There appears to be a bug (or a severe inefficiency) in the nvcc compiler. I am trying to compile a computational kernel that is essentially a large case statement. Using tcc, it will compile in 2.38 seconds… when using nvcc to compile, the same basic code has been compiling for over 4 days (with optimization of -O0 which should be equivalent to tcc) so far, and still hasn’t finished. (smaller case statements seem to work, but this slow down (assuming it will ever finished), seems extraordinarily bad. The memory used by the compiler is small, so it is not matter of resources.
Day 13 of the compile…. Giving up on it ever finishing
2.38 seconds vs 13+ days…
I have moved this post to CUDA NVCC Compiler section for better traction.
If you’re not using the latest CUDA version, then try that first.
If it still seems slow, and you can provide an actual complete code example that demonstrates the slowness, then I would suggest filing a bug.