| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1869.1 | DXML 2.7 on Alpha/OSF available | HPCGRP::LOFGREN | | Mon Jun 13 1994 11:34 | 139 |
|
Digital eXtended Math Library (DXML) Version 2.7 for DEC OSF/1 AXP is now
available on the network. This version was submitted to CSERL and SSB on
June 10, 1994. It will be available to customers with the Q2CY94 consolidated
layered product CDROM later this month.
The kits and user documentation can be copied from
stcunx::"/usr/ftp/pub/dxml/v27/DXML270DOC.tar.Z" User docs (/docs)
stcunx::"/usr/ftp/pub/dxml/v27/DXML270DVKIT.tar" Dev kit and manpages
stcunx::"/usr/ftp/pub/dxml/v27/DXML270RTKIT.tar" Run Time kit
The Development kit allows you to link and execute programs with calls to
DXML. The Run time allows you to execute programs that have been linked
on a different system. Only one of the two kits should be installed on
your system.
The file stcunx::"/usr/ftp/pub/dxml/v27/README" explains how to copy and
install the network kits.
Please report problems in the HPCGRP::DXML notesfile. For convenience, the
release notes are attached.
Release Notes for
Digital Extended Math Library (DXML) V2.7 for DEC OSF/1 AXP
June 1994
1 PURPOSE OF RELEASE NOTES
These release notes provide information about DXML Version 2.7 and
user documentation. Please read these notes before using the
software and documentation.
1.1 DXML DISTRIBUTION
DXML is distributed in two different forms:
1. The DXML V2.7 Development Kit allows you to link an application
program with the DXML library and then run the executable image.
The name of the development kit is XMDLOA270.
2. The DXML V2.7 Run-time Kit allows an executable image that has
been linked on a different system to be executed on the target
system. The name of the run time kit is XMRLOA270.
You should install only one of the DXML kits on your system.
2 CHANGES SINCE DXML V2.5
A bug in the signal processing component involving DXMLDEF has been
fixed.
A bug in the BLAS3 routines, CTRSM and ZTRSM, has been fixed.
The Installation Guide and the Reference Manual have been updated
and a number of minor problems have been fixed.
Several minor bugs in the installation script have been fixed.
The DXML manpages have been further improved, and manpages for all
LAPACK routines are included.
Code to improve performance of BLAS2 routines, SGEMV and DGEMV, has
been added.
Version 1.1 of LAPACK has been included with this release. This
version completely replaces LAPACK 1.0.
LAPACK 1.1 added 72 new routines and made minor modifications to 176
routines. The new routines include:
DXML V2.7 Release Notes Page 2
- generalized nonsymmetric eigenvalue solver (_gegs.f)
- generalized nonsymmetric left/right eigenvector solver (_gegv.f)
- generalized linear regression model (_ggglm.f)
- linear equality constrained least square (_gglse.f)
- GSVD (_ggsvd.f)
- norm functions (_lanht.f)
BLAS3 routine _TRSM has been modified to improve performance when
the B matrix contains fewer than four columns.
Significant new functionality has been added to the Iterative Solver
package, as follows:
1. The constraint on the main diagonal being stored in the first
column of the array for SDIA and UDIA storage schemes has been
removed.
2. A driver routine (DITSOL_DRIVER) has been added.
3. Routine DMATVEC_UDIA has been optimized.
4. Some of the routine names have been changed. The old names will
still work in DXML V2.7.
The installation procedure will continue and complete even if a
valid PAK is not installed at the time of installation. However,
the IVP will still fail, and any use of DXML without a valid PAK
will fail.
All routines have been compiled with -automatic, which places all
local variables on the stack and ensures that the code is thread
safe.
We have included the public domain program called equivalence, which
provides a quick and easy conversion of a Linpack subroutine name to
the corresponding LAPACK name. The program is installed at
/usr/opt/XMDLOA270/dxml/equivalence_lapack (and .c)
This version of DXML was built on DEC OSF/1 V2.0 using DEC Fortran
V3.4-480 and DEC C T1.4-023.
AXP and Alpha AXP are registered as trademarks of Digital
Equipment Corporation.
Copyright (c) Digital Equipment Corporation 1994. All Rights Reserved.
|
| 1869.2 | DXML 3.0 SSB kit | HPCGRP::LOFGREN | | Tue Dec 06 1994 11:21 | 293 |
|
The SSB kit of DXML V3.0 for DEC OSF/1 AXP is now available on the network.
This release includes various maintenance fixes as well as new functionality
such as:
- archive library
- parallel library with some of the most frequently called
subroutines parallelized for shared memory (SMP) configurations
- Cos and sin transforms added to the signal processing library
- QMR solver added to the iterative solver library
- substantial skyline solver optimizations
- all BLAS2 routines have been hand-optimized
- support of dataless installation
The parallel library should be installed only on multiprocessor systems
running OSF/1 V3.0 or later. The serial library has been tested and verified
to work on OSF/1 V2.1 and V3.0.
DXML 3.0 for Alpha/OSF is planned to ship from SSB on December 30 with the
Q4CY94 consolidated layered product CDROM.
