| [eb26a03] | 1 | #ifdef _CIVL
|
|---|
| 2 | #include <civlc.cvh>
|
|---|
| 3 | #endif
|
|---|
| 4 | /* diffusion1d.c: parallel 1d-diffusion solver with constant boundary.
|
|---|
| 5 | * To execute: mpicc diffusion1d.c ; mpiexec -n 4 ./a.out
|
|---|
| 6 | * Or replace "4" with however many procs you want to use.
|
|---|
| 7 | * To verify: civl verify diffusion1d.c
|
|---|
| 8 | */
|
|---|
| 9 | #include <stdlib.h>
|
|---|
| 10 | #include <stdio.h>
|
|---|
| 11 | #include <assert.h>
|
|---|
| 12 | #include <mpi.h>
|
|---|
| 13 | #define OWNER(index) ((nprocs*(index+1)-1)/nx)
|
|---|
| 14 |
|
|---|
| 15 | #ifdef _CIVL
|
|---|
| 16 |
|
|---|
| 17 | $input int nx; // global number of points excl. boundary
|
|---|
| 18 | $assume(1<=nx && nx<=NXB);
|
|---|
| 19 | $input double U_INIT[nx+2]; // initial values for temperature incl. boundary
|
|---|
| 20 | $input double k; // the constant D*dt/(dx*dx)
|
|---|
| 21 | $assume(k>0 && k<.5);
|
|---|
| 22 | $input int NSTEPS_BOUND=5; // upper bound on nsteps
|
|---|
| 23 | $input int nsteps; // number of time steps
|
|---|
| 24 | $assume(1<=nsteps && nsteps<=NSTEPS_BOUND);
|
|---|
| 25 | $input int wstep; // write frame every this many time steps
|
|---|
| 26 | $assume(1<=wstep && wstep<=nsteps);
|
|---|
| 27 | double oracle[nsteps][nx+2]; // solution computed sequentially, proc 0 only
|
|---|
| 28 |
|
|---|
| 29 | #else
|
|---|
| 30 |
|
|---|
| 31 | int nx;
|
|---|
| 32 | double k;
|
|---|
| 33 | int nsteps;
|
|---|
| 34 | int wstep;
|
|---|
| 35 |
|
|---|
| 36 | #endif
|
|---|
| 37 |
|
|---|
| 38 | /* Global variables */
|
|---|
| 39 |
|
|---|
| 40 | double lbound; /* left fixed boundary value */
|
|---|
| 41 | double rbound; /* right fixed boundary value */
|
|---|
| 42 | double *u; /* temperature function, local */
|
|---|
| 43 | double *u_new; /* second copy of temperature function, local */
|
|---|
| 44 | int nprocs; /* number of processes */
|
|---|
| 45 | int rank; /* the rank of this process */
|
|---|
| 46 | int left; /* rank of left neighbor */
|
|---|
| 47 | int right; /* rank of right neighbor on torus */
|
|---|
| 48 | int nxl; /* horizontal extent of one process */
|
|---|
| 49 | int first; /* global index for local index 0 */
|
|---|
| 50 | double *buf; /* temp. buffer used on proc 0 only */
|
|---|
| 51 | int print_pos; /* number of cells printed on current line */
|
|---|
| 52 | int time=0; /* current time step */
|
|---|
| 53 |
|
|---|
| 54 |
|
|---|
| 55 | /* Returns the global index of the first cell owned
|
|---|
| 56 | * by the process with given rank */
|
|---|
| 57 | int firstForProc(int rank) {
|
|---|
| 58 | return (rank*nx)/nprocs;
|
|---|
| 59 | }
|
|---|
| 60 |
|
|---|
| 61 | /* Returns the number of cells owned by the process
|
|---|
| 62 | * of the given rank (excluding ghosts) */
|
|---|
| 63 | int countForProc(int rank) {
|
|---|
| 64 | int a = firstForProc(rank+1);
|
|---|
| 65 | int b = firstForProc(rank);
|
|---|
| 66 |
|
|---|
| 67 | return a-b;
|
|---|
| 68 | }
|
|---|
| 69 |
|
|---|
| 70 | /* Initializes the global variables.
