perfsuite-users — Mailing list for users of PerfSuite software

Re: [PerfSuite-users] PerfSuite

[Rick K. <rk...@nc...>]

MeMo,

On Wed, 21 Dec 2005, Mehmet Belgin wrote:

> Hello everyone,
> 
> I just installed PerfSuite, but psrun gives the following error:
> 
> #psrun ./ps_test
> ERROR: ld.so: object 'libpsrun.so.0' from LD_PRELOAD cannot be preloaded:
> ignored.

I recall seeing this at some point but am embarrassed to say that I don't 
remember what the particular source of the ld.so complaint was.  Can you 
say anything more about what the contents of ps_test are?  That might help 
reproduce/diagnose the case.

> 
> I installed PerfSuite with PAPI and tDOM support. I already had a recompiled
> kernel for perfctr support with PAPI installed. I can see 'libpsrun.so.0' under
> /usr/local/lib. I have /usr/local/lib listed in /etc/ld.so.conf (as usual)
> and I ran ldconfig after the installation just to be sure. This is all I can
> do, so I need your help please!
>

This sounds as though you have set up everything properly.  I should 
mention that the current PerfSuite releases (0.6.1 or 0.6.2a2) do not 
support the 2.6 kernel, so you may run into problems related to that
if you are running that kernel version.  "uname -r" should report your
kernel version.

2.6.x support is on its way - there have been a few inquiries/reports to 
this list about it - plans had been to release prior to year end, however
some activities at NCSA have reduced available time in the short term, so 
it's looking like early January is a good bet for the first release 
(0.6.2a3) with 2.6 support.  It is a substantial release with 
numerous changes.  Thanks to all for your patience on this.

Rick

> Thank you in advance... > 
> MeMo
> 

[PerfSuite-users] PerfSuite

[Mehmet B. <meh...@gm...>]

Hello everyone,

I just installed PerfSuite, but psrun gives the following error:

#psrun ./ps_test
ERROR: ld.so: object 'libpsrun.so.0' from LD_PRELOAD cannot be preloaded:
ignored.

I installed PerfSuite with PAPI and tDOM support. I already had a recompile=
d
kernel for perfctr support with PAPI installed. I can see 'libpsrun.so.0' u=
nder
/usr/local/lib. I have /usr/local/lib listed in /etc/ld.so.conf (as usual)
and I ran ldconfig after the installation just to be sure. This is all I ca=
n
do, so I need your help please!

Thank you in advance...

MeMo

Re: [PerfSuite-users] perfsuite and mpi

[Rick K. <rk...@nc...>]

> If you add calls to pshwpc (as you have done):
> 
>    -lpshwpc -lperfsuite -lpapi
> 
> If you do not, and it is an MPI program:
> 
>    -lpshwpc_mpi -lpshwpc -lperfsuite -lpapi

Giuseppe,

I neglected to mention one other library that may be necessary when 
linking: it is the Expat parser that is used for parsing XML documents.
If you still have missing external routines, with names like 
XML_ParserCreate, then this indicates that Expat is needed.  To include it 
in your link line, try adding -lexpat at the end.

Apologies - I think much of this confusion is due to the need for better 
documentation of PerfSuite.

Rick

Re: [PerfSuite-users] perfsuite and mpi

[Rick K. <rk...@nc...>]

Giuseppe,

You will want to include -lpshwpc and -lperfsuite in your link line,
after the -lpshwpc_mpi.  Those libraries have the externals that are 
missing.

Also, just to clarify, libpshwpc_mpi is a PMPI-based interface.  There is
no need to add any calls to PerfSuite if you link with that (it's likely
it will result in a runtime error, actually).

So in summary:

If you add calls to pshwpc (as you have done):

   -lpshwpc -lperfsuite -lpapi

If you do not, and it is an MPI program:

   -lpshwpc_mpi -lpshwpc -lperfsuite -lpapi

Plus you may need to adjust the MPI libraries linked in for the MPI 
profiling interface, which is independent of PerfSuite.

Good luck,
Rick

On Wed, 14 Dec 2005, Giuseppe Grieco wrote:

> Dear Rick,
> 
> I am following to try to use perfsuite with mpi, but without success. 
> What I do is editing the file cpi-mpif.f present in the folder 
> /examples/cpi/ by inserting calls to the four subroutine of psrun. It 
> makes the object file but does not links correctly. What I am doing is 
> the following
> 
> % /usr/local/mpich-1.2.7/build/linux_intel/bin/mpif90  cpi-mpif.f 
> -static -o cpi -I/opt/perfsuite_intel/include -L/opt/perfsuite_intel/lib 
> -L/usr/local/lib -lpshwpc_mpi -lpapi
> 
> The answer is
> 
> IPO Error: unresolved : psf_hwpc_init_
>         Referenced in cpi-mpif.o
> IPO Error: unresolved : psf_hwpc_start_
>         Referenced in cpi-mpif.o
> IPO Error: unresolved : psf_hwpc_stop_
>         Referenced in cpi-mpif.o
> IPO Error: unresolved : psf_hwpc_shutdown_
>         Referenced in cpi-mpif.o
> IPO Error: unresolved : ps_hwpc_init
>         Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
> IPO Error: unresolved : ps_hwpc_start
>         Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
> IPO Error: unresolved : _ps_malloc
>         Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
> IPO Error: unresolved : ps_hwpc_stop
>         Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
> IPO Error: unresolved : ps_hwpc_shutdown
>         Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
> IPO Error: unresolved : _ps_getargs
>         Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
> cpi-mpif.o(.text+0x3d): In function `MAIN__':
> : undefined reference to `psf_hwpc_init_'
> cpi-mpif.o(.text+0x10e): In function `MAIN__':
> : undefined reference to `psf_hwpc_start_'
> cpi-mpif.o(.text+0x1a7): In function `MAIN__':
> : undefined reference to `psf_hwpc_stop_'
> cpi-mpif.o(.text+0x1c0): In function `MAIN__':
> : undefined reference to `psf_hwpc_shutdown_'
> /opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0x2d): 
> In function `MPI_Init':
> /opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:88: undefined reference to 
> `ps_hwpc_init'
> /opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0x36):/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:90: 
> undefined reference to `ps_hwpc_start'
> /opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0xc3): 
> In function `MPI_Finalize':
> /opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:122: undefined reference 
> to `_ps_malloc'
> /opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0xf0):/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:131: 
> undefined reference to `ps_hwpc_stop'
> /opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0x10d):/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:133: 
> undefined reference to `ps_hwpc_shutdown'
> /opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0x137): 
> In function `mpi_init_':
> /opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:155: undefined reference 
> to `_ps_getargs'
> 
> Thanks, Giuseppe.
> 
> 
> 
> 
> -------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
> for problems?  Stop!  Download the new AJAX search engine that makes
> searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
> http://ads.osdn.com/?ad_id=7637&alloc_id=16865&op=click
> _______________________________________________
> PerfSuite-users mailing list
> Per...@li...
> https://lists.sourceforge.net/lists/listinfo/perfsuite-users
> 

