Parallelize three_for_loop in C++ by openmp -

-

i have code. here, a, b, c, a1, b1, c1 vectors in 3 dimensional. a, b, c independent , a1, b1, c1 independent together. want parallelize calculate using openmp. however, run openmp, "segmentation fault" error.could me problem? thank in advance. 

#include <omp.h> #include<math.h>  #include<cmath>  #include<vector>     #include<iostream>  using namespace std; int main () {  int nx=801;              // number of grid in x direction int ny=501;               int nz=401;    float pi=3.14159265358979323846; unsigned int i,j,k; vector<vector<vector<float> > > (nx,vector<vector<float> >(ny,vector <float>(nz,0.0))); vector<vector<vector<float> > > b (nx,vector<vector<float> >(ny,vector <float>(nz,0.0))); vector<vector<vector<float> > > c (nx,vector<vector<float> >(ny,vector <float>(nz,0.0))); vector<vector<vector<float> > > a1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0))); vector<vector<vector<float> > > b1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0))); vector<vector<vector<float> > > c1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));  cout<<"start"<<endl; #pragma omp parallel private (j) shared(a,b,c,i,k,nx,ny,nz)  (i=0;i<nx;i++)     (j=0;j<ny;j++)         (k=0;k<nz;k++)         {             a[i][j][k]=sin(2.0*pi/float(nx*ny*nz)*float(i*j*k));             b[i][j][k]=cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));             c[i][j][k]=sin(2.0*pi/float(nx*ny*nz))*cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));         }  #pragma omp parallel private (j) shared(a1,b1,c1,a,b,c,i,k,nx,ny,nz)  (i=1;i<nx-1;i++)     (j=1;j<ny-1;j++)         (k=1;k<nz-1;k++)         {             a1[i][j][k]=c[i+1][j][k]*cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));             b1[i][j][k]=a[i][j][k]+b[i][j][k]+c[i][j][k]*cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));             c1[i][j][k]=16.0*a[i][j][k]*cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));         } cout<<"finish"<<endl;   return 0; }

this code easy parallelise openmp. however, did mistakes in own attempt, notably trying declare i , k shared whereas should private. better even, don't declare variables in advance , declare them inside for loop. way, have automatically right scope, preventing mixing up.

here give:

#include <omp.h> #include<math.h>  #include<cmath>  #include<vector>     #include<iostream>  using namespace std; int main () {      int nx=801;              // number of grid in x direction     int ny=501;                   int nz=401;        float pi=3.14159265358979323846;     vector<vector<vector<float> > > (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > b (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > c (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > a1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > b1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > c1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));      cout<<"start"<<endl;     #pragma omp parallel     (int i=0;i<nx;i++)         (int j=0;j<ny;j++)             (int k=0;k<nz;k++)             {                 a[i][j][k]=sin(2.0*pi/float(nx*ny*nz)*float(i*j*k));                 b[i][j][k]=cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));                 c[i][j][k]=sin(2.0*pi/float(nx*ny*nz))*cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));             }      #pragma omp parallel      (int i=1;i<nx-1;i++)         (int j=1;j<ny-1;j++)             (int k=1;k<nz-1;k++)             {                 a1[i][j][k]=c[i+1][j][k]*cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));                 b1[i][j][k]=a[i][j][k]+b[i][j][k]+c[i][j][k]*cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));                 c1[i][j][k]=16.0*a[i][j][k]*cos(5.0*pi/float(nx*ny*nz)*float(i*j*k));             }     cout<<"finish"<<endl;      return 0; }

now, since ask parallelising code, guess interested in performance. nothing should prevent implementing 1 or 2 basic performance optimisations this:

