[PATCH] D18583: Cloning: Reduce complexity of debug info cloning and fix correctness issue.

Duncan P. N. Exon Smith via llvm-commits llvm-commits at lists.llvm.org
Wed Mar 30 14:53:52 PDT 2016


LGTM.

> On 2016-Mar-30, at 14:34, Peter Collingbourne <peter at pcc.me.uk> wrote:
> 
> pcc added a subscriber: tstellarAMD.
> pcc added a comment.
> 
> The problems were related to the calls to `CloneFunction` in lib/CodeGen/CGVTables.cpp` and in `lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp`. In both cases we were wrongly cloning distinct debug info, which confused later codegen/verifier passes. This update fixes the inconsistency. Please take another look.
> 
> @tstellarAMD it's a little surprising that the only test coverage for the always inline pass in the case where there's debug info is in Clang (specifically in `test/CodeGen/backend-unsupported-error.ll`, which seems to test something else). Might be something worth adding coverage for in LLVM?
> 
> 
> http://reviews.llvm.org/D18583
> 
> 
> 



More information about the llvm-commits mailing list