warp.tile_cholesky_solve_inplace#

warp.tile_cholesky_solve_inplace(
L: Tile[Float, tuple[int, int]],
y: Tile[Float, tuple[int]],
) → None#

  • Kernel

With L such that LL^T = A, solve for x in Ax = y by overwriting y with x

Note: This inplace variant does not support automatic differentiation (adjoint computation), but avoids allocating shared memory for the output x by reusing y’s memory.

Supported datatypes are:

  • float32

  • float64

param L:

A square, lower triangular, matrix, such that LL^T = A

param y:

A 1D or 2D tile of length M that gets overwritten by x where LL^T x = y