From 3604dfae01896fa69a8327b313cf2f5da11bc4c1 Mon Sep 17 00:00:00 2001 From: kmazouzi Date: Wed, 8 Apr 2015 20:10:43 +0200 Subject: [PATCH] MPI version --- lab3/Makefile | 14 ++- lab3/mpi_heat2D.c | 343 +++++++++++++++++++++++++++++++++++++++++++++++++++++ lab3/omp_heat2D.c | 86 +++++++++----- lab3/ser_heat2D.c | 71 ++++++----- 4 files changed, 451 insertions(+), 63 deletions(-) create mode 100644 lab3/mpi_heat2D.c diff --git a/lab3/Makefile b/lab3/Makefile index 7eb3091..7297cfc 100644 --- a/lab3/Makefile +++ b/lab3/Makefile @@ -2,21 +2,25 @@ GCC = gcc CFLAGS = -O3 -fopenmp OMP_FLAG = -fopenmp RM = rm -rf - - -EXE = omp_heat2D ser_heat2D +MPI = mpicc +MPI_FLAG = -O1 -g +EXE = omp_heat2D ser_heat2D mpi_heat2D all : $(EXE) #.PHONY: all clean purge -pi_ser: ser_heat2D.o +ser_heat2D: ser_heat2D.o $(GCC) $(CFLAGS) -o $@ $^ -pi_task: omp_heat2D.o +omp_heat2D: omp_heat2D.o $(GCC) $(CFLAGS) -o $@ $^ +mpi_heat2D: + $(MPI) $(MPI_FLAG) mpi_heat2D.c -o $@ + + %.o :%.c $(GCC) $(CFLAGS) -c -o $@ $< diff --git a/lab3/mpi_heat2D.c b/lab3/mpi_heat2D.c new file mode 100644 index 0000000..57d10d0 --- /dev/null +++ b/lab3/mpi_heat2D.c @@ -0,0 +1,343 @@ +/**************************************************************************** + * DESCRIPTION: + * Serial HEAT2D Example - C Version + * This example is based on a simplified + * two-dimensional heat equation domain decomposition. The initial + * temperature is computed to be high in the middle of the domain and + * zero at the boundaries. The boundaries are held at zero throughout + * the simulation. During the time-stepping, an array containing two + * domains is used; these domains alternate between old data and new data. + * + * The physical region, and the boundary conditions, are suggested + by this diagram; + + u = 0 + +------------------+ + | | + u = 100 | | u = 100 + | | + | | + | | + | | + +------------------+ + u = 100 + +Interrior point : + u[Central] = (1/4) * ( u[North] + u[South] + u[East] + u[West] ) + + +PARALLEL MPI VERSION : + + +-------------------+ + | | P0 m=(n-2)/P +2 + +-------------------+ + | | P1 + +-------------------+ + n | | .. + +-------------------+ + | | Pq + +-------------------+ + + <-------- n --------> + <-------n-2 ------> + + + ****************************************************************************/ +#include +#include +#include +#include +#define NN 50 +#define MM 50 + +#define RING 100 +#define ITER_PRINT 100 +#define PRINT_DATA 1 +#define MAX_ITER 1000 +#define _EPSILON 0.01 + + +float update(int rank,int size, int nx,int ny, float *u, float *unew); +void inidat(int rank, int size, int nx, int ny, float *u, float *unew); +void prtdat(int rank, int size, int nx, int ny, float *u,const char *fnam); + + + + +int main(int argc, char *argv[]) +{ + int N=NN,M=MM; + + int rank,size; + + float EPSILON=_EPSILON; + + + /* INITIALIZE MPI */ + MPI_Init(&argc, &argv); + + /* GET THE PROCESSOR ID AND NUMBER OF PROCESSORS */ + MPI_Comm_rank(MPI_COMM_WORLD, &rank); + MPI_Comm_size(MPI_COMM_WORLD, &size); + + //Only Rank 0 read application parameters + if(rank==0) { + + if(argc !=3) + { + fprintf(stderr,"usage %s N EPSILON\n ", argv[0]); + fprintf(stderr,"\t\twhere N is GRID size, EPSILON is Tolerance\n"); + fprintf(stderr,"\t\texample N = 100, EPSILON = 0.1\n"); + return -1; + } + + N = M = atoi(argv[1]); + EPSILON = atof(argv[2]); + + if(N % size!