[PerfSuite-users] perfsuite and mpi

[Giuseppe G. <gg9...@un...>]

Dear Rick,

I am following to try to use perfsuite with mpi, but without success. 
What I do is editing the file cpi-mpif.f present in the folder 
/examples/cpi/ by inserting calls to the four subroutine of psrun. It 
makes the object file but does not links correctly. What I am doing is 
the following

% /usr/local/mpich-1.2.7/build/linux_intel/bin/mpif90  cpi-mpif.f 
-static -o cpi -I/opt/perfsuite_intel/include -L/opt/perfsuite_intel/lib 
-L/usr/local/lib -lpshwpc_mpi -lpapi

The answer is

IPO Error: unresolved : psf_hwpc_init_
        Referenced in cpi-mpif.o
IPO Error: unresolved : psf_hwpc_start_
        Referenced in cpi-mpif.o
IPO Error: unresolved : psf_hwpc_stop_
        Referenced in cpi-mpif.o
IPO Error: unresolved : psf_hwpc_shutdown_
        Referenced in cpi-mpif.o
IPO Error: unresolved : ps_hwpc_init
        Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
IPO Error: unresolved : ps_hwpc_start
        Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
IPO Error: unresolved : _ps_malloc
        Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
IPO Error: unresolved : ps_hwpc_stop
        Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
IPO Error: unresolved : ps_hwpc_shutdown
        Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
IPO Error: unresolved : _ps_getargs
        Referenced in libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)
cpi-mpif.o(.text+0x3d): In function `MAIN__':
: undefined reference to `psf_hwpc_init_'
cpi-mpif.o(.text+0x10e): In function `MAIN__':
: undefined reference to `psf_hwpc_start_'
cpi-mpif.o(.text+0x1a7): In function `MAIN__':
: undefined reference to `psf_hwpc_stop_'
cpi-mpif.o(.text+0x1c0): In function `MAIN__':
: undefined reference to `psf_hwpc_shutdown_'
/opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0x2d): 
In function `MPI_Init':
/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:88: undefined reference to 
`ps_hwpc_init'
/opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0x36):/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:90: 
undefined reference to `ps_hwpc_start'
/opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0xc3): 
In function `MPI_Finalize':
/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:122: undefined reference 
to `_ps_malloc'
/opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0xf0):/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:131: 
undefined reference to `ps_hwpc_stop'
/opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0x10d):/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:133: 
undefined reference to `ps_hwpc_shutdown'
/opt/perfsuite_intel/lib/libpshwpc_mpi.a(libpshwpc_mpi_a-hwpc-mpi.o)(.text+0x137): 
In function `mpi_init_':
/opt/perfsuite-0.6.1/src/libpshwpc/hwpc-mpi.c:155: undefined reference 
to `_ps_getargs'

Thanks, Giuseppe.

Re: [PerfSuite-users] mpirun with perfsuite

[Rick K. <rk...@nc...>]

On Wed, 7 Dec 2005, Giuseppe Grieco wrote:

> Dear Rick,
> I tried to run =A0
>=20
>  % /usr/local/mpich-1.2.7/bin/mpirun -np 2 /usr/local/bin/psrun cpi-mpi
>=20
> as I have ch_p4 device for MPICH installed. As soon as I will try the o=
ther way you suggested me, I will let you know the result.
> Now I would like to make you another question:
> in an old mail I asked you if it was possible install perfsuite with di=
fferent fortran compilers and part of your answer was:
>=20
> [ ... ]
>=20
> I receive errors. I try to install perfsuite with mpich that has differ=
ent installation for GNU compilers and intel compilers. So when I set MPI=
CPPFLAGS for step 4 is different than for step 1. Does it cause problems?=
 If yes, is it possible to install different versions of perfsuite, as fo=
r mpich?
> Always thanks for your support.

Giuseppe,

Since you are using MPICH ch_p4, I would guess that psrun will not be=20
effective for you for the reasons that I mentioned previously (command=20
line getting mangled).

Regarding how to install multiple versions of PerfSuite on one system,=20
each using different configuration options: PerfSuite uses GNU Autotools=20
for its build system, and that is quite flexible, so most combinations ar=
e=20
possible and should work with multiple Fortran compilers, multiple MPI=20
implementations, etc.  The primary limitation is on how much disk space=20
you can afford to use for each additional installation.

We routinely use multiple compiler vendors (GNU, Intel, Open64), and=20
multiple MPI implementations (MPICH ch_p4/GM, ChaMPIon/Pro), so it is=20
definitely possible to do this, but does take some planning and=20
familiarity with software installation, calling conventions, etc.=20

Probably the most straightforward way to maintain different versions of=20
PerfSuite is to use a distinct "--prefix" option for each variant.  For=20
example, let's say you want GNU and Intel compiler support with MPICH and=
=20
LAM as MPI layers and you want to mix and match.  This would be a total o=
f=20
4 combinations.  You could do:

./configure --prefix=3D/opt/perfsuite-gnu-mpich --enable-mpi \
    MPICPPFLAGS=3D"-I/opt/mpich/include"
make && make install

./configure --prefix=3D/opt/perfsuite-gnu-lam --enable-mpi \
    MPICPPFLAGS=3D"-I/opt/lam/include"
make && make install

./configure --prefix=3D/opt/perfsuite-intel-mpich --enable-mpi \
    MPICPPFLAGS=3D"-I/opt/mpich/include" F77=3Difort
make && make install

./configure --prefix=3D/opt/perfsuite-intel-lam --enable-mpi \
    MPICPPFLAGS=3D"-I/opt/lam/include" F77=3Difort
make && make install

To use any one of these versions, you would then have to make sure you=20
have "source"d the proper "psenv" script and probably have to arrange for=
=20
PATH/LD_LIBRARY_PATH settings so that remote MPI tasks, when launched=20
through rsh or ssh are set properly.  There are numerous site-specific=20
things that are likely to crop up, and for that you have to use some=20
investigative skills to resolve.

