ComputeCache not invalidated when include files change


I’ve found that when I change a header file that I’ve #included from a CL C source file, the online CL compiler is still fetching a stale binary from its cache - it’s as if the cache is indexed just using the supplied source file, instead of the preprocessed source. Once I nuke ~/.nv/ComputeCache or set CUDA_CACHE_DISABLE=1 it will then compile correctly.

Is this a known issue or should I file a bug?

You could file a bug if you are not sure. Our engineers will look into it.