gecko-dev/media/libjpeg/simd/jfdctflt-sse-64.asm
Ryan VanderMeulen 50bc583645 Bug 1061240 - Update libjpeg-turbo to version 1.4. r=jmuizelaar, r=glandium, f=rankov
--HG--
rename : media/libjpeg/config.h => media/libjpeg/jconfigint.h
rename : media/libjpeg/simd/jcclrmmx.asm => media/libjpeg/simd/jccolext-mmx.asm
rename : media/libjpeg/simd/jcclrss2-64.asm => media/libjpeg/simd/jccolext-sse2-64.asm
rename : media/libjpeg/simd/jcclrss2.asm => media/libjpeg/simd/jccolext-sse2.asm
rename : media/libjpeg/simd/jccolmmx.asm => media/libjpeg/simd/jccolor-mmx.asm
rename : media/libjpeg/simd/jccolss2-64.asm => media/libjpeg/simd/jccolor-sse2-64.asm
rename : media/libjpeg/simd/jccolss2.asm => media/libjpeg/simd/jccolor-sse2.asm
rename : media/libjpeg/simd/jcgrammx.asm => media/libjpeg/simd/jcgray-mmx.asm
rename : media/libjpeg/simd/jcgrass2-64.asm => media/libjpeg/simd/jcgray-sse2-64.asm
rename : media/libjpeg/simd/jcgrass2.asm => media/libjpeg/simd/jcgray-sse2.asm
rename : media/libjpeg/simd/jcgrymmx.asm => media/libjpeg/simd/jcgryext-mmx.asm
rename : media/libjpeg/simd/jcgryss2-64.asm => media/libjpeg/simd/jcgryext-sse2-64.asm
rename : media/libjpeg/simd/jcgryss2.asm => media/libjpeg/simd/jcgryext-sse2.asm
rename : media/libjpeg/simd/jcsammmx.asm => media/libjpeg/simd/jcsample-mmx.asm
rename : media/libjpeg/simd/jcsamss2-64.asm => media/libjpeg/simd/jcsample-sse2-64.asm
rename : media/libjpeg/simd/jcsamss2.asm => media/libjpeg/simd/jcsample-sse2.asm
rename : media/libjpeg/simd/jdclrmmx.asm => media/libjpeg/simd/jdcolext-mmx.asm
rename : media/libjpeg/simd/jdclrss2-64.asm => media/libjpeg/simd/jdcolext-sse2-64.asm
rename : media/libjpeg/simd/jdclrss2.asm => media/libjpeg/simd/jdcolext-sse2.asm
rename : media/libjpeg/simd/jdcolmmx.asm => media/libjpeg/simd/jdcolor-mmx.asm
rename : media/libjpeg/simd/jdcolss2-64.asm => media/libjpeg/simd/jdcolor-sse2-64.asm
rename : media/libjpeg/simd/jdcolss2.asm => media/libjpeg/simd/jdcolor-sse2.asm
rename : media/libjpeg/simd/jdmermmx.asm => media/libjpeg/simd/jdmerge-mmx.asm
rename : media/libjpeg/simd/jdmerss2-64.asm => media/libjpeg/simd/jdmerge-sse2-64.asm
rename : media/libjpeg/simd/jdmerss2.asm => media/libjpeg/simd/jdmerge-sse2.asm
rename : media/libjpeg/simd/jdmrgmmx.asm => media/libjpeg/simd/jdmrgext-mmx.asm
rename : media/libjpeg/simd/jdmrgss2-64.asm => media/libjpeg/simd/jdmrgext-sse2-64.asm
rename : media/libjpeg/simd/jdmrgss2.asm => media/libjpeg/simd/jdmrgext-sse2.asm
rename : media/libjpeg/simd/jdsammmx.asm => media/libjpeg/simd/jdsample-mmx.asm
rename : media/libjpeg/simd/jdsamss2-64.asm => media/libjpeg/simd/jdsample-sse2-64.asm
rename : media/libjpeg/simd/jdsamss2.asm => media/libjpeg/simd/jdsample-sse2.asm
rename : media/libjpeg/simd/jf3dnflt.asm => media/libjpeg/simd/jfdctflt-3dn.asm
rename : media/libjpeg/simd/jfsseflt-64.asm => media/libjpeg/simd/jfdctflt-sse-64.asm
rename : media/libjpeg/simd/jfsseflt.asm => media/libjpeg/simd/jfdctflt-sse.asm
rename : media/libjpeg/simd/jfmmxfst.asm => media/libjpeg/simd/jfdctfst-mmx.asm
rename : media/libjpeg/simd/jfss2fst-64.asm => media/libjpeg/simd/jfdctfst-sse2-64.asm
rename : media/libjpeg/simd/jfss2fst.asm => media/libjpeg/simd/jfdctfst-sse2.asm
rename : media/libjpeg/simd/jfmmxint.asm => media/libjpeg/simd/jfdctint-mmx.asm
