lapack installation troubles...

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

lapack installation troubles...

Postby mj » Fri Apr 08, 2005 9:24 am

Hello,

We are trying to install LAPACK package on our PC,whose
main characteristics can be found below, together with the main parameters
we have used for the package compilation. Nevertheless, we have experienced
some troubles during LAPACK testing process. System error messages are showed below.
We would be very thankful if you could help us to solve these problems.

mj

=======================================================================
Intel Pentium 4 2.80GHz
cygwin ver. 1.5.14-1
gcc ver. 3.4.1-1

CLAPACK, version 3.0

####################################################################
# LAPACK make include file. #
# LAPACK, Version 3.0 #
# June 30, 1999 #
####################################################################
#
SHELL = /bin/sh

PLAT = _LINUX

CC = gcc
CFLAGS = -funroll-all-loops -O3
LOADER = gcc
LOADOPTS = $(CFLAGS)
NOOPT =
DRVCFLAGS = $(CFLAGS)
F2CCFLAGS = $(CFLAGS) -DNO_ONEXIT
ARCH = ar
ARCHFLAGS= cr
RANLIB = ranlib
BLASLIB = ../../blas$(PLAT).a
LAPACKLIB = lapack$(PLAT).a
F2CLIB = ../../F2CLIBS/libF77.a ../../F2CLIBS/libI77.a
TMGLIB = tmglib$(PLAT).a
EIGSRCLIB = eigsrc$(PLAT).a
LINSRCLIB = linsrc$(PLAT).a


----
sgd.out
-----
SXV routines passed the tests of the error exits ( 87 tests done)

SXV -- Real Expert Eigenvalue/vector problem driver
Matrix types:

TYPE 1: Da is diagonal, Db is identity,
A = Y^(-H) Da X^(-1), B = Y^(-H) Db X^(-1)
YH and X are left and right eigenvectors.

TYPE 2: Da is quasi-diagonal, Db is identity,
A = Y^(-H) Da X^(-1), B = Y^(-H) Db X^(-1)
YH and X are left and right eigenvectors.


Tests performed:
a is alpha, b is beta, l is a left eigenvector,
r is a right eigenvector and ' means transpose.
1 = max | ( b A - a B )' l | / const.
2 = max | ( b A - a B ) r | / const.
3 = max ( Sest/Stru, Stru/Sest ) over all eigenvalues
4 = max( DIFest/DIFtru, DIFtru/DIFest ) over the 1st and 5th eigenvectors

Type= 2, IWA= 5, IWB= 1, IWX= 1, IWY= 1, result 4 is 1448.19
.........
.......
Type= 2, IWA= 5, IWB= 4, IWX= 2, IWY= 2, result 4 is 148.82
SXV drivers: 37 out of 5000 tests failed to pass the threshold

-----------------------------------------------------------------------
---------
csep.out
------------


All tests for CST passed the threshold ( 4662 tests run)

CST -- Complex Hermitian eigenvalue problem
Matrix types (see xDRVST for details):

Special Matrices:
1=Zero matrix. 5=Diagonal: clustered entries.
2=Identity matrix. 6=Diagonal: large, evenly spaced.
3=Diagonal: evenly spaced entries. 7=Diagonal: small, evenly spaced.
4=Diagonal: geometr. spaced entries.
Dense Hermitian Matrices:
8=Evenly spaced eigenvals. 12=Small, evenly spaced eigenvals.
9=Geometrically spaced eigenvals. 13=Matrix with random O(1) entries.
10=Clustered eigenvalues. 14=Matrix with large random entries.
11=Large, evenly spaced eigenvals. 15=Matrix with small random entries.

Tests performed: See cdrvst.f
Matrix order= 20, type= 9, seed=1494,3156,1807,2209, result 101 is 3047.00
CST drivers: 1 out of 11664 tests failed to pass the threshold


SEP: NB = 10, NBMIN = 2, NX = 1

All tests for CST passed the threshold ( 4662 tests run)

All tests for CST drivers passed the threshold ( 11664 tests run)

------------------------------------------------------------------------------
----
dsvd.out
-----

SVD: NB = 1, NBMIN = 2, NX = 1, NRHS = 2
DCHKBD: DBDSDC(vects) returned INFO= 1.
M= 30, N= 40, JTYPE= 12, ISEED=( 2195, 634, 3653, 1853)

DBD -- Real Singular Value Decomposition
Matrix types (see xCHKBD for details):
Diagonal matrices:
1: Zero 5: Clustered entries
2: Identity 6: Large, evenly spaced entries
3: Evenly spaced entries 7: Small, evenly spaced entries
4: Geometrically spaced entries
General matrices:
8: Evenly spaced sing. vals. 12: Small, evenly spaced sing vals
9: Geometrically spaced sing vals 13: Random, O(1) entries
10: Clustered sing. vals. 14: Random, scaled near overflow
11: Large, evenly spaced sing vals 15: Random, scaled near underflow

