mindspore/tests/st/sparse/test_control_flow.py

219 lines
6.8 KiB
Python

# Copyright 2022 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""smoke tests for Sparse control flow cases"""
import pytest
import numpy as np
from mindspore import Tensor, CSRTensor, Parameter, nn, context
from mindspore.common import dtype as mstype
import mindspore.ops.operations as P
from .sparse_utils import compare_csr, get_csr_tensor, csr_add, get_csr_components, get_csr_from_scalar, \
forward_grad_net
@pytest.mark.level0
@pytest.mark.platform_x86_gpu_training
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_while_tensor_as_condition_forward_and_backward():
"""
Feature: Test CSRTensor in while.
Description: Test CSRTensor computation in while loop.
Expectation: Success.
"""
class Net(nn.Cell):
def construct(self, x, y):
out = y
while x < 1:
out = csr_add(out, out.values)
x += 1
return out
x = Tensor(-2, dtype=mstype.int32)
y = get_csr_tensor()
net = Net()
csr1, grad_py = forward_grad_net(net, x, y, mode=context.PYNATIVE_MODE)
csr2, grad_graph = forward_grad_net(net, x, y, mode=context.GRAPH_MODE)
# Compare results
compare_csr(csr1, csr2)
assert (csr1.values.asnumpy() == np.array([8, 16], dtype=np.float32)).all()
assert len(grad_py) == 2
assert len(grad_graph) == 2
assert isinstance(grad_graph[1], CSRTensor)
@pytest.mark.level0
@pytest.mark.platform_x86_gpu_training
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_control_flow_while_if_continue_not_relevant_gt():
"""
Feature: Test CSRTensor in while.
Description: Test CSRTensor computation in while loop.
Expectation: Success.
"""
class Net(nn.Cell):
def __init__(self):
super().__init__()
self.addn = P.AddN()
def construct(self, x):
s = x
t = csr_add(x, 1)
tensor_list = [x.values, x.values]
while len(tensor_list) < 4:
tensor_list.append(x.values)
a = self.addn(tensor_list)
x = csr_add(x, 1)
if t.values in tensor_list:
continue
s = csr_add(s, a)
return s
num = Tensor(-2, dtype=mstype.float32)
x = get_csr_from_scalar(num)
net = Net()
csr1, grad_py = forward_grad_net(net, x, mode=context.PYNATIVE_MODE)
csr2, grad_graph = forward_grad_net(net, x, mode=context.GRAPH_MODE)
# Compare results
compare_csr(csr1, csr2)
assert (csr1.values.asnumpy() == np.array([-8], dtype=np.float32)).all()
assert len(grad_py) == 1
assert len(grad_graph) == 1
assert isinstance(grad_graph[0], CSRTensor)
@pytest.mark.level0
@pytest.mark.platform_x86_gpu_training
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_control_flow_for_while_return_in_while_no():
"""
Feature: Test CSRTensor in while.
Description: Test CSRTensor computation in while loop.
Expectation: Success.
"""
class Net(nn.Cell):
def construct(self, x, y):
out = y
for _ in range(3):
out = csr_add(out, y.values)
while x < 5:
out = csr_add(out, out.values)
if x > 1:
return out
x += 1
return out
x = Tensor(-2, dtype=mstype.int32)
y = get_csr_tensor()
net = Net()
csr1, grad_py = forward_grad_net(net, x, y, mode=context.PYNATIVE_MODE)
csr2, grad_graph = forward_grad_net(net, x, y, mode=context.GRAPH_MODE)
# Compare results
compare_csr(csr1, csr2)
assert (csr1.values.asnumpy() == np.array([128, 256], dtype=np.float32)).all()
assert len(grad_py) == 2
assert len(grad_graph) == 2
assert isinstance(grad_graph[1], CSRTensor)
@pytest.mark.level0
@pytest.mark.platform_x86_gpu_training
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_control_flow_for_enumerate_if_continue():
"""
Feature: Test CSRTensor in while.
Description: Test CSRTensor computation in while loop.
Expectation: Success.
"""
class Net(nn.Cell):
def __init__(self, t1, t2):
super().__init__()
self.p1 = Parameter(Tensor(t1, mstype.float32), name="a")
self.p2 = Parameter(Tensor(t2, mstype.float32), name="b")
self.assignadd = P.AssignAdd()
def construct(self, x):
plist = [self.p1, self.p2]
out = x
for i, t in enumerate(plist):
if t > 2:
continue
self.assignadd(t, 1)
out = csr_add(out, i * t)
return out
t1 = 1
t2 = 2
x = get_csr_tensor()
csr1, grad_py = forward_grad_net(Net(t1, t2), x, mode=context.PYNATIVE_MODE)
csr2, grad_graph = forward_grad_net(Net(t1, t2), x, mode=context.GRAPH_MODE)
# Compare results
compare_csr(csr1, csr2)
assert (csr1.values.asnumpy() == np.array([4, 5], dtype=np.float32)).all()
assert len(grad_py) == 1
assert len(grad_graph) == 1
assert isinstance(grad_graph[0], CSRTensor)
@pytest.mark.level0
@pytest.mark.platform_x86_gpu_training
@pytest.mark.platform_x86_cpu
@pytest.mark.env_onecard
def test_multi_csr_in_if_else():
"""
Feature: Test multiple CSRTensors in if-else.
Description: Test CSRTensor computation in control flow.
Expectation: Success.
"""
class Net(nn.Cell):
def __init__(self, shape):
super().__init__()
self.shape = shape
def construct(self, indptr, indices, values, a, b):
x = CSRTensor(indptr, indices, values, self.shape)
if a > b:
x1 = x.abs()
x2 = x.astype(mstype.float16)
x3 = x.to_tuple()
else:
x1 = x.abs()
x2 = x.astype(mstype.float16)
x3 = x.to_tuple()
return x1, x2, x3
a = Tensor(1, mstype.float32)
b = Tensor(0, mstype.float32)
indptr, indices, values, shape = get_csr_components()
net = Net(shape)
forward_grad_net(net, indptr, indices, values, a, b, mode=context.PYNATIVE_MODE)
forward_grad_net(net, indptr, indices, values, a, b, mode=context.GRAPH_MODE)