rename : media/libjpeg/simd/jfss2int-64.asm => media/libjpeg/simd/jfdctint-sse2-64.asm
rename : media/libjpeg/simd/jfss2int.asm => media/libjpeg/simd/jfdctint-sse2.asm
rename : media/libjpeg/simd/ji3dnflt.asm => media/libjpeg/simd/jidctflt-3dn.asm
rename : media/libjpeg/simd/jisseflt.asm => media/libjpeg/simd/jidctflt-sse.asm
rename : media/libjpeg/simd/jiss2flt-64.asm => media/libjpeg/simd/jidctflt-sse2-64.asm
rename : media/libjpeg/simd/jiss2flt.asm => media/libjpeg/simd/jidctflt-sse2.asm
rename : media/libjpeg/simd/jimmxfst.asm => media/libjpeg/simd/jidctfst-mmx.asm
rename : media/libjpeg/simd/jiss2fst-64.asm => media/libjpeg/simd/jidctfst-sse2-64.asm
rename : media/libjpeg/simd/jiss2fst.asm => media/libjpeg/simd/jidctfst-sse2.asm
rename : media/libjpeg/simd/jimmxint.asm => media/libjpeg/simd/jidctint-mmx.asm
rename : media/libjpeg/simd/jiss2int-64.asm => media/libjpeg/simd/jidctint-sse2-64.asm
rename : media/libjpeg/simd/jiss2int.asm => media/libjpeg/simd/jidctint-sse2.asm
rename : media/libjpeg/simd/jimmxred.asm => media/libjpeg/simd/jidctred-mmx.asm
rename : media/libjpeg/simd/jiss2red-64.asm => media/libjpeg/simd/jidctred-sse2-64.asm
rename : media/libjpeg/simd/jiss2red.asm => media/libjpeg/simd/jidctred-sse2.asm
rename : media/libjpeg/simd/jcqnt3dn.asm => media/libjpeg/simd/jquant-3dn.asm
rename : media/libjpeg/simd/jcqntmmx.asm => media/libjpeg/simd/jquant-mmx.asm
rename : media/libjpeg/simd/jcqntsse.asm => media/libjpeg/simd/jquant-sse.asm
rename : media/libjpeg/simd/jcqnts2f-64.asm => media/libjpeg/simd/jquantf-sse2-64.asm
rename : media/libjpeg/simd/jcqnts2f.asm => media/libjpeg/simd/jquantf-sse2.asm
rename : media/libjpeg/simd/jcqnts2i-64.asm => media/libjpeg/simd/jquanti-sse2-64.asm
rename : media/libjpeg/simd/jcqnts2i.asm => media/libjpeg/simd/jquanti-sse2.asm
2015-01-20 22:19:57 -05:00

359 lines
14 KiB
NASM

;
; jfdctflt.asm - floating-point FDCT (64-bit SSE)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright 2009 D. R. Commander
;
; Based on
; x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
; For conditions of distribution and use, see copyright notice in jsimdext.inc
;
; This file should be assembled with NASM (Netwide Assembler),
; can *not* be assembled with Microsoft's MASM or any compatible
; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; This file contains a floating-point implementation of the forward DCT
; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
;
; [TAB8]
%include "jsimdext.inc"
%include "jdct.inc"
; --------------------------------------------------------------------------
%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
shufps %1,%2,0x44
%endmacro
%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
shufps %1,%2,0xEE
%endmacro
; --------------------------------------------------------------------------
SECTION SEG_CONST
alignz 16
global EXTN(jconst_fdct_float_sse)
EXTN(jconst_fdct_float_sse):
PD_0_382 times 4 dd 0.382683432365089771728460
PD_0_707 times 4 dd 0.707106781186547524400844
PD_0_541 times 4 dd 0.541196100146196984399723
PD_1_306 times 4 dd 1.306562964876376527856643
alignz 16
; --------------------------------------------------------------------------
SECTION SEG_TEXT
BITS 64
;
; Perform the forward DCT on one block of samples.
