1 |
On 13/06/08 14:29 -0700, Bryan Green wrote: |
2 |
> Bryan Green writes: |
3 |
> > Hello, |
4 |
> > |
5 |
> > Now that I am using empi/eselect-mpi, I would like to install boost-1.35.0 |
6 |
> > with its MPI library (it has an 'mpi' use flag). |
7 |
> > When I try to emerge it, it trys to pull in sys-cluster/openmpi as a |
8 |
> > dependency. I tried adding sys-cluster/openmpi to |
9 |
> > /etc/portage/package.provided, but for some reason that does not work. |
10 |
> > |
11 |
> > Any suggestions on how to get boost-mpi to build against my eselected |
12 |
> > MPI? I also wonder if it would really be necessary build a version of boost |
13 |
> > for each MPI package installed, using 'empi --add ...'. That seems like |
14 |
> > overkill. |
15 |
> |
16 |
> Having added openmpi to /etc/portage/profile/package.provided, |
17 |
> I can emerge boost with mpi, but the mpi autodetect fails during the |
18 |
> compilation with this message: |
19 |
> |
20 |
> MPI auto-detection failed: unknown wrapper compiler mpic++ |
21 |
> |
22 |
> The same occurs if I run 'ebuild ... compile'. |
23 |
> But if I run the same command as root from the resulting /var/tmp/.../work |
24 |
> directory, the detection is successful. |
25 |
> |
26 |
> Any ideas what might be going on here? What part of the ebuild environment |
27 |
> changes such that it does not see mpic++ anymore? |
28 |
> |
29 |
> -bryan |
30 |
> |
31 |
|
32 |
I haven't looked at this, but if you only have openmpi as installed by empi, |
33 |
then that's the missing part. The emerge environment isn't getting the |
34 |
variables set with eselect-mpi. |
35 |
|
36 |
Also, if boost is actually linking against mpi, then you will need to get an |
37 |
mpi.eclass enabled ebuild to use it, which does mean building it for each |
38 |
implementation you have installed. |
39 |
|
40 |
I can take a look at the ebuild this weekend or early next week, and |
41 |
hopefully a port wouldn't be too much work. |
42 |
|
43 |
-- |
44 |
Justin Bronder |