These are chat archives for elemental/chat

28th
Jul 2016
Jack Poulson
@poulson
Jul 28 2016 01:37
I agree that settling on how to handle dependencies is the deeper issue
LAPACK is up on GitHub officially as of this week
Perhaps we can get ScaLAPACK there as well
Metis and ParMetis are hopeless though
My conjecture is that if we iron out dependencies everything else is easy
Ryan H. Lewis
@rhl-
Jul 28 2016 03:21
@poulson metis is available in most package managers
lapack is in most package managers
looks like scalapack is starting to be packaged for more recent fedora and centos: https://www.rpmfind.net/linux/rpm2html/search.php?query=scalapack
it looks like all those packages are also in debian.
Jack Poulson
@poulson
Jul 28 2016 04:01
the ParMETIS modifications aren't
and it is helpful but not necessary to have some bugs fixed in ScaLAPACK
Ryan H. Lewis
@rhl-
Jul 28 2016 04:11
those things should be contributed to those libraries.
Jack Poulson
@poulson
Jul 28 2016 04:12
just because you send it to the project doesn't mean it's accepted...
ParMETIS is notoriously difficult to get patches into
and I have sent the patch, for what it's worth
Ryan H. Lewis
@rhl-
Jul 28 2016 04:12
yeah, that sucks
you can also send the patch to packagers
and explain that
and they can apply them slightly downstream.
the changes are the commit logs
Ryan H. Lewis
@rhl-
Jul 28 2016 04:13
what I am saying is that fedora packagers could apply the patch
why does ParMETIS ppl accept the patches?
rather not accept them
also, i've started an elemental spec file
Jack Poulson
@poulson
Jul 28 2016 04:14
ParMETIS is run by one professor
and he is busy
Ryan H. Lewis
@rhl-
Jul 28 2016 04:14
I know, Karypis
he used to be against having open licenses for this stuff period
I've talked to him :)
Jack Poulson
@poulson
Jul 28 2016 04:15
given that 99.99% of people only take, it's hard to really argue against
Ryan H. Lewis
@rhl-
Jul 28 2016 04:16
haha
so i'm having a problem with the cmake build with this uber new gcc
Jack Poulson
@poulson
Jul 28 2016 04:16
open source burnout is a serious issue
what is the problem?
docker image is using GCC 6.1.1 which honestly, i'm surprised exists as a version.
Jack Poulson
@poulson
Jul 28 2016 04:17
that is fixed with the current HEAD
0.85 is really, really old
Ryan H. Lewis
@rhl-
Jul 28 2016 04:17
frustratingly I cant point a spec file at HEAD
so I pointed it at the release.
Jack Poulson
@poulson
Jul 28 2016 04:17
fair enough
but, FWIW, that issue has been fixed for months
Ryan H. Lewis
@rhl-
Jul 28 2016 04:18
i mean, i believe you
but, how do you want me to fix it?
Jack Poulson
@poulson
Jul 28 2016 04:18
can you not point to a branch?
Ryan H. Lewis
@rhl-
Jul 28 2016 04:18
oh, I forgot to mention
I can point to a tarball off a URL
Jack Poulson
@poulson
Jul 28 2016 04:18
master is more stable than 0.85
Ryan H. Lewis
@rhl-
Jul 28 2016 04:18
then ship it :)
call it 0.9
so, i'll use git and make a tarball for now
Jack Poulson
@poulson
Jul 28 2016 04:19
I want D&C SVD really bad
most of the way there
Ryan H. Lewis
@rhl-
Jul 28 2016 04:19
infact, I'll do this, i'll make a release on my own elemental fork on github.
Jack Poulson
@poulson
Jul 28 2016 04:19
people have been benchmarking against the SVD and this will help
Ryan H. Lewis
@rhl-
Jul 28 2016 04:19
and then we can move it to the official one
Jack Poulson
@poulson
Jul 28 2016 04:19
ok
I'm not against that
it's important that there not be too much of a bus factor
Ryan H. Lewis
@rhl-
Jul 28 2016 04:20
weird
0.86-rc1 is on github
is that much better?
