[GenABEL-dev] Fwd: [genabel-Bugs][1889] Different results when running ProbABEL/palinear with mmscore on 32bit or 64bit Linux

[Yurii Aulchenko]

Dear All,

For these of you who use --mmscore option on 64-bit platform, the bug
message below is highly relevant.

Many thanks, Lennart, for spotting this problem and coming up with
potential solution.

Best wishes,
Yurii

----------------------
Yurii Aulchenko
Independent consultant
(sent from mobile device)

> -------- Original Message --------
> Subject: [genabel-Bugs][1889] Different results when running
> ProbABEL/palinear with mmscore on 32bit or 64bit Linux
> Date: Sun, 18 Mar 2012 11:08:57 +0100 (CET)
> From: <genabel-bugs at r-forge.wu-wien.ac.at>
> Reply-To: genabel-bugs at r-forge.wu-wien.ac.at
> To: noreply at r-forge.wu-wien.ac.at
>
> Bugs item #1889, was changed at 2012-03-18 10:23 by Lennart Karssen
> You can respond by visiting:
> https://r-forge.r-project.org/tracker/?func=detail&atid=2058&aid=1889&group_id=505
>
> Status: Open
> Priority: 5
> Submitted By: Lennart Karssen (lckarssen)
> Assigned to: Lennart Karssen (lckarssen)
> Summary: Different results when running ProbABEL/palinear with mmscore
> on 32bit or 64bit Linux
>> Resolution: Fixed
> Operating System: Linux (64bit)
> Severity: critical
> Hardware: All
>> Version: other
> Component: ProbABEL
> URL:
>
>
> Initial Comment:
> As pointed out in the following post on the mailing list:
> http://lists.r-forge.r-project.org/pipermail/genabel-devel/2012-March/000392.html,
> ProbABEL's palinear gives different results when running on a 32bit or a
> 64bit platform.
> This bug was found in ProbABEL 0.2.0-beta, but checks with older
> versions of ProbABEL show that it has been present also in earlier
> version. At first sight it seems to have something to do with using
> FileVector files as input and using the mmscore option, because the
> examples/example_mms.sh test script fails (on 64bit Linux, it passes on
> 32bit Linux). This was tested on Ubuntu 11.04 (64 and 32 bit) and Ubuntu
> 11.10 (64bit).
>
> Attached the output of the first test in the example_mms.sh script.
>
> ----------------------------------------------------------------------
>
>> Comment By: Lennart Karssen (lckarssen)
> Date: 2012-03-18 11:08
>
> Message:
> This bug is fixed in svn r. 852.
>
> It turned out to be a problem with the inverse variance matrix and how
> that was passed on to the regression code.
> It seems that this bug has been present ever since the reg1.h file has
> been in svn (r.17).
>
> Now ProbABEL passes "make check" and "make distcheck" also on 64bit
> platforms.
>
> ----------------------------------------------------------------------
>
> You can respond by visiting:
> https://r-forge.r-project.org/tracker/?func=detail&atid=2058&aid=1889&group_id=505
>