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

SSE3 optimization

Browse Source
This commit is contained in:
chudov
2008-10-14 17:13:34 +00:00
parent c585467f70
commit 05074774ea
1 changed files with 25 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
MAC_SDK/Source/MACLib/Assembly/Assembly64.nas
View File

@@ -29,19 +29,19 @@ AdaptAddLoop0%1:
je short AdaptAddLoopULast%1 je short AdaptAddLoopULast%1
AdaptAddLoopU%1: AdaptAddLoopU%1:
movdqa xmm0, [rcx] movdqa xmm0, [rcx]
movdqu xmm1, [rdx] lddqu xmm1, [rdx]
%1 xmm0, xmm1 %1 xmm0, xmm1
movdqa [rcx], xmm0 movdqa [rcx], xmm0
movdqa xmm2, [rcx + 16] movdqa xmm2, [rcx + 16]
movdqu xmm3, [rdx + 16] lddqu xmm3, [rdx + 16]
%1 xmm2, xmm3 %1 xmm2, xmm3
movdqa [rcx + 16], xmm2 movdqa [rcx + 16], xmm2
movdqa xmm4, [rcx+32] movdqa xmm4, [rcx+32]
movdqu xmm5, [rdx+32] lddqu xmm5, [rdx+32]
%1 xmm4, xmm5 %1 xmm4, xmm5
movdqa [rcx+32], xmm4 movdqa [rcx+32], xmm4
movdqa xmm6, [rcx + 48] movdqa xmm6, [rcx + 48]
movdqu xmm7, [rdx + 48] lddqu xmm7, [rdx + 48]
%1 xmm6, xmm7 %1 xmm6, xmm7
movdqa [rcx + 48], xmm6 movdqa [rcx + 48], xmm6
add rcx, byte 64 add rcx, byte 64
@@ -52,11 +52,11 @@ AdaptAddLoopULast%1:
cmp r8d, byte 0 cmp r8d, byte 0
je short AdaptAddLoopUEnd%1 je short AdaptAddLoopUEnd%1
movdqa xmm0, [rcx] movdqa xmm0, [rcx]
movdqu xmm1, [rdx] lddqu xmm1, [rdx]
%1 xmm0, xmm1 %1 xmm0, xmm1
movdqa [rcx], xmm0 movdqa [rcx], xmm0
movdqa xmm2, [rcx + 16] movdqa xmm2, [rcx + 16]
movdqu xmm3, [rdx + 16] lddqu xmm3, [rdx + 16]
%1 xmm2, xmm3 %1 xmm2, xmm3
movdqa [rcx + 16], xmm2 movdqa [rcx + 16], xmm2
AdaptAddLoopUEnd%1: AdaptAddLoopUEnd%1:
@@ -135,16 +135,16 @@ proc CalculateDotProduct
je DotNonEven je DotNonEven
LoopDotEven: LoopDotEven:
movdqu xmm0, [rcx] ;pA lddqu xmm0, [rcx] ;pA
pmaddwd xmm0, [rdx] ;pB pmaddwd xmm0, [rdx] ;pB
paddd xmm7, xmm0 paddd xmm7, xmm0
movdqu xmm1, [rcx + 16] lddqu xmm1, [rcx + 16]
pmaddwd xmm1, [rdx + 16] pmaddwd xmm1, [rdx + 16]
paddd xmm7, xmm1 paddd xmm7, xmm1
movdqu xmm2, [rcx + 32] lddqu xmm2, [rcx + 32]
pmaddwd xmm2, [rdx + 32] pmaddwd xmm2, [rdx + 32]
paddd xmm7, xmm2 paddd xmm7, xmm2
movdqu xmm3, [rcx + 48] lddqu xmm3, [rcx + 48]
pmaddwd xmm3, [rdx + 48] pmaddwd xmm3, [rdx + 48]
add rcx, byte 64 add rcx, byte 64
add rdx, byte 64 add rdx, byte 64
@@ -156,10 +156,10 @@ DotNonEven:
cmp r9d, byte 0 cmp r9d, byte 0
je DotFinal je DotFinal
movdqu xmm0, [rcx] ;pA lddqu xmm0, [rcx] ;pA
pmaddwd xmm0, [rdx] ;pB pmaddwd xmm0, [rdx] ;pB
paddd xmm7, xmm0 paddd xmm7, xmm0
movdqu xmm1, [rcx + 16] lddqu xmm1, [rcx + 16]
pmaddwd xmm1, [rdx + 16] pmaddwd xmm1, [rdx + 16]
paddd xmm7, xmm1 paddd xmm7, xmm1
@@ -167,8 +167,7 @@ DotFinal:
movdqa xmm6, xmm7 movdqa xmm6, xmm7
psrldq xmm6, 8 psrldq xmm6, 8
paddd xmm7, xmm6 paddd xmm7, xmm6
movdqa xmm6, xmm7 movshdup xmm6, xmm7
psrldq xmm6, 4
paddd xmm7, xmm6 paddd xmm7, xmm6
movd eax, xmm7 movd eax, xmm7
emms emms
@@ -180,5 +179,16 @@ endproc
; ;
proc GetMMXAvailable proc GetMMXAvailable
push rbx
push rcx
push rdx
mov eax,1 mov eax,1
CPUID
test ecx, 1 ; actually, testing for SSE3
setnz al
and eax, byte 1
pop rdx
pop rcx
pop rbx
endproc endproc