|
|---|
| 71 | * Precondition: the configuration parameters have
|
|---|
| 72 | * already been set. */
|
|---|
| 73 | void init_globals() {
|
|---|
| 74 | MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
|
|---|
| 75 | MPI_Comm_rank(MPI_COMM_WORLD, &rank);
|
|---|
| 76 | // nxl: number actual points (incl. end-points)
|
|---|
| 77 | // nxl+2: size of array (incl. ghost cells)
|
|---|
| 78 | first = firstForProc(rank);
|
|---|
| 79 | nxl = countForProc(rank);
|
|---|
| 80 | left = first==0 || nxl==0 ? MPI_PROC_NULL : OWNER(first-1);
|
|---|
| 81 | right = first+nxl >= nx || nxl == 0 ? MPI_PROC_NULL : OWNER(first+nxl);
|
|---|
| 82 | u = (double*)malloc((nxl+2)*sizeof(double));
|
|---|
| 83 | assert(u);
|
|---|
| 84 | u_new = (double*)malloc((nxl+2)*sizeof(double));
|
|---|
| 85 | assert(u_new);
|
|---|
| 86 | if (rank == 0)
|
|---|
| 87 | buf = (double*)malloc((1+nx/nprocs)*sizeof(double));
|
|---|
| 88 | }
|
|---|
| 89 |
|
|---|
| 90 | void initialize() {
|
|---|
| 91 | #ifdef _CIVL
|
|---|
| 92 | //elaborate nx to concrete value...
|
|---|
| [841d45a] | 93 | //for (int i=0; i<nx; i++);
|
|---|
| [eb26a03] | 94 | // initialize globals and u...
|
|---|
| 95 | init_globals();
|
|---|
| 96 | lbound = U_INIT[0];
|
|---|
| 97 | rbound = U_INIT[nx+1];
|
|---|
| 98 | for (int i=1; i<=nxl; i++)
|
|---|
| 99 | u[i] = U_INIT[first+i];
|
|---|
| 100 | if (rank == 0) {
|
|---|
| 101 | // compute the oracle...
|
|---|
| 102 | for (int i=0; i<nx+2; i++)
|
|---|
| 103 | oracle[0][i]=U_INIT[i];
|
|---|
| 104 | for (int t=1; t<nsteps; t++) {
|
|---|
| 105 | oracle[t][0] = lbound;
|
|---|
| 106 | for (int i=1; i<=nx; i++)
|
|---|
| 107 | oracle[t][i] = oracle[t-1][i] +
|
|---|
| 108 | k*(oracle[t-1][i+1] + oracle[t-1][i-1]
|
|---|
| 109 | - 2*oracle[t-1][i]);
|
|---|
| 110 | oracle[t][nx+1] = rbound;
|
|---|
| 111 | }
|
|---|
| 112 | }
|
|---|
| 113 | #else
|
|---|
| 114 | nx = 10;
|
|---|
| 115 | k = 0.2;
|
|---|
| 116 | nsteps = 20;
|
|---|
| 117 | wstep = 2;
|
|---|
| 118 | lbound = rbound = 0.0;
|
|---|
| 119 | init_globals();
|
|---|
| 120 | for (int i=1; i<=nxl; i++)
|
|---|
| 121 | u[i]=100.0;
|
|---|
| 122 | #endif
|
|---|
| 123 | if (nx>=1 && rank == OWNER(0))
|
|---|
| 124 | u[0] = u_new[0] = lbound;
|
|---|
| 125 | if (nx>=1 && rank == OWNER(nx-1))
|
|---|
| 126 | u[nxl+1] = u_new[nxl+1] = rbound;
|
|---|
| 127 | //if (rank == 0)
|
|---|
| 128 | //printf("nx=%d, k=%lf, nsteps=%d, wstep=%d, nprocs=%d\n",
|
|---|
| 129 | // nx, k, nsteps, wstep, nprocs);
|
|---|
| 130 | }
|
|---|
| 131 |
|
|---|
| 132 | /* Prints header for time step. Called by proc 0 only */
|
|---|
| 133 | void print_time_header() {
|
|---|
| 134 | //printf("======= Time %d =======\n", time);
|
|---|
| 135 | print_pos = 0;
|
|---|
| 136 | }
|
|---|
| 137 |
|
|---|
| 138 | /* Prints one cell. Called by proc 0 only. */
|
|---|
| 139 | void print_cell(double value) {
|
|---|
| 140 | //printf("%7.2f\n", value);
|
|---|
| 141 | #pragma CIVL $assert(value == oracle[time][print_pos], \
|
|---|
| 142 | "Error: disagreement at time %d position %d: saw %lf, expected %lf", \
|
|---|
| 143 | time, print_pos, value, oracle[time][print_pos]);
|
