BLAS testing Fatal Errors - 1st time install

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

BLAS testing Fatal Errors - 1st time install

Postby stevekuo » Wed Mar 31, 2010 7:51 pm

Hello all,

I'm new to BLAS and LAPACK and have run into an issue with the internal testing routines. I was wondering if anyone could help shed light on this for me. Here is the beginning of the output of sblat2.out. This is step #4 of the README.install instructions. I saw no compile errors when making f2clib or blaslib.

--------------------------------------------------------------

TESTS OF THE REAL LEVEL 2 BLAS

THE FOLLOWING PARAMETER VALUES WILL BE USED:
FOR N 0 1 2 3 5 9
FOR K 0 1 2 4
FOR INCX AND INCY 1 2 -1 -2
FOR ALPHA .0 1.0 .7
FOR BETA .0 1.0 .9

ROUTINES PASS COMPUTATIONAL TESTS IF TEST RATIO IS LESS THAN 16.00

RELATIVE MACHINE PRECISION IS TAKEN TO BE 1.1E-19

SGEMV PASSED THE TESTS OF ERROR-EXITS

******* FATAL ERROR - COMPUTED RESULT IS LESS THAN HALF ACCURATE *******
EXPECTED RESULT COMPUTED RESULT
1 .449874 .449874
2 -.583124 -.583124
******* SGEMV FAILED ON CALL NUMBER:
8: SGEMV ('N', 2, 1, 1.0, A, 3, X, 1, 1.0, Y, 1) .

SGBMV PASSED THE TESTS OF ERROR-EXITS

******* FATAL ERROR - COMPUTED RESULT IS LESS THAN HALF ACCURATE *******
EXPECTED RESULT COMPUTED RESULT
1 .254575 .254575
2 .346653 .346653
******* SGBMV FAILED ON CALL NUMBER:
17: SGBMV ('N', 2, 1, 0, 0, 1.0, A, 2, X, 1, 1.0, Y, 2) .

(This repeats for all the functions through the rest of the output)
----------------------------------------------------------

I'm running UBUNTU 8.04 on an Intel core i7. Also tried with Ubuntu 9.10 on a core2duo Macbook. I made no changes to the make.inc file other than renaming the example one. It didn't seem like I needed to change any of the settings, but I could be wrong here...

As an aside (which will probably come in handy later), the README talks about optimized BLAS libraries for different machines. Does anyone know which library I should look into for these machines? My goal for this project is to get a speedy SVD routine running on the core i7.

Thanks so much,

steve
stevekuo
 
Posts: 4
Joined: Wed Mar 31, 2010 7:41 pm

Re: BLAS testing Fatal Errors - 1st time install

Postby sven » Thu Apr 01, 2010 7:57 am

Dear Steve,

I think that UBUNTU has the BLAS and LAPACK available as an 'extra' (as with most of the Linux distributions). The BLAS are probably based upon ATLAS, which should give you good efficiency. An alternative is the MKL Library from Intel, which used to be free on Linux, but I suspect that they now charge.

Your tester is giving quite the wrong value of the relative machine precision. For single precision it should be of order 10^(-8). So, you probably need to turn down the level of optimization when compiling sblat2.

Best wishes,

Sven Hammarling.
sven
 
Posts: 144
Joined: Wed Dec 22, 2004 4:28 am

Re: BLAS testing Fatal Errors - 1st time install

Postby stevekuo » Fri Apr 02, 2010 5:04 pm

My NOOPT flag is set to -O0 in the make.inc
I don't see any other places to reduce the optimizations. -O0 is the lowest level correct?

steve
stevekuo
 
Posts: 4
Joined: Wed Mar 31, 2010 7:41 pm

Re: BLAS testing Fatal Errors - 1st time install

Postby admin » Fri Apr 02, 2010 6:20 pm

Steve,
On Macs, Lapack and BLAS come pre-installed in the Veclib framework. You can link against them by using "-framework vecLib".
You will get very good performance. Otherwise under Linux, I would recommend atlas, goto, mkl or acml.
If you send me you make.inc, I may be able to help you. I really do not understand where the problem comes from.
You could download the BLAS and LAPACK packages directly from netlib and give it another try.
Those tests work fine on all my machines (ubuntu, mac, etc...)
Julie
admin
Site Admin
 
Posts: 488
Joined: Wed Dec 08, 2004 7:07 pm


Return to User Discussion

Who is online

Users browsing this forum: Google [Bot], Yahoo [Bot] and 1 guest

cron