| Re: [eigen] Index type for PermutationMatrix and Transpositions |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen@xxxxxxxxxxxxxxxxxxx
- Subject: Re: [eigen] Index type for PermutationMatrix and Transpositions
- From: Benoit Jacob <jacob.benoit.1@xxxxxxxxx>
- Date: Tue, 25 Jan 2011 10:18:37 -0500
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:in-reply-to:references:date :message-id:subject:from:to:content-type:content-transfer-encoding; bh=aYhglpqGIet703vLwm9O2nHEmvHei0M6y7lO65bC9Q0=; b=POgW8AhTGIH89llAL+waC44fgSbRZ62GNBM9i+2yeswgvfLi0PtNCJwoKjrXq+XBAP PjfUHT2tU4QPNrnQI7g8KrVHhDUwNUCr67Yqr9zJPC20qwliYqV5XBnW9C4/mme+dUfx Yz2SSzDRiEgv3AOXsQEGnUn60ljRsPQmEApok=
- 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=qxhKDN+PmRDEMg5QlDvZBoUkucrbqOOnuO8kQ6Hv3857hS7E8E3WlEgtnw9gnf3Bab bgyGfqBtmsAvR4SsxPj+VXNRdFj1qK0l/13XR6P8Vdvy4FdjW8Eyp3alG+fVS6AiQF4b dE9xkkocF8M676KoRLp1A0fnYFWK+lfIfB9y8=
2011/1/25 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
> On Tue, Jan 25, 2011 at 1:54 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
>> 2011/1/25 Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx>:
>>> On 25.01.2011 10:29, Helmut Jarausch wrote:
>>>> On 01/25/11 10:11:22, Gael Guennebaud wrote:
>>>>> Since it is unbelievable
>>>>> that one day someone will perform a LU or LDLT dec on a matrix
>>>>> requiring 64bits integers for the indices (think about the size of
>>>>> the
>>>>> matrix!), I think it is safe to use int there too.
>>>
>>> Just like 640k ought to be enough for everyone ;)
>>>
>>>> I agree for dense matrices, but a tridiagonal matrix of size 2^32
>>>> isn't out of range (IMHO). So the situation might be different for
>>>> sparse matrices - even in the near future.
>>>
>>> I agree on that. Why not make it default to MatrixType::Index and tell
>>> users to use an appropriate IndexType in the first place? This could be
>>> even smaller if the matrix never reaches sizes above 2^16 (2^15?).
>>
>> I agree with Christian here. DenseIndex is the most sane defaut, and
>> let the user specify smaller integer types.
>>
>> The problem with defaulting to int is that it is neither what dense
>> matrices use, nor the smallest integer type.
>
> The problem is that the index type of Matrix is not really
> controllable.
What do you mean? The user who is ready to break the ABI can control
the default index type for dense objects by defining
EIGEN_DEFAULT_DENSE_INDEX_TYPE.
>Now think about the idea of computing the LU or LDLT
> decomposition of a *dense* matrix larger than 2^32 x 2^32 ...
Not sure that I follow :) you would need a very big computer, and the
result would be 42 anyway :)
Benoit
>
>
> gael
>
>
>> Benoit
>>
>>>
>>> Christoph
>>>
>>> --
>>> ----------------------------------------------
>>> Dipl.-Inf. Christoph Hertzberg
>>> Cartesium 0.051
>>> Universität Bremen
>>> Enrique-Schmidt-Straße 5
>>> 28359 Bremen
>>>
>>> Tel: (+49) 421-218-64252
>>> ----------------------------------------------
>>>
>>>
>>>
>>
>>
>>
>
>
>