Re: [eigen] Help needed to run a benchmark on many machines

[Benoit Jacob <jacob.benoit.1@xxxxxxxxx>]

2015-02-18 16:46 GMT-05:00 Ilja Honkonen <ilja.j.honkonen@xxxxxxxxx>:

         double starttime = time();
         for (int i = 0; i < iters_at_a_time; i++) {
           c = a * b;
         }
         double endtime = time();

....

(4M) on a typical ARM CPU (1M cache). But indeed, my benchmark also aims
to measure much smaller cases, so that's a problem. Any suggestions
welcome on how to address this!

Maybe fill vector<T> a, b, c; with more data than e.g. 16 MB and use c[i] = a[i] * b[i] in the loop? If that doesn't show a large difference then perhaps it's not something to worry about at this stage.

That's interesting, thanks. With this, i can change from simulating a situation where we were using the same matrix over and over again, to a situation where we're never using twice the same matrix. But I don't know what would be the most realistic thing to simulate. Not knowing that, and seeing that already with just one set of a,b,c matrices I'm seeing very significant timing differences with different blocking parameters, it is tempting to stick to that.

Benoit

 

Ilja