[PATCH] D122890: [2/3] TLS loads optimization (hoist)

Fri Apr 1 04:11:42 PDT 2022

LuoYuanke added a comment.

In D122890#3421961 <https://reviews.llvm.org/D122890#3421961>, @lebedev.ri wrote:

> In D122890#3421946 <https://reviews.llvm.org/D122890#3421946>, @LuoYuanke wrote:
>
>> In D122890#3421936 <https://reviews.llvm.org/D122890#3421936>, @lebedev.ri wrote:
>>
>>> I'm not really sure that attribute should be there, especially after the optimization is enabled by default and cl::opt is inverted.
>>
>> I think we can leave end user a chance to enable it or not. Maybe for debug purpose.
>
> I don't believe that is consistent with the current situation, i.e. is there any prior art for this, an attribute that specifically tells to perform some optimization?
> How should it play with inliner, should it stop the inlining if they mismatch on the caller-callee, or should it be propagated to callee? Etc.

I think this pass runs after inlining.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D122890/new/

https://reviews.llvm.org/D122890