def iddr_rid(queue, m, n, matvect, krank):
id_srand = util.setup_rand(queue)
blas.setup()
dtype = 'float64'
clx = cl_array.zeros(queue, m, dtype)
rnorms = cl_array.Array(queue, n, dtype)
lst = cl_array.Array(queue, n, np.int32)
proj = cl_array.Array(queue, (krank+2,n), dtype)
l = krank + 2
for i in range(l):
id_srand(m, clx)
matvect(clx, proj[i,:])
iddr_id(queue, l, n, proj, krank, lst, rnorms)
blas.teardown()
return lst, proj
rng = np.random.RandomState(1) # change the seed to see different data
A[...] = rng.uniform(-1, 1, size=A.shape)
x[...] = rng.uniform(-1, 1, size=x.shape)
y[...] = rng.uniform(-1, 1, size=y.shape)
# allocate OpenCL memory on the device
clA = Array(queue, A.shape, A.dtype)
clx = Array(queue, x.shape, x.dtype)
cly = Array(queue, y.shape, y.dtype)
# copy data to device
clA.set(A)
clx.set(x)
# compute a matrix-vector product (gemv)
blas.gemv(queue, clA, clx, cly)
# check the result
print("Expected: ", np.dot(A, x))
print("Actual: ", cly.get())
# try a matrix-vector product with the transpose
cly.set(y)
blas.gemv(queue, clA, cly, clx, transA=True)
print("Expected: ", np.dot(A.T, y))
print("Actual: ", clx.get())
# tidy up the BLAS
blas.teardown()