;
; GLOBAL(void)
; jsimd_fdct_float_sse (FAST_FLOAT * data)
;
; r10 = FAST_FLOAT * data
%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
%define WK_NUM 2
align 16
global EXTN(jsimd_fdct_float_sse)
EXTN(jsimd_fdct_float_sse):
push rbp
mov rax,rsp ; rax = original rbp
sub rsp, byte 4
and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
mov [rsp],rax
mov rbp,rsp ; rbp = aligned rbp
lea rsp, [wk(0)]
collect_args
; ---- Pass 1: process rows.
mov rdx, r10 ; (FAST_FLOAT *)
mov rcx, DCTSIZE/4
.rowloop:
movaps xmm0, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm2, XMMWORD [XMMBLOCK(2,1,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm3, XMMWORD [XMMBLOCK(3,1,rdx,SIZEOF_FAST_FLOAT)]
; xmm0=(20 21 22 23), xmm2=(24 25 26 27)
; xmm1=(30 31 32 33), xmm3=(34 35 36 37)
movaps xmm4,xmm0 ; transpose coefficients(phase 1)
unpcklps xmm0,xmm1 ; xmm0=(20 30 21 31)
unpckhps xmm4,xmm1 ; xmm4=(22 32 23 33)
movaps xmm5,xmm2 ; transpose coefficients(phase 1)
unpcklps xmm2,xmm3 ; xmm2=(24 34 25 35)
unpckhps xmm5,xmm3 ; xmm5=(26 36 27 37)
movaps xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm7, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)]
; xmm6=(00 01 02 03), xmm1=(04 05 06 07)
; xmm7=(10 11 12 13), xmm3=(14 15 16 17)
movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 32 23 33)
movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(24 34 25 35)
movaps xmm4,xmm6 ; transpose coefficients(phase 1)
unpcklps xmm6,xmm7 ; xmm6=(00 10 01 11)
unpckhps xmm4,xmm7 ; xmm4=(02 12 03 13)
movaps xmm2,xmm1 ; transpose coefficients(phase 1)
unpcklps xmm1,xmm3 ; xmm1=(04 14 05 15)
unpckhps xmm2,xmm3 ; xmm2=(06 16 07 17)
movaps xmm7,xmm6 ; transpose coefficients(phase 2)
unpcklps2 xmm6,xmm0 ; xmm6=(00 10 20 30)=data0
unpckhps2 xmm7,xmm0 ; xmm7=(01 11 21 31)=data1
movaps xmm3,xmm2 ; transpose coefficients(phase 2)
unpcklps2 xmm2,xmm5 ; xmm2=(06 16 26 36)=data6
unpckhps2 xmm3,xmm5 ; xmm3=(07 17 27 37)=data7
movaps xmm0,xmm7
movaps xmm5,xmm6
subps xmm7,xmm2 ; xmm7=data1-data6=tmp6
subps xmm6,xmm3 ; xmm6=data0-data7=tmp7
addps xmm0,xmm2 ; xmm0=data1+data6=tmp1
addps xmm5,xmm3 ; xmm5=data0+data7=tmp0
movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 32 23 33)
movaps xmm3, XMMWORD [wk(1)] ; xmm3=(24 34 25 35)
movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
movaps xmm7,xmm4 ; transpose coefficients(phase 2)
unpcklps2 xmm4,xmm2 ; xmm4=(02 12 22 32)=data2
unpckhps2 xmm7,xmm2 ; xmm7=(03 13 23 33)=data3
movaps xmm6,xmm1 ; transpose coefficients(phase 2)
unpcklps2 xmm1,xmm3 ; xmm1=(04 14 24 34)=data4
unpckhps2 xmm6,xmm3 ; xmm6=(05 15 25 35)=data5
movaps xmm2,xmm7
movaps xmm3,xmm4
addps xmm7,xmm1 ; xmm7=data3+data4=tmp3
addps xmm4,xmm6 ; xmm4=data2+data5=tmp2
subps xmm2,xmm1 ; xmm2=data3-data4=tmp4
subps xmm3,xmm6 ; xmm3=data2-data5=tmp5
; -- Even part
movaps xmm1,xmm5
movaps xmm6,xmm0
subps xmm5,xmm7 ; xmm5=tmp13
subps xmm0,xmm4 ; xmm0=tmp12
addps xmm1,xmm7 ; xmm1=tmp10
addps xmm6,xmm4 ; xmm6=tmp11
addps xmm0,xmm5