The Development kit and user documentation can be copied from
stcunx::"/usr/ftp/pub/dxml/v30/dxml300.tar"
stcunx::"/usr/ftp/pub/dxml/v30/docs/dxml_30_osf_rel_notes.txt" (Rel. notes)
stcunx::"/usr/ftp/pub/dxml/v30/docs/dxml_30_osf_install.ps" (Install Guide)
stcunx::"/usr/ftp/pub/dxml/v30/docs/dxml_30_osf_ref.ps" (Ref. Manual)
stcunx::"/usr/ftp/pub/dxml/v30/docs/dxml_30_osf_spd.ps" (SPD)
To copy any of the network files (eg. the software kit) to your current
directory
> ftp stcunx (or 16.124.32.134)
Name: ftp
Password: ident
ftp> bin
ftp> get pub/dxml/v30/dxml300.tar dxml300.tar
To untar and install
> tar -xvf dxml300.tar .
> su
# setld -l [directory of the expanded tar file] XMDLOA300 (serial library)
# setld -l [directory of the expanded tar file] XMDMAN300 (manpages)
# setld -l [directory of the expanded tar file] XMDPLL300 (parallel library)
Your old PAK should work.
For convenience, the release notes are attached.
Digital Extended Math Library (DXML) V3.0 for DEC OSF/1 AXP
December 1994
1 PURPOSE OF RELEASE NOTES
These release notes provide information about DXML Version V3.0 and about
user documentation. A fair number of changes are introduced with this
release, so please read these notes carefully before using the software
and documentation.
1.1 DXML DISTRIBUTION
DXML is distributed in two different forms:
1. The DXML V3.0 Development Kit allows you to link an application
program with the DXML library and then run the executable image. The
LMF name for the development option is EXT-MATH-LIB. The kit
includes the following subsets:
o XMDLOA300 - includes shared library, archive library, release
notes, IVP, examples and various DXML utilities.
o XMDMAN300 - includes the DXML manpages.
o XMDPLL300 - includes shared library for the parallel version of
the library and the parallel version of the IVP. This subset
should only be installed on SMP configurations running DEC OSF/1
V3.0.
2. The DXML V3.0 Run-time Kit allows an executable image that has been
linked on a different system to be executed on the target system.
The LMF-name for the run time option is EXT-MATH-LIB-RT. The kit
includes the following subset:
o XMRLOA300 - includes the run time version of the shared library,
release notes, and IVP.
You should install either the development kit or the run time kit on your
system - never both.
2 CHANGES SINCE DXML V2.7
2.1 Archive Library
The archive library of DXML is included with this release. A valid
Development license is required to load the archive library.
DXML V3.0 Release Notes Page 2
2.2 Parallel DXML
The parallel version of DXML (libdxmlp.so) is introduced with this
release. It includes exactly the same number of subroutines with the
identical names and calling interfaces as the serial library
(libdxml.so). The following routines in the parallel library have been
modified to take advantage of additional CPUs in shared memory
configurations.
{C,D,S,Z}GEMM BLAS3
{C,D,S,Z}GETRF LAPACK
{C,D,S,Z}POTRF
{C,D,S,Z}FFT Signal Processing
{C,D,S,Z}FFT_INIT
{C,D,S,Z}FFT_APPLY
{C,D,S,Z}FFT_EXIT
{C,D,S,Z}FFT_2D
{C,D,S,Z}FFT_INIT_2D
{C,D,S,Z}FFT_APPLY_2D
{C,D,S,Z}FFT_EXIT_2D
{C,D,S,Z}FFT_3D
{C,D,S,Z}FFT_INIT_3D
{C,D,S,Z}FFT_APPLY_3D
{C,D,S,Z}FFT_EXIT_3D
To use the parallel version of DXML, the following environment variables
must be defined.
PARALLEL
The PARALLEL environment variable must be set to a numeric value
defining the number of threads to be used by the KAP run-time
library. The parallel versions of DXML BLAS and LAPACK routines use
the KAP run-time library.
Eg. > setenv PARALLEL 3
KMP_STACKSIZE
The KMP_STACKSIZE environment variable must be set to a numeric
value. KMP_STACKSIZE defines the size of each thread's private
stack. The value must be at least 262144 which defines a 256K byte
private stack.
Eg. > setenv KMP_STACKSIZE 262144
Please note that the use of the parallel library for small (<4K) FFTs may
result in a degradation in performance. This is because the overhead to
create the parallel structure for smaller data sets may be larger than
the time needed for the computation. The exact crossover point where
employing parallelism will improve performance will depend on the
particular model of your system.
DXML V3.0 Release Notes Page 3
For information on linking and executing with the parallel library,
please refer to the DXML Reference Manual or the DXML manpages.
2.3 BLAS2
A serious computational bug in {D,S}GEMV, which was introduced in the
DXML 2.7 release, has been fixed.
All BLAS2 routines have been hand-optimized, and will perform better in
most cases. In our internal testing we have noticed almost 2X
improvement for the _GEMV routine. The performance gains for your
particular application may be different.
2.4 BLAS3
The internal data copy performance of {C,D,S,Z}GEMM has been improved
2.5 LAPACK
LAPACK V1.1 is included in this release of DXML.
2.6 Iterative Solver
The transpose-free version of the QMR solver has been added to the
iterative solver package. The name of the new routine is DITSOL_PTFQMR.
See the manpages or the DXML Reference Manual for details.
The driver routine has also been updated to include the TFQMR solver.
This solver can be called by setting the value isolve=6.
Some routines names were changed in DXML 2.7 and both new and old names
were supported in that release. DXML 3.0 supports only the new names.
2.7 Skyline Solver
The skyline solver package has been optimized. With our internal test
cases we saw an improvement of 2X, but the performance of the skyline
solver is very strongly dependent on the structure of the matrix.
