def test_stack():
fname = os.path.join(this_file_dir,
'runtime_analysis/{}/general/stack.txt'.format(DIM))
if os.path.exists(fname):
os.remove(fname)
for lib, call in [
(l, c) for l, c in helpers.calls
if c not in [helpers.tf_graph_call, helpers.mx_graph_call]
]:
time_lib = LIB_DICT[lib]
append_to_file(fname, '{}'.format(lib))
x0 = ivy_gen.tensor([random.uniform(0, 1) for _ in range(DIM)], f=lib)
x1 = ivy_gen.tensor([random.uniform(0, 1) for _ in range(DIM)], f=lib)
ivy_gen.stack([x0, x1], f=lib)
ivy_gen_w_time.stack([x0, x1], f=time_lib)
TIMES_DICT.clear()
for _ in range(100):
log_time(fname, 'tb0')
ivy_gen_w_time.stack([x0, x1], 0, f=time_lib)
log_time(fname, 'tb4', time_at_start=True)
log_time(fname, 'tt0')
ivy_gen.stack([x0, x1], 0, f=lib)
log_time(fname, 'tt1', time_at_start=True)
write_times()
append_to_file(fname, 'end of analysis')
def test_linear():
fname = os.path.join(this_file_dir,
'runtime_analysis/{}/layers/linear.txt'.format(DIM))
if os.path.exists(fname):
os.remove(fname)
for lib, call in [
(l, c) for l, c in helpers.calls
if c not in [helpers.tf_graph_call, helpers.mx_graph_call]
]:
time_lib = LIB_DICT[lib]
append_to_file(fname, '{}'.format(lib))
x0 = ivy_gen.tensor([[random.uniform(0, 1) for _ in range(DIM)]],
f=lib)
weight = ivy_gen.tensor([[random.uniform(0, 1) for _ in range(DIM)],
[random.uniform(0, 1) for _ in range(DIM)]],
f=lib)
bias = ivy_gen.tensor(
[random.uniform(0, 1), random.uniform(0, 1)], f=lib)
ivy_layers.linear(x0, weight, bias, num_hidden=2, f=lib)
ivy_layers_w_time.linear(x0, weight, bias, num_hidden=2, f=time_lib)
TIMES_DICT.clear()
for _ in range(100):
log_time(fname, 'tb0')
ivy_layers_w_time.linear(x0,
weight,
bias,
num_hidden=2,
f=time_lib)
log_time(fname, 'tb4', time_at_start=True)
log_time(fname, 'tt0')
ivy_layers.linear(x0, weight, bias, num_hidden=2, f=lib)
log_time(fname, 'tt1', time_at_start=True)
write_times()
append_to_file(fname, 'end of analysis')
def test_gradient_descent_update():
fname = os.path.join(
this_file_dir,
'runtime_analysis/{}/gradients/gradient_descent_update.txt'.format(
DIM))
if os.path.exists(fname):
os.remove(fname)
for lib, call in [
(l, c) for l, c in helpers.calls
if c not in [helpers.tf_graph_call, helpers.mx_graph_call]
]:
time_lib = LIB_DICT[lib]
append_to_file(fname, '{}'.format(lib))
x0 = ivy_gen.tensor([random.uniform(0, 1) for _ in range(DIM)], f=lib)
x1 = ivy_gen.tensor([random.uniform(0, 1) for _ in range(DIM)], f=lib)
ws = [ivy_grad.variable(x0)]
dcdws = [x1]
ivy_grad.gradient_descent_update(ws, dcdws, 0.1, f=lib)
ivy_grad_w_time.gradient_descent_update(ws, dcdws, 0.1, f=time_lib)
TIMES_DICT.clear()
for _ in range(100):
log_time(fname, 'tb0')
ivy_grad_w_time.gradient_descent_update(ws, dcdws, 0.1, f=time_lib)
log_time(fname, 'tb4', time_at_start=True)
log_time(fname, 'tt0')
ivy_grad.gradient_descent_update(ws, dcdws, 0.1, f=lib)
log_time(fname, 'tt1', time_at_start=True)
write_times()
append_to_file(fname, 'end of analysis')
def test_vector_to_skew_symmetric_matrix():
fname = os.path.join(
this_file_dir,
'runtime_analysis/{}/linalg/vector_to_skew_symmetric_matrix.txt'.
