BUILD: fix GPU arch flags in amdclang build

openucx / ucc

Unified Collective Communication Library

https://openucx.github.io/ucc/

BSD 3-Clause "New" or "Revised" License

177 stars 85 forks source link

BUILD: fix GPU arch flags in amdclang build #914

Closed nileshnegi closed 4 months ago

nileshnegi commented 5 months ago

What

Modified GPU architecture flags for compatibility with AMD GPUs.
Add -O3 for improving performance.

Why ?

Incompatibility between explicitly specified GPU arch flags (like --target-offload=gfx90a:xnack+) and --target-offload=native led to build errors on some AMD GPUs.

swx-jenkins3 commented 5 months ago

Can one of the admins verify this patch?

Sergei-Lebedev commented 5 months ago

ok to test