===================
ERRATA in ScaLAPACK
===================
VERSION 1.0 : February 28, 1995
PATCHES:
VERSION 1.1 : March 20, 1995
VERSION 1.2 : May 10, 1996
PATCHES:
VERSION 1.3 : June 5, 1996
VERSION 1.4 : November 17, 1996
VERSION 1.5 : May 1, 1997
PATCHES:
VERSION 1.6 : November 15, 1997
DATE: July 31, 2001
This errata file lists errata for the ScaLAPACK Users' Guide and errata
for the ScaLAPACK code itself. It also lists machine-specific installation
hints, details of the updates, and any outstanding bugs to be fixed in the
upcoming release.
===============================================
Errata in ScaLAPACK Users' Guide, First Edition
===============================================
page
----
8: Change to netlib format ( .tar.gz --> .tgz )
Section 1.7 change scalapack.tar.gz -> scalapack.tgz
9: Change to netlib format ( .tar.gz --> .tgz )
Section 1.8 change manpages.tar.gz -> manpages.tgz
13-14: Change to netlib format ( .tar.gz --> .tgz )
Section 2.1, numerous references to .tar.gz files
must be replaced by .tgz
14: Change to netlib format ( .shar -> .tgz )
http://www.netlib.org/blas/blas.shar
should be changed to
http://www.netilb.org/blas/blas.tgz
and
sh blas.shar
should be changed to
gunzip -c blas.tgz | tar xvf -
20: Typographical error in the 8th row of the coefficient
matrix A. The 8th row is listed as
-s -c -a -l -a -p -a c k
and should be replaced by:
-s +c -a -l -a -p -a c k
Likewise, this correction must be propogated to the
8th row of Figure 2.1.
-s +c -a -l -a -p -a c k
72: Section 4.4.2
In the text beginning "Table 4.10 illustrates ...",
it erroneously says NB=3. This should be NB=8, as noted
in the caption of Table 4.10.
90: Section 4.6.7, third paragraph
inherit --> inherent
116: Erroneous use of BLAS instead of BLACS in the first sentence
of Section 5.4.1. The first sentence should say:
"Users should choose vendor-supplied BLACS optimized for their
computer; these BLACS will be the fastest BLACS implementation."
===================
Errata in ScaLAPACK
===================
We have tested ScaLAPACK on heterogeneous clusters of workstations
using pvm3.4 and mpich version 1.1, and on the IBM SP-2,
Intel Paragon, SGI Power Challenge, SGI Origin 2000, and Cray T3E. We
tested on the IBM SP-2 using IBM MPI, on the SGI Power Challenge
and Origin 2000 using SGI MPI v3.0, and on the Cray T3E using Cray
MPI (mpt.1.2.0.0.6beta). The workstations used were
AIX46K, ALPHA, HPPA, LINUX, SGI64, SUN4, and SUN4SOL2.
*** !! IMPORTANT !! *********************************************
Note to PGF77 and NAGF90 compiler users: We have unconfirmed reports
of Fortran/C interoperability problems (string handling) with
ScaLAPACK on the nagf90 and pg77 compilers. Until this is resolved,
the user will have to avoid using these two compilers in conjunction
with ScaLAPACK and the BLACS.
*** !! IMPORTANT !! *********************************************
#######################################
# MACHINE-SPECIFIC INSTALLATION HINTS #
#######################################
1. For HPPA, you may need to compile all routines in
SCALAPACK/PBLAS/SRC/INTERNAL/PBBLAS/ with NO optimization.
And, the user must modify SCALAPACK/SRC/pslabad.f and pdlabad.f
to remove the IF-conditional and always take the square root.
2. For NEC SX-4, the user must modify SCALAPACK/SRC/pslabad.f and
pdlabad.f to remove the IF-conditional and always take the
square root. And, you must include the -DNOIEEE flag in
the CDEFS definition in the SLmake.inc.
3. For Cray T3E, the user must modify SCALAPACK/SRC/pslabad.f and
pdlabad.f to remove the IF-conditional and always take the
square root. And, you must include the -DNOIEEE flag in
the CDEFS definition in the SLmake.inc.
