Reduce the number of symbolic analyses #104

cvanaret · 2024-11-19T00:14:25Z

Tasks:

pass WarmstartInformation object to factorize_matrix();
compute symbolic analysis only when the Hessian sparsity changes;
figure out why MUMPS behaves differently when there is a single analysis at the beginning (apparently a different scaling).

… moment) + minor refactoring

…roblem has changed

…ing in MUMPS to recompute scaling before each factorization

cvanaret changed the title ~~Reduce the number of symbolical factorizations~~ WIP: Reduce the number of symbolical factorizations Nov 19, 2024

cvanaret changed the title ~~WIP: Reduce the number of symbolical factorizations~~ WIP: Reduce the number of symbolic factorizations Nov 20, 2024

cvanaret force-pushed the improve_factorizations branch from ddc1589 to 7cb8e6a Compare December 3, 2024 21:33

cvanaret changed the title ~~WIP: Reduce the number of symbolic factorizations~~ Reduce the number of symbolic analyses Dec 4, 2024

cvanaret added 11 commits December 5, 2024 08:19

Removed fixed_hessian_sparsity flag in Model

2462129

Minor refactoring

11af9bb

Pass warmstart object to factorize_matrix() (without using it for the…

efc3fd5

… moment) + minor refactoring

Trying out other MUMPS parameters

d74fa6f

Restored previous MUMPS settings

fadf598

MUMPSSolver: replaced member COO_matrix with two vectors

de1c6d6

MUMPSSolver: disable scaling during analysis

87ecfdd

Relaxed threshold for unbounded problems

97acb41

SymmetricIndefiniteLinearSystem: perform the analysis only when the p…

5f08f59

…roblem has changed

Temporarily restore analysis at each iteration

30c7304

Perform analysis only when the problem structure changes + set a sett…

5670b8e

…ing in MUMPS to recompute scaling before each factorization

cvanaret force-pushed the improve_factorizations branch from d912ef8 to 5670b8e Compare December 5, 2024 07:20

cvanaret merged commit 3a92daa into main Dec 5, 2024
7 checks passed

cvanaret deleted the improve_factorizations branch December 5, 2024 08:12

Provide feedback