1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
|
# Copyright 2017 The TensorFlow 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.
# ==============================================================================
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import numpy as np
import tensorflow as tf
import tensorflow.contrib.mpi_collectives as mpi
from tensorflow.python.platform import test
average_allgather = False
class AllgatherTest(test.TestCase):
def checkAllgather(self, num_ranks, all_gathered, local_gathered):
# Ensure that indices match.
all_gat_ind = np.sort(all_gathered.indices)
loc_gat_ind = np.sort(local_gathered.indices)
assert(len(loc_gat_ind) == len(all_gat_ind))
for i in range(len(loc_gat_ind)):
assert(loc_gat_ind[i] == all_gat_ind[i])
# For each index, verify same values.
local_checked = []
for i in range(len(local_gathered.indices)):
local_checked.append(False)
for i in range(len(all_gathered.indices)):
all_index = all_gathered.indices[i]
# TODO(jthestness): Make this lookup quicker using sorting.
loc_index = -1
for j in range(len(local_gathered.indices)):
if local_gathered.indices[j] == all_index and not local_checked[j]:
loc_index = j
local_checked[j] = True
break
assert(loc_index >= 0)
correct_output = local_gathered.values[loc_index][0]
if average_allgather:
correct_output = correct_output / float(num_ranks)
assert(all_gathered.values[i][0] == correct_output)
def test_mpi_allgather(self):
# Get MPI rank
my_rank = int(os.environ['PMI_RANK'])
num_ranks = int(os.environ['PMI_SIZE'])
indices_per_rank = 100
tensor_width = 10
# Create IndexedSlices for each rank, some with overlapping indices.
to_gather_indices = []
to_gather_values = []
to_gather = []
for rank_id in range(num_ranks):
indices = []
values = []
my_multiple = rank_id + 1
current_index = my_multiple
for i in range(indices_per_rank):
indices.append(current_index)
ones_tensor = tf.ones([tensor_width])
values.append(tf.multiply(ones_tensor,
tf.fill(ones_tensor.get_shape(),
float(current_index))))
current_index += my_multiple
concat_ind = tf.stack(indices)
concat_vals = tf.stack(values)
to_gather_indices.append(concat_ind)
to_gather_values.append(concat_vals)
to_gather.append(tf.IndexedSlices(concat_vals, concat_ind))
# Collect the local IndexedSlices (indices and values) to create
# correct IndexedSlices output.
correct_gather_indices = tf.concat(to_gather_indices, 0)
correct_gather_values = tf.concat(to_gather_values, 0)
correct_gather = tf.IndexedSlices(correct_gather_values,
correct_gather_indices)
all_gather = mpi.allreduce(to_gather[my_rank], average_allgather)
# NOTE: This assumes that device IDs are numbered the same as ranks.
gpu_options = tf.GPUOptions(visible_device_list=str(my_rank))
config = tf.ConfigProto(gpu_options=gpu_options)
# MPI Session to test allgather.
with mpi.Session(config=config) as sess:
sess.run(tf.global_variables_initializer())
all_gathered, local_gathered = sess.run([all_gather, correct_gather])
# Compare all_gathered with local_gathered.
self.checkAllgather(num_ranks, all_gathered, local_gathered)
if __name__ == '__main__':
test.main()
|