maarquitos14 wrote: > I find the opencl naming a bit odd, but do not have something better to > suggest. > > I find it odd in the sense that opencl is not really an OS.
I agree, OpenCL is not an OS, but neither is Vulkan, as far as I know. I remember @Keenuts explaining why they did go for making Vulkan an `OSType`, but can't find the reference now. I think the point was that they needed the environment type for the shading stages (e.g. `spirv-unknown-vulkan-pixel`, `spirv-unknown-vulkan-vertex`). Please, @Keenuts, correct me if I'm wrong. This PR just aligns `OpenCL` with `Vulkan`. > I'm also wondering how this could eventually compose with other targets. For > example, if I'm compiling for amd, we could use amdgcn-amd-opencl instead of > --amdhsa? and would that enable some simplification somewhere? Are other > targets communicating some OpenCL specific information in the IR through > other means, like metadata, attributes... that we could replace? You could use `amdgcn-amd-opencl`, but I don't think it would replace `--amdhsa` at the moment. I don't really know, but that wasn't the intent of this PR, and I guess it would require some more work elsewhere. Like I said before, the main driver for this PR was to align `OpenCL` with `Vulkan`, and facilitate how `SPIRVSubtarget` chooses the environment. https://github.com/llvm/llvm-project/pull/170297 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