Also for the Cray T3E ONLY, the user must reset the defined size
of an integer and a real in the ScaLAPACK Test Suite. Set the
variable INTGSZ = 8, REALSZ = 8, and CPLXSZ = 16 in all
SCALAPACK/TESTING/LIN/p*driver.f, SCALAPACK/TESTING/EIG/p*driver.f,
SCALAPACK/TESTING/EIG/p*sepreq.f, SCALAPACK/TESTING/EIG/p*gsepreq.f,
SCALAPACK/PBLAS/TESTING/p*tst.f, and SCALAPACK/PBLAS/TIMING/p*tim.f
testing/timing programs! By default, INTGSZ = 4, REALSZ = 4, and
CPLXSZ = 8 for all other architectures.
###################################
# TO BE FIXED IN THE NEXT RELEASE #
###################################
------------- ------- -----------------------------------------
DIRECTORY ROUTINE DESCRIPTION OF CHANGE
------------- ------- -----------------------------------------
SCALAPACK/PBLAS/SRC/psaxpy_.c On lines 362-363, there is a call to its
pdaxpy_.c corresponding serial routine _axpy. The
pcaxpy_.c very last argument in the call should be
pzaxpy_.c "incy", not "&desc_Y[LLD_]".
pcdotc_.c On line 471, &ixcol should be &iycol.
pzdotc_.c
pcdotu_.c On line 472, &ixcol should be &iycol.
pzdotu_.c
SCALAPACK/PBLAS/SRC/INTERNAL/PBBLAS/
Bug description:
================
In the rank-k and ran-2k Hermitian updates PBBLAS routines, when C is just
a block, BETA is one, and ( ALPHA is zero or NQ/NP is zero ), the call to the
Level 3 BLAS _HERK or HER2K does not zero the imaginary part of the diagonals
of C, and simply returns.
This causes a problem for example when the process owning C does not own
any of A/B; In such a case one has to zero those imaginary parts.
==========================================================================
pbcherk.f
line 283:
INTRINSIC CMPLX, MAX, MIN
should be
INTRINSIC CMPLX, MAX, MIN, REAL
line 770:
CALL CHERK( UPLO, TRANS, N, NQ, ALPHA, A,LDA, BETA, C,LDC )
should be
CALL CHERK( UPLO, TRANS, N, NQ, ALPHA, A,LDA, BETA, C,LDC )
IF( ( BETA.EQ.RONE ).AND.
$ ( ( ALPHA.EQ.RZERO ).OR.( NQ.EQ.0 ) ) ) THEN
DO 80 JJ = 1, N
C( JJ, JJ ) = CMPLX( REAL( C( JJ, JJ ) ), RZERO )
80 CONTINUE
END IF
line 824:
CALL CHERK( UPLO, TRANS, N, NP, ALPHA, A,LDA, BETA, C,LDC )
should be
CALL CHERK( UPLO, TRANS, N, NP, ALPHA, A,LDA, BETA, C,LDC )
IF( ( BETA.EQ.RONE ).AND.
$ ( ( ALPHA.EQ.RZERO ).OR.( NP.EQ.0 ) ) ) THEN
DO 90 JJ = 1, N
C( JJ, JJ ) = CMPLX( REAL( C( JJ, JJ ) ), RZERO )
90 CONTINUE
END IF
==========================================================================
pbcher2k.f
line 332
INTRINSIC CONJG, CMPLX, MAX, MIN
should be
INTRINSIC CONJG, CMPLX, MAX, MIN, REAL
line 1164:
CALL CHER2K( UPLO, TRANS, N, NQ, ALPHA, A, LDA, B, LDB,
$ BETA, C, LDC )
should be
CALL CHER2K( UPLO, TRANS, N, NQ, ALPHA, A, LDA, B, LDB,
$ BETA, C, LDC )
IF( ( BETA.EQ.RONE ).AND.
$ ( ( ALPHA.EQ.ZERO ).OR.( NQ.EQ.0 ) ) ) THEN
DO 150 JJ = 1, N
C( JJ, JJ ) = CMPLX( REAL( C( JJ, JJ ) ), RZERO )
150 CONTINUE
END IF
line 1223
CALL CHER2K( UPLO, TRANS, N, NP, ALPHA, A, LDA, B, LDB,
$ BETA, C, LDC )
should be
CALL CHER2K( UPLO, TRANS, N, NP, ALPHA, A, LDA, B, LDB,
$ BETA, C, LDC )
IF( ( BETA.EQ.RONE ).AND.
