================ @@ -1,8 +1,33 @@ // REQUIRES: amdgpu-registered-target // RUN: %clang_cc1 -verify -cl-std=CL1.2 -triple amdgcn-amd-amdhsa -Wno-unused-value %s // RUN: %clang_cc1 -verify -cl-std=CL2.0 -triple amdgcn-amd-amdhsa -Wno-unused-value %s +// RUN: %clang_cc1 -fcuda-is-device -cl-std=CL2.0 -verify -triple x86_64 -aux-triple amdgcn -Wno-unused-value %s ---------------- yxsamliu wrote:
This RUN line looks inconsistent. If this is meant to test host-side compilation with an AMDGPU aux target, it should not use `-fcuda-is-device`. If it is meant to test device-side compilation, the triples look backwards: AMDGPU should be the primary triple and the host should be the aux triple. For OpenCL, I do not think we should add this aux-target RUN line at all since OpenCL is not a single-source host/device language. https://github.com/llvm/llvm-project/pull/207687 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
