Chinaunix

标题: 谁有mpich-gm啊？ [打印本页]
作者: 我爱钓鱼    时间: 2006-06-14 07:28
标题: 谁有mpich-gm啊？
官方网站上下载不了。。。要注册，并一定要有Myrinet的产品才给用户名和密码，是不是的确必须要有他们的产品才能用这个软件呢？
在el4上装mpich2(mpich2-1.0.3.tar.gz)和mpich1(mpich-1.2.7p1)都能顺利安装，但运行
mpif77时提示
  1. /usr/lib/gcc/i386-redhat-linux/3.4.3/libfrtbegin.a(frtbegin.o)(.text+0x35): In function `main':
  2. : undefined reference to `MAIN__'
  3. collect2: ld returned 1 exit status
复制代码

运行mpicc时提示
  1. mpicc
  2. /usr/lib/gcc/i386-redhat-linux/3.4.3/../../../crt1.o(.text+0x18): In function `_start':
  3. : undefined reference to `main'
  4. collect2: ld returned 1 exit status
复制代码

google了一下，说是gcc的版本太新。。。但看到正在运行的曙光集群中mpif77用的是mpich-1.2.5/ch_p4/gm-pgi/mpif77，而不是直接用的mpich的mpif77，看来可能是mpich的问题，但没有下载到mpich-gm，哪位哥哥有的话，可否共享一个，让我测试一下？
或者有经验的高手指点一下，gcc的版本应该用哪个比较合适一点儿？
BTW：今天我会测试redhat linux 8.0和mpich-1.2.7p1，然后把结果给大家。。。
作者: 我爱钓鱼    时间: 2006-06-14 07:29
BTW:mpich的官方网站也没有提到这个问题，也没有说他们对应的gcc版本号。。。郁闷。。。是不是我没有找到？今天再认真的看看官方网站。。。
作者: 我爱钓鱼    时间: 2006-06-14 14:24
看来的确是gcc版本的问题....在rh8和rh7.3上测试成功....
作者: nntp    时间: 2006-06-15 02:14
I suggest you to choose RHEL WS and don't waste your time on those version conflict issues. It doesn't make sense.

Simply download RHEL WS from redhat's official website and run your mpi environment upon . This will  pull everybody there back to the computing research, not research infrastructure itself.

good luck,
作者: 我爱钓鱼    时间: 2006-06-15 09:54
原帖由 nntp 于 2006-6-15 02:14 发表
I suggest you to choose RHEL WS and don't waste your time on those version conflict issues. It doesn't make sense.

Simply download RHEL WS from redhat's official website and run your mpi environ ...


Thank you very much,but by the way,why not as/es? I try it on as4,but i'm failed.... I'll try it on ws soon,thanks again...
作者: 我爱钓鱼    时间: 2006-06-15 12:54
I try it on gcc version 3.2.3 20030502 (Red Hat Linux 3.2.3-42) with 1.27p1,and successful.But there still a warinnig:
mpif77
  1. /usr/lib/gcc-lib/i386-redhat-linux/3.2.3/libfrtbegin.a(frtbegin.o)(.text+0x32): In function `main':
  2. : undefined reference to `MAIN__'
  3. collect2: ld returned 1 exit status
复制代码

