1 |
Alexey Shvetsov wrote: |
2 |
> hi all! |
3 |
> |
4 |
> May be it will be better if we can choose combinations mpi_imp/compiler |
5 |
> for example |
6 |
> mpi-openmpi-gcc |
7 |
> mpi-openmpi-intel |
8 |
> and so on |
9 |
> |
10 |
> for now it can be done manualy specifing CC, Fc and so on env |
11 |
> may be it will be better if it will be done from empi? |
12 |
> |
13 |
You might want to go read: http://bugs.gentoo.org/show_bug.cgi?id=44132 |
14 |
|
15 |
> 2008/6/16 Eric Thibodeau <kyron@××××××××.com>: |
16 |
> |
17 |
>> Justin Bronder wrote: |
18 |
>> |
19 |
>> On 12/06/08 11:28 -0700, Bryan Green wrote: |
20 |
>> |
21 |
>> |
22 |
>> Hello, |
23 |
>> |
24 |
>> I'm beginning to give empi/eselect-mpi a try. Bravo, Justin, for making |
25 |
>> these! |
26 |
>> |
27 |
>> |
28 |
>> Thanks for trying them out, I'm hoping someone else finds them as useful as |
29 |
>> I |
30 |
>> have. |
31 |
>> |
32 |
>> |
33 |
>> It is planned to be part of the Gentoo Clustering LiveCD ;) |
34 |
>> |
35 |
>> |
36 |
>> |
37 |
>> I've run into one bug in eselect-mpi that effects csh users (a popular |
38 |
>> shell where I work). There is simply one setenv line that has bash syntax |
39 |
>> rather than csh syntax. I already submitted a bug: its bug 226105. |
40 |
>> Here is the patch to fix it: |
41 |
>> |
42 |
>> ================================================ |
43 |
>> --- files/mpi.eselect-0.0.3 (revision 1137) |
44 |
>> +++ files/mpi.eselect-0.0.3 (working copy) |
45 |
>> @@ -161,7 +161,7 @@ |
46 |
>> setenv PATH "${binpath}" |
47 |
>> setenv MANPATH "${manpath}" |
48 |
>> setenv LD_LIBRARY_PATH "${lld}" |
49 |
>> -setenv ESELECT_MPI_IMP="${1}" |
50 |
>> +setenv ESELECT_MPI_IMP "${1}" |
51 |
>> EOF |
52 |
>> |
53 |
>> echo "Remember to source ${user_ev_sh} or ${user_ev_csh}" |
54 |
>> ================================================ |
55 |
>> |
56 |
>> |
57 |
>> Fixed in eselect-0.0.3-r1. Also found another mistake when setting PATH for |
58 |
>> c shells and fixed that as well. |
59 |
>> |
60 |
>> |
61 |
>> |
62 |
>> Also, in the README file, the for loop in step 6 is incorrect: |
63 |
>> |
64 |
>> for i in $(ls ${HOME}/.env.d/*); do |
65 |
>> source ${HOME}/.env.d/${i} |
66 |
>> done |
67 |
>> |
68 |
>> Results in: |
69 |
>> |
70 |
>> -bash: /home/bgreen/.env.d//home/bgreen/.env.d/mpi.csh: No such file or |
71 |
>> directory |
72 |
>> -bash: /home/bgreen/.env.d//home/bgreen/.env.d/mpi.sh: No such file or |
73 |
>> directoryA |
74 |
>> |
75 |
>> |
76 |
>> Also fixed. You might also want to check out the doc I've been working on |
77 |
>> lately, http://dev.gentoo.org/~jsbronder/empi.xml. Hopefully it's a little |
78 |
>> more up to date. |
79 |
>> |
80 |
>> |
81 |
>> Have to remember that one ;) |
82 |
>> |
83 |
>> |
84 |
>> |
85 |
>> On the system where I'm using empi, I've created files in /etc/profile.d to |
86 |
>> source the user's .env.d files when they log in. What do you think about |
87 |
>> having the eselect-mpi ebuild install these, so mpi users dont have to do |
88 |
>> that part manually? They would instead just have to source /etc/profile or |
89 |
>> /etc/csh.login after running 'eselect mpi'. |
90 |
>> |
91 |
>> |
92 |
>> Love it, I had not even thought of using this mechanism before even though I |
93 |
>> should have. It will also simplify switching from an mpi environment to one |
94 |
>> without it. I'll try to get something checked into the overlay soon, |
95 |
>> shouldn't be more than a few days.< |
96 |
>> |
97 |
>> |
98 |
>> Keep us posted because I will definitely want that version! |
99 |
>> |
100 |
>> |
101 |
>> |
102 |
>> Here are my versions of the files in /etc/profile.d: |
103 |
>> |
104 |
>> ============== |
105 |
>> mpi-config.sh |
106 |
>> ============== |
107 |
>> if [ -d ${HOME}/.env.d ]; then |
108 |
>> for i in ${HOME}/.env.d/*.sh ; do |
109 |
>> . "${i}" |
110 |
>> done |
111 |
>> unset i |
112 |
>> fi |
113 |
>> |
114 |
>> ============== |
115 |
>> mpi-config.csh |
116 |
>> ============== |
117 |
>> if ( -d ${HOME}/.env.d ) then |
118 |
>> set nonomatch |
119 |
>> foreach i ( ${HOME}/.env.d/*.csh ) |
120 |
>> source ${i} |
121 |
>> end |
122 |
>> unset i nonomatch |
123 |
>> endif |
124 |
>> |
125 |
>> ============== |
126 |
>> |
127 |
>> -bryan |
128 |
>> |
129 |
>> |
130 |
>> Thanks again for testing and your comments, they're much appreciated. |
131 |
>> |
132 |
>> |
133 |
>> Yeah, thanks you both, multi-mpi-on-1-system has always plagued me too ;) |
134 |
>> |
135 |
>> |
136 |
> |
137 |
> |
138 |
> |
139 |
> |