JonChesterfield added a comment. In D107971#2941666 <https://reviews.llvm.org/D107971#2941666>, @jdoerfert wrote:
> Assuming this causes us to generate an `alloc as(5)` for `__tmp`, LG Yes, with the running example and this patch we get the perfect: %__tmp = alloca double, align 8, addrspace(5) %call = call double @__ocml_sincos_f64(double %__x, double addrspace(5)* %__tmp) Before, we get: %__tmp = call fastcc i8* @__kmpc_alloc_shared() %__tmp_on_stack = bitcast i8* %__tmp to double* %__tmp_on_stack.ascast = addrspacecast double* %__tmp_on_stack to double addrspace(5)* %call = call double @__ocml_sincos_f64(double %__x, double addrspace(5)* %__tmp_on_stack.ascast) which interacts badly with the addrspace(5) annotation on the ocml function Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D107971/new/ https://reviews.llvm.org/D107971 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits