PWR034: avoid strided array access to improve performance

Issue

Strided array access may impact performance.

Relevance

Accessing an array using a non-unit stride is less efficient than accessing consecutive positions because the latter improves the  locality of reference. Note that C/C++ column-wise matrix access is an example non-unit stride, where the stride is the column width.

Actions

Consider using techniques like loop fusion, loop interchange, loop tiling or changing the data layout to avoid non-sequential access in hot loops.

Code example

The following code accesses both variables a and b with a stride LEN.

        for (int i = 0; i < LEN; ++i) {
            for (int j = 1; j < LEN; j++) {
                a[j * LEN + i] = a[j * LEN + i - 1] + b[j * LEN + i];
            }
        }

By doing loop interchange of loops over a and b, we get rid of the strided accesses with regard to the innermost loop.

 for (int j = 1; j < LEN; ++j) {
            for (int i = 0; i < LEN; i++) {
                a[j * LEN + i] = a[j * LEN + i - 1] + b[j * LEN + i];
            }
        }