Jack Poulson
@poulson
Jul 28 2016 04:20
yes, and it's very old
no, it is not
I have been bad at the release cycle
changing jobs and all that
Ryan H. Lewis
@rhl-
Jul 28 2016 04:28
i understand
so, for the moment I am past that
one thing to worry about for packages
are which dependencies
when there are choices
like OpenMPI or MPICH?
for my purposes, I would like mpich
so maybe ill start there
and we can fix it later
Jack Poulson
@poulson
Jul 28 2016 04:30
unfortunately everyone has their own favorite
the honest answer is to generate a different package for each compiler and MPI combination...
Ryan H. Lewis
@rhl-
Jul 28 2016 04:30
there is a way to do that.. i just dont know offhand
Jack Poulson
@poulson
Jul 28 2016 04:30
MPI packaging is a complete mess
Ryan H. Lewis
@rhl-
Jul 28 2016 04:31
so lets just start with somethign and fix it.
Jack Poulson
@poulson
Jul 28 2016 04:35
so OpenMPI and MPICH seem to be the options for Fedora
Ryan H. Lewis
@rhl-
Jul 28 2016 04:36
yeah.
Jack Poulson
@poulson
Jul 28 2016 04:36
that doesn't include Intel MPI, but I guess that makes sense
or MVAPICH
Ryan H. Lewis
@rhl-
Jul 28 2016 04:36
If intel produced rpms we could make elemental packages for it
Free OSS Build infra
how do you tell CMAKE about the specific MPI Compiler
and is there a way to build elemental without mpi ?
(answer can be no)
Jack Poulson
@poulson
Jul 28 2016 04:39
not at this point, but that is a common question
and I've given it serious thought
though I'm honestly afraid that it would be 10x more popular without the MPI
Ryan H. Lewis
@rhl-
Jul 28 2016 04:40
haha
apparently part of fedora packaging guidelines
provide it if its supported
Jack Poulson
@poulson
Jul 28 2016 04:40
adding support for that would be a solid weekend project
Ryan H. Lewis
@rhl-
Jul 28 2016 04:40
its not important.
Jack Poulson
@poulson
Jul 28 2016 04:41
I might look into it after adding D&C SVD and Hermitian EVD
Ryan H. Lewis
@rhl-
Jul 28 2016 04:45
ok more important
how do I turn this off
[ 60%] Built target metis
[100%] Built target parmetis
Jack Poulson
@poulson
Jul 28 2016 04:45
turn what off?
the CMake output?
Ryan H. Lewis
@rhl-
Jul 28 2016 04:45
i dont want it downloading dependencies
i want it to use stuff in repos
Jack Poulson
@poulson
Jul 28 2016 04:46
you have to have provided the libraries manually to CMake
or at least their locations, and they need to be detected
otherwise the library cannot build
Ryan H. Lewis
@rhl-
Jul 28 2016 04:47
find_package(..) should work without modification when RPM's are installed.
Jack Poulson
@poulson
Jul 28 2016 04:47
...
so what directory does the RPM install into?
e.g., where is libmetis.so?
Ryan H. Lewis
@rhl-
Jul 28 2016 04:48
[makerpm@d79005f25ca2 SPECS]$ rpm -ql metis-devel
/usr/include/metis.h
/usr/lib64/libmetis.so
Jack Poulson
@poulson
Jul 28 2016 04:48
ugh
assuming a lib64/ subfolder is awful
there should have been output from whether or not one of the other was found
Ryan H. Lewis
@rhl-
Jul 28 2016 04:49
looks btw like parmetis is not really packaged. we can package that separetly.
Jack Poulson
@poulson
Jul 28 2016 04:49
I would assume it would be if /usr/lib/libmetis.so was the path
parmetis is not important for 99% of people
if not more
Ryan H. Lewis
@rhl-
Jul 28 2016 04:49
fedora follows the file heirarchy standard
Jack Poulson
@poulson
Jul 28 2016 04:50
CMake apparently does not then
cmake isn't auto finding metis?
Jack Poulson
@poulson
Jul 28 2016 04:51
apparently not if it built it
oh, I know what happened
Ryan H. Lewis
@rhl-
Jul 28 2016 04:51
can we turn off parmetis stuff?
