[lworld] RFR: 8242210: [lworld] TestCallingConvention::test36 spuriously fails

[Frederic Parain]

On Tue, 7 Apr 2020 16:17:19 GMT, Tobias Hartmann <thartmann at openjdk.org> wrote:

> Problems:
> - G1 pre- and post-barriers used when (un-)packing arguments for the calling convention can call into the runtime which
>   screws up argument registers. Save all registers until JDK-8232094 is fixed in mainline (it's the slow path anyway).
> - SignatureStream::as_value_klass triggers a SystemDictionary lookup which acquires the ProtectionDomainSet_lock. When
>   used from fieldDescriptor::print_on_for when some debug printing flags are enabled, this conflicts with the tty_lock.
>   We should simply use get_value_field_klass instead. Also, we should handle null as a vale for non-flattened fields.
>   TraceDeoptimization needs to handle re-allocation of the inline type return value.
> 
> I've also added a new StressCC option to the ValueTypeTest suite to randomly restrict some compilation to C1 and
> thereby stress test the calling convention.

Looks good to me, however, I haven't reviewed the tests.

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

Marked as reviewed by fparain (Committer).

PR: https://git.openjdk.java.net/valhalla/pull/13