As far as I know, there was a sample named glang in older versions of CUDA which was able to produce NVVM IR (.ll file) from a cu source file. What has happened to glang in newer versions? I was wondering if there are some alternatives…
Actually I’m trying to write an optimization for Cuda Kernels. Is there any other way to deal with cu files and extract the kernels as llvm objects? (something like LLVM passes)
Any comment would be appreciated.