#include <omp.h> #include<math.h>  #include<cmath>  #include<vector>     #include<iostream>  using namespace std; int main () {      int nx=801;              // number of grid in x direction     int ny=501;                   int nz=401;        float pi=3.14159265358979323846;     vector<vector<vector<float> > > (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > b (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > c (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > a1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > b1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));     vector<vector<vector<float> > > c1 (nx,vector<vector<float> >(ny,vector <float>(nz,0.0)));      const float pioversize = pi/(nx*ny*nz);     const float sin2pioversize = sin(2.0f*pioversize);     cout<<"start"<<endl;     double tbeg = omp_get_wtime();     #pragma omp parallel     {     #pragma omp     (int i=0;i<nx;i++)         (int j=0;j<ny;j++)         {             float ijpioversize=i*j*pioversize;             (int k=0;k<nz;k++)             {                 a[i][j][k]=sin(2.0f*ijpioversize*k);                 b[i][j][k]=cos(5.0f*ijpioversize*k);                 c[i][j][k]=sin2pioversize*cos(5.0f*ijpioversize*k);             }          }     #pragma omp      (int i=1;i<nx-1;i++)         (int j=1;j<ny-1;j++)         {             float ijpioversize=i*j*pioversize;             (int k=1;k<nz-1;k++)             {                 a1[i][j][k]=c[i+1][j][k]*cos(5.0f*ijpioversize*k);                 b1[i][j][k]=a[i][j][k]+b[i][j][k]+c[i][j][k]*cos(5.0f*ijpioversize*k);                 c1[i][j][k]=16.0f*a[i][j][k]*cos(5.0f*ijpioversize*k);             }         }     }     double time = omp_get_wtime() - tbeg;     cout<<"finish in "<<time<<" seconds"<<endl;      return 0; }

with this, code should faster.


Comments

Post a Comment

Popular posts from this blog

c++ - llvm function pass ReplaceInstWithInst malloc -

-
#include "llvm/pass.h" #include "llvm/ir/module.h" #include "llvm/ir/function.h" #include "llvm/support/raw_ostream.h" #include "llvm/ir/legacypassmanager.h" #include "llvm/ir/instrtypes.h" #include "llvm/transforms/ipo/passmanagerbuilder.h" #include "llvm/ir/irbuilder.h" #include "llvm/transforms/utils/basicblockutils.h" using namespace llvm; namespace { struct replacepass : public functionpass { static char id; replacepass() : functionpass(id) {} virtual bool runonfunction(function &f) { allocainst* insttoreplace = ??? basicblock::iterator ii(insttoreplace); replaceinstwithinst(insttoreplace->getparent()->getinstlist(), ii, new allocainst(type::int32ty, 0, insttoreplace)); return true; } }; } char replacepass::id = 0; static void registerreplacepass(const passmanagerbuilder &, legacy::passmanagerbase &pm) { pm.add(new re...
Read more

Cross-Compiling Linux Kernel for Raspberry Pi - ${CCPREFIX}gcc -v does not work -

-
i'm trying follow this guide . i'm running both ubuntu 12.04.5 lts (gnu/linux 3.13.0-74-generic x86_64) on "real" hardware , 14.04.1 via virtualbox on mac. problem don't past step 1: hoffmann@angl99:~$ export ccprefix=/home/hoffmann/tools/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian/bin/arm-linux-gnueabihf- hoffmann@angl99:~$ ${ccprefix}gcc -v i'm getting following error: -bash: /home/hoffmann/tools/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian/bin/arm-linux-gnueabihf-gcc: no such file or directory however, file i'm told missing there: hoffmann@angl99:~$ less /home/hoffmann/tools/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian/bin/arm-linux-gnueabihf-gcc "/home/hoffmann/tools/arm-bcm2708/gcc-linaro-arm-linux-gnueabihf-raspbian/bin/arm-linux-gnueabihf-gcc" may binary file. see anyway? this result of basic error/misconception. suggest solution? thanks! sebastian ok - i've worked out (with of person po...
Read more

java.lang.NoClassDefFoundError When Creating New Android Project -

-
i've tried installing (and re-installing) android studio several times now, , each time, studio isn't able create new android project. when attempt initial android studio window (ie. basic 'create', 'open', 'import' etc options), i'm asked of questions in wizard, brief progress dialog, followed nothing @ all. if open incomplete project process creates choosing 'open' list, can try choosing create new android project menu. this has same issue, error message in output console time... java.lang.noclassdeffounderror: com/sun/xml/internal/bind/v2/model/annotation/annotationreader (wrong name: com/sun/xml/internal/bind/v0/model/qnnotation/annotationreader) note: when installed android studio, installed incompatible jdk (1.8, not 1.7). have had go default project settings change jdk uses 1.8. does know how can avoid this? there place should downloading jxb classes? missing in java paths? this appears have been caused cor...
Read more