[PATCH] D58463: [CUDA]Delayed diagnostics for the asm instructions.

[Artem Belevich via Phabricator via llvm-commits]

tra added a comment.

There's a new quirk we've ran into after this patch landed. Consider this code:

  int foo() {
    int prev;
    __asm__ __volatile__("whatever" : "=a" (prev)::);
    return prev;
  }

When we compile for device, asm constraint is not valid for NVPTX, we emit delayed diag and move on. The function is never code-gen'ed so the diag never shows up. So far so good.

Now we add `-Werror -Wininitialized` and things break -- because we bail out early, `prev` is left uninitialized and is reported as such.

  $ bin/clang++ -c --cuda-gpu-arch=sm_35 asm.cu -nocudainc --cuda-device-only -Wuninitialized -Werror
  asm.cu:4:10: error: variable 'prev' is uninitialized when used here [-Werror,-Wuninitialized]
    return prev;
           ^~~~
  asm.cu:2:11: note: initialize the variable 'prev' to silence this warning
    int prev;
            ^
             = 0
  1 error generated when compiling for sm_35.

I think this should also show up in the test case in this patch, too, if you add -Wuninitialized


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D58463/new/

https://reviews.llvm.org/D58463