Re: [eigen] Re: Release 3.2.1? |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen <eigen@xxxxxxxxxxxxxxxxxxx>
- Subject: Re: [eigen] Re: Release 3.2.1?
- From: Gael Guennebaud <gael.guennebaud@xxxxxxxxx>
- Date: Mon, 24 Feb 2014 17:22:45 +0100
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:in-reply-to:references:from:date:message-id:subject:to :content-type; bh=h3PWev1E4I/haxGrmfoagc8TD5ODuAlmUptUrsODmk4=; b=1Hczq21Jl85IRC00AGOQGqvFs+Q2cesI/l+sw4WYkl79T2KTnEUvVQVbPTvY+dRSvW 8osgvc2najGvBGHwwGnsw3SX44sZoQA6rE4+p9AHpq332YipkUuw9h40gbJa5ym48cw+ KNAPK8c9+TUgVIBcASAYYwiXuSDhtEZMUZOj7OZjPbQ81X/nLTeDGVyAIm69IyO2LfvT 0IqWhc5d+3NpteKQvOEzDEzJ58zDzxs1qRnz2Oh4GNkr0zELujAIhh5evlYT9Ghndeq6 JC40ZY5PlYI0nqyKsSQhZzP1FmIpc1O80AQunTFlXTmRZv08WuzT1eL+pMsqKQ1FI+83 QjrA==
Alright, the problem is that with clang, when an argument is not used,
then the ABI to call this function is not the same. In Eigen,
depending on template parameters, some arguments are only used through
assertions making the lapack interface incompatible with a testing
program compiled in debug mode and that instantiate the same function.
Yes, that's far fetched...
Anyway, the unused argument has really to used, even after dead code
elimination, so the only workaround I found is quite ugly:
https://bitbucket.org/eigen/eigen/commits/1e19a9cb83f0/
Gael.
On Mon, Feb 24, 2014 at 4:09 PM, Gael Guennebaud
<gael.guennebaud@xxxxxxxxx> wrote:
> There only one issue I'd like to investigate with the
> cholmod_support_1 unit test which trigger an assertion through the
> Eigen's lapack interface (dpotrf_):
>
> Assertion failed: (resIncr==1), function run, file
> /Users/guenneba/Eigen/eigen3.2/Eigen/src/Core/products/GeneralMatrixVector.h,
> line 73.
>
> gael
>
>
> On Mon, Feb 24, 2014 at 3:29 PM, Jitse Niesen <jitseniesen@xxxxxxxxx> wrote:
>> On Wed, 12 Feb 2014, I wrote:
>>
>>> Given that it's half a year ago since 3.2.0 is released and a number of
>>> fixes have gone in, how about releasing a new version? I could do this on
>>> Saturday if there are no problems. I just built and ran all tests and got
>>> no failures except for two tests in unsupported (gmres_2 and minres_1;
>>> also reported in bug 715).
>>
>> As you must have noticed, this did not happen. Partially this was because I
>> had less time as expected, but there was also a rush of very welcome
>> bug-fixing which I did not want to interrupt. It seems to have calmed down a
>> bit, so I plan to do the release tomorrow or on Wednesday.
>>
>>
>> Jitse
>>
>>
>>