| Re: [eigen] slides of my talk |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen@xxxxxxxxxxxxxxxxxxx
- Subject: Re: [eigen] slides of my talk
- From: Gael Guennebaud <gael.guennebaud@xxxxxxxxx>
- Date: Mon, 26 Jan 2009 17:45:09 +0100
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:received:in-reply-to:references :date:message-id:subject:from:to:content-type :content-transfer-encoding; bh=q0Y/QcoQOfUGpp3wQ/yPaWlXNArJUU2DJZtoyRMaG50=; b=RhUps+UfkM6AevawS/4ylZErAygmlUWS7dXhDF8D8hxwqHAItiUHhxBW3xcCDoSsfL +kXjr5wlOgWaiuqp1Gslb3fuIk1OFyDxeUZIVndmmfjnrV8vYJl8XysSsiBJs1fmKsw/ RBnRg2Kr4gGquMqmbgOs/g8Mdd2YldIq1hpJ8=
- Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type:content-transfer-encoding; b=VRINa+Jm59sYBhtuUgTShHgYJyqWwHiPnKp/e2pm+g4GBX6q7RXqy+eIxmN7B1lK8I RcFZ7ySFaOyTOM6Gh1SPcmqcgCLV3fHzcwwatATTJIS4eirFzE8Qxs+I8kSJiZLyKtcl zuacVcfWZquJxjgQ7mTla9dG7NUBEPvWT8SE4=
well if you have for instance:
matrix44 data[a lot];
for each mat in data do
foo(mat);
I expect a JIT to optimize foo for 4x4 matrices. but perhaps I'm
expecting too much !
On Mon, Jan 26, 2009 at 5:24 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
> 2009/1/26 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
>> On Mon, Jan 26, 2009 at 4:53 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
>>> The point is that even if they optimize some common operations for
>>> small fixed sizes they can only do it for a small fixed number of
>>> these.
>>
>> actually if they have a good JIT, they only have the overhead of the
>> parsing and compilation and then the algorithm would be as fast as
>> Eigen !
>
> For that, they need not only a good JIT, they need their whole matrix
> framework to support fixed size like Eigen does (as opposed to dynamic
> size). This isn't going to be easy for them as the Matlab syntax
> doesn't let the user specify which sizes are fixed. So I guess that it
> would still be possible for Matlab to handle literals as fixed values,
> but that wouldn't work if the user makes a function taking a matrix as
> parameter...
>
>
>