add ut/st for TensorScatter ops in ascend

Merge pull request  from yuchaojie/op_dev2
This commit is contained in:
i-robot 2022-05-09 06:50:47 +00:00 committed by Gitee
commit 6b385fe951
No known key found for this signature in database
GPG Key ID: 173E9B9CA92EEF8F
2 changed files with 211 additions and 0 deletions
tests

View File

@ -0,0 +1,199 @@
# 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.
# ============================================================================
import pytest
import numpy as np
import mindspore.nn as nn
import mindspore.ops as ops
import mindspore.context as context
from mindspore.common import dtype as mstype
from mindspore.common import Tensor, Parameter
op_map = {
"add": ops.TensorScatterAdd,
"sub": ops.TensorScatterSub,
"max": ops.TensorScatterMax,
"min": ops.TensorScatterMin,
}
func_map = {
"add": ops.tensor_scatter_add,
}
np_func_map = {
"mul": lambda a, b: a * b,
"div": lambda a, b: a / b,
"add": lambda a, b: a + b,
"sub": lambda a, b: a - b,
"max": np.maximum,
"min": np.minimum,
}
class TestTensorScatterArithmeticNet(nn.Cell):
def __init__(self, func, input_x, indices, updates):
super(TestTensorScatterArithmeticNet, self).__init__()
self.scatter_func = op_map.get(func)()
self.input_x = Parameter(input_x, name="input_x")
self.indices = Parameter(indices, name="indices")
self.updates = Parameter(updates, name="updates")
def construct(self):
output = self.scatter_func(self.input_x, self.indices, self.updates)
return output
def tensor_scatter_np(func, input_x, indices, updates):
result = input_x.asnumpy().copy()
indices_np = indices.asnumpy().copy()
updates_np = updates.asnumpy().copy()
f = np_func_map.get(func)
for idx, _ in np.ndenumerate(np.zeros(indices.shape[:-1])):
upd_idx = tuple(idx)
out_idx = tuple(indices_np[upd_idx])
result[out_idx] = f(result[out_idx], updates_np[upd_idx])
return result
def compare_with_numpy(func, input_x, indices, updates):
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
graph_output = TestTensorScatterArithmeticNet(func, input_x, indices, updates)()
expected = tensor_scatter_np(func, input_x, indices, updates)
np.testing.assert_array_almost_equal(graph_output.asnumpy(), expected)
context.set_context(mode=context.PYNATIVE_MODE, device_target="Ascend")
pynative_output = TestTensorScatterArithmeticNet(func, input_x, indices, updates)()
np.testing.assert_array_almost_equal(pynative_output.asnumpy(), expected)
@pytest.mark.level0
@pytest.mark.platform_arm_ascend_training
@pytest.mark.platform_x86_ascend_training
@pytest.mark.env_onecard
@pytest.mark.parametrize('func', ['add', 'sub'])
@pytest.mark.parametrize('data_type', [mstype.float16, mstype.float32])
@pytest.mark.parametrize('index_type', [mstype.int32])
def test_tensor_scatter_arithmetic_small_float(func, data_type, index_type):
"""
Feature: TensorScatter* operators.
Description: test cases for TensorScatter* operator
Expectation: the result match numpy implementation.
"""
input_x = Tensor(np.array([[-0.1, 0.3, 3.6], [0.4, 0.5, -3.2]]), data_type)
indices = Tensor(np.array([[0, 0], [1, 1]]), index_type)
updates = Tensor(np.array([1.0, 2.2]), data_type)
compare_with_numpy(func, input_x, indices, updates)
@pytest.mark.level0
@pytest.mark.platform_arm_ascend_training
@pytest.mark.platform_x86_ascend_training
@pytest.mark.env_onecard
@pytest.mark.parametrize('func', ['add', 'sub'])
@pytest.mark.parametrize('data_type', [mstype.int8, mstype.uint8, mstype.int32, mstype.int64])
@pytest.mark.parametrize('index_type', [mstype.int32])
def test_tensor_scatter_arithmetic_small_int(func, data_type, index_type):
"""
Feature: TensorScatter* operators.
Description: test cases for TensorScatter* operator
Expectation: the result match numpy implementation.
"""
input_x = Tensor(np.array([5, 6, 7, 8, 9, 10, 11, 12]), data_type)
indices = Tensor(np.array([[4], [3], [1], [7]]), index_type)
updates = Tensor(np.array([1, 2, 3, 4]), data_type)
compare_with_numpy(func, input_x, indices, updates)
@pytest.mark.level0
@pytest.mark.platform_arm_ascend_training
@pytest.mark.platform_x86_ascend_training
@pytest.mark.env_onecard
@pytest.mark.parametrize('func', ['add', 'sub'])
@pytest.mark.parametrize('data_type', [mstype.int32, mstype.float32])
@pytest.mark.parametrize('index_type', [mstype.int32])
def test_tensor_scatter_arithmetic_multi_dims(func, data_type, index_type):
"""
Feature: TensorScatter* operators.
Description: test cases for TensorScatter* operator
Expectation: the result match numpy implementation.
"""
input_x = Tensor(np.ones((4, 4, 4)) * 10, data_type)
indices = Tensor(np.array([[0], [2]]), index_type)
updates = Tensor(
np.array(
[
[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]],
[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]],
]
),
data_type,
)
compare_with_numpy(func, input_x, indices, updates)
@pytest.mark.level0
@pytest.mark.platform_arm_ascend_training
@pytest.mark.platform_x86_ascend_training
@pytest.mark.env_onecard
@pytest.mark.parametrize('func', ['add'])
@pytest.mark.parametrize('data_type', [mstype.float32])
@pytest.mark.parametrize('index_type', [mstype.int32])
def test_tensor_scatter_arithmetic_function_op(func, data_type, index_type):
"""
Feature: TensorScatter* functional operators.
Description: test cases for ops.tensor_scatter_* api
Expectation: the result match numpy implementation.
"""
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
input_x = Tensor(np.array([[-0.1, 0.3, 3.6], [0.4, 0.5, -3.2]]), data_type)
indices = Tensor(np.array([[0, 1]]), index_type)
updates = Tensor(np.array([1.0]), data_type)
expected = tensor_scatter_np(func, input_x, indices, updates)
output = func_map.get(func)(input_x, indices, updates)
np.testing.assert_allclose(output.asnumpy(), expected, rtol=1e-6)
@pytest.mark.level0
@pytest.mark.platform_arm_ascend_training
@pytest.mark.platform_x86_ascend_training
@pytest.mark.env_onecard
@pytest.mark.parametrize('func', ['add'])
@pytest.mark.parametrize('data_type', [mstype.float32])
@pytest.mark.parametrize('index_type', [mstype.int32])
def test_tensor_scatter_arithmetic_tensor_op(func, data_type, index_type):
"""
Feature: TensorScatter* tensor operators.
Description: test cases for tensor.tensor_scatter_* api
Expectation: the result match numpy implementation.
"""
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
input_x = Tensor(np.array([[-0.1, 0.3, 3.6], [0.4, 0.5, -3.2]]), data_type)
indices = Tensor(np.array([[0, 1]]), index_type)
updates = Tensor(np.array([1.0]), data_type)
expected = tensor_scatter_np(func, input_x, indices, updates)
if func == 'add':
output = input_x.tensor_scatter_add(indices, updates)
np.testing.assert_allclose(output.asnumpy(), expected, rtol=1e-6)

