Closed
Description
Consider the following snippet
using SparseArrays
using LinearAlgebra
function setup(n)
e = ones(n)
e2 = -ones(n-1)
K = spdiagm(n, n, 0 => 2.0e, 1 => e2, -1 => e2)
Id = sparse(1.0*I, n, n)
return kron(K, Id) + kron(Id, K)
end
function solve(n)
A = setup(n)
F = ones(n^2)
@time A \ F
endIt just builds a discretization of a PDE and solves it. Now let us execute
solve(1156);
solve(1157);Using julia 1.9.3 I get
2.095628 seconds (66 allocations: 1.339 GiB, 0.08% gc time)
2.134111 seconds (66 allocations: 1.339 GiB, 0.08% gc time)
and all is fine. With julia 1.10.4 instead I get
1.966967 seconds (69 allocations: 1.339 GiB, 0.17% gc time)
7.775528 seconds (1.99 M allocations: 9.856 GiB, 2.97% gc time)
Thus, the second solve on a slightly larger system is suddenly much slower and allocates memory. The profiling shows that cholmod_l_analyze runs much longer. The same happens with the RC of 1.11.
Digging a little bit deeper with
A = setup(1157);
B = SparseArrays.CHOLMOD.Sparse(A);
@time F = SparseArrays.CHOLMOD.symbolic(B, perm=nothing);
@time cholesky!(F, A; shift = 1., check = true);
and looking at the output of top, one can see that symbolic runs only on a single core and allocates memory, cholesky! seems to be fine (and seems to run in parallel).
I also tried to fiddle around with
SparseArrays.CHOLMOD.getcommon()[].nthreads_max
(which is 1) after startup, but it does not seem to change anything.
Metadata
Metadata
Assignees
Labels
No labels