mulps xmm0,[rel PD_0_707] ; xmm0=z1
movaps xmm7,xmm1
movaps xmm4,xmm5
subps xmm1,xmm6 ; xmm1=data4
subps xmm5,xmm0 ; xmm5=data6
addps xmm7,xmm6 ; xmm7=data0
addps xmm4,xmm0 ; xmm4=data2
movaps XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)], xmm1
movaps XMMWORD [XMMBLOCK(2,1,rdx,SIZEOF_FAST_FLOAT)], xmm5
movaps XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
movaps XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
; -- Odd part
movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
addps xmm2,xmm3 ; xmm2=tmp10
addps xmm3,xmm6 ; xmm3=tmp11
addps xmm6,xmm0 ; xmm6=tmp12, xmm0=tmp7
mulps xmm3,[rel PD_0_707] ; xmm3=z3
movaps xmm1,xmm2 ; xmm1=tmp10
subps xmm2,xmm6
mulps xmm2,[rel PD_0_382] ; xmm2=z5
mulps xmm1,[rel PD_0_541] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
mulps xmm6,[rel PD_1_306] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
addps xmm1,xmm2 ; xmm1=z2
addps xmm6,xmm2 ; xmm6=z4
movaps xmm5,xmm0
subps xmm0,xmm3 ; xmm0=z13
addps xmm5,xmm3 ; xmm5=z11
movaps xmm7,xmm0
movaps xmm4,xmm5
subps xmm0,xmm1 ; xmm0=data3
subps xmm5,xmm6 ; xmm5=data7
addps xmm7,xmm1 ; xmm7=data5
addps xmm4,xmm6 ; xmm4=data1
movaps XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)], xmm0
movaps XMMWORD [XMMBLOCK(3,1,rdx,SIZEOF_FAST_FLOAT)], xmm5
movaps XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)], xmm7
movaps XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
add rdx, 4*DCTSIZE*SIZEOF_FAST_FLOAT
dec rcx
jnz near .rowloop
; ---- Pass 2: process columns.
mov rdx, r10 ; (FAST_FLOAT *)
mov rcx, DCTSIZE/4
.columnloop:
movaps xmm0, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm2, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FAST_FLOAT)]
; xmm0=(02 12 22 32), xmm2=(42 52 62 72)
; xmm1=(03 13 23 33), xmm3=(43 53 63 73)
movaps xmm4,xmm0 ; transpose coefficients(phase 1)
unpcklps xmm0,xmm1 ; xmm0=(02 03 12 13)
unpckhps xmm4,xmm1 ; xmm4=(22 23 32 33)
movaps xmm5,xmm2 ; transpose coefficients(phase 1)
unpcklps xmm2,xmm3 ; xmm2=(42 43 52 53)
unpckhps xmm5,xmm3 ; xmm5=(62 63 72 73)
movaps xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm7, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm1, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm3, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FAST_FLOAT)]
; xmm6=(00 10 20 30), xmm1=(40 50 60 70)
; xmm7=(01 11 21 31), xmm3=(41 51 61 71)
movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 23 32 33)
movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(42 43 52 53)
movaps xmm4,xmm6 ; transpose coefficients(phase 1)
unpcklps xmm6,xmm7 ; xmm6=(00 01 10 11)
unpckhps xmm4,xmm7 ; xmm4=(20 21 30 31)
movaps xmm2,xmm1 ; transpose coefficients(phase 1)
unpcklps xmm1,xmm3 ; xmm1=(40 41 50 51)
unpckhps xmm2,xmm3 ; xmm2=(60 61 70 71)
movaps xmm7,xmm6 ; transpose coefficients(phase 2)
unpcklps2 xmm6,xmm0 ; xmm6=(00 01 02 03)=data0
unpckhps2 xmm7,xmm0 ; xmm7=(10 11 12 13)=data1
movaps xmm3,xmm2 ; transpose coefficients(phase 2)
