================ @@ -0,0 +1,97 @@ +// RUN: echo -n "GPU binary would be here." > %t + +// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fcuda-is-device -fclangir \ +// RUN: -x hip -emit-cir %s -o %t.dev.cir +// RUN: FileCheck --check-prefix=CIR-DEV --input-file=%t.dev.cir %s + +// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fcuda-is-device -fclangir \ +// RUN: -x hip -emit-llvm %s -o %t.dev.cir.ll +// RUN: FileCheck --check-prefix=DEV --input-file=%t.dev.cir.ll %s + +// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -fcuda-is-device \ +// RUN: -x hip -emit-llvm %s -o %t.dev.ll +// RUN: FileCheck --check-prefix=DEV --input-file=%t.dev.ll %s + +// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir \ +// RUN: -x hip -fhip-new-launch-api -fcuda-include-gpubinary %t \ +// RUN: -emit-cir %s -o %t.host.cir +// RUN: FileCheck --check-prefix=CIR-HOST --input-file=%t.host.cir %s + +// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir \ +// RUN: -x hip -fhip-new-launch-api -fcuda-include-gpubinary %t \ +// RUN: -emit-llvm %s -o %t.host.cir.ll +// RUN: FileCheck --check-prefix=HOST --input-file=%t.host.cir.ll %s + +// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu \ +// RUN: -x hip -fhip-new-launch-api -fcuda-include-gpubinary %t \ +// RUN: -emit-llvm %s -o %t.host.ll +// RUN: FileCheck --check-prefix=HOST --input-file=%t.host.ll %s + +#include "../CodeGenCUDA/Inputs/cuda.h" ---------------- Sirraide wrote:
> @Sirraide - Do you know the history here? Seems like there is an open issues > (#40874 & #147373), but they seem old and somewhat stale. I’m aware of the issues existing but I haven’t checked on the state of `__managed__` in Clang; it would be good if we could enable that for CUDA as well assuming the implementation actually matches NVCC's semantics (within reason at least) https://github.com/llvm/llvm-project/pull/230910 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
