Thanks for this!

Thanks for this!

SGTM. Thanks for the review!

Thanks! If it's easy to come up with a small regression test (like David mentioned on the bug, reverse_iteration may be a great help), please do so and commit it with this. Otherwise, LGTM as-is.

LGTM with one change -- thanks!

thanks for this!

just a few drive-by nits/comments from me. as usual, not super familiar with clang-tidy, so i won't be able to stamp this.

LGTM

Thanks for this!

Thanks!

Thanks!

Thanks!

just two stylistic nits; lgtm afterward. thanks for this!

Thanks!

ping :)

ping :)

Thanks for adding this!

Please also add a test before committing if doing so is reasonably possible. If the test is super convoluted, I'm OK with this as-is.

Fair enough; thanks again!

Will submit once gribozavr indicates that they're happy with the new test names. Thanks again for working on this!

Thanks you both! LGTM with a test, assuming that adding the test is straightforward.

Thanks for this!

Thanks for the feedback!

Addressed feedback, modulo the constant foldable comment thread.

Address comments -- thanks!

neonfp isn't passed as a feature in the first place; there's a separate API setFPMath which is used for that. We translate it into a target feature for the sake of the backend. So I'm not sure what you're proposing.

Rebased

I'm happy to give rebasing it a shot later this week. My recollection of the prior state of this patch was that we wanted some backend work done to double-check that no illegal ops get generated by optimizations and such, since these checks are purely done in the frontend. I don't foresee myself having time in the near future to make that happen, so is that something that we want to continue to block this patch on? If so, then someone else is probably going to need to do that piece. Otherwise, I think people were happy enough with this patch as-is?

Thanks for this!

just letting George mark it as accepted when he's happy too

Thanks!

LGTM after two nits; thanks!

MSSA bits LGTM modulo a few nits. Feel free to commit when Chandler is happy.

I don't mean to interrupt the ongoing discussion; just wanted to send a few nits about the MSSA-specific bits. Thanks!

This sounds like people are fine with me doing this...

I can't find any users of this flag, so I'm in favor of s/EnableEarlyCSEMemSSA/true/ if that simplifies things.

Thanks!

LGTM after a nit is addressed. Thanks!

LGTM modulo one nit; thanks again!

Thanks again!

Thanks for this!

Only a few more nits on my side, and this LGTM. WDYT, arichardson?

(Forgot to refresh before pressing 'Submit', so maybe efriedma's comment answers all of the questions in mine ;) )

We have warnings like -Wdivision-by-zero that take reachability into account: https://godbolt.org/z/3PD-eF. I don't immediately know how it's all done (e.g. in batch because CFG construction is expensive? ...), but looking there for inspiration may be useful.

Thanks for this!

Thanks for this!

The reason is that MemorySSA has to be moveable

LGTM; thanks again!

Just a few style nits for you, and this LGTM. I assume rnk and serge-sans-paille are content, so I'm happy to check this in for you once these are addressed.

MSSA bits LGTM (to reiterate, as do the CFLAA bits). Leaving the rest of the patch to the other reviewers

LGTM; thanks!

Thanks for this!

Overall approach LGTM for this selective caching behavior.

This LGTM; feel free to submit after Aaron stamps this.

I think that addresses all of the concerns people have put forward; given rnk's comment about one more round of fixes, this LGTM. Will check this in for you shortly.

I think we should do one more round of fixes, we can commit that for you, and then move on to the next steps

LGTM; thanks again!

Thanks for this! Functionally, this looks good. My comments are mostly just simplicity/readability nitpicks.

Looks solid to me overall; just a few nits.

I think @george.burgess.iv is the right person because he appears a lot in git blame :)

Thanks for working on this!

Thanks again!

Thanks for this!

Thank you all for helping with this!

Thanks again!

Thanks for doing this! Just two nits for you :)

Deferring to Chandler's LGTM for the approach + first patch; changes since then LGTM.

LGTM with one more nit. Thanks again!

Thanks for this!

LGTM; thanks!

Thanks! Just a tiny nit

Thanks for this!

LGTM after one nit is addressed.

I would prefer to do the flag-flip separately though, in case it unearths some problematic usage.

Thanks for this!

Thanks for this!

Thanks for this!