AuburnSounds / intel-intrinsics

The Dlang SIMD library
https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#techs=MMX,SSE,SSE2,SSE3,SSSE3,SSE4_1
Boost Software License 1.0
68 stars 11 forks source link

How would be a way to have divergent code path for say using AVX? #110

Closed p0nce closed 1 year ago

p0nce commented 1 year ago
p0nce commented 1 year ago

It all stems for the decision to make it work all the time. It the intrinsics were forcing instruction set usage locally, it would force to enforce instruciton set support externally at application level, thus allowing divergent code path.

At the same time, one wouldn't want to mixin the whole intrinsics definition at the point where we know the instruction set needed, that would be a really big mixin.

p0nce commented 1 year ago

Perhaps this is why Walter think badly about "emulating vectors"

p0nce commented 1 year ago

This makes intel-intrinsics kind of biased against consumer software, and biased pro-server software