$ ( ( ALPHA.EQ.ZERO ).OR.( NP.EQ.0 ) ) ) THEN
DO 160 JJ = 1, N
C( JJ, JJ ) = CMPLX( REAL( C( JJ, JJ ) ), RZERO )
160 CONTINUE
END IF
==========================================================================
for pbzherk.f and pbzher2k.f, changes are the same with essentially the
following caveats CMPLX -> DCMPLX, REAL -> DBLE
==========================================================================
SCALAPACK/INSTALL/SLmake.LINUX Accidently missing definitions of F77, CC,
CCFLAGS, SRCFLAG, F77LOADER, CCLOADER,
F77LOADFLAGS, CCLOADFLAGS.
SCALAPACK/SRC/ psgebrd.f Avoid 0-th element reference on line 376
Replace with:
IF( JS.GT.0 )
$ CALL PSELSET( A, I+JB-1, J+JB, DESCA, E( JS ) )
Avoid 0-th element reference on line 380
Replace with:
IF( JS.GT.0 )
$ CALL PSELSET( A, I+JB, J+JB-1, DESCA, E( JS ) )
pcgebrd.f Avoid 0-th element reference on line 378.
Replace with:
IF( JS.GT.0 )
$ CALL PCELSET( A, I+JB-1, J+JB, DESCA, CMPLX( E( JS ) ) )
Avoid 0-th element reference on line 382.
Replace with:
IF( JS.GT.0 )
$ CALL PCELSET( A, I+JB, J+JB-1, DESCA, CMPLX( E( JS ) ) )
SCALAPACK/SRC/ psgeqpf.f Avoid 0-th element reference on lines 460 and
and 465. Replace lines 459-468 with:
IF( (JJA+NQ-JJ).GT.0 ) THEN
IF( MYROW.EQ.ICURROW ) THEN
CALL SGEBS2D( ICTXT, 'Columnwise', ' ', 1, JJA+NQ-JJ,
$ A( II+( MIN( JJA+NQ-1, JJ )-1 )*LDA ),
$ LDA )
CALL SCOPY( JJA+NQ-JJ, A( II+( MIN( JJA+NQ-1, JJ )
$ -1)*LDA ), LDA, WORK( IPW+MIN( JJA+NQ-1,
$ JJ )-1 ), 1 )
ELSE
CALL SGEBR2D( ICTXT, 'Columnwise', ' ', JJA+NQ-JJ, 1,
$ WORK( IPW+MIN( JJA+NQ-1, JJ )-1 ),
$ MAX( 1, NQ ), ICURROW, MYCOL )
END IF
END IF
SCALAPACK/SRC/ pcgeqpf.f Avoid 0-th element reference on lines 485 and
and 490. Replace lines 484-493 with:
IF( (JJA+NQ-JJ).GT.0 ) THEN
IF( MYROW.EQ.ICURROW ) THEN
CALL CGEBS2D( ICTXT, 'Columnwise', ' ', 1, JJA+NQ-JJ,
$ A( II+( MIN( JJA+NQ-1, JJ )-1 )*LDA ),
$ LDA )
CALL CCOPY( JJA+NQ-JJ, A( II+( MIN( JJA+NQ-1, JJ )
$ -1)*LDA ), LDA, WORK( MIN( JJA+NQ-1, JJ ) ),
$ 1 )
ELSE
CALL CGEBR2D( ICTXT, 'Columnwise', ' ', JJA+NQ-JJ, 1,
$ WORK( MIN( JJA+NQ-1, JJ ) ), MAX( 1, NQ ),
$ ICURROW, MYCOL )
END IF
END IF
SCALAPACK/SRC/ psgeqpf.f/ Line 484:
pcgeqpf.f
$ J+LL-JJ+1, DESCA, 1 )
should be
$ J+LL-JJ+2, DESCA, 1 )
psgesvx.f/ Line 502 of psgesvx.f should be
pcgesvx.f
NQ = NUMROC( N+ICOFFA, DESCA( NB_ ), MYCOL, IACOL,
$ NPCOL )
instead of
NQ = NUMROC( N+ICOFFA, DESCB( NB_ ), MYCOL, IACOL,
$ NPCOL )
Likewise, line 503-504 of pcgesvx.f should be
NQ = NUMROC( N+ICOFFA, DESCA( NB_ ), MYCOL, IACOL,
$ NPCOL )
instead of
NQ = NUMROC( N+ICOFFA, DESCB( NB_ ), MYCOL, IACOL,
$ NPCOL )
SCALAPACK/SRC/ pslahqr.f Copy the H11 line so that it is defined on both
sides of the loop. That is, line 2440 should be
the same as line 2435.
