mindspore/tests/ut/python/parallel/test_gather_field_split.py

124 lines
4.8 KiB
Python

# Copyright 2020 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 numpy as np
import mindspore as ms
from mindspore import context, Tensor, Parameter
from mindspore.common.api import _cell_graph_executor
from mindspore.nn import Cell
from mindspore.ops import operations as P
from mindspore.common.initializer import initializer
from parallel.utils.utils import ParallelValidator
class Net(Cell):
def __init__(self, strategy1, strategy2, split_tuple, param_shape, mul_weight_shape):
super().__init__()
self.gatherv2 = P.Gather().shard(strategy1)
self.gatherv2.add_prim_attr("manual_split", split_tuple)
self.mul = P.Mul().shard(strategy2)
self.param = Parameter(initializer("ones", param_shape, ms.float32), name="gather_param")
self.mul_weight = Parameter(initializer("ones", mul_weight_shape, ms.float32), name="mul_weight")
def construct(self, x, b):
out = self.gatherv2(self.param, x, 0)
out = self.mul(out, self.mul_weight)
return out
def compile_net(net, x):
net.set_train()
b = Tensor(np.ones([64, 8]), dtype=ms.float32)
phase, _ = _cell_graph_executor.compile(net, x, b)
context.reset_auto_parallel_context()
return phase
def test_field_split_dim_1x1():
"""
Feature: test field split
Description: param dim is 1, indices dim is 1
Expectation: compile success
"""
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=1)
strategy1 = ((8,), (8,))
strategy2 = ((8,), (1,))
split_tuple = (5, 6, 7, 8, 9, 10, 11, 8)
param_shape = (64)
mul_weight_shape = (1)
x = Tensor(np.ones([16 // 8]), dtype=ms.int32)
net = Net(strategy1, strategy2, split_tuple, param_shape, mul_weight_shape)
phase = compile_net(net, x)
validator = ParallelValidator(net, phase)
# check layout, dev-matrix/tensor-map/slice_shape/field_size/uniform_split/opt_shard_group
gather_param_layout = ([8], [0], [6], 0, False, '')
assert validator.check_parameter_layout('gather_param', gather_param_layout)
# check inputs
sub_expect_inputs = ['TupleGetItem', 'value=5']
assert validator.check_node_inputs_fuzzy_match('Sub-0', sub_expect_inputs)
def test_field_split_dim_2x1():
"""
Feature: test field split
Description: param dim is 2, indices dim is 1
Expectation: compile success
"""
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=2)
strategy1 = ((4, 2), (4,))
strategy2 = ((4, 2), (1,))
split_tuple = (10, 20, 30, 4)
param_shape = (64, 32)
mul_weight_shape = (1)
x = Tensor(np.ones([16 // 8]), dtype=ms.int32)
net = Net(strategy1, strategy2, split_tuple, param_shape, mul_weight_shape)
phase = compile_net(net, x)
validator = ParallelValidator(net, phase)
# check layout, dev-matrix/tensor-map/slice_shape/field_size/uniform_split/opt_shard_group
gather_param_layout = ([4, 2], [1, 0], [20, 16], 0, False, '')
assert validator.check_parameter_layout('gather_param', gather_param_layout)
# check inputs
sub_expect_inputs = ['Reshape', 'value=10']
assert validator.check_node_inputs_fuzzy_match('Sub-0', sub_expect_inputs)
def test_field_split_dim_3x1():
"""
Feature: test field split
Description: param dim is 3, indices dim is 1
Expectation: compile success
"""
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel", device_num=8, global_rank=2)
strategy1 = ((4, 2, 1), (4,))
strategy2 = ((4, 2, 1), (1,))
split_tuple = (10, 20, 30, 4)
param_shape = (64, 32, 16)
mul_weight_shape = (1)
x = Tensor(np.ones([16 // 8]), dtype=ms.int32)
net = Net(strategy1, strategy2, split_tuple, param_shape, mul_weight_shape)
phase = compile_net(net, x)
validator = ParallelValidator(net, phase)
# check layout, dev-matrix/tensor-map/slice_shape/field_size/uniform_split/opt_shard_group
gather_param_layout = ([4, 2], [1, 0, -1], [20, 16, 16], 0, False, '')
assert validator.check_parameter_layout('gather_param', gather_param_layout)
# check inputs
sub_expect_inputs = ['Reshape', 'value=10']
assert validator.check_node_inputs_fuzzy_match('Sub-0', sub_expect_inputs)