Module driver 

Shared CUDA driver presence/loading helpers used by every cuBLAS / cuSPARSE / cuSOLVER routing module.

The GPU path uses ONE context model: cudarc’s device PRIMARY context (cuDevicePrimaryCtxRetain, bound in device_runtime::cuda_context_for). cuBLAS/cuSOLVER/cuSPARSE handles attach to that current context; there is no separate user cuCtxCreate context (its removal fixed the #1017 NOT_INITIALIZED handle failures). This module keeps only the libcuda presence probes, byte-size/layout helpers, and the check_cuda status wrap.

Functions

array_from_row_major

Wrap a row-major flat buffer of shape (rows, cols) as an Array2<f64> without permutation. The buffer is consumed (no copy when its length matches). Returns None on a length mismatch.

check_cuda

cuda_driver_available

Lossless CUDA-driver presence probe. Ok(false) means every candidate was genuinely absent; loader/ABI/transitive-dependency faults remain Err.

cuda_library_candidates

from_col_major

from_col_major_inplace

Convert a column-major flat buffer back into row-major Array2<f64>.

preload_cuda_driver

require_cuda_compute_library

Require that the platform loader can open the named CUDA compute library (cublas, cusolver, cusparse) from the one selected userspace stack.

to_col_major

Repack a 2D ndarray::ArrayBase (row-major) into the column-major layout expected by every cuBLAS / cuSOLVER entry point.

to_i32

to_row_major

Borrow (or pack) a 2D array’s buffer in ROW-major (C) order.

Type Aliases

CuResult