Install lammps with clmath

LAMMPS LAMMPS Mailing List Mirror
1

Hi Lammps people!
I try appml / clMath (clFft) compiled Lammps, but after I added the
appropriate environment make linux given, it should be about changing the
FFTW result. I would like to ask you, whether by replacing the FFTW
library that implements the AMD GPU computing.
lammps version lammps-30Oct13
libclFFT.so.2.1.0==>libfftw.so
clAmdFft.h==>fftw.h
Makefile
CC = mpicc
CCFLAGS = -O
SHFLAGS = -fPIC
DEPFLAGS = -M

LINK = mpicc
LINKFLAGS = -O
LIB = -lstdc++
SIZE = size

ARCHIVE = ar
ARFLAGS = -rc
SHLIBFLAGS = -shared
LMP_INC = -DLAMMPS_GZIP -DLAMMPS_JPEG

MPI_INC = -DMPICH_SKIP_MPICXX
MPI_PATH =
MPI_LIB = -lmpich -lmpl -lpthread
FFT_INC = -DFFT_FFTW
-I/home/zpg/Downloads/clFFT-develop/src/include/ -I/opt/AMDAPP/include/
FFT_PATH =
FFT_LIB = /home/zpg/Downloads/clFFT-develop/src/library/libfftw.so

Error message：

mpicc -O -DLAMMPS_GZIP -DLAMMPS_JPEG -DMPICH_SKIP_MPICXX -DFFT_FFTW
-I/home/zpg/Downloads/clFFT-develop/src/include/ -I/opt/AMDAPP/include/
-c ../ewald_disp.cpp
mpicc -O -DLAMMPS_GZIP -DLAMMPS_JPEG -DMPICH_SKIP_MPICXX -DFFT_FFTW
-I/home/zpg/Downloads/clFFT-develop/src/include/ -I/opt/AMDAPP/include/
-c ../fft3d.cpp
In file included from ../fft3d.cpp:25:
../fft3d.h:214: 错误：‘FFTW_COMPLEX’没有命名一个类型
../fft3d.h:250: 错误：ISO C++ 不允许声明无类型的‘FFT_DATA’
../fft3d.h:250: 错误：expected ‘;’ before ‘*’ token
../fft3d.h:251: 错误：ISO C++ 不允许声明无类型的‘FFT_DATA’
../fft3d.h:251: 错误：expected ‘;’ before ‘*’ token
../fft3d.h:292: 错误：‘fftw_plan’没有命名一个类型
../fft3d.h:293: 错误：‘fftw_plan’没有命名一个类型
../fft3d.h:294: 错误：‘fftw_plan’没有命名一个类型
../fft3d.h:295: 错误：‘fftw_plan’没有命名一个类型
../fft3d.h:296: 错误：‘fftw_plan’没有命名一个类型
../fft3d.h:297: 错误：‘fftw_plan’没有命名一个类型
../fft3d.h:318: 错误：变量或字段‘fft_3d’声明为 void
../fft3d.h:318: 错误：‘FFT_DATA’在此作用域中尚未声明
../fft3d.h:318: 错误：expected primary-expression before ‘,’ token
../fft3d.h:318: 错误：‘FFT_DATA’在此作用域中尚未声明
../fft3d.h:318: 错误：expected primary-expression before ‘,’ token
../fft3d.h:318: 错误：expected primary-expression before ‘int’
../fft3d.h:318: 错误：expected primary-expression before ‘struct’
../fft3d.h:326: 错误：变量或字段‘fft_1d_only’声明为 void
../fft3d.h:326: 错误：‘FFT_DATA’在此作用域中尚未声明
../fft3d.h:326: 错误：expected primary-expression before ‘,’ token
../fft3d.h:326: 错误：expected primary-expression before ‘int’
../fft3d.h:326: 错误：expected primary-expression before ‘int’
../fft3d.h:326: 错误：expected primary-expression before ‘struct’
../fft3d.cpp:66: 错误：变量或字段‘fft_3d’声明为 void
../fft3d.cpp:66: 错误：‘FFT_DATA’在此作用域中尚未声明
../fft3d.cpp:66: 错误：‘in’在此作用域中尚未声明
../fft3d.cpp:66: 错误：‘FFT_DATA’在此作用域中尚未声明
../fft3d.cpp:66: 错误：‘out’在此作用域中尚未声明
../fft3d.cpp:66: 错误：expected primary-expression before ‘int’
../fft3d.cpp:66: 错误：expected primary-expression before ‘struct’ make[1]:
*** [fft3d.o] 错误 1
make[1]: Leaving directory
`/home/zpg/Downloads/lammps-30Oct13/src/Obj_linux' make: *** [linux] 错误 2

2

Hi Lammps people!
I try appml / clMath (clFft) compiled Lammps, but after I added the
appropriate environment make linux given, it should be about changing the
FFTW result. I would like to ask you, whether by replacing the FFTW
library that implements the AMD GPU computing.

this is a fairly pointless undertaking.
have you ever heard of a thing called Amdahl's Law?
have you looked at how much time your calculations actually spend in doing FFTs?
have you looked at how LAMMPS implements its parallel FFTs?
most likely you will *de*celerate your calculation.

a much more productive activity would be to experiment with compiler
flags for the CPU. with the settings below, you are missing quite a
few options to speed things up that the compiler can do for you.

axel.