[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[Paul Garlick]

* gnu/packages/maths.scm (petsc-mumps-openmpi): New variable.
---
 gnu/packages/maths.scm | 42 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)

diff --git a/gnu/packages/maths.scm b/gnu/packages/maths.scm
index 736bac1..b2fff54 100644
--- a/gnu/packages/maths.scm
+++ b/gnu/packages/maths.scm
@@ -1724,6 +1724,48 @@ scientific applications modeled by partial differential equations.")
            ,@(delete "--with-mpi=0" ,cf)))))
     (synopsis "Library to solve PDEs (with complex scalars and MPI support)")))
 
+(define-public petsc-mumps-openmpi
+  (package (inherit petsc-openmpi)
+    (name "petsc-mumps-openmpi")
+    (inputs
+     `(("metis" ,metis)
+       ("mumps" ,mumps-openmpi)
+       ("scalapack" ,scalapack)
+       ("scotch" ,pt-scotch)
+       ,@(package-inputs petsc-openmpi)))
+    (arguments
+     (substitute-keyword-arguments (package-arguments petsc-openmpi)
+       ((#:configure-flags cf)
+        ``(,(string-append "--with-metis-include="
+                           (assoc-ref %build-inputs "metis") "/include")
+           ,(string-append "--with-metis-lib="
+                           (assoc-ref %build-inputs "metis") "/lib/"
+                           "libmetis.so")
+           ,(string-append "--with-mumps-include="
+                           (assoc-ref %build-inputs "mumps") "/include")
+           ,(string-append "--with-mumps-lib=["
+                           (assoc-ref %build-inputs "mumps") "/lib/"
+                           "libdmumps.a,"
+                           "libzmumps.a,"
+                           "libsmumps.a,"
+                           "libcmumps.a,"
+                           "libmumps_common.a,"
+                           "libpord.a]")
+           ,(string-append "--with-scalapack-lib="
+                           (assoc-ref %build-inputs "scalapack") "/lib/"
+                           "libscalapack.so")
+           ,(string-append "--with-ptscotch-include="
+                           (assoc-ref %build-inputs "scotch") "/include")
+           ,(string-append "--with-ptscotch-lib=["
+                           (assoc-ref %build-inputs "scotch") "/lib/"
+                           "libesmumps.a,"
+                           "libscotch.a,"
+                           "libscotcherr.a,"
+                           "libptesmumps.a,"
+                           "libptscotch.a,"
+                           "libptscotcherr.a]")
+           ,@,cf))))
+    (synopsis "Library to solve PDEs (with MUMPS and MPI support)")))
 
 (define-public python-kiwisolver
   (package
-- 
1.8.3.1

[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[Ludovic Courtès]

Hi,

Paul Garlick <pgarlick@tourbillion-technology.com> skribis:

> * gnu/packages/maths.scm (petsc-mumps-openmpi): New variable.

On current master (e72702475da2379e10ca09e724e10a9bceceef1b), it fails
like this:

--8<---------------cut here---------------start------------->8---
TESTING: checkDependencies from config.packages.PTScotch(/tmp/guix-build-petsc-mumps-openmpi-3.9.3.drv-0/petsc-3.9.3/config/BuildSystem/config/package.py:721)
TESTING: configureLibrary from config.packages.PTScotch(/tmp/guix-build-petsc-mumps-openmpi-3.9.3.drv-0/petsc-3.9.3/config/BuildSystem/config/package.py:746)
TESTING: check from config.libraries(/tmp/guix-build-petsc-mumps-openmpi-3.9.3.drv-0/petsc-3.9.3/config/BuildSystem/config/libraries.py:155)
*******************************************************************************
         UNABLE to CONFIGURE with GIVEN OPTIONS    (see configure.log for details):
-------------------------------------------------------------------------------
--with-ptscotch-lib=['/gnu/store/rnbax3lpgai4mg6hgxp9j80jkxhavini-pt-scotch-6.0.5a/lib/libesmumps.a', 'libscotch.a', 'libscotcherr.a', 'libptesmumps.a', 'libptscotch.a', 'libptscotcherr.a'] and 
--with-ptscotch-include=['/gnu/store/rnbax3lpgai4mg6hgxp9j80jkxhavini-pt-scotch-6.0.5a/include'] did not work
*******************************************************************************
--8<---------------cut here---------------end--------------->8---

Any ideas?

Thanks,
Ludo’.

[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[Paul Garlick]

Hi Ludo,

> On current master (e72702475da2379e10ca09e724e10a9bceceef1b), it
> fails

That's curious.  The build is succeeding for me (on today's
commit 5c8031f...).

The configure.log should help.  The last error reported is the one to
look for.  For me, the path to a previous failed-build log file was:

/tmp/guix-build-petsc-mumps-openmpi-3.9.3.drv-6/petsc-3.9.3/arch-
linux2-c-debug/lib/petsc/conf/configure.log

My build commands are:

$ ./pre-inst-env guix environment --pure petsc-mumps-openmpi --ad-hoc
guile-gcrypt guile-git 

$ GUILE_LOAD_PATH=$GUIX_ENVIRONMENT/share/guile/site/2.2 ./pre-inst-env 
guix build -K petsc-mumps-openmpi

Can you see in your configure.log a clue, such as 'undefined
reference', near to the end?  

