librsb-users — The mailing list of librsb users

[Librsb-users] librsb-1.2.0-rc2.tar.gz available

[Michele M. <mic...@us...>]

Dear librsb users,

Release 1.2.0-rc2 of librsb is available from
 http://sourceforge.net/projects/librsb/

This update fixes cache size recognition code when using
librsb in the Cygwin environment, e.g. together with
the GNU Octave librsb plugin "sparsersb"
 http://octave.sourceforge.net/sparsersb/index.html

Thanks go to Marco Atzeri for helping out with this !

Michele

Re: [Librsb-users] librsb-1.2.0-rc1.tar.gz followup ?

[Michele M. <mic...@ip...>]

On 20151019@19:00, marco atzeri wrote:
> Hi Michele,
> any plan, short term,  for a rc2 or more ?
> 
> Regards
> Marco

Hi Marco,

indeed, it's the moment for it.
So very soon !

[Librsb-users] librsb-1.2.0-rc1.tar.gz followup ?

[marco a. <mar...@gm...>]

Hi Michele,
any plan, short term,  for a rc2 or more ?

Regards
Marco

Re: [Librsb-users] confusing config error on cygwin

[Michele M. <mic...@us...>]

On 20150622@16:36, Marco Atzeri wrote:
> On 6/22/2015 3:38 PM, Michele Martone wrote:
> >On 20150622@15:26, Marco Atzeri wrote:
> >>On 6/22/2015 1:57 PM, Michele Martone wrote:
> >>>On 20150622@13:34, Marco Atzeri wrote:
> >>
> >>>
> >>>If you have a suggestion about how to replicate the problem or how to
> >>>fix it, I'll use it...
> >>
> >>I will look on it.
> >Thanks.
> >
> 
> the build problem is due to the lack of usage of $(EXEEXT) in the
> Makefile.am for the binary programs.
> 
> See patch as example, I can still miss some places in it.
Ok: --- I'll apply $(EXEEXT) where needed --- thanks !

> I was forced to disable the "scripts/readme-tests.sh"
> as the rule was failing and creating a crap script.
> We need to find a better way for it.
If the generated script contains garbage is probably
because one of the two:
 - the rule contains a non-standard expression
 - the tools involved (echo,grep,sed) behave out of standard

> Unfortunately the real test fails immediately :
> ...
> ERROR 0xffffe000 : Memory hierarchy info failed to be detected. You
> can bypass this by setting a meaningful
> RSB_USER_SET_MEM_HIERARCHY_INFO environment variable.
> ./scripts/readme-tests.sh: line 2:  8500 Segmentation fault
> (core dumped) ./rsbench -oa -Ob -f ${srcdir}/A.mtx -qH -R -n1 -t100
> --verbose
> Makefile:3052: recipe for target 'mtests' failed
> 
> -----------------------------------------------------
> 
> Suggestions ?
> Marco

The following:
 export RSB_USER_SET_MEM_HIERARCHY_INFO="L2:4/64/512K,L1:8/64/32K"
shall trick "ERROR 0xffffe000" not to occur.
If this does not work I'll have to look more carefully into it.
Please let me know.

Re: [Librsb-users] confusing config error on cygwin

[Marco A. <mar...@gm...>]

On 6/22/2015 3:38 PM, Michele Martone wrote:
> On 20150622@15:26, Marco Atzeri wrote:
>> On 6/22/2015 1:57 PM, Michele Martone wrote:
>>> On 20150622@13:34, Marco Atzeri wrote:
>>
>>>
>>> If you have a suggestion about how to replicate the problem or how to
>>> fix it, I'll use it...
>>
>> I will look on it.
> Thanks.
>

the build problem is due to the lack of usage of $(EXEEXT) in the 
Makefile.am for the binary programs.

See patch as example, I can still miss some places in it.

I was forced to disable the "scripts/readme-tests.sh"
as the rule was failing and creating a crap script.
We need to find a better way for it.

