During building opencv-4.1.0 I realize that nvcc (10.1) just compiles *.cu files in sequence one by one. That delays the building progress significantly.
I wonder if nvcc supports compiling in parallel? If it does not, what is reason behind it and if it does, what is proper way to enable?
Thanks.
It doesn’t. I don’t know why.
FWIW I don’t think g++ does this, and at least on linux, nvcc depends on g++.
[url]c++ - Compiling with g++ using multiple cores - Stack Overflow
[url]c++ - Cuda parallel code generation in Visual Studio - Stack Overflow
If you’d like to see a change, you’re welcome to request that using the bug reporting system (perhaps add RFE keyword to description): the instructions are linked to a sticky post at the top of this forum.
Thank you for reply,
I am confused that according to you, nvcc does not support compiling in parallel, how external utils could make nvcc become supported?
I am quite surprised that CUDA has been popular for years but still it does not support multi-core world.
Perhaps you should study how make -j works. That has nothing to do with CUDA, but it can certainly work with nvcc.
I guess the same could be said for g++
–threads option was added to nvcc in CUDA 11.3