oski-users — OSKI end-users post general questions, problems/bug reports here.

Re: [oski-users] CSR reordering

[Pavanakumar M. <m.p...@gm...>]

Dear Rich,

> Thanks for your question. Reordering is not implemented in the library
> that has been distributed, so if that's an optimization you need,
> you'll have to try it separately.

Thanks for the quick reply. I will actually benefit a lot from
ordering. I am attaching the matrix structure before and after
re-ordering using RCM. The TSP ordering can actually make the matrix
block structured (unlike RCM which reduces only the bandwidth), which
I think will be more beneficial. But implementing it is beyond my
scope. Is it possible to share the TSP re-ordering code which is not
distributed with OSKI ?

> (The TSP reordering cited there can
> be quite expensive; without a clear model of when it can be applied,
> we opted not to implement it.)

The time spent in tuning and ordering is justified in my case because
the matrix is a geometric one and is constructed just one time. The
solution evolves around this sparse geometric matrix.

> You may wish to try "pOSKI," which is the multicore version of OSKI.
> It is maintained as a separate library; see bebop.cs.berkeley.edu for
> more details.

Yes. I have both OSKI and pOSKI installed on our cluster. I am also
trying to find which one gives the best performance for my problem.

>
> -- Rich
>

> On Wed, Jul 11, 2012 at 4:20 AM, Pavanakumar Mohanamuraly
> <m.p...@gm...> wrote:
>> Hi
>>
>> I am trying to tune my matrix in stored in CSR format. It is obtained
>> from an irregular structure and is quite random.
>>
>> It is mentioned in the users guide to OSKI that "The library may
>> optimize kernel performance by permuting the rows and columns of the
>> matrix to reduce the bandwidth [27, 44, 13, 23, 14] or to create dense
>> block structure [36]." I found the "TSP ordering[36] implementation"
>> in the TODO list. Does this mean that when I give a CSR matrix OSKI
>> will not re-order it and make it a tuned BCSR mat ?
>>
>> Some pointer in this regards will be greatly appreciated.
>>
>> Regards,
>>
>> --
>> Pavanakumar Mohanamuraly
>>
>> ------------------------------------------------------------------------------
>> Live Security Virtual Conference
>> Exclusive live event will cover all the ways today's security and
>> threat landscape has changed and how IT managers can respond. Discussions
>> will include endpoint security, mobile security and the latest in malware
>> threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
>> _______________________________________________
>> oski-users mailing list
>> osk...@li...
>> https://lists.sourceforge.net/lists/listinfo/oski-users



-- 
Pavanakumar Mohanamuraly

Re: [oski-users] CSR reordering

[rich v. <res...@vu...>]

Pavanakumar,

Thanks for your question. Reordering is not implemented in the library
that has been distributed, so if that's an optimization you need,
you'll have to try it separately. (The TSP reordering cited there can
be quite expensive; without a clear model of when it can be applied,
we opted not to implement it.)

You may wish to try "pOSKI," which is the multicore version of OSKI.
It is maintained as a separate library; see bebop.cs.berkeley.edu for
more details.

-- Rich

On Wed, Jul 11, 2012 at 4:20 AM, Pavanakumar Mohanamuraly
<m.p...@gm...> wrote:
> Hi
>
> I am trying to tune my matrix in stored in CSR format. It is obtained
> from an irregular structure and is quite random.
>
> It is mentioned in the users guide to OSKI that "The library may
> optimize kernel performance by permuting the rows and columns of the
> matrix to reduce the bandwidth [27, 44, 13, 23, 14] or to create dense
> block structure [36]." I found the "TSP ordering[36] implementation"
> in the TODO list. Does this mean that when I give a CSR matrix OSKI
> will not re-order it and make it a tuned BCSR mat ?
>
> Some pointer in this regards will be greatly appreciated.
>
> Regards,
>
> --
> Pavanakumar Mohanamuraly
>
> ------------------------------------------------------------------------------
> Live Security Virtual Conference
> Exclusive live event will cover all the ways today's security and
> threat landscape has changed and how IT managers can respond. Discussions
> will include endpoint security, mobile security and the latest in malware
> threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
> _______________________________________________
> oski-users mailing list
> osk...@li...
> https://lists.sourceforge.net/lists/listinfo/oski-users