Unfortunately the real test fails immediately :
-----------------------------------------------------
  [*] beginning quick test...
make mtests -C .
make[2]: Entering directory 
'/cygdrive/e/cyg_pub/devel/librsb/build32/librsb-1.2.0-rc1'
srcdir=. /bin/sh ./scripts/readme-tests.sh
# beginning run at 1434983205
# 
/cygdrive/e/cyg_pub/devel/librsb/build32/librsb-1.2.0-rc1/./.libs/rsbench -oa 
-Ob -f ./A.mtx -qH -R -n1 -t100 --verbose
# compiled with: CC=gcc CFLAGS=-g -O2 -fopenmp -std=c99
# average timer granularity: 1.11e-07 s
# will NOT perform ancillary tests.
# will flush cache memory:  between each operation measurement series, 
and NOT between each operation.
# will keep any zero encountered in the matrix.
# user specified a verbosity level of 1 (each --verbose occurrence 
counts +1)
Checking environment RSB_USER_SET_MEM_HIERARCHY_INFO variable.
Checking hardcoded RSB_USER_SET_MEM_HIERARCHY_INFO symbol
Checking hardcoded RSB_DETECTED_MEM_HIERARCHY_INFO symbol
Available memory hierarchy info string: ""
ERROR 0xffffe000 : Memory hierarchy info failed to be detected. You can 
bypass this by setting a meaningful RSB_USER_SET_MEM_HIERARCHY_INFO 
environment variable.
./scripts/readme-tests.sh: line 2:  8500 Segmentation fault      (core 
dumped) ./rsbench -oa -Ob -f ${srcdir}/A.mtx -qH -R -n1 -t100 --verbose
Makefile:3052: recipe for target 'mtests' failed

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

Suggestions ?
Marco

Re: [Librsb-users] confusing config error on cygwin

[Michele M. <mic...@us...>]

On 20150622@15:26, Marco Atzeri wrote:
> On 6/22/2015 1:57 PM, Michele Martone wrote:
> >On 20150622@13:34, Marco Atzeri wrote:
> 
> >I've just made an out-of-tree build on Linux and it does not have such
> >problems.
> >I don't have cygwin so I don't know how to replicate the problem you
> >encounter.
> >Now, sbtc is not part of the library but a test program which according
> >to the config.log files you've sent me shall be even empty.
> >Indeed, you shall have librsb.a built I suppose.
> >
> >If you have a suggestion about how to replicate the problem or how to
> >fix it, I'll use it...
> 
> I will look on it.
Thanks.
 
> Question: do you have a repository ?
> It will make easier to send patch versus latest code.
No public repository.

> ----------------------------------------
> Please note
> 
> $ head -n 2 Makefile.am
> ACLOCAL_AMFLAGS=-I m4
> SUBDIRS= . examples scripts bench blas_sparse doc
> 
> Autoconf/Autoreconf will look for a directory named m4,
> but you have all *.m4 files in the tree root.
> 
> Also copying all the files in a m4 directory, I get:
> 
> $ autoreconf -ifv
> autoreconf-2.69: Entering directory `.'
> autoreconf-2.69: configure.ac: not using Gettext
> autoreconf-2.69: running: aclocal --force -I m4
> svn: E155007:
> '/cygdrive/e/cyg_pub/devel/librsb/build32/librsb-1.2.0-rc1' is not a
> working copy
> basename: missing operand
> Try 'basename --help' for more information.
It's meant to be this way for libtool.

Re: [Librsb-users] confusing config error on cygwin

[Marco A. <mar...@gm...>]

On 6/22/2015 1:57 PM, Michele Martone wrote:
> On 20150622@13:34, Marco Atzeri wrote:

> I've just made an out-of-tree build on Linux and it does not have such
> problems.
> I don't have cygwin so I don't know how to replicate the problem you
> encounter.
> Now, sbtc is not part of the library but a test program which according
> to the config.log files you've sent me shall be even empty.
> Indeed, you shall have librsb.a built I suppose.
>
> If you have a suggestion about how to replicate the problem or how to
> fix it, I'll use it...

I will look on it.

Question: do you have a repository ?
It will make easier to send patch versus latest code.

----------------------------------------
Please note

$ head -n 2 Makefile.am
ACLOCAL_AMFLAGS=-I m4
SUBDIRS= . examples scripts bench blas_sparse doc

Autoconf/Autoreconf will look for a directory named m4,
but you have all *.m4 files in the tree root.

