View page source

warp.tile_cholesky_inplace#

warp.tile_cholesky_inplace(
A: Tile[Float, tuple[int, int]],
fill_mode: str,
) → None#

  • Kernel

Compute the Cholesky factorization of a symmetric positive-definite matrix A inplace.

When fill_mode="lower" (default), the lower triangle of A is replaced by L such that LL^T = A; the upper triangle is set to zero. When fill_mode="upper", the upper triangle of A is replaced by U such that U^T U = A; the lower triangle is set to zero.

The fill_mode parameter must be a compile-time constant.

Note: This inplace variant does not support automatic differentiation (adjoint computation), but offers improved performance and uses half the shared memory compared to the standard version.

Supported datatypes are:

  • float32

  • float64

Parameters:

  • A – A square, symmetric positive-definite matrix.

  • fill_mode – "lower" (default) or "upper". Must be a compile-time constant.