unpcklps2 xmm2,xmm5 ; xmm2=(60 61 62 63)=data6
unpckhps2 xmm3,xmm5 ; xmm3=(70 71 72 73)=data7
movaps xmm0,xmm7
movaps xmm5,xmm6
subps xmm7,xmm2 ; xmm7=data1-data6=tmp6
subps xmm6,xmm3 ; xmm6=data0-data7=tmp7
addps xmm0,xmm2 ; xmm0=data1+data6=tmp1
addps xmm5,xmm3 ; xmm5=data0+data7=tmp0
movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 23 32 33)
movaps xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53)
movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6
movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
movaps xmm7,xmm4 ; transpose coefficients(phase 2)
unpcklps2 xmm4,xmm2 ; xmm4=(20 21 22 23)=data2
unpckhps2 xmm7,xmm2 ; xmm7=(30 31 32 33)=data3
movaps xmm6,xmm1 ; transpose coefficients(phase 2)
unpcklps2 xmm1,xmm3 ; xmm1=(40 41 42 43)=data4
unpckhps2 xmm6,xmm3 ; xmm6=(50 51 52 53)=data5
movaps xmm2,xmm7
movaps xmm3,xmm4
addps xmm7,xmm1 ; xmm7=data3+data4=tmp3
addps xmm4,xmm6 ; xmm4=data2+data5=tmp2
subps xmm2,xmm1 ; xmm2=data3-data4=tmp4
subps xmm3,xmm6 ; xmm3=data2-data5=tmp5
; -- Even part
movaps xmm1,xmm5
movaps xmm6,xmm0
subps xmm5,xmm7 ; xmm5=tmp13
subps xmm0,xmm4 ; xmm0=tmp12
addps xmm1,xmm7 ; xmm1=tmp10
addps xmm6,xmm4 ; xmm6=tmp11
addps xmm0,xmm5
mulps xmm0,[rel PD_0_707] ; xmm0=z1
movaps xmm7,xmm1
movaps xmm4,xmm5
subps xmm1,xmm6 ; xmm1=data4
subps xmm5,xmm0 ; xmm5=data6
addps xmm7,xmm6 ; xmm7=data0
addps xmm4,xmm0 ; xmm4=data2
movaps XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FAST_FLOAT)], xmm1
movaps XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FAST_FLOAT)], xmm5
movaps XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
movaps XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
; -- Odd part
movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6
movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7
addps xmm2,xmm3 ; xmm2=tmp10
addps xmm3,xmm6 ; xmm3=tmp11
addps xmm6,xmm0 ; xmm6=tmp12, xmm0=tmp7
mulps xmm3,[rel PD_0_707] ; xmm3=z3
movaps xmm1,xmm2 ; xmm1=tmp10
subps xmm2,xmm6
mulps xmm2,[rel PD_0_382] ; xmm2=z5
mulps xmm1,[rel PD_0_541] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
mulps xmm6,[rel PD_1_306] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
addps xmm1,xmm2 ; xmm1=z2
addps xmm6,xmm2 ; xmm6=z4
movaps xmm5,xmm0
subps xmm0,xmm3 ; xmm0=z13
addps xmm5,xmm3 ; xmm5=z11
movaps xmm7,xmm0
movaps xmm4,xmm5
subps xmm0,xmm1 ; xmm0=data3
subps xmm5,xmm6 ; xmm5=data7
addps xmm7,xmm1 ; xmm7=data5
addps xmm4,xmm6 ; xmm4=data1
movaps XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)], xmm0
movaps XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FAST_FLOAT)], xmm5
movaps XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FAST_FLOAT)], xmm7
movaps XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)], xmm4
add rdx, byte 4*SIZEOF_FAST_FLOAT
dec rcx
jnz near .columnloop
uncollect_args
mov rsp,rbp ; rsp <- aligned rbp
pop rsp ; rsp <- original rbp
pop rbp
ret
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
align 16