But f77 can use now.
  1. [root@As3test examples]# /usr/local/mpich1//bin/mpif77 -o pi3 pi3.o
  2. [root@As3test examples]# ll
  3. total 1148
  4. -rw-r--r--    1 root     root         1640 Jun 15 12:09 cpi.c
  5. -rwxr-xr-x    1 root     root       353848 Jun 15 12:12 cpilog
  6. -rw-r--r--    1 root     root         2721 Jun 15 12:09 cpilog.c
  7. -rw-r--r--    1 root     root        65536 Jun 15 12:13 cpilog.clog
  8. -rw-r--r--    1 root     root         2992 Jun 15 12:12 cpilog.o
  9. -rw-r--r--    1 root     root         1650 Jun 15 12:09 cpip.c
  10. -rw-r--r--    1 root     root         1547 Jun 15 12:09 hello++.cc
  11. -rw-r--r--    1 root     root         2803 Jun 15 12:09 Makefile
  12. -rw-r--r--    1 root     root         2748 Jun 15 12:09 Makefile.in
  13. drwxr-xr-x    2 root     root         4096 Jun 15 12:09 MPI-2-C++
  14. lrwxrwxrwx    1 root     root           13 Jun 15 12:09 mpirun -> ../bin/mpirun
  15. -rwxr-xr-x    1 root     root       337519 Jun 15 12:11 pi3
  16. -rw-r--r--    1 root     root         2335 Jun 15 12:09 pi3.f
  17. -rw-r--r--    1 root     root         3232 Jun 15 12:11 pi3.o
  18. -rwxr-xr-x    1 root     root       337520 Jun 15 12:12 pi3p
  19. -rw-r--r--    1 root     root         2341 Jun 15 12:09 pi3p.f
  20. -rw-r--r--    1 root     root         3236 Jun 15 12:12 pi3p.o
  21. -rw-r--r--    1 root     root          993 Jun 15 12:09 README
  22. -rw-r--r--    1 root     root         2658 Jun 15 12:09 simpleio.c

  24. [root@As3test examples]# ./pi3 1
  25. Process  0 of  1 is alive
  26. Enter the number of intervals: (0 quits)
  27. 1
  28.   pi is approximately: 3.2000000000000002  Error is: 0.0584073464102071
  29. Enter the number of intervals: (0 quits)
  30. 2
  31.   pi is approximately: 3.1623529411764704  Error is: 0.0207602875866773
  32. Enter the number of intervals: (0 quits)
  33. 3
  34.   pi is approximately: 3.1508492098656036  Error is: 0.0092565562758105
  35. Enter the number of intervals: (0 quits)
  36. 4
  37.   pi is approximately: 3.1468005183939427  Error is: 0.0052078648041496
  38. Enter the number of intervals: (0 quits)
  39. 0
复制代码
作者: 我爱钓鱼    时间: 2006-06-15 13:03
BTW:I'm downloading ws3.7,And I'll get it soon...
作者: 我爱钓鱼    时间: 2006-06-15 14:58
I try it on Red Hat Enterprise Linux WS release 3 (Taroon Update 7),and get the same result  as on Red Hat Enterprise Linux AS release 3 (Taroon Update 3)
作者: nntp    时间: 2006-06-16 00:51
did you choose the backtowards develop packages when install the WS?  that is necessary options.
作者: 我爱钓鱼    时间: 2006-06-16 07:43
原帖由 nntp 于 2006-6-16 00:51 发表
did you choose the backtowards develop packages when install the WS?  that is necessary options.



backtowards ? I'll try it soon.
By the way,the application must use pgi too.But I have no pgi  software untill yesterday night. I'll try it today....
作者: nntp    时间: 2006-06-16 11:20
原帖由 我爱钓鱼 于 2006-6-16 07:43 发表



backtowards ? I'll try it soon.
By the way,the application must use pgi too.But I have no pgi  software untill yesterday night. I'll try it today....



hi buddy, PGI is commercial software, you've to pay for it.

EKO pathscale is the best and pgi is the second, if you have latest EKO pathscale(purchased), i can exchange with you.(ofcoz only for personal and research use only).  
作者: 我爱钓鱼    时间: 2006-06-16 11:35
原帖由 nntp 于 2006-6-16 11:20 发表



hi buddy, PGI is commercial software, you've to pay for it.

EKO pathscale is the best and pgi is the second, if you have latest EKO pathscale(purchased), i can exchange with you.(ofcoz onl ...

Thanks,some guys will pay for it...because there applications must use it... I downloading it now.And I can get a 15 days license,I'll try it soon...

[ 本帖最后由 我爱钓鱼 于 2006-6-20 07:45 编辑 ]
作者: 我爱钓鱼    时间: 2006-06-16 15:18
Fiant...I use pgi 6.1,but it still failed...   :em12::em12:
  1. mpif77
  2. /App/pgi/linux86/6.1/lib/pgfmain.o(.text+0x69): In function `main':
  3. : undefined reference to `MAIN_'
复制代码
作者: nntp    时间: 2006-06-17 08:27
原帖由 我爱钓鱼 于 2006-6-16 07:43 发表



backtowards ? I'll try it soon.
By the way,the application must use pgi too.But I have no pgi  software untill yesterday night. I'll try it today....



