Charm++ parallel programming system

[pellegrini <pellegrini AT ill.fr>]

Hi Abhimav,

I tried to rebuild charm++ by removing the ibverbs flag using this time 
the following instructions:

1 cd charm-6.1.3
2 env MPICXX=mpicxx ./build charm++ net-linux-x86_64 pgcc
--basedir=/usr/pgi/linux86-64/2010/mpi/mvapich --no-build-shared pgf90
--no-build-shared -j32 -O2 -DCMK_OTIMIZE
3 cd tests/charm++/megatests
4 make clean
5 make pgm
6 charmrun ++local ++verbose ./pgm

now we get the following error:

master charm++/megatest> charmrun ++local ++verbose ./pgm
Charmrun> charmrun started...
Charmrun> adding client 0: "127.0.0.1", IP:127.0.0.1
Charmrun> Charmrun = 127.0.0.1, port = 59250
Charmrun> IBVERBS version of charmrun
Charmrun> start 0 node program on localhost.
Charmrun> node programs all started
Charmrun> Waiting for 0-th client to connect.
charmrun: charmrun.c:1958: req_one_client_partinit: Assertion 
`(__extension__ (__builtin_constant_p (8) && ((__builtin_constant_p 
(partStartMsg.header.type) && strlen (partStartMsg.header.type) < 
((size_t) (8))) || (__builtin_constant_p ("partinit") && strlen 
("partinit") < ((size_t) (8)))) ? __extension__ ({ size_t __s1_len, 
__s2_len; (__builtin_constant_p (partStartMsg.header.type) && 
__builtin_constant_p ("partinit") && (__s1_len = strlen 
(partStartMsg.header.type), __s2_len = strlen ("partinit"), 
(!((size_t)(const void *)((partStartMsg.header.type) + 1) - 
(size_t)(const void *)(partStartMsg.header.type) == 1) || __s1_len >= 4) 
&& (!((size_t)(const void *)(("partinit") + 1) - (size_t)(const void 
*)("partinit") == 1) || __s2_len >= 4)) ? __builtin_strcmp 
(partStartMsg.header.type, "partinit") : (__builtin_constant_p 
(partStartMsg.header.type) && ((size_t)(const void 
*)((partStartMsg.header.type) + 1) - (size_t)(const void 
*)(partStartMsg.header.type) == 1) && (__s1_len = strlen 
(partStartMsg.header.type), __s1_len < 4) ? (__builtin_constant_p 
("partinit") && ((size_t)(const void *)(("partinit") + 1) - 
(size_t)(const void *)("partinit") == 1) ? __builtin_strcmp 
(partStartMsg.header.type, "partinit") : (__extension__ ({ __const 
unsigned char *__s2 = (__const unsigned char *) (__const char *) 
("partinit"); register int __result = (((__const unsigned char *) 
(__const char *) (partStartMsg.header.type))[0] - __s2[0]); if (__s1_len 
> 0 && __result == 0) { __result = (((__const unsigned char *) (__const 
char *) (partStartMsg.header.type))[1] - __s2[1]); if (__s1_len > 1 && 
__result == 0) { __result = (((__const unsigned char *) (__const char *) 
(partStartMsg.header.type))[2] - __s2[2]); if (__s1_len > 2 && __result 
== 0) __result = (((__const unsigned char *) (__const char *) 
(partStartMsg.header.type))[3] - __s2[3]); } } __result; }))) : 
(__builtin_constant_p ("partinit") && ((size_t)(const void 
*)(("partinit") + 1) - (size_t)(const void *)("partinit") == 1) && 
(__s2_len = strlen ("partinit"), __s2_len < 4) ? (__builtin_constant_p 
(partStartMsg.header.type) && ((size_t)(const void 
*)((partStartMsg.header.type) + 1) - (size_t)(const void 
*)(partStartMsg.header.type) == 1) ? __builtin_strcmp 
(partStartMsg.header.type, "partinit") : (__extension__ ({ __const 
unsigned char *__s1 = (__const unsigned char *) (__const char *) 
(partStartMsg.header.type); register int __result = __s1[0] - ((__const 
unsigned char *) (__const char *) ("partinit"))[0]; if (__s2_len > 0 && 
__result == 0) { __result = (__s1[1] - ((__const unsigned char *) 
(__const char *) ("partinit"))[1]); if (__s2_len > 1 && __result == 0) { 
__result = (__s1[2] - ((__const unsigned char *) (__const char *) 
("partinit"))[2]); if (__s2_len > 2 && __result == 0) __result = 
(__s1[3] - ((__const unsigned char *) (__const char *) 
("partinit"))[3]); } } __result; }))) : __builtin_strcmp 
(partStartMsg.header.type, "partinit")))); }) : strncmp 
(partStartMsg.header.type, "partinit", 8))) == 0' failed.
Abort

does it tell you something ?

thank you very much for your support

Eric



Abhinav S Bhatele a écrit :
> Hello Eric,
>
> The ibverbs issue you pointed out might take some time before it gets 
> resolved.
>
> In the meantime, if you want to get NAMD up and running, I would 
> recommend removing
> ibverbs from your build line (at step 2) and compiling afresh.
>
> - Abhinav
>
>
> On Thu, Jun 10, 2010 at 11:11 AM, pellegrini <pellegrini AT ill.fr 
> <mailto:pellegrini AT ill.fr>> wrote:
>
>     Hello,
>
>     we are trying to install NAMD on our new DALCO cluster made of 43
>     8-cores Intel Xeon cpus (Nehalem) connected through InfiniBand. We
>     followed the installation notes but we fails to make charm++ work. The
>     build was successful but when performing some tests we got no results.
>     We are using charm-6.1.3 that comes along
>     with NAMD-2.7b2. We did the following from the NAMD directory:
>
>     1 cd charm-6.1.3
>     2 env MPICXX=mpicxx ./build charm++ net-linux-x86_64 pgcc
>     --basedir=/usr/pgi/linux86-64/2010/mpi/mvapich --no-build-shared pgf90
>     ibverbs --no-build-shared -j32 -O2 -DCMK_OTIMIZE --> the build was
>     successful
>     3 cd tests/charm++/megatests
>     4 make clean
>     5 make pgm --> the make was successfull
>     6 charmrun ++local +p4 ./pgm and from here we get:
>
>        Charmrun> IBVERBS version of charmrun
>        Charmrun: Bad initnode data length. Aborting
>
>     Based on this, the build of NAMD works but once again the tests are
>     failing. We think that both failures are related.
>
>     Would you have any idea of what is going wrong or if we did some
>     mistakes when building charm++ ?
>
>     thank you very much
>
>     best regards
>
>     Eric Pellegrini
>
>
>
>
>     _______________________________________________
>     charm mailing list
>     
> charm AT cs.uiuc.edu
>  
> <mailto:charm AT cs.uiuc.edu>
>     http://lists.cs.uiuc.edu/mailman/listinfo/charm
>
>
>
>
> --
> Abhinav S Bhatele
> My work: http://charm.cs.illinois.edu/~bhatele/phd 
> <http://charm.cs.illinois.edu/%7Ebhatele/phd>