[CPMD-list] installation of parallelized version for MAC darwin

[Yun-an Yan]

Dear Axel,

Thank you so much for your quick response. I got some ideas from it.

Best wishes,
Yun-an

On Apr 17, 2008, at 8:29 PM, Axel Kohlmeyer wrote:

> On Thu, 17 Apr 2008, Yun-an Yan wrote:
>
>
> YY> Dear all,
> YY>
> YY> when I try to install the parallelized version of CPMD in MAC  
> OS, the
> YY> executable
> YY> is produced successfully and the speed of it is 10% slower than  
> the single
>
> yep. about 10% is the penalty for a single task when comparing the
> serial version of the code to the parallel version.
>
> YY> CPU version. But from the output there are something unpleasure.  
> Please check
> YY> the following for details.
>
> YY> COMPILER: MPICH2, Intel C/Fortran compiler
> YY>
> YY> $ more Makefile
> YY>  
> #----------------------------------------------------------------------------
> YY> #Personal Configuration
> YY>  
> #----------------------------------------------------------------------------
> YY> LD = mpif90 -O1
> YY> LFLAGS = -Xlinker -lsvml -L. \
> YY>                  -L/opt/intel/mkl/lib/em64t -llapack -lblas \
> YY>                   -lguide -Vaxlib
> YY> #LFLAGS =  -llapack -lblas -lm -lpthread -lifport -lifcore
> YY> CPP = /usr/bin/cpp -P -C -traditional
> YY> CPPFLAGS = -D__alpha -DFFT_DEFAULT -DPOINTER8  -DALPHALINUX \
> YY>                           -DPARALLEL -DMYRINET -DLINUX_IA64_INTEL
>
> hmmmm.... this combination of preprocessor flags will give you
> some weird behavior at times. luckily this will mostly affect
> modules you are not likely to use (e.g. k-points with swap files).
>
> YY> CC = mpicc -c -O1
> YY> CFLAGS = -O1
> YY> FC = mpif90 -c -I. -I$(SRC)
> YY> FFLAGS = -O1
> YY>
> YY> The code cannot be compiled with the option -D__OSX and I have to
> YY> use the stupid thing like -DALPHALINUX for MAC system.
> YY>
> YY> $ mpirun -np 2 ./cpmd.x CPMD_inp.run .
> YY> ........
> YY> ........
> YY> ***  RINFORCE| SIZE OF THE PROGRAM IS (unknown) kBYTES ***
> YY> ***    FFTPRP| SIZE OF THE PROGRAM IS (unknown) kBYTES ***
> YY> ***   CLUSTER| SIZE OF THE PROGRAM IS (unknown) kBYTES ***
> YY> ........
> YY> ........
> YY> ***   INTERPT| SIZE OF THE PROGRAM IS (unknown) kBYTES ***
> YY> ***     PHFAC| SIZE OF THE PROGRAM IS (unknown) kBYTES ***
> YY> ***    ATOMWF| SIZE OF THE PROGRAM IS (unknown) kBYTES ***
> YY> ......
> YY>
> YY> One sees that there are many unknowns.
> YY>
> YY> who has idea for that?
>
> because of setting for alpha/linux the memory size detection
> routine tries to read the current memory size of the executable
> from /proc/self/status, which is a linux-only feature. -> unknown.
>
> you machine is currently not directly supported. one has to
> carefully check the code and place a new define (e.g. __OSX_INTEL)
> at the places where needed. you are currently including random
> code from different platforms...
>
> cheers,
>   axel.
>
>
>
> YY>
> YY> Thank you so much for your consideration.
> YY>
> YY>
> YY> Best wishes,
> YY> Yun-an Yan
> YY>
> YY>
>
> -- 
> = 
> ======================================================================
> Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http:// 
> www.cmm.upenn.edu
>   Center for Molecular Modeling   --   University of Pennsylvania
> Department of Chemistry, 231 S.34th Street, Philadelphia, PA  
> 19104-6323
> tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
> = 
> ======================================================================
> If you make something idiot-proof, the universe creates a better  
> idiot.