Also copying all the files in a m4 directory, I get:

$ autoreconf -ifv
autoreconf-2.69: Entering directory `.'
autoreconf-2.69: configure.ac: not using Gettext
autoreconf-2.69: running: aclocal --force -I m4
svn: E155007: 
'/cygdrive/e/cyg_pub/devel/librsb/build32/librsb-1.2.0-rc1' is not a 
working copy
basename: missing operand
Try 'basename --help' for more information.
-------------------------------------------

Regards
Marco

Re: [Librsb-users] confusing config error on cygwin

[Michele M. <mic...@us...>]

On 20150622@13:34, Marco Atzeri wrote:
> On 6/22/2015 12:56 PM, Michele Martone wrote:
> >On 20150622@12:21, Marco Atzeri wrote:
> >>  "make test" fails also using same tree for build and source..
> >>...
> >>gcc -g -O2 -fopenmp -std=c99    sbtc.c   -o sbtc
> >>sbtc.c:39:25: fatal error: blas_sparse.h: No such file or directory
> >>  #include <blas_sparse.h>
> >>                          ^
> >>compilation terminated.
> >For some reason "-DHAVE_CONFIG_H -I." is not present in your line
> >compiling sbtc.c. What comes out if you 'grep ^DEFS Makefile' ?
> 
> same in both case
> 
> $ grep ^DEFS Makefile
> DEFS = -DHAVE_CONFIG_H

This seems correct.
However, it is strange that we don't have this flag in the gcc -O2..
line above.

I've just made an out-of-tree build on Linux and it does not have such
problems.
I don't have cygwin so I don't know how to replicate the problem you
encounter.
Now, sbtc is not part of the library but a test program which according
to the config.log files you've sent me shall be even empty.
Indeed, you shall have librsb.a built I suppose.

If you have a suggestion about how to replicate the problem or how to
fix it, I'll use it...

Re: [Librsb-users] confusing config error on cygwin

[Marco A. <mar...@gm...>]

On 6/22/2015 12:56 PM, Michele Martone wrote:
> On 20150622@12:21, Marco Atzeri wrote:
>>   "make test" fails also using same tree for build and source..
>> ...
>> gcc -g -O2 -fopenmp -std=c99    sbtc.c   -o sbtc
>> sbtc.c:39:25: fatal error: blas_sparse.h: No such file or directory
>>   #include <blas_sparse.h>
>>                           ^
>> compilation terminated.
> For some reason "-DHAVE_CONFIG_H -I." is not present in your line
> compiling sbtc.c. What comes out if you 'grep ^DEFS Makefile' ?

same in both case

$ grep ^DEFS Makefile
DEFS = -DHAVE_CONFIG_H

Re: [Librsb-users] confusing config error on cygwin

[Michele M. <mic...@us...>]

On 20150622@12:21, Marco Atzeri wrote:
> .. 
> ... 
> your configure help say that the default is "all", so may be you
> need to clarify the point.
Where did you read that ?
The default is the four BLAS types:

 ./configure --help # gives:
   --enable-matrix-types   Generate kernels for specified types
                          (default:"double,float,float complex,double complex") (you can specify 'all' to
                          get all of them, or 'blas' for Sparse BLAS ones)
> >>...
>  "make test" fails also using same tree for build and source..
> ...
> gcc -g -O2 -fopenmp -std=c99    sbtc.c   -o sbtc
> sbtc.c:39:25: fatal error: blas_sparse.h: No such file or directory
>  #include <blas_sparse.h>
>                          ^
> compilation terminated.
For some reason "-DHAVE_CONFIG_H -I." is not present in your line
compiling sbtc.c. What comes out if you 'grep ^DEFS Makefile' ?

Re: [Librsb-users] confusing config error on cygwin

[Marco A. <mar...@gm...>]

On 6/22/2015 12:06 PM, Michele Martone wrote:
> On 20150622@11:53, Marco Atzeri wrote:
>> On 6/22/2015 10:42 AM, Michele Martone wrote:
>>> On 20150622@10:26, Marco Atzeri wrote:

>>
>> surely is misleading for me.
>> It looks a config error, or at least a warning
> In the configure scripts,
>   - that message has been generated by AC_MSG_NOTICE;
>   - warnings are generated by AC_MSG_WARN;
>   - errors are generated by AC_MSG_ERROR.
> If you have less scary messages or wording to suggest, you're
> very welcome :-) !
>
>> but octave and int are available.
> Even if octave has been detected, the 'int' type has been not
> enabled (by you): grep your config.log files for "Numerical types":
>
>   $ bzgrep Numerical.types ~/config.log_i686.bz2
>          All Numerical types          : "int,double,float,float complex,double complex"
>          Numerical types              : "double,float,float complex,double complex" vs "double,float,float complex,double complex"
>   $ bzgrep Numerical.types config.log_x86_64.bz2
>          All Numerical types          : "int,double,float,float complex,double complex"
>          Numerical types              : "double,float,float complex,double complex" vs "double,float,float complex,double complex"
>
> You can enable 'int' at configure time by e.g.
>   --enable-matrix-types="int, double, double complex"

your configure help say that the default is "all", so may be you need to 
clarify the point.