[oski-users] CSR reordering

[Pavanakumar M. <m.p...@gm...>]

Hi

I am trying to tune my matrix in stored in CSR format. It is obtained
from an irregular structure and is quite random.

It is mentioned in the users guide to OSKI that "The library may
optimize kernel performance by permuting the rows and columns of the
matrix to reduce the bandwidth [27, 44, 13, 23, 14] or to create dense
block structure [36]." I found the "TSP ordering[36] implementation"
in the TODO list. Does this mean that when I give a CSR matrix OSKI
will not re-order it and make it a tuned BCSR mat ?

Some pointer in this regards will be greatly appreciated.

Regards,

-- 
Pavanakumar Mohanamuraly

[oski-users] Failed to build OSKI-1.0.1h

[Warren A. <pt...@gm...>]

Hi,

I recently decided to build OSKI on a Core2 Duo running Gentoo Linux.
I downloaded the OSKI-1.0.1h tarball, unpacked it and configured it.
However,
I cannot get it to build, due to problems with the Fortran libraries.
I am using gcc version 4.4.3.  I configured OSKI as:

> cd build-1.0.1h/
>  ../oski-1.0.1h/configure --prefix=/usr/local/lib/ --with-blas='-L/usr/local/lib -Wl,-rpath,/usr/local/lib -lgoto2'

The configuration seems to work, and comes up with the Fortran linking flags
-L/usr/lib/gcc/i686-pc-linux-gnu/4.4.3
-L/usr/lib/gcc/i686-pc-linux-gnu/4.4.3/../../../../i686-pc-linux-gnu/lib
-L/usr/lib/gcc/i686-pc-linux-gnu/4.4.3/../../..
-lgfortranbegin -lgfortran -lm

However, when I type 'make' in the top level build directory, I
(eventually) get the following error message:

cc -O3 -std=c99 -o .libs/oskitimer .libs/oskitimerS.o -Wl,-rpath
-Wl,/usr/local/lib oskitimer-timer.o -Wl,--export-dynamic
-L/usr/local/lib -L/usr/lib/gcc/i686-pc-linux-gnu/4.4.3
-L/usr/lib/gcc/i686-pc-linux-gnu/4.4.3/../../../../i686-pc-linux-gnu/lib
-L/usr/lib/gcc/i686-pc-linux-gnu/4.4.3/../../..
/usr/lib/gcc/i686-pc-linux-gnu/4.4.3/libgfortranbegin.a
../src/.libs/liboski.so -lgoto2
/usr/lib/gcc/i686-pc-linux-gnu/4.4.3/libgfortran.so -lm -ldl
-Wl,--rpath -Wl,/usr/local/lib/lib/oski
../src/.libs/liboski.so: undefined reference to `MAIN__'
collect2: ld returned 1 exit status
make[1]: *** [oskitimer] Error 1
make[1]: Leaving directory `/home/woza/oski/build-1.0.1h/bench'
make: *** [all-recursive] Error 1


I have tried passing --disable-fortran to the configure command, but
that causes configuration to fail because it cannot link Fortran
programs.

Is anyone able to help?

Cheers,
Warren Armstrong

Re: [oski-users] OSKI Mac build

[Heroux, M. A <ma...@sa...>]

Thanks Rich.  The build is chugging along now...

Mike


On 7/30/09 2:31 PM, "Rich Vuduc" <ri...@cc...> wrote:

Mike,

This problem is related to OSKI configure's attempt to link
against a native BLAS, which picks up the default Mac OS X
BLAS in /usr/lib. If you only need SpMV, the quickest hack
is, at configure time, to skip the BLAS:

   configure ... --with-blas=no ...