|---|
| 144 | print_pos++;
|
|---|
| 145 | }
|
|---|
| 146 |
|
|---|
| 147 | /* Prints the current values of u. */
|
|---|
| 148 | void write_frame() {
|
|---|
| 149 | if (rank != 0) {
|
|---|
| 150 | MPI_Send(u+1, nxl, MPI_DOUBLE, 0, 0, MPI_COMM_WORLD);
|
|---|
| 151 | } else {
|
|---|
| 152 | print_time_header();
|
|---|
| 153 | print_cell(lbound); // left boundary
|
|---|
| 154 | for (int source = 0; source < nprocs; source++) {
|
|---|
| 155 | int count;
|
|---|
| 156 |
|
|---|
| 157 | if (source != 0) {
|
|---|
| 158 | MPI_Status status;
|
|---|
| 159 |
|
|---|
| 160 | MPI_Recv(buf, 1+nx/nprocs, MPI_DOUBLE, source, 0, MPI_COMM_WORLD,
|
|---|
| 161 | &status);
|
|---|
| 162 | MPI_Get_count(&status, MPI_DOUBLE, &count);
|
|---|
| 163 | } else {
|
|---|
| 164 | for (int i = 1; i <= nxl; i++)
|
|---|
| 165 | buf[i-1] = u[i];
|
|---|
| 166 | count = nxl;
|
|---|
| 167 | }
|
|---|
| 168 | for (int i = 0; i < count; i++)
|
|---|
| 169 | print_cell(buf[i]);
|
|---|
| 170 | }
|
|---|
| 171 | print_cell(rbound); // right boundary
|
|---|
| 172 | //printf("\n");
|
|---|
| 173 | }
|
|---|
| 174 | }
|
|---|
| 175 |
|
|---|
| 176 | /* exchange_ghost_cells: updates ghost cells using MPI communication */
|
|---|
| 177 | void exchange_ghost_cells() {
|
|---|
| 178 | MPI_Sendrecv(&u[1], 1, MPI_DOUBLE, left, 0,
|
|---|
| 179 | &u[nxl+1], 1, MPI_DOUBLE, right, 0,
|
|---|
| 180 | MPI_COMM_WORLD, MPI_STATUS_IGNORE);
|
|---|
| 181 | MPI_Sendrecv(&u[nxl], 1, MPI_DOUBLE, right, 0,
|
|---|
| 182 | &u[0], 1, MPI_DOUBLE, left, 0,
|
|---|
| 183 | MPI_COMM_WORLD, MPI_STATUS_IGNORE);
|
|---|
| 184 | }
|
|---|
| 185 |
|
|---|
| 186 | /* Updates u_new using u, then swaps u and u_new.
|
|---|
| 187 | * Reads the ghost cells in u. Purely local operation. */
|
|---|
| 188 | void update() {
|
|---|
| 189 | for (int i = 1; i <= nxl; i++)
|
|---|
| 190 | u_new[i] = u[i] + k*(u[i+1] + u[i-1] - 2*u[i]);
|
|---|
| 191 | double * tmp = u_new; u_new=u; u=tmp;
|
|---|
| 192 | }
|
|---|
| 193 |
|
|---|
| 194 | /* Executes the simulation. */
|
|---|
| 195 | int main(int argc, char *argv[]) {
|
|---|
| 196 | MPI_Init(&argc, &argv);
|
|---|
| 197 | initialize();
|
|---|
| 198 | write_frame();
|
|---|
| 199 | printf("nx = %d", nx);
|
|---|
| 200 | for (time=1; time < nsteps; time++) {
|
|---|
| 201 | exchange_ghost_cells();
|
|---|
| 202 | update();
|
|---|
| 203 | if (time%wstep==0)
|
|---|
| 204 | write_frame();
|
|---|
| 205 | }
|
|---|
| 206 | MPI_Finalize();
|
|---|
| 207 | free(u);
|
|---|
| 208 | free(u_new);
|
|---|
| 209 | if (rank == 0)
|
|---|
| 210 | free(buf);
|
|---|
| 211 | return 0;
|
|---|
| 212 | }
|
|---|
| 213 |
|
|---|
| 214 |
|
|---|
| 215 |
|
|---|
| 216 |
|
|---|
| 217 | // ************************ Model **********************
|
|---|
| 218 |
|
|---|
| 219 | /* This is a model of a process. It should exhibit the same
|
|---|
| 220 | * interface as the real process, where the interface is interpreted
|
|---|
| 221 | * to mean the sequence of send/recv events.
|
|---|
| 222 | */
|
|---|
| 223 | void getNextSend(int dest, void *buf, MPI_Datatype *type, int* count);
|
|---|
| 224 | processRecv(source, message) : void
|
|---|