>
>>> Michele
>>
>> About testing, for what I see:
>> - "make check" do nothing excepts this warning:
>>
>> make[2]: Entering directory
>> '/cygdrive/e/cyg_pub/devel/librsb/librsb-1.2.0-rc1_build32'
>> if test -f ch2icfb ; then
>> /pub/devel/librsb/librsb-1.2.0-rc1/scripts/rsb_h_to_rsb_fi.sh
>> /pub/devel/librsb/librsb-1.2.0-rc1 > rsb.F90 ; else echo "Warning:
>> Your system did not build ch2icfb for some reason --- skipping
>> rebuild of "rsb.F90 ;true ; fi
> There is no rule written by me for "make check" in that Makefile.
>
> I'll reply to the rest of your email separately.
>> ...


  "make test" fails also using same tree for build and source..

  $ make test |& tee make_test.log
gcc -g -O2 -fopenmp -std=c99    ot.c   -o ot
ot.c: In function ‘main’:
ot.c:1:1: warning: implicit declaration of function ‘printf’ 
[-Wimplicit-function-declaration]
  int main(){printf("sorry, you did not install octave, so the octave 
based tester is disabled\n");return 0;}
  ^
ot.c:1:12: warning: incompatible implicit declaration of built-in 
function ‘printf’
  int main(){printf("sorry, you did not install octave, so the octave 
based tester is disabled\n");return 0;}
             ^
gcc -g -O2 -fopenmp -std=c99    sbtc.c   -o sbtc
sbtc.c:39:25: fatal error: blas_sparse.h: No such file or directory
  #include <blas_sparse.h>
                          ^
compilation terminated.
<builtin>: recipe for target 'sbtc' failed
make: *** [sbtc] Error 1


$ find . -name blas_sparse.h
./blas_sparse/blas_sparse.h
./blas_sparse.h

Re: [Librsb-users] confusing config error on cygwin

[Michele M. <mic...@us...>]

On 20150622@11:53, Marco Atzeri wrote:
> On 6/22/2015 10:42 AM, Michele Martone wrote:
> >On 20150622@10:26, Marco Atzeri wrote:
> >>Hi Michele
> >>librsb-1.2.0-rc1  on cygwin (32 bit and 64 bit)
> >>configure:25969: You seem to not have GNU Octave or have disabled
> >>'int' type. Part of the test suite will
> >>not be generated. If you want more testing capabilities, you should
> >>enable the 'int' type as well.
> >>configure:25988: Successfully configured librsb version "1.2.0-rc1".
> >>
> >>Attached config.log's
> >>
> >>Regards
> >>Marco
> >>
> >
> >Hi Marco,
> >
> >If you are referring to the message from "configure:25969:", that is not
> >an error (and neither a warning) but just a notice.
> >Did the library build went well afterwards ?
> 
> surely is misleading for me.
> It looks a config error, or at least a warning
In the configure scripts,
 - that message has been generated by AC_MSG_NOTICE;
 - warnings are generated by AC_MSG_WARN;
 - errors are generated by AC_MSG_ERROR.
If you have less scary messages or wording to suggest, you're
very welcome :-) !

> but octave and int are available.
Even if octave has been detected, the 'int' type has been not 
enabled (by you): grep your config.log files for "Numerical types":

 $ bzgrep Numerical.types ~/config.log_i686.bz2
        All Numerical types          : "int,double,float,float complex,double complex"
        Numerical types              : "double,float,float complex,double complex" vs "double,float,float complex,double complex"
 $ bzgrep Numerical.types config.log_x86_64.bz2
        All Numerical types          : "int,double,float,float complex,double complex"
        Numerical types              : "double,float,float complex,double complex" vs "double,float,float complex,double complex"

You can enable 'int' at configure time by e.g. 
 --enable-matrix-types="int, double, double complex"

> >Michele
> 
> About testing, for what I see:
> - "make check" do nothing excepts this warning:
> 
> make[2]: Entering directory
> '/cygdrive/e/cyg_pub/devel/librsb/librsb-1.2.0-rc1_build32'
> if test -f ch2icfb ; then
> /pub/devel/librsb/librsb-1.2.0-rc1/scripts/rsb_h_to_rsb_fi.sh
> /pub/devel/librsb/librsb-1.2.0-rc1 > rsb.F90 ; else echo "Warning:
> Your system did not build ch2icfb for some reason --- skipping
> rebuild of "rsb.F90 ;true ; fi
There is no rule written by me for "make check" in that Makefile.

