I know there are many tips online about installing mpich2 in linux/ubuntu, but I just do not know how to add path so that each .c/.f90 file can see the 'mpif.h'.
I think I have successfully installed mpich2 on my machine. Because when I run
mpiexec -np 3 ./cpi
It works. cpi is a application in the 'example' file. But when I tried to run gcc -o mpitest hellow.c. It always said undefined reference to MPI_INIT and some similar errors.
The mpiexec mpirun are in /usr/local/bin and mpif.h mpi.hare in /usr/local/include. So how to add path?
Thanks in advance!
You don't need to add a anything to your PATH. What you need to do is to link your application with the MPI libraries.
MPI installations provide a useful wrapper for this. Simply replace gcc by mpicc:
mpicc -o mpitest hellow.c
Related
First of all: I'm on linux mint 17.3 x64
What I've done so far:
Guide to install Open MPI 1.8
Guide to install MPI
Attemp to remove MPI executing: sudo apt-get install libcr-dev mpich2 mpich2-doc (Actually the should be not installed)
What I can see from terminal:
output of: echo $PATH
/path/to/mpj//bin:/home/timmy/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/home/timmy/.openmpi/bin
(I immagine that I've to remove /path/to/mpj/ (not exists) and /home/timmy/.openmpi/bin (I want to remove previous version of ompi))
output of: echo $LD_LIBRARY_PATH
(nothing)
Really, doesn't appear anything!
output of mpirun
--------------------------------------------------------------------------
mpirun could not find anything to do.
It is possible that you forgot to specify how many processes to run
via the "-np" argument.
--------------------------------------------------------------------------
Why I want to remove Open MPI and reinstall it
I've a project to do using both MPI and OpenMP and with the actual installation of MPI I cannot compile using the following command: mpicc -openmp "test_omp.c" -o "test_omp". It gives me the following error: Not defined function omp_get_thread_num(); and moreover, it ignore my #pragma commands.
Your problem is that you are giving the compiler the wrong option to enable the OpenMP support. -openmp is only understood by the (commercial) Intel compiler, which is probably the tool-set installed on the site you've referred to in your other question. Most Linux distributions come with GCC and one is to assume that mpicc will use GCC (check with mpicc -showme).
The option to enable OpenMP support in GCC is -fopenmp (notice the f).
I am installing Open MPI v1.8.8 with CUDA v7.5 on my Linux Debian.
I have tested CUDA and it works, tested OpenMPI and it works too. But when i try to combine them into a program, i meet an error: cannot find cuda.h file . This is my scenario:
My program source code include these .h file
include "cuda.h"
include "mpi.h"
I run command:
mpicc <filePath> -o test
And error appear:cuda.h: No such file or directory
#include "cuda.h"`
omp_info give me : mca:mpi:base:param:mpi_built_with_cuda_support:value:false
I have googled , and i followed some methods i found:
./configure --with-cuda
./configure --with-cuda=/usr/local/cuda-7.5
( source link : http://mirror.its.dal.ca/openmpi/faq/?category=buildcuda)
After that, i remake all , remake install Open Mpi. I run: mpicc or mpirun, the compiler give me error : mpirun error mca: base: component find: unable to open /usr/local/lib/openmpi/mca_mpool_sm
I set up soft link : ln -s /usr/local/cuda/include /usr/include ( describe in link : Building CUDA-aware openMPI on Ubuntu 12.04 cannot find cuda.h).
But it cannot fix my issue.
Does anyone successfully install it? Please help me or share your experience.
Thanks.
I think you are confusing installation problems with incorrect compiler options. It will be necessary to explicity specify the include paths, library paths, and libraries for CUDA when compiling and linking host code with your mpi wrapped host compiler.
Something like:
mpicc -I/usr/local/cuda-7.5/include -L/usr/local/cuda-7.5/lib -o test <filePath> -lcuda
would be the normal way to build a simple MPI program which call the cuda driver APIs. You will need to add nvcc compilation for device code and host code which uses the runtime API.
The apparent lack of CUDA support in your MPI flavour is a separate question and one you should probably take up in another forum (like the user mailing list of the MPI flavour you use).
I'm trying to compile Caffe (http://caffe.berkeleyvision.org/installation.html) and I get the following errors:
/usr/bin/ld: cannot find -lcblas
/usr/bin/ld: cannot find -latlas
However, I have these libraries installed (libatlas).
My LD_LIBRARY_PATH contains the path /usr/lib/atlas-base and it contains the files libcblas.so and libatlas.so (and some other files as well).
Why ld can't find these libraries?
Thanks.
tl;dr: Caffe makefile looks for libblas.so in /usr/lib. If missing, update-alternatives creates a symbolic link /usr/lib/libblas.so to the location where it is installed. Same applies to libcblas.so. LD_LIBRARY_PATH is for runtime, and doesn't have anything to do with this.
LD_LIBRARY_PATH doesn't really help you when compiling. It only provides directories to look for shared libraries when executing programs that rely on them, after they are compiled. Still, when linking during the compilation, the compiler needs to find these shared libraries, and does so by other means than LD_LIBRARY_PATH.
More to the point: if compiling with gcc or clang, the directories in which to look for libraries to link with are provided using the -L flag, and it does not consider the LD_LIBRARY_PATH environment variable.
Common locations for libblas.so are /usr/lib/atlas-base/ and /usr/lib/libblas/. The Makefile for caffe doesn't do anything particular to try and locate these subdirectories, and relies on these libraries being in the default library directory /usr/lib/. Typically a symbolic link /usr/lib/libblas.so exists, and points to the real location of the shared library. For some reason, this wasn't the case in your initial configuration.
