wehub-resource-sync
470 lines
18 KiB
Python
470 lines
18 KiB
Python
# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# 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 unittest
|
|
|
|
from paddle.distributed.auto_parallel.static.dist_attribute import (
|
|
DistTensorSpec,
|
|
TensorDistAttr,
|
|
)
|
|
from paddle.distributed.fleet import auto
|
|
from paddle.framework import core
|
|
|
|
|
|
class TestConv3dSPMDRule(unittest.TestCase):
|
|
def setUp(self):
|
|
self.rule = core.get_phi_spmd_rule("conv3d")
|
|
|
|
def test_conv3d_ncdhw_infer_forward(self):
|
|
# forward setup
|
|
input_shape = [2, 4, 6, 8, 8]
|
|
self.data_format = "NCDHW"
|
|
filter_shape = [10, 4, 2, 3, 3]
|
|
process_mesh = auto.ProcessMesh(
|
|
mesh=[[[0, 1], [2, 3]], [[4, 5], [6, 7]]]
|
|
)
|
|
|
|
input_tensor_dist_attr = TensorDistAttr()
|
|
input_tensor_dist_attr.dims_mapping = [0, -1, -1, -1, -1]
|
|
input_tensor_dist_attr.process_mesh = process_mesh
|
|
self.input_dist_tensor_spec = DistTensorSpec(
|
|
input_shape, input_tensor_dist_attr
|
|
)
|
|
|
|
filter_tensor_dist_attr = TensorDistAttr()
|
|
filter_tensor_dist_attr.dims_mapping = [-1, -1, -1, -1, -1]
|
|
filter_tensor_dist_attr.process_mesh = process_mesh
|
|
self.filter_dist_tensor_spec = DistTensorSpec(
|
|
filter_shape, filter_tensor_dist_attr
|
|
)
|
|
|
|
self.strides = [1, 1, 1]
|
|
self.paddings = [0, 0, 0]
|
|
self.padding_algorithm = "EXPLICIT"
|
|
self.group = 1
|
|
self.dilations = [1, 1, 1]
|
|
# case 1
|
|
# input: NCDinHinWin[0, -1, -1, -1, -1], filter: MCDkHkWk[-1, -1, -1, -1, -1]
|
|
# ---> output: NMDoutHoutWout[0, -1, -1, -1, -1]
|
|
result_dist_attrs = self.rule.infer_forward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 1)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [0, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [-1, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [0, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), False)
|
|
|
|
# case 2
|
|
# input: NCDinHinWin[-1, -1, -1, -1, -1], filter: MCDkHkWk[0, -1, -1, -1, -1]
|
|
# ---> output: NMDoutHoutWout[-1, 0, -1, -1, -1]
|
|
self.input_dist_tensor_spec.set_dims_mapping([-1, -1, -1, -1, -1])
|
|
self.filter_dist_tensor_spec.set_dims_mapping([0, -1, -1, -1, -1])
|
|
|
|
result_dist_attrs = self.rule.infer_forward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 1)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [-1, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [0, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [-1, 0, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), False)
|
|
|
|
# case 3
|
|
# input: NCDinHinWin[0, -1, -1, -1, -1], filter: MCDkHkWk[1, -1, -1, -1, -1] --->
|
|
# output: NMDoutHoutWout[0, 1, -1, -1, -1]
|
|
self.input_dist_tensor_spec.set_dims_mapping([0, -1, -1, -1, -1])
|
|
self.filter_dist_tensor_spec.set_dims_mapping([1, -1, -1, -1, -1])
|
|
|
|
result_dist_attrs = self.rule.infer_forward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 1)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [0, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [1, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [0, 1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), False)
|
|
|
|
# case 4
|
|
# input: NCDinHinWin[-1, 0, -1, -1, -1], filter: MCDkHkWk[-1, 0, -1, -1, -1] --->
|
|
# output: NMDoutHoutWout[-1, -1, -1, -1, -1]
|
|
self.input_dist_tensor_spec.set_dims_mapping([-1, 0, -1, -1, -1])
|
|
self.filter_dist_tensor_spec.set_dims_mapping([-1, 0, -1, -1, -1])
|
|
|
|
result_dist_attrs = self.rule.infer_forward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 1)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [-1, 0, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [-1, 0, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [-1, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), True)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._partial_dims(), {0})
