develooper Front page | perl.perl5.porters | Postings from February 2013

Re: OP_PADSV_NOLV

Thread Previous | Thread Next
From:
Nicholas Clark
Date:
February 27, 2013 20:51
Subject:
Re: OP_PADSV_NOLV
Message ID:
20130227205058.GD5653@plum.flirble.org
On Wed, Feb 27, 2013 at 09:42:16PM +0100, Steffen Mueller wrote:
> Sorry for the late reply.
> 
> On 02/27/2013 10:08 AM, bulk88 wrote:
> > Why? remove a branch and turn it into a separate non branching opcode?
> > (not saying I dont like your idea).
> 
> Yes, that was the intention. Apart from cache-size issues, we basically 
> *know* that it ought to be faster that way. On top of that, PADSVs are 
> very common, too.
> 
> Initial artificial benchmarking showed an improvement on the order of 
> 2-3%. Further (marginally more sophisticated) benchmarking showed that 
> that is compatible with the noise and bias induced by the environment 
> and matters such as linker order and alignment.
> 
> It's really just the usual frustration: Benchmarking micro-optimizations 
> is nigh-on impossible.
> 
> Since this particular change is non-trivial (new OP etc), I won't go 
> ahead and clean it up/push it. But I do believe that there is an 
> opportunity in analysing hot OPs (not necessarily the same as pp_hot) 
> for similar opportunities and looking at the performance impact of 
> specializing a few more OPs.

Yes, I thought this. I certainly tried it for one of the ops a long time
ago. I hit exactly the problem you did - I couldn't measure the difference.

Nicholas Clark

Thread Previous | Thread Next


nntp.perl.org: Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at ask@perl.org | Group listing | About