Benchmark with New libcpml.a for Alpha/Linux

Benchmark with New libcpml.a for Alpha/Linux

Post by NewsMa » Sat, 06 Mar 1999 04:00:00



Sorry, I have not done it yet.

BTW, there is no instruction for proper installation
of libcpml.a.
I see /lib/libm.so.* and /usr/lib/libm.a in the original RedHat
distribution.  So what should I do to use libcpml.a?

Thank you.

Those who wants to benchmark the new library with the
Polyhedron's F77BENCH.ZIP on a Unix-like machine will find
ftp://penguin.kjist.ac.kr/pub/fortran.d/f77bench.tar.gz
useful and much more convenient than the original
zip file.  It's got a Makefile!

G. Hugh Song

--
Newsman

PS: Reply to ghsong "at" kjist "dot" ac "dot" kr, if you really have to.

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by Steve Lion » Sat, 06 Mar 1999 04:00:00


On Fri, 05 Mar 1999 16:09:53 +0900, NewsMan


>BTW, there is no instruction for proper installation
>of libcpml.a.
>I see /lib/libm.so.* and /usr/lib/libm.a in the original RedHat
>distribution.  So what should I do to use libcpml.a?

The way it was explained to me is that you're expected to just stick
libcpml.a in /usr/lib and use -lcpml in your link.  If you want to go
further and actually replace the Linux libm with libcpml (renaming it
in the process), go ahead, but we're not suggesting/recommending that
approach.


Fortran Engineering
Compaq Computer Corporation, Nashua NH

Compaq Fortran web site: http://www.compaq.com/fortran

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by NewsMa » Sun, 07 Mar 1999 04:00:00



> Sorry, I have not done it yet.

> BTW, there is no instruction for proper installation
> of libcpml.a.
> I see /lib/libm.so.* and /usr/lib/libm.a in the original RedHat
> distribution.  So what should I do to use libcpml.a?

> Thank you.

> Those who wants to benchmark the new library with the
> Polyhedron's F77BENCH.ZIP on a Unix-like machine will find
> ftp://penguin.kjist.ac.kr/pub/fortran.d/f77bench.tar.gz
> useful and much more convenient than the original
> zip file.  It's got a Makefile!

> G. Hugh Song

> --
> Newsman

> PS: Reply to ghsong "at" kjist "dot" ac "dot" kr, if you really have to.

I have read Steve Lionel's reply on the above question of mine.
In fact, I replaced /usr/lib/libm.a with libcpml.a.
I am not sure.  But, I don't see any change (from libffm.a) in
the benchmark result after using libcpml.a.  Maybe I am
not using it properly.  BTW, I have installed GOTO's
other custom BLAS routines also.

Because the untared file is ready to compile and run, I would like
to ask people to obtain the benchmark results for all of us.

Thank you.

The original f77bench is found at http://www.polyhedron.co.uk/

G. Hugh Song

--
Newsman

PS: Reply to ghsong "at" kjist "dot" ac "dot" kr, if you really have to.

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by Robert J. Goodin » Mon, 08 Mar 1999 04:00:00


Certain speed advantages follow from linking different parts of
Joachim Wesner's libffm with cpml; the most dramatic that I have found comes
from pow.o. E.g.

g77 -o t.e t1.o t2.o -lcpml /usr/local/libffm.0.28/pow.o -lfm

is twice as fast for codes with intensive use of the pow routine.
Note that adding fm is crucial for allowing pow.o to find the
requisite calls (not that I understand this ordering of linkings).

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by Lee Bo-sun » Tue, 09 Mar 1999 04:00:00


The result of my fortran code with various math libraries.
I used egcs-1.1.1 and single precision and '-O3' option only.

- with 'libm'

87.123u 0.086s 1:27.54 99.6%    0+0k 0+0io 119pf+0w

- with 'libffm'

54.054u 0.053s 0:54.21 99.7%    0+0k 0+0io 120pf+0w

- with 'libcpml'

54.438u 0.074s 0:54.75 99.5%    0+0k 0+0io 122pf+0w

- on CrayT3E-900(450 MHz Alpha) with Cray's 'libfm' and CF90 compiler.
default double precision


62.93u 2.14s 1:13.12 89.0%

There is no difference between 'libffm' and 'libcpml' and two are still
slower than Digital Visual Fortran on Win/NT-alpha.

--
Lee, Bo-sung
Ph.D Candidate in Aerospace Eng. Seoul National Univ.
Invited Reseacher in ETRI/Supercomputing Center


http://www.shinbiro.com/~bslee6

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by NewsMa » Tue, 09 Mar 1999 04:00:00



> Certain speed advantages follow from linking different parts of
> Joachim Wesner's libffm with cpml; the most dramatic that I have found comes
> from pow.o. E.g.

> g77 -o t.e t1.o t2.o -lcpml /usr/local/libffm.0.28/pow.o -lfm

> is twice as fast for codes with intensive use of the pow routine.
> Note that adding fm is crucial for allowing pow.o to find the
> requisite calls (not that I understand this ordering of linkings).

libfm.a used to exist for libffm.a 0.21.
libfm does not seem to exist in libffm.a version 0.28.
Are you sure that we still need libfm.a with libffm.a 0.28?
There was a newer verison of libm.a that is supposed to
be used with libffm.a 0.28.

So you are saying  that we need libfm.a back.
What is going on?

Thanks.

