LAPACK3E release notes
Created:  September 25, 2002

Tests of xGGESX and xGGEVX fail on every known platform.  These are
produced by the test runs
   xeigtsts < sgd.in
   xeigtstd < dgd.in
   xeigtstc < cgd.in
   xeigtstz < zgd.in
No suitable explanation has ever been provided by the LAPACK developers.
Failures in those tests are ignored in the following notes.  All other
standard tests pass except as noted below.

------------------------------
IBM SP, AIX 4.3.3, xlf 7.1.0.2
------------------------------

BLAS/SRC:  Compile snrm2.f, srotm.f, scnrm2.f, dnrm2.f, drotm.f, and
   dznrm2.f as shown in Makefile.ibm.  ESSL is missing SROTM and DROTM.
INSTALL:   Replace second.f and dsecnd.f with second.f.ibm and
   dsecnd.f.ibm.  These versions call TIMEF instead of ETIME_ for
   greater accuracy.
SRC:  The xlf compiler can't compile la_lapack.f with any optimization
   options.  Enter make -f Makefile.ibm in the SRC directory to compile
   that module without optimization.

-----------------------------------------
CRAY T3E, Unicos/mk 2.0.5.49, f90 3.4.0.2
-----------------------------------------

The "rename" program in rename.f was used to modify all the la_*.f
module files to include only the 64-bit real and complex specific names.
LAPACK3E was designed to support 32-bit real and complex naming
conventions on Cray systems, using a leading H in the real case and a
leading G in the complex case, but testing of this option has been
deferred.

You can save some time compiling if you set the environment variable
NPROC to 4, instead of the default value of 1.

BLAS/SRC:  Compile snrm2.f and scnrm2.f into libblas.a as shown in
   Makefile.t3e.
INSTALL:  In the tests, it's assumed that only the 64-bit REAL and
   COMPLEX versions are compiled for the CRAY T3E.  But the special
   makefile Makefile.t3e shows how the 32-bit option would work by
   providing tests of both SLAMCH (the 64-bit version) and HLAMCH
   (the 32-bit version) using the common source module slamch.F.
SRC:  Save the la*.f files somewhere and copy the files in lafiles.t3e
   to this directory as described in the README file.
TESTING/LIN and TESTING/EIG:  Because of the unsafe complex divide,
   test problems must be scaled between sqrt(underflow) and
   sqrt(overflow) in the complex case.  Special versions of SLABAD and
   DLABAD in clabad.f and zlabad.f are referenced from Makefile.t3e
   in these directories.

--------------------------------
Sun Ultra-4, Forte 6.0 compilers
--------------------------------

INSTALL:   Replace second.f and dsecnd.f with second.f.sun and
   dsecnd.f.sun.  These versions call the high-resolution clock
   GETHRTIME instead of ETIME for greater accuracy.  Unfortunately,
   these versions of SECOND and DSECND contain a SAVE statement, but
   none of the LAPACK3E routines call SECOND or DSECND so the thread-
   safe feature of the LAPACK3E library is not compromised.

   The default versions of SECOND/DSECND using ETIME can deliver higher
   resolution if run under "ptime".  For the tests in this directory,
   the syntax would be
      /usr/proc/bin/ptime testsecond
      /usr/proc/bin/ptime testdsecnd

BLAS/SRC:  Compile snrm2.f, scnrm2.f, dnrm2.f, and dznrm2.f into the
   LAPACK3E library as shown in Makefile.  One test failure in tests of
   STRSEN can be avoided by also compiling sdot.f into libblas.a as
   shown in Makefile.sun.
TESTING/EIG:  Use make -f Makefile.sun to avoid a problem when compiling
   xdrvst.f in this directory.

With the netlib version of SDOT, one test exceeded its threshold test,
but it was small enough not to cause undue concern:

 Tests of the Symmetric Eigenvalue Problem routines

 LAPACK VERSION 3.0, released June 30, 1999 

 The following parameter values will be used:
    M:         0     1     2     3     5    20
    N:         0     1     2     3     5    20
    NB:        1     3     3     3    10
    NBMIN:     2     2     2     2     2
    NX:        1     0     5     9     1
 
 Relative machine underflow is taken to be    0.117549E-37
 Relative machine overflow  is taken to be    0.340282E+39
 Relative machine precision is taken to be    0.119209E-06

 Routines pass computational tests if test ratio is less than   50.00

 
 SST routines passed the tests of the error exits (147 tests done)


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

 All tests for SST routines passed the threshold (  3276 tests run)

 All tests for SST drivers  passed the threshold ( 14256 tests run)


 SEP:  NB =   3, NBMIN =   2, NX =   0

 All tests for SST routines passed the threshold (  3276 tests run)

 All tests for SST drivers  passed the threshold ( 14256 tests run)


 SEP:  NB =   3, NBMIN =   2, NX =   5

 All tests for SST routines passed the threshold (  3276 tests run)

 All tests for SST drivers  passed the threshold ( 14256 tests run)


 SEP:  NB =   3, NBMIN =   2, NX =   9

 All tests for SST routines passed the threshold (  3276 tests run)

 All tests for SST drivers  passed the threshold ( 14256 tests run)


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

 All tests for SST routines passed the threshold (  3276 tests run)

 SST -- Real Symmetric 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 Symmetric 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 sdrvst.f
 Matrix order=   20, type=10, seed= 396,1967, 588,3009, result  89 is   61.99
 SST drivers:      1 out of  14256 tests failed to pass the threshold


 End of tests
 Total time used =         9.37 seconds