SCALAPACK/SRC/ pslarf.f Avoid 0-th element reference on line 487.
Replace with
IF( IOFFC.GT.0 )
$ CALL SGEMV( 'Transpose', MP, NQ, ONE,
$ C( IOFFC ), LDC, WORK, 1, ZERO,
$ WORK( IPW ), 1 )
Avoid 0-th element reference on line 501.
Replace with
IF( IOFFC.GT.0 )
$ CALL SGER( MP, NQ, -TAULOC, WORK, 1, WORK( IPW ),
$ 1, C( IOFFC ), LDC )
Avoid 0-th element reference on line 592.
Replace with
IF( IOFFV.GT.0 .AND. IOFFC.GT.0 )
$ CALL SGER( MP, NQ, -TAULOC, WORK, 1,
$ V( IOFFV ), LDV, C( IOFFC ), LDC )
Avoid 0-th element reference on line 531.
Replace with:
IF( IOFFC.GT.0 )
$ CALL SGEMV( 'Transpose', MP, NQ, ONE,
$ C( IOFFC ), LDC, WORK, 1, ZERO,
$ WORK( IPW ), 1 )
Avoid 0-th element reference on line 545.
Replace with:
IF( IOFFC.GT.0 )
$ CALL SGER( MP, NQ, -TAULOC, WORK, 1, WORK( IPW ),
$ 1, C( IOFFC ), LDC )
Avoid 0-th element reference on line 712.
Replace with:
IF( IOFFV.GT.0 )
$ CALL SCOPY( NQ, V( IOFFV ), LDV, WORK, 1 )
Avoid 0-th element reference on line 746.
Replace with:
IF( IOFFC.GT.0 )
$ CALL SGER( MP, NQ, -TAULOC, WORK( IPW ), 1, WORK,
$ 1, C( IOFFC ), LDC )
pclarf.f Avoid 0-th element reference on line 488.
Replace with:
IF( IOFFC.GT.0 )
$ CALL CGEMV( 'Conjugate transpose', MP, NQ, ONE,
$ C( IOFFC ), LDC, WORK, 1, ZERO,
$ WORK( IPW ), 1 )
Avoid 0-th element reference on line 502.
Replace with:
IF( IOFFC.GT.0 )
$ CALL CGERC( MP, NQ, -TAULOC, WORK, 1, WORK( IPW ),
$ 1, C( IOFFC ), LDC )
pclarf.f Avoid 0-th element reference on line 595.
Replace with:
IF( IOFFV.GT.0 .AND. IOFFC.GT.0 )
$ CALL CGERC( MP, NQ, -TAULOC, WORK, 1,
$ V( IOFFV ), LDV, C( IOFFC ),
$ LDC )
Avoid 0-th element reference on line 531.
Replace with:
IF( IOFFC.GT.0 )
$ CALL CGEMV( 'Conjugate transpose', MP, NQ, ONE,
$ C( IOFFC ), LDC, WORK, 1, ZERO,
$ WORK( IPW ), 1 )
Avoid 0-th element reference on line 547.
Replace with:
IF( IOFFC.GT.0 )
$ CALL CGERC( MP, NQ, -TAULOC, WORK, 1, WORK( IPW ),
$ 1, C( IOFFC ), LDC )
Avoid 0-th element reference on line 715.
Replace with:
IF( IOFFV.GT.0 )
$ CALL CCOPY( NQ, V( IOFFV ), LDV, WORK, 1 )
Avoid 0-th element reference on line 749.
Replace with:
IF( IOFFC.GT.0 )
$ CALL CGERC( MP, NQ, -TAULOC, WORK( IPW ), 1, WORK,
$ 1, C( IOFFC ), LDC )
SCALAPACK/SRC/ pslarz.f Avoid 0-th element reference on line 676.