=0) + { + fprintf(stderr,"Grid Size MUST be divisible by the number of processors !"); + return -1; + } + + } + + //Wait for rank 0 , all process start here + MPI_Barrier(MPI_COMM_WORLD); + + //Exchange N + MPI_Bcast(&N , 1, MPI_FLOAT, 0 , MPI_COMM_WORLD); + //Exchange EPSILON + MPI_Bcast(&EPSILON , 1, MPI_FLOAT, 0 , MPI_COMM_WORLD); + + //local size + M = (N-2)/size + 2; + + float *u = (float *)malloc(N * M * sizeof(float)); + float *unew = (float *)malloc(N * M * sizeof(float)); + + if(u==0 || unew ==0) + { + perror("Can't allocated data\n"); + return -1; + } + + if(rank==0) { + + printf ( "\n" ); + printf ( "HEATED_PLATE\n" ); + printf ( " Parallel MPI version using %d processors \n",size ); + printf ( " A program to solve for the steady state temperature distribution\n" ); + printf ( " over a rectangular plate.\n" ); + printf ( "\n" ); + printf ( " Spatial grid of %d by %d points.\n", N, N ); + printf ( " Each processor will use grid of %d +2 by %d points.\n", M-2, N ); + printf ( " The iteration will end until tolerance <= %f\n\n",EPSILON); + + } + + /* Initialize grid and create input file + * each process initialize its part + * */ + + inidat(rank,size,M,N,u,unew); + + prtdat(rank,size,M,N,u, "initial.dat"); + + + /* + * iterate until the new solution unew differs from the old solution u + * by no more than EPSILON. + * */ + + float diff=1.0; + int iter=0; + + while(diff> EPSILON) { + + diff= update(rank,size,M,N, u, unew); + + if(rank==0) + if(iter%ITER_PRINT==0) + printf("Processor #%d, iteration %d, epsilon = %f\n ", rank,iter,diff); + iter++; + } + + prtdat(rank,size,M,N, u, "final.dat"); + free(u); + free(unew); + MPI_Finalize(); +} + + + +/**************************************************************************** + * subroutine update + ****************************************************************************/ +float update(int rank, int size, int nx,int ny, float *u, float *unew){ + int ix, iy; + float diff=0.0; + float globaldiff; + MPI_Status status; + + + /* + * EXCHANGE GHOST CELL + */ + if (rank > 0 && rank< size-1) + { + MPI_Sendrecv(&u[ny*(nx-2)], ny, MPI_FLOAT, rank+1, 0, + &u[ny*0], ny, MPI_FLOAT, rank-1, 0, MPI_COMM_WORLD, &status); + MPI_Sendrecv(&u[ny*1], ny, MPI_FLOAT, rank-1, 1, + &u[ny*(nx-1)], ny, MPI_FLOAT, rank+1, 1, MPI_COMM_WORLD, &status); + } + + else if (rank == 0 && rank< size-1) + MPI_Sendrecv(&u[ny*(nx-2)], ny, MPI_FLOAT, rank+1, 0, + &u[ny*(nx-1)], ny, MPI_FLOAT, rank+1, 1, MPI_COMM_WORLD, &status); + else if ( rank> 0 && rank == size-1) + MPI_Sendrecv(&u[ny*1], ny, MPI_FLOAT, rank-1, 1, + &u[ny*0], ny, MPI_FLOAT, rank-1, 0, MPI_COMM_WORLD, &status); + + + + /** + * PERFORM LOCAL COMPUTATION + * */ + + for (ix = 1; ix < nx-1; ix++) { + for (iy = 1; iy < ny-1; iy++) { + unew[ix*ny+iy] = (u[(ix+1)*ny+iy] + u[(ix-1)*ny+iy] + u[ix*ny+iy+1] + u[ix*ny+iy-1] )/4.0 + ; + if (diff < fabs (unew[ix*ny+iy] - u[ix*ny+iy] )) + { + diff = fabs ( unew[ix*ny+iy] - u[ix*ny+iy] ); + } + } + + } + + + /** + * COMPUTE GLOBAL CONVERGENCE + * + * */ + + MPI_Allreduce(&diff, &globaldiff , 1, MPI_FLOAT, MPI_MAX, MPI_COMM_WORLD); + + + /** + * COPY OLD DATA + * */ + + + for (ix = 1; ix < nx-1; ix++) { + for (iy = 1; iy < ny-1; iy++) { + u[ix*ny+iy] = unew[ix*ny+iy]; + } + } + + + return globaldiff; +} + +/***************************************************************************** + * Initialize Data + *****************************************************************************/ + +void inidat(int rank, int size,int nx, int ny, float *u, float *unew) +{ + int ix, iy; + + /* + *Set boundary data and interrior values + * */ + + + // interior points + for (ix = 1; ix < nx-1; ix++) + for (iy = 1; iy < ny-1; iy++) { + u[ix*ny+iy]=0.0; + } + + //boundary left + for (ix = 1; ix < nx-1; ix++){ + u[ix*ny]=100.0; + + } + + //boundary right + for (ix = 1; ix < nx-1; ix++){ + u[ix*ny+ (ny-1)]=100.0; + + } + + //boundary down + for (iy = 0; iy < ny; iy++){ + + if(rank==size-1) { + u[(nx-1)*(ny)+iy]=100.0; + }else + { + u[(nx-1)*(ny)+iy]=0.0; + } + } + + //boundary top + for (iy = 0; iy < ny; iy++){ + u[iy]=0.0; + + } + +} + + +/*************************************************************************** + * Print Data to files + **************************************************************************/ + +void print2file(int rank, int nx, int ny, float *u,const char *fname) +{ + int ix, iy; + FILE *fp; + + char str[255]; + + sprintf(str, "%d_%s", rank,fname); + + fp = fopen(str, "w"); + + for (ix = 0 ; ix < nx; ix++) { + for (iy =0; iy < ny; iy++) { + + fprintf(fp, "%6.2f ", u[ix*ny+iy]); + } + fputc ( '\n', fp); + } + + fclose(fp); + +} + + + +void prtdat(int rank, int size,int nx, int ny, float *u,const char *fname) +{ + + + if(ITER_PRINT==0)return; + + + /** + * USE TOKEN RING to write to unique file + * 0 will rite first and then 1, 2, ... size-1 + * + * * */ + + + print2file(rank,nx,ny,u,fname); + +} + + + diff --git a/lab3/omp_heat2D.c b/lab3/omp_heat2D.c index 2744257..ae85c38 100644 --- a/lab3/omp_heat2D.c +++ b/lab3/omp_heat2D.c @@ -37,7 +37,7 @@ Interrior point : #define _EPSILON 0.001 -void update(int nx,int ny, float *u, float *unew, float * diff); +float update(int nx,int ny, float *u, float *unew); void inidat(int nx, int ny, float *u, float *unew); void prtdat(int nx, int ny, float *u,const char *fnam); @@ -97,7 +97,7 @@ int main(int argc, char *argv[]) while(diff> EPSILON) { - update(N, M, u, unew,&diff); + diff = update(N, M, u, unew); if(iter%ITER_PRINT==0) printf("Iteration %d, diff = %f\n ", iter,diff); @@ -116,10 +116,10 @@ int main(int argc, char *argv[]) /**************************************************************************** * subroutine update ****************************************************************************/ -void update(int nx,int ny, float *u, float *unew, float * diff) +float update(int nx,int ny, float *u, float *unew) { int ix, iy; - *diff=0.0; + float diff=0.0; #pragma omp parallel for shared(nx,ny,u,unew) private (ix,iy) for (ix = 1; ix < nx-1; ix++) { @@ -136,6 +136,10 @@ void update(int nx,int ny, float *u, float *unew, float * diff) float mydiff; +/** + * IMPLEMENT OMP REDUCE MAX + */ + #pragma omp parallel shared(nx,ny,u,unew, diff) private (ix,iy,mydiff) { mydiff=0.0; @@ -150,15 +154,17 @@ void update(int nx,int ny, float *u, float *unew, float * diff) } -# pragma omp critical +#pragma omp critical { - if (*diff < mydiff ) + if (diff < mydiff ) { - *diff = mydiff; + diff = mydiff; } } - +/* + * COPY OLD DATA + */ #pragma omp for for (ix = 1; ix < nx-1; ix++) { for (iy = 1; iy < ny-1; iy++) { @@ -166,11 +172,14 @@ void update(int nx,int ny, float *u, float *unew, float * diff) } } } + + return diff; } /***************************************************************************** * Initialize Data *****************************************************************************/ + void inidat(int nx, int ny, float *u, float *unew) { int ix, iy; @@ -178,33 +187,52 @@ void inidat(int nx, int ny, float *u, float *unew) /* *Set boundary data and interrior values * */ - for (ix = 0; ix < nx; ix++) - for (iy = 0; iy < ny; iy++) { - if(ix==0) - { - u[ix*ny+iy]=0.0; - } - else - if(iy==0 && ix!