RFR: 8318889: C2: add bailout after assert Bad graph detected in build_loop_late

Emanuel Peter epeter at openjdk.org
Thu Oct 26 11:53:33 UTC 2023


On Thu, 26 Oct 2023 10:47:28 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

> Ensure we do not crash with SIGSEGV in product mode where we do not assert.
> The check is very cheap, and a bailout an adequate solution.
> In debug, we still assert and can fix the bug.

> Analyzing the call graph fanning out from build_loop_late, there are more bailouts missing.
> 
> We really need a better way for this. More motivation to work on #16289, but I am out of time for this week and the next one.

Looking up, I see:
build_and_optimize, and then the PhaseIdealLoop constructor. So I did not add any new cases really. And you are fixing all the existing cases, am I right?

-------------

PR Comment: https://git.openjdk.org/jdk/pull/16375#issuecomment-1780960242


More information about the hotspot-compiler-dev mailing list