On Thursday, 3 November 2016 01:40:29 UTC, Nigel Tao wrote: > > > Another ignorant question from me, but what do you mean exactly by > universal binary? >
Apologies for the confusing and nonsensical term. What I meant was a binary that works for a number of CPUs within an architecture, with or without SSE[n], AVX[n] etc. > > FWIW, the golang.org/x/image/vector package is portable, in that it > has an pure Go implementation. If compiled on GOARCH=amd64, it has > SIMD optimizations, guarded at runtime by a very cheap check (look for > haveFixedAccumulateSIMD). But that check is made once per e.g. > accumulate call (a higher level function composed of many SIMD ops), > not once per e.g. 128-bit add. > That's the sort of cheap checks that I had mind in the very first post when I talked about "I envisaged a call to CPUID and then some bool tests along the way to utilise SSE[2-4]/AVX[2] (or NEON on ARM) if available. All in a static, portable package." Thanks for a good example of that. O. -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
