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

[ Thread Index | Date Index | More Archives ]

2015-02-18 15:51 GMT-05:00 Benoit Jacob <jacob.benoit.1@xxxxxxxxxx>:
If anyone could walk me through the relevant bits of ATLAS code, I would be very interested. I just don't have experience with it, and it's large enough that it's not trivial to explore.

I'd also be very interested in how one can empty caches, and how it can have an effect (I thought that any existing cache entry would be evicted as needed).

Oh - I guess you mean in case a matrix got allocated at the same address as a previous one, and not initialized. But the benchmark initializes all the matrix coefficients anyway, so IIUC it should be pretty deterministic in this respect.



2015-02-18 15:46 GMT-05:00 Ilja Honkonen <ilja.j.honkonen@xxxxxxxx>:

(see here: Maybe there is a profiling pro over
here who can shed some light on this?

Whaley knows a lot about profiling and optimizing linear algebra libraries and the source code is also available: Making sure the caches are empty before every operation/measurement seems to be important and I guess atlas has separate functions for that also.

Mail converted by MHonArc 2.6.19+