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

Investigate cost of failed inlined intrinsics that do not turn to a good sequence #106

Closed p0nce closed 1 year ago

p0nce commented 1 year ago

Following from #105 , it turns out some intrinsics that optimize correctly in Godbolt might not once inlined into another function. This is because intel-intrinsics, in a lot of cases, is forced to use the backend optimizations and no builtin are available.

Sometimes no-builtins will be for the best, but sometimes not. Does that even happen often?

p0nce commented 1 year ago

And on the other hand, oftentime use of builtin inhibits optimizations... especially with GDC which requires more builtins.