openmp parallel for loop with two or more reductions
Solution 1
You can do reduction by specifying more than one variable separated by a comma, i.e. a list:
#pragma omp parallel for default(shared) reduction(+:sum,result) ...
Private thread variables will be created for sum
and result
that will be combined using +
and assigned to the original global variables at the end of the thread block.
Also, variable y
should be marked private.
See https://computing.llnl.gov/tutorials/openMP/#REDUCTION
Solution 2
You can simply add another reduction
clause:
#include <iostream>
#include <cmath>
int main(){
double sum_i = 0, max_i = -1;
#pragma omp parallel for reduction(+:sum_i) reduction(max:max_i)
for (int i=0; i<5000; i++){
sum_i += i;
if (i > max_i)
max_i = i;
}
std::cout << "Sum = " << sum_i << std::endl;
std::cout << "Max = " << max_i << std::endl;
return 0;
}
From OpenMP 4.5 Complete Specifications (Nov 2015)
Any number of reduction clauses can be specified on the directive, but a list item can appear only once in the reduction clauses for that directive.
The same works on Visual C++ that uses oMP v2.0: reduction VC++
Related videos on Youtube
pyCthon
Updated on July 09, 2022Comments
-
pyCthon almost 2 years
Hi just wondering if this is the right way to go going about having a regular for loop but with two reductions , is this the right approach below? Would this work with more then two reductions as well. Is there a better way to do this? also is there any chance to integrate this with an MPI_ALLREDUCE command?
heres the psuedo code #pragma omp parallel for \ default(shared) private(i) \ //todo first reduction(+:sum) //todo second reduction(+:result) for loop i < n; i ++; { y = fun(x,z,i) sum += fun2(y,x) result += fun3(y,z) }
-
worenga over 8 yearsWhat if there are different operations to be performed e.g.
+
andmax
? -
Azmisov about 8 years@mightyuhu See my answer
-
Royi over 6 yearsIt doesn't work in GCC 7.1. Any idea how to apply 2 different reductions?
-
Marv almost 6 years@Royi maybe it's because of the comma after the first reduction term? I don't recall commas being used to distinguish omp parameters.