The factor and solve routines in the skyline package have also been
optimized. As a result, some of the other routines such as the condition
number estimator, the iterative refinement routine, etc., will also see
an improvement.
DXML V3.0 Release Notes Page 4
2.8 Signal Processing
Subroutines for fast cosine and sine transforms have been added to the
signal processing library. The new subroutines are:
Cosine Transform
{S,D}FCT
{S,D}FCT_INIY
{S,D}FCT_APPLY
{S,D}FCT_EXIT
Sine Transform
{S,D}FST
{S,D}FST_INIT
{S,D}FST_APPLY
{S,D}FST_EXIT
Performance of large sized 1D-FFT, GRP-FFT routines (n>8192) and
multi-dimensional FFT routines have all been further optimized.
2.9 Kit
DXML 3.0 Development kit supports installation in a dataless
client/server environment. For more information see the DXML
Installation Guide.
The required disk space for the full development kit has increased from
9MB to approximately 27 MB. The reason is that the kit now includes an
archive version of the library as well as the parallel library.
This version of DXML was built on DEC OSF/1 V3.0 (rev 347) using DEC
Fortran T3.5-542 and DEC C V4.0-005.
3 USER DOCUMENTATION
The DXML Reference Manual and the Installation Guide have been updated
for this release.
AXP and Alpha AXP are registered as trademarks of Digital
Equipment Corporation.
Copyright (c) Digital Equipment Corporation 1994. All Rights Reserved.
|
| 1869.3 | DXML V2.8 SSB kit | HPCGRP::LOFGREN | | Thu Jan 26 1995 16:56 | 357 |
|
The SSB kit of DXML Version 2.8 for OpenVMS Alpha is now available on the
network.
The product will ship from SSB on February 24 on the Q1CY95 layered product
CDROM.
WHAT IS NEW
===========
This release includes various maintenance fixes as well as new
functionality and new optimizations:
- LAPACK 2.0
- Cos and sin transforms added to the signal processing library
- QMR solver added to the iterative solver library
- skyline solver routines have been optimized
- all BLAS2 routines have been optimized
Additional HELP
===============
To facilitate access and use of the library we have prepared a
DXML$HELP.HLB library which complements the regular DCL HELP. In
terms of content it corresponds to the DXML manpages on Alpha/OSF.
We decided to not include the file with the external kit because
it added over 10% to the size of the saveset. While distribution
at this time is limited to internal users, we don't mind if it
finds its way to external DXML users.
If you install the file at SYS$COMMON:[SYSHLP]DXML$HELP.HLB you
can get online, detailed information about all 1418 subroutines
in the library.
After installation you could define a system level logical name
$ DEFINE HLP$LIBRARY SYS$COMMON:[SYSHLP]DXML$HELP.HLB
To get a brief description and parameter list for the BLAS1
routine SAXPY you type:
$ HELP SAXPY
KIT LOCATION
============
The DXML V2.8 kit and user documentation can be copied from:
Directory 4HEAD::DXML:[AXP_VMS.V28]
DXMLDVAV028.A Development Kit
DXMLRTAV028.A Run time kit
DXML_28_VMS_INSTALL.PS (or .TXT) Installation Guide
DXML_28_VMS_REF.PS Reference Manual
DXML_28_VMS_SPD.PS (or .TXT) SPD
DXML028.RELEASE_NOTES Release Notes
DXML$HELP.HLB DXML help library
The Development kit allows linking and execution, while the Run Time
kit allows execution only. You should not install both kits on the
same system.
LICENSE PAK
===========
Your old license PAKs will still work.
RELEASE NOTES
=============
Attached.
Digital Extended Math Library (DXML) V2.8 for OpenVMS Alpha
January 1995
1 PURPOSE OF RELEASE NOTES
These release notes provide information about DXML Version 2.8 and about
user documentation.
1.1 DXML DISTRIBUTION
DXML is distributed in two different forms:
1. The DXML V2.8 Development Kit allows you to link an application
program with the DXML library and then run the executable image. The
name of the saveset for the development kit is DXMLDVAV028.A.
2. The DXML V2.8 Run-time Kit allows an executable image that has been
linked on a different system to be executed on the target system.
The name of the saveset for the run time kit is DXMLRTAV028.A.
You should install either the development kit or the run time kit on your
system - never both.
2 CHANGES SINCE DXML V2.7
2.1 BLAS1
A problem in the routine {S,D,C,Z}NRM2 has been fixed.
2.2 BLAS2
A serious computational bug in {D,S}GEMV, which was introduced in the
DXML V2.7 release, has been fixed.
All BLAS2 routines have been hand-optimized and will perform better in
most cases. In our internal testing we have noticed several cases with
speed-ups in the order of 50 - 60% for the DGEMV routine. The
performance gains for your particular application may be different.
2.3 BLAS3
The internal data copy performance of {C,D,S,Z}GEMM has been improved.
The routine {C,Z}HER2K has been slightly modified to adhere to the
defacto standard in LAPACK V2.0.
DXML V2.8 Release Notes Page 2
2.4 LAPACK
Version 2.0 of LAPACK was released from the LAPACK project to the public
on September 30, 1994. DXML V2.8 includes this new version of LAPACK.
There is a total of 84 new LAPACK routines, and 374 routines include code
that has changed since version 1.1 of LAPACK.
The following information is extracted from documents in the LAPACK
Version 2.0 distribution kit.