Thanks for reporting this problem, which I'll add to the bug
fix list.

-- Rich

Heroux, Michael A wrote:
> Hi Rich,
>
> Sorry to bother you with a simple question, but I am hoping you have a
> quick pointer. I am trying to build OSKI 1.0.1h on my Leopard Mac and
> get an error part way through the build. It is identical to the error
> found in the discussion on this website:
>
> http://lists.apple.com/archives/Fortran-dev/2008/Mar/msg00017.html
>
> I am wondering if there is a quick fix. I tried using a newer version of
> autoconf/automake, but there are many error message generated and the
> resulting configure file does not work. I am using GCC 4.4.0.
>
> Thanks.
>
> Mike

Re: [oski-users] OSKI Mac build

[Richard (R. V. <ri...@cc...>]

Mike,

This problem is related to OSKI configure's attempt to link 
against a native BLAS, which picks up the default Mac OS X 
BLAS in /usr/lib. If you only need SpMV, the quickest hack 
is, at configure time, to skip the BLAS:

   configure ... --with-blas=no ...

Thanks for reporting this problem, which I'll add to the bug 
fix list.

-- Rich

Heroux, Michael A wrote:
> Hi Rich,
>
> Sorry to bother you with a simple question, but I am hoping you have a
> quick pointer. I am trying to build OSKI 1.0.1h on my Leopard Mac and
> get an error part way through the build. It is identical to the error
> found in the discussion on this website:
>
> http://lists.apple.com/archives/Fortran-dev/2008/Mar/msg00017.html
>
> I am wondering if there is a quick fix. I tried using a newer version of
> autoconf/automake, but there are many error message generated and the
> resulting configure file does not work. I am using GCC 4.4.0.
>
> Thanks.
>
> Mike

Re: [oski-users] make check failed

[Kai W. <was...@ma...>]

I'm running OpenSuSE 11.0 with gcc 4.3.1

I also ran 'make check' one a system with OpenSuSE 11.1 and gcc 4.3.2, 
which resulted in the same error messages.

Kai

On Sat, 25 Jul 2009, Richard Vuduc wrote:

> Kai,
>
> What platform are you running on (OS, compiler)?
>
> --r
>

Re: [oski-users] make check failed

[Richard V. <ri...@cs...>]

Kai,

What platform are you running on (OS, compiler)?

--r

On Jul 24, 2009, at 4:45 AM, Kai Wassmuss wrote:

