RFR(M): 7197327: 40% regression on 8 b41 comp 8 b40 on specjvm2008.mpegaudio on oob

[Vladimir Kozlov]

On 2/1/13 8:46 AM, Roland Westrelin wrote:
> Thanks for the review, Vladimir.
> I'm working on your suggestions.
>
>> Why cleanup_expensive_nodes() is not using cmp_expensive_nodes()? It will reduce time spent in should_optimize_expensive_nodes() later. May be it should call should_optimize_expensive_nodes() and if it returns 'true' do cleanup using cmp_expensive_nodes(). And if it returns 'false' emptying whole list (and set NULL control).
>
> 2 PowD nodes may be unrelated when cleanup_expensive_nodes() is called but later optimizations may change the inputs of the PowD nodes and make them candidates for optimization. That's why cleanup_expensive_nodes() only check node's types and not their inputs.

You are right, we should be conservative here.

Thanks,
Vladimir

>
> Roland.
>