Version 2.0 of LAPACK introduces new routines and extends the
functionality of existing routines. The most significant new routines
and functions are:
1. the generalized nonsymmetric eigenproblem Ax = lambda Bx
2. the generalized definite banded eigenproblem Ax = lambda Bx
3. two generalized linear least squares problems (least squares with
linear constraints, and the Gauss-Markov linear model problem)
4. the generalized (quotient) singular value decomposition
5. faster routines for symmetric eigenproblems, based on
divide-and-conquer
6. a faster bidiagonal SVD (singular values only)
7. reduction of a band matrix to bidiagonal form
8. improved linear system drivers to better reuse existing
factorizations and return the pivot growth factor
9. the least squares drivers now return the residual root-sum-of-squares
Numerous bug fixes, code modifications, and/or commenting changes have
occurred also. For more details, see below.
Bug fixes and/or code modifications:
- Removed {S,D,C,Z}LAZRO; replaced with calls to {S,D,C,Z}LASET
- Fixed {S,D,C,Z}LAGS2 (AVB12 not initialized in .not. upper case)
- Fixed SLARNV (incorrect value of TWOPI)
- Fixed hermitian indefinite routines to ignore the imaginary parts of
the diagonal elements.
- Fixed SBDSQR (THRESH)
- Fixed SLAS2
- Modification to {S,D,C,Z}PTEQR and SSTERF
DXML V2.8 Release Notes Page 3
- Fixed SSTEBZ
- Fixed {S,D,C,Z}STEQR
- Modified {S,D,C,Z}TGEVC workspace WORK to 1-dimensional to avoid
WORK(N,*) problem when N=0. Various modifications to fix floating
point arithmetic problems. SAFETY parameter added to
reduce/eliminate xgg.out failures.
- {S,D,C,Z}HGEQZ SAFETY parameter added to reduce/eliminate xgg.out
failures.
New Routines:
o {S,D,C,Z}GBBRD, computational routines for banded reduction to
bidiagonal form.
o SSBGV/CHBGV, driver routine for generalized
symmetric/Hermitian-definite banded eigenproblem.
o SSBGST/CHBGST, computational routine to reduces a
symmetric/Hermitian-definite banded generalized eigenproblem to
standard form.
o SPBSTF/CPBSTF, computational routine to compute a split Cholesky
factorization of a symmetric/Hermitian positive definite band matrix.
o SSBTRD/CHBTRD, computational routine to reduce a band symmetric/-
Hermitian matrix to tridiagonal form.
o SSYEVD/CHEEVD, SSBEVD/CHBEVD, SSPEVD/CHPEVD, and SSTEVD -- divide and
conquer driver routines for symmetric eigenproblem, with accompanying
computational and auxiliary routines CLACRM, SLAED0/CLAED0, SLAED1,
SLAED2, SLAED3, SLAED4, SLAED5, SLAED6, SLAED7/CLAED7, SLAED8/CLAED8,
SLAED9, SLAEDA, SLAMRG, SLASQ1, SLASQ2, SLASQ3, SLASQ4, SLASRT, and
SSTEDC/CSTEDC.
2.4.1 LAPACK Documentation -
With this release of DXML we have included four files with a brief
description of each of the LAPACK callable routines. The information is
copied from the LAPACK manpages for Unix and has been only slightly
edited.
These files are installed at
SYS$HELP:DXML_LAPACK_C.TXT [Single Precision, Complex]
SYS$HELP:DXML_LAPACK_D.TXT [Double Precision, Real]
SYS$HELP:DXML_LAPACK_S.TXT [Single Precision, Real]
SYS$HELP:DXML_LAPACK_Z.TXT [Double Precision, Complex]
DXML V2.8 Release Notes Page 4
2.4.2 Equivalence Between LAPACK And LINPACK/EISPACK Routines -
DXML_EQUIVALENCE_LAPACK.C provides (after compilation and linking) the
name and parameter list for the routine that is equivalent to a LINPACK
or EISPACK routine.
This source program from public domain is installed in the
SYS$COMMON:[syslib] directory. After compiling and linking the program
we suggest that you define a symbol to facilitate execution.
Eg.
$ EQUIVALENCE_LAPACK :== "$device:[directory]DXML_EQUIVALENCE_LAPACK.EXE"
If you want to find the corresponding LAPACK routine for any
LINPACK/EISPACK routine, eg. DGECO, you would type:
$ EQUIVALENCE_LAPACK DGECO
2.5 Iterative Solver
The transpose-free version of the QMR solver has been added to the
iterative solver package. The name of the new routine is DITSOL_PTFQMR.
Please see the DXML Reference Manual for details.
The driver routine has also been updated to include the TFQMR solver.
This solver can be called by setting the value isolve=6.
Some routines' names were changed in DXML V2.7 and both new and old names
were supported in that release. DXML V2.8 supports only the new names.
2.6 Skyline Solver
The skyline solver package has been optimized. With our internal test
cases we have seen speed-ups of up to 2X, but the performance of the
skyline solver is very strongly dependent on the structure of the matrix.
The factor and solve routines in the skyline package have also been
optimized. As a result, some of the other routines, such as the
condition number estimator and the iterative refinement routine, are also
improved.
2.7 Signal Processing
Subroutines for fast cosine and sine transforms have been added to the
signal processing library. The new subroutines are:
Cosine Transform
{S,D}FCT
DXML V2.8 Release Notes Page 5
{S,D}FCT_INIY
{S,D}FCT_APPLY
{S,D}FCT_EXIT
Sine Transform
{S,D}FST
{S,D}FST_INIT
{S,D}FST_APPLY
{S,D}FST_EXIT
Performance of large sized 1D-FFT, GRP-FFT routines (n>8192) and
multi-dimensional FFT routines have all been further optimized.
