GSL 0.9 Fails in test with VC++ 6.0

GSL 0.9 Fails in test with VC++ 6.0

[Jose Miguel Buenaposada Biencinto]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 3200 bytes --]

Hi again Brian,

Here we are trying to build a functional library of GSL 0.9 with Visual
C++ 6.0.

When we build it in debug configuration there is no errors at all in the
test cases, but
when we build the release versions of the libraries and the test
programs we
get 28 failures (See below). It seems to be related with the
optimisation flags
("maximizate speed" choosen, /O2 flag) because if we build all, in the
release
configuration, but changing the optimization to "default" then we get 0
errors.

This is a issue to investigate in more depth, I think ;-).

Best regards.
José Miguel.

--------------------->8--------------------------------------------------------

FAIL: qagiu(f455) smooth elist (3.1474288913331199e-014
            observed vs 3.1473804321981764e-014 expected)

[...]

FAIL: filip gsl_fit_multilinear cov00 (88854.396979064302 observed vs
88854.387618829263 expected)
FAIL: filip gsl_fit_multilinear cov11 (313353.53093073762 observed vs
313353.49779115315 expected)
FAIL: filip gsl_fit_multilinear cov22 (217601.34836382096 observed vs
217601.32529824311 expected)
FAIL: filip gsl_fit_multilinear cov33 (51621.78781601938 observed vs
51621.782340961217 expected)
FAIL: filip gsl_fit_multilinear cov44 (5133.4172587447338 observed vs
5133.4167149056166 expected)
FAIL: filip gsl_fit_multilinear cov55 (233.77549738541126 observed vs
233.7754726891439 expected)
FAIL: filip gsl_fit_multilinear cov66 (5.0037740242150477 observed vs
5.0037734979842003 expected)
FAIL: filip gsl_fit_multilinear cov77 (0.049117345206935024 observed vs
0.049117340072805889 expected)
FAIL: filip gsl_fit_multilinear cov88 (0.0002026744316204776 observed vs
0.00020267441059574351 expected)
FAIL: filip gsl_fit_multilinear cov99 (2.8688603819991648e-007 observed
vs 2.8688600870584576e-007 expected)
FAIL: filip gsl_fit_multilinear cov1010 (8.039505268652566e-011 observed
vs 8.039504450571134e-011 expected)

[...]

FAIL: gsl_sf_legendre_Plm_e(100, 5, 0.999, &r)
  expected:   -1983161080.380621
  obtained:    -1983161080.38064    0.02135698560189238  1.07692e-011
  fracdiff: 4.628527347686955e-015
  value not within tolerance of expected value
     -1983161080.3806396    0.021356985601892382

[...]

FAIL: gsl_sf_legendre_sphPlm_e(10, 0, 0.999, &r)
  expected:    1.222575412279739
  obtained:    1.222575412279737   2.307463332491048e-015  1.88738e-015
  fracdiff: 5.448611251987747e-016
  value not within tolerance of expected value
      1.2225754122797372  2.307463332491048e-015

[...]

FAIL: gsl_isnan(nan) (0 observed vs 1 expected)
[...]
FAIL: gsl_finite(nan) (1 observed vs 0 expected)









--
-------------------------------------------------------------------
Jose Miguel Buenaposada Biencinto
Facultad de Informática                  | Computer Science School
Universidad Politécnica de Madrid  (UPM) | Madrid Tech. University.
Campus de Montegancedo s/n
28660 Boadilla del Monte, Madrid (ESPAÑA)| (SPAIN)
Voice    +34 91 336 69 47
Fax      +34 91 352 48 19
e-mail:  jmbuena@dia.fi.upm.es
web   :  http://www.dia.fi.upm.es/~jmbuena
-------------------------------------------------------------------
Yo soy Precario FPU, ¿ y tu ?: http://www.precarios.org

Re: GSL 0.9 Fails in test with VC++ 6.0

[Brian Gough]

Jose Miguel Buenaposada Biencinto writes:
 >  Here we are trying to build a functional library of GSL 0.9 with
 > Visual C++ 6.0.
 >  When we build it in debug configuration there is no errors at all
 > in the test cases, but when we build the release versions of the
 > libraries and the test programs we get 28 failures (See below). It
 > seems to be related with the optimisation flags ("maximizate speed"
 > choosen, /O2 flag) because if we build all, in the release
 > configuration, but changing the optimization to "default" then we
 > get 0 errors.

I have checked in for the first two problems (integration and
multifit).

