[PATCH] D61634: [clang/llvm] Allow efficient implementation of libc's memory functions in C/C++

[Guillaume Chatelet via Phabricator via cfe-commits]

gchatelet added a comment.

In D61634#1502201 <https://reviews.llvm.org/D61634#1502201>, @tejohnson wrote:

> Using function level attributes instead of module flags does provide finer grained control and avoids the conservativeness when merging IR for LTO. The downsides I see, mostly just in terms of the engineering effort to get this to work, are:
>
> - need to prevent inlining with different attributes


IIUC this is needed regardless of the proposed change. Correct?

> - currently the TargetLibraryInfo is constructed on a per-module basis. Presumably it would instead need to be created per Function - this one in particular seems like it would require fairly extensive changes.

Yes this one is a bit worrying.
I think we can discard right away any solution that would mutate or creating a TLI on a per function basis.
Another design could be the following:

  auto FunctionTLI = ModuleTLI.createCustomizedTLI(F);

`FunctionTLI` would either return the function customizations or delegate to the modules TLI. WDYT?

I'm unsure if we want to support function level attribute right away or if it's OK to be in an intermediate state with only module level attributes.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61634/new/

https://reviews.llvm.org/D61634