Parallel Computing for Data Science

5.8. FURTHER CACHE/VIRTUAL MEMORY ISSUES 145

ful of such issues when we write shared-memory code. And remember, the

problem is especially acute in multicore settings, due to cache coherency

issues (Section 2.5.1.1).

To make this idea concrete, we’ll look at two OpenMP programs to do

in-place matrix transpose. Here’s the first:

// CacheRow . c

#include <omp. h>

#include <s t d l i b . h>

#include <s t d i o . h>

// t r a n s l a t e from 2−D to 1−D i n d i c e s

int onedim ( int n , int i , int j ) { return n ∗ i + j ; }

void t r a n sp ( int ∗m, int n)

{

#pragma omp p a r a l l e l

{ int i , j , tmp ;

// walk through a l l the above−dia g on a l elements ,