[PATCH] D120781: [IRLinker] materialize Functions before moving any

[Duncan P. N. Exon Smith via Phabricator via llvm-commits]

dexonsmith added a subscriber: lhames.
dexonsmith added a comment.

In D120781#3380046 <https://reviews.llvm.org/D120781#3380046>, @nickdesaulniers wrote:

> In D120781#3376069 <https://reviews.llvm.org/D120781#3376069>, @dexonsmith wrote:
>
>> - The only way to avoid a materialization explosion is to record backedges.
>
> Unless I can make @tejohnson 's suggestion work regarding making `IRLinker::AddLazyFor` optional, then at that point I agree.

Is that just going to fix ThinLTO, or will it also fix other lazy-materialization use cases as well (like the potential JIT case I described)?

@lhames, do you know of concrete JIT uses cases that rely on lazy materialization for performance? Is there someone we could add to the review to help get a testcase to check, if @nickdesaulniers  is able to get the `AddLazyFor` fix working for ThinLTO?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D120781/new/

https://reviews.llvm.org/D120781