unitstride

Unitstride, or unit stride, is a term used in computing to describe a memory access pattern in which successive elements of a sequence are located at consecutive memory addresses; the stride, defined as the index difference between successive elements, is 1.

In multi-dimensional arrays, unit stride refers to iterating along an axis such that each successive element is stored next to the previous one in memory. For languages with row-major order (such as C), iterating across columns within a row yields unit stride; iterating down a column yields non-unit stride. In column-major languages (such as Fortran), the inverse holds, with unit stride along the appropriate dimension.

Unit-stride access benefits from spatial locality, improving cache efficiency and enabling compiler vectorization. Non-unit stride can cause cache misses and lower performance. For example, in a row-major 2D array A, a row-wise traversal (for i = 0 to rows-1; for j = 0 to cols-1; sum += A[i][j];) accesses memory with unit stride along the inner loop, whereas a column-wise traversal (for j = 0 to cols-1; for i = 0 to rows-1; sum += A[i][j];) incurs non-unit stride.

In practice, programmers aim to write loops that access arrays with unit stride to improve performance. Some languages and compilers offer tools or annotations to help enforce or detect stride-1 patterns, and many vectorized or parallel operations assume unit-stride access for efficiency. Understanding stride and contiguity is important when optimizing numerical kernels and memory-heavy algorithms.