By -O1 I'm assuming you mean just -O? ("Optimize for Speed" in Xcode). Or -Onone?

Is this a project you can share? This probably isn't enough information to track down the issue.

Comment by Tomas Srna (JIRA)

@belkadan The functioning optimization is "Fast [-O, O1]". As for the code, I'll try to isolate the problem and upload a mini project this week.

That's not a Swift optimization setting; changing it does not affect how the Swift code in your target is built.

Comment by Tomas Srna (JIRA)

@belkadan for Swift Compiler - Code Generation it is -Onone that works and -O that doesn't

Comment by Tomas Srna (JIRA)

I tried to isolate the problem, but couldn't reproduce outside our project, which is not public, under heavy NDA. I'll try and send the crashlog, maybe it could help. 2019-05-22_08-26-28.8305_+0200-ec3f7a17da0884e8a195cc8c543c67da8b004207.crash

Thanks. @eeckstein, shajrawi (JIRA User), does this sound familiar?

That's really hard to say without the project.
srna (JIRA User) If you cannot isolate the problem, there is another possibility to track this down: You can find the offending optimization by bisecting.
When you compile with -Osize, you can add the compiler option
-Xllvm -sil-opt-pass-count=<n>
where n the number of optimizations executed, e.g. 0 means all optimizations turned off.
If you can find the n where the crash occurs and does not occur with n-1, this would be very useful for us (I'll give you more instructions then).

Comment by Tomas Srna (JIRA)

I tried it and the result is rather weird.

`-Xllvm -sil-opt-pass-count=1` — works
`-Xllvm -sil-opt-pass-count=2` — works
`-Xllvm -sil-opt-pass-count=3` — works
`-Xllvm -sil-opt-pass-count=4` — works
`-Xllvm -sil-opt-pass-count=8` — works
`-Xllvm -sil-opt-pass-count=16` — works
`-Xllvm -sil-opt-pass-count=128` — works
`-Xllvm -sil-opt-pass-count=1024` — works
without this flag — crashes

@eeckstein Can you give me a hint of how large <n> should I try?

n can be quite large (> 100000)
I took the opportunity and documented the whole procedure here: https://github.com/apple/swift/blob/master/docs/DebuggingTheCompiler.rst#identifying-an-optimizer-bug

Could you do that exercise and provide the name of the bad optimization and the SIL before and after the optimization?
(If you don't want to attach the SIL to the public Jira, you can also file a radar and attach it there).

Comment by Tomas Srna (JIRA)

@eeckstein Thank you. To keep you posted, I am slowly approaching the N, currently the spread is +/- 50. The number is really big, at approximately 5 mil.

Comment by Tomas Srna (JIRA)

So the crashing N=4916513, N-1 doesn't crash. I'll do the second step in near future.