|
|
|
|
# case 5
|
|
# input: NCDinHinWin[0, 2, -1, -1, -1], filter: MCDkHkWk[1, 2, -1, -1, -1] --->
|
|
# output: NMDoutHoutWout[0, 1, -1, -1, -1]
|
|
self.input_dist_tensor_spec.set_dims_mapping([0, 2, -1, -1, -1])
|
|
self.filter_dist_tensor_spec.set_dims_mapping([1, 2, -1, -1, -1])
|
|
|
|
result_dist_attrs = self.rule.infer_forward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 1)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [0, 2, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [1, 2, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [0, 1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), True)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._partial_dims(), {2})
|
|
|
|
def test_conv3d_ndhwc_infer_forward(self):
|
|
# forward setup
|
|
input_shape = [2, 6, 8, 8, 4]
|
|
self.data_format = "NDHWC"
|
|
filter_shape = [10, 4, 2, 3, 3]
|
|
process_mesh = auto.ProcessMesh(
|
|
mesh=[[[0, 1], [2, 3]], [[4, 5], [6, 7]]]
|
|
)
|
|
|
|
input_tensor_dist_attr = TensorDistAttr()
|
|
input_tensor_dist_attr.dims_mapping = [-1, -1, -1, -1, -1]
|
|
input_tensor_dist_attr.process_mesh = process_mesh
|
|
self.input_dist_tensor_spec = DistTensorSpec(
|
|
input_shape, input_tensor_dist_attr
|
|
)
|
|
|
|
filter_tensor_dist_attr = TensorDistAttr()
|
|
filter_tensor_dist_attr.dims_mapping = [-1, -1, -1, -1, -1]
|
|
filter_tensor_dist_attr.process_mesh = process_mesh
|
|
self.filter_dist_tensor_spec = DistTensorSpec(
|
|
filter_shape, filter_tensor_dist_attr
|
|
)
|
|
|
|
self.strides = [1, 1, 1]
|
|
self.paddings = [0, 0, 0]
|
|
self.padding_algorithm = "EXPLICIT"
|
|
self.group = 1
|
|
self.dilations = [1, 1, 1]
|
|
# case 6
|
|
# input: NDinHinWinC[-1, -1, -1, -1, 0], filter: MCDkHkWk[-1, 0, -1, -1, -1] --->
|
|
# output: NMDoutHoutWout[-1, -1, -1, -1, -1]
|
|
self.input_dist_tensor_spec.set_dims_mapping([-1, -1, -1, -1, 0])
|
|
self.filter_dist_tensor_spec.set_dims_mapping([-1, 0, -1, -1, -1])
|
|
|
|
result_dist_attrs = self.rule.infer_forward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 1)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [-1, -1, -1, -1, 0]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [-1, 0, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [-1, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), True)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._partial_dims(), {0})
|
|
|
|
# case 7
|
|
# input: NDinHinWinC[0, -1, -1, -1, 2], filter: MCDkHkWk[1, 2, -1, -1, -1] --->
|
|
# output: NMDoutHoutWout[0, 1, -1, -1, -1]
|
|
self.input_dist_tensor_spec.set_dims_mapping([0, -1, -1, -1, 2])
|
|
self.filter_dist_tensor_spec.set_dims_mapping([1, 2, -1, -1, -1])
|
|
|
|
result_dist_attrs = self.rule.infer_forward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 1)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [0, -1, -1, -1, 2]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [1, 2, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [0, 1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), True)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._partial_dims(), {2})
|
|
|
|
def test_conv3d_ncdhw_infer_backward(self):
|
|
# backward setup
|
|
input_shape = [2, 4, 8, 8, 8]
|
|
self.data_format = "NCDHW"
|
|
filter_shape = [10, 4, 3, 3, 3]
|
|
output_shape = [2, 10, 6, 6, 6]
|
|
process_mesh = auto.ProcessMesh(
|
|
mesh=[[[0, 1], [2, 3]], [[4, 5], [6, 7]]]
|
|
)
|
|
|
|
input_tensor_dist_attr = TensorDistAttr()
|
|
input_tensor_dist_attr.dims_mapping = [-1, -1, -1, -1, -1]
|
|
input_tensor_dist_attr.process_mesh = process_mesh
|
|
self.input_dist_tensor_spec = DistTensorSpec(
|
|
input_shape, input_tensor_dist_attr
|
|
)
|
|
|
|
filter_tensor_dist_attr = TensorDistAttr()
|
|