> Hi,
> in my installation of oski, 'make check' gives the following errors:
>
> ------------------------------------------------------------------
>
> PASS: test133.sh
> *** Checking ./test142-hbmatmult using a custom script ***
> *** Error condition 'err != 0' detected ***
>
> ***********************************************************
> * OSKI Error -17 : An error occurred
> * Occurred at/near '../../oski-1.0.1h/tests/hbmatmult.c', line 70.
>
> Additional information:
>   In call to 'check_MatMult()': Computed an incorrect value or values.
> **********************************************************
> FAIL: test142.sh
> *** Checking ./test151-hbata using a custom script ***
> *** Error condition 'err != 0' detected ***
>
> ***********************************************************
> * OSKI Error -17 : An error occurred
> * Occurred at/near '../../oski-1.0.1h/tests/hbata.c', line 74.
>
> Additional information:
>   In call to 'check_MatTransMatMult()': Computed an incorrect value  
> or values.
> **********************************************************
> FAIL: test151.sh
> *** Checking ./test160-hba_and_at using a custom script ***
> *** Error condition 'err != 0' detected ***
>
> ***********************************************************
> * OSKI Error -17 : An error occurred
> * Occurred at/near '../../oski-1.0.1h/tests/hba_and_at.c', line 69.
>
> Additional information:
>   In call to 'check_MatMultAndMatTransMult()': Computed an incorrect  
> value or
> va
> lues.
> **********************************************************
> FAIL: test160.sh
> *** Checking ./test169-trisolve_xform using a custom script ***
> PASS: test169.sh
> ==================================
> 3 of 25 tests failed
> Please report to osk...@sf...
> ==================================
> make[2]: *** [check-TESTS] Fehler 1
> make[2]: Leaving directory `/tmp/build-1.0.1h/tests'
> make[1]: *** [check-am] Fehler 2
> make[1]: Leaving directory `/tmp/build-1.0.1h/tests'
> make: *** [check-recursive] Fehler 1
>
> ---------------------------------------------------------------------
>
> the configuration was:
>
> /tmp/build-1.0.1h # ../oski-1.0.1h/configure CFLAGS='-fPIC'
>
> additionally, the example call to
>
> oski_CreateMatCSR( Aptr, Aind, Aval, 3, 3, SHARE_INPUTMAT, 3,  
> INDEX_ZERO_BASED,
> MAT_TRI_LOWER, MAT_UNIT_DIAG_IMPLICIT );
>
> which was given in the manual fails with the error
>
> ***********************************************************
> * OSKI Error -9 : Feature not yet implemented
> * Occurred at/near '../../oski-1.0.1h/src/matcreate.c', line 290.
>
> Additional information:
>   Can't find CSR wrapper, liboski_Tid_LTX_oski_WrapCSR()
> **********************************************************
>
> when called from within my program while it works with a short test  
> program
> that does nothing else.
> While this could very well be a problem within my own program, I  
> wonder if this
> error might be related to the errors from 'make check'.
>
> Thank you in advance for your response, any hints are welcome
> Kai Wassmuss
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> oski-users mailing list
> osk...@li...
> https://lists.sourceforge.net/lists/listinfo/oski-users

[oski-users] make check failed

[Kai W. <was...@ma...>]

Hi,
in my installation of oski, 'make check' gives the following errors:

------------------------------------------------------------------

PASS: test133.sh
*** Checking ./test142-hbmatmult using a custom script ***
*** Error condition 'err != 0' detected ***

***********************************************************
* OSKI Error -17 : An error occurred
* Occurred at/near '../../oski-1.0.1h/tests/hbmatmult.c', line 70.

Additional information:
   In call to 'check_MatMult()': Computed an incorrect value or values.
**********************************************************
FAIL: test142.sh
*** Checking ./test151-hbata using a custom script ***
*** Error condition 'err != 0' detected ***

***********************************************************
* OSKI Error -17 : An error occurred
* Occurred at/near '../../oski-1.0.1h/tests/hbata.c', line 74.

Additional information:
   In call to 'check_MatTransMatMult()': Computed an incorrect value or values.
**********************************************************
FAIL: test151.sh
*** Checking ./test160-hba_and_at using a custom script ***
*** Error condition 'err != 0' detected ***

***********************************************************
* OSKI Error -17 : An error occurred
* Occurred at/near '../../oski-1.0.1h/tests/hba_and_at.c', line 69.

Additional information:
   In call to 'check_MatMultAndMatTransMult()': Computed an incorrect value or 
va
lues.
**********************************************************
FAIL: test160.sh
*** Checking ./test169-trisolve_xform using a custom script ***
PASS: test169.sh
==================================
3 of 25 tests failed
Please report to osk...@sf...
==================================
make[2]: *** [check-TESTS] Fehler 1
make[2]: Leaving directory `/tmp/build-1.0.1h/tests'
make[1]: *** [check-am] Fehler 2
make[1]: Leaving directory `/tmp/build-1.0.1h/tests'
make: *** [check-recursive] Fehler 1

---------------------------------------------------------------------

the configuration was:

/tmp/build-1.0.1h # ../oski-1.0.1h/configure CFLAGS='-fPIC'

additionally, the example call to

oski_CreateMatCSR( Aptr, Aind, Aval, 3, 3, SHARE_INPUTMAT, 3, INDEX_ZERO_BASED, 
MAT_TRI_LOWER, MAT_UNIT_DIAG_IMPLICIT );

which was given in the manual fails with the error

***********************************************************
* OSKI Error -9 : Feature not yet implemented
* Occurred at/near '../../oski-1.0.1h/src/matcreate.c', line 290.