2.8 Kit
The installation procedure has been slightly changed.
The kit size has increased from approximately 36K to 43K blocks. The
increase is caused by the new version of LAPACK and the added
functionality to the Signal Processing component.
This version of DXML was built on OpenVMS Alpha V6.1 using DEC Fortran
X6.2-614 and DEC C T1.4-023.
3 USER DOCUMENTATION
The DXML Reference Manual and the Installation Guide have been updated
for this release.
AlphaGeneration, OpenVMS, DEC Fortran are registered as trademarks of
Digital Equipment Corporation.
Copyright (c) Digital Equipment Corporation 1995. All Rights Reserved.
|
| 1869.4 | DXML V3.1 | HPCGRP::LOFGREN | | Tue Jun 20 1995 15:46 | 168 |
| (This is cross-posted from the HPCGRP::DXML Notes Conf.)
DXML Version 3.1 for Digital UNIX was submitted to cdrom consolidation on
June 19. The software is now available on the net and from SSB on the Q3CY95
layered product cdrom around July 14.
The significant changes since Version 3.0 in December 1994 are:
o The DXML V3.1 distribution kit includes two versions of the library;
(1) DXML compiled and optimized for EV5. This makes a big difference
on EV5 systems.
(2) DXML compiled and optimized for EV4.
The installation script determines if you are on an EV4 or EV5 system,
and suggests that version of the library to be installed as default.
It is possible to override the default, but it is not recommended as
the user will lose performance.
Because of the two libraries the distribution kit is quite a bit larger
than last time and additional disk space is required during the
installation. Once DXML is installed, the required disk space is
approximately the same as for the previous release.
o Performance of many routines on EV5 has been dramatically improved.
o The archive library is included with the development kit. The license
checking in the archive library routines has been disabled. This
has been done to facilitate for application developers to include and
distribute DXML routines in their products. Please see the SPD (41.86)
for details. See also note 232 in the DXML notes conference.
o A serious computational bug in the BLAS2 routine ctrsv (and ztrsv) has
been fixed.
o LAPACK 2.0 is included with this version of DXML. See Release Note for
details.
o The installation will be aborted if there is no valid license PAK
installed at the time of the installation. In the past the installation
continued, but the IVP failed.
o We have made a number of updates to the DXML manpages and included the
new Lapack manpages.
o The DXML Reference Manual and Installation Guides have been updated for
this release.
The release notes are attached.
The kits can be copied from
stcunx:/usr/ftp/pub/dxml/v31/dxml313.tar Development kit
stcunx:/usr/ftp/pub/dxml/v31/dxml313_rt.tar Run time kit
The development kit includes:
o shared (serial) libraries (libdxml_ev4.so, libdxml_ev5.so)
o archived (serial) libraries (libdxml_ev4.a. libdxml_ev5.a)
o shared (parallel) libraries (libdxmlp_ev4.so. libdxmlp_ev5.so)
o manpages
The Run Time kit includes
o shared (serial) libraries (libdxml_ev4.so, libdxml_ev5.so)
The development kit allows you to link and execute program with calls to DXML.
The run time kit allows execution only. You should never install both
kits on the same system
The parallel library should be installed on SMP configurations only.
Use anonymous ftp to copy the development kit to your current directory.
> ftp stcunx (or 16.124.32.134)
Name: ftp
Password: ident
ftp> bin
ftp> get pub/dxml/v31/dxml313.tar dxml313.tar
ftp> quit
To untar and install
> tar -xvf dxml313.tar
> su
# setld -l [directory of the expanded tar file] XMDLOA313 (serial library)
# setld -l [directory of the expanded tar file] XMDMAN313 (manpages)
# setld -l [directory of the expanded tar file] XMDPLL313 (parallel library)
Digital Extended Math Library (DXML) Version 3.1 for Digital UNIX
Final release - June 1995
1 PURPOSE OF RELEASE NOTES
These release notes provide information about DXML Version 3.1.
2 DXML DISTRIBUTION
DXML is distributed in two different forms:
1. The DXML Version 3.1 Development Kit allows you to link an
application program with the DXML library and then run the executable
image. The LMF-name for the development option is EXT-MATH-LIB. The
kit includes the following subsets:
o XMDLOA313 - includes shared library, archive library, release
notes, IVP, examples, and various DXML utilities.
o XMDMAN313 - includes the DXML manpages.
o XMDPLL313 - includes shared library for the parallel version of
the library and the parallel version of the IVP. This subset
should only be installed on SMP configurations running Digital
UNIX Version 3.0 or later.
2. The DXML Version 3.1 Run Time Kit allows an executable image that has
been linked on a different system to be executed on the target
system. The LMF-name for the run time option is EXT-MATH-LIB-RT.
The kit includes the following subset:
o XMRLOA313 - includes the run time version of the shared library,
release notes, and IVP.
Install either the development kit or the run time kit on your system -
never both.
4 REPORTING PROBLEMS AND COMMENTS TO DIGITAL
Comments or problem reports on DXML can be sent to [email protected].
The DXML developers welcome your feedback.
AlphaGeneration, DEC, AlphaServer, AlphaStation, the DIGITAL logo,
DEC Fortran are registered as trademarks of Digital Equipment
Corporation.
