Skip to main content

Module gpu_kernels

Module gpu_kernels 

Source

Modules§

arrow_schur
Fully GPU-resident batched Arrow-Schur dense Cholesky solver.
batched_k1
Batched independent K=1 border solves — the post-SAC GPU arrow-Schur target.
pirls_row
Generic GPU PIRLS row-reweight primitives.
reml_trace
GPU Hutchinson stochastic trace estimator for the REML/LAML logdet gradient, per math team block 2 (sections 12–18 of the V100 design).
sae_resident
Device-resident SAE inner-iteration workspace for issue #1017.