View File

@ -3114,6 +3114,18 @@ test_case_other_ops = [
Tensor(np.array([[0, 1], [1, 2]], np.int32)),
Tensor(np.ones([2, 5], np.float32) * 99)),
'desc_bprop': [([3, 4, 5], {'dtype': np.float32})]}),
('TensorScatterAdd', {
'block': P.TensorScatterAdd(),
'desc_inputs': (Tensor(np.arange(3 * 4 * 5).reshape((3, 4, 5)), mstype.float32),
Tensor(np.array([[0, 1], [1, 2]], np.int32)),
Tensor(np.ones([2, 5], np.float32) * 99)),
'desc_bprop': [([3, 4, 5], {'dtype': np.float32})]}),
('TensorScatterSub', {
'block': P.TensorScatterSub(),
'desc_inputs': (Tensor(np.arange(3 * 4 * 5).reshape((3, 4, 5)), mstype.float32),
Tensor(np.array([[0, 1], [1, 2]], np.int32)),
Tensor(np.ones([2, 5], np.float32) * 99)),
'desc_bprop': [([3, 4, 5], {'dtype': np.float32})]}),
('ScatterMaxUseLocking', {
'block': ScatterMax(use_locking=True),
'desc_inputs': (Tensor(np.array([1, 0], np.int32)),