RFR: 8331736: C2: Live Node limit exceeded limit after JDK-8316991

[Tobias Hartmann]

On Tue, 4 Jun 2024 16:32:15 GMT, Cesar Soares Lucas <cslucas at openjdk.org> wrote:

> Please, consider this patch to interrupt execution of `split_unique_types` (SUT) when number of live nodes reaches 3/4 of `max_live_nodes`.
> 
> The included test case reproduces the problem. The number of live nodes before running phase 3 of SUT is ~20k, after processing about 750 mergemem nodes the number of live nodes is over 70k. This problem was first encountered when running an old `.jar` file that was created before `invokedynamic` optimizations - that's why I disable string optimizations in the test case. The test case is strongly based on [the method that was originally triggering the problem](https://github.com/Unidata/netcdf-java/blob/c782ef80ab54a09befd6d5065c6baeed54949222/cdm/radial/src/main/java/ucar/nc2/iosp/nids/Nidsheader.java#L2174).
> 
> Tested on Linux, Win, Mac x86_64 tier1-3 and GHA.

Looks good to me. We have similar logic for split-if (see `must_throttle_split_if`).

src/hotspot/share/opto/escape.cpp line 4807:

> 4805:         _compile->record_failure(C2Compiler::retry_no_escape_analysis());
> 4806:       }
> 4807:       return ;

Suggestion:

      return;

test/hotspot/jtreg/compiler/c2/TestScalarReplacementMaxLiveNodes.java line 41:

> 39:  *                   -XX:DesiredMethodLimit=100000
> 40:  *                   compiler.c2.TestScalarReplacementMaxLiveNodes
> 41:  */

Please add a run without any arguments (`@run main/othervm ...`) so that we get some additional coverage with other VM args through this test.

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

Marked as reviewed by thartmann (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/19545#pullrequestreview-2098456843
PR Review Comment: https://git.openjdk.org/jdk/pull/19545#discussion_r1627291547
PR Review Comment: https://git.openjdk.org/jdk/pull/19545#discussion_r1627296582