def schur(self, A, schur_list, full_matrix=True):
"""
Setup the solver object to work in Schur complement mode
full_matrix: boolean, default True. If full_matrix is False
and a symmetric factorization is used, only the lower part of
the Schur matrix is stored in self.S
"""
self.A = A
self.spmA = spm.spmatrix(A)
if self.verbose:
self.spmA.printInfo()
self.schur_list = schur_list + self.spmA.findBase()
setSchurUnknownList(self.pastix_data, self.schur_list)
task_analyze(self.pastix_data, self.spmA)
task_numfact(self.pastix_data, self.spmA)
nschur = len(schur_list)
self.S = np.zeros((nschur, nschur), order='F', dtype=A.dtype)
getSchur(self.pastix_data, self.S)
if full_matrix and (self.factotype != factotype.LU):
self.S += la.tril(self.S, -1).T
def setup(self, A):
"""
Setup the solver object for the classic case w/o Schur complement
"""
self.A = A
self.spmA = spm.spmatrix(A)
if self.verbose:
self.spmA.printInfo()
task_analyze(self.pastix_data, self.spmA)
task_numfact(self.pastix_data, self.spmA)
Univ. Bordeaux. All rights reserved.
@version 6.0.0
@author Pierre Ramet
@author Mathieu Faverge
@author Louis Poirel
@date 2017-05-04
"""
import spm
import scipy.sparse as sps
import numpy as np
# Hack to make sure that the mkl is loaded
tmp = np.eye(2).dot(np.ones(2))
# Set the problem
n = 9
A = sps.spdiags([np.ones(n) * i for i in [4, -1, -1, -1, -1]],
[0, 1, 3, -1, -3], n, n)
x0 = np.arange(n)
b = np.zeros(n)
spmA = spm.spmatrix(A)
# Multiply A by x
spmA.mult(x0, b, trans=spm.trans.NoTrans, alpha=1., beta=0.)
# Check that A * x = b
spmA.checkAxb(None, b, x0)
@version 6.0.0 @author Pierre Ramet @author Mathieu Faverge @author Louis Poirel @date 2017-05-04 """ import spm import numpy as np # Hack to make sure that the mkl is loaded tmp = np.eye(2).dot(np.ones(2)) # Load a sparse matrix from the Laplacian driver A = spm.spmatrix(None, driver=spm.driver.Laplacian, filename="10:10:10:2.:1.") # Example from a HB file #A = spm( None, driver=driver.HB, filename="$PASTIX_DIR/test/matrix/orsirr.rua" ) A.printInfo() # Scale A for low-rank: A / ||A||_f norm = A.norm() A.scale(1. / norm) # Generate b and x0 vectors such that A * x0 = b nrhs = 10 x0, b = A.genRHS(spm.rhstype.RndX, nrhs) # Check that A * x = b
# Hack to make sure that the mkl is loaded
tmp = np.eye(2).dot(np.ones(2))
# Initialize parameters to default values
iparm, dparm = pastix.initParam()
# Startup PaStiX
pastix_data = pastix.init(iparm, dparm)
# Change some parameters
iparm[pastix.iparm.factorization] = pastix.factotype.LU
for nspb in range(1, 3):
# Load a sparse matrix from HB driver
spmA = spm.spmatrix(None,
driver=spm.driver.Laplacian,
filename=("%d:%d:%d:4.:1." %
(nspb * 5, nspb * 5, nspb * 5)))
#spmA = spm.spmatrix( None, driver=driver.HB, filename="$PASTIX_DIR/test/matrix/orsirr.rua" )
spmA.printInfo()
# Scale A for low-rank: A / ||A||_f
norm = spmA.norm()
spmA.scale(1. / norm)
# Perform analyze
pastix.task_analyze(pastix_data, spmA)
for nfact in range(1, 3):
# Perform numerical factorization
pastix.task_numfact(pastix_data, spmA)
