[PATCH] D123353: [CUDA][HIP] Externalize kernels in anonymous name space

Artem Belevich via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Fri Apr 8 14:17:16 PDT 2022


tra accepted this revision.
tra added a comment.
This revision is now accepted and ready to land.

LGTM overall.



================
Comment at: clang/test/CodeGenCUDA/kernel-in-anon-ns.cu:13
+
+// CHECK: define weak_odr {{.*}}void @[[KERN:_ZN12_GLOBAL__N_16kernelEv\.anon\..*]](
+// CHECK: @[[STR:.*]] = {{.*}} c"[[KERN]]\00"
----------------
yaxunl wrote:
> tra wrote:
> > Will the externalized names be uniquified as well?
> > 
> > E.g. if we compile with -fgpu-rdc, we do want the kernels to be externally visible, but we also don't want the names to clash if we have two TUs having the same external name for them.
> Yes, the kernel name is uniquified with a hash of the source path and compile options.
Then we should probably include the unique suffix in the CHECK line. The RUN lines already provide specific cuid, so the fuffix will always be the same.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D123353/new/

https://reviews.llvm.org/D123353



More information about the cfe-commits mailing list