Best regards,

Paul.

[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[Eric Bavier]

[-- Attachment #1: Type: text/plain, Size: 3089 bytes --]

On Mon, 24 Sep 2018 11:32:44 +0100
Paul Garlick <pgarlick@tourbillion-technology.com> wrote:

> * gnu/packages/maths.scm (petsc-mumps-openmpi): New variable.
> ---
>  gnu/packages/maths.scm | 42 ++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 42 insertions(+)
> 
> diff --git a/gnu/packages/maths.scm b/gnu/packages/maths.scm
> index 736bac1..b2fff54 100644
> --- a/gnu/packages/maths.scm
> +++ b/gnu/packages/maths.scm
> @@ -1724,6 +1724,48 @@ scientific applications modeled by partial differential equations.")
>             ,@(delete "--with-mpi=0" ,cf)))))
>      (synopsis "Library to solve PDEs (with complex scalars and MPI support)")))
>  
> +(define-public petsc-mumps-openmpi
> +  (package (inherit petsc-openmpi)
> +    (name "petsc-mumps-openmpi")
> +    (inputs
> +     `(("metis" ,metis)
> +       ("mumps" ,mumps-openmpi)
> +       ("scalapack" ,scalapack)
> +       ("scotch" ,pt-scotch)
> +       ,@(package-inputs petsc-openmpi)))

I don't see why we'd add this as a new package, rather than simply
adjust the capabilities/inputs of the "petsc-openmpi" package.

> +    (arguments
> +     (substitute-keyword-arguments (package-arguments petsc-openmpi)
> +       ((#:configure-flags cf)
> +        ``(,(string-append "--with-metis-include="
> +                           (assoc-ref %build-inputs "metis") "/include")
> +           ,(string-append "--with-metis-lib="
> +                           (assoc-ref %build-inputs "metis") "/lib/"
> +                           "libmetis.so")
> +           ,(string-append "--with-mumps-include="
> +                           (assoc-ref %build-inputs "mumps") "/include")
> +           ,(string-append "--with-mumps-lib=["
> +                           (assoc-ref %build-inputs "mumps") "/lib/"
> +                           "libdmumps.a,"
> +                           "libzmumps.a,"
> +                           "libsmumps.a,"
> +                           "libcmumps.a,"
> +                           "libmumps_common.a,"
> +                           "libpord.a]")
> +           ,(string-append "--with-scalapack-lib="
> +                           (assoc-ref %build-inputs "scalapack") "/lib/"
> +                           "libscalapack.so")
> +           ,(string-append "--with-ptscotch-include="
> +                           (assoc-ref %build-inputs "scotch") "/include")
> +           ,(string-append "--with-ptscotch-lib=["
> +                           (assoc-ref %build-inputs "scotch") "/lib/"
> +                           "libesmumps.a,"
> +                           "libscotch.a,"
> +                           "libscotcherr.a,"
> +                           "libptesmumps.a,"
> +                           "libptscotch.a,"
> +                           "libptscotcherr.a]")

I believe PETSc's configure script will handle much of the flags on its
own if it's provided with "--with-metis", "--with-scotch",
"--with-mumps", "--with-scalapack" flags.

I have not tried building this yet, but I will give it a try later.

`~Eric

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[Paul Garlick]

Hi Eric,

> I don't see why we'd add this as a new package, rather than simply
> adjust the capabilities/inputs of the "petsc-openmpi" package.

Yes, that was my Plan A.  However, I spotted that the package dealii-
openmpi has petsc-openmpi as an input but also mumps-metis-openmpi,
instead of mumps-openmpi.  If the mumps configuration in petsc-mumps-
openmpi is included in petsc-openmpi then dealii-openmpi may not build.
 WDYT?

> I believe PETSc's configure script will handle much of the flags on
> its
> own if it's provided with "--with-metis", "--with-scotch",
> "--with-mumps", "--with-scalapack" flags.

This would be handy for sure.  I will give it a try.

Best regards,

Paul.

[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[Paul Garlick]

Hi Eric,

>  I will give it a try.

The build is successful with the "--with-PACKAGENAME=1" flags.  The
PETSc configure script picks up the same set of libraries that I had
specified by the "--with-PACKAGENAME-lib" flags.  Thank you for the
tip!

Actually, I would like to add a "--with-hypre=1" flag too to add in
support for the HYPRE preconditioner.  However, this is related to the
petsc-mumps-openmpi vs petsc-openmpi naming question.  As far as I can
see the presence of HYPRE would not interfere with dealii-openmpi so
perhaps it would be better to include it in petsc-openmpi instead.

Paul.

