void enumerated_subrange<GRID,GT>::construct(CELLIT c)
{
REQUIRE_ALWAYS(cells.empty(), "enumerated_subrange<GRID>::construct: cells must be empty!",1);
while(! c.IsDone()) {
append_cell(c.handle());
++c;
}
init();
init_counts();
}
DOUBLE doAlign(unsigned char *seqA, unsigned char *seqB, int lenA, int lenB,
DOUBLE seqA_enc[][ALPHA_SIZE], DOUBLE seqA_cum[],
DOUBLE seqB_enc[][ALPHA_SIZE], DOUBLE seqB_cum[],
DOUBLE *final_counts /* If the caller wants the final counts */
)
{
DOUBLE mdlCost;
struct cell D[2][lenB+1];
int i,j;
struct val_counts empty, final;
#ifdef DO_COUNTS
init_counts(empty.counts);
init_counts(final.counts);
#endif
final.val = INFINITY;
int main (int argc, char *argv[])
{
int myid, numprocs, i;
int size;
float **s_buf, **r_buf;
double t_start = 0.0, t_end = 0.0;
if (argc == 2) {
root_id = atoi(argv[1]);
}
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &myid);
if(myid == root_id) {
fprintf(stderr, "Running test with root id %d and %d proceses\n", root_id, numprocs);
fflush(0);
}
init_counts(myid);
s_buf = (float **)malloc(numprocs*sizeof(float *));
r_buf = (float **)malloc(numprocs*sizeof(float *));
if (myid == root_id) {
for (i = 1; i < numprocs; i++) {
s_buf[i] = (float *)calloc(1, send_counts[(i-1)%26]*sizeof(float));
r_buf[i] = (float *)calloc(1, send_counts[(i-1)%26]*sizeof(float));
}
}
else {
s_buf[0] = (float *)calloc(1, send_counts[(myid-1)%26]*sizeof(float));
r_buf[0] = (float *)calloc(1, send_counts[(myid-1)%26]*sizeof(float));
}
loop = LOOP_LARGE;
skip = SKIP_LARGE;
MPI_Barrier(MPI_COMM_WORLD);
MPI_Request req[2*numprocs];
MPI_Status statuses[2*numprocs];
HPM hpm;
if(myid == root_id) {
for(i = 0; i < loop + skip; i++) {
if(i == skip) t_start = MPI_Wtime();
int rq, r, n;
if (i >= skip) hpm.HPM_Start("loop");
rq = 0;
for (n = 0; n < numprocs; n++) {
if (n == root_id) continue;
MPI_Irecv(r_buf[n], send_counts[(n-1)%26], MPI_FLOAT, n, 1, MPI_COMM_WORLD, &req[rq]);
rq++;
}
for (n = 0; n < numprocs; n++) {
if (n == root_id) continue;
MPI_Isend(s_buf[n], send_counts[(n-1)%26], MPI_FLOAT, n, 1, MPI_COMM_WORLD, &req[rq]);
rq++;
}
MPI_Waitall(rq, req, statuses);
if (i >= skip) hpm.HPM_Stop("loop");
}
t_end = MPI_Wtime();
}
else if (myid != root_id) {
for(i = 0; i < loop + skip; i++) {
int r, rq;
rq=0;
MPI_Irecv(r_buf[0], send_counts[(myid-1)%26], MPI_FLOAT, root_id, 1, MPI_COMM_WORLD, &req[rq]);
rq++;
#if 1
MPI_Isend(s_buf[0], send_counts[(myid-1)%26], MPI_FLOAT, root_id, 1, MPI_COMM_WORLD, &req[rq]);
rq++;
#else
MPI_Send(s_buf[0], send_counts[(myid-1)%26], MPI_FLOAT, root_id, 1, MPI_COMM_WORLD);
#endif
MPI_Waitall(rq, req, statuses);
}
}
if(myid == root_id) {
double latency = (t_end - t_start) * 1e6 / (1.0 * loop);
fprintf(stdout, "latency = %f\n", latency);
fflush(stdout);
hpm.HPM_Stats();
}
MPI_Finalize();
return 0;
}
