search

NVIDIA / nccl

Optimized primitives for collective multi-GPU communication
Other
3.22k stars 812 forks source link

build: allow to have ccbin != CXX #751

Open zougloub opened 1 year ago

zougloub commented 1 year ago

A simple indirection in makefiles/common.mk such as follows:

NVCU_CXX ?= $(CXX)
... -ccbin $(NVCU_CXX) ...

would be nice.

zougloub commented 1 year ago

Hadn't seen https://github.com/NVIDIA/nccl/pull/267, sorry.

sjeaugey commented 1 year ago

Why not just set CXX to the compiler you want to use with CUDA? Do you need to build the rest of NCCL with another compiler?