Any SIMD experts?

[ponce via Digitalmars-d]

On Monday, 8 December 2014 at 16:32:50 UTC, Martin Nowak wrote:
> I want to do bounds checking of 2 (4 on avx) ulongs (64-bit) at 
> a time.
>
> ulong2 vval = [v0, v1];
> ulong2 vlow = [low, low];
> ulong2 vhigh = [high, high];
>
> int res = PMOVMSKB(vval >= vlow & vval < vhigh);
>
> I figured out sort of a solution, but it seems way too 
> complicated, because there is only signed comparison.
>
> Usually (scalar) I'd use this, which makes use of unsigned wrap 
> to safe one conditional
>
> immutable size = cast(ulong)(vhigh - vlow);
> if (cast(ulong)(v0 - vlow) < size) {}
> if (cast(ulong)(v1 - vlow) < size) {}
>
> over
>
> if (v0 >= vlow && v0 < vhigh) {}
>
> Maybe this can be used on SIMD too (saturated sub or so)?
>
> -Martin

Another solution, in SSE pseudo-code:

   values <- [vval0, vval1] // two value to bound-check, would be 
4 in AVX

   low <- [vlow, vlow]
   high <- [vhigh, vhigh]

   subl <- values - low // using the PSUBQ instruction (SSE2)
   subh <- values - high // using the PSUBQ instruction (SSE2)
   mask <- subh andnot subl // using the PANDN instruction (SSE2)

   // only logical shift is available for 64-bit integers until 
AVX-512
   // using PSRLQ to shift the sign bit to the right (SSE2)
   vresult <- mask >> 31 (for each dword)


Now, each 64-bit word in vresult contains exactly 1 if bound were 
checked, or 0 else.