[Help-gsl] Help with "make check" failures

[David Gunter]

I am trying to build GSL 1.14 for our Cray XE system using the Cray compilers. 
The build completes but some checks fail. I would like to know how critical 
these failures are. For example, the backward diff functions seem to be 
deprecated so I'm not worried about them.

The Gnu build is fine and passes all tests, of course.

Here are the failures:

FAIL: gsl_ldexp DBL_MAX (inf observed vs 1.79769313486231571e+308 expected) [81]
FAIL: gsl_ldexp(DBL_MIN,2000) (2.55467559620431174e+294 observed vs 
2.55467559620444136e+294 expected) [83]
FAIL:   SV_decomp bigsparse [212]
FAIL: Singular Value Decomposition [66886]
FAIL: gsl_sf_bessel_J0_e(1.0e+10, &r) [78]
FAIL: gsl_sf_bessel_J1_e(1.0e+10, &r) [82]
FAIL: gsl_sf_bessel_Y0_e(4294967296.0, &r) [92]
FAIL: gsl_sf_bessel_Y1_e(4294967296.0, &r) [96]
FAIL: gsl_sf_bessel_Yn_e(100, 4294967296.0, &r) [100]
FAIL: gsl_sf_bessel_Yn_e(1000, 4294967296.0, &r) [101]
FAIL: Bessel Functions [398]
FAIL: gsl_sf_conicalP_half_e(1.0e+8, 1.1, &r) [2405]
FAIL: gsl_sf_conicalP_half_e(1.0e+8, 100.0, &r) [2406]
FAIL: gsl_sf_conicalP_mhalf_e(1.0e+8, 1.1, &r) [2422]
FAIL: gsl_sf_conicalP_mhalf_e(1.0e+8, 100.0, &r) [2423]
FAIL: gsl_sf_legendre_H3d_0_e(1.0e+08, 1.0, &r) [2480]
FAIL: gsl_sf_legendre_H3d_0_e(1.0e+08, 100.0, &r) [2481]
FAIL: Legendre Functions [2551]
FAIL: exp(-x^2), x=0.5, backward diff (-0.778800202757120141 observed vs 
-0.778800783071404878 expected) [23]
FAIL: exp(-x^2), x=0.5, backward diff, valid error estimate [24]

-david
--
David Gunter
HPC-3: Infrastructure Team
Los Alamos National Laboratory