Replace with:
IF( MPC2.GT.0 .AND. NQV.GT.0 )
$ CALL SGER( MPC2, NQV, -TAULOC, WORK, 1,
$ V( IOFFV ), LDV, C( IOFFC2 ),
$ LDC )
SCALAPACK/SRC/ pclarz.f Avoid 0-th element reference on line 677.
Replace with:
IF( MPC2.GT.0 .AND. NQV.GT.0 )
$ CALL CGERC( MPC2, NQV, -TAULOC, WORK, 1,
$ V( IOFFV ), LDV, C( IOFFC2 ),
$ LDC )
SCALAPACK/SRC/ pslarzb.f Avoid 0-th element reference on line 602.
Replace with:
IF( IOFFC2.GT.0 )
$ CALL SGEMM( 'No transpose', 'No transpose', MPC2, NQC2, K,
$ -ONE, WORK( IPW ), LW, WORK( IPV ), LV, ONE,
$ C( IOFFC2 ), LDC )
SCALAPACK/SRC/ pclarzb.f Avoid 0-th element reference on line 615.
Replace with:
IF( IOFFC2.GT.0 )
$ CALL CGEMM( 'No transpose', 'No transpose', MPC2, NQC2, K,
$ -ONE, WORK( IPW ), LW, WORK( IPV ), LV, ONE,
$ C( IOFFC2 ), LDC )
SCALAPACK/SRC/ psorg2l.f/ Code correction as follows:
pcung2l.f
Line 258:
TAUJ = TAU( MIN( JJ, NQA0 ) )
should be:
IACOL = INDXG2P( J, DESCA( NB_ ), MYCOL, DESCA( CSRC_ ),
$ NPCOL )
IF( MYCOL.EQ.IACOL )
$ TAUJ = TAU( MIN( JJ, NQA0 ) )
psorgl2.f/
pcungl2.f
Line 257:
TAUI = TAU( MIN( II, KP ) )
should be:
IAROW = INDXG2P( I, DESCA( MB_ ), MYROW, DESCA( RSRC_ ),
$ NPROW )
IF( MYROW.EQ.IAROW )
$ TAUI = TAU( MIN( II, KP ) )
psorg2r.f/
pcung2r.f
Line 258:
TAUJ = TAU( MIN( JJ, KQ ) )
should be:
IACOL = INDXG2P( J, DESCA( NB_ ), MYCOL, DESCA( CSRC_ ),
$ NPCOL )
IF( MYCOL.EQ.IACOL )
$ TAUJ = TAU( MIN( JJ, KQ ) )
psorgr2.f/
pcungr2.f
Line 258:
TAUI = TAU( MIN( II, MP ) )
should be:
IAROW = INDXG2P( I, DESCA( MB_ ), MYROW, DESCA( RSRC_ ),
$ NPROW )
IF( MYROW.EQ.IAROW )
$ TAUI = TAU( MIN( II, MP ) )
SCALAPACK/SRC/ psposvx.f/ Line 426 of psposvx.f should be
pcposvx.f
NQ = NUMROC( N+ICOFFA, DESCA( NB_ ), MYCOL, IACOL, NPCOL )
instead of
NQ = NUMROC( N+ICOFFA, DESCB( NB_ ), MYCOL, IACOL, NPCOL )
Likewise, line 425 of pcposvx.f should be
NQ = NUMROC( N+ICOFFA, DESCA( NB_ ), MYCOL, IACOL, NPCOL )
instead of
NQ = NUMROC( N+ICOFFA, DESCB( NB_ ), MYCOL, IACOL, NPCOL )
SCALAPACK/SRC/ pssyevx.f/ Input error checking corrections:
pdsyevx.f
Line 513 should be replaced by:
INFO = -( 800+CTXT_ )
Line 516 should be replaced by:
INFO = -( 2100+CTXT_ )
Line 520 should be replaced by:
CALL CHK1MAT( N, 4, N, 4, IA, JA, DESCA, 8, INFO )
Line 522 should be replaced by:
$ CALL CHK1MAT( N, 4, N, 4, IZ, JZ, DESCZ, 21, INFO )
Line 702-703 should be replaced by:
CALL PCHK2MAT( N, 4, N, 4, IA, JA, DESCA, 8, N, 4, N, 4, IZ,
$ JZ, DESCZ, 21, 4, IDUM1, IDUM2, INFO )
Line 706 should be replaced by:
CALL PCHK1MAT( N, 4, N, 4, IA, JA, DESCA, 8, 4, IDUM1,