--
Newsman

PS: Reply to ghsong "at" kjist "dot" ac "dot" kr, if you really have to.

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by Martin Kahle » Tue, 09 Mar 1999 04:00:00





> I have read Steve Lionel's reply on the above question of mine.
> In fact, I replaced /usr/lib/libm.a with libcpml.a.
> I am not sure.  But, I don't see any change (from libffm.a) in
> the benchmark result after using libcpml.a.  Maybe I am
> not using it properly.  BTW, I have installed GOTO's
> other custom BLAS routines also.

The compilers default to  link with the shared lib libm.so.*
Did you use the statically linked version libm.a?
(Try: ldd your_prog, there should be no libm.so.*)

Bye,
Martin.

--
Your mouse has moved. Windows must be restarted for the change
to take effect. Reboot now?

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by News Ma » Tue, 09 Mar 1999 04:00:00


I just realized why nobody responded to my request for a
benchmark result.  My penguin did not accept anonymous
ftp login.  I thus moved the tar file to a new location.

ftp://infcom.kjist.ac.kr/pub/fortran.d/f77bench.tar.gz

Please try.

BTW, my penguin accepts anonymous ftp logins within
our domain.  So, I thought it would accept ftp-logins from
outside our domain.  Does it not?  Appaerenly not.
Redhat-5.1 was put on my Penguin.  What should I do
to accept anonymous ftp logins from outside of our
domain?

Thank you.



> > Sorry, I have not done it yet.

> > BTW, there is no instruction for proper installation
> > of libcpml.a.
> > I see /lib/libm.so.* and /usr/lib/libm.a in the original RedHat
> > distribution.  So what should I do to use libcpml.a?

> > Thank you.

> > Those who wants to benchmark the new library with the
> > Polyhedron's F77BENCH.ZIP on a Unix-like machine will find
> > ftp://penguin.kjist.ac.kr/pub/fortran.d/f77bench.tar.gz
> > useful and much more convenient than the original
> > zip file.  It's got a Makefile!

> > G. Hugh Song

> > --
> > Newsman

> > PS: Reply to ghsong "at" kjist "dot" ac "dot" kr, if you really have to.

> I have read Steve Lionel's reply on the above question of mine.
> In fact, I replaced /usr/lib/libm.a with libcpml.a.
> I am not sure.  But, I don't see any change (from libffm.a) in
> the benchmark result after using libcpml.a.  Maybe I am
> not using it properly.  BTW, I have installed GOTO's
> other custom BLAS routines also.

> Because the untared file is ready to compile and run, I would like
> to ask people to obtain the benchmark results for all of us.

> Thank you.

> The original f77bench is found at http://www.polyhedron.co.uk/

> G. Hugh Song

> --
> Newsman

> PS: Reply to ghsong "at" kjist "dot" ac "dot" kr, if you really have to.

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by Rob Kom » Tue, 09 Mar 1999 04:00:00


: BTW, my penguin accepts anonymous ftp logins within
: our domain.  So, I thought it would accept ftp-logins from
: outside our domain.  Does it not?  Appaerenly not.
: Redhat-5.1 was put on my Penguin.  What should I do
: to accept anonymous ftp logins from outside of our
: domain?

You probably have to play with the /etc/hosts.allow and
/etc/hosts.deny files.  On my Slackware box, the man
pages are found under hosts_access(5).

Cheers,
Rob Komar

 
 
 

Benchmark with New libcpml.a for Alpha/Linux

Post by bg.. » Tue, 09 Mar 1999 04:00:00



> BTW, my penguin accepts anonymous ftp logins within
> our domain.  So, I thought it would accept ftp-logins from
> outside our domain.  Does it not?  Appaerenly not.

Comp.os.linux.setup would seem a more appropriate forum in
which to ask that question. However, I'd recommend asking
your local Linux gurus first. Two obvious places to look at
are /etc/hosts.{allow,deny} and (if you are using wu-ftpd)
the ftpaccess file. (I'm no longer sure of all the details,
and can't look them up right now since the system I'm writing this
on doesn't have wu-ftpd installed.)

Quote:> Redhat-5.1 was put on my Penguin.  What should I do
> to accept anonymous ftp logins from outside of our
> domain?

It is also conceivable that your domain sits behind a firewall, with
direct inbound ftp disabled. Ask your network administrators.
 
 
 

1. anonymous ftp (was: Benchmark with New libcpml.a for Alpha/Linux)

Comp.os.linux.setup would seem a more appropriate forum in
which to ask that question. However, I'd recommend asking
your local Linux gurus first. Two obvious places to look at
are /etc/hosts.{allow,deny} and (if you are using wu-ftpd)
the ftpaccess file. (I'm no longer sure of all the details,
and can't look them up right now since the system I'm writing this
on doesn't have wu-ftpd installed.)

It is also conceivable that your domain sits behind a firewall, with
direct inbound ftp disabled. Ask your network administrators.

2. USENIX '03 Call for papers

3. Still Disappointing libcpml.a for Alpha/Linux

4. Jumpstart-Question DHCP

5. benchmarks on linux/alpha?

6. NIS client set up?

7. iptables question

8. Athlon, Alpha linux benchmarks...

9. New Linux BYTE benchmark

10. Alpha and 2xPPro Benchmarks?

11. Dec Alpha vs. Pentium II benchmarks???

12. rwsem benchmarks [Re: generic rwsem [Re: Alpha "process table hang"]]