mathnet-numerics

Commit Graph

Author	SHA1	Message	Date
Christoph Ruegg	b4a0cc97ed	LA Perf: replace parallel copy loops with native copy II	14 years ago
Christoph Ruegg	2f0aabbe28	LA Perf: replace parallel copy loops with native copy	14 years ago
Christoph Ruegg	d91acbf523	LA Perf: rangesize tweaks	14 years ago
Christoph Ruegg	d3dc54b4c1	Reworking common parallelization routines #92	14 years ago
Iain McDonald	8e1cd36027	Alter parallelisation of matrix multiplication I empirically noticed that matrix multiplication was extremely slow for the matrices I was using i.e. multiplications were taking minutes rather than seconds. After tracking down the root of the problem, I found it was in the CacheObliviousMatrixMultiply() implementation, specifically, where the sum of the column/row dimensions were compared with the Control.ParallelizeOrder variable. For significantly skewed matrix multiplication (wide * tall), the parallelised redistribution greatly reduces performance. By removing the 'k' parameter from the comparison (where k is the number of columns in the left matrix and the number of rows in the right) the performance is improved drastically. For perfectly square matrices, this change reduces performance slightly.	14 years ago
Christoph Ruegg	f8c7e4a68d	Comments: fix some inline xml doc; Drop redundant resources in Numerics.IO	14 years ago
Marcus Cuda	fe9e158777	added thin qr	14 years ago
Marcus Cuda	08f1483e87	tweaked cholesky solve fixed issue with LU solve for complex types using parallel element copy instead of Array.Copy	15 years ago
Marcus Cuda	57da4a187e	vector parallelization tweaks	15 years ago
Marcus Cuda	b4c8440e57	tweaked vector scale routines	16 years ago
Marcus Cuda	89aa96eec8	started optimizing the parallel code for small matrices and vectors	16 years ago
Marcus Cuda	435eaff230	fixed matrix multiple bug when using more than two threads	16 years ago
Abratiychuk	37526cb8d9	Change MatrixMultuplyWithUpdate to cache-oblivious multiplication.	16 years ago
Marcus Cuda	34bc517324	native: added SVD and modified the linear algebra provider as needed	16 years ago
Marcus Cuda	6beb3b6ace	native: added QR solve, tweaking the provider interface as needed	16 years ago
Marcus Cuda	aba8c988cb	native: added QR factor	16 years ago
Marcus Cuda	de4eb2a7a5	native: added Cholesky solve and removed redundant number of B parameter from its interface	16 years ago
Marcus Cuda	60f4ee5cc5	native: added LU solve api: removed the LU solve methods with the transpose option from LA provider interface for consistency with other solvers added SecuritySafeCritical and SecurityCritical attributes native provider and p/invoke methods	16 years ago
Marcus Cuda	2cf99d8184	commented out CommonParallel.Aggregate modified the affected code	16 years ago
Marcus Cuda	289dad9181	removed redundant copyto's	16 years ago
Marcus Cuda	f4b3f0c6bf	native: a little more reorganizing and stylecop cleanup	16 years ago
Marcus Cuda	71ccf00c9b	native: native providers now extend mananaged provider split native providers it type based templates	16 years ago
Marcus Cuda	0693388f8c	matrix: moved dense norm code into managed la provider	16 years ago

23 Commits (5f45c842c5220d7072cb959f01efeaf6908d62e3)