Digital UNIX Version 3.2 is an X/Open UNIX 93 product
UNIX is a registered trademark in the United States and other countries
licensed exclusively through X/Open Company, Limited.
Copyright (c) Digital Equipment Corporation 1995. All Rights Reserved.
|
| 1869.5 | DXML V3.2 | HPCGRP::LOFGREN | | Wed Dec 20 1995 10:36 | 393 |
|
DXML Version 3.2 (baselevel 24) for Digital UNIX is now available on the net.
It will ship from SSB on the Q1CY96 consolidated layered product cdrom around
January 12.
The significant changes since Version 3.1 in June 1995 are:
o We have parallelized the frequently called BLAS2 routine {S,D,C,Z}GEMV.
o We have parallelized almost all of the iterative solver routines.
o We have fixed a bug in the BLAS1 routines DZNRM2 and SCNRM2.
o We added code to {S,D,C,Z}GEMM to detect and take advantage of
cases when one of the matrices is very sparse.
o For all BLAS routines we have systematically searched for the compiler
switches that yield the best performance.
o We fixed a bug to allow the signal processing routines to work
properly when the environment variable PARALLEL is set to any number
between 0 and 14. If you try to set it to a number larger and 14 it will
default to 14. To set PARALLEL to any number larger than the number of
processors in the configuration is allowed, but will hurt performance.
o Support of VxWorks real time monitor from Wind River Systems, Inc. for
AXPvme embedded Alpha systems. The release notes include detailed
information about extent of support and instructions for linking, loading
and execution.
o The user documentation and manpages have been updated. The updated
DXML Reference Manual includes a new chapter on "Using the Parallel
Library" as well as much improved description of the FFT routines.
o We have added two C++ programming examples to the online examples
directory.
The release note for this baselevel is attached.
The kits can be copied from stcunx (16.124.32.134)
stcunx:/usr/ftp/pub/dxml/v32/dxml320.tar Development kit
stcunx:/usr/ftp/pub/dxml/v32/dxml320_rt.tar Run time kit
stcunx:/usr/ftp/pub/dxml/v32/docs/ User Documentation
The development kit includes:
o shared (serial) libraries (libdxml_ev4.so, libdxml_ev5.so)
o archived (serial) libraries (libdxml_ev4.a. libdxml_ev5.a)
o shared (parallel) libraries (libdxmlp_ev4.so. libdxmlp_ev5.so)
o manpages
The Run Time lit includes
o shared (serial) libraries (libdxml_ev4.so, libdxml_ev5.so)
o shared (parallel) libraries (libdxmlp_ev4.so. libdxmlp_ev5.so)
The development kit allows you to link and execute program with calls to DXML.
The run time kit allows execution only. You should never install both
kits on the same system
The parallel library should be installed on SMP configurations only.
Use anonymous ftp to copy the development kit to your current directory.
> ftp stcunx (or 16.124.32.134)
Name: ftp
Password: ident
ftp> bin
ftp> get pub/dxml/v32/dxml320.tar dxml320.tar
ftp> quit
To untar and install
> tar -xvf dxml320.tar
> su
# setld -l [directory of the expanded tar file] XMDLOA320 (serial library)
# setld -l [directory of the expanded tar file] XMDMAN320 (manpages)
# setld -l [directory of the expanded tar file] XMDPLL320 (parallel library)
Digital Extended Math Library (DXML)
Version 3.2 for Digital UNIX Release Notes
January 1996
1 Purpose of Release Notes
These release notes provide information about the DXML Version
3.2 release.
2 DXML Distribution
DXML is distributed in two different forms:
1. The DXML Version 3.2 Development Kit allows you to link an
application program with the DXML library and then run the
executable image. The LMF-name for the development option is
EXT-MATH-LIB. The kit includes the following subsets:
o XMDLOA320 - includes shared library, archive library,
release notes, IVP, examples, and various DXML utilities.
o XMDMAN320 - includes the DXML manpages.
o XMDPLL320 - includes shared library for the parallel ver-
sion of the library and the parallel version of the IVP.
This subset should only be installed on SMP configurations
running Digital UNIX Version 3.0 or later.
2. The DXML Version V3.2 Run Time Kit allows an executable image
that has been linked on a different system to be executed on
the target system. The LMF-name for the run time option is
EXT-MATH-LIB-RT. The kit includes the following subset:
o XMRLOA320 - includes the run time version of the serial
shared library, release notes, and IVP.
o XMRPLL320 - includes the run time version of the parallel
shared library, release notes, and IVP.
Install either the development kit or the run time kit on your
system - never both.
In April 1995, Digital announced a set of new Alpha Server
and Alpha Station products based on the 21164 microprocessor.
The new microprocessor and the new products are functionally
compatible with the previous generation of Alpha products.
However, as the performance characteristics of the new systems
are different from the previous systems, we are including two
different versions of the libraries with this release. The
versions are:
o DXML compiled and optimized for Alpha 21164 (EV5). This
version of the library allows users to take full advantage of
the new technology on the new systems.
o DXML compiled and optimized for the previous generation of
Alpha microprocessor such as 21064 (EV4).
The installation script will determine what system it is running
on and then prompt the installer with one of the following two
questions:
Do you want to install the EV5 library [y]?
or
Do you want to install the EV4 library [y]?
A carriage return or a y will always install the EV5 library
on an EV5 system, and the EV4 library on EV4 systems. Only
one library will be installed. It is possible to override the
default, but it is not recommended.
3 Changes Since Last Release
The previous DXML Version 3.1 was released in July 1995.
