ELL (ELLPACK)

E839062

data structure matrix representation sparse matrix storage format

ELL (ELLPACK) is a sparse matrix storage format that stores each row with a fixed number of nonzero elements, enabling efficient and regular memory access patterns on parallel architectures like GPUs.

Try in SPARQL Jump to: Statements Referenced by

Statements (46)

Predicate	Object
instanceOf	data structure ⓘ matrix representation ⓘ sparse matrix storage format ⓘ
advantageOver	CSR for regular GPU SpMV due to more regular memory access ⓘ
alsoKnownAs	ELL format ⓘ ELLPACK sparse format ⓘ
comparedTo	COO (Coordinate) format ⓘ CSR (Compressed Sparse Row) format ⓘ
componentOf	hybrid sparse formats such as HYB (ELL+COO) ⓘ
dataStored	column indices of nonzero values ⓘ nonzero values ⓘ
designedFor	efficient parallel sparse matrix-vector multiplication ⓘ
disadvantageComparedTo	CSR for highly irregular sparsity patterns ⓘ
fullName	ELLPACK NERFINISHED ⓘ
hasProperty	best suited for matrices with nearly uniform number of nonzeros per row ⓘ each thread can process one row independently ⓘ enables coalesced memory accesses on GPUs ⓘ fixed-stride access pattern ⓘ good cache utilization for regular sparsity patterns ⓘ indexing typically uses (row, k) where k is position within fixed row length ⓘ inefficient for matrices with highly variable row lengths ⓘ may incur storage overhead due to padding ⓘ memory layout is typically column-major for the ELL arrays ⓘ no per-row pointer array is required ⓘ often used in GPU-accelerated linear algebra libraries ⓘ padding overhead increases with row length imbalance ⓘ row length equals maximum number of nonzeros in any row ⓘ rows are padded with zeros or dummy entries to a uniform length ⓘ simple indexing arithmetic ⓘ stores each matrix row with a fixed number of nonzero entries ⓘ supports efficient vectorized access across rows ⓘ supports straightforward parallelization over rows ⓘ uses regular memory access patterns ⓘ uses two main arrays: values and column indices ⓘ well-suited for GPU architectures ⓘ well-suited for SIMD architectures ⓘ well-suited for many-core processors ⓘ
originField	numerical linear algebra ⓘ
requires	knowledge of maximum nonzeros per row at construction time ⓘ
usedFor	SpMV NERFINISHED ⓘ sparse matrix-vector multiplication ⓘ storing sparse matrices ⓘ
usedIn	GPU computing ⓘ high-performance computing ⓘ iterative solvers for sparse linear systems ⓘ scientific computing ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

cuSPARSE → supportsMatrixFormat → ELL (ELLPACK) ⓘ