llvm-project

6d7962db - [clang][CUDA] Avoid ambiguity in host/device template specializations (#201049)

[clang][CUDA] Avoid ambiguity in host/device template specializations (#201049) This commit changes SemaOverload to resolve an otherwise diagnosed ambiguity between addresses of template specializations of functions that are overloaded for both device and host. Similar to how it works for non-templated function overloads, these changes prioritizes the specializations that corresponds to the target of the owning function, i.e. if compiling for host, the address of the host specialization takes precedence over the device specialization and vice versa. Fixes https://github.com/llvm/llvm-project/issues/199299 --------- Signed-off-by: Steffen Holst Larsen <sholstla@amd.com>