3.1 BLAS
{S,D,C,Z}GEMV has been parallelized and added to the parallel
library.
We fixed a bug in routines DZNRM2 and SCNRM2.
Code has been added to {S,D,C,Z}GEMM to detect and take advan-
tage of cases where one of the matrices is very sparse.
For all BLAS routines, we have systematically searched for the
compiler switches that yield the best performance.
3.2 Signal Processing
We fixed a bug to allow the signal processing routines to work
properly when the environment variable PARALLEL is set to any
number between 0 and 14. If you try to set PARALLEL to a number
larger than 14, it will default to 14. Setting PARALLEL to any
number larger than the number of processors in the configuration
will hurt performance.
2
3.3 Sparse Solver Package
The following routines in the iterative solver package have been
parallelized:
ditsol_driver
dmatvec_driver
dpcondl_driver
dpcondr_driver
ditsol_pbcg
ditsol_pcg
ditsol_pcgs
ditsol_pgmres
ditsol_plscg
ditsol_ptfqmr
dmatvec_genr
dmatvec_sdia
dmatvec_udia
dcreate_diag_genr
dcreate_diag_sdia
dcreate_diag_udia
dapply_diag_all
dcreate_poly_genr
dcreate_poly_sdia
dcreate_poly_udia
dapply_poly_genr
dapply_poly_sdia
dapply_poly_udia
The serial versions of the following routines have been opti-
mized:
dapply_ilu_genr_l
dapply_ilu_sdia
dapply_ilu_udia_l
dapply_ilu_udia_u
dmatvec_poly_sdia
dmatvec_poly_udia
dmatvec_sdia
3.4 Online Examples
We have added two new C++ examples
example_itsol_1.cxx
example_skysol_1.cxx
3
3.5 Manpages
We have made a number of minor updates and corrections to the
DXML manpages. We have added a manpage on VxWorks.
3.6 DXML for the VxWorks
DXML Version 3.2 supports development of applications targeted
for AXPvme embedded systems running under VxWorks.
The archived DXML library must be installed on the host sys-
tem. Only routines from BLAS, Lapack and Signal Processing are
available for the VxWorks environment.
Operating instructions for compiling, linking and loading
applications with calls to DXML routines are as follows:
o To compile on the host, enter the following command:
cc -c -I/[path-spec] -non_shared -std1 -DCPU=21064 f1.c
Where:
[path-spec] -- the access path to include files in
specified directories
DCPU=21064 -- the Alpha chip
f1.c -- the file name of the example
This command creates the object file f1.o.
o To create a VxWorks loadable image on the host, enter the
following command:
ld -non_shared -o output.o -r f1.o \
/usr/opt/XMDLOA320/dxml/end.o -ldxml -lots \
lib21064osfmath.a
Where:
-r -- retains relocation entries in the output file,
output.o, for the VxWorks loader.
end.o -- a dummy object file provided in the DXML kit to let
the VxWorks loader bypass some symbols it cannot resolve.
lib21064osfmath.a -- the VxWorks provided version of
libm.a.
This command creates the output file output.o.
4
o To load the image on the target, enter the following command:
ld < output.o
After the image is successfully loaded, it is ready to run.
You can verify the newly loaded image output.o by entering:
moduleShow
For details and specifics on building, loading and unloading
application modules, please refer to the Cross Development
chapter of the VxWorks Programmer's Guide.
4 Kit
This version of DXML was built on Digital UNIX V3.2C (rev 214)
using DEC Fortran T3.8-711, DEC C V5.0-106. KAP for DEC Fortran
Version 2.1 was used to parallelize the _GEMV, _GEMM, _GETRF,
_POTRF and iterative solver routines.
5 User Documentation
The DXML Reference Manual and Installation Guides have been
updated for this release. We have added a chapter on parallel
processing and improved the signal processing chapter.
6 Reporting Problems and Comments to Digital
Comments or problem reports on DXML can be sent to
[email protected]. The DXML developers welcome your feedback.
AlphaGeneration, AlphaServer, Bookreader, and the DIGITAL logo
are trademarks of Digital Equipment Corporation.
VxWorks is a registered trademark of Wind River Systems, Inc.
Copyright Digital Equipment Corporation 1996. All rights reserved
5
|
| 1869.6 | DXML 2.9 on OpenVMS Alpha | HPCGRP::LOFGREN | | Mon Apr 29 1996 11:11 | 247 |
|
DXML Version 2.9 (baselevel 24) for OpenVMS Alpha is now available on the net.
The kit has been submitted and will ship from SSB on the Q2CY96 consolidated
layered product cdrom for OpenVMS Alpha in the middle of June.
The significant changes since Version 2.8 in January 1995 are:
o The distribution kit on cdrom includes one save set optimized for EV4 and
a different save set optimized for EV5. The installation procedure has
been updated to determine if your system is EV4 or EV5 and to install the
corresponding library automatically. If you copy the kit over the network
you need to copy only the .a and .b save sets for an EV4 system, or
.a and .c for EV5.
o The ctrsv (ztrsv) bug mentioned in note 219 in this notes file is fixed.
o We have fixed a bug in the BLAS1 routines DZNRM2 and SCNRM2.
o We added code to {S,D,C,Z}GEMM to detect and take advantage of
cases when one of the matrices is very sparse.
o For all BLAS routines we have systematically searched for the compiler
switches that yield the best performance.
o We have added two C++ programming examples to the online examples
directory.
o We fixed a bug in the Skyline Solver routines for matrix sizes less than
12 (see note 276).