Please note that it is <mpi.h> and the setting of F77 that is most=20
relevant to the PerfSuite library.  It is up to the user and/or the mpicc=
=20
or mpif77 command to select the proper MPI library itself when building a=
n=20
MPI application.  That is, nothing in the PerfSuite build actually links=20
with an existing MPI library... this is deferred until the MPI applicatio=
n=20
is actually built or run.

I would say that, for the person just getting started using PerfSuite,=20
it's probably easiest to choose just one preferred Fortran compiler and=20
MPI installation and work exclusively with that first, so that you can ge=
t=20
familiar with PerfSuite and MPI without worrying about multiple installat=
ions. =20
Then once you have the basics down, expand to additional versions as=20
needed.

Good luck,
Rick

Re: [PerfSuite-users] mpirun with perfsuite

[Rick K. <rk...@nc...>]

Giuseppe,

There is one other thing I forgot to mention here:

> /usr/local/mpich-1.2.7/bin/mpirun -np 2 /usr/local/bin/psrun cpi-mpi

If this option still doesn't work for you, it might be that the MPI 
implementation is changing the command-line arguments supplied to psrun 
when the MPI program actually starts on the remote nodes.  This would be 
the case, for example, if you are using the "ch_p4" version of MPICH as 
supplied by Argonne.  What is happening there is that psrun isn't actually 
seeing the command "cpi-mpi", but something else altogether that is put 
back in order by the MPI routine MPI_Init.  However, that is too late for 
psrun to be able to do its work.

The only workarounds available for this problem are:

1. link the MPI application with the PMPI version of PerfSuite's libpshwpc
library, which is built if you supplied the configure-time option
--enable-mpi.  The library is called "libpshwpc_mpi" and will be installed
in the standard PerfSuite library directory.

2. Insert calls to libpshwpc in the MPI application itself.

Both of these workarounds eliminate the need for using psrun altogether, 
so you should omit psrun in your mpirun command line if you use them.

This topic was discussed some months back in this mailing list but I 
thought it would be helpful to recap.

Thanks,
Rick

Re: [PerfSuite-users] mpirun with perfsuite

[Rick K. <rk...@nc...>]

On Tue, 6 Dec 2005 gg...@un... wrote:

> I am using perfsuite-0.6.1 with GNU compilers gcc and g77 and with mpich.1.2.7.
> It seems that all works properly, but when I run
> 
> % /usr/local/mpich-1.2.7/bin/mpirun -np 2 psrun cpi-mpi
> 
> the answer is
> 
> Unrecognized argument psrun ignored.
> Process 0 on gemelli
> Process 1 on gemelli
> pi is approximately 3.1415926535899814, Error is 0.0000000000001883
> wall clock time = 1.040470
> 
> What happens? Thanks in advance.

Giuseppe,

This is a situation that depends on the version of MPI you are using and 
is a common problem.  The output "Unrecognized argument" implies that the 
MPI launch command mpirun is interpreting psrun as an option (which it 
doesn't recognize and therefore ignores).  You might try supplying the 
full pathname to psrun, for example:

/usr/local/mpich-1.2.7/bin/mpirun -np 2 /usr/local/bin/psrun cpi-mpi

Hope this helps, 
Rick

p.s. I am redirecting your question from perfsuite-bugs to perfsuite-users
because I think this is of general interest.

[PerfSuite-users] mpirun with perfsuite

[<gg...@un...>]

I am using perfsuite-0.6.1 with GNU compilers gcc and g77 and with mpich.=
1.2.7.
It seems that all works properly, but when I run

% /usr/local/mpich-1.2.7/bin/mpirun -np 2 psrun cpi-mpi

the answer is

Unrecognized argument psrun ignored.
Process 0 on gemelli
Process 1 on gemelli
pi is approximately 3.1415926535899814, Error is 0.0000000000001883
wall clock time =3D 1.040470

What happens? Thanks in advance.

Dr. Giuseppe Grieco
DIFA Universit=E0 della Basilicata

Re: [PerfSuite-users] forrtl: severe (168)

[Rick K. <rk...@nc...>]

Giuseppe,

On Thu, 24 Nov 2005 gg...@un... wrote:

> Dear all,
> I installed perfsuite successfully, but when I run the examples in matvec folder,
> after 
> 
> % psrun ./matvec-r4
>  
> I receive the following message
> 
> forrtl: severe (168): Program Exception - illegal instruction
> Image              PC        Routine            Line        Source
> matvec-r4          0804A335  Unknown               Unknown  Unknown
> matvec-r4          0804A2C5  Unknown               Unknown  Unknown
> libc.so.6          40243A07  Unknown               Unknown  Unknown
> matvec-r4          0804A1F1  Unknown               Unknown  Unknown
> 
> I do not know what it happens. Is there somebody who can help me? Thanks in
> advance, Giuseppe.
> 

It's a little hard to tell what's going on without some additional 
information, such as the type of system on which you are running 
(processor, OS), the version of the Fortran compiler you are using (I'm
guessing Intel, if so you can use the "-V" option of the compiler to
retrieve compiler version). If running with PAPI, then version info there 
might also be useful.  Lastly, the output of the PerfSuite test suite can 
have some useful information (the result of "make check").

If using the Intel compiler, we have seen that for some versions (not 
all), standard profiling is broken and generates a run-time error.  
If that's the case with your version, then psrun using the "profil" style
of profiling will also fail.  You can check on this by compiling the example
program for profiling and seeing if it runs properly.  Check the "-help"
output of the compiler for the flags to use for compilation with gprof-style
profiling enabled.

Configuration options used when building PerfSuite are stored at the 
beginning of the file "config.log" after running the "configure" script, 
that can also help provide details for diagnosing.

If you can send that info along, we can see what might be happening.  You 
can send to per...@li... if you like, to minimize 
traffic here.

Thanks,
Rick

Re: [PerfSuite-users] OT: what contributes to 'poll_idle'

[Rick K. <rk...@nc...>]

Harry,

> Apologies for the somewhat OT post, but failing to be able to run perfsuite, 
> I've had to fall back for now on oprofile.  This may be of interest here as 
> well.  Advice and RTFM's welcome as long as they come with URLs.
> 

Not a problem, but I see that the best source of advice for this one (the 
OProfile people themselves), have responded with some info.  For those who 
are interested, the mail thread is available here:

http://sourceforge.net/mailarchive/forum.php?thread_id=9050286&forum_id=3252

As an aside, I think the OProfile manual is a good source of information, 
not only for that project but on the general topic of sampling profilers 
and some things to consider when working with one and interpreting the 
results (PerfSuite uses statistical sampling also).