[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[Eric Bavier]

[-- Attachment #1: Type: text/plain, Size: 990 bytes --]

On Tue, 25 Sep 2018 14:40:52 +0100
Paul Garlick <pgarlick@tourbillion-technology.com> wrote:

> Hi Eric,
> 
> > I don't see why we'd add this as a new package, rather than simply
> > adjust the capabilities/inputs of the "petsc-openmpi" package.  
> 
> Yes, that was my Plan A.  However, I spotted that the package dealii-
> openmpi has petsc-openmpi as an input but also mumps-metis-openmpi,
> instead of mumps-openmpi.  If the mumps configuration in petsc-mumps-
> openmpi is included in petsc-openmpi then dealii-openmpi may not build.
>  WDYT?

Good point.  I'm not sure how dealii would react to that.  It's also
possible that the state of affairs has changed in more recent versions
of dealii (also on my todo list).  I suppose we'd need to give it a try
to know.  If it does fail, I'd say a patch to dealii is called for,
since the presense of scotch ordering capability in MUMPS should not
prevent dealii from using the metis orderings in any way.

`~Eric

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[bug#32818] [PATCH] gnu: Add petsc-mumps-openmpi.

[Paul Garlick]

Hi Eric,

> Good point.  I'm not sure how dealii would react to that.  It's also
> possible that the state of affairs has changed in more recent
> versions
> of dealii (also on my todo list).  I suppose we'd need to give it a
> try
> to know.  If it does fail, I'd say a patch to dealii is called for,
> since the presense of scotch ordering capability in MUMPS should not
> prevent dealii from using the metis orderings in any way.
> 

OK, let's go for the 'add-mumps-to-petsc-openmpi' plan.  

I shall re-submit this patch (#32818) to add MUMPS support to the
existing petsc-openmpi package.

Best regards,

Paul.

[bug#32818] [PATCH] gnu: petsc-openmpi: Configure with support for MUMPS solver.

[Paul Garlick]

* gnu/packages/maths.scm (petsc-openmpi)[inputs]: Add metis,
mumps-openmpi, scalapack and pt-scotch.
[arguments]: Add "--with-PACKAGENAME=1" flags.  Move 'mpi-setup' phase
ahead of 'configure' phase.
---
 gnu/packages/maths.scm | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/gnu/packages/maths.scm b/gnu/packages/maths.scm
index 636b485..6adddc0 100644
--- a/gnu/packages/maths.scm
+++ b/gnu/packages/maths.scm
@@ -1695,13 +1695,21 @@ scientific applications modeled by partial differential equations.")
   (package (inherit petsc)
     (name "petsc-openmpi")
     (inputs
-     `(("openmpi" ,openmpi)
-       ("hdf5" ,hdf5-parallel-openmpi)
+     `(("hdf5" ,hdf5-parallel-openmpi)
+       ("metis" ,metis)
+       ("mumps" ,mumps-openmpi)
+       ("openmpi" ,openmpi)
+       ("scalapack" ,scalapack)
+       ("scotch" ,pt-scotch)
        ,@(package-inputs petsc)))
     (arguments
      (substitute-keyword-arguments (package-arguments petsc)
        ((#:configure-flags cf)
         ``("--with-mpiexec=mpirun"
+           "--with-metis=1"
+           "--with-mumps=1"
+           "--with-scalapack=1"
+           "--with-ptscotch=1"
            ,(string-append "--with-mpi-dir="
                            (assoc-ref %build-inputs "openmpi"))
            ,(string-append "--with-hdf5-include="
@@ -1711,9 +1719,9 @@ scientific applications modeled by partial differential equations.")
            ,@(delete "--with-mpi=0" ,cf)))
        ((#:phases phases)
         `(modify-phases ,phases
-           (add-before 'check 'mpi-setup
+           (add-before 'configure 'mpi-setup
              ,%openmpi-setup)))))
-    (synopsis "Library to solve PDEs (with MPI support)")))
+    (synopsis "Library to solve PDEs (with MUMPS and MPI support)")))
 
 (define-public petsc-complex-openmpi
   (package (inherit petsc-complex)
@@ -1730,7 +1738,6 @@ scientific applications modeled by partial differential equations.")
            ,@(delete "--with-mpi=0" ,cf)))))
     (synopsis "Library to solve PDEs (with complex scalars and MPI support)")))
 
-
 (define-public python-kiwisolver
   (package
     (name "python-kiwisolver")
-- 
1.8.3.1

[bug#32818] [PATCH] gnu: petsc-openmpi: Configure with support for MUMPS solver.

[Paul Garlick]

Hi Guix,

Are there any comments on the revised patch? 

I  have re-checked that it builds successfully.  It does.

Best regards,

Paul.

bug#32818: [PATCH] gnu: petsc-openmpi: Configure with support for MUMPS solver.

[Leo Famulari]

[-- Attachment #1: Type: text/plain, Size: 258 bytes --]

On Thu, Oct 04, 2018 at 02:53:16PM +0100, Paul Garlick wrote:
> Hi Guix,
> 
> Are there any comments on the revised patch? 
> 
> I  have re-checked that it builds successfully.  It does.

I pushed as 9e2aa70bad8c06cdf9a44065a31f12b3205e92e0. Thanks!

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]