warp.tile_lower_solve_inplace#

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

  • Kernel

Solve for z in Lz = y, where L is a lower triangular matrix by overwriting y with z.

This performs general forward substitution for a lower triangular system inplace.

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

Supported datatypes are:

  • float32

  • float64

param L:

A square, non-singular, lower triangular matrix

param y:

A 1D or 2D tile with compatible shape that gets overwritten by z where Lz = y