VectorAPI Testing Infrastructure custom loop trip count

[Halimi, Jean-Philippe]

Razvan,

It looks like you are right. I just double-checked and C2 does not seem to be triggered if the iteration count is 10, however I see generated code if when it is set to 100.

As this is the only change, I believe I can change the value and merge it as is, is it OK with you?

Thanks,

Jp

-----Original Message-----
From: Lupusoru, Razvan A 
Sent: Thursday, April 5, 2018 11:05 AM
To: Paul Sandoz <paul.sandoz at oracle.com>; Halimi, Jean-Philippe <jean-philippe.halimi at intel.com>
Cc: panama-dev at openjdk.java.net
Subject: RE: VectorAPI Testing Infrastructure custom loop trip count

Initial tests I created locally for Vector API were done using a mix of both techniques - namely both an outer loop repeating the computation body loop and multiple invocation counts. This was done to ensure tests run for an adequate amount of time to allow compilation to happen while also allowing exercising of both on-stack replacements and whole method computation of code with Vector API. Thus I still have some preference for multiple invocation counts but I think we could probably do without. That said, I worry that the current trip count of 10 is too small and won't have the same chances of successful compilation as 10 invocations - I do like that it can be changed dynamically but I wonder if we can select a higher default, like 100.

--Razvan

-----Original Message-----
From: panama-dev [mailto:panama-dev-bounces at openjdk.java.net] On Behalf Of Paul Sandoz
Sent: Thursday, April 05, 2018 8:29 AM
To: Halimi, Jean-Philippe <jean-philippe.halimi at intel.com>
Cc: panama-dev at openjdk.java.net
Subject: Re: VectorAPI Testing Infrastructure custom loop trip count



> On Apr 5, 2018, at 7:40 AM, Halimi, Jean-Philippe <jean-philippe.halimi at intel.com> wrote:
> 
> Sorry for the delayed answer. I totally agree. For some reason, I 
> assumed that the iterations had some kind of write dependencies but you are right, they don’t. J I have merged the original patch and I have rebased this latest revision with the fixes you suggested.
>  
> As for the compare tests, I agree. If I recall correctly, @Razvan made these tests this way for one specific reason. Any thoughts Razvan?
>  
> http://cr.openjdk.java.net/~jphalimi/webrev_testing_trip_count_v1.2/we
> brev/
> <http://cr.openjdk.java.net/~jphalimi/webrev_testing_trip_count_v1.2/w
> ebrev/>
>  

+1

Paul.