[Pw_forum] Fwd: Error when compiling GPU version of QE
mustardt at onid.orst.edu
Fri Sep 21 15:02:29 CEST 2012
I have gotten the MPI CPU and serial GPU versions to install nicely, but I
am having some problems with the MPI GPU version.
I configured with this:
> ./configure --enable-cuda --enable-parallel --with-internal-blas
--with-internal-lapack --with-gpu-arch=20 --with-cuda-dir=/usr/local/cuda
> make pw -j 6
Slightly off topic but if I --enable-magma I lose my FFT libraries...What?!
I have compiled the CPU version with Intel MKL libraries and have no real
reason for using the internal libraries. It does not matter which library
sources I use I will always get this cuda_env_para.o error. (Sometimes in
conjuction with a cuda memory error as well.)
My error occurs like this...
> mpif90 -O3 -g -fopenmp -x f95-cpp-input -fopenmp -D__GFORTRAN -D__STD_F95
-D__FFTW -D__MPI -D__PARA -D__CUDA -D__GPU_NVIDIA_20 -D__PHIGEMM
-D__CUDA_QE_TIMING -D__OPENMP -I../include
-I../iotk/src -I. -c cuda_env_para.f90
> In file cuda_env_para.f90:40
> Error: Syntax error in argument list at (1)
> make: *** [cuda_env_para.o] Error 1
> make: *** Waiting for unfinished jobs....
> make: Leaving directory `/root/espresso-5.0.1-GPU/Modules'
> make: *** [mods] Error 2
Anything would be grateful.
Thomas J. L. Mustard
Department of Chemistry
Oregon State University
P.S. I have intel 11.1 C/CPP/ifort installed as well as the mkl. Also fftw3
in double precision and CUDA 4.0.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the users