class OperatorLibsTestCase(OptkitCTestCase):
"""TODO: docstring"""
@classmethod
def setUpClass(self):
self.env_orig = os.getenv('OPTKIT_USE_LOCALLIBS', '0')
os.environ['OPTKIT_USE_LOCALLIBS'] = '1'
self.libs = OperatorLibs()
self.A_test = self.A_test_gen
self.A_test_sparse = self.A_test_sparse_gen
@classmethod
def tearDownClass(self):
os.environ['OPTKIT_USE_LOCALLIBS'] = self.env_orig
def setUp(self):
self.x_test = np.random.rand(self.shape[1])
def tearDown(self):
self.free_all_vars()
self.exit_call()
def validate_operator(self, operator_, m, n, OPERATOR_KIND):
o = operator_
self.assertEqual( o.size1, m )
self.assertEqual( o.size2, n )
self.assertEqual( o.kind, OPERATOR_KIND )
self.assertNotEqual( o.data, 0 )
self.assertNotEqual( o.apply, 0 )
self.assertNotEqual( o.adjoint, 0 )
self.assertNotEqual( o.fused_apply, 0 )
self.assertNotEqual( o.fused_adjoint, 0 )
self.assertNotEqual( o.free, 0 )
def exercise_operator(self, lib, operator_, A_py, TOL):
o = operator_
m, n = A_py.shape
RTOL = TOL
ATOLM = TOL * m**0.5
ATOLN = TOL * n**0.5
alpha = np.random.rand()
beta = np.random.rand()
# allocate vectors x, y
x, x_, x_ptr = self.register_vector(lib, n, 'x')
y, y_, y_ptr = self.register_vector(lib, m, 'y')
x_ += self.x_test
self.assertCall( lib.vector_memcpy_va(x, x_ptr, 1) )
# test Ax
Ax = A_py.dot(x_)
self.assertCall( o.apply(o.data, x, y) )
self.assertCall( lib.vector_memcpy_av(y_ptr, y, 1) )
self.assertVecEqual( y_, Ax, ATOLM, RTOL )
# test A'y
y_[:] = Ax[:] # (update for consistency)
Aty = A_py.T.dot(y_)
self.assertCall( o.adjoint(o.data, y, x) )
self.assertCall( lib.vector_memcpy_av(x_ptr, x, 1) )
self.assertVecEqual( x_, Aty, ATOLN, RTOL )
# test Axpy
x_[:] = Aty[:] # (update for consistency)
Axpy = alpha * A_py.dot(x_) + beta * y_
self.assertCall( o.fused_apply(o.data, alpha, x, beta, y) )
self.assertCall( lib.vector_memcpy_av(y_ptr, y, 1) )
self.assertVecEqual( y_, Axpy, ATOLM, RTOL )
# test A'ypx
y_[:] = Axpy[:] # (update for consistency)
Atypx = alpha * A_py.T.dot(y_) + beta * x_
self.assertCall( o.fused_adjoint(o.data, alpha, y, beta, x) )
self.assertCall( lib.vector_memcpy_av(x_ptr, x, 1) )
self.assertVecEqual( x_, Atypx, ATOLN, RTOL )
self.free_vars('x', 'y')
def test_libs_exist(self):
libs = []
for (gpu, single_precision) in self.CONDITIONS:
libs.append(self. libs.get(single_precision=single_precision,
gpu=gpu))
self.assertTrue(any(libs))
def test_dense_alloc_free(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
for order in (lib.enums.CblasRowMajor, lib.enums.CblasColMajor):
A, _, _, = self.register_matrix(lib, m, n, order, 'A')
o = lib.dense_operator_alloc(A)
self.register_var('o', o.contents.data, o.contents.free)
self.validate_operator(o.contents, m, n, lib.enums.DENSE)
self.free_vars('o', 'A')
self.assertCall( lib.ok_device_reset() )
def test_dense_operator(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
DIGITS = 7 - 2 * single_precision - 1 * gpu
TOL = 10**(-DIGITS)
for rowmajor in (True, False):
order = lib.enums.CblasRowMajor if rowmajor else \
lib.enums.CblasColMajor
A, A_, A_ptr = self.register_matrix(lib, m, n, order, 'A')
A_ += self.A_test
self.assertCall( lib.matrix_memcpy_ma(A, A_ptr, order) )
o = lib.dense_operator_alloc(A)
self.register_var('o', o.contents.data, o.contents.free)
self.exercise_operator(lib, o.contents, A_, TOL)
self.free_vars('o', 'A')
self.assertCall( lib.ok_device_reset() )
def test_sparse_alloc_free(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
for rowmajor in (True, False):
order = lib.enums.CblasRowMajor if rowmajor else \
lib.enums.CblasColMajor
enum = lib.enums.SPARSE_CSR if rowmajor else \
lib.enums.SPARSE_CSC
A, _, _, _, _, _ = self.register_sparsemat(
lib, self.A_test_sparse, order, 'A')
o = lib.sparse_operator_alloc(A)
self.register_var('o', o.contents.data, o.contents.free)
self.validate_operator(o.contents, m, n, enum)