Jack Poulson
@poulson
Jul 28 2016 04:51
it defaulted to ParMETIS, which internally builds METIS
-D EL_DISABLE_PARMETIS=ON
Ryan H. Lewis
@rhl-
Jul 28 2016 04:52
ok, now its downloading suite sparse.
lets see about that
ugh, suitesparse depends on tbb
-D EL_DISABLE_SUITESPARSE
?
Jack Poulson
@poulson
Jul 28 2016 04:55
it isn't downloading suite sparse
it is included
my modification is anyway
Ryan H. Lewis
@rhl-
Jul 28 2016 04:55
oh, so it doesn't depend on suitesparse?
Jack Poulson
@poulson
Jul 28 2016 04:55
it is a very significant extension and it has its own symbols
Ryan H. Lewis
@rhl-
Jul 28 2016 04:55
by that I mean
Jack Poulson
@poulson
Jul 28 2016 04:55
no
Ryan H. Lewis
@rhl-
Jul 28 2016 04:55
it doesn't depend on suitesparse the rpm
Jack Poulson
@poulson
Jul 28 2016 04:55
it is just using an LDL that was similar enough to Tim Davis's that I made sure to give credit
no it does not
see external/
Ryan H. Lewis
@rhl-
Jul 28 2016 04:56
its not even using git submodules? thats just code in the repos?
Jack Poulson
@poulson
Jul 28 2016 04:56
git submodules are broken on github
literally completely broken
Ryan H. Lewis
@rhl-
Jul 28 2016 04:56
CMake Error at CMAKE/ExternalOpenBLAS.cmake:31 (message):
Could not find gfortran library; please consider setting the GFORTRAN_LIB
variable.
Jack Poulson
@poulson
Jul 28 2016 04:56
yes
Ryan H. Lewis
@rhl-
Jul 28 2016 04:56
i dont think I've installed a BLAS or LAPACK
yet
Jack Poulson
@poulson
Jul 28 2016 04:57
please consider setting GFORTRAN_LIB
OpenBLAS involves linking C code and Fortran code together
it was going to build OpenBLAS for you
Ryan H. Lewis
@rhl-
Jul 28 2016 04:57
right, we are going to grab openblas from an rpm
Jack Poulson
@poulson
Jul 28 2016 04:57
but libgfortran was apparently not in the normal location
then you need to manually specify EL_MATH_LIBS
Ryan H. Lewis
@rhl-
Jul 28 2016 04:58
forgot to install openblas though, so lets ee
they are here: /usr/lib/gcc/x86_64-redhat-linux/6.1.1/32/libgfortran.so
/usr/lib/gcc/x86_64-redhat-linux/6.1.1/libgfortran.so
CMake should be able to find that..
what is BLACS
is it also an external dependency?
Jack Poulson
@poulson
Jul 28 2016 05:01
it is the Basic Linear Algebra Communication Subprograms and is a piece of ScaLAPACK
This message was deleted
Ryan H. Lewis
@rhl-
Jul 28 2016 05:02
do we also not need scalapack ?
Jack Poulson
@poulson
Jul 28 2016 05:03
it depends if you want the distributed Schur decomposition and pseudospectra packaged
the one routine Elemental uses from ScaLAPACK is the distributed Hessenberg QR algorithm
Ryan H. Lewis
@rhl-
Jul 28 2016 05:03
well I mean, its just version one
Jack Poulson
@poulson
Jul 28 2016 05:03
it is extremely buggy but I haven't written my own yet
Ryan H. Lewis
@rhl-
Jul 28 2016 05:03
i just want to get through a build and then I can fix the problems
scalapack is packaged in EPEL
Im not sure why cmake is not finding openblas, lapack, or gfortran libs
Jack Poulson
@poulson
Jul 28 2016 05:08
what directories are they in?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:08
[makerpm@d79005f25ca2 SPECS]$ rpm -ql openblas-devel
/usr/include/openblas
/usr/include/openblas/cblas.h
/usr/include/openblas/f77blas.h
/usr/include/openblas/lapacke.h
/usr/include/openblas/lapacke_config.h
/usr/include/openblas/lapacke_mangling.h
/usr/include/openblas/lapacke_utils.h
/usr/include/openblas/openblasconfig.h
/usr/lib64/libopenblas.so
/usr/lib64/libopenblas64.so
/usr/lib64/libopenblas64
.so
/usr/lib64/libopenblaso.so
/usr/lib64/libopenblaso64.so
/usr/lib64/libopenblaso64.so
/usr/lib64/libopenblasp.so
/usr/lib64/libopenblasp64.so
/usr/lib64/libopenblasp64
.so
there are bunch more packages too
[makerpm@d79005f25ca2 SPECS]$ rpm -ql openblas
openblas openblas-openmp openblas-openmp64 openblas-serial64 openblas-threads64
openblas-devel openblas-openmp64 openblas-serial64 openblas-threads openblas-threads64_
[makerpm@d79005f25ca2 SPECS]$ rpm -ql openblas-threadss64
package openblas-threadss64 is not installed
[makerpm@d79005f25ca2 SPECS]$ rpm -ql openblas-threads64
/usr/lib64/libopenblasp64-r0.2.18.so
/usr/lib64/libopenblasp64.so.0
The CMake build system is currently somewhat of a tire fire, but the relevant logic is contained in that file
aaand GitHub's servers are failing on me
but at least they're serving me a picture of an angry unicorn instead
Ryan H. Lewis
@rhl-
Jul 28 2016 05:13
lol I see it
were you trying to commit fixed to OpenBLAS.cmake
?
