How to install MAGMA on a login node

[shmcu]

Hi guys,

I am trying to install MAGMA on a login node of a cluster. The login node does not have physical GPU card attached, it just has CUDA module installed. I can compile cuda codes on the login node and run the cuda programs on the cluster nodes with real GPU hardware. The problem is when I tried to install MAGMA, I typed:

Code: Select all

make

it failed and reported the following error:

Code: Select all

[hsu335@dlxlogin2 magma_1.1.0]$ make
( cd control        && make )
make[1]: Entering directory `/home/hsu224/MAGMA/magma_1.1.0/control'
icc -O3 -DADD_ -DGPUSHMEM=130 -I/usr/share/Modules/modulefiles/cuda/include -I../include -I../quark/include -c magma_zf77.cpp -o magma_zf77.o
magma_zf77.cpp(21): catastrophic error: could not open source file "/usr/share/Modules/modulefiles/cuda/4.0.17/include/cuda_runtime_api.h"
  #include </usr/share/Modules/modulefiles/cuda/4.0.17/include/cuda_runtime_api.h>
                                                                                  ^

compilation aborted for magma_zf77.cpp (code 4)
make[1]: *** [magma_zf77.o] Error 4
make[1]: Leaving directory `/home/hsu224/MAGMA/magma_1.1.0/control'
make: *** [libmagma] Error 2

I guess the file "cuda_runtime_api.h" is under the installation directory of CUDA, but I am not authorized to access it. I can search it, but the system said something like this:

Code: Select all

find / -name cuda_runtime_api.h
find: /panfs/pan01/home/aalex2/.nv: Permission denied
find: /panfs/pan01/home/aalex2/.lsbatch: Permission denied ...
...

Please help ! Thanks!

[mgates3]

Yes, that header is in the CUDA install, for instance,
/usr/local/cuda/include/cuda_runtime_api.h

Based on your compile line, I would expect that in
/usr/share/Modules/modulefiles/cuda/include/cuda_runtime_api.h

If you can't read that directory, you need to consult with your system administrator. Check permissions with ls:
ls -ld /usr/share/Modules/modulefiles/cuda
ls -ld /usr/share/Modules/modulefiles/cuda/include
ls -ld /usr/share/Modules/modulefiles/cuda/include/cuda_runtime_api.h

The files are usually world searchable and readable, as below:
> ls -ld /usr/local/cuda
drwxr-xr-x 9 root wheel 306B May 13 2011 /usr/local/cuda/
> ls -ld /usr/local/cuda/include
drwxr-xr-x 54 root wheel 1.8K Sep 7 2011 /usr/local/cuda/include/
> ls -ld /usr/local/cuda/include/cuda_r
-rw-r--r-- 1 root wheel 175K May 13 2011 /usr/local/cuda/include/cuda_runtime_api.h

-mark