Could you try the following patch and let me know if it works for the
specfunc tests. Thanks.


Index: test_legendre.c
===================================================================
RCS file: /cvs/gsl/gsl/specfunc/test_legendre.c,v
retrieving revision 1.18
diff -u -r1.18 test_legendre.c
--- test_legendre.c	2001/07/12 14:26:32	1.18
+++ test_legendre.c	2001/07/30 16:51:45
@@ -96,7 +96,7 @@
   TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, -0.5, &r), -6.617107444248382171e+08, TEST_TOL0, GSL_SUCCESS);
   TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, 1.0e-08, &r), 817.8987598063712851, TEST_TOL0, GSL_SUCCESS);
   TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, 0.5, &r), 6.617107444248382171e+08, TEST_TOL0, GSL_SUCCESS);
-  TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, 0.999, &r), -1.9831610803806212189e+09, TEST_TOL1, GSL_SUCCESS);
+  TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, 0.999, &r), -1.9831610803806212189e+09, TEST_TOL2, GSL_SUCCESS);
 
   sa = 0;
   gsl_sf_legendre_Plm_array(100, 5, 0.5, L);
@@ -108,7 +108,7 @@
 
   TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 0, -0.5, &r), -0.24332702369300133776, TEST_TOL0, GSL_SUCCESS);
   TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 0, 0.5, &r), -0.24332702369300133776, TEST_TOL0, GSL_SUCCESS);
-  TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 0, 0.999, &r), 1.2225754122797385990, TEST_TOL0, GSL_SUCCESS);
+  TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 0, 0.999, &r), 1.2225754122797385990, TEST_TOL1, GSL_SUCCESS);
 
   TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 5, -0.5, &r),    -0.3725739049803293972,     TEST_TOL0, GSL_SUCCESS);
   TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 5, 1.0e-08, &r), -3.1377233589376792243e-08, TEST_TOL0, GSL_SUCCESS);

Re: GSL 0.9 Fails in test with VC++ 6.0

[Brian Gough]

Jose Miguel Buenaposada Biencinto writes:
 >  When we build it in debug configuration there is no errors at all
 > in the test cases, but when we build the release versions of the
 > libraries and the test programs we get 28 failures (See below). It
 > seems to be related with the optimisation flags ("maximizate speed"
 > choosen, /O2 flag) because if we build all, in the release
 > configuration, but changing the optimization to "default" then we
 > get 0 errors.
 >  This is a issue to investigate in more depth, I think ;-).

Thanks for the bug reports.

I only have MSVC++ 'Standard Edition', which does not include
optimization, so I will have to rely on you here.  The 'Professional
Edition' with optimization was too expensive for me.

Most of those errors look ok, and we can fix them by
increasing the tolerances in the tests.

 >  FAIL: gsl_isnan(nan) (0 observed vs 1 expected) [...]  
 >  FAIL: gsl_finite(nan) (1 observed vs 0 expected)

This error is bad.  If you can find the cause please let me know, as I
cannot debug it myself.

Brian

Re: GSL 0.9 Fails in test with VC++ 6.0

[Jose Miguel Buenaposada Biencinto]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1932 bytes --]

Brian Gough wrote:

> Jose Miguel Buenaposada Biencinto writes:
>  >  Here we are trying to build a functional library of GSL 0.9 with
>  > Visual C++ 6.0.
>  >  When we build it in debug configuration there is no errors at all
>  > in the test cases, but when we build the release versions of the
>  > libraries and the test programs we get 28 failures (See below). It
>  > seems to be related with the optimisation flags ("maximizate speed"
>  > choosen, /O2 flag) because if we build all, in the release
>  > configuration, but changing the optimization to "default" then we
>  > get 0 errors.
>
> I have checked in for the first two problems (integration and
> multifit).
>
> Could you try the following patch and let me know if it works for the
> specfunc tests. Thanks.

Ok, I have checked out current gsl version from cvs and applied the
patch (increasing the tolerances in some places of test_legendre.c)
that you sent, by hand.

When I compile the whole thing (libs and test programs) with the
"Maximize speed" optimisation option (/O2 flag) the only errors I get
now are:

[...]
FAIL: gsl_isnan(nan) (0 observed vs 1 expected)
[...]
FAIL: gsl_finite(nan) (1 observed vs 0 expected)

So only the difficult problems remains ...

I can not try it today and I'm going to go out for three months so, maybe
somebody else can debug it.

Regards.

