-
Notifications
You must be signed in to change notification settings - Fork 2
/
sparse_ls.py
97 lines (95 loc) · 3.19 KB
/
sparse_ls.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
#
# Author(s): Pierre Jolivet <[email protected]>
# Date: 2021-07-11
#
# Copyright (C) 2021- Centre National de la Recherche Scientifique
#
# This script is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
#
# If you use this script, you are kindly asked to cite the following article:
#
# "A Robust Algebraic Domain Decomposition Preconditioner for Sparse Normal Equations",
# H. Al Daas, P. Jolivet, and J. A. Scott (2022).
#
import petsc4py,sys
petsc4py.init(sys.argv)
from petsc4py import PETSc
A = PETSc.Mat().create(PETSc.COMM_WORLD)
# begin loading phase (section 4.1.1)
viewer = PETSc.Viewer().createBinary(PETSc.Options().getString('-mat_name'))
A.load(viewer)
viewer.destroy()
# end loading phase
x = PETSc.Vec()
b = PETSc.Vec()
c = PETSc.Vec()
ksp = PETSc.KSP().create(PETSc.COMM_WORLD)
ksp.setFromOptions()
pc = ksp.getPC()
normal = False
if pc.getType() == PETSc.PC.Type.QR:
ksp.setOperators(A,A)
ksp.setType(PETSc.KSP.Type.PREONLY)
else:
# begin partitioning phase (section 4.1.1)
B = A.transposeMatMult(A)
mpart = PETSc.MatPartitioning().create(PETSc.COMM_WORLD)
mpart.setFromOptions()
mpart.setAdjacency(B)
isp = PETSc.IS()
mpart.apply(isp)
rows = PETSc.IS()
rows = isp.buildTwoSided()
B = B.createSubMatrix(rows,rows)
(m,n) = A.getOwnershipRange()
isp = PETSc.IS().createStride(n-m,m,1,PETSc.COMM_WORLD)
A = A.createSubMatrix(isp,rows)
for obj in [mpart,isp,rows]:
obj.destroy()
# end partitioning phase
# begin setup phase (section 4.1.2)
B.setOption(PETSc.Mat.Option.NEW_NONZERO_ALLOCATION_ERR,False)
B.setOption(PETSc.Mat.Option.SYMMETRIC,True)
B.shift(B.norm() * 1.0e-10)
C = PETSc.Mat()
if not ksp.getType() == PETSc.KSP.Type.LSQR:
normal = True
C.createNormal(A)
ksp.setOperators(C if normal else A,B)
if pc.getType() == PETSc.PC.Type.HPDDM:
(m,n) = B.getOwnershipRange()
cols = PETSc.IS().createStride(n-m,m,1,PETSc.COMM_SELF)
(m,n) = A.getSize()
rows.createStride(m,0,1,PETSc.COMM_SELF)
A.setOption(PETSc.Mat.Option.SUBMAT_SINGLEIS,True)
Neumann = A.createSubMatrices(rows,cols)
isp = Neumann[0].findZeroRows()
B.increaseOverlap(cols)
Neumann = A.createSubMatrices(rows,cols)
Neumann[0].zeroRows(isp,0.0)
aux = PETSc.Mat()
aux = Neumann[0].transposeMatMult(Neumann[0])
aux.setOption(PETSc.Mat.Option.NEW_NONZERO_ALLOCATION_ERR,False)
aux.shift(aux.norm() * 1.0e-8)
pc.setHPDDMAuxiliaryMat(cols,aux)
pc.setHPDDMHasNeumannMat(True)
if not normal:
correction_type = pc.getHPDDMCoarseCorrectionType()
if correction_type == PETSc.PC.HPDDMCoarseCorrectionType.DEFLATED:
pc.setHPDDMCoarseCorrectionType(PETSc.PC.HPDDMCoarseCorrectionType.BALANCED)
# end setup phase
# begin solution phase (section 4.1.3)
(x,b) = A.createVecs()
b.setRandom()
c = x.duplicate()
A.multTranspose(b,c)
ksp.solve(c if normal else b,x)
# end solution phase (section 4.1.3)
b.scale(-1.0)
A.multAdd(x,b,b)
A.multTranspose(b,c)
norm = [b.norm(),c.norm()]
if not PETSc.COMM_WORLD.Get_rank():
print("||A^T(Ax-b)|| / ||Ax-b|| = "+str(norm[1])+" / "+str(norm[0])+" = "+str(norm[1]/norm[0]))