On Tue, 7 Feb 2023 20:47:28 GMT, Roger Riggs <rri...@openjdk.org> wrote:
> I'm going to backup a bit; so another question... This test is a fairly
> simple test that should not be taxing on GC or any other resource so I'm
> surprised the the default configuration is not sufficient. It might be worth
> looking a bit deeper into the GC vs Xcomp conditions and the interplay with
> "looongaarch64". I can't help much there, not being familiar with details of
> GC or the compiler.
Using `-Xcomp` necessarily results in more methods being compiled. And in
hotspot, the codecache does not trigger the GC after it is full, but triggers
the GC when it reaches a certain threshold with `SweeperThreshold`.
<pre><code class="cpp">
// ./src/hotspot/share/code/codeCache.cpp
818 if (allocated_since_last_ratio > threshold) { // threshold =
SweeperThreshold / 100.0;
819 // In case the GC is concurrent, we make sure only one thread requests
the GC.
820 if (Atomic::cmpxchg(&_unloading_threshold_gc_requested, false, true)
== false) {
821 log_info(codecache)("Triggering threshold (%.3f%%) GC due to
allocating %.3f%% since last unloading (%.3f%% used -> %.3f%% used)",
822 threshold * 100.0, allocated_since_last_ratio *
100.0, last_used_ratio * 100.0, used_ratio * 100.0);
823 Universe::heap()->collect(GCCause::_codecache_GC_threshold);
824 }
825 }
</code></pre>
So, when I adjust `-XX:SweeperThreshold=30`, the test passes also.
-------------
PR: https://git.openjdk.org/jdk/pull/12399
