bug-gsl
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Bug-gsl] make check fails for Intel, PGI and Pathscale Compilers


From: Ulf Markwardt
Subject: [Bug-gsl] make check fails for Intel, PGI and Pathscale Compilers
Date: Fri, 19 May 2006 17:36:57 +0200
User-agent: KMail/1.8.2

Hello there, 

I start with the second question: Is there a convenient way to let GSL use MKL 
or ACML instead of the native BLAS?

The more serious problem is this: 
I am trying to install GSL v.1.8 on an Opteron system (SLES 9). After 
compiling just with -O2 the "make check" fails with the following messages. 
----------------------------------------------------------------
PGI:
----------------------------------------------------------------
FAIL: gsl_complex_polar imag part at (r=-1.5,t=0) (-1.5 observed vs -0 
expected) [2]
...
Thes failures are reported for gsl_complex_log, gsl_complex_log10, 
gsl_complex_exp, gsl_complex_sinh, gsl_complex_cosh, gsl_complex_csch, 
gsl_complex_sech as well.

----------------------------------------------------------------
Pathscale 2.4 :
----------------------------------------------------------------
FAIL: gsl_vector_char_scale by 2 stride=1, N=87 [7519]
FAIL: gsl_vector_char_add_constant stride=1, N=87 [7520]
...
The same for .._uchar_..

----------------------------------------------------------------
Intel 9.1
----------------------------------------------------------------
FAIL:   SV_decomp nan5 [197]
FAIL: Singular Value Decomposition [66871]
singular value 4 =                      5 vs previous                      0
FAIL:   SV_decomp_mod nan5 [167097]
FAIL: Singular Value Decomposition (Mod) [233771]
FAIL: test

===========================

I did some double-checks and forced the compilers to strict IEEE arithmetics:

PGI: -Kieee shows no different results
       -O0 -Kieee reports:
                FAIL: gsl_block_char_fprintf and fscanf [136]
                FAIL: gsl_block_char_write and read [149]

Pathsale: -fno_unsafe-math-optimizations  shows no different results
Intel: -mp WORKS WELL! - No failures reported.


Do you have any idea what goes wrong? 

Thank You
Ulf


-- 
______________________________________________
Dr. Ulf Markwardt

TU Dresden
Center for Information Services
and High Performance Computing (ZIH)
D-01062 Dresden
Germany

Phone:    (+49) 351/463-33640
Fax:      (+49) 351/463-37773

e-mail: address@hidden
WWW:    http://www.tu-dresden.de/zih




reply via email to

[Prev in Thread] Current Thread [Next in Thread]