Sorry, the official naming is " Legacy develop packages" during the installation of your RHEL WS.

i do think you should check your development environment and put it to the higher priority. I never have such kind issues after deployed several hpc clusters for my customers in China.

Good Luck,
作者: 我爱钓鱼    时间: 2006-06-18 15:41
原帖由 nntp 于 2006-6-17 08:27 发表



Sorry, the official naming is " Legacy develop packages" during the installation of your RHEL WS.

i do think you should check your development environment and put it to the highe ...



thanks again... Now the cluster is working....but there's only one cpu can be work,I'll try again....
作者: nntp    时间: 2006-06-19 06:33
原帖由 我爱钓鱼 于 2006-6-18 15:41 发表



thanks again... Now the cluster is working....but there's only one cpu can be work,I'll try again....


It depends on your kernel version,  pay attention to kernel-EL, kernel-SMP, kernel-hugemem as well.

cat /proc/cpuinfo from one compute node , that is the first place for you to confirm your processor status.
作者: 我爱钓鱼    时间: 2006-06-19 09:18
原帖由 nntp 于 2006-6-19 06:33 发表


It depends on your kernel version,  pay attention to kernel-EL, kernel-SMP, kernel-hugemem as well.

cat /proc/cpuinfo from one compute node , that is the first place for you to confirm your  ...


Thanks,kernel is ok.Because I can see four cpu is working. I will try it again after a while.
作者: 我爱钓鱼    时间: 2006-06-19 13:57
BTW:It's duble kernel and duble cpu what I'm useing,The cpu's  model is AMD operton 270,Now mpi can use two cpus,but can't use four cpus,maybe mpi can't know the new cpu modle,so....
作者: nntp    时间: 2006-06-19 19:34
原帖由 我爱钓鱼 于 2006-6-19 13:57 发表
BTW:It's duble kernel and duble cpu what I'm useing,The cpu's  model is AMD operton 270,Now mpi can use two cpus,but can't use four cpus,maybe mpi can't know the new cpu modle,so....



No, MPI doesn't know what type of processor you're exactly running. OS handles SMP or multi-core processor. Ofcoz the compiler will perform some optimization if you lab equip latest dual core system.
i've compiled and executed MPI based parallel codes upon dual core IA64(Itanium2) cluster and that's cool.
By the way, if you don't have 4-way(dual core) or up standing system, disable NUMA support by sending parameters to your kernel boot options. And SuSE has better performance than Redhat or/centOS on AMD Opteron system.

Good luck,

[ 本帖最后由 nntp 于 2006-6-19 19:37 编辑 ]
作者: 我爱钓鱼    时间: 2006-06-20 07:43
原帖由 nntp 于 2006-6-19 19:34 发表



No, MPI doesn't know what type of processor you're exactly running. OS handles SMP or multi-core processor. Ofcoz the compiler will perform some optimization if you lab equip latest dual core ...



yeah,I agree with you.Because there is a YSE certification for AMD .Maybe I can try use suse.

There is a good news:Every thing is work NOW!The problem is mpirun must use rsh/ssh even on local computer.I haven't install rsh-server on it and I haven't configure mpi with ssh either,so there's something wrong.Yesterday night,I install rsh-server on the node,every thing is ok!

BTW:pgi 6.x have a tools for AMD 64 too,It's working very well.

Thanks for your advice!
作者: 我爱钓鱼    时间: 2006-06-20 07:55
原帖由 nntp 于 2006-6-17 08:27 发表



Sorry, the official naming is " Legacy develop packages" during the installation of your RHEL WS.

i do think you should check your development environment and put it to the highe ...


To tell you the truth,after installed it,I have the same error message.This packages have five rpms,but the develop packages have the similar packages.So I have the same error message.I think so,What do you think about it?

I readed a lot of document,And I found almost every guys use pgi/XLF(IBM) instead g77.Maybe there is something uncompatible between mpi and gcc.I don't know what's the truth,I guess so.

[ 本帖最后由 我爱钓鱼 于 2006-6-20 07:57 编辑 ]



欢迎光临 Chinaunix (http://bbs.chinaunix.net/) Powered by Discuz! X3.2