Samples

Runnable examples, one project per feature area. Each sample verifies its own results and exits with a non-zero code on failure, so they double as smoke tests. They need a CUDA-capable GPU and the native libraries for the modules they use (see Installing the native libraries). The tensor and cuDSS parts skip themselves with a message when cuTENSOR or cuDSS is not installed.

dotnet run --project samples/FftAdvancedSample

Sample	Shows
FftSample	Basic 1-D FFT and the one-shot helpers
FftAdvancedSample	N-D and batched transforms, leading-axis layouts, real-to-complex round trips, output normalization
MatmulSample	A first cuBLASLt matrix product
MatmulAdvancedSample	Bias epilog, strided-batch matmul, autotuning with the plan cache, complex matmul with conjugate transpose
ChainingSample	Chaining operations on one stream, multi-stream patterns
SparseSample	SpMV and SpMM on a CSR matrix
SparseAdvancedSample	Triangular solves, SpGEMM, CSR/CSC/COO and transpose
DirectSolverSample	cuDSS sparse direct solver: analyse / factorise / solve
TensorSample	A cuTENSOR contraction
TensorOpsSample	Permutation, reduction, element-wise binary and contraction with mode labels
SolverSample	LU, Cholesky, SVD and eigenvalues
LeastSquaresSample	QR, polynomial least-squares fitting, factor-once/solve-many LU, batched LU and SVD
ComplexSample	Complex dense solvers, SVD, Hermitian eigenproblem, complex sparse products, Hermitian cuDSS solve
RandSample	cuRAND generators and distributions
MemoryAndLoggingSample	The caching memory manager, DeviceMemory.Default and the NVMathLog hook

Layout reminder: BLAS-style modules (matmul, solvers, sparse dense operands) and tensors are column-major; the FFT module follows cuFFT and is row-major (the last axis is contiguous).