Rick

[PerfSuite-users] forrtl: severe (168)

[<gg...@un...>]

Dear all,
I installed perfsuite successfully, but when I run the examples in matvec=
 folder,
after=20

% psrun ./matvec-r4
=20
I receive the following message

forrtl: severe (168): Program Exception - illegal instruction
Image              PC        Routine            Line        Source
matvec-r4          0804A335  Unknown               Unknown  Unknown
matvec-r4          0804A2C5  Unknown               Unknown  Unknown
libc.so.6          40243A07  Unknown               Unknown  Unknown
matvec-r4          0804A1F1  Unknown               Unknown  Unknown

I do not know what it happens. Is there somebody who can help me? Thanks =
in
advance, Giuseppe.

Dr. Giuseppe Grieco
DIFA Universit=E0 della Basilicata

[PerfSuite-users] OT: what contributes to 'poll_idle'

[Harry M. <hj...@ta...>]

Apologies for the somewhat OT post, but failing to be able to run perfsuite, 
I've had to fall back for now on oprofile.  This may be of interest here as 
well.  Advice and RTFM's welcome as long as they come with URLs.

In examining the oprofile output for several related programs, I was struck by 
how much time as being wasted in 'poll_idle'.  Because these apps typically
consumed very large (1GB) files, I originally thought that this was all due
to disk i/o, but now I'm not so sure.  

In trying to grasp how different programs behaved in terms of I/O and how
that related to 'poll_idle', I ran a few other apps.  Below are the
oprofile outputs for 3 of them, one is for a compilation of a fairly large
lib, 2 others are bioinformatics apps with which I'm familiar.

Especially notable is the oprofile output for clustalw.  It ingested and
egested relatively tiny files, ran for about 8 minutes, but oprofile
suggests that 50% of the time, the CPU was idle.  This cannot be disk I/O
(the system was essentially idle except for this test), so can memory
operations account for this much idle time (cache misses, moving data to &
from main mem, page faults, etc)?  And if so, how would I go about
quantifying this?  Is this the point where I have to dive into tau?

1) compilation of the hdf5 library (which I needed to do anyway). 
 CPU: AMD64 processors, speed 1990.96 MHz (estimated) 
 Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a unit
mask of 0x00 (No unit mask) count 100000 
 samples % app name symbol name 
 1495583 30.8198 cc1 (no symbols) 
 870380 17.9361 vmlinux poll_idle 
 486332 10.0219 libc-2.3.2.so (no symbols) 
 241578 4.9782 bash (no symbols) 
 197392 4.0677 vmlinux unmap_vmas 
 160895 3.3156 vmlinux copy_page_range 
 127658 2.6307 vmlinux copy_page 
 108131 2.2283 ld-2.3.2.so (no symbols) 
 101516 2.0920 as (no symbols) 
 99807 2.0567 vmlinux clear_page 
 49755 1.0253 libbfd-2.15.so (no symbols) 
 48223 0.9937 vmlinux page_fault 
  
  
 2) a run of clustalw, a multiple alignment app from bioinformatics which
takes a small amount of data (227KB) and performs a large amount of integer
computation on it, and then emits a relatively small amount of data (3
files totalling 180KB). The program is fairly chatty tho (lots of screen
output, redirected to a file in this case, but the total is included in the
180K output). 
  
 samples % app name symbol name 
 8950726 48.0420 vmlinux poll_idle 
 7498500 40.2473 clustalw forward_pass 
 832632 4.4691 clustalw prfscore 
 382780 2.0545 clustalw pdiff 
 247423 1.3280 clustalw reverse_pass 
 99006 0.5314 clustalw diff 
 77909 0.4182 vmlinux timer_interrupt 
 77141 0.4140 clustalw open_penalty2 
 73940 0.3969 clustalw calc_prf1 
 59589 0.3198 clustalw open_penalty1 
 49189 0.2640 clustalw ext_penalty1 
 47153 0.2531 clustalw ext_penalty2 
  
  
 3) a run of tacg, another bioinformatics sequence matching app that
consumes a larger amount of data (~5MB) and performs a fair amount of
integer arithmetic on it and then emits a file of ~117MB) 
  
 samples % app name symbol name 
 212135 53.1937 vmlinux poll_idle 
 111922 28.0649 libc-2.3.2.so (no symbols) 
 17894 4.4870 tacg hash 
 7601 1.9060 tacg Anti_Par 
 6468 1.6219 tacg Translate 
 3451 0.8654 tacg Cutting 
 2421 0.6071 tacg Degen_Cmp 
 2292 0.5747 libm-2.3.2.so (no symbols) 
 2034 0.5100 tacg HorribleAccounting 
 1913 0.4797 tacg LinearMap 
 1847 0.4631 tacg GetSequence2 
 1638 0.4107 vmlinux timer_interrupt 
  




-- 
Cheers, Harry
Harry J Mangalam - 949 856 2847 (vox; email for fax) - hj...@ta... 
            <<plain text preferred>>

Re: [PerfSuite-users] psrun fatal errors: hardware event not available

[Rick K. <rk...@nc...>]

> ..And while waiting for the 2.6 version of perfsuite, what opensource tool 
> would you suggest using to approximate the functionality of perfsuite?  ie 
> Name your top competitor ;)

That's a loaded question :)

I'm less concerned about talking about alternatives than I am about 
leaving someone out...

It seems to me that for profiling, oprofile is very widely used and 
maintained.  There is a project called TAU, based at the University of 
Oregon, that is also a very good alternative for portable profiling.  
Additionally, there has been an ongoing project at Rice for statistical 
profiling that is called HPC Toolkit (I believe it uses PAPI and/or 
oprofile for the actual sampling). For aggregate counting, you'll find a
utility called papiex, written by Phil Mucci, that does similar things to 
what psrun in counting mode does.

You can find pointers to some of these at the NCSA site:

http://perfsuite.ncsa.uiuc.edu/perftools/

SGI also announced an open source version of their OpenSpeedShop tool set, 
but I don't have any experience with it, so can't comment (the IRIX 
version worked very well).

I'd be curious to hear what other people think on this subject.

Thanks again,
Rick

Re: [PerfSuite-users] psrun fatal errors: hardware event not available

[Rick K. <rk...@nc...>]

Harry,

> Thanks for the zippy reply - much appreciated.

You're very welcome, although I'm sorry the reply didn't exactly offer a 
current fix!