Additional information:
   Can't find CSR wrapper, liboski_Tid_LTX_oski_WrapCSR()
**********************************************************

when called from within my program while it works with a short test program 
that does nothing else.
While this could very well be a problem within my own program, I wonder if this 
error might be related to the errors from 'make check'.

Thank you in advance for your response, any hints are welcome
Kai Wassmuss

[oski-users] Tech Report about Epetra/OSKI interface

[Ian K. <Ian...@Co...>]

Hi all,

Jonathan Hu and my tech report from our work developing and testing an 
interface to include OSKI within the Trilinos package Epetra is 
available at: http://www.cs.sandia.gov/CSRI/Proceedings/index.htm.  The 
report is included in the 2008 CSRI proceedings.

Thanks,

Ian

[oski-users] Compile Error on timer.c

[Keita T. <ke...@cr...>]

Hi,

 

I am having the following error at timer.c.  I'd like to have an advice for a workaround.

 

Thank you,

 

 

nid00004 src/timer> make

source='timer.c' object='liboski_tempo_la-timer.lo' libtool=yes \

DEPDIR=.deps depmode=none /bin/sh ../../config/depcomp \

/bin/sh ../../libtool --tag=CC --mode=compile cc -DHAVE_CONFIG_H -DWITH_SLOW_TIMER -I. -I. -I../../config -DOSKI_BUILD_TIME -I../.. -I../../include -I../.. -I../../include   -DMOD_NAME=liboski_tempo -O3 -fastsse -fast -c99 -c -o liboski_tempo_la-timer.lo `test -f 'timer.c' || echo './'`timer.c

 cc -DHAVE_CONFIG_H -DWITH_SLOW_TIMER -I. -I. -I../../config -DOSKI_BUILD_TIME -I../.. -I../../include -I../.. -I../../include -DMOD_NAME=liboski_tempo -O3 -fastsse -fast -c99 -c timer.c

/opt/xt-asyncpe/1.1/bin/cc: INFO: linux target is being used

PGC-S-0040-Illegal use of symbol, ticks (timer.c: 141)

PGC-W-0156-Type not specified, 'int' assumed (timer.c: 141)

PGC-S-0040-Illegal use of symbol, ticks (timer.c: 142)

PGC-W-0156-Type not specified, 'int' assumed (timer.c: 142)

PGC-S-0141-Wrong number of parameters to function (timer.c: 177)

PGC-S-0040-Illegal use of symbol, ticks (timer.c: 235)

PGC-W-0156-Type not specified, 'int' assumed (timer.c: 235)

PGC-S-0094-Illegal type conversion required (timer.c: 238)

PGC-S-0094-Illegal type conversion required (timer.c: 240)

PGC-S-0094-Illegal type conversion required (timer.c: 245)

PGC/x86-64 Linux 7.2-3: compilation completed with severe errors

 

================================
 Keita Teranishi
 Math Software Group
 Cray, Inc.
 ke...@cr...
================================

[oski-users] Epetra/Oski interface code and paper.

[Ian K. <Ian...@Co...>]

Hi all,

Over the summer at Sandia I have developed an interface to OSKI within Epetra, which is a package in Trilinos.  The interface should be released as part of Trilinos 9.0.  Also I have written along with Jonathan Hu a tech report summarizing our work along with some performance tests run.  This will be published as part of the CSRI summer proceedings and should be available sometime this fall at: http://www.cs.sandia.gov/CSRI/Proceedings/index.htm.  I will send another e-mail to this list when it becomes available, but if you have any questions about this work or Trilinos feel free to contact me or Jonathan at jh...@sa....

