Re: Population count in SSE2, again

"Maarten Kronenburg"  wrote in message
>
>
Optimizing the loop instructions improves from 2.8 to 2.4 cycles/32-bit.
The first lines in the loop should be:
 movdqa xmm2, xmm4
 pandn xmm2, xmm1
 psrld xmm2, 1
 psubd xmm1, xmm2
 movdqa xmm2, xmm5
 pandn xmm2, xmm1
 psrld xmm2, 2
 pand xmm1, xmm5
 paddd xmm1, xmm2
; ...
Another function that may be optimized this way is bit reversal.