LAPACK 3.1.0 stalling during xlintsts test

Open discussion regarding features, bugs, issues, vendors, etc.

LAPACK 3.1.0 stalling during xlintsts test

Postby anmunoz » Wed Feb 21, 2007 6:36 pm

Hi Julien:

- I'm installing LAPACK 3.1.0 on a Sun UltraSPARC III Cu.

- The compiler I'm using is f77: Sun WorkShop 6 update 2 FORTRAN 77 5.3 2001/05/15

- The options I use are -u -f -dalign -native -xO5, although I tried with -xO4 and -xO3

- The BLAS libraries I use are those of ATLAS 3.6.0 which I compiled and tested succesfully

My problem is that the testing stalls in the line

./xlintsts < stest.in > stest.out 2>&1

if I run ./xlintsts < stest.in

it will stall after this lines:

M = 50, N = 10, type 5, test 3, ratio = 0.55326E+06
M = 50, N = 10, type 6, test 1, ratio = 0.49428E+06
M = 50, N = 10, type 6, test 2, ratio = 0.18626E+06
M = 50, N = 10, type 6, test 3, ratio = 0.72327E+06

A user called Dominique told me that there could be a problem with the optimization options I used (if I understood correctly) and when I tried without optimization options i.e. f77 -u -f the test stalls in a different place:

TRANS='N', M= 5, N= 10, NRHS= 15, NB= 3, type 2, test( 2)= 0.18796E+07
TRANS='T', M= 5, N= 10, NRHS= 15, NB= 3, type 2, test( 2)= 0.11708E+07
TRANS='N', M= 5, N= 10, NRHS= 15, NB= 20, type 2, test( 2)= 0.18971E+07
TRANS='T', M= 5, N= 10, NRHS= 15, NB= 20, type 2, test( 2)= 0.82252E+06

but stalls nonetheless, another user mentioned that switching from f77 to g77 solve the problem and that's what I'm going to try no exept that f77 is specially tailored for my architecture (or so I read) and I would very much like to use it if possible.

Do you have any ideas about what the problem might be or about what might I be doing wrong?

thank you very much,

Andrés
anmunoz
 
Posts: 15
Joined: Fri May 12, 2006 4:18 pm

Postby anmunoz » Wed Feb 21, 2007 6:40 pm

I forgot to ask you, how much time I should let pass before deciding it stalls? is four hours in a single line enough?

thanks!

Andrés
anmunoz
 
Posts: 15
Joined: Fri May 12, 2006 4:18 pm

Postby anmunoz » Wed Feb 21, 2007 7:56 pm

Hi Julien:

I tried g77 with the options:

-O3 -funroll-loops

And all the test worked fine so the problem is with f77
anmunoz
 
Posts: 15
Joined: Fri May 12, 2006 4:18 pm

Postby Julien Langou » Wed Feb 21, 2007 8:19 pm

Thanks Andres for trying g77.

Well, fine, I'm happy that you get one way to have lapack installed but we'll need to have
a look to see if we can reproduce the problem with f77 and then look at it. This should
work... Can you try f77 on your Sun but with the reference BLAS, please?

That will spare us some time.

(
type make blaslib
then edit the make.inc to change BLASLIB
then go in TESTING, remove xlintsts
go in LIN, type make (this should create xlintsts back but with the reference BLAS)
go one directory up in TESTING and type
./xlintsts < stest.in
)

To answer your question, after 4 hours of stalling in the LAPACK testing, you can defintely
hit CTRL+C and post on the forum :). More seriously, each executables should not take
more than 5 minutes with the reference BLAS (so less with an optimized BLAS). If
you want to follow even more closely where the stalling happen, you can edit the file
stest.in and change the 'Threshold value of test ratio' to 0.0. All the test will then fail and
this will print you one output per test. Each test is extremely fast.

Julien.
Julien Langou
 
Posts: 835
Joined: Thu Dec 09, 2004 12:32 pm
Location: Denver, CO, USA

Postby Julie » Wed Feb 21, 2007 9:02 pm

Hi Andres

Could you try the following flags without any optimization (I do not think it will make a difference...):
Code: Select all
-silent -u -f -dalign 

These work fine for me for my 2 Sun machines with the atlas library:
Code: Select all
*** First Sun Machine ***
uname -a
SunOS beaker 5.9 Generic_118558-09 sun4u sparc SUNW,Sun-Blade-100
config.guess
beaker-sparc-sun-solaris2.9
f77 -V
NOTICE: Invoking /usr/local/SUNWspro/bin/f90 -f77 -ftrap=%none -V
f90: Sun Fortran 95 8.0 2004/07/15
***************************
*** Second Sun Machine ***
uname -a
SunOS enterprise 5.9 Generic_117171-13 sun4u sparc SUNW,Ultra-4
config.guess
enterprise-sparc-sun-solaris2.9
f77 -V
NOTICE: Invoking /usr/local/SUNWspro/bin/f90 -f77 -ftrap=%none -V
f90: Sun Fortran 95 8.0 2004/07/15


If you could use the Sun f90 compiler (same than me), that may solve your problem.