SCALAPACK/SRC/ pcheevx.f/ Input error checking corrections:
pzheevx.f
Line 532 should be replaced by:
INFO = -( 800+CTXT_ )
Line 535 should be replaced by:
INFO = -( 2100+CTXT_ )
Line 538 should be replaced by:
CALL CHK1MAT( N, 4, N, 4, IA, JA, DESCA, 8, INFO )
Line 541 should be replaced by:
$ CALL CHK1MAT( N, 4, N, 4, IZ, JZ, DESCZ, 21, INFO )
Line 732-33 should be replaced by:
CALL PCHK2MAT( N, 4, N, 4, IA, JA, DESCA, 8, N, 4, N, 4, IZ,
$ JZ, DESCZ, 21, 4, IDUM1, IDUM2, INFO )
Line 735 should be replaced by:
CALL PCHK1MAT( N, 4, N, 4, IA, JA, DESCA, 8, 4, IDUM1,
SCALAPACK/TESTING/EIG/ pssepsubtst.f Avoid 0-th element reference
by replacing line 254 with:
NORMWIN = SAFMIN / EPS
IF( N.GE.1 )
$ NORMWIN = MAX( ABS( WIN( 1 ) ), ABS( WIN( N ) ), NORMWIN )
SCALAPACK/TESTING/EIG/ pzbrdinfo.f Typo in FORMAT statement line 257
(complex single --> complex double)
pzhrdinfo.f Typo in FORMAT statement line 261
(complex single --> complex double)
pztrdinfo.f Typo in FORMAT statement line 262
(complex single --> complex double)
*******************************************************************
* Changes to ScaLAPACK code to make it run on T3E *
*******************************************************************
SCALAPACK/SLmake.inc:
add -DNO_IEEE to CDEFS
add -dp to compilation flags
SCALAPACK/SRC/p?stein.f:
as first executable line, add:
onenrm = 0.0E0
SCALAPACK/SRC/p?latrz.f:
add
AII = ZERO
right after quick return so it is initialized on nodes that do not do
stuff in P_LARFG
SCALAPACK/SRC/p?labad.f:
comment out
IF( LOG10( LARGE ).GT.2000.D0 ) THEN
because T3E has IEEE hardware, but not arithmetic
SCALAPACK/SRC/pslaiect.c
as first line, add:
#define float double
SCALAPACK/PBLAS/SRC/pblas.h
in original file, move lines 92 and 93 to line 119 (:92,93mo119)
this moves the typedef for complex below the #define float double . .
at line 27, add:
#ifdef T3E
#define _MACH_ _T3D_
#endif
SCALAPACK/PBLAS/TESTING/p*tst.f:
Set REALSZ = 8 and CPLXSZ = 16!!
SCALAPACK/PBLAS/TIMING/p*tim.f:
Set REALSZ = 8 and CPLXSZ = 16!!
SCALAPACK/TESTING/LIN/p*driver.f:
Set INTGSZ = 8, REALSZ = 8, and CPLXSZ = 16!!
SCALAPACK/TESTING/EIG/p*driver.f:
Set INTGSZ = 8, REALSZ = 8, and CPLXSZ = 16!!
SCALAPACK/TESTING/EIG/p*sepreq.f:
Set INTGSZ = 8, REALSZ = 8, and CPLXSZ = 16!!
SCALAPACK/TESTING/EIG/p*gsepreq.f:
Set INTGSZ = 8, REALSZ = 8, and CPLXSZ = 16!!
SCALAPACK/TESTING/EIG/p[c,z]sepsubtst.f:
change line 737 from:
IF( MISSSMALLEST .AND. ( WIN( MYIL-1 ).LT.VL+NORMWIN*
$ FIVE*THRESH*EPS ) )MISSSMALLEST = .FALSE.
to:
if (misssmallest)
$ misssmallest = win(myil-1) .ge.
