8 ; t2 t1 t0 lt .. .. .. ..
+ movd m1, [r1+r2*2-4] ; l3
+ punpcklbw m1, [r1+r2*1-4] ; l2 l3
+ movd m2, [r0+r2*2-4] ; l1
+ punpcklbw m2, [r0+r2*1-4] ; l0 l1
+ punpckhwd m1, m2 ; l0 l1 l2 l3
+ punpckhdq m1, m0 ; t2 t1 t0 lt l0 l1 l2 l3
+ movq m0, m1
+ movq m2, m1
+ movq m5, m1
+ psrlq m0, 16 ; .. .. t2 t1 t0 lt l0 l1
+ psrlq m2, 8 ; .. t2 t1 t0 lt l0 l1 l2
+ pavgb m5, m2
+ PRED4x4_LOWPASS m3, m1, m0, m2, m4
+ punpcklbw m5, m3
+ psrlq m3, 32
+ PALIGNR m3, m5, 6, m4
+ movh [r1+r2*2], m5
+ psrlq m5, 16
+ movh [r1+r2*1], m5
+ psrlq m5, 16
+ movh [r0+r2*2], m5
+ movh [r0+r2*1], m3
+ RET
+
+;-----------------------------------------------------------------------------
+; void pred4x4_vertical_right_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred4x4_vertical_right_8, 3,3
+ sub r0, r2
+ lea r1, [r0+r2*2]
+ movh m0, [r0] ; ........t3t2t1t0
+ movq m5, m0
+ PALIGNR m0, [r0-8], 7, m1 ; ......t3t2t1t0lt
+ pavgb m5, m0
+ PALIGNR m0, [r0+r2*1-8], 7, m1 ; ....t3t2t1t0ltl0
+ movq m1, m0
+ PALIGNR m0, [r0+r2*2-8], 7, m2 ; ..t3t2t1t0ltl0l1
+ movq m2, m0
+ PALIGNR m0, [r1+r2*1-8], 7, m3 ; t3t2t1t0ltl0l1l2
+ PRED4x4_LOWPASS m3, m1, m0, m2, m4
+ movq m1, m3
+ psrlq m3, 16
+ psllq m1, 48
+ movh [r0+r2*1], m5
+ movh [r0+r2*2], m3
+ PALIGNR m5, m1, 7, m2
+ psllq m1, 8
+ movh [r1+r2*1], m5
+ PALIGNR m3, m1, 7, m1
+ movh [r1+r2*2], m3
+ RET
+
+;-----------------------------------------------------------------------------
+; void pred4x4_down_right_8_mmxext(uint8_t *src, const uint8_t *topright, int stride)
+;-----------------------------------------------------------------------------
+
+INIT_MMX mmxext
+cglobal pred4x4_down_right_8, 3,3
+ sub r0, r2
+ lea r1, [r0+r2*2]
+ movq m1, [r1-8]
+ movq m2, [r0+r2*1-8]
+ punpckhbw m2, [r0-8]
+ movh m3, [r0]
+ punpckhwd m1, m2
+ PALIGNR m3, m1, 5, m1
+ movq m1, m3
+ PALIGNR m3, [r1+r2*1-8], 7, m4
+ movq m2, m3
+ PALIGNR m3, [r1+r2*2-8], 7, m4
+ PRED4x4_LOWPASS m0, m3, m1, m2, m4
+ movh [r1+r2*2], m0
+ psrlq m0, 8
+ movh [r1+r2*1], m0
+ psrlq m0, 8
+ movh [r0+r2*2], m0
+ psrlq m0, 8
+ movh [r0+r2*1], m0
+ RET
projects / ffmpeg / blobdiff