If you really want the best from your machine, why not give it a try to the sunperf library from Sun..this will be for sure the most optimized code you can run on your machine. (it is based right now on the LAPACK 3.0)

Hope it helps
Julie
Julie
 
Posts: 299
Joined: Wed Feb 23, 2005 12:32 am
Location: ICL, Denver. Colorado

Postby anmunoz » Thu Feb 22, 2007 1:40 pm

Hi Julie:

Thank you very much for your help I didn't know that all the libraries where already in sunperf, I will certainly give it a try, and also I will try using f90 -f77.

I will keep also trying with f77 with Julien's help to see if we can find the problem but I will use your suggestions and tell you how did it go.

thanks,

Andrés
anmunoz
 
Posts: 15
Joined: Fri May 12, 2006 4:18 pm

Postby anmunoz » Thu Feb 22, 2007 1:47 pm

Hi Julien:

I tried what you told me and there are several things I want to tell you about:

- the first think I did was to I build the LAPACK and BLAS libraries using f77 with the options I mentioned initially in this post

- Then what happens is that when I use the make command on /TESTING/LIN I get an error:

Undefined first referenced
symbol in file
c_abs ../../blas_Earth.a(crotg.o)
ld: fatal: Symbol referencing errors. No output written to ../xlintstc
*** Error code 1
make: Fatal error: Command failed for target `../xlintstc'

the funny thing is that if I use my ATLAS libraries or the Sunperf libraries that Julie mentioned I get a completely different error:

"derrab.f", line 45: Error: syntax error at or near symbol (
"derrab.f", line 50: Error: attempt to use undefined variable "srnamt"
*** Error code 1
make: Fatal error: Command failed for target `derrab.o'

since those errors take place after xlintsts is built I went ahead and changed the threshold value of test ratio to 0.0 and it didn't stall at all, it just went all the way into the end and said that it failed all tests.

what do you think of this?

thank you very much for your help,

Andrés
anmunoz
 
Posts: 15
Joined: Fri May 12, 2006 4:18 pm

Postby anmunoz » Thu Feb 22, 2007 1:49 pm

I forgot to mention that even though it doesn't take forever there are some lines it takes much longer than the rest comparatively speaking, what I will do in the meantime is to increase the threshold gradually and see what happens

Andrés
anmunoz
 
Posts: 15
Joined: Fri May 12, 2006 4:18 pm

Postby Julie » Thu Feb 22, 2007 2:00 pm

Andres

Undefined first referenced
symbol in file
c_abs ../../blas_Earth.a(crotg.o)
ld: fatal: Symbol referencing errors. No output written to ../xlintstc
*** Error code 1
make: Fatal error: Command failed for target `../xlintstc'

That error is easy to fix it comes from the call to the Intrinsic Function CABS.
I do not understand why your FORTRAN 77 compiler does not have CABS....
Anyway you can fix it by adding -lg2c in the linking sequence

"derrab.f", line 45: Error: syntax error at or near symbol (
"derrab.f", line 50: Error: attempt to use undefined variable "srnamt"
*** Error code 1
make: Fatal error: Command failed for target `derrab.o'

that one is also easy to fix..
at line 45 replace
Code: Select all
CHARACTER(32)      SRNAMT
by
CHARACTER*32      SRNAMT


These 2 errors shows that your are not using a real F77 complier..your seems to use a wrapper from C. Sometime they called something f77 but this is not a true FORTRAN compiler.

Hope it helps
Julie
Julie
 
Posts: 299
Joined: Wed Feb 23, 2005 12:32 am
Location: ICL, Denver. Colorado

Postby anmunoz » Thu Feb 22, 2007 2:07 pm

Hi Julie:

Thanks for your answer, do you know if there is a way of finding out wether my f77 is a wrapper for C besides the errors it gives me?

Andrés
anmunoz
 
Posts: 15
Joined: Fri May 12, 2006 4:18 pm

Postby anmunoz » Fri Mar 02, 2007 6:07 pm

Hi Julie and Julien:

I think I give up at least for the moment in finding the problem with f77 since g77 has no problem and thanks to Julie I learned that the libraries are already available in the SunOS which I used to compile the software I need.

Thank you very much for all your help.

Andrés
anmunoz
 
Posts: 15
Joined: Fri May 12, 2006 4:18 pm


Return to User Discussion

Who is online

Users browsing this forum: No registered users and 4 guests