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> 41 static void SliceHelper(int iters, int size) { 42 testing::StopTiming(); 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 testing::BytesProcessed(static_cast<int64>(iters) * kDim * size * sizeof(T)); 74 testing::StartTiming(); 75 test::Benchmark("cpu", g).Run(iters); 76 testing::UseRealTime(); 77 } 78 79 static void BM_SliceFloat(int iters, int dim2) { 80 SliceHelper<float>(iters, dim2); 81 } 82 83 BENCHMARK(BM_SliceFloat)->Arg(100)->Arg(1000)->Arg(10000); 84 85 static void BM_SliceComplex64(int iters, int dim2) { 86 SliceHelper<std::complex<float>>(iters, dim2); 87 } 88 89 BENCHMARK(BM_SliceComplex64)->Arg(100)->Arg(1000)->Arg(10000); 90 91 static void BM_SliceBFloat16(int iters, int dim2) { 92 SliceHelper<bfloat16>(iters, dim2); 93 } 94 95 BENCHMARK(BM_SliceBFloat16)->Arg(100)->Arg(1000)->Arg(10000); 96 97 static void BM_ValidateStridedSliceOp(int iters) { 98 testing::StopTiming(); 99 int kDim = 100; 100 int kMaxSize = 15000; 101 int size = 100; 102 Tensor begin = test::AsTensor<int32>({10, 10}); 103 Tensor end = test::AsTensor<int32>({10 + kDim, 10 + size}); 104 Tensor strides = test::AsTensor<int32>({1, 1}); 105 TensorShape input_shape({2 * kDim, kMaxSize}); 106 107 testing::StartTiming(); 108 for (int i = 0; i < iters; ++i) { 109 TensorShape processing_shape, final_shape; 110 bool is_identity = true, slice_dim0 = true, is_simple_slice = true; 111 gtl::InlinedVector<int64, 4> begin_out, end_out, strides_out; 112 const int32 begin_mask = 0; 113 const int32 end_mask = 0; 114 const int32 ellipsis_mask = 0; 115 const int32 new_axis_mask = 0; 116 const int32 shrink_axis_mask = 0; 117 118 TF_CHECK_OK(ValidateStridedSliceOp( 119 &begin, &end, strides, input_shape, begin_mask, end_mask, ellipsis_mask, 120 new_axis_mask, shrink_axis_mask, &processing_shape, &final_shape, 121 &is_identity, &is_simple_slice, &slice_dim0, &begin_out, &end_out, 122 &strides_out)); 123 } 124 } 125 126 BENCHMARK(BM_ValidateStridedSliceOp); 127 128 } // namespace 129 } // namespace tensorflow 130