mesytec-mnode/external/taskflow-3.8.0/benchmarks/wavefront/omp.cpp
2025-01-04 01:25:05 +01:00

82 lines
2.1 KiB
C++

#include "matrix.hpp"
#include <omp.h>
int **D {nullptr};
// wavefront computation
void wavefront_omp(unsigned num_threads) {
// set up the dependency matrix
D = new int *[MB];
for(int i=0; i<MB; ++i) D[i] = new int [NB];
for(int i=0; i<MB; ++i){
for(int j=0; j<NB; ++j){
D[i][j] = 0;
}
}
omp_set_num_threads(num_threads);
#pragma omp parallel
{
#pragma omp single
{
matrix[M-1][N-1] = 0;
for( int k=1; k <= 2*MB-1; k++) {
int i, j;
if(k <= MB){
i = k-1;
j = 0;
}
else{
//assume matrix is square
i = MB-1;
j = k-MB;
}
for(; (k <= MB && i>=0) || (k > MB && j <= NB-1) ; i--, j++){
if(i > 0 && j > 0){
#pragma omp task depend(in:D[i-1][j], D[i][j-1]) depend(out:D[i][j]) firstprivate(i, j)
block_computation(i, j);
}
//top left corner
else if(i == 0 && j == 0){
#pragma omp task depend(out:D[i][j]) firstprivate(i, j)
block_computation(i, j);
}
//top edge
else if(j+1 <= NB && i == 0 && j > 0){
#pragma omp task depend(in:D[i][j-1]) depend(out:D[i][j]) firstprivate(i, j)
block_computation(i, j);
}
//left edge
else if(i+1 <= MB && i > 0 && j == 0){
#pragma omp task depend(in:D[i-1][j]) depend(out:D[i][j]) firstprivate(i, j)
block_computation(i, j);
}
//bottom right corner
else if(i == MB-1 && j == NB-1){
#pragma omp task depend(in:D[i-1][j] ,D[i][j-1]) firstprivate(i, j)
block_computation(i, j);
}
else{
assert(false);
}
}
}
#pragma omp taskwait
}
}
for ( int i = 0; i < MB; ++i ) delete [] D[i];
delete [] D;
}
std::chrono::microseconds measure_time_omp(unsigned num_threads) {
auto beg = std::chrono::high_resolution_clock::now();
wavefront_omp(num_threads);
auto end = std::chrono::high_resolution_clock::now();
return std::chrono::duration_cast<std::chrono::microseconds>(end - beg);
}