--
-------------------------------------------------------------------
Jose Miguel Buenaposada Biencinto
Facultad de Informática                  | Computer Science School
Universidad Politécnica de Madrid  (UPM) | Madrid Tech. University.
Campus de Montegancedo s/n
28660 Boadilla del Monte, Madrid (ESPAÑA)| (SPAIN)
Voice    +34 91 336 69 47
Fax      +34 91 352 48 19
e-mail:  jmbuena@dia.fi.upm.es
web   :  http://www.dia.fi.upm.es/~jmbuena
-------------------------------------------------------------------
Yo soy Precario FPU, ¿ y tu ?: http://www.precarios.org

Re: GSL 0.9 Fails in test with VC++ 6.0

[Jose Miguel Buenaposada Biencinto]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 3325 bytes --]

Brian Gough wrote:

> Jose Miguel Buenaposada Biencinto writes:
>  >  Here we are trying to build a functional library of GSL 0.9 with
>  > Visual C++ 6.0.
>  >  When we build it in debug configuration there is no errors at all
>  > in the test cases, but when we build the release versions of the
>  > libraries and the test programs we get 28 failures (See below). It
>  > seems to be related with the optimisation flags ("maximizate speed"
>  > choosen, /O2 flag) because if we build all, in the release
>  > configuration, but changing the optimization to "default" then we
>  > get 0 errors.
>
> I have checked in for the first two problems (integration and
> multifit).
>
> Could you try the following patch and let me know if it works for the
> specfunc tests. Thanks.

Yes, I'll make a try and see if that works ... of course I can try to solve the other
problem with isNan and isInf when using optimisation and MSVC++ 6.0. The
problem is that I'm going to stay for three months out there and I don't know
if I'm goint to have access to MSVC++ 6.0 with optimisations ...

Regards.


>
>
> Index: test_legendre.c
> ===================================================================
> RCS file: /cvs/gsl/gsl/specfunc/test_legendre.c,v
> retrieving revision 1.18
> diff -u -r1.18 test_legendre.c
> --- test_legendre.c     2001/07/12 14:26:32     1.18
> +++ test_legendre.c     2001/07/30 16:51:45
> @@ -96,7 +96,7 @@
>    TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, -0.5, &r), -6.617107444248382171e+08, TEST_TOL0, GSL_SUCCESS);
>    TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, 1.0e-08, &r), 817.8987598063712851, TEST_TOL0, GSL_SUCCESS);
>    TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, 0.5, &r), 6.617107444248382171e+08, TEST_TOL0, GSL_SUCCESS);
> -  TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, 0.999, &r), -1.9831610803806212189e+09, TEST_TOL1, GSL_SUCCESS);
> +  TEST_SF(s, gsl_sf_legendre_Plm_e, (100, 5, 0.999, &r), -1.9831610803806212189e+09, TEST_TOL2, GSL_SUCCESS);
>
>    sa = 0;
>    gsl_sf_legendre_Plm_array(100, 5, 0.5, L);
> @@ -108,7 +108,7 @@
>
>    TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 0, -0.5, &r), -0.24332702369300133776, TEST_TOL0, GSL_SUCCESS);
>    TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 0, 0.5, &r), -0.24332702369300133776, TEST_TOL0, GSL_SUCCESS);
> -  TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 0, 0.999, &r), 1.2225754122797385990, TEST_TOL0, GSL_SUCCESS);
> +  TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 0, 0.999, &r), 1.2225754122797385990, TEST_TOL1, GSL_SUCCESS);
>
>    TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 5, -0.5, &r),    -0.3725739049803293972,     TEST_TOL0, GSL_SUCCESS);
>    TEST_SF(s, gsl_sf_legendre_sphPlm_e, (10, 5, 1.0e-08, &r), -3.1377233589376792243e-08, TEST_TOL0, GSL_SUCCESS);

--
-------------------------------------------------------------------
Jose Miguel Buenaposada Biencinto
Facultad de Informática                  | Computer Science School
Universidad Politécnica de Madrid  (UPM) | Madrid Tech. University.
Campus de Montegancedo s/n
28660 Boadilla del Monte, Madrid (ESPAÑA)| (SPAIN)
Voice    +34 91 336 69 47
Fax      +34 91 352 48 19
e-mail:  jmbuena@dia.fi.upm.es
web   :  http://www.dia.fi.upm.es/~jmbuena
-------------------------------------------------------------------
Yo soy Precario FPU, ¿ y tu ?: http://www.precarios.org