I'm trying to write a matrix multiplication code in UPC. If I don't use b_local and directly use b, its working fine. But when I use b_local via memget function, it crashes at the line "upc_memget" with the above error.
#define N 10 //Input Matrix A = N*P
#define P 10 //Input Matrix B = P*M
#define M 10 //Result Matrix C = N*M
shared [N*P /THREADS] double a[N][P] , c[N][M];
shared [M / THREADS] double b[P][M] ;
double b_local[P][M];
int main() {
//Initialization
if(MYTHREAD==0)
gettimeofday(&start_time,NULL);
upc_barrier;
upc_memget(b_local, b, P*M*sizeof(double));
for (k=0; k<ITER; k++) {
/* UPC_FORALL work-sharing construct for matrix multiplication */
upc_forall(i=0;i<N;i++;&a[i][0]) {
// &a[i][0] determines affinity
for (j=0; j<M; j++) {
c[i][j] = 0;
for(l=0; l< P; l++) c[i][j] +=a[i][l]*b_local[l][j];
}
}
}
upc_barrier;
if(MYTHREAD==0)
gettimeofday(&end_time,NULL);
}
upc_memget fetches a contiguous block of memory with affinity to a single thread. Given your declaration of b, there are roughly P*M/THREADS
elements with affinity to thread 0 and your call tries to fetch P*M
elements from that one thread - the crash occurs because you are trying to copy unallocated memory beyond the end of the elements with affinity to thread 0.