[Insight-developers] Thread-safety issues - netlib

[Miller, James V (Research)]

A first pass at removing the statics:

The only vnl test that does NOT pass is one of the 30 subtests in test_qr.  I am tracking 
this down. The calculation error increases by one decimal place without the statics. I 
have run all the ITK as well and ImageRegistration9 does not pass but that may be due to 
other changes I have on my system.

Of the 3 routines I mentioned early that used COMMON blocks, it turns out that there
are only two.  One, the lbfgs really only uses the COMMON blocks to share read only 
variables.  I thinnk that can be converted without an issue. The other, dsrc2c uses
several COMMON blocks and reads and writes to these blocks.  I'll have to dig deeper to
see if there is a single entry point to these routines that could create these
"global" structs and pass them to the other routines.

Jim



-----Original Message-----
From: insight-developers-bounces+millerjv=crd.ge.com at itk.org
[mailto:insight-developers-bounces+millerjv=crd.ge.com at itk.org]On Behalf
Of Miller, James V (Research)
Sent: Monday, July 11, 2005 12:53 PM
To: Karthik Krishnan
Cc: Insight Developers List
Subject: RE: [Insight-developers] Thread-safety issues - netlib


Karthik,

Dan Blezek and I are looking at the f2c output.  We might be able to put together an 
awk script to convert the existing code.  I look through the original fortran code
and there are only 3 routines that use COMMON blocks.  So it should be safe to 
replace the statics on the rest of them.

We'll make a pass at converting the files and see if things compile and the current
tests pass.

Jim

-----Original Message-----
From: Karthik Krishnan [mailto:Karthik.Krishnan at kitware.com]
Sent: Monday, July 11, 2005 12:08 PM
To: Miller, James V (Research)
Cc: Insight Developers List
Subject: Re: [Insight-developers] Thread-safety issues - netlib


Yes, I am going to weed out some statics and replace them with const 
where appropriate to get at least vnl_svd to work. I was worried about 
going through the code to account for instances where static is really 
needed. (eg a recursive implementation where a function is called 
multiple times, but I don't think that is ever done in netlib).

Thanks
regards
karthik


Miller, James V (Research) wrote:

>Karthik, 
>
>When you use f2c to generate a file without statics, are the differences
>between that file and the one in vxl just the "statics"?  Or is the code
>different in another manner?
>  
>
The code is mostly the same expect for removing statics and some other 
minor changes like
replacing 2.f with (float) 2.0  and variable expansion in places.. etc.

>We should be able to dig up the original fortran.  But if the only difference
>between the f2c generated code with and without statics is to not use the
>keyword static on those lines, when we should just be able to edit them out.
>
>I am not sure how much editting the vxl folks did on the f2c generated code.
>I am aware of one change where they change the mapping of a "long".
>
>Is it possible to run a quick experiment where we edit f2c.h and insert a
>
>#define static
>
>Jim
>
>
>-----Original Message-----
>From: insight-developers-bounces+millerjv=crd.ge.com at itk.org
>[mailto:insight-developers-bounces+millerjv=crd.ge.com at itk.org]On Behalf
>Of Karthik Krishnan
>Sent: Sunday, July 10, 2005 2:31 AM
>To: Insight Developers List
>Subject: [Insight-developers] Thread-safety issues - netlib
>
>
>Hi,
>
>This discussion has cropped up before. I was having some issues with
>thread safety of LAPACK. 
>
>The recently added class itkDiffusionTensor3DReconstructionImageFilter
>is threaded, but the threading does not work (at present). Users of the
>filter need to SetNumberOfThreads(1). Turns out this is because of the
>static defs in dsvdc.c and other related files.
>
>LAPACK's thread safety (therefore vnl's thread safety):
>http://sourceforge.net/mailarchive/forum.php?thread_id=7325117&forum_id=2132
>
>In a recent mail from Andrew Fitzgabbon, he suspects not all statics
>must be removed.
>http://www.itk.org/mailman/private/insight-developers/2005-May/006637.html
>
>The issue has been on vxl's todo list for a while
>http://cvs.sourceforge.net/viewcvs.py/vxl/vxl/TODO.txt?rev=1.12
>
>f2c seems to have some options to generate code without static defs and
>I've been able to compile and generate a few LAPACK C routines from the
>provided FORTRAN files in netlib. 
>
>Invoke f2c with the following options
>f2c -a -C++ file.f
>
>This does produce cleaner code. However some of the vxl folks seem to
>have modified some of the .C files after conversion. Besides I can't
>seem to find all the .f corresponding to the .c files. 
>
>I am wondering if the solution is simply to start weeding out static
>defs from netlib or to look for better faster libraries.. 
>
>Thanks
>Regards
>Karthik
>
>_______________________________________________
>Insight-developers mailing list
>Insight-developers at itk.org
>http://www.itk.org/mailman/listinfo/insight-developers
>_______________________________________________
>Insight-developers mailing list
>Insight-developers at itk.org
>http://www.itk.org/mailman/listinfo/insight-developers
>
>  
>
_______________________________________________
Insight-developers mailing list
Insight-developers at itk.org
http://www.itk.org/mailman/listinfo/insight-developers