[NewGVN] Assertion "Seem to have processed the same Value a lot"' failed

[Quuxplusone]

Bugzilla Link	PR36500
Status	NEW
Importance	P enhancement
Reported by	Jonas Paulsson (paulsson@linux.vnet.ibm.com)
Reported on	2018-02-24 00:51:59 -0800
Last modified on	2018-10-29 03:31:56 -0700
Version	trunk
Hardware	PC Linux
CC	florian_hahn@apple.com, llvm-bugs@lists.llvm.org, paulsson@linux.vnet.ibm.com, uweigand@de.ibm.com
Fixed by commit(s)
Attachments	tc_newgvn.ll (1901 bytes, text/plain) tc_newgvn_ProcessedCount.ll (5217 bytes, text/plain) crash0.i (227560 bytes, text/plain)
Blocks
Blocked by
See also	PR34472

Created attachment 19948
reduced testcase

bin/opt -mtriple=s390x-linux-gnu -march=z13 tc_newgvn.ll -o a.opt.ll -sroa -
instcombine -loop-rotate -indvars -loop-unroll -newgvn

[Quuxplusone]

Attached tc_newgvn.ll (1901 bytes, text/plain): reduced testcase

[Quuxplusone]

I cannot reproduce the failure with r341688, maybe because one of the passes does not generate IR that was causing the assertion. Jonas, could you check if you can reproduce the issue with the un-reduced test case?

There is a know issue and PR34472 contains more details.

[Quuxplusone]

(In reply to Florian Hahn from comment #1)
> I cannot reproduce the failure with r341688, maybe because one of the passes
> does not generate IR that was causing the assertion. Jonas, could you check
> if you can reproduce the issue with the un-reduced test case?
>
> There is a know issue and PR34472 contains more details.

Sorry for late response. I used to see this frequently, so I didn't save the
original test case, unfortunately. I reported this and then stopped enabling
newgvn in testing temporarily. I'll enable newgvn again and see what I get...

[Quuxplusone]

Attached tc_newgvn_ProcessedCount.ll (5217 bytes, text/plain): reduced testcase

[Quuxplusone]

Attached crash0.i (227560 bytes, text/plain): clang unreduced test case

[Quuxplusone]

Probably phi-of-ops related, passes with -enable-phi-of-ops=false