I'll reply to the rest of your email separately.
> ...

Re: [Librsb-users] confusing config error on cygwin

[Marco A. <mar...@gm...>]

On 6/22/2015 10:42 AM, Michele Martone wrote:
> On 20150622@10:26, Marco Atzeri wrote:
>> Hi Michele
>> librsb-1.2.0-rc1  on cygwin (32 bit and 64 bit)
>> configure:25969: You seem to not have GNU Octave or have disabled
>> 'int' type. Part of the test suite will
>> not be generated. If you want more testing capabilities, you should
>> enable the 'int' type as well.
>> configure:25988: Successfully configured librsb version "1.2.0-rc1".
>>
>> Attached config.log's
>>
>> Regards
>> Marco
>>
>
> Hi Marco,
>
> If you are referring to the message from "configure:25969:", that is not
> an error (and neither a warning) but just a notice.
> Did the library build went well afterwards ?

surely is misleading for me.
It looks a config error, or at least a warning but octave and int are 
available.

> Michele

About testing, for what I see:
- "make check" do nothing excepts this warning:

make[2]: Entering directory 
'/cygdrive/e/cyg_pub/devel/librsb/librsb-1.2.0-rc1_build32'
if test -f ch2icfb ; then 
/pub/devel/librsb/librsb-1.2.0-rc1/scripts/rsb_h_to_rsb_fi.sh 
/pub/devel/librsb/librsb-1.2.0-rc1 > rsb.F90 ; else echo "Warning: Your 
system did not build ch2icfb for some reason --- skipping rebuild of 
"rsb.F90 ;true ; fi


- "make test" fails, I suspect as source and build are on different
   trees
   (in addition, printf usage requires #include <stdio.h> )

$ echo $PWD
/pub/devel/librsb/librsb-1.2.0-rc1_build32

$ make test
gcc -g -O2 -fopenmp -std=c99    /pub/devel/librsb/librsb-1.2.0-rc1/ot.c 
   -o ot
/pub/devel/librsb/librsb-1.2.0-rc1/ot.c: In function ‘main’:
/pub/devel/librsb/librsb-1.2.0-rc1/ot.c:1:1: warning: implicit 
declaration of function ‘printf’ [-Wimplicit-function-declaration]
  int main(){printf("sorry, you did not install octave, so the octave 
based tester is disabled\n");return 0;}
  ^
/pub/devel/librsb/librsb-1.2.0-rc1/ot.c:1:12: warning: incompatible 
implicit declaration of built-in function ‘printf’
  int main(){printf("sorry, you did not install octave, so the octave 
based tester is disabled\n");return 0;}
             ^
gcc -g -O2 -fopenmp -std=c99 
/pub/devel/librsb/librsb-1.2.0-rc1/sbtc.c   -o sbtc
/pub/devel/librsb/librsb-1.2.0-rc1/sbtc.c:39:25: fatal error: 
blas_sparse.h: No such file or directory
  #include <blas_sparse.h>
                          ^
compilation terminated.
<builtin>: recipe for target 'sbtc' failed
make: *** [sbtc] Error 1

Re: [Librsb-users] confusing config error on cygwin

[Michele M. <mic...@us...>]

On 20150622@10:26, Marco Atzeri wrote:
> Hi Michele
> librsb-1.2.0-rc1  on cygwin (32 bit and 64 bit)
> I see a confusing error message as reported on config.log
> 
> ..
>        OCTAVE                 : octave
> ..
>         All Numerical types          : "int,double,float,float
> complex,double complex"
> ..
> configure:25969: You seem to not have GNU Octave or have disabled
> 'int' type. Part of the test suite will
> not be generated. If you want more testing capabilities, you should
> enable the 'int' type as well.
> configure:25988: Successfully configured librsb version "1.2.0-rc1".
> 
> Attached config.log's
> 
> Regards
> Marco
> 

Hi Marco,

If you are referring to the message from "configure:25969:", that is not
an error (and neither a warning) but just a notice.
Did the library build went well afterwards ?

