Skip to main content

Module gemv_transpose_strided

Module gemv_transpose_strided 

Source
Expand description

Generic runtime-dispatch transposed sub-matrix GEMV (y += Aᵀ · x, row stride lda).

Generalizes super::gemv_transpose() to a row-major sub-matrix: nrows × ncols with leading dimension lda ≥ ncols. lda = ncols recovers the packed transpose. Computes Σᵢ xᵢ·A[i,:] (sum of the strided rows scaled by x), vectorizing across the ncols output lanes with no horizontal reduction. Admits the Aᵀ·x reduction over a trailing/leading block of a larger buffer — e.g. forming Aw = Σⱼ wⱼ·colⱼ in a reflector apply — without copying it out. Accumulates into y.