================ @@ -0,0 +1,501 @@ +//===- LowerGPUIntrinsic.cpp ----------------------------------------------===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// +// +// Lower the llvm.gpu intrinsics to target specific code sequences. +// Can be called from clang if building for a specific GPU or from the backend +// as part of a SPIRV lowering pipeline. Initial pass can lower to amdgcn or ---------------- JonChesterfield wrote:
They're optionally lowered to amdgpu or to nvptx as they're the only two architectures in tree, or they come out the other end in IR to be handled later in the pipeline. I'd like lowering for intel too but that will need to be out of tree until intel comes in tree. The semantics are hopefully fairly general purpose but I expect they'll need iterative improvement as one learns more about other targets - the first draft is the basis set of llvm libc. I don't really understand the spirv pipeline. The one I've been working with is the amdgpu / hip one, which passes llvm intrinsics through unchanged, and ends up creating amdgcn-- LLVM IR to feed back into the backend. My hope is that in the first instance the SPIR-V module will contain the llvm.gpu intrinsic call, and then sometime later it'll pick up a proper SPIRV name of some sort, but where conversion back to llvm with that intrinsic in it works losslessly. https://github.com/llvm/llvm-project/pull/131190 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
