[PATCH] D39005: [OpenMP] Clean up variable and function names for NVPTX backend
Hal Finkel via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Mon Oct 23 17:50:37 PDT 2017
hfinkel added a comment.
In https://reviews.llvm.org/D39005#900973, @jlebar wrote:
> > The first question that comes to mind is what is the link between data layout and name mangling conventions?
> I pulled up http://llvm.org/doxygen/classllvm_1_1DataLayout.html and searched for "mangling" -- presumably this is what they were referring to. We also don't need to speculate, rnk still works on LLVM. :)
DataLayout generally holds information that the target-independent optimizer needs in order to simplify the IR into our canonical form. This is as opposed to TargetTransformInfo, which provides data necessary to optimize the IR in target-aware ways (e.g., do things that are orthogonal to canonicalization such as inlining and vectorization). It is also as opposed to external utility functions that might be used by the frontend (e.g., llvm::sys::getHostCPUName()). If I recall correctly, this is information that would be used by the frontend when generating the IR, and the function results are controlled by the triple. As a result, I think that a general utility function somewhere would be fine.
More information about the cfe-commits