Re: [eigen] 3 dimensional arrays |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen@xxxxxxxxxxxxxxxxxxx
- Subject: Re: [eigen] 3 dimensional arrays
- From: Rodrigo Benenson <rodrigo.benenson@xxxxxxxxx>
- Date: Wed, 26 Jan 2011 18:16:47 +0100
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:in-reply-to:references:from:date :message-id:subject:to:content-type:content-transfer-encoding; bh=2w/HNVzDlgdZkLGakHCdaFgZ7C9VtrIIQZFD+IlPi0c=; b=dJ9nWCho0Ymb4bj8E9ErKk8GhuI9KS1TkqZc/tGNg+h/4RrCMRGzQC2j4vDtvl2neq bR34oAcNE4pK7zdYpyovBRoDOw6jbhxo1yZ1Mro7oq0lQZ7eB2Ha8QuPd2ryKbNKHDde Sxq1wxklhT01DM715J+fTFy2RIHQy1pzasFcE=
- 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:content-transfer-encoding; b=c68iqky3P9hFD55OxNDnQF+x8Umov9zIZv809T4ASA+04+D/Rsov7nYukDQdu/gUgF /MlKRs05Pz0puKBGGPrli1zUrs1tU30HB6qyg0uDIrzkAAm/v0Mxfm/gwA+UcL1BzW7u l5UJT/vXjseDq5uN1HX7JyUVnR1LVlZw5cnN8=
thanks for the hint.
regards,
rodrigob.
On Tue, Jan 25, 2011 at 3:24 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
> Out of the top of my head, here are 2 libraries that offer
> higher-dimensional arrays and claim high performance:
> - NT2
> - LTL
>
> Benoit
>
> 2011/1/25 Rodrigo Benenson <rodrigo.benenson@xxxxxxxxx>:
>> It is there a recommended "Eigen like" library for 3d arrays manipulation ?
>> I am looking for something to could be compared performance-wise with Eigen...
>>
>> Regards,
>> rodrigob.
>>
>> On Tue, Jun 29, 2010 at 2:50 PM, Mani chandra <mchandra@xxxxxxxxxx> wrote:
>>> We do have a GPU port of our code, where about 70% of the computation is
>>> done on the GPU. We're getting nice speed ups but we can do much better.. But
>>> at the moment it runs only on one tesla card. We want to change to Eigen
>>> since all development on Blitz++ has stopped.
>>>
>>> Mani
>>>
>>> On 06/29/2010 06:03 PM, Rohit Garg wrote:
>>>>
>>>> If it has a lot of array based computation aka element wise stuff, may
>>>> be should look into gpu's as well.
>>>>
>>>> On Tue, Jun 29, 2010 at 4:58 PM, Mani chandra<mchandra@xxxxxxxxxx> wrote:
>>>>>
>>>>> Nice. We really want to port our CFD code written using Blitz++ to Eigen
>>>>> and manipulating 3D arrays is essential. So I guess we wait then? With
>>>>> the
>>>>> 3D data structure that you are suggesting, can the arrays then be used
>>>>> just
>>>>> like 3D arrays?
>>>>>
>>>>> Mani
>>>>> On 06/29/2010 04:40 PM, Benoit Jacob wrote:
>>>>>>
>>>>>> 2010/6/29 Mani chandra<mchandra@xxxxxxxxxx>:
>>>>>>>
>>>>>>> Can one use Eigen to perform operations similar to Blitz++ on
>>>>>>> multidimensional arrays?
>>>>>>
>>>>>> Not currently. Eigen is definitely focused on the cases of 1D
>>>>>> (vectors) and 2D (matrices).
>>>>>>
>>>>>> However, as Hauke mentions (but his example is inside out :) ), we
>>>>>> support using Array as Scalar type, e.g.
>>>>>>
>>>>>> Matrix<ArrayXd, Dynamic, Dynamic>
>>>>>>
>>>>>> will give you a 3D data structure. Currently this isn't well
>>>>>> optimized, and we don't offer a tensor API, but both issues can be
>>>>>> overcome (the speed issue hopefully by the time 3.0 or 3.1 is
>>>>>> released, the tensor API would be a nice extension module).
>>>>>>
>>>>>> Benoit
>>>>>>
>>>>>>> Mani
>>>>>>> On 06/29/2010 01:12 PM, Hauke Heibel wrote:
>>>>>>>>
>>>>>>>> On Tue, Jun 29, 2010 at 9:23 AM, Mani chandra<mchandra@xxxxxxxxxx>
>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>> Any plans to support multi-dimensional arrays in Eigen in the
>>>>>>>>> future?
>>>>>>>>
>>>>>>>> Only to some extend. It is unlikely that you'll ever be seeing
>>>>>>>> constructors like
>>>>>>>>
>>>>>>>> ArrayXXXd m(rows,cols,slices)
>>>>>>>>
>>>>>>>> but we plan to support (maybe that is already partially working in the
>>>>>>>> development branch) so called "arrays of structures", i.e.
>>>>>>>>
>>>>>>>> Array<VectorXd, Rows, Cols>
>>>>>>>>
>>>>>>>> Our core devs, Benoit& Gael know more.
>>>>>>>>
>>>>>>>> - Hauke
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>
>>>
>>>
>>>
>>
>>
>>
>
>
>