$ vl+normwin*five*thresh*eps
SCALAPACK/TESTING/EIG/p[d,s]sepsubtst.f:
change line 714 from:
IF( MISSSMALLEST .AND. ( WIN( MYIL-1 ).LT.VL+NORMWIN*
$ FIVE*THRESH*EPS ) )MISSSMALLEST = .FALSE.
to:
if (misssmallest)
$ misssmallest = win(myil-1) .ge.
$ vl+normwin*five*thresh*eps
SCALAPACK/TESTING/EIG/p*gsepsubtst.f:
Same change as for p*sepsubtst.f.
#######################################
# HETEROGENEOUS NETWORKS OF COMPUTERS #
#######################################
4. Unresolved issues remain in heterogeneous computing. As a
result, the ScaLAPACK library is not as robust in a heterogeneous
environment as it is in a homogeneous environment. The
following is a list of known problems when ScaLAPACK is
run on a heterogeneous network of computers. For further
details, please refer to LAPACK Working Note 112:
http://www.netlib.org/lapack/lawns/lawn112.ps
5. Because the ALPHA handles denormalized numbers differently
than other architectures, the least squares executables
SCALAPACK/TESTING/x_ls will fail when run on a heterogeneous
network. However, the least squares executables will run
homogeneously on ALPHAs.
6. The symmetric eigensolvers may have trouble on heterogeneous
networks when a subset of eigenvalues is chosen by value
(i.e. RANGE='V') and one of the limits of that range (VL or VU)
is within a couple ulps of an actual eigenvalue. This
is not a problem on homogeneous systems. When this happens,
some processes will return INFO <> 0. This can happen
when running the test code. In every case that we have seen,
the answer is correct despite the spurious error message.
7. The symmetric eigensolver test will often hang on a heterogeneous
system which includes an ALPHA. The symmetric eigensolver test will
occasionally hang on a network of heterogeneous computers
which does not include an ALPHA, such as a network which
includes an HPPA and an RS6K.
###################################
# DETAILS OF UPDATES #
###################################
Note: Unless otherwise stated, changes to single precision routines (names
beginning with PS or PC) apply also to the corresponding double precision
routines (names beginning with PD and PZ).
UPDATE (Version 1.6)
------------ ------- ------------- --------------------------
DIRECTORY ROUTINE LAST MODIFIED DESCRIPTION OF CHANGE
------------ ------- ------------- --------------------------
SRC/ psdb*.f/ Several modifications to source code
pcdb*.f
psdt*.f/ Several modifications to source code
pcdt*.f
psgb*.f/ Several modifications to source code
pcgb*.f
psgeqlf.f/ Missing from input error-checking:
pcgeqlf.f IF( LWORK.EQ.-1 ) THEN
IDUM1( 1 ) = -1
ELSE
IDUM1( 1 ) = 1
END IF
IDUM2( 1 ) = 9
psgerfs.f/ Uninitialized variables sometimes caused
pcgerfs.f failures in p_gerfs on SGI systems. The
fix is to change:
IDUM2( 3 ) = 3
IF( LWORK.EQ.-1 ) THEN
IDUM1( 5 ) = -1
ELSE
IDUM1( 5 ) = 1
END IF
IDUM2( 5 ) = 24
to:
IDUM2( 3 ) = 3
IF( LWORK.EQ.-1 ) THEN
IDUM1( 4 ) = -1
ELSE
IDUM1( 4 ) = 1
END IF
IDUM2( 4 ) = 24
psgesvd.f Corrections to input error-checking.
pslacp2.f/ Error when UPLO = 'All', and the operands A
pclacp2.f and B are not exactly distributed the same
way, but still correctly aligned, PxLACP2
fails for certain values of the parameters
M and N.
pslapiv.f/ Incorrectly used DESCA instead of DESCIP.
pclapiv.f Lines 234 and 235 should be:
IPT = MOD( JP-1, DESCIP(NB_) )
DESCPT(M_) = N + IPT + NPROW*DESCIP(NB_)
Line 237 should be:
DESCPT(MB_) = DESCIP(NB_)
Line 294 should be:
JPT = MOD( IP-1, DESCIP(MB_) )
Line 296 should be:
DESCPT(N_) = M + JPT + NPCOL*DESCIP(MB_)
Line 298 should be:
DESCPT(NB_) = DESCIP(MB_)
psorm2l.f/ Incorrectly used DESCA instead of DESCC.
pcunm2l.f Line 416 of psorm2l.f
$ JC, DESCA, WORK )
should be
$ JC, DESCC, WORK )
Lines 425 and 428 of pcunm2l.f
$ IC, JC, DESCA, WORK )
should be
$ IC, JC, DESCC, WORK )
pcporfs.f BERR and FERR were erroneously declared
as COMPLEX instead of REAL!
psstebz.f Error in LWORK=-1 return, should retain
INFO=0 for LWORK=-1 or LIWORK=-1.