filter_tensor_dist_attr.dims_mapping = [-1, -1, -1, -1, -1]
|
|
filter_tensor_dist_attr.process_mesh = process_mesh
|
|
self.filter_dist_tensor_spec = DistTensorSpec(
|
|
filter_shape, filter_tensor_dist_attr
|
|
)
|
|
|
|
output_tensor_dist_attr = TensorDistAttr()
|
|
output_tensor_dist_attr.dims_mapping = [0, 1, -1, -1, -1]
|
|
output_tensor_dist_attr.process_mesh = process_mesh
|
|
self.output_dist_tensor_spec = DistTensorSpec(
|
|
output_shape, output_tensor_dist_attr
|
|
)
|
|
|
|
self.strides = [1, 1, 1]
|
|
self.paddings = [0, 0, 0]
|
|
self.padding_algorithm = "EXPLICIT"
|
|
self.group = 1
|
|
self.dilations = [1, 1, 1]
|
|
|
|
# case 1:
|
|
# Output: NMDoutHoutWout[0, 1, -1, -1, -1] --->
|
|
# input: NCDinHinWin[0, -1, -1, -1, -1], filter: MCDkHkWk[1, -1, -1, -1, -1]
|
|
# input_grad: NCDinHinWin[0, -1, -1, -1, -1], filter_grad: MCDkHkWk[1, -1, -1, -1, -1]
|
|
result_dist_attrs = self.rule.infer_backward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.output_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 3)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 2)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [0, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [1, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[2].dims_mapping, [0, 1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [0, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[1].dims_mapping, [1, -1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_input_dist_attrs[0]._is_partial(), False)
|
|
self.assertEqual(inferred_input_dist_attrs[1]._is_partial(), False)
|
|
self.assertEqual(inferred_input_dist_attrs[2]._is_partial(), False)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), True)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._partial_dims(), {1})
|
|
self.assertEqual(inferred_output_dist_attrs[1]._is_partial(), True)
|
|
self.assertEqual(inferred_output_dist_attrs[1]._partial_dims(), {0})
|
|
|
|
# case 2:
|
|
# Output: NMDoutHoutWout[0, 1, -1, -1, -1] partial_dim=[2]--->
|
|
# input: NCDinHinWin[0, 2, -1, -1, -1], filter: MCDkHkWk[1, 2, -1, -1, -1]
|
|
# input_grad: NCDinHinWin[0, 2, -1, -1, -1], filter_grad: MCDkHkWk[1, 2, -1, -1, -1]
|
|
output_tensor_dist_attr._set_partial_dims([2])
|
|
self.output_dist_tensor_spec = DistTensorSpec(
|
|
output_shape, output_tensor_dist_attr
|
|
)
|
|
self.input_dist_tensor_spec.set_dims_mapping([0, 2, -1, -1, -1])
|
|
self.filter_dist_tensor_spec.set_dims_mapping([1, 2, -1, -1, -1])
|
|
result_dist_attrs = self.rule.infer_backward(
|
|
self.input_dist_tensor_spec,
|
|
self.filter_dist_tensor_spec,
|
|
self.output_dist_tensor_spec,
|
|
self.strides,
|
|
self.paddings,
|
|
self.padding_algorithm,
|
|
self.group,
|
|
self.dilations,
|
|
self.data_format,
|
|
)
|
|
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 3)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 2)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [0, 2, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [1, 2, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[2].dims_mapping, [0, 1, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [0, 2, -1, -1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[1].dims_mapping, [1, 2, -1, -1, -1]
|
|
)
|
|
self.assertEqual(inferred_input_dist_attrs[0]._is_partial(), False)
|
|
self.assertEqual(inferred_input_dist_attrs[1]._is_partial(), False)
|
|
self.assertEqual(inferred_input_dist_attrs[2]._is_partial(), True)
|
|
self.assertEqual(inferred_input_dist_attrs[2]._partial_dims(), {2})
|
|
self.assertEqual(inferred_output_dist_attrs[0]._is_partial(), True)
|
|
self.assertEqual(inferred_output_dist_attrs[0]._partial_dims(), {1})
|
|
self.assertEqual(inferred_output_dist_attrs[1]._is_partial(), True)
|
|
self.assertEqual(inferred_output_dist_attrs[1]._partial_dims(), {0})
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|