Request for review correction of regression test compiler/6857159/Test6857159

[Tom Rodriguez]

On Nov 3, 2010, at 10:30 AM, Pavel Tisnovsky wrote:

> Hi all,
> 
> I prepared correction for regression test compiler/6857159/Test6857159.
> This test (i.e. original version) usually runs fine on my test machine,
> but it sometimes fails due to timeout which is set to 240 seconds (2
> minutes) for all reg.tests by default.
> 
> My test machine is quite powerfull (Two Quad-Core AMD Opteron Processor
> 2350 = it's possible to run 8 threads concurrently, 8 GB memory), but
> the test takes about 260 seconds on average. So I prepared two variants
> of correction:
> 
> http://cr.openjdk.java.net/~ptisnovs/Test6857159correction1/
> - the timeout is increased to ten minutes (600 seconds)
> 
> http://cr.openjdk.java.net/~ptisnovs/Test6857159correction2/
> - the number of iterations is decreased to one half of original value
> (which IMHO does not change the authenticity of this test but I may be
> wrong of course)
> 
> Can anybody please review these corrections and decide which correction
> (if any :) is better?

I think the second is better.  It could probably be lowered further to 20000.  It's being run -Xbatch and as long as the counter is greater than CompileThreshold it will be a valid test.  Setting the CompileThreshold to something like 2000 in the test script would allow running it for even less which might be a good idea.

tom

> 
> Thank you in advance,
> Pavel T.