New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add some more fastmath functions #2047

Merged

maleadt merged 1 commit into JuliaGPU:master from Zentrik:fastmath-intrinsics

Aug 22, 2023

Contributor

Zentrik commented Aug 20, 2023

This prevents falling back to slow functions and integrates with the new fastmath kwarg.

exp2_fast doesn't fall back to a julia implementation anymore.
inv_fast no longer falls back to inv.
When the new fastmath kwarg is not set, sqrt_fast loses the fast flag in llvm ir (call float @llvm.sqrt.f32 vs call fast float @llvm.sqrt.f32) but the ptx doesn't seem to be affected by the flag. Is there a way to set the fast math flag in a ccall?


          Add some more fastmath functions

23b84f2

Member

maleadt commented Aug 22, 2023

Is there a way to set the fast math flag in a ccall?

That would be #1991, so not right now. Well, unless you want to use inline LLVM IR with llvmcall, see e.g. https://github.com/vchuravy/FastmathOverlay.jl/blob/main/src/FastmathOverlay.jl.

maleadt added cuda kernels performance labels

maleadt merged commit f89e1ab into JuliaGPU:master

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

cuda kernels performance