> Just a suggestion - one of the tests that should run (maybe it does and I 
> missed it) is one that checks for the kernel version.  It seems odd that NONE 
> of the tests fail (I thought that originally the threads test failed, but in 
> the output below, they pass).  Seems like a pretty good test result for 
> something that won't run :)

That's a good idea and as you point out, no such check exists in the 
0.6.2a2 releases or earlier.  The upcoming version does have a check for 
kernel major and minor versions in the configure script, not only to 
possibly bail on error but also because there are a few things for which a 
different code path has to be taken based on kernel-specific things (like 
in /proc).

> I read the Linux Journal article which had no mention of the 2.6 kernel 
> prohibition and I must have missed the part where this was noted.  In the 
> INSTALL, it mentions:
> 
>     124       Currently, the latest 2.6.x release of perfctr is the best 
> choice of
>     125       versions for new installations of the driver.  Please read the
>     126       instructions in the perfctr distribution carefully.
> 
> which implies that the 2.6 kernel is supported.

Yes, this is kind of confusing.  The release numbers of perfctr, which is 
what is being referred to in the INSTALL file, are not in lockstep with 
the kernel versions.  That is, you can run perfctr 2.6.x on a 2.4.x 
kernel.  Also, you didn't miss anything in the LJ article... it just 
wasn't addressed :(

> I can't argue with free SW, and I'm trembling with antici.....pation :) for 
> the 2.6 supporting version, but it might have been made clearer.

I agree.  The web sites *and* the INSTALL instructions need a revisit.  
And, of course, a newer release for current kernels will solve the real 
issue :) 

> If there's currently a cvs/svn version of perfsuite, I'd be willing to give it 
> a try.

Thanks much for the offer.  There isn't a publicly-accessible CVS tree, 
but the next (hopefully final alpha) release is on its way.  What is going 
on right now are updates for the Perfmon (IA-64) driver (which *is* tied 
to the kernel major version).  Other generic 2.6 updates are in place but 
it makes sense to package it all together.

> Also, the difference between root and user was only due to permissions 
> on /dev/perfctr.  changing to a+r allowed a normal user to obtain the same 
> results.

Ah, that's good to know.  Thanks for clarifying.

> Do you have any idea if the new version of PAPI will allow perfsuite to run on 
> 2.6 (ie is the reason it won't run due to PAPI or due to perfsuite?)

It's not a PAPI-specific thing, it's in PerfSuite.  Specifically, 
necessary mods are in libperfsuite, not libpshwpc.  It is possible that 
psrun, in counting mode, might run OK on a 2.6 kernel with PAPI 3, but 
it's been a while since I checked.  I'm certain that profiling would be 
affected, though, and that would be the case whether profiling with 
hardware counters or with interval timers or profil().  Counting with 
Perfmon directly also would fail (but that is not applicable to Opteron at 
this point in time).

I've lopped off your test suite results (thanks for sending though).  The 
test suite is best at detecting fatal (core dump-inducing) type errors and 
has to be a little generic, because it has to be able to run whether or 
not the kernel has been patched and PAPI is installed.  But you can check 
whether you're getting anything meaningful at all through PAPI like this:

$ cd src/libpshwpc/tests
$ ./testsuite -v -k count

You should get "verbose" output from the test cases that try to count 
events, and if the results are reasonable (non-zero), you might be OK 
(again, for counting only... not profiling).

Thanks again for the feedback,
Rick

Re: [PerfSuite-users] psrun fatal errors: hardware event not available

[Harry M. <hj...@ta...>]

..And while waiting for the 2.6 version of perfsuite, what opensource tool 
would you suggest using to approximate the functionality of perfsuite?  ie 
Name your top competitor ;)

Thanks!

-- 
Cheers, Harry
Harry J Mangalam - 949 856 2847 (vox; email for fax) - hj...@ta... 
            <<plain text preferred>>

Re: [PerfSuite-users] psrun fatal errors: hardware event not available

[Harry M. <hj...@ta...>]

Hi Rick,

Thanks for the zippy reply - much appreciated.

Just a suggestion - one of the tests that should run (maybe it does and I 
missed it) is one that checks for the kernel version.  It seems odd that NONE 
of the tests fail (I thought that originally the threads test failed, but in 
the output below, they pass).  Seems like a pretty good test result for 
something that won't run :)

I read the Linux Journal article which had no mention of the 2.6 kernel 
prohibition and I must have missed the part where this was noted.  In the 
INSTALL, it mentions:

    124       Currently, the latest 2.6.x release of perfctr is the best 
choice of
    125       versions for new installations of the driver.  Please read the
    126       instructions in the perfctr distribution carefully.

which implies that the 2.6 kernel is supported.

I can't argue with free SW, and I'm trembling with antici.....pation :) for 
the 2.6 supporting version, but it might have been made clearer.

If there's currently a cvs/svn version of perfsuite, I'd be willing to give it 
a try.

Also, the difference between root and user was only due to permissions 
on /dev/perfctr.  changing to a+r allowed a normal user to obtain the same 
results.

Do you have any idea if the new version of PAPI will allow perfsuite to run on 
2.6 (ie is the reason it won't run due to PAPI or due to perfsuite?)

FYI,  below is the output of the perfsuite tests on my system.


hjm


19:42:49 hjm@clay ~/perfsuite-0.6.2a2
502 $ sudo make -s check
Password:
Making check in src
Making check in libperfsuite
Making check in .
Making check in tests
## ------------------------------------------- ##
## PerfSuite 0.6.2a2 test suite: Core Library. ##
## ------------------------------------------- ##
  1: Hello, world!                                ok
  2: CPU identification                           ok
  3: CPU speed information                        ok
  4: CPU speed information (Fortran)              ok
  5: CPU usage                                    ok
  6: error reporting                              ok
  7: error reporting (Fortran)                    ok
  8: argument processing                          ok
  9: proc filesystem                              ok
 10: memory usage reporting                       ok
 11: memory usage reporting (Fortran)             ok
 12: dynamic memory usage reporting               ok
 13: hash tables                                  ok
 14: load map address inquiry                     ok