self.free_vars('o', 'A')
self.assertCall( lib.ok_device_reset() )
def test_sparse_operator(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
DIGITS = 7 - 2 * single_precision - 1 * gpu
TOL = 10**(-DIGITS)
for order in (lib.enums.CblasRowMajor, lib.enums.CblasColMajor):
hdl = self.register_sparse_handle(lib, 'hdl')
A, A_, A_sp, A_val, A_ind, A_ptr = self.register_sparsemat(
lib, self.A_test_sparse, order, 'A')
self.assertCall( lib.sp_matrix_memcpy_ma(hdl, A, A_val, A_ind,
A_ptr) )
o = lib.sparse_operator_alloc(A)
self.register_var('o', o.contents.data, o.contents.free)
self.exercise_operator(lib, o.contents, A_, TOL)
self.free_vars('o', 'A', 'hdl')
self.assertCall( lib.ok_device_reset() )
def test_diagonal_alloc_free(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
for rowmajor in (True, False):
d = lib.vector(0, 0, None)
self.assertCall( lib.vector_calloc(d, n) )
self.register_var('d', d, lib.vector_free)
o = lib.diagonal_operator_alloc(d)
self.register_var('o', o.contents.data, o.contents.free)
self.validate_operator(o.contents, n, n, lib.enums.DIAGONAL)
self.free_vars('o', 'd')
self.assertCall( lib.ok_device_reset() )
def test_diagonal_operator(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
DIGITS = 7 - 2 * single_precision - 1 * gpu
TOL = 10**(-DIGITS)
for rowmajor in (True, False):
d, d_, d_ptr = self.register_vector(lib, n, 'd')
d_ += self.A_test[0, :]
self.assertCall( lib.vector_memcpy_va(d, d_ptr, 1) )
o = lib.diagonal_operator_alloc(d)
self.register_var('o', o.contents.data, o.contents.free)
self.exercise_operator(lib, o.contents, np.diag(d_), TOL)
self.free_vars('o', 'd')
self.assertCall( lib.ok_device_reset() )
class OperatorLibsTestCase(OptkitCTestCase):
"""TODO: docstring"""
@classmethod
def setUpClass(self):
self.env_orig = os.getenv("OPTKIT_USE_LOCALLIBS", "0")
os.environ["OPTKIT_USE_LOCALLIBS"] = "1"
self.libs = OperatorLibs()
self.A_test = self.A_test_gen
self.A_test_sparse = self.A_test_sparse_gen
@classmethod
def tearDownClass(self):
os.environ["OPTKIT_USE_LOCALLIBS"] = self.env_orig
def setUp(self):
self.x_test = np.random.rand(self.shape[1])
def tearDown(self):
self.free_all_vars()
self.exit_call()
def validate_operator(self, operator_, m, n, OPERATOR_KIND):
o = operator_
self.assertEqual(o.size1, m)
self.assertEqual(o.size2, n)
self.assertEqual(o.kind, OPERATOR_KIND)
self.assertNotEqual(o.data, 0)
self.assertNotEqual(o.apply, 0)
self.assertNotEqual(o.adjoint, 0)
self.assertNotEqual(o.fused_apply, 0)
self.assertNotEqual(o.fused_adjoint, 0)
self.assertNotEqual(o.free, 0)
def exercise_operator(self, lib, operator_, A_py, TOL):
o = operator_
m, n = A_py.shape
RTOL = TOL
ATOLM = TOL * m ** 0.5
ATOLN = TOL * n ** 0.5
alpha = np.random.rand()
beta = np.random.rand()
# allocate vectors x, y
x, x_, x_ptr = self.register_vector(lib, n, "x")
y, y_, y_ptr = self.register_vector(lib, m, "y")
x_ += self.x_test
self.assertCall(lib.vector_memcpy_va(x, x_ptr, 1))
# test Ax
Ax = A_py.dot(x_)
self.assertCall(o.apply(o.data, x, y))
self.assertCall(lib.vector_memcpy_av(y_ptr, y, 1))
self.assertVecEqual(y_, Ax, ATOLM, RTOL)
# test A'y
y_[:] = Ax[:] # (update for consistency)
Aty = A_py.T.dot(y_)
self.assertCall(o.adjoint(o.data, y, x))
self.assertCall(lib.vector_memcpy_av(x_ptr, x, 1))
self.assertVecEqual(x_, Aty, ATOLN, RTOL)
# test Axpy
x_[:] = Aty[:] # (update for consistency)
Axpy = alpha * A_py.dot(x_) + beta * y_
self.assertCall(o.fused_apply(o.data, alpha, x, beta, y))
self.assertCall(lib.vector_memcpy_av(y_ptr, y, 1))
self.assertVecEqual(y_, Axpy, ATOLM, RTOL)
# test A'ypx
y_[:] = Axpy[:] # (update for consistency)
Atypx = alpha * A_py.T.dot(y_) + beta * x_
self.assertCall(o.fused_adjoint(o.data, alpha, y, beta, x))
self.assertCall(lib.vector_memcpy_av(x_ptr, x, 1))
self.assertVecEqual(x_, Atypx, ATOLN, RTOL)
self.free_vars("x", "y")
def test_libs_exist(self):
