[Libclc-dev] [PATCH] math: Implement remainder(x, y)

[Matt Arsenault via Libclc-dev]

> On Jan 17, 2017, at 19:30, Aaron Watry via Libclc-dev <libclc-dev at lists.llvm.org> wrote:
> 
> Mostly ported from the amd-builtins branch.
> 
> The amd-builtins branch uses __amdil_improved_fdiv_f32 and FTZ which aren't
> available in generic CLC.
> 
> __amdil_improved_fdiv_f32 points to native_divide which does
> native_recip(y)*x.
> 
> Since we don't have native_divide or native_recip yet, I've just stuck an
> actual division here.

Those can just be trivially implemented as the regular divide. Refinements would be to use the intrinsic (should only be needed for f64, f32/f16 it should should just happen for amdgcn)


> 
> I've taken a shot at a replacement for FTZ(x), but feel free to suggest
> alternatives.
> 

A denormal flush function should be implemented with llvm.canonicalize (for subtargets where f32 denormals are off by default, which is always true now but will probably change for VI). It’s only implemented for amdgcn in the backend currently

-Matt