[PATCH] D138958: [clang] Better UX for Clang’s unwind-affecting attributes

[Sven van Haastregt via Phabricator via llvm-commits]

svenvh added inline comments.


================
Comment at: clang/test/CodeGenOpenCL/fdeclare-opencl-builtins.cl:20
 // Test that Attr.Const from OpenCLBuiltins.td is lowered to a readnone attribute.
+// FIXME: we don't, though.
 // CHECK-LABEL: @test_const_attr
----------------
lebedev.ri wrote:
> I've looked, and i really don't understand how D64319 works.
> It seems like the AST is then serialized into a header?
> Because just adding a new attribute without spelling does not solve the issue.
We're not setting the `NoUnwind` attribute for OpenCL (yet).  The following quick-and-dirty patch appears to fix this test for your patch (but will cause other tests to fail).  If you think it's time to add `NoUnwind` now, I can try putting up a review for that.

```
diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 276d91fa2758..1ea3c11fbe03 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -1972,7 +1972,7 @@ void CodeGenModule::getDefaultFunctionAttributes(StringRef Name,
   // TODO: NoUnwind attribute should be added for other GPU modes OpenCL, HIP,
   // SYCL, OpenMP offload. AFAIK, none of them support exceptions in device
   // code.
-  if (getLangOpts().CUDA && getLangOpts().CUDAIsDevice) {
+  if ((getLangOpts().CUDA && getLangOpts().CUDAIsDevice) || getLangOpts().OpenCL) {
     // Exceptions aren't supported in CUDA device code.
     FuncAttrs.addAttribute(llvm::Attribute::NoUnwind);
   }
```


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D138958/new/

https://reviews.llvm.org/D138958