openucx/ucc

Include #847 in v1.3.0 RC

sandeepd-nv opened this issue · 2 comments

sandeepd-nv commented

Hi, please consider including PR #847 in v1.3.0 RC. I am not able to build v1.2.0 or v.1.3.0-rc1 with CUDA. Either nvcc hangs on certain files or it errors out due to the issue addressed in the aforementioned PR.

Sergei-Lebedev commented

Hi @sandeepd-nv, this PR already included into v1.3.0. Please keep in mind that if you build UCC with default options it might take a lot of time to compile it since we need to build all combinations of reductions and GPU archs. You can use --with-nvcc-gencode option to reduce compilation time (https://github.com/openucx/ucc/wiki/FAQ#15-how-to-compile-ucc-for-a-specific-gpu-architecture)