1 /* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
2
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15
16 #include <functional>
17 #include <memory>
18
19 #include "tensorflow/core/common_runtime/kernel_benchmark_testlib.h"
20 #include "tensorflow/core/framework/allocator.h"
21 #include "tensorflow/core/framework/op_kernel.h"
22 #include "tensorflow/core/framework/tensor.h"
23 #include "tensorflow/core/framework/tensor_testutil.h"
24 #include "tensorflow/core/framework/types.h"
25 #include "tensorflow/core/framework/types.pb.h"
26 #include "tensorflow/core/graph/node_builder.h"
27 #include "tensorflow/core/graph/testlib.h"
28 #include "tensorflow/core/kernels/ops_testutil.h"
29 #include "tensorflow/core/kernels/ops_util.h"
30 #include "tensorflow/core/lib/core/status_test_util.h"
31 #include "tensorflow/core/platform/test.h"
32 #include "tensorflow/core/platform/test_benchmark.h"
33 #include "tensorflow/core/util/strided_slice_op.h"
34
35 namespace tensorflow {
36 namespace {
37
38 // For the benchmark, we set up two 2-dimensional tensors, each kDim1 x 'dim'
39 // in size, and concat them together along "concat_dimension"
40 template <typename T>
SliceHelper(::testing::benchmark::State & state)41 static void SliceHelper(::testing::benchmark::State& state) {
42 const int size = state.range(0);
43 Graph* g = new Graph(OpRegistry::Global());
44 DataType dt = DataTypeToEnum<T>::v();
45 int kDim = 100;
46 int kMaxSize = 15000;
47 CHECK_LT(size, kMaxSize);
48
49 Tensor begin(DT_INT32, TensorShape({2}));
50 begin.flat<int32>()(0) = 10;
51 begin.flat<int32>()(1) = 10;
52
53 Tensor end(DT_INT32, TensorShape({2}));
54 end.flat<int32>()(0) = 10 + kDim;
55 end.flat<int32>()(1) = 10 + size;
56
57 Tensor strides(DT_INT32, TensorShape({2}));
58 strides.flat<int32>()(0) = 1;
59 strides.flat<int32>()(1) = 1;
60
61 Tensor input(dt, TensorShape({2 * kDim, kMaxSize}));
62 input.flat<T>().setRandom();
63
64 Node* node;
65 TF_CHECK_OK(NodeBuilder(g->NewName("n"), "StridedSlice")
66 .Input(test::graph::Constant(g, input))
67 .Input(test::graph::Constant(g, begin))
68 .Input(test::graph::Constant(g, end))
69 .Input(test::graph::Constant(g, strides))
70 .Attr("T", dt)
71 .Finalize(g, &node));
72
73 test::Benchmark("cpu", g, /*old_benchmark_api*/ false).Run(state);
74 state.SetBytesProcessed(static_cast<int64>(state.iterations()) * kDim * size *
75 sizeof(T));
76 }
77
BM_SliceFloat(::testing::benchmark::State & state)78 void BM_SliceFloat(::testing::benchmark::State& state) {
79 SliceHelper<float>(state);
80 }
81
82 BENCHMARK(BM_SliceFloat)->UseRealTime()->Arg(100)->Arg(1000)->Arg(10000);
83
BM_SliceComplex64(::testing::benchmark::State & state)84 void BM_SliceComplex64(::testing::benchmark::State& state) {
85 SliceHelper<std::complex<float>>(state);
86 }
87
88 BENCHMARK(BM_SliceComplex64)->UseRealTime()->Arg(100)->Arg(1000)->Arg(10000);
89
BM_SliceBFloat16(::testing::benchmark::State & state)90 void BM_SliceBFloat16(::testing::benchmark::State& state) {
91 SliceHelper<bfloat16>(state);
92 }
93
94 BENCHMARK(BM_SliceBFloat16)->UseRealTime()->Arg(100)->Arg(1000)->Arg(10000);
95
BM_ValidateStridedSliceOp(::testing::benchmark::State & state)96 void BM_ValidateStridedSliceOp(::testing::benchmark::State& state) {
97 int kDim = 100;
98 int kMaxSize = 15000;
99 int size = 100;
100 Tensor begin = test::AsTensor<int32>({10, 10});
101 Tensor end = test::AsTensor<int32>({10 + kDim, 10 + size});
102 Tensor strides = test::AsTensor<int32>({1, 1});
103 TensorShape input_shape({2 * kDim, kMaxSize});
104
105 for (auto s : state) {
106 TensorShape processing_shape, final_shape;
107 bool is_identity = true, slice_dim0 = true, is_simple_slice = true;
108 gtl::InlinedVector<int64, 4> begin_out, end_out, strides_out;
109 const int32_t begin_mask = 0;
110 const int32_t end_mask = 0;
111 const int32_t ellipsis_mask = 0;
112 const int32_t new_axis_mask = 0;
113 const int32_t shrink_axis_mask = 0;
114
115 TF_CHECK_OK(ValidateStridedSliceOp(
116 &begin, &end, strides, input_shape, begin_mask, end_mask, ellipsis_mask,
117 new_axis_mask, shrink_axis_mask, &processing_shape, &final_shape,
118 &is_identity, &is_simple_slice, &slice_dim0, &begin_out, &end_out,
119 &strides_out));
120 }
121 }
122
123 BENCHMARK(BM_ValidateStridedSliceOp);
124
125 } // namespace
126 } // namespace tensorflow
127