format(DIM))
if os.path.exists(fname):
os.remove(fname)
for lib, call in [
(l, c) for l, c in helpers.calls
if c not in [helpers.tf_graph_call, helpers.mx_graph_call]
]:
time_lib = LIB_DICT[lib]
append_to_file(fname, '{}'.format(lib))
x0 = ivy_gen.tensor([[
random.uniform(0, 1),
random.uniform(0, 1),
random.uniform(0, 1)
] for _ in range(DIM)],
f=lib)
ivy_linalg.vector_to_skew_symmetric_matrix(x0, f=lib)
ivy_linalg_w_time.vector_to_skew_symmetric_matrix(x0, f=time_lib)
TIMES_DICT.clear()
for _ in range(100):
log_time(fname, 'tb0')
ivy_linalg_w_time.vector_to_skew_symmetric_matrix(x0, f=time_lib)
log_time(fname, 'tb4', time_at_start=True)
log_time(fname, 'tt0')
ivy_linalg.vector_to_skew_symmetric_matrix(x0, f=lib)
log_time(fname, 'tt1', time_at_start=True)
write_times()
append_to_file(fname, 'end of analysis')
def test_compile_fn():
fname = os.path.join(
this_file_dir,
'runtime_analysis/{}/general/compile_fn.txt'.format(DIM))
if os.path.exists(fname):
os.remove(fname)
for lib, call in [
(l, c) for l, c in helpers.calls
if c not in [helpers.tf_graph_call, helpers.mx_graph_call]
]:
time_lib = LIB_DICT[lib]
if call is helpers.mx_call:
continue
append_to_file(fname, '{}'.format(lib))
some_fn = lambda x: x**2
x0 = ivy_gen.tensor([random.uniform(0, 1) for _ in range(DIM)], f=lib)
ivy_gen.dtype(x0, f=lib)
ivy_gen_w_time.dtype(x0, f=time_lib)
TIMES_DICT.clear()
for _ in range(100):
log_time(fname, 'tb0')
ivy_gen_w_time.compile_fn(some_fn, x0, f=time_lib)
log_time(fname, 'tb4', time_at_start=True)
log_time(fname, 'tt0')
ivy_gen.compile_fn(some_fn, x0, f=lib)
log_time(fname, 'tt1', time_at_start=True)
write_times()
append_to_file(fname, 'end of analysis')
def test_one_hot():
fname = os.path.join(this_file_dir,
'runtime_analysis/{}/general/one_hot.txt'.format(DIM))
if os.path.exists(fname):
os.remove(fname)
for lib, call in [
(l, c) for l, c in helpers.calls
if c not in [helpers.tf_graph_call, helpers.mx_graph_call]
]:
time_lib = LIB_DICT[lib]
append_to_file(fname, '{}'.format(lib))
x0 = ivy_gen.cast(
ivy_gen.tensor([
random.randint(0,
int(DIM**0.5) - 1) for _ in range(int(DIM**0.5))
],
f=lib), 'int64')
ivy_gen.one_hot(x0, int(DIM**0.5), f=lib)
ivy_gen_w_time.one_hot(x0, int(DIM**0.5), f=time_lib)
TIMES_DICT.clear()
for _ in range(100):
log_time(fname, 'tb0')
ivy_gen_w_time.one_hot(x0, int(DIM**0.5), f=time_lib)
log_time(fname, 'tb4', time_at_start=True)
log_time(fname, 'tt0')
ivy_gen.one_hot(x0, int(DIM**0.5), f=lib)
log_time(fname, 'tt1', time_at_start=True)
write_times()
append_to_file(fname, 'end of analysis')
def test_execute_with_gradients():
fname = os.path.join(
this_file_dir,
'runtime_analysis/{}/gradients/execute_with_gradients.txt'.format(DIM))
