StrideB

StrideB is a term used in computer science to denote the memory distance between consecutive elements along the B dimension in a multi-dimensional array when stored in linear memory. In a three-dimensional array with axes A, B, and C stored in row-major (C-order) layout, the memory address of element (a, b, c) is base + ((a * B) + b) * C + c, all multiplied by the element size. From this, the typical strides are StrideA = B * C, StrideB = C, and StrideC = 1. StrideB thus represents how many contiguous data units you must skip to move from one B-indexed element to the next while keeping A constant and varying C.

In general, for a tensor with shape [D0, D1, D2, ..., Dn-1] stored in C-order, the stride for axis i equals the product of the sizes of all later axes. StrideB corresponds to the product of the dimensions to the right of B. If the array is transposed, sliced with a step, or otherwise reorganized, StrideB may differ from its contiguous-default value, reflecting a non-contiguous memory layout.

Usage and implications: StrideB is important for indexing, pointer arithmetic, and performance. Access patterns that use unit strides (where successive B elements are adjacent in memory) enable vectorization and cache-friendly memory access. Frameworks and languages expose strides as properties of arrays or tensors, and compilers optimize code based on actual stride information to maximize throughput.

Note: StrideB is not a universal standard term; in many contexts, strides are described as a vector for all axes. StrideB is a shorthand emphasis used when discussing memory layout with respect to the B dimension.