Possible to inline is_zombie check?

[Azeem Jiva]

The change itself looks OK, although a little more detail in the 
performance gains would be useful.  How many times did you run it?  
What's the STD DEV.  Did you calculate a Student's TTest on it to see if 
the gains were consistent?  What system did you run on?  64bit? 32bit?

Azeem Jiva
@javawithjiva

On 08/22/2012 09:57 AM, Eric Caspole wrote:
> Hi hotspot team,
> While profiling a customer app, I saw "nmethod::is_zombie() const" 
> surprisingly high in the profile. It turns out that is_zombie() is 
> virtual in CodeBlob, so it is always a call. But there is only one 
> "guarantee" in CodeCache that takes advantage of the virtualness that 
> can be coded around.
>
> I have a simple microbenchmark, attached, mimicking the real app, 
> attached, where "nmethod::is_zombie() const" is exercised due to 
> frequent stack-walking. This webrev removes is_zombie as a virtual 
> call so it can be inlined, and improves the performance of the 
> microbenchmark by a few percent.
>
>  http://cr.openjdk.java.net/~ecaspole/inline_zombie_check/webrev.00/
>
> Regards,
> Eric
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20120822/2c5e7b1d/attachment.html