Michele

[Librsb-users] confusing config error on cygwin

[Marco A. <mar...@gm...>]

Hi Michele
librsb-1.2.0-rc1  on cygwin (32 bit and 64 bit)
I see a confusing error message as reported on config.log

..
        OCTAVE                 : octave
..
         All Numerical types          : "int,double,float,float 
complex,double complex"
..
configure:25969: You seem to not have GNU Octave or have disabled 'int' 
type. Part of the test suite will
not be generated. If you want more testing capabilities, you should 
enable the 'int' type as well.
configure:25988: Successfully configured librsb version "1.2.0-rc1".

Attached config.log's

Regards
Marco

[Librsb-users] [carandraug@octave.org: sparsersb 1.0.0 released]

[Michele M. <mic...@ip...>]

FYI

----- Forwarded message from Carnë Draug <car...@oc...> -----

From: Carnë Draug <car...@oc...>
Subject: sparsersb 1.0.0 released
To: octave maintainers mailing list <mai...@oc...>, Octave Help <he...@oc...>
Date: Sun, 31 May 2015 13:41:00 +0100

Hi everyone

the first release of the sparsersb package [1] is out, by Michele Martone.

The sparsersb package provides an Octave interface to librsb [2] which
implements the RSB sparse matrix format.  librsb is a library for sparse
matrix computations featuring the Recursive Sparse Blocks (RSB) matrix format.
This format allows cache efficient and multi-threaded (that is, shared memory
parallel) operations on large sparse matrices.  The most common operations
necessary to iterative solvers are available, e.g.: matrix-vector
multiplication, triangular solution, rows/columns scaling, diagonal
extraction / setting, blocks extraction, norm computation, formats
conversion.  The RSB format is especially well suited for symmetric and
transposed multiplication variants. On these variants, librsb has been found
to be faster than Intel MKL's implementation for CSR.  Most numerical kernels
code is auto generated, and the supported numerical types can be chosen by
the user at build time.

Enjoy Octave responsibly,
Carnë

[1] http://octave.sourceforge.net/sparsersb/
[2] https://sourceforge.net/projects/librsb/

----- End forwarded message -----

[Librsb-users] librsb-1.2.0 release candidate 0 is out and available for download

[Michele M. <mic...@us...>]

Dear users and friends of librsb,

I write to inform you that librsb-1.2.0-rc0 is out and available
for download.

Main improvements of librsb-1.2 are:

 * faster runtime autotuning of structure and threads for sparse
   matrix-vector/matrix-matrix multiply and triangular solve:
   now uses a parallel merge/subdivision based algorithm 
 * speeded up triangular solution of a matrix by multi-vector
 * introduced extension BLAS property blas_rsb_autotune_next_operation
   to trigger auto tuning at the next USMV/USMM call
 * collection/report of time spent in librsb routines now possible
   (a simple built-in profiler)
 * further Encapsulated PostScript (EPS) matrix rendering options
 * improved benchmark program for large parameter studies (converts
   and reuses matrices rather than re-reading them from disk)
 * binary performance records write/read/combine support, also with
   Intel MKL CSR built-in comparison; LaTeX tables output and stats

Source code, documentation, benchmarks and full change log on:
 http://sourceforge.net/projects/librsb

As always, you are encouraged to provide suggestions, criticism and
test cases.

[Librsb-users] librsb-1.1.0-rc0 is out

[Michele M. <mic...@us...>]

Dear users and friends of librsb,

I write to inform you that librsb-1.1.0-rc0 is out.

Main improvements of librsb-1.1 are:

 * runtime autotuning of optimal threads / structure for sparse
   matrix-vector multiply and triangular solve;
   also available as an extension of the Sparse BLAS interface
 * Fortran ISO C Binding interface in rsb.F90/rsb.mod is now 100%
   compatible with the C one in rsb.h
 * slightly improved performance of matrix by vector / multi-vector
   multiplication
 * significantly improved documentation
 * collection of time spent in librsb routines now possible
 * better debug options (e.g.: max allocatable memory, verbose
   operation of rsb.h or Sparse BLAS routines on rebuild / request )

On a wide sample of matrices, the new autotuning feature has been
observed to improve execution speed up between a few and a few hundred
percent.

Source code, documentation, benchmarks and change log on:
 http://sourceforge.net/projects/librsb

As always, you are encouraged to provide suggestions and criticism.