## ------------- ##
## Test results. ##
## ------------- ##

All 14 tests were successful.
Making check in libpshwpc
Making check in .
Making check in arch-defaults
Making check in doc
Making check in xml
Making check in css
Making check in dtds
Making check in xsl
Making check in tests
## ----------------------------------------------------------- ##
## PerfSuite 0.6.2a2 test suite: Hardware Performance Library. ##
## ----------------------------------------------------------- ##
  1: base PAPI functionality                      ok
  2: PAPI compatibility                           ok
  3: PAPI to PerfSuite conversion (C/Fortran)     ok
  4: PAPI with POSIX threads (non-multiplexed)    ok
  5: PAPI with POSIX threads (multiplexed)        ok
  6: XML input document parsing (counting)        ok
  7: XML input document parsing (profiling)       ok
  8: inquiry (numevents)                          ok
  9: inquiry (numevents) error handling           ok
 10: inquiry (numevents = 0)                      ok
 11: serial program (counting)                    ok
 12: serial program (profiling)                   ok
 13: serial program (profil)                      ok
 14: serial program (itimer)                      ok
 15: serial program (counting auto)               ok
 16: serial program (profiling auto)              ok
 17: serial program (profil auto)                 ok
 18: serial program (itimer auto)                 ok
 19: serial program (counting Fortran)            ok
 20: serial program (profiling Fortran)           ok
 21: serial program (profil Fortran)              ok
 22: serial program (counting Fortran auto)       ok
 23: serial program (profiling Fortran auto)      ok
 24: serial program (profil Fortran auto)         ok
 25: serial program (itimer Fortran auto)         ok
 26: fork (counting)                              ok
 27: fork (profiling)                             ok
 28: POSIX threads program (counting)             ok
 29: POSIX threads program (profiling)            ok

## ------------- ##
## Test results. ##
## ------------- ##

All 29 tests were successful.
Making check in tcllib
Making check in psutils
Making check in pshwpc
Making check in metrics
Making check in xml
Making check in dtds
Making check in papi
Making check in msgs
Making check in perfmon
Making check in psxml
Generating top-level PerfSuite Tcl package index
Making check in tools
Making check in psenv
Making check in doc
Making check in psinv
Making check in .
Making check in doc
Making check in tests
## ------------------------------------ ##
## PerfSuite 0.6.2a2 test suite: psinv. ##
## ------------------------------------ ##
  1: psinv (basic)                                ok
  2: psinv (features)                             ok
  3: psinv (hardware performance events)          ok
  4: psinv (verbose)                              ok
  5: psinv (version)                              ok
  6: psinv (help)                                 ok

## ------------- ##
## Test results. ##
## ------------- ##

All 6 tests were successful.
Making check in psprocess
Making check in tcl
Making check in doc
Making check in msgs
Making check in tests
## ---------------------------------------- ##
## PerfSuite 0.6.2a2 test suite: psprocess. ##
## ---------------------------------------- ##
  1: psprocess (help)                             ok

## ------------- ##
## Test results. ##
## ------------- ##

1 test was successful.
Making check in PSConfig
Making check in doc
Making check in psrun
Making check in .
Making check in doc
Making check in xml
Making check in dtds
Making check in tests
## ------------------------------------ ##
## PerfSuite 0.6.2a2 test suite: psrun. ##
## ------------------------------------ ##
  1: psrun (basic)                                ok
  2: psrun (profil)                               ok
  3: psrun (itimer)                               ok
  4: psrun (count)                                ok
  5: psrun (count text)                           ok
  6: psrun (profile)                              ok
  7: psrun (resource)                             ok
  8: psrun (resource text)                        ok

## ------------- ##
## Test results. ##
## ------------- ##

All 8 tests were successful.
Making check in ext
Making check in tklib
Making check in plotchart
Making check in examples
Making check in cpi
Making check in hl
Making check in matvec
Making check in misc
Making check in psrun-raw
Making check in pstcl
Making check in threadrun

On Monday 21 November 2005 06:38 pm, Rick Kufrin wrote:
> Harry,
>
> > Like a recent poster here on the list, I cannot get psrun to work.
> >
> > I'm running Kubuntu on a dual opteron, have patched the kernel (2.6.11)
> > for both oprofile & papi and all PAPI tests pass.  All the perfsuite
> > tests pass as well (except the thread tests, which I'm not concerned
> > about just now).
>
> Thanks for forging ahead with trying to get things running.  It sounds
> like you've gotten all the prerequisites going just fine.  But I would
> guess that the reason why you are running into problems is that you are
> working with a 2.6 kernel.  To date, there hasn't been a PerfSuite release
> meant for use with anything but 2.4.
>
> That's the bad news, but the good news is that an updated release should
> be coming out shortly, and that is being developed against 2.6.  So I'd
> advise you to wait until available, and try again.
>
> The different behavior when run as a normal user and as root is a little
> worrisome, though, as there should be no need to have any special
> privileges to run psrun (unless you are working with things like SUID
> binaries).
>
> The "hardware event not available" error message means pretty much what it
> says - that the event listed in the configuration file is not available on
> that platform.  There have been a few updates to PAPI over the past
> several months and a new release out just today, for which events listed
> in the configuration files will be checked against the current PAPI
> support before the next PerfSuite release.
>
> Thanks much for the report,
> Rick

-- 
Cheers, Harry
Harry J Mangalam - 949 856 2847 (vox; email for fax) - hj...@ta... 
            <<plain text preferred>>

Re: [PerfSuite-users] psrun fatal errors: hardware event not available

[Rick K. <rk...@nc...>]

Harry,

> Like a recent poster here on the list, I cannot get psrun to work.  
> 
> I'm running Kubuntu on a dual opteron, have patched the kernel (2.6.11) for 
> both oprofile & papi and all PAPI tests pass.  All the perfsuite tests pass 
> as well (except the thread tests, which I'm not concerned about just now).

Thanks for forging ahead with trying to get things running.  It sounds 
like you've gotten all the prerequisites going just fine.  But I would 
guess that the reason why you are running into problems is that you are 
working with a 2.6 kernel.  To date, there hasn't been a PerfSuite release 
meant for use with anything but 2.4.

That's the bad news, but the good news is that an updated release should 
be coming out shortly, and that is being developed against 2.6.  So I'd 
advise you to wait until available, and try again.

The different behavior when run as a normal user and as root is a little 
worrisome, though, as there should be no need to have any special 
privileges to run psrun (unless you are working with things like SUID 
binaries).

The "hardware event not available" error message means pretty much what it 
says - that the event listed in the configuration file is not available on 
that platform.  There have been a few updates to PAPI over the past 
several months and a new release out just today, for which events listed 
in the configuration files will be checked against the current PAPI 
support before the next PerfSuite release.

Thanks much for the report,
Rick

[PerfSuite-users] psrun fatal errors: hardware event not available

[Harry M. <hj...@ta...>]

Like a recent poster here on the list, I cannot get psrun to work.  

I'm running Kubuntu on a dual opteron, have patched the kernel (2.6.11) for 
both oprofile & papi and all PAPI tests pass.  All the perfsuite tests pass 
as well (except the thread tests, which I'm not concerned about just now).