Jack Poulson
@poulson
Jul 28 2016 05:17
no, I was trying to open the link to make sure it worked
Ryan H. Lewis
@rhl-
Jul 28 2016 05:18
what about find_package(BLAS)
Jack Poulson
@poulson
Jul 28 2016 05:18
that requires Fortran
which, in some circumstances, Elemental does not
e.g., if you have a prebuilt BLAS library
Ryan H. Lewis
@rhl-
Jul 28 2016 05:18
ok, im really confused wtf is wrong
Jack Poulson
@poulson
Jul 28 2016 05:19
what is the full output?
from CMake
at least in the section where it is looking for OpenBLAS
it might be a linking issue
also, just figured out how to keep scalapack
with installed scalapack: -- Searching for previously installed ScaLAPACK
-- Looking for dtrsm
-- Looking for dtrsm - not found
-- Looking for dtrsm
-- Looking for dtrsm
- not found
-- Looking for dsytrd
-- Looking for dsytrd - not found
-- Looking for dsytrd
-- Looking for dsytrd
- not found
-- Looking for pdsyngst
-- Looking for pdsyngst - not found
-- Looking for pdsyngst
-- Looking for pdsyngst
- found
-- Looking for Csys2blacs_handle
-- Looking for Csys2blacs_handle - not found
-- Could not link ScaLAPACK
Jack Poulson
@poulson
Jul 28 2016 05:20
the issue is that it could not find GFORTRAN_LIB
where is libgfortran.so on your system?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:20
yeah, i showed you were eariler
Jack Poulson
@poulson
Jul 28 2016 05:21
ah, /usr/lib/gcc/x86_64-redhat-linux/6.1.1/32/libgfortran.so
saying "CMake should be able to find that" is not a productive statement...
Ryan H. Lewis
@rhl-
Jul 28 2016 05:22
there is also the non 32 bit version
Jack Poulson
@poulson
Jul 28 2016 05:22
what is your proposed strategy for guessing that path?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:22
ive done this before
gfortran should know its companion library path, no?
Jack Poulson
@poulson
Jul 28 2016 05:22
only if you told Elemental where gfortran was
Ryan H. Lewis
@rhl-
Jul 28 2016 05:23
its on the path
Jack Poulson
@poulson
Jul 28 2016 05:23
you can link to OpenBLAS without doing that
in your case it is
are you proposing shelling out to a gfortran command and parsing the output?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:24
yeah, ive done that
but, for some reason this gfortran doesnt provide that
it provides:
[makerpm@d79005f25ca2 SPECS]$ gfortran -print-libgcc-file-name
/usr/lib/gcc/x86_64-redhat-linux/6.1.1/libgcc.a
here we go: [makerpm@d79005f25ca2 SPECS]$ gfortran -print-file-name=libgfortran.so
/usr/lib/gcc/x86_64-redhat-linux/6.1.1/libgfortran.so
Jack Poulson
@poulson
Jul 28 2016 05:25
that is reasonable
what to do if gfortran is not in the path?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:25
[makerpm@d79005f25ca2 SPECS]$ dirname $(gfortran -print-file-name=libgfortran.so)
/usr/lib/gcc/x86_64-redhat-linux/6.1.1
Jack Poulson
@poulson
Jul 28 2016 05:26
or, more importantly, if the wrong gfortran is in the path
Ryan H. Lewis
@rhl-
Jul 28 2016 05:26
well, luckily, this is an rpm
so thhat is not possible.
