Other |
---|
tbdUnresolved |
Relates :
|
A DESCRIPTION OF THE PROBLEM : The Math class lacks intrinsic implementations of the double precision fp rounding methods: ceil, floor, rint. On x86 SSE/AVX systems; It would be preferable to utilize instructions 'roundsd' rather than the default jdk implementation. Since there's also a packed variation, '(v)roundpd', the use of intrinsics may enable auto-vectorization where currently such optimization is obstructed by the default implementations of Math.ceil, floor, rint.