=0) - { - u[ix*ny+iy]=100.0; - }else +#pragma omp parallel private (ix,iy) + { + // interior points + #pragma omp for + for (ix = 1; ix < nx-1; ix++) + for (iy = 1; iy < ny-1; iy++) { + u[ix*ny+iy]=5.0; + } + + //boundary left + #pragma omp for + for (ix = 1; ix < nx-1; ix++){ + u[ix*ny]=100.0; - if(ix==nx-1) - { - u[ix*ny+iy]=100.0; - }else + } - if(iy==ny-1 && ix!=0) - { - u[ix*ny+iy]=100.0; - }else + //boundary right + #pragma omp for + for (ix = 1; ix < nx-1; ix++){ + u[ix*ny+ (ny-1)]=100.0; + + } + + //boundary down + #pragma omp for + for (iy = 0; iy < ny; iy++){ + u[(nx-1)*(ny)+iy]=100.0; + + } + + //boundary top + #pragma omp for + for (iy = 0; iy < ny; iy++){ + u[iy]=0.0; + + } + + } - u[ix*ny+iy]=0.0; - } } + + + + /************************************************************************** * Print Data to files **************************************************************************/ diff --git a/lab3/ser_heat2D.c b/lab3/ser_heat2D.c index 176bd1c..23637d5 100644 --- a/lab3/ser_heat2D.c +++ b/lab3/ser_heat2D.c @@ -36,7 +36,7 @@ Interrior point : #define _EPSILON 0.01 -void update(int nx,int ny, float *u, float *unew, float * diff); +float update(int nx,int ny, float *u, float *unew); void inidat(int nx, int ny, float *u, float *unew); void prtdat(int nx, int ny, float *u,const char *fnam); @@ -98,7 +98,7 @@ int main(int argc, char *argv[]) while(diff> EPSILON) { - update(N, M, u, unew,&diff); + diff= update(N, M, u, unew); if(iter%ITER_PRINT==0) @@ -118,30 +118,33 @@ int main(int argc, char *argv[]) /**************************************************************************** * subroutine update ****************************************************************************/ -void update(int nx,int ny, float *u, float *unew, float * diff) +float update(int nx,int ny, float *u, float *unew) { int ix, iy; - *diff=0.0; + float diff=0.0; for (ix = 1; ix < nx-1; ix++) { for (iy = 1; iy < ny-1; iy++) { unew[ix*ny+iy] = (u[(ix+1)*ny+iy] + u[(ix-1)*ny+iy] + u[ix*ny+iy+1] + u[ix*ny+iy-1] )/4.0 ; - if (*diff < fabs (unew[ix*ny+iy] - u[ix*ny+iy] )) + if (diff < fabs (unew[ix*ny+iy] - u[ix*ny+iy] )) { - *diff = fabs ( unew[ix*ny+iy] - u[ix*ny+iy] ); + diff = fabs ( unew[ix*ny+iy] - u[ix*ny+iy] ); } } } - + /** + * COPY OLD DATA + */ for (ix = 1; ix < nx-1; ix++) { for (iy = 1; iy < ny-1; iy++) { u[ix*ny+iy] = unew[ix*ny+iy]; } } + return diff; } /***************************************************************************** @@ -154,33 +157,43 @@ void inidat(int nx, int ny, float *u, float *unew) /* *Set boundary data and interrior values * */ - for (ix = 0; ix < nx; ix++) - for (iy = 0; iy < ny; iy++) { - - if(ix==0) - { - u[ix*ny+iy]=0.0; - } - else - if(iy==0 && ix!=0) - { - u[ix*ny+iy]=100.0; - }else - - if(ix==nx-1) - { - u[ix*ny+iy]=100.0; - }else - if(iy==ny-1 && ix!=0) - { - u[ix*ny+iy]=100.0; - }else - u[ix*ny+iy]=0.0; + // interior points + for (ix = 1; ix < nx-1; ix++) + for (iy = 1; iy < ny-1; iy++) { + u[ix*ny+iy]=5.0; } + + //boundary left + for (ix = 1; ix < nx-1; ix++){ + u[ix*ny]=100.0; + + } + + //boundary right + for (ix = 1; ix < nx-1; ix++){ + u[ix*ny+ (ny-1)]=100.0; + + } + + //boundary down + for (iy = 0; iy < ny; iy++){ + u[(nx-1)*(ny)+iy]=100.0; + + } + + //boundary top + for (iy = 0; iy < ny; iy++){ + u[iy]=0.0; + + } + + } + + /************************************************************************** * Print Data to files **************************************************************************/ -- 1.7.10.4