Add the line
LQUERY = ( LWORK.EQ.-1 .OR. LIWORK.EQ.-1 )
Change line 346 from
ELSE IF( LWORK.LT.MAX( 5*N, 7 ) ) THEN
to
ELSE IF( LWORK.LT.MAX( 5*N, 7 ).AND..NOT.LQUERY) THEN
Change line 348 from
ELSE IF( LIWORK.LT.MAX( 4*N, 14, NPROW*NPCOL ) ) THEN
to
ELSE IF( LIWORK.LT.MAX( 4*N, 14, NPROW*NPCOL )
$ .AND. .NOT.LQUERY ) THEN
psstein.f/ Error in LWORK=-1 return, should retain
pcstein.f INFO=0 if LWORK=-1.
Add the line
LQUERY = ( LWORK.EQ.-1 .OR. LIWORK.EQ.-1 )
Change line 347 from
ELSE IF( MAXVEC.LT.LOAD ) THEN
to
ELSE IF( MAXVEC.LT.LOAD.AND..NOT.LQUERY) THEN
Change line 349 from
ELSE IF( LIWORK.LT.3*N+P+1 ) THEN
to
ELSE IF( LIWORK.LT.3*N+P+1.AND..NOT.LQUERY) THEN
pssyev.f Documentation corrections and corrections
to the input error-checking.
pssyevx.f/ Documentation corrections and corrections
pcheevx.f to the input error-checking.
TOOLS/ SL_gridreshape.c The following include statement should
be added:
#include <stdlib.h>
reshape.c The following include statement should
be added:
#include <stdlib.h>
TESTING/LIN/ p*db*.f/ Miscellaneous corrections
p*dt*.f/
p*gb*.f/
p*pb*.f/
p*pt*.f
TESTING/EIG/ pshrddriver.f/ Error in workspace calculation
pchrddriver.f Change Line 281 from:
WORKSIZ = NB*NB + NB*IHLP + ITEMP + IPOSTPAD
to
WORKSIZ = MAX( NB*NB + NB*IHLP + ITEMP, NB * NP ) +
$ IPOSTPAD
pssepchk.f/ Error in documentation describing
pcsepchk.f tests performed and LWORK calculation
pssepsubtst.f/ "
pcsepsubtst.f
pssqpsubtst.f/ "
pcsqpsubtst.f
###################################
# REPORTED PROBLEMS #
###################################
The following testing problems have been noted, but their cause is
currently unknown.
1. For ALPHA, testing failures in SCALAPACK/TESTING/xsnep, xdnep,
xssep, and/or xdsep, depending upon the version of OSF, under
investigation.
2. During testing on the IBM SP-2, using IBM MPI, we encountered
testing failures in SCALAPACK/TESTING/xsnep.
The following bugs have been identified.
1. On the SP-2, using MPLBLACS, hangs occur in x_pbllt, x_ptllt,
x_dtlu, x_dblu due to an error in MPL. These hangs do not occur when using
the MPIBLACS. Therefore, users of the banded codes are urged to use the
MPIBLACS. Here is a small MPL routine demonstrating the problem:
*****
program tst
integer k, iam, Np, ictxt, i, j
call mpc_environ(Np, Iam);
k = Iam + 100
print*,'start'
if (iam.eq.1) then
call mp_send(Iam, 4, 0, 2, i)
call mp_send(k, 4, 0, 3, j)
print*,mp_status(i)
print*,mp_status(j)
else if (iam .eq. 0) then
call mp_brecv(k, 4, 1, 3, j)
call mp_brecv(k, 4, 1, 2, j)
end if
print*,'done'
stop
end
*****