Batched QR factorization of sparse matrices on GPUs. Wrapper of the CUDA library cusolverSpDcsrqrsvBatched()
apt-get install libsuitesparse-dev
pip install GBQsparseThis package solves multiple linear systems, A x=b, sharing the same sparsity pattern.
Below are some of the resources I have used in writing GBQsparse:
High-level discussion of Batched QR factorization
from GBQsparse import MSparse
import scipy.sparse as sp
import numpy as np
import time
import scipy.sparse.linalg as sla
N = 10
nbatch = 10
m = sp.diags([1, -2, 1], [-1, 0, 1], shape=(N, N),format='coo')
A = np.random.random_sample((nbatch,m.nnz))
B = np.random.random_sample((nbatch,N))
m = MSparse(m.row,m.col,N,nbatch,reordering=True)
m.add_LHS(A)
t1 = time.time()
X = m.solve(B)
t2 = time.time()
m.free_memory()
xs = []
for i in range(nbatch):
S = sp.csr_matrix((A[i],(m.row,m.col)),shape=(N,N),dtype=float)
x = sla.spsolve(S,B[i])
xs.append(x)
xs=np.array(xs)
t3 = time.time()
print(t2-t1)
print(t3-t2)
print(np.allclose(xs,X,rtol=1e-01,atol=1e-1))