I'm not arguing that passes should not look into unreachable code.

A same/similar question about this code in BasicBlock was raised on llvm-dev here:
http://lists.llvm.org/pipermail/llvm-dev/2015-August/089091.html

Unfortunately, I don't think the legality of the invalid SSA is documented anywhere. But as we can see, it passes the IR verifier.

As noted in D69823, I've made several patches to other passes to work-around the issue. It would be great if we didn't have to do that, but AFAIK, we're stuck.

In D69865#1736280, @jdoerfert wrote:

Unreachable code is generally allowed to be in non-SSA form.

Here are the 2 follow-up patches: D69862, D69865

Thanks for getting back to this.
GVNSink and SimplifyCFG only sink common tails of 2 BBs. I.e. they will not cherry-pick any individual stores from the middle of a BB.
That was the motivation for a little improvement. According to my testing it doesn't have measurable performance impact. I saw fluctuations within 1% which I think is caused by code placement.
I will try to find someone to review.

Kind reminder. @bjope , please review.

Kind reminder. @bjope , please review.

@bjope, please review.

Ping.

I introduced an option that controls splitting the footer BB which is disabled by default.
So, no functional change at the moment.

Fixed Bjorn's comments.

Ran update_test_checks.py

Closed by rL366387

In D64887#1590334, @thegameg wrote:

LGTM, thanks! Not sure how I missed this test.

Closed by https://reviews.llvm.org/rG92fb5f2e479e

There should be also positive impact on LSR pass from this change, since LSR likes equality comparisons.

@qcolombet , can you please commit this patch on my behalf?

Fixed comments.

@reuk , can you please commit this patch on my behalf?

I think "LINKER Stream" formatting doesn't look as intended now:
https://llvm.org/docs/PDB/index.html

I also found performance regression as a result of this commit:
http://lists.llvm.org/pipermail/llvm-dev/2019-March/131308.html

Added "-m elf_x86_64" to the gold invocation.

Updated title, added test and removed copying files.
I decided not to add '.s' to the filename, because then user can also name it as "out.s" and it will be "out.s.s" which seems unnatural.

@spatel, @lebedev.ri, if no other objections, can you please commit it on my behalf?

Fixed comments.

Changed the comment.