if os.path.exists(fname):
os.remove(fname)
for lib, call in [
(l, c) for l, c in helpers.calls
if c not in [helpers.tf_graph_call, helpers.mx_graph_call]
]:
time_lib = LIB_DICT[lib]
append_to_file(fname, '{}'.format(lib))
x0 = ivy_gen.tensor([random.uniform(0, 1) for _ in range(DIM)], f=lib)
func = lambda xs_in: (xs_in[0] * xs_in[0])[0]
xs = [ivy_grad.variable(x0)]
ivy_grad.execute_with_gradients(func, xs, f=lib)
ivy_grad_w_time.execute_with_gradients(func, xs, f=time_lib)
TIMES_DICT.clear()
for _ in range(100):
log_time(fname, 'tb0')
ivy_grad_w_time.execute_with_gradients(func, xs, f=time_lib)
log_time(fname, 'tb4', time_at_start=True)
log_time(fname, 'tt0')
ivy_grad.execute_with_gradients(func, xs, f=lib)
log_time(fname, 'tt1', time_at_start=True)
write_times()
append_to_file(fname, 'end of analysis')
def test_conv3d_transpose():
fname = os.path.join(
this_file_dir,
'runtime_analysis/{}/layers/conv3d_transpose.txt'.format(DIM))
if os.path.exists(fname):
os.remove(fname)
for lib, call in [
(l, c) for l, c in helpers.calls
if c not in [helpers.tf_graph_call, helpers.mx_graph_call]
]:
time_lib = LIB_DICT[lib]
if call in [
helpers.np_call, helpers.jnp_call, helpers.mx_call,
helpers.mx_graph_call
]:
# numpy and jax do not yet support 3d convolutions, and mxnet only supports with CUDNN
continue
x0 = ivy_gen.tensor(
[[[[[random.uniform(0, 1)], [random.uniform(0, 1)]],
[[random.uniform(0, 1)], [random.uniform(0, 1)]]],
[[[random.uniform(0, 1)], [random.uniform(0, 1)]],
[[random.uniform(0, 1)], [random.uniform(0, 1)]]]]
for _ in range(DIM)],
f=lib)
if call in [helpers.tf_call, helpers.tf_graph_call]:
data_format = "NDHWC"
else:
x0 = ivy_gen.reshape(x0, (DIM, 1, 2, 2, 2))
data_format = "NCDHW"
append_to_file(fname, '{}'.format(lib))
filters = ivy_gen.tensor([[[[[0.]], [[0.]]], [[[1.]], [[1.]]]],
[[[[1.]], [[1.]]], [[[0.]], [[0.]]]]],
f=lib)
ivy_layers.conv3d_transpose(x0,
filters,
1,
"SAME", (DIM, 2, 2, 2, 1),
data_format,
filter_shape=[2, 2, 2],
num_filters=1,
f=lib)
ivy_layers_w_time.conv3d_transpose(x0,
filters,
1,
"SAME", (DIM, 2, 2, 2, 1),
data_format,
filter_shape=[2, 2, 2],
num_filters=1,
f=time_lib)
TIMES_DICT.clear()
for _ in range(100):
log_time(fname, 'tb0')
ivy_layers_w_time.conv3d_transpose(x0,
filters,
1,
"SAME", (DIM, 2, 2, 2, 1),
data_format,
filter_shape=[2, 2, 2],
num_filters=1,
f=time_lib)
log_time(fname, 'tb4', time_at_start=True)
log_time(fname, 'tt0')
ivy_layers.conv3d_transpose(x0,
filters,
1,
"SAME", (DIM, 2, 2, 2, 1),
data_format,
filter_shape=[2, 2, 2],
num_filters=1,
f=lib)
log_time(fname, 'tt1', time_at_start=True)
write_times()
append_to_file(fname, 'end of analysis')