Efficiently computing the parity of the Hamming weight of a binary sequence

[rcgldr]

X86 parity flag implementation.

Based on some comments in some old threads at other forums, I think this is similar to what gcc probably does. It takes 4.4 seconds on my system (note it's 1/2 the number of iterations I used for my popcnt test).

        mov     ecx,07blackfh
        xor     edx,edx                 ;set edx = parity
par0:   mov     eax,ecx
        xor     al,ah
        jpe     par1
        not     edx
par1:   shr     eax,16
        xor     al,ah
        jpe     par2
        not     edx
par2:   loop    par0
        neg     edx

 

[uart]

Hey good idea to add the "xor al,ah" to check 16 bits at once, that nearly doubles the speed. That reduced my times from 14.7 sec down to about 8.6 seconds.

note it's 1/2 the number of iterations I used for my popcnt test

Yeah I noticed that you had 2^32-1 for that previous test. At first I didn't see it, and I was seriously puzzled as to why my much slower computer seemed to be doing the tight loop at the same speed or even slightly faster than yours (about 5.6 sec). Then I noticed you were doing twice as many iterations.

 

[rcgldr]

Wasn't sure what the return value is supposed to be. At the end of the code the "neg edx", returns 0 for even parity, 1 for odd parity, to reverse this use "inc edx" instead (also the return value would normally be in eax, not edx).