Accéder au contenu.
Menu Sympa

starpu-devel - Re: [Starpu-devel] Problem with performance model calibration

Objet : Developers list for StarPU

Archives de la liste

Re: [Starpu-devel] Problem with performance model calibration


Chronologique Discussions 
  • From: Mirko Myllykoski <mirkom@cs.umu.se>
  • To: Samuel Thibault <samuel.thibault@inria.fr>, starpu-devel@lists.gforge.inria.fr
  • Subject: Re: [Starpu-devel] Problem with performance model calibration
  • Date: Tue, 9 Jun 2020 13:54:25 +0200
  • Authentication-results: mail3-smtp-sop.national.inria.fr; spf=None smtp.pra=mirkom@cs.umu.se; spf=Pass smtp.mailfrom=mirkom@cs.umu.se; spf=None smtp.helo=postmaster@mail.cs.umu.se
  • Ironport-phdr: 9a23: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
  • List-archive: <http://lists.gforge.inria.fr/pipermail/starpu-devel/>
  • List-id: "Developers list. For discussion of new features, code changes, etc." <starpu-devel.lists.gforge.inria.fr>
  • Organization: Umeå University / HPC2N



On 2020-06-09 13:16, Samuel Thibault wrote:
Mirko Myllykoski, le mar. 09 juin 2020 12:44:00 +0200, a ecrit:
In particular, you can see that the "c" parameter is quite small for the
CUDA implementation:

Indeed, and also for the CPU case c is bogus, even if to a lesser
extent. Looking at _starpu_regression_non_linear_power, I see that it
starts its dichotomy for c in [0, min(averages)]. But with the noisy
results that you have, taking the minimum is not a good idea: e.g. in
the CUDA case the minimum average is around 0.004-0.005, but only a few
of the averages have this value. Taking the minimum was meant to make
the dichotomy faster, but with noisy data it doesn't seem like a good
idea. We still need some initial maximum value for the dichotomy, but we
could use the maximum average, or even a coefficiented average of the
average.

Could you give a try to converting find_list_min into find_list_max, and
using that for the initial value of cmax?

That is going to cause problems due to the log(y[i]-c) statement in the test_r, compute_b and compute_a functions (negative argument for log).

- Mirko






Archives gérées par MHonArc 2.6.19+.

Haut de le page