Both PAPI and perfsuite libs have been installed in the /usr/local/lib tree 
and ldconfig -v finds them appropriately

The problem comes when trying to run psrun on anything except itself.

As a user, I get:

$ psrun ls
psrun fatal error: error reported by performance software layer


And as root, I get:

 $ sudo psrun ls
psrun fatal error: hardware event not available

Any suggestions?


-- 
Cheers, Harry
Harry J Mangalam - 949 856 2847 (vox; email for fax) - hj...@ta... 
            <<plain text preferred>>

Re: [PerfSuite-users] request of information

[Rick K. <rk...@nc...>]

Giuseppe,

Sorry to hear you are experiencing problems but hopefully they are nothing 
too serious.  Here is some feedback:

> 1. I am using PerfSuite 0.6.1 on double processor Intel Pentium III, OS
> RedHat/Linux 7.1, kernel 2.4.20-18.9 whit patch perfctr 2.6.0. I tried to
> install PAPI but no one seems to be compatible with may OS. Do you have a
> solution for me?

This is a problem best directed to the PAPI mailing list, but I think you 
already know since I see that you sent a mail to the PAPI team as well.  I 
think you'll get good advice from that source.  For anyone on this list 
who is not aware of the PAPI lists, there is an archive of postings that 
can be very helpful in seeing about problems and questions regarding PAPI 
here:

http://www.ptools.org/archive/perfapi/maillist.html

> 2. I tried to use psrun because the INSTALL file says that it is possible to use
> it without PAPI. I used it in the example matvec. I changed the compiler in
> Makefile.matvec because I have just g77 and not ifc.
> In my prompt I digit the following line
> 
>    psrun matvec-r4
> 
> and the answer is 
> 
> matvec-r4: error while loading shared libraries: libpsrun_r.so: cannot open
> shared object file: No such file or directory
> 
> What is happening? Thank you in advance. Giuseppe Grieco.

This usually means that the runtime linker cannot find shared libraries 
that it is looking for, in this case a PerfSuite shared library called 
"libpsrun_r.so".  On Linux, the environment variable LD_LIBRARY_PATH is 
what's used to provide additional directories to search (see the man page 
for ld.so).

PerfSuite includes two convenience scripts that are meant to set up 
environment variables like LD_LIBRARY_PATH properly for your installation 
based on the configure-time "--prefix" option.  There is one script for 
Bourne (bash) like shells called psenv.sh, and another for C (tcsh) like 
shells called psenv.csh.  Choose the one according to the shell you use 
and execute "source FILENAME" to incorporate the settings.

Another point to make is that this assumes that you have actually 
installed PerfSuite after building it by executing "make install".  It's 
not easy to get things set up properly from within the build directories.

Finally, please keep in mind that without hardware counter support such as 
through PAPI, you will only get time-based (gprof-like) profiling from 
psrun and libpshwpc.

Hope this helps and good luck,
Rick

[PerfSuite-users] request of information

[<gg9...@un...>]

Hi, my name is Giuseppe

I have some problems with PerfSuite:

1. I am using PerfSuite 0.6.1 on double processor Intel Pentium III, OS
RedHat/Linux 7.1, kernel 2.4.20-18.9 whit patch perfctr 2.6.0. I tried to
install PAPI but no one seems to be compatible with may OS. Do you have a
solution for me?

2. I tried to use psrun because the INSTALL file says that it is possible to use
it without PAPI. I used it in the example matvec. I changed the compiler in
Makefile.matvec because I have just g77 and not ifc.
In my prompt I digit the following line

   psrun matvec-r4

and the answer is 

matvec-r4: error while loading shared libraries: libpsrun_r.so: cannot open
shared object file: No such file or directory

What is happening? Thank you in advance. Giuseppe Grieco.

Re: [PerfSuite-users] Using perfsuite without PAPI

[Rick K. <rk...@nc...>]

Frederique,

> I am trying to use Perfsuite to profile a multi-threaded program on a=20
> quadripro. I am particularly interested in knowing how much time a=20
> thread is spending on each processor. I believe that perfsuite can do=20
> that but I can't install PAPI (I am not root  on the quadripro and so I=
=20
> can't path the kernel with perfctr).

You're right in thinking that PerfSuite (psrun and libpshwpc) should be=20
able to do this, however you are also right in thinking that it requires=20
some other mechanism for sampling than the basic support that is availabl=
e=20
in the absence of access to the performance counters.  I am guessing that=
=20
you are on an x86 system since you mention perfctr.

PerfSuite (version 0.6.2) can access performance counters using either=20
PAPI or Perfmon, the latter being relevant to Itanium (IA-64) systems.
So if you're on an Itanium system, you don't need PAPI, you can get by=20
with using Perfmon directly.  On x86, though, you need the kernel patched=
=20
with perfctr and PAPI installed (I am guessing you know this, but just=20
clarifying).

Without any counter support, you can only do profiling and that only in=20
one of two ways: either by the profil() routine available in glibc (this=20
would be the default in such an installation, using a configuration file=20
that is called "profil.xml"), or through interval timers and interrupt
handlers that PerfSuite installs.  The itimer approach is more flexible
and can be selected using a sample XML configuration file called=20
"itimer.xml" that should be installed in PREFIX/share/perfsuite/xml/pshwp=
c

These two methods use signals to trigger the interrupt, though, and=20
signals are a process-wide mechanism, and so they are really not useful i=
n=20
the case of a multithreaded application (at least not in the present=20
versions of PerfSuite).  The counter-based techniques interrupt on counte=
r=20
overflow, and that can be kept isolated within a particular thread since=20
the underlying drivers support this.

I am guessing that this is the major problem you are running into, and=20
unfortunately I don't have a solution apart from having to patch your=20
kernel.

The other issue (not getting file/line information) I think might be=20
related to not compiling the application with symbols enabled (-g option).

Finally, about resource collection: that is done, when using the -r optio=
n=20
to psrun, for the main thread only, and does not track time spent on a=20
per-CPU basis.  So I am afraid that again, it is not what you might=20
have been hoping for.

Sorry for the bad news - please let me know if I have misunderstood the=20
question....