welcome to the world of fixing the base environment, it is nice :)
Jack Poulson
@poulson
Jul 28 2016 05:26
many MPI developers frequently have multiple versions of compilers installed
Ryan H. Lewis
@rhl-
Jul 28 2016 05:26
thats what conan is for
rpms are built for OS package managers
they fix the toolchain
Jack Poulson
@poulson
Jul 28 2016 05:27
but if something like this is to be added into the CMake build, I'd like to think of the edge cases
it will happen, unfortunately
Ryan H. Lewis
@rhl-
Jul 28 2016 05:27
im adding it to the rpm spec
Jack Poulson
@poulson
Jul 28 2016 05:28
perhaps it is okay to assume that the first found gfortran is the correct one
I guess most searches do so
Ryan H. Lewis
@rhl-
Jul 28 2016 05:29
so in this case
there is one gfortran installed
rpm's are precompiled
is GFORTRAN_LIB an env variable or a cmake one?
UGH

-- A library with BLAS API not found. Please specify library location.
-- LAPACK requires BLAS
-- A library with LAPACK API not found. Please specify library location.
-- Searching for previously installed OpenBLAS+LAPACK
-- Looking for dgemm
-- Looking for dgemm - not found
-- Looking for dgemm
-- Looking for dgemm
- not found
-- Looking for dsytrd
-- Looking for dsytrd - not found
-- Looking for dsytrd
-- Looking for dsytrd
- not found
CMake Warning at cmake/external_projects/ElMath/OpenBLAS.cmake:78 (message):
OpenBLAS was found as /usr/lib64/libopenblas.so, but BLAS support was not
detected
Call Stack (most recent call first):
cmake/external_projects/ElMath.cmake:251 (include)
CMakeLists.txt:358 (include)

CMake Warning at cmake/external_projects/ElMath/OpenBLAS.cmake:83 (message):
OpenBLAS was found as /usr/lib64/libopenblas.so, but LAPACK support was not
detected
Call Stack (most recent call first):
cmake/external_projects/ElMath.cmake:251 (include)
CMakeLists.txt:358 (include)

-- Will pull OpenBLAS from https://github.com/xianyi/OpenBLAS.git

Jack Poulson
@poulson
Jul 28 2016 05:32
CMake
and CMakeFiles/CMakeError.log probably reveals what is going on
whatever the undefined references are
Ryan H. Lewis
@rhl-
Jul 28 2016 05:35
ah sweet
fixed it :)
Jack Poulson
@poulson
Jul 28 2016 05:35
what was the issue?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:35
messed up the variable GFORTRAN_LIB
Jack Poulson
@poulson
Jul 28 2016 05:35
ah
Ryan H. Lewis
@rhl-
Jul 28 2016 05:35
looks like it got into building El now. solid.