The release notes are attached for your convenience.
The kits can be copied from HPCGRP::KITS:[DXML.ALPHA.V29]:
DXMLDVAV029.A Development kit - common routines
DXMLDVAV029.B Development kit - EV4 routines
DXMLDVAV029.C Development kit - EV5 routines
DXMLRTAV029.A Run Time kit - common routines
DXMLRTAV029.B Run Time kit - EV4 routines
DXMLRTAV029.C Run Time kit - EV5 routines
DXML_29_IG.PS Installation Guide
DXML_29_IG.TXT
DXML_29_REF.PS Reference Manual (BIG - 600+ pages)
DXML_29_SPD.PS SPD
DXML_29_SPD.TXT
The Development kit allows you to link and execute programs with calls to DXML.
The Run Time kit allows execution only.
You should never install both the development and the run time kit.
DXML license PAKs from previous release will still work. Otherwise PAKs are
available from VTX PAK. The LMF name is EXT-MATH-LIB for the development kit.
Digital Extended Math Library (DXML)
Version 2.9 for OpenVMS Alpha Release Notes
April 1996
1 Purpose of Release Notes
These release notes provide information about DXML Version 2.9
and about user documentation. DXML Version 2.9 is a maintenance
release without any significant technical changes.
2 DXML Distribution
DXML is distributed in two different forms:
1. The DXML Version 2.9 Development Kit allows you to link an
application program with the DXML library and then run the
executable image. The names of the save sets for the develop-
ment kit are DXMLDVAV029.A, DXMLDVAV029.B (DXML optimized for
EV4, for example, the AlphaServer 4/nnn), and DXMLDVAV029.C
(DXML optimized for EV5, for example, the AlphaServer 5/nnn).
2. The DXML Version 2.9 Run-Time Kit allows an executable image
that has been linked on a different system to be executed on
the target system. The names of the save sets for the run-
time kit are DXMLRTAV029.A, DXMLRTAV029.B (DXML optimized for
EV4, for example, the AlphaServer 4/nnn), and DXMLRTAV029.C
(DXML optimized for EV5, for example, the AlphaServer 5/nnn).
You should install either the development kit or the run-time
kit on your system - never both.
3 Changes Since DXML Version 2.8
The previous DXML Version 2.8 was released in January 1995.
Changes made for Version 2.9 are as follows:
o In April 1995 Digital introduced the new Alpha processor
21164 (also known as EV5) and several products based on the
new processor, such as the AlphaServer 5/nnn. The new pro-
cessor and products are functionally compatible with the
previous Alpha products, but they offer enhanced performance
of many DXML routines. The Version 2.9 distribution kit in-
cludes a version of the DXML library optimized for the EV5
processor as well as a version optimized for the EV4 pro-
cessor. The installation procedure automatically determines
whether your system is an EV4 or an EV5 and then installs the
corresponding save set.
o A bug in the BLAS 1 routines DZNRM2 and SCNRM2 has been
fixed.
o A serious computational bug in the BLAS 2 routines CTRSV
and ZTRSV has been fixed. The problem occurred only when the
routines were called with uplo='L' and trans='N' and only
for certain array dimensions. As CTRSV and ZTRSV are called
directly or indirectly by a number of other LAPACK and BLAS
routines, the incorrect results may have appeared to come
from some other routines.
o Code has been added to the BLAS 3 {S,D,C,Z}GEMM to detect
and take advantage of cases where one of the matrices is very
sparse.
o All BLAS routines have been systematically searched for the
compiler switches that yield the best performance.
o A bug in the Skyline Solver routines for matrix sizes less
than 12 has been fixed.
o The following Sparse Solver Package routines have been opti-
mized:
dapply_ilu_genr_l
dapply_ilu_genr_u
dapply_ilu_sdia
dapply_ilu_udia_l
dapply_ilu_udia_u
dmatvec_poly_sdia
dmatvec_poly_udia
dmatvec_sdia
o New examples of calling DXML from programs written in C and
C++ have been added.
4 LAPACK Documentation
With this release of DXML we have again included four files with
a brief description of each of the LAPACK callable routines. The
information is copied from the LAPACK manpages for UNIX and has
been edited only slightly.
These files are installed at:
SYS$HELP:DXML_LAPACK_C.TXT [Single Precision, Complex]
SYS$HELP:DXML_LAPACK_D.TXT [Double Precision, Real]
SYS$HELP:DXML_LAPACK_S.TXT [Single Precision, Real]
SYS$HELP:DXML_LAPACK_Z.TXT [Double Precision, Complex]
5 Kit
The installation procedure has been slightly changed, as ex-
plained in Section 3.
This version of DXML was built on OpenVMS Alpha V6.2-FT3 using
Digital Fortran T7.0-805 and DEC C T5.2-003.
2
6 User Documentation
The DXML Reference Manual and the Installation Guide have been
updated for this release.
7 Reporting Problems and Comments to Digital
Comments or problem reports on DXML can be sent to
[email protected]. The DXML developers welcome your feedback.
AlphaGeneration, AlphaServer, AlphaStation, OpenVMS, and the
DIGITAL logo are registered as trademarks of Digital Equipment
Corporation.
�Digital Equipment Corporation. 1996. All Rights Reserved.
3
|
| 1869.7 | DXML 3.3 on Digital UNIX | HPCGRP::LOFGREN | | Wed Sep 18 1996 16:40 | 468
|