starpu-max/README.dev

Installing StarPU on windows
----------------------------

1. Install cygwin 

http://cygwin.com/install.html

Make sure the following packages are available:
- svn
- libtool
- gcc
- make
- your favorite editor (vi, emacs, ...)
- gdb
- zip
- pkg-config

2. Install mingw

http://sourceforge.net/projects/mingw/

3. Install hwloc (not mandatory)

http://www.open-mpi.org/projects/hwloc

4. Install Microsoft Visual C++ 2008 Studio Express

   http://www.microsoft.com/express/Downloads

   On cygwin (and mingw), add in your path the following directories:

   export PATH=/cygdrive/c/Program Files/Microsoft Visual Studio 9.0/Common7/IDE:$PATH
   export PATH=/cygdrive/c/Program Files/Microsoft Visual Studio 9.0/VC/bin:$PATH
   
   Try to call <lib.exe> and <link.exe> from MinGW, else no .def or .lib file
   will be produced.

5. Install GPU Drivers (not mandatory)

  5.1 Install Cuda

      http://developer.nvidia.com/object/cuda_3_2_downloads.html

      You need to install at least the CUDA toolkit.

      libtool is not able to find the libraries automatically, you
      need to make some copies:

      copy c:\cuda\lib\cuda.lib c:\cuda\lib\libcuda.lib
      copy c:\cuda\lib\cudart.lib c:\cuda\lib\libcudart.lib
      copy c:\cuda\lib\cublas.lib c:\cuda\lib\libcublas.lib
      copy c:\cuda\lib\cufft.lib c:\cuda\lib\libcufft.lib
      copy c:\cuda\lib\OpenCL.lib c:\cuda\lib\libOpenCL.lib

      (and if the version of your CUDA driver is >= 3.2)

      copy c:\cuda\lib\curand.lib c:\cuda\lib\libcurand.lib

      Add the CUDA bin directory in your path

      export PATH=/cygdrive/c/CUDA/bin:$PATH

  5.2 Install OpenCL

      http://developer.nvidia.com/object/opencl-download.html

      You need to download the NVIDIA Drivers for your version of
      Windows. Executing the file will extract all files in a given
      directory. The the driver installation will start, it will fail
      if no compatibles drivers can be found on your system.

      Anyway, you should copy the *.dl_ files from the directory
      (extraction path) in the bin directory of the CUDA installation
      directory (the directory should be v3.2/bin/)

  5.3 Install MsCompress

      http://gnuwin32.sourceforge.net/packages/mscompress.htm

      Go in the CUDA bin directory, uncompress .dl_ files and rename
      them in .dll files

      cp /cygdrive/c/NVIDIA/DisplayDriver/190.89/International/*.dl_ .
      for i in *.dl_ ; do /cygdrive/c/Program\ Files/GnuWin32/bin/msexpand.exe  $i ; mv ${i%_} ${i%_}l ; done

6. Start autogen.sh from cygwin

   cd starpu-trunk
   ./autogen.sh

7. Start a MinGW shell

   /cygdrive/c/MinGW/msys/1.0/bin/sh.exe --login -i

8. Configure, make, install from MinGW

   If you have a non-english version of windows, use

     export LANG=C

   else libtool has troubles parsing the translated output of the toolchain.

   cd starpu-trunk
   mkdir build
   cd build
   ../configure --prefix=$PWD/target --disable-default-drand48
        --with-hwloc=<HWLOC installation directory> \
        --with-cuda-dir=<CUDA installation directory> \
        --with-cuda-lib-dir=<CUDA installation directory>/lib/Win32 \
        --enable-opencl
   make
   make install

   Also convert a couple of files to CRLF:

   sed -e 's/$/'$'\015'/ < README > $PWD/target/README.txt
   sed -e 's/$/'$'\015'/ < AUTHORS > $PWD/target/AUTHORS.txt
   sed -e 's/$/'$'\015'/ < COPYING.LGPL > $PWD/target/COPYING.LGPL.txt

9. If you want your StarPU installation to be standalone, you need to
   copy the DLL files from hwloc, Cuda, and OpenCL into the StarPU
   installation bin directory, as well as MinGW/bin/libpthread*dll

   cp <CUDA directory>/bin/*dll target/bin
   cp <HWLOC directory>/bin/*dll target/bin
   cp /cygdrive/c/MinGW/bin/libpthread*dll target/bin

   and set the StarPU bin directory in your path.

   export PATH=<StarPU installation directory>/bin:$PATH


Gcc-compiled CUDA
-----------------

If you tell nvcc to build CUDA code with gcc instead of Visual studio, a fix is
needed: c:\cuda\include\host_defines.h has a bogus CUDARTAPI definition which
makes

  linking fail completely. Replace the first occurence of

    #define CUDARTAPI
    
  with
    
    #ifdef _WIN32
    #define CUDARTAPI __stdcall
    #else
    #define CUDARTAPI
    #endif

  While at it, you can also comment the __cdecl definition to avoid spurious
  warnings.