Jack Poulson
@poulson
Jul 28 2016 05:35
kickass
Ryan H. Lewis
@rhl-
Jul 28 2016 05:36
and its going to do openmpi and mpich
it wasn't detecting scalapack correctltly but i wont if fortran fixed it
so: -- Searching for previously installed ScaLAPACK
-- Looking for dtrsm
-- Looking for dtrsm - not found
-- Looking for dtrsm
-- Looking for dtrsm
- not found
-- Looking for dsytrd
-- Looking for dsytrd - not found
-- Looking for dsytrd
-- Looking for dsytrd
- not found
-- Looking for pdsyngst
-- Looking for pdsyngst - not found
-- Looking for pdsyngst
-- Looking for pdsyngst
- found
-- Looking for Csys2blacs_handle
-- Looking for Csys2blacs_handle - not found
-- Could not link ScaLAPACK
-- Will pull ScaLAPACK from https://github.com/scibuilder/scalapack.git
Jack Poulson
@poulson
Jul 28 2016 05:37
perhaps CMakeFiles/CMakeError.log's undefined references have something to say
Ryan H. Lewis
@rhl-
Jul 28 2016 05:38
looks like :
CMakeFiles/cmTC_10b8b.dir/CheckFunctionExists.c.o: In function main': CheckFunctionExists.c:(.text+0x15): undefined reference topdsyngst'
collect2: error: ld returned 1 exit status
CMakeFiles/cmTC_10b8b.dir/build.make:99: recipe for target 'cmTC_10b8b' failed
Jack Poulson
@poulson
Jul 28 2016 05:38
is there another undefined reference to pdsyngst_?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:38
ah its a DSO issue
/usr/bin/ld: CMakeFiles/cmTC_623b0.dir/CheckFunctionExists.c.o: undefined reference to symbol 'Csys2blacs_handle'
/usr/lib64/openmpi/lib/libmpiblacs.so.2: error adding symbols: DSO missing from command line
this might be an El build problem
Jack Poulson
@poulson
Jul 28 2016 05:41
how so?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:44
well, something somewhere has mis-specified its link line
it looks like the problem is in CHeckFunctionExists
Jack Poulson
@poulson
Jul 28 2016 05:46
the definition of CheckFunctionExists or the call site?
Ryan H. Lewis
@rhl-
Jul 28 2016 05:46
im not sure
i cant find yet where that function is defined
i mean, the callsite is one of the ones for scalapack
Jack Poulson
@poulson
Jul 28 2016 05:46
it is intrinsic to CMake
oh
that is CMakeErrorLog
Jack Poulson
@poulson
Jul 28 2016 05:51
hmm, it seems it is a possibility that libraries are being specified in the wrong order: http://stackoverflow.com/questions/19901934/strange-linking-error-dso-missing-from-command-line
Ryan H. Lewis
@rhl-
Jul 28 2016 05:51
ill replicate on a clean build one sec.
Jack Poulson
@poulson
Jul 28 2016 05:52
I don't see libmpiblacs in the link line
Ryan H. Lewis
@rhl-
Jul 28 2016 05:52
yeah, thats probably the issue
Jack Poulson
@poulson
Jul 28 2016 05:53
it must be a very old version of ScaLAPACK
very old
newer versions switched to a single blacs library
and even newer versions eliminated it entirely and merged it into libscalapack
unless libmpiblacs is packed directly into libscalapack
for the patches and stuff they apply: http://koji.fedoraproject.org/koji/rpminfo?rpmID=7350585
Jack Poulson
@poulson
Jul 28 2016 05:55
hmm, that is the newest version
Ryan H. Lewis
@rhl-
Jul 28 2016 05:56
it is picking up scalapack on the system
so is it a problem in FindSCALAPACK
that it doesn't produce the libraries line correctly?
Jack Poulson
@poulson
Jul 28 2016 05:57
it might be helpful to try to directly execute /usr/lib64/openmpi/bin/mpicc -DCHECK_FUNCTION_EXISTS=Csys2blacs_handle -Wl,-rpath -Wl,/usr/lib64/openmpi/lib -Wl,--enable-new-dtags CMakeFiles/cmTC_623b0.dir/CheckFunctionExists.c.o -o cmTC_623b0 -rdynamic /usr/lib64/openmpi/lib/libscalapack.so /usr/lib64/openmpi/lib/libmpi.so from within the appropriate directory
Ryan H. Lewis
@rhl-
Jul 28 2016 05:57
yeah im trying
Jack Poulson
@poulson
Jul 28 2016 05:57
or to try to manually link to ScaLAPACK
Ryan H. Lewis
@rhl-
Jul 28 2016 05:57
Fatal error: can't create CMakeFiles/cmTC_bb4c6.dir/CheckFunctionExists.c.o: No such file or director
Jack Poulson
@poulson
Jul 28 2016 05:58
yea, they delete those files
or it is in a subdirectory
Ryan H. Lewis
@rhl-
Jul 28 2016 05:59
i mean
that is the error
if we run the cmd
it will show us that
Jack Poulson
@poulson
Jul 28 2016 05:59
does libmpiblacs.so exist within /usr/lib64/openmpi/lib?