libs = []
for (gpu, single_precision) in self.CONDITIONS:
libs.append(self.libs.get(single_precision=single_precision, gpu=gpu))
self.assertTrue(any(libs))
def test_dense_alloc_free(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
for order in (lib.enums.CblasRowMajor, lib.enums.CblasColMajor):
A, _, _, = self.register_matrix(lib, m, n, order, "A")
o = lib.dense_operator_alloc(A)
self.register_var("o", o.contents.data, o.contents.free)
self.validate_operator(o.contents, m, n, lib.enums.DENSE)
self.free_vars("o", "A")
self.assertCall(lib.ok_device_reset())
def test_dense_operator(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
DIGITS = 7 - 2 * single_precision - 1 * gpu
TOL = 10 ** (-DIGITS)
for rowmajor in (True, False):
order = lib.enums.CblasRowMajor if rowmajor else lib.enums.CblasColMajor
A, A_, A_ptr = self.register_matrix(lib, m, n, order, "A")
A_ += self.A_test
self.assertCall(lib.matrix_memcpy_ma(A, A_ptr, order))
o = lib.dense_operator_alloc(A)
self.register_var("o", o.contents.data, o.contents.free)
self.exercise_operator(lib, o.contents, A_, TOL)
self.free_vars("o", "A")
self.assertCall(lib.ok_device_reset())
def test_sparse_alloc_free(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
for rowmajor in (True, False):
order = lib.enums.CblasRowMajor if rowmajor else lib.enums.CblasColMajor
enum = lib.enums.SPARSE_CSR if rowmajor else lib.enums.SPARSE_CSC
A, _, _, _, _, _ = self.register_sparsemat(lib, self.A_test_sparse, order, "A")
o = lib.sparse_operator_alloc(A)
self.register_var("o", o.contents.data, o.contents.free)
self.validate_operator(o.contents, m, n, enum)
self.free_vars("o", "A")
self.assertCall(lib.ok_device_reset())
def test_sparse_operator(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
DIGITS = 7 - 2 * single_precision - 1 * gpu
TOL = 10 ** (-DIGITS)
for order in (lib.enums.CblasRowMajor, lib.enums.CblasColMajor):
hdl = self.register_sparse_handle(lib, "hdl")
A, A_, A_sp, A_val, A_ind, A_ptr = self.register_sparsemat(lib, self.A_test_sparse, order, "A")
self.assertCall(lib.sp_matrix_memcpy_ma(hdl, A, A_val, A_ind, A_ptr))
o = lib.sparse_operator_alloc(A)
self.register_var("o", o.contents.data, o.contents.free)
self.exercise_operator(lib, o.contents, A_, TOL)
self.free_vars("o", "A", "hdl")
self.assertCall(lib.ok_device_reset())
def test_diagonal_alloc_free(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
for rowmajor in (True, False):
d = lib.vector(0, 0, None)
self.assertCall(lib.vector_calloc(d, n))
self.register_var("d", d, lib.vector_free)
o = lib.diagonal_operator_alloc(d)
self.register_var("o", o.contents.data, o.contents.free)
self.validate_operator(o.contents, n, n, lib.enums.DIAGONAL)
self.free_vars("o", "d")
self.assertCall(lib.ok_device_reset())
def test_diagonal_operator(self):
m, n = self.shape
for (gpu, single_precision) in self.CONDITIONS:
lib = self.libs.get(single_precision=single_precision, gpu=gpu)
if lib is None:
continue
self.register_exit(lib.ok_device_reset)
DIGITS = 7 - 2 * single_precision - 1 * gpu
TOL = 10 ** (-DIGITS)
for rowmajor in (True, False):
d, d_, d_ptr = self.register_vector(lib, n, "d")
d_ += self.A_test[0, :]
self.assertCall(lib.vector_memcpy_va(d, d_ptr, 1))
o = lib.diagonal_operator_alloc(d)
self.register_var("o", o.contents.data, o.contents.free)
self.exercise_operator(lib, o.contents, np.diag(d_), TOL)
self.free_vars("o", "d")
self.assertCall(lib.ok_device_reset())
def setUpClass(self):
self.env_orig = os.getenv('OPTKIT_USE_LOCALLIBS', '0')
os.environ['OPTKIT_USE_LOCALLIBS'] = '1'
self.libs = OperatorLibs()
self.A_test = self.A_test_gen
self.A_test_sparse = self.A_test_sparse_gen
def setUpClass(self):
self.env_orig = os.getenv("OPTKIT_USE_LOCALLIBS", "0")
os.environ["OPTKIT_USE_LOCALLIBS"] = "1"
self.libs = OperatorLibs()
self.A_test = self.A_test_gen
self.A_test_sparse = self.A_test_sparse_gen
