claunia/cuetools.net
1
0
Fork 0
mirror of https://github.com/claunia/cuetools.net.git synced 2025-12-16 18:14:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

playing with SSE2

Browse Source
This commit is contained in:
chudov
2008-10-14 00:19:30 +00:00
parent 30a26ae9a8
commit f1f8e8e308
5 changed files with 35 additions and 327 deletions
Download Patch File Download Diff File Expand all files Collapse all files

141
MAC_SDK/Source/MACLib/Assembly/Assembly64.nas
View File

@@ -95,10 +95,6 @@ endproc
;
; int CalculateDotProduct ( const short* pA, const short* pB, int nOrder )
;
; [esp+12] nOrder
; [esp+ 8] pB
; [esp+ 4] pA
; [esp+ 0] Return Address
align 16
nop
@@ -118,90 +114,48 @@ endproc
proc CalculateDotProduct
shr r8d, 4
pxor mm7, mm7
loopDot: movq mm0, [rcx] ;pA
pmaddwd mm0, [rdx] ;pB
paddd mm7, mm0
movq mm1, [rcx + 8]
pmaddwd mm1, [rdx + 8]
paddd mm7, mm1
movq mm2, [rcx + 16]
pmaddwd mm2, [rdx + 16]
paddd mm7, mm2
movq mm3, [rcx + 24]
pmaddwd mm3, [rdx + 24]
add rcx, byte 32
add rdx, byte 32
paddd mm7, mm3
dec r8d
jnz loopDot
movq mm6, mm7
psrlq mm7, 32
paddd mm6, mm7
movd eax, mm6
emms
endproc
;
; int CalculateDotProduct ( const short* pA, const short* pB, int nOrder )
;
; [esp+12] nOrder
; [esp+ 8] pB
; [esp+ 4] pA
; [esp+ 0] Return Address
align 16
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
nop
proc CalculateDotProductXMM
shr r8d, 4
pxor xmm7, xmm7
mov r9d, r8d
and r9d, 1
shr r8d, 1
loopDotXMM: movdqu xmm0, [rcx] ;pA
LoopDotEven:
movdqu xmm0, [rcx] ;pA
pmaddwd xmm0, [rdx] ;pB
paddd xmm7, xmm0
movdqu xmm1, [rcx + 16]
pmaddwd xmm1, [rdx + 16]
; paddd xmm7, xmm1
; movq xmm2, [rcx + 32]
; pmaddwd xmm2, [rdx + 32]
; paddd xmm7, mm2
; movq xmm3, [rcx + 48]
; pmaddwd xmm3, [rdx + 48]
add rcx, byte 32
add rdx, byte 32
; paddd xmm7, xmm3
movdqu xmm1, [rcx + 16]
pmaddwd xmm1, [rdx + 16]
paddd xmm7, xmm1
movdqu xmm2, [rcx + 32]
pmaddwd xmm2, [rdx + 32]
paddd xmm7, xmm2
movdqu xmm3, [rcx + 48]
pmaddwd xmm3, [rdx + 48]
add rcx, byte 64
add rdx, byte 64
paddd xmm7, xmm3
dec r8d
jnz loopDotXMM
jnz short LoopDotEven
movq xmm5, xmm7
psrldq xmm5, 16
movq xmm4, xmm5
psrlq xmm5, 32
movq xmm6, xmm7
psrlq xmm7, 32
paddd xmm6, xmm4
paddd xmm6, xmm5
paddd xmm6, xmm7
movd eax, xmm6
cmp r9d, byte 0
je DotFinal
movdqu xmm0, [rcx] ;pA
pmaddwd xmm0, [rdx] ;pB
paddd xmm7, xmm0
movdqu xmm1, [rcx + 16]
pmaddwd xmm1, [rdx + 16]
paddd xmm7, xmm1
DotFinal:
movdqa xmm6, xmm7
psrldq xmm6, 8
paddd xmm7, xmm6
movdqa xmm6, xmm7
psrldq xmm6, 4
paddd xmm7, xmm6
movd eax, xmm7
emms
endproc
@@ -211,30 +165,5 @@ endproc
;
proc GetMMXAvailable
push rax
push rcx
push rdx
push rbx
pushfq
pop rax
mov rcx, rax
xor rax, 0x200000
push rax
popfq
pushfq
pop rax
cmp rax, rcx
jz short return ; no CPUID command, so no MMX
mov rax,1
CPUID
test rdx,0x800000
return: pop rbx
pop rdx
pop rcx
pop rax
setnz al
and eax, byte 1
mov eax, 1
endproc
; end