https://github.com/michaelrj-google commented: @jhuber6 is this okay for GPUs? Right now the new headergen design doesn't have support for different macros on different platforms. https://github.com/llvm/llvm-project/pull/98092