Thanks,

Ian

[oski-users] Oski Compile Error

[Karlin, I. <ik...@sa...>]

Hi,

When I run a default configure then make on a Clovertown system with gcc 4.1.1 and redhat 4.1.1 I think running 2.6.18-1.2798.fc6 kernel and get the following errors on the following command:

if /bin/sh ../../libtool --tag=CC --mode=compile cc -DHAVE_CONFIG_H -I. -I../../../oski-1.0.1h/lua/src -I../../config -DOSKI_BUILD_TIME -I../../../oski-1.0.1h -I../../../oski-1.0.1h/include -I../.. -I../../include   -I../../../oski-1.0.1h/lua/include -DLUA_PATH_DEFAULT="\"/home/ikarlin/Oski/install/lib/oski/?.lua;/home/ikarlin/Oski/install/lib/oski/?\"" -O3 -malign-double -march=pentium4 -msse2 -std=c99 -MT liblua_la-lapi.lo -MD -MP -MF ".deps/liblua_la-lapi.Tpo" -c -o liblua_la-lapi.lo `test -f 'lapi.c' || echo '../../../oski-1.0.1h/lua/src/'`lapi.c; \
        then mv -f ".deps/liblua_la-lapi.Tpo" ".deps/liblua_la-lapi.Plo"; else rm -f ".deps/liblua_la-lapi.Tpo"; exit 1; fi
 cc -DHAVE_CONFIG_H -I. -I../../../oski-1.0.1h/lua/src -I../../config -DOSKI_BUILD_TIME -I../../../oski-1.0.1h -I../../../oski-1.0.1h/include -I../.. -I../../include -I../../../oski-1.0.1h/lua/include "-DLUA_PATH_DEFAULT=\"/home/ikarlin/Oski/install/lib/oski/?.lua;/home/ikarlin/Oski/install/lib/oski/?\"" -O3 -malign-double -march=pentium4 -msse2 -std=c99 -MT liblua_la-lapi.lo -MD -MP -MF .deps/liblua_la-lapi.Tpo -c ../../../oski-1.0.1h/lua/src/lapi.c  -fPIC -DPIC -o .libs/liblua_la-lapi.o
../../../oski-1.0.1h/lua/src/lapi.c:1: error: CPU you selected does not support x86-64 instruction set
../../../oski-1.0.1h/lua/src/lapi.c:1: error: CPU you selected does not support x86-64 instruction set
../../../oski-1.0.1h/lua/src/lapi.c:1: error: -malign-double makes no sense in the 64bit mode

Any insite as to what I can do to fix this.  I can provide more information too if that would help.

Thanks,

Ian

Re: [oski-users] Another small error in the User's guide

[Richard V. <ri...@cs...>]

Ian, thanks again. I've made the changes in CVS.

It's nice to know at least one person is out there reading the  
documentation! :-)

--rich

On Jun 17, 2008, at 5:35 PM, Karlin, Ian wrote:

> Hi Rich,
>
> The same error occurs as well in the doc oski_SetHintMatTransMatMult  
> and oski_SetHintMatPowMult.
>
> Ian
> ________________________________________
> From: Richard Vuduc [ri...@cs...]
> Sent: Tuesday, June 17, 2008 10:26 AM
> To: Karlin, Ian
> Cc: osk...@li...
> Subject: Re: [oski-users] Another small error in the User's guide
>
> Ian,
>
> Thanks for catching this as well. I've made the change in CVS.
>
> --rich
>
> On Jun 17, 2008, at 1:11 PM, Karlin, Ian wrote:
>
>> Hi Rich,
>>
>> In the documentation for oski_MatTransMatMult the argument t_view is
>> listed as optional and able to take in INVALID_MAT as a parameter.
>> When I do not include this argument I get a compile error and the
>> INVALID_MAT should be INVALID_VEC as far as I can tell.
>>
>> Ian
>
>
>

Re: [oski-users] Another small error in the User's guide

[Richard V. <ri...@cs...>]

Ian,

Thanks for catching this as well. I've made the change in CVS.

