RFR (S) 8216302: StackTraceElement::fill_in can use cached Class.name

[coleen.phillimore at oracle.com]

On 1/8/19 2:41 PM, Aleksey Shipilev wrote:
> On 1/8/19 7:20 PM, coleen.phillimore at oracle.com wrote:
>> Having the oops on the stack and getting a safepoint is the definition of "unhandled oops"..    If
>> the oop isn't used below any calls involving it, it is safe.  If there are intervening calls
>> (especially ones with TRAPS), they should be put in Handle so that they are known safe without
>> having to know the implementation details of all the functions that are called.  There is little
>> overhead if there is a HandleMark near the Handles.
> Thanks for the explanation. So, something like this would do?
>    http://cr.openjdk.java.net/~shade/8216302/webrev.03/

Yes, this looks better.  I can't imagine why it would be measurably 
slower, but beats faster with low frequency bugs.

http://cr.openjdk.java.net/~shade/8216302/webrev.03/test/hotspot/jtreg/runtime/StackTrace/StackTraceClassCache.java.html

I have heard that the copyright should also have the Oracle line on it.

Thanks,
Coleen
>
> Passes new test and hotspot tier1. If this patch looks okay, I would put it through jdk-submit.
>
> It is a bit slower than previous version, but still much better than the current baseline:
>
> Benchmark             (depth)  Mode  Cnt    Score   Error  Units
>
> # Old patch
> StackTraceBench.test        1  avgt   15   14.450 ± 0.136  us/op
> StackTraceBench.test       10  avgt   15   20.182 ± 0.088  us/op
> StackTraceBench.test      100  avgt   15   77.107 ± 0.632  us/op
> StackTraceBench.test     1000  avgt   15  647.128 ± 6.159  us/op
>
> # New patch
> StackTraceBench.test        1  avgt   15   14.841 ± 0.095  us/op
> StackTraceBench.test       10  avgt   15   20.708 ± 0.114  us/op
> StackTraceBench.test      100  avgt   15   78.628 ± 0.621  us/op
> StackTraceBench.test     1000  avgt   15  659.408 ± 6.015  us/op
>
> -Aleksey
>