parallel_for

parallel_for

Hi,

I'm reading ITBB and there is something I cant get about the "load-balanced execution".

Supose one want to apply a function wich complexity depends upon the index of the array, e.g.,

class ApplyFoo {
float *const my_a;
public:
void operator() ( const blocked_range& r ) const {
float *a = my_a;
for( size_t i=r.begin();i!=r.end();i++){
for(size_t j=0;j!=i*i;j++)//complexity n^2
a[i] += j;
}
}
ApplyFoo( float a[] ) :
my_a(a)
{}
};

The cuestion is how this will be "load-balanced" if one have enough cores to make all the task run indepently?

thanks,
sebastian

2 Beiträge / 0 neu
Letzter Beitrag
Nähere Informationen zur Compiler-Optimierung finden Sie in unserem Optimierungshinweis.

The blocked_range used above is a "splittable" range. Some thread is going to get stuck with the upper end of i and be bogged down added j to a[something] but the other threads will be working on the other end of the range and finish their work more quickly. Then they will come looking for work and start stealing parts of the high-range task's assignment range. Thus the work will be balanced among the available threads.

Melden Sie sich an, um einen Kommentar zu hinterlassen.