llvm dev - Mar 2021 - NVPTX codegen for llvm.sin (and friends)

Artem, Justin,

I am running into a problem and I'm curious if I'm missing something or 
if the support is simply missing.
Am I correct to assume the NVPTX backend does not deal with `llvm.sin` 
and friends?

This is what I see, with some variations: https://godbolt.org/z/PxsEWs

If this is missing in the backend, is there a plan to get this working, 
I'd really like to have the
intrinsics in the middle end rather than __nv_cos, not to mention that 
-ffast-math does emit intrinsics
and crashes.

~ Johannes


-- 
───────────────────
∽ Johannes (he/his)

On Wed, Mar 10, 2021 at 11:41 AM Johannes Doerfert <
johannesdoerfert at gmail.com> wrote:
> Artem, Justin,
>
> I am running into a problem and I'm curious if I'm missing
something or
> if the support is simply missing.
> Am I correct to assume the NVPTX backend does not deal with `llvm.sin`
> and friends?
>
Correct. It can't deal with anything that may need to lower to a standard
library call.
>
> This is what I see, with some variations: https://godbolt.org/z/PxsEWs
>
> If this is missing in the backend, is there a plan to get this working,
> I'd really like to have the
> intrinsics in the middle end rather than __nv_cos, not to mention that
> -ffast-math does emit intrinsics
> and crashes.
>
It all boils down to the fact that PTX does not have the standard libc/libm
which LLVM could lower the calls to, nor does it have a 'linking' phase
where we could link such a library in, if we had it.

Libdevice bitcode does provide the implementations for some of the
functions (though with a __nv_ prefix) and clang links it in in order to
avoid generating IR that LLVM can't handle, but that's a workaround that
does not help LLVM itself.

--Artem


>
> ~ Johannes
>
>
> --
> ───────────────────
> ∽ Johannes (he/his)
>
>
-- 
--Artem Belevich
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20210310/07f048e4/attachment.html>