Bug 138447 - The lapack package is broken
Summary: The lapack package is broken
Keywords:
Status: CLOSED RAWHIDE
Alias: None
Product: Fedora
Classification: Fedora
Component: lapack
Version: 3
Hardware: i386
OS: Linux
medium
high
Target Milestone: ---
Assignee: Ivana Varekova
QA Contact:
URL:
Whiteboard:
: 139404 146447 (view as bug list)
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2004-11-09 12:00 UTC by Terje Røsten
Modified: 2007-11-30 22:10 UTC (History)
6 users (show)

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of:
Environment:
Last Closed: 2005-05-05 07:56:56 UTC
Type: ---
Embargoed:


Attachments (Terms of Use)
Patch to lapack.spec (539 bytes, patch)
2004-11-09 12:06 UTC, Terje Røsten
no flags Details | Diff

Description Terje Røsten 2004-11-09 12:00:58 UTC
Description of problem:

 Two files in the in Lapack package is build with too aggressive
 optimaze flags.

 The files in question is dlamch.f and slamch.f.

 These files are used to determine single and 
 double precision machine parameters.

Version-Release number of selected component (if applicable):

 lapack-3.0-25

How reproducible:

 Try to run the dlamctst program from the Lapack tarball.

Steps to Reproduce:

 $ g77 -o dlamchtst dlamchtst.f -lm -lblas -llapack

 $ ./dlamchtst

Actual results:

  It just hangs.

Expected results:

 Produce this output:
  Epsilon                      =   2.22044605E-16
  Safe minimum                 =   2.22507386E-308
  Base                         =   2.
  Precision                    =   4.4408921E-16
  Number of digits in mantissa =   53.
  Rounding mode                =   0.
  Minimum exponent             =  -1021.
  Underflow threshold          =   2.22507386E-308
  Largest exponent             =   1024.
  Overflow threshold           =   1.79769313E+308
  Reciprocal of safe minimum   =   4.49423284E+307

Additional info:

 This works:

  $ g77 -c -Os dlamch.f
  
  $ g77 -o dlamchtst dlamchtst.f -lm -lblas dlamch.o
  $ ./dlamchtst 
  Epsilon                      =   2.22044605E-16
  Safe minimum                 =   2.22507386E-308
  Base                         =   2.
  Precision                    =   4.4408921E-16
  Number of digits in mantissa =   53.
  Rounding mode                =   0.
  Minimum exponent             =  -1021.
  Underflow threshold          =   2.22507386E-308
  Largest exponent             =   1024.
  Overflow threshold           =   1.79769313E+308
  Reciprocal of safe minimum   =   4.49423284E+307

 The rpm package is compiled with the -O2 flag.
 A patch to the spec file which fix the problem is attached.

Comment 1 Terje Røsten 2004-11-09 12:06:15 UTC
Created attachment 106324 [details]
Patch to lapack.spec

Patch to lapack.spec to fix the problem.

Comment 3 Ivana Varekova 2004-11-12 09:11:22 UTC
Thank you for your notice.
The problem was fixed.
IV

Comment 4 Dmitri A. Sergatskov 2004-11-12 15:59:09 UTC
The -O2 is "safe" optimization and should work. (It used to with 
previous versions of gcc.) In my opinion the
actual issue is that gcc-3.4 is not ISO-compliant with regard to 
excessive presision. I think that reducing optimization helps here is
purely fortuous. The fix would be to add -ffloat-store flag to
FFLAGS and keep -O2. See bugs:
 https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=138791
and
 https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=138683

I still think that this is a gcc issue.

Sincerely,

Dmitri.


Comment 5 Ivana Varekova 2004-12-07 08:20:47 UTC
Solved with -27, closing now.



Comment 6 Ivana Varekova 2005-01-10 15:03:50 UTC
*** Bug 139404 has been marked as a duplicate of this bug. ***

Comment 7 Ivana Varekova 2005-01-28 12:05:13 UTC
*** Bug 146447 has been marked as a duplicate of this bug. ***

Comment 8 Matthew Miller 2005-03-23 14:46:21 UTC
Could this be released as an update to FC3, please? Thanks.

Comment 9 Matthew Miller 2005-03-23 16:26:10 UTC
Especially since this has been dropped from FC4, and so "upgrade to FC4" will
never be a solution, and since there is already a corrected package, making it
work in FC3 seems like the right thing to do.

Given the "is broken" status and the several dups, I'm going to take the liberty
of reopening this.

Comment 10 petrosyan 2005-04-29 17:14:51 UTC
This bug is still present in Fedora Core 3.
A scientific application that uses lapack stopped working in FC3.
This is a very serious bug, since Linux is only good for scientific applications
and most of them use lapack.

Comment 11 Dmitri A. Sergatskov 2005-04-29 17:20:02 UTC
RedHat gave up on scientists (they dropped lapack from upcoming FC4 release).
In a mean time you can get a working copy from:

ftp://coffee.phys.unm.edu/pub/octave/RPMS/i386/

Dmitri.
-- 


Comment 12 Matthew Miller 2005-04-29 17:35:27 UTC
Dmitri -- lapack isn't gone. It's in Fedora Extras now. However, it'd still be
nice for there to be an update for FC3.

Comment 13 petrosyan 2005-05-04 20:25:31 UTC
this bug has been fixed in the latest lapack-3.0-26.fc3 update

Comment 14 Matthew Miller 2005-05-04 20:40:11 UTC
Cool, thank you very very much, Red Hat folks.

Comment 15 Ivana Varekova 2005-05-05 07:56:56 UTC
The new fc3 version lapack-3.0-26.fc3 fixes this problem (and two other bugs) in
fc3.
Thank you for your notice.
Ivana Varekova

Comment 16 Matthew Miller 2005-05-05 12:12:45 UTC
Not to harass, but will there be an announcement mail for the fc3 update? It is
very helpful to end-users when every update that appears in the tree has a
matching mail message. Thanks!

Comment 17 Matthew Miller 2005-05-05 14:32:07 UTC
Got the announcement. Thank you again! I hope this puts "Red Hat gave up on
scientists" to rest.


Note You need to log in before you can comment on or make changes to this bug.