Rick

>=20
> I then installed perfsuite without PAPI. And I am getting some results=20
> running:
>=20
> $ psrun -r -p myprogram
>=20
> The resulting files are for example myprogram.*.19416.mymachine.xml and=
=20
> also myprogram.19416.mymachine.res.xml.
>=20
> When I watch the results with psprocess
>=20
> $ psprocess -e myprogram myprogram.*.19416.mymachine.xml
>=20
> I get this:
>=20
>=20
> File Summary
> -----------------------------------------------------------------------=
---------Samples=20
>    Self %  Total %  File
>=20
>     1519  100.00%  100.00%  ??
>=20
> Function Summary
> -----------------------------------------------------------------------=
---------Samples=20
>    Self %  Total %  Function
>=20
>     1519  100.00%  100.00%  ??
>=20
> Function:File:Line Summary
> -----------------------------------------------------------------------=
---------Samples=20
>    Self %  Total %  Function:File:Line
>=20
>     1519  100.00%  100.00%  ??:??:0
>=20
> So there is no detail.
>=20
> And when I watched the resource utilization, I get this:
>=20
> $ psprocess myprogram.19416.mymachine.res.xml
> CPU time (user, seconds)     : 3.69
> CPU time (system, seconds)   : 0.10
>=20
> but there is no distinction between CPUs and no distinction between thr=
eads.
>=20
> Is there any way for me to get more information using perfsuite without=
=20
> PAPI than what I already have now? Especially with the names of files,=20
> functions and the function lines?
>=20
> Is there a way to get the time spent by each thread on each processor?
>=20
> Thank you,
> Fr=E9d=E9rique.
>=20

[PerfSuite-users] Using perfsuite without PAPI

[Frederique Silber-C. <fre...@in...>]

Hello,

I am trying to use Perfsuite to profile a multi-threaded program on a=20
quadripro. I am particularly interested in knowing how much time a=20
thread is spending on each processor. I believe that perfsuite can do=20
that but I can't install PAPI (I am not root  on the quadripro and so I=20
can't path the kernel with perfctr).

I then installed perfsuite without PAPI. And I am getting some results=20
running:

$ psrun -r -p myprogram

The resulting files are for example myprogram.*.19416.mymachine.xml and=20
also myprogram.19416.mymachine.res.xml.

When I watch the results with psprocess

$ psprocess -e myprogram myprogram.*.19416.mymachine.xml

I get this:


File Summary
-------------------------------------------------------------------------=
-------Samples=20
   Self %  Total %  File

    1519  100.00%  100.00%  ??

Function Summary
-------------------------------------------------------------------------=
-------Samples=20
   Self %  Total %  Function

    1519  100.00%  100.00%  ??

Function:File:Line Summary
-------------------------------------------------------------------------=
-------Samples=20
   Self %  Total %  Function:File:Line

    1519  100.00%  100.00%  ??:??:0

So there is no detail.

And when I watched the resource utilization, I get this:

$ psprocess myprogram.19416.mymachine.res.xml
CPU time (user, seconds)     : 3.69
CPU time (system, seconds)   : 0.10

but there is no distinction between CPUs and no distinction between threa=
ds.

Is there any way for me to get more information using perfsuite without=20
PAPI than what I already have now? Especially with the names of files,=20
functions and the function lines?

Is there a way to get the time spent by each thread on each processor?

Thank you,
Fr=E9d=E9rique.
--=20
Fr=E9d=E9rique Silber-Chaussumier T=E9l: +33(0)160764565
D=E9partement INFormatique - =C9quipe Parall=E9lisme
INT - 9 rue C. Fourier 91011 Evry - France=09

[PerfSuite-users] LCI-2005 PerfSuite paper available

[Rick K. <rk...@nc...>]

The paper "PerfSuite: An Accessible, Open Source, Performance Analysis 
Environment for Linux", presented at the 6th International Conference on 
Linux Clusters (LCI-2005) in April 2005, is now available online. This 
overview paper describes the motivation, implementation, and design of 
PerfSuite's libraries and tools.

http://perfsuite.ncsa.uiuc.edu/publications/LCI-2005.pdf

This is the first formal technical paper about the software and gives
information about internals and organization that may be of interest
to anyone who wonders how it works.

Rick

[PerfSuite-users] Re: PerfSuite: psrun and 32bit code on x86_64

[Rick K. <rk...@nc...>]

Alex (and others),

Here's a followup on your question below regarding using PerfSuite with 
32-bit binaries on AMD64.  In general, it seems that it won't be too 
difficult to incorporate this in an upcoming release although the current 
versions won't build cleanly as-is and will have a few adjustments made 
to deal with things better.  As far as I can tell, there isn't a standard 
for filesystem names for 32- or 64-bit libraries, so it'll be the choice 
of the person building it ("--libdir=" configure-time option).

I'm working on RHEL4 (Nahant Update 1), Opteron 248, and currently have 
run into an issue that I'm hoping you or others on this list might be able
to offer a suggestion/workaround.  Specifically, linking a program compiled
for 32-bit fails in finding the Expat library (-lexpat, which is actually 
present in both of /usr/lib and /usr/lib64).  It seems to me that the 
linker search path isn't what it should be, even though -m32 is on the 
compile/link line.  This, in turn, is messing up a creation of the 
"libpshwpc" shared library which is where Expat is used in PerfSuite.

What I'm after are any suggestions that would successfully compile and 
link the following program as a 32-bit app on AMD64.  Once this gets 
solved, things should be on their way.  If there isn't a clean way to fix 
it, I think it'll still be possible to have 32-bit support but would 
require hand-editing automake-generated makefiles, which is naturally not 
too pleasant...

Thanks in advance for any suggestions!

Rick

Error output from the link is:

# gcc -m32 -o expattest expattest.c -lexpat
/usr/bin/ld: skipping incompatible /usr/lib64/libexpat.so when searching for -lexpat
/usr/bin/ld: skipping incompatible /usr/lib64/libexpat.a when searching for -lexpat
/usr/bin/ld: cannot find -lexpat
collect2: ld returned 1 exit status

Explicitly adding -L/usr/lib or setting LDEMULATION don't seem to do the 
trick for me.  The only thing I've found that works is to add 
/usr/lib/libexpat.so.0 on the link line instead of -lexpat :(

-----------------------------
#include <stdlib.h>
#include <stdio.h>
#include <expat.h>

main(int argc, char **argv)
{
    XML_Parser p;

    p = XML_ParserCreate(NULL);
    if ( ! p ) {
        fputs("Error in creating parser\n", stderr);
        exit(1);
    }
    exit(0);
}


On Mon, 18 Jul 2005, Alex Mericas wrote:

> I am running Suse on an Athlon64.  I can use psrun without problems on binaries compiled for 64bit, but if I try to run 32bit binaries psrun cannot find its runtime library.  I'm guessing psrun would need both a 32bit and a 64bit library to work this way.  Is this possible?  Do you have plans to implement?  Thanks!
>