When dealing with multiple alternatives for packages, update-alternatives comes in handy. In the case of libblas.so it let's you easily switch between multiple implementations (libblas, openblas) you might have installed, and does so by changing out the symbolic links.
sudo update-alternatives --config libblas.so created this symbolic link when it was missing, which in turn let the compiler find the shared library, solving your problem. This is indicated by the output of the command:
$ sudo update-alternatives --config libblas.so
There is only one alternative in link group libblas.so (providing /usr/lib/libblas.so): /usr/lib/libblas/libblas.so
Nothing to configure.
Same kind of reasoning applies to libcblas.so.
It turns out I had to run
sudo update-alternatives --config libblas.so
sudo update-alternatives --config liblapack.so
and to select libatlas .
I have no idea why,. If anyone can explain this me I will give him the answer.
Thanks.
sudo apt-get install libatlas-base-dev worked for me, it removed both missing dependencies.
See this thread for additional details https://github.com/BVLC/caffe/issues/559
As an addendum to #Ran's answer, Ubuntu in particular has an odd package structure for what's needed with Caffe. I just came across this post in fixing this same issue on my own machine, and here's some help if others are stuck. (Ubuntu 14.04).
libatlas-dev does NOT have libatlas-base-dev as a dependency! Caffe seems to like the libraries from the latter only. Install it.
Then, run the commands suggested by #Ran and select the libraries from the atlas-base directory under /usr/lib. With just libatlas-dev installed, update-alternatives will have the output at the bottom of #swalog's post, but does not actually link an atlas library that caffe seems to approve of! It needs to be the one from atlas-base. Hope this helps!
I'm attempting to compile a relatively simple Fortran executable so that it can be passed around to other Windows users that don't have Cygwin (or something of the sort) installed, however, I'm unable to get the executable to operate as a standalone. I've tried gfortran -static file.f and gfortran -static-libgfortran file.f, however other users always encounter this error:
The program can’t start because cygwin1.dll is missing from your computer. Try reinstalling the program to fix this problem.
From what I've read online (e.g. here), the -static option should be sufficient. I have verified that running the executable from my machine (DOS prompt) does work.
I have gcc (gfortran) version 4.7.3. I should also point out this is my first attempt at compiling portable Fortran.
Update
After realizing that this isn't a gfortran-specific issue (thanks to replies here), searches led me to related posts here and here
This is partially explained in the Cygwin FAQ. The solution is to install the mingw64-i686-gcc-fortran package with its dependencies, and cross-compile your code with i686-w64-mingw32-gfortran -static.
Just package the cygwin1.dll along with your binary file (both in the same folder) then it will run just fine.
While building gcc, I get this error:
In file included from /usr/include/bits/errno.h:25,
from /usr/include/errno.h:36,
from ../.././gcc/tsystem.h:96,
from ../.././gcc/crtstuff.c:68:
/usr/include/linux/errno.h:4:23: error: asm/errno.h: No such file or directory
make[2]: *** [crtbegin.o] Error 1
make[2]: Leaving directory `/opt/gcc-4.1.2/host-x86_64-unknown-linux-gnu/gcc'
I am building gcc 4.1 from source. I think I have to install build-essential. However installing that package in ubuntu 12.04 will automatically download and install gcc 4.6 and I don't want that.
Is there any other way?
I think the package you want is linux-libc-dev . I encountered this when building 32-on-64; so I needed linux-libc-dev:i386 .
This worked for me:
ln -s /usr/include/asm-generic /usr/include/asm
This worked for me:
sudo ln -s /usr/include/asm-generic /usr/include/asm
The reason being that what GCC expects to be called /usr/include/asm is renamed to /usr/include/asm-generic in some distros.
This fixed it for me.
sudo apt-get install linux-libc-dev:i386
This solved it for me on Debian 10, even though I was compiling with an LLVM-based compiler:
sudo apt install gcc-multilib
/usr/include/asm/errno.h is part of the linux headers. I can't speak directly to Ubuntu 12.04, but in general you can download the linux sources as a package for your distro and it shouldn't require you to download/install gcc. Failing that, you can manually download the linux headers for the version of your kernel (uname -a) and use an include directive to CFLAGS to specify the directory to look for those.
Edit: sudo apt-get install linux-headers-generic may work for you.
You are missing part of the development packages. I don't know Ubuntu, but you should be able to ask it's package management system to install the package containing /usr/include/asm/errno.h.
Do not copy some file with a similar name from somewhere on your system (or, even worse, from somewhere else). Missing files might mean that some package is damaged; again, ask your package manager to check everything and (re)install missing/broken pieces.
Unless you are running some LTS release, upgrade. Your Ubuntu is some 2 years old, i.e., ancient.
While we are at this, why on this beautiful planet are you building such an ancient compiler? Current GCC is just released 4.9.0, anything before 4.7 is ancient history, not longer supported.
On Ubuntu 16.04 x86_64 you could try this:
ln -s /usr/include/x86_64-linux-gnu/asm /usr/include/asm
This works on my server.
If you want to use errno.h that is in the asm file, simply go to /usr/(ctrl + l, type /usr/) and then search for errno.h and errno-base.h. Once you did find them, copy the code in these two files, and place them in your include folder. And be careful, in "errno.h" the file includes "errno-base.h" as:
#include <asm-generic/errno-base.h>
Either create a directory with the same name above or change the code above to something different which is suitable for you to use.
If you can find:
usr/include/asm-generic/errno.h
by executing:
find /usr/include -name errno.h
then try to execute:
cp --archive /usr/include/asm-generic /usr/include/asm
It may fix that problem.
I had this issue while compiling Asterisk 1.8.24.0 and solved it with:
mkdir /usr/include/asm-generic
cp /usr/include/asm/errno-base.h /usr/include/asm-generic/
Don't know if it is the "right way" but i've read the comments above and that gave me the idea... and it worked :)