Test ratios: (B: bidiagonal, S: diagonal, Q, P, U, and V: orthogonal
X: m x nrhs, Y = Q' X, and Z = U' Y)
1: norm( A - Q B P' ) / ( norm(A) max(m,n) ulp )
2: norm( I - Q' Q ) / ( m ulp )
3: norm( I - P' P ) / ( n ulp )
4: norm( B - U S V' ) / ( norm(B) min(m,n) ulp )
5: norm( Y - U Z ) / ( norm(Z) max(min(m,n),k) ulp )
6: norm( I - U' U ) / ( min(m,n) ulp )
7: norm( I - V' V ) / ( min(m,n) ulp )
8: Test ordering of S (0 if nondecreasing, 1/ulp otherwise)
9: norm( S - S2 ) / ( norm(S) ulp ), where S2 is computed
without computing U and V'
10: Sturm sequence test (0 if sing. vals of B within THRESH of S)
11: norm( A - (QU) S (V' P') ) / ( norm(A) max(m,n) ulp )
12: norm( X - (QU) Z ) / ( |X| max(M,k) ulp )
13: norm( I - (QU)'(QU) ) / ( M ulp )
14: norm( I - (V' P') (P V) ) / ( N ulp )
M= 30, N= 40, type 12, seed=2195, 634,3653,1853, test(15)= .4504E+16
DBD: 1 out of 5510 tests failed to pass the threshold
*** Error code from DCHKBD = 1

All tests for DBD drivers passed the threshold ( 5320 tests run)
......

------
dgd.out
-----


DXV routines passed the tests of the error exits ( 87 tests done)

DXV -- Real Expert Eigenvalue/vector problem driver
Matrix types:

TYPE 1: Da is diagonal, Db is identity,
A = Y^(-H) Da X^(-1), B = Y^(-H) Db X^(-1)
YH and X are left and right eigenvectors.

TYPE 2: Da is quasi-diagonal, Db is identity,
A = Y^(-H) Da X^(-1), B = Y^(-H) Db X^(-1)
YH and X are left and right eigenvectors.


Tests performed:
a is alpha, b is beta, l is a left eigenvector,
r is a right eigenvector and ' means transpose.
1 = max | ( b A - a B )' l | / const.
2 = max | ( b A - a B ) r | / const.
3 = max ( Sest/Stru, Stru/Sest ) over all eigenvalues
4 = max( DIFest/DIFtru, DIFtru/DIFest ) over the 1st and 5th eigenvectors

Type= 2, IWA= 1, IWB= 1, IWX= 5, IWY= 1, result 4 is 3766.18
.......
.......
Type= 2, IWA= 5, IWB= 5, IWX= 4, IWY= 4, result 4 is 577.87
DXV drivers: 200 out of 5000 tests failed to pass the threshold

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

------
zgbak.out
-------
.. test output of ZGGBAK ..
value of largest test error = .796E+04
example number where ZGGBAL info is not 0 = 0
example number where ZGGBAK(L) info is not 0 = 0
example number where ZGGBAK(R) info is not 0 = 0
example number having largest error = 5
number of examples where info is not 0 = 0
total number of examples tested = 10


End of tests
Total time used = .01 seconds

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

------
zgd.out
------

ZXV routines passed the tests of the error exits ( 85 tests done)

ZXV -- Complex Expert Eigenvalue/vector problem driver
Matrix types:

TYPE 1: Da is diagonal, Db is identity,
A = Y^(-H) Da X^(-1), B = Y^(-H) Db X^(-1)
YH and X are left and right eigenvectors.

TYPE 2: Da is quasi-diagonal, Db is identity,
A = Y^(-H) Da X^(-1), B = Y^(-H) Db X^(-1)
YH and X are left and right eigenvectors.


Tests performed:
a is alpha, b is beta, l is a left eigenvector,
r is a right eigenvector and ' means transpose.
1 = max | ( b A - a B )' l | / const.
2 = max | ( b A - a B ) r | / const.
3 = max ( Sest/Stru, Stru/Sest ) over all eigenvalues
4 = max( DIFest/DIFtru, DIFtru/DIFest ) over the 1st and 5th eigenvectors

Type= 2, IWA= 1, IWB= 1, IWX= 1, IWY= 5, result 4 is 2364.97
............
............
Type= 2, IWA= 5, IWB= 4, IWX= 4, IWY= 5, result 4 is 507.92
ZXV drivers: 24 out of 5000 tests failed to pass the threshold

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

Thank you very much in advance for your interest. :cry:
mj
 
Posts: 1
Joined: Fri Apr 08, 2005 8:42 am

Postby Julien Langou » Tue Jul 26, 2005 11:23 am

Hello,

yes about How do I interpret LAPACK testing failures?"
please have also a look at http://www.netlib.org/lapack/faq.html#1.23

  1. no problem to reproduce all your outputs,
  2. about the 37 (resp. 0, 200, 24) failures to pass the threshold over 5000 tests in SXV (resp. CXV, DXV, ZXV) in the file sgd.out (resp. cgd.out, dgd.out, zgd.out),
    yes we know, this is 'normal', the test are really hard for the routines and the routines fail to satisfy the thresholds in some cases. The problems are explained in LAPACK Working Note 87,
  3. zgbak.out -- value of largest test error = .796E+04
    yes that's a bit high .... we will try to have a better understanding about this
  4. dsvd.out -- M= 30, N= 40, type 12, seed=2195, 634,3653,1853, test(15)= .4504E+16
    This one is much more problematic, the testing is running ok with LAPACK (g77 compiler) and with CLAPACK it exhibit a severe failure. We do not know what's going on, we'll try to have a look


Finally, yes there are some failures, just one is really severe, the other can be considered as minor.

Julien Langou

Julien
Julien Langou
 
Posts: 733
Joined: Thu Dec 09, 2004 12:32 pm
Location: Denver, CO, USA


Return to User Discussion

Who is online

Users browsing this forum: Bing [Bot] and 1 guest