Re: [eigen] Eigen Blocked QR algorithm |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen@xxxxxxxxxxxxxxxxxxx
- Subject: Re: [eigen] Eigen Blocked QR algorithm
- From: Gael Guennebaud <gael.guennebaud@xxxxxxxxx>
- Date: Thu, 17 Jun 2010 19:42:20 +0200
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:received:mime-version:received:in-reply-to :references:from:date:message-id:subject:to:content-type; bh=FoMz1/Jmy3ON3v6Baa3usDm3rFD+pVeitD2szQMgd7s=; b=vsRNivDlKMn768A9WXQg0pvsydT4+b8t5Xet3Uk1PM8PpQVKX0d7tENgUd7CsoQCBh wcCXTzFojxgDaRjyX2d9CIp4zxK8FWzcpH3g42ICuZQ3V/y6+qf+ieNI3n7MlvdItNSP dRB+hRO2qXExI9tc8Vxw3nU9GQJQmnciusd7M=
- Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:from:date:message-id:subject:to :content-type; b=HrITjkH+0uq8iw7RsATwzirp+KtPsvUC/FLVLuwDg1dCUcWzWOIJ69odPcHYtqt1CD 0DjNgqyEBMJLrYn98/Jf+5SkPbVwvifdhWmTisbhA7fY5yt1Wb/82aFxb52hmdA2Kc6m TSJx0ifQUX/lnJfFU79c2+TAGmRTDC1HfuZiQ=
Hi list,
I''ve just integrated and committed Vincent's code to our
HouseholderQR class. There is no novel API, it is automatically
enabled.
In the future we should:
- add a mechanism to control the blocking (for all decompositions)
- make the underlying inplace functions available to the public API
(for all decompositions)
- slightly redesign the Householder module to better integrate block
Householder transformations (I suggest to mark this entire module
\internal)
However, all this only concerns very advanced uses of the library, and
so all these changes can be delayed for the 3.1 release.
As a side note, I plane to remove our custom \experimental tag in
favor of \internal since purely experimental stuffs go to
unsupported/
cheers,
gael.
On Thu, Jun 17, 2010 at 1:46 PM, <vincent.lejeune@xxxxxxxxxx> wrote:
>
> Hi,
>
>
>
> I worked on the algorithm of blocked QR decomposition. It can handles now
>
> complex<...> number, and matrix of size that are not multiple of blocksize.
>
> It computes between 2 and 3 times faster than non blocked QR algorithm.
>
>
>
> May it be possible to exhaustivly test it and to include it in the next
>
> version of Eigen ?
>
>
>