This is the mail archive of the
mailing list for the GCC project.
Re: Autotuning parameters/heuristics within gcc - best place to start?
- From: <gfursin at gmail dot com>
- To: "Andrew Bennett" <Andrew dot Bennett at imgtec dot com>, "Gary Funck" <gary at intrepid dot com>, "Andi Kleen" <andi at firstfloor dot org>
- Cc: "Robert Stevenson" <9golive9 at gmail dot com>, <gcc at gcc dot gnu dot org>, "Collective Mind" <collective-mind at googlegroups dot com>, "ctuning-discussions" <ctuning-discussions at googlegroups dot com>
- Date: Mon, 6 Oct 2014 12:10:39 +0200
- Subject: Re: Autotuning parameters/heuristics within gcc - best place to start?
- Authentication-results: sourceware.org; auth=none
Actually MILEPOST link is quite outdated. We now have a new
python-based Collective Mind framework which includes universal
multi-dimensional multi-objective auto-tuner with a web-based interface.
If you are interested, you may find some optimization flags and parameters
for different versions of GCC in JSON format for this framework here:
The framework itself and our long-term vision is available here:
By the way, we are not advertising this framework much since it became
quite large in the past few years and includes many different proof-of-concept
plugins for our current work on reproducible experimentation and predictive modeling -
therefore it's in our plans to release a lighter and simpler version at some point ;) ...
Hope it's of any help and take care,
From: Andrew Bennett
Sent: Thursday, October 02, 2014 12:11 PM
To: Gary Funck ; Andi Kleen
Cc: Robert Stevenson ; email@example.com
Subject: RE: Autotuning parameters/heuristics within gcc - best place to start?
From: firstname.lastname@example.org [mailto:email@example.com] On Behalf Of Gary
Sent: 28 September 2014 20:02
To: Andi Kleen
Cc: Robert Stevenson; firstname.lastname@example.org
Subject: Re: Autotuning parameters/heuristics within gcc - best place to
On 09/26/14 07:47:05, Andi Kleen wrote:
> One example of an existing autotuner is the gccflags tuner in opentuner.
Although dated, ACOVEA might offer up some ideas.
The Milepost project might also be of interest: