I won’t annoy you repeating myself.
I have reached this point where i am actually desperate. I feel like i am the only entity of this universe which is currently trying to build the Fortran CUBLAS example (Fortran_Cuda_Blas.tgz) under Windows XP (yes I know it’s not the case, everybody knows how to do it, it’s just me ← stupid noob)… <img src=‘http://hqnveipbwb20/public/style_emoticons/<#EMO_DIR#>/crying.gif’ class=‘bbc_emoticon’ alt=‘:’(’ /> <img src=‘http://hqnveipbwb20/public/style_emoticons/<#EMO_DIR#>/crying.gif’ class=‘bbc_emoticon’ alt=‘:’(’ /> <img src=‘http://hqnveipbwb20/public/style_emoticons/<#EMO_DIR#>/crying.gif’ class=‘bbc_emoticon’ alt=‘:’(’ />
I have a hammer, and i know that at least one CUDA programmer will feel concerned about the likely interaction between this everyday tool and one brand new 8800 GTX… External Image
Sorry, I can’t help because I don’t use Fortran, CUBLAS, or XP for my project, but the threat of hardware violence made it worth a reply. Thanks for the laugh!
EDIT: I had one idea - are you sure you have the CUBLAS library in your linker command?
With the Fortran CUBLAS example a small wrapper “fortran.c” is provided, my guess is that all my problems come from my inability to compile and link this wrapper with my “sgemm_speed.f90” code (I have tried under Intel Visual Fortran and Compaq Visual Fortran). Have you ever done such things ? Do i have to change of compiler ? If yes, can I compile CUDA under cygwin with g95?
A lot of questions that may or may not find an answer in our next episode : “Hard Times for Noob”.
I’ve never used fortran in windows and only once used it long ago in linux. There, I could link in C code by compiling the C code to an object file (gcc -c fortran.c) and link it into the fortran executable (fortran-compiler -o exec fortran_code.f fortran.o).
In visual studio, I’m not sure if you can link object files like that directly into the executable. You may have better luck compiling the c wrapper as a static library, then linking that static library to your fortran code.
Why is it impossible to use CUDA from cygwin? Assuming you use the command line versions of the microsoft/intel compilers, I don’t see how it is really any different calling them from a command line than using the GUI. But I also haven’t actually tried, so please enlighten me.
