[llvm] [NVPTX] Custom lower ADDRSPACECAST (PR #125607)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> I guess the short answer is if we know that we can't generate sensible code for the given IR, we should diagnose it, the sooner, the better.

I think we need to never error on addrspacecast, and add an addrspacecast sanitizer.

Consider this OpenCLish example:

```

void foo(volatile generic int* x) {
  __builtin_assume(is_shared(x));
 *x = 4;
}

void bar() {
  private int y;
  foo(&y); // violation, wrong address space
}

```

After inlining + infer address spaces, we could end up with an invalid private->shared addrspacecast. This will of course be optimization level dependent. This cast should produce poison, and should not be a hard compiler error. 



https://github.com/llvm/llvm-project/pull/125607