--rich

On Jun 17, 2008, at 1:11 PM, Karlin, Ian wrote:

> Hi Rich,
>
> In the documentation for oski_MatTransMatMult the argument t_view is  
> listed as optional and able to take in INVALID_MAT as a parameter.   
> When I do not include this argument I get a compile error and the  
> INVALID_MAT should be INVALID_VEC as far as I can tell.
>
> Ian

[oski-users] Another small error in the User's guide

[Karlin, I. <ik...@sa...>]

Hi Rich,

In the documentation for oski_MatTransMatMult the argument t_view is listed as optional and able to take in INVALID_MAT as a parameter.  When I do not include this argument I get a compile error and the INVALID_MAT should be INVALID_VEC as far as I can tell.

Ian

Re: [oski-users] How to implement multithreaded applications using OSKI.

[Richard V. <ri...@cs...>]

Parneeth,

The official OSKI distribution does not currently use multiple  
threads. However, we are this summer working on a thin layer that  
would sit on top of OSKI that is multithreaded and NUMA-aware, based  
on the work by Williams, et al., that appeared in Supercomputing 2007  
[see: http://www.cs.berkeley.edu/~samw/research/papers/sc07.pdf ]

--Rich

On Jun 17, 2008, at 12:16 AM, pra...@cr... wrote:

>
> Dear Sir,
>
> We have implemented Sparse-Matrix X Vector multiplication using OSKI  
> routines.
> We want to run the same application using different number of  
> threads. We have searched for a way to do it in the OSKI user guide,  
> but we couldn't figure out how to do it.
> Can you please send us the information about how to use OSKI for  
> multithreaded applications? Also, it would be of great help if you  
> can send us some sample examples for the same.
>
>
> Thanks & Regards,
> Praneeth Ravikanti
> ------------------------------------------------
> (Member of Technical Staff)
> Computational Research Laboratories,
> Pune,  
> India 
> .-------------------------------------------------------------------------

[oski-users] How to implement multithreaded applications using OSKI.

[<pra...@cr...>]

Dear Sir,

We have implemented Sparse-Matrix X Vector multiplication using OSKI 
routines.
We want to run the same application using different number of threads. We 
have searched for a way to do it in the OSKI user guide, but we couldn't 
figure out how to do it. 
Can you please send us the information about how to use OSKI for 
multithreaded applications? Also, it would be of great help if you can 
send us some sample examples for the same.


Thanks & Regards,
Praneeth Ravikanti
------------------------------------------------
(Member of Technical Staff)
Computational Research Laboratories,
Pune, India.

Re: [oski-users] Is it a mispell?

[Richard V. <ri...@cs...>]

Xianyi,

You are right; thanks for catching this error.

--Rich

On Jun 16, 2008, at 7:46 AM, Xianyi Zhang wrote:

> Hi,guys
>
> Please check out src/MBCSR/ata.c file 35 rows. I think it should be
> "SubmatReprTransSubmatReprMult" :-)
>
>
> Thanks
>
> Xianyi Zhang

[oski-users] Is it a mispell?

[Xianyi Z. <zx...@ma...>]

Hi,guys

Please check out src/MBCSR/ata.c file 35 rows. I think it should be
"SubmatReprTransSubmatReprMult" :-)


Thanks

Xianyi Zhang

[oski-users] Compiling OSKI with GOTO on MacOSX

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

Hi there,

I am trying to install the latest OSKI (oski-1.0.1h) to a G5 system
with Mac OS X Tiger. Here's my "configure" line:

./configure --disable-static --prefix=/Users/mehmetb/Libs/OSKI
--with-blas='-L/Users/mehmetb/Libs/32_GotoBLAS -lgoto'

When compilation reaches to the point:

Extracting /sw/lib/libg2c.a
... Lots of object files ...
...
...

