[PATCH] D129541: [llvm][IPO] Add IR function attribute fine_grained_bitfields

[John McIver via Phabricator via llvm-commits]

jmciver added a comment.

@nikic thanks for the feedback.

In D129541#3646365 <https://reviews.llvm.org/D129541#3646365>, @nikic wrote:

> LLVM should not have any knowledge about "fine-grained bitfields".

I agree that strong division in tooling purpose is paramount. Ideally I would like to perform analysis during `opt` that would identify potential poison conflicts as to be agnostic to the particular pass (as there are four function inline pass types by my count). It appears post Clang CodeGen there is limited context for identifying if the code was generated in support of a bit-field initialization. Do you have any ideas?

In D129541#3646365 <https://reviews.llvm.org/D129541#3646365>, @nikic wrote:

> This also doesn't seem principled, in the sense that other IPO transforms (say, IPSCCP) could end up causing issues, not just inlining.

I am aware of SCCP, but did not know of the interprocedural version. Thanks for bringing it to my attention. I'll take a look and discuss with my mentor @nlopes.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D129541/new/

https://reviews.llvm.org/D129541