yxsamliu wrote:
I have some doubt whether this will work for CUDA/HIP since their AST is not
aware of address space. When translated to IR, any pointer is expected to
pointing to default addr space. That is why alloca is immediately casted to
generic pointer and used as generic pointer to match other parts of AST. For
example:
```
__device__ void f(int*);
__device__ void g() {
int a[10];
int *p = a;
f(a);
}
```
both p and f is expecting generic pointer in IR. Actually anywhere a is used,
it is expected to be a generic pointer. That's why we need to cast it to
generic pointer and use it as a generic pointer in the beginning.
https://github.com/llvm/llvm-project/pull/129837
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
