3 ms·

I've had a similar hypothesis, but it didn't work out for me, only bloating the codebase. Please lmk if you have a different outcome :)

• anonymoushn 3y ago
  I don't know about this code in particular but if you're mostly doing math you tend to run out of architectural registers before you achieve the full throughput of the instructions you care about :(
  • guyomes 3y ago
    That makes sense. I did observe significant speed up using this technic for evaluating a polynomial on several values. I used Hörner algorithm and indeed the number of registers is very small in this case. If the lack of registers is really the bottleneck, a variant might to use both zmm registers for some cumulative sums and ymm registers for some others. In this case the speed up might be less spectacular though. Edit: actually, I just discovered that zmm registers overlap the ymm registers, so the only registers left are the ones from the FPU.
    • gpderetta 3y ago
      AVX512 added 16 more vector registers. Isn't 32 registers enough to unroll by 8?