Very nice! That's essentially all I want from a cuda runtime. It should be possible to run llvm libc unit tests against this, which might then justify a corresp

Very nice! That's essentially all I want from a cuda runtime. It should be possible to run llvm libc unit tests against this, which might then justify a corresponding amd library that does the same direct to syscall approach.