Re: Population count in SSE2, again

"Maarten Kronenburg"  wrote in message
>

As always there is some room for improvement,
below the loop runs the other way which is a bit better:

PROC    __int_popcnt
ARG     pdata:dword, ndata:dword
push ebp
mov ebp, esp
 push esi
 push edi
 mov esi, [pdata]
 mov edi, [ndata]
 pxor xmm7, xmm7
 mov eax, 0F0F0F0Fh
 movd xmm6, eax
 punpckldq xmm6, xmm6
 unpcklpd xmm6, xmm6
 movdqa xmm5, xmm6
 pslld xmm5, 2
 pxor xmm5, xmm6
 movdqa xmm4, xmm5
 pslld xmm4, 1
 pxor xmm4, xmm5
 pxor xmm0, xmm0
 mov ecx, edi
 and edi, 0FFFFFFFCh
 lea esi, [esi+4*edi]
 neg edi
 jz short @[EMAIL PROTECTED]
 movdqu xmm1, [esi+4*edi]
@[EMAIL PROTECTED]
 movdqa xmm2, xmm1
 psrld xmm2, 1
 pand xmm1, xmm4
 pand xmm2, xmm4
 paddd xmm1, xmm2
 movdqa xmm2, xmm1
 psrld xmm2, 2
 pand xmm1, xmm5
 pand xmm2, xmm5
 paddd xmm1, xmm2
 movdqa xmm2, xmm1
 psrld xmm2, 4
 paddd xmm1, xmm2
 pand xmm1, xmm6
 psadbw xmm1, xmm7
 paddd xmm0, xmm1
 add edi, 4
 jnz short @[EMAIL PROTECTED]
 and ecx, 3
 jz short @[EMAIL PROTECTED]
 mov edi, 0FFFFFFFCh
 dec ecx
 jnz short @[EMAIL PROTECTED]
 movd xmm1, [esi]
 jmp short @[EMAIL PROTECTED]
 movq xmm1, [esi]
 dec ecx
 jz short @[EMAIL PROTECTED]
 movd xmm2, [esi+8]
 movlhps xmm1, xmm2
 dec ecx
 jmp short @[EMAIL PROTECTED]
 movhlps xmm1, xmm0
 paddd xmm0, xmm1
 movd eax, xmm0
 pop edi
 pop esi
pop ebp
ret
ENDP    __int_popcnt