Ryan H. Lewis
@rhl-
Jul 28 2016 06:00
yeap
well
[makerpm@d79005f25ca2 CMakeTmp]$ ls /usr/lib64/openmpi/lib/libmpiblacs.so.2
/usr/lib64/openmpi/lib/libmpiblacs.so.2
Jack Poulson
@poulson
Jul 28 2016 06:01
there is no such library for my version of ScaLAPACK
Ryan H. Lewis
@rhl-
Jul 28 2016 06:01
odd
click on 'spec file'
thats how they make it
Jack Poulson
@poulson
Jul 28 2016 06:03
According to the release notes for 2.0 and on, BLACS is combined with libscalapack
umm, are they making their own libmpiblacs for backwards compatibility reasons? https://pkgs.fedoraproject.org/cgit/rpms/scalapack.git/tree/scalapack-2.0.2-shared-blacs.patch
Ryan H. Lewis
@rhl-
Jul 28 2016 06:06
yeah, thats what I was just about to show
not sure why
let me ask in #fedora-devel
Jack Poulson
@poulson
Jul 28 2016 06:06
I'm going to say that this is on them
Ryan H. Lewis
@rhl-
Jul 28 2016 06:06
is there a workaround?
Jack Poulson
@poulson
Jul 28 2016 06:07
need to understand their patches better first
not sure what line 9 is doing at the moment
a comment would have been nice
Ryan H. Lewis
@rhl-
Jul 28 2016 06:10
they are building blacs by hand
Jack Poulson
@poulson
Jul 28 2016 06:11
and then removing it from libscalapack.so?
Ryan H. Lewis
@rhl-
Jul 28 2016 06:12
i dont see any removing
Jack Poulson
@poulson
Jul 28 2016 06:12
then it shouldn't be needed to be linked
Ryan H. Lewis
@rhl-
Jul 28 2016 06:12
well not quite
perhaps those patches can just be removed?
Ryan H. Lewis
@rhl-
Jul 28 2016 06:19
ok, well none of those people are online
so i cant really ask
we could file a bug
maybe tomorrow
but there should be some easy workaround
like lets just add that SO to the link like when it exists in the same place..
can you point me at the scalapack release notes?
Jack Poulson
@poulson
Jul 28 2016 06:24
Ryan H. Lewis
@rhl-
Jul 28 2016 06:25
i dont see the note you mentioned about blacs
Ryan H. Lewis
@rhl-
Jul 28 2016 06:25
hm?
Jack Poulson
@poulson
Jul 28 2016 06:26
"""- The BLACS is now part of ScaLAPACK, and is compiled into the ScaLAPACK
library. It is no longer necessary to link against BLACS libraries."""
ok so here is a workaround
for now I will disable scalapack
what about these: -- Could NOT find QD (missing: QD_INCLUDES QD_LIBRARIES)
-- Could NOT find GMP (missing: GMP_INCLUDES GMP_LIBRARIES GMP_VERSION_OK) (Required is at least version "6.0.0")
Ryan H. Lewis
@rhl-
Jul 28 2016 06:36
ok, ive got a build going
im going to let it try and do its thing
Jack Poulson
@poulson
Jul 28 2016 06:36
awesome
Ryan H. Lewis
@rhl-
Jul 28 2016 06:36
which is taking forever
Jack Poulson
@poulson
Jul 28 2016 06:37
trying to respond to someone complaining about something being broken
Ryan H. Lewis
@rhl-
Jul 28 2016 06:37
its not a bad first go at it I think
Ryan H. Lewis
@rhl-
Jul 28 2016 14:14

pasted from IRC:

<spot> you pinged me several hours ago?
<rhl> Hi spot, I ended up filing a bug. I believe you maintain scalapack ? It seems the fedora package is applying a patch to create a shared object which was supposed to be removed in 2.0
<rhl> i'm trying to package a dependent library and its failing because it doesn't expect that new SO
<spot> ah yes, I saw that. I'm just kicking chromium through the build system again, lemme look at that scalapack issue next.
<rhl> we can talk in #fedora-devel if it is easier
<rhl> great :)

Jack Poulson
@poulson
Jul 28 2016 15:13
awesome
thanks for chasing that down
Ryan H. Lewis
@rhl-
Jul 28 2016 16:34
@poulson the elemental v0 rpm build looks good. I need to sync with you on some other details of it, can do that tonight. Should be done pretty quickly.