Loop Interchange

Loop interchange is an optimization technique that swaps the order of nested loops, making the inner loop the outer loop and vice versa. This transformation can improve data locality, optimize memory access patterns, and expose opportunities for parallelization and vectorization. By reordering loops, loop interchange can enhance cache efficiency and reduce execution time, especially in matrix operations or scenarios where data access patterns impact performance. It is particularly useful in optimizing code for modern processors and parallel computing architectures.

Loop Interchange Transformation in emmtrix Studio

emmtrix Studio implements loop interchange using #pragma directives or via the GUI. Loop interchange is a transformation that swaps two nested loops with each other, so that the inner loop becomes the outer one and vice versa.

Typical Usage and Benefits

Loop interchange is used to change granularity of the outer loops which has parallelization implications. Another important application of loop interchange is in scenarios in which vectorization possibilities could be exposed. By interchanging the loops some previously scattered data elements could now be localized.

Example

/* The following code tests loop interchange transformation applied to a for loop: */   

#define N 5
int main(void) {
    int i1, i2;
    int a[N][N] = { 0 };
    #pragma EMX_TRANSFORMATION LoopInterchange
    for (i1 = 0; i1 < N; i1++) {
        for (i2 = 0; i2 < N; i2++) {
            a[i1][i2] = a[i1][i2] + 5;
        }
    }
    return 0;
}

/* The generated code:*/

int main(void) {
int i1 ;
int i2 ;
int a [5][5] = {0};
for (i2 = 0; i2 < 5; i2 = i2 + 1) {
for (i1 = 0; i1 < 5; i1 = i1 + 1) {
a[i1 ][ i2 ] = a[i1 ][ i2 ] + 5;
}
}
return 0;
}

Loop Interchange