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
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
|
/* Copyright 2015 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.
==============================================================================*/
#include <functional>
#include <vector>
#include "tensorflow/core/common_runtime/device.h"
#include "tensorflow/core/common_runtime/device_factory.h"
#include "tensorflow/core/common_runtime/kernel_benchmark_testlib.h"
#include "tensorflow/core/framework/allocator.h"
#include "tensorflow/core/framework/fake_input.h"
#include "tensorflow/core/framework/node_def_builder.h"
#include "tensorflow/core/framework/op_kernel.h"
#include "tensorflow/core/framework/tensor.h"
#include "tensorflow/core/framework/tensor_testutil.h"
#include "tensorflow/core/framework/types.h"
#include "tensorflow/core/framework/types.pb.h"
#include "tensorflow/core/graph/node_builder.h"
#include "tensorflow/core/graph/testlib.h"
#include "tensorflow/core/kernels/ops_testutil.h"
#include "tensorflow/core/kernels/ops_util.h"
#include "tensorflow/core/platform/test.h"
#include "tensorflow/core/platform/test_benchmark.h"
#include "tensorflow/core/public/session_options.h"
#include "tensorflow/core/public/version.h"
namespace tensorflow {
template <typename Index>
static void BM_SegmentReduction(int iters, const string& reduction,
Index num_rows, Index num_cols,
Index segment_size) {
testing::StopTiming();
std::unique_ptr<Device> device(
DeviceFactory::NewDevice("CPU", {}, "/job:a/replica:0/task:0"));
// Create inputs
gtl::InlinedVector<TensorValue, 4> reduction_inputs;
TensorShape shape1({num_rows, num_cols});
Tensor input1(DT_FLOAT, shape1);
reduction_inputs.push_back({nullptr, &input1});
TensorShape shape2({num_rows});
Tensor input2(DataTypeToEnum<Index>::v(), shape2);
test::FillFn<Index>(&input2, [&num_rows, &segment_size](Index i) -> Index {
return std::min(i / segment_size, num_rows - 1);
});
reduction_inputs.push_back({nullptr, &input2});
NodeDef reduction_node_def;
TF_CHECK_OK(NodeDefBuilder(reduction, reduction)
.Input(FakeInput(DT_FLOAT))
.Input(FakeInput(DataTypeToEnum<Index>::v()))
.Finalize(&reduction_node_def));
Status status;
std::unique_ptr<OpKernel> reduction_op(
CreateOpKernel(DEVICE_CPU, device.get(), cpu_allocator(),
reduction_node_def, TF_GRAPH_DEF_VERSION, &status));
OpKernelContext::Params params;
params.device = device.get();
params.frame_iter = FrameAndIter(0, 0);
params.inputs = &reduction_inputs;
params.op_kernel = reduction_op.get();
std::vector<AllocatorAttributes> attrs;
test::SetOutputAttrs(¶ms, &attrs);
std::unique_ptr<OpKernelContext> reduction_context(
new OpKernelContext(¶ms));
reduction_op->Compute(reduction_context.get());
TF_CHECK_OK(reduction_context->status());
testing::StartTiming();
for (int i = 0; i < iters; ++i) {
delete reduction_context->release_output(0).tensor;
reduction_op->Compute(reduction_context.get());
}
int64 bytes_per_iter =
static_cast<int64>(num_rows * num_cols * sizeof(float));
testing::BytesProcessed(bytes_per_iter * iters);
}
#define BM_Reduce(O, R, C, S) \
static void BM_Reduce_##O##_##R##_##C##_##S##_int32(int iters) { \
BM_SegmentReduction<int32>(iters, #O, R, C, S); \
} \
static void BM_Reduce_##O##_##R##_##C##_##S##_int64(int iters) { \
BM_SegmentReduction<int64>(iters, #O, R, C, S); \
} \
BENCHMARK(BM_Reduce_##O##_##R##_##C##_##S##_int32); \
BENCHMARK(BM_Reduce_##O##_##R##_##C##_##S##_int64);
#define BM_Reduce_Arg(R, C, S) \
BM_Reduce(SegmentSum, R, C, S); \
BM_Reduce(SegmentMean, R, C, S);
BM_Reduce_Arg(64, 32, 1);
BM_Reduce_Arg(4096, 128, 1);
BM_Reduce_Arg(16, 8, 2);
BM_Reduce_Arg(64, 32, 2);
BM_Reduce_Arg(4096, 32, 2);
BM_Reduce_Arg(4096, 128, 2);
static void SparseSegmentMeanGradHelper(int iters, float uniqueness, int size) {
testing::StopTiming();
Graph* g = new Graph(OpRegistry::Global());
CHECK_LE(uniqueness, 1.0);
CHECK_GT(uniqueness, 0.0);
const int kNumIndices = size;
Tensor indices(DT_INT32, TensorShape({kNumIndices}));
auto indices_flat = indices.flat<int32>();
Tensor segments(DT_INT32, TensorShape({kNumIndices}));
auto segments_flat = segments.flat<int32>();
int kUniqueIndices = uniqueness * kNumIndices;
Tensor output_dim0(DT_INT32, TensorShape({}));
output_dim0.scalar<int32>()() = kUniqueIndices;
for (int i = 0; i < kNumIndices; ++i) {
indices_flat(i) = (i * 31) % kUniqueIndices;
segments_flat(i) = i * .8;
}
const int kDim1 = segments_flat(kNumIndices - 1) + 1;
const int kDim2 = 128;
Tensor input(DT_FLOAT, TensorShape({kDim1, kDim2}));
input.flat<float>().setRandom();
Node* node;
TF_CHECK_OK(NodeBuilder(g->NewName("n"), "SparseSegmentMeanGrad")
.Input(test::graph::Constant(g, input))
.Input(test::graph::Constant(g, indices))
.Input(test::graph::Constant(g, segments))
.Input(test::graph::Constant(g, output_dim0))
.Attr("T", DT_FLOAT)
.Finalize(g, &node));
testing::UseRealTime();
testing::BytesProcessed(static_cast<int64>(iters) * (kDim1 * kDim2) *
sizeof(float));
testing::StartTiming();
test::Benchmark("cpu", g).Run(iters);
}
static void BM_SparseSegmentMeanGrad_Low(int iters, int size) {
return SparseSegmentMeanGradHelper(iters, 1.0, size);
}
static void BM_SparseSegmentMeanGrad_High(int iters, int size) {
return SparseSegmentMeanGradHelper(iters, 0.01, size);
}
BENCHMARK(BM_SparseSegmentMeanGrad_Low)->Arg(1000)->Arg(100000);
BENCHMARK(BM_SparseSegmentMeanGrad_High)->Arg(1000)->Arg(100000);
} // namespace tensorflow
|
