Exporting ArrayFire CUDA kernels to files

https://arrayfire.com/ is a great library providing an abstraction layer for different kinds of acceleration technologies: CPU, OpenCL and CUDA. You write your code once and the code will be compiled with a JIT and executed on the active backend.

Sometimes you need to dig a little deeper to figure out what’s going on. There is an environment variable you can set before starting your process AF_JIT_KERNEL_TRACE

The expected values are:

  • AF_JIT_KERNEL_TRACE = stdout
  • AF_JIT_KERNEL_TRACE = stderror
  • AF_JIT_KERNEL_TRACE = path to an existing folder on the file system

Leave a Reply

Your email address will not be published.