MRCQR stabilizes Cholesky-QR via a subsampled randomized trigonometric transform preconditioner computed in FP32 or FP16, achieving orthogonality error O(u) up to condition number 10^16 while outperforming rand-cholQR and cuSOLVER geqrf on H100 GPUs.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.NA 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Numerically Stable Cholesky-QR on GPU via Mixed-Precision Randomized Preconditioning
MRCQR stabilizes Cholesky-QR via a subsampled randomized trigonometric transform preconditioner computed in FP32 or FP16, achieving orthogonality error O(u) up to condition number 10^16 while outperforming rand-cholQR and cuSOLVER geqrf on H100 GPUs.