https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125109
Thomas Schwinge <tschwinge at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|AMD GPU offloading |AMD GPU offloading
|performance pessimized if |performance pessimized if
|also compiling for NVIDIA |also compiling for NVIDIA
|GPU offloading |GPU offloading (OpenMP)
--- Comment #4 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
That one was quick to try:
(In reply to Andrew Stubbs from comment #2)
> If the "return 32" is replaced with "return 64", what affect does that have
> on each device?
>
> I.e. Might that help AMD while leaving NVPTX unharmed?
AMD GPU is even worse: ~523.3 -> ~535.4 s, and NVIDIA GPU is unchanged. (One
run only, each.)