I have these errors:

ld: Undefined symbols:
_G77_date_y2kbuggy_0
_G77_vxtidate_y2kbuggy_0
/usr/libexec/gcc/powerpc-apple-darwin8/4.0.1/libtool: internal link
edit command failed
make[1]: *** [liboski.la] Error 1
make: *** [all-recursive] Error 1

Any ideas why I am getting these?

As a completely separate question, should I give vecLib a try as well?
I chose GOTO because it performs better in many cases (in my
experience) and vecLib may also cause numerical inaccuracies (can you
believe??)...

Thank you very much in advance!
-Memo



PS: The configuration output is as follows:

------------------------------------------------------------------------
Configuration

  Source code location:

  Host (CPU architecture):
    powerpc-apple-darwin8.11.0 (G5)
  Compiler:
    cc
  Compiler flags:
    -O3 -mcpu=G5 -std=c99
  Linker flags:
     -lm
  Fortran linker flags:
     -L/sw/lib/gcc/powerpc-apple-darwin8.4.0/3.4.3
-L/sw/lib/gcc/powerpc-apple-darwin8.4.0/3.4.3/../../.. -lfrtbegin
-lg2c -lSystemStubs -lmx
/usr/lib/gcc/powerpc-apple-darwin8/4.0.1/libgcc.a
  Install path:
    /Users/mehmetb/Libs/OSKI

Re: [oski-users] oski_MatMult with an untuned matrix

[Rich V. <ri...@cc...>]

Hi Jim,

Yes, that=B9s correct.

--rich

On 10/24/07 11:42 PM, "James Willenbring" <jm...@sa...> wrote:

> Hello,
> =20
> If I use oski_CreateMatCSR to create an oski matrix using SHARE_INPUTMAT =
and
> perform an oski_MatMult without tuning, will the matrix-vector multiplica=
tion
> be performed in place?  I think that is what the documentation is saying.=
  It
> sounds like the matrix is not replicated until some sort of tuning takes
> place.  Am I correct?
> =20
> Thanks!
> =20
> Jim Willenbring
>=20
>=20
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc.
> Still grepping through log files to find problems?  Stop.
> Now Search log events and configuration files using AJAX and a browser.
> Download your FREE copy of Splunk now >> http://get.splunk.com/
>=20
> _______________________________________________
> oski-users mailing list
> osk...@li...
> https://lists.sourceforge.net/lists/listinfo/oski-users

[oski-users] oski_MatMult with an untuned matrix

[James W. <jm...@sa...>]

Hello,

 

If I use oski_CreateMatCSR to create an oski matrix using SHARE_INPUTMAT and
perform an oski_MatMult without tuning, will the matrix-vector
multiplication be performed in place?  I think that is what the
documentation is saying.  It sounds like the matrix is not replicated until
some sort of tuning takes place.  Am I correct?

 

Thanks!

 

Jim Willenbring

Re: [oski-users] Implicit tuning

[Richard V. <ri...@cs...>]

Eduardo,

This listing is just an example of what a solver writer 
might write, and so the loops are just dummy loops. The main 
idea is that you can call oski_TuneMat() repeatedly whenever 
you like.

--rich

Eduardo Correa wrote:
>  I am investigating the oski library for my master thesis...and in the 
> user guide i didnt understand what means
> "while( ! converged ) " , example listing4. Which converge test do you 
> use in implicit tuning.?
> Best,
> Correa Eduardo

[oski-users] Implicit tuning

[Eduardo C. <edu...@gm...>]

 I am investigating the oski library for my master thesis...and in the user
guide i didnt understand what means
"while( ! converged ) " , example listing4. Which converge test do you use
in implicit tuning.?
Best,
Correa Eduardo