warp.tile_upper_solve_inplace

warp.tile_upper_solve_inplace(

U: Tile[Float, tuple[int, int]],

z: Tile[Float, tuple[int]],

) → None

Kernel

Solve for x in Ux = z, where U is an upper triangular matrix by overwriting z with x.

This performs general back substitution for upper triangular systems inplace.

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

Supported datatypes are:

• float32

• float64

param U:

A square, non-singular, upper triangular matrix

param z:

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