1 # Copyright 2017 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 """Tests for matmul_benchmark.py.""" 16 17 from __future__ import absolute_import 18 from __future__ import division 19 from __future__ import print_function 20 21 import itertools 22 import numpy as np 23 24 from tensorflow.core.framework import graph_pb2 25 from tensorflow.core.framework import node_def_pb2 26 from tensorflow.python.framework import ops 27 from tensorflow.python.ops import matmul_benchmark 28 from tensorflow.python.platform import test as googletest 29 from tensorflow.python.platform import tf_logging 30 31 32 def BuildGraphTest(n, m, k, transpose_a, transpose_b, dtype): 33 34 def Test(self): 35 if not googletest.is_gpu_available(): 36 tf_logging.info("Skipping BuildGraphTest %s", 37 (n, m, k, transpose_a, transpose_b)) 38 return 39 tf_logging.info("Testing BuildGraphTest %s", 40 (n, m, k, transpose_a, transpose_b)) 41 self._VerifyBuildGraph(n, m, k, transpose_a, transpose_b, dtype) 42 43 return Test 44 45 46 def RunGraphTest(n, m, k, transpose_a, transpose_b, dtype): 47 48 def Test(self): 49 if not googletest.is_gpu_available(): 50 tf_logging.info("Skipping RunGraphTest %s", 51 (n, m, k, transpose_a, transpose_b)) 52 return 53 tf_logging.info("Testing RunGraphTest %s", 54 (n, m, k, transpose_a, transpose_b)) 55 self._VerifyRunGraph(n, m, k, transpose_a, transpose_b, dtype) 56 57 return Test 58 59 60 class MatmulBenchmarkTest(googletest.TestCase): 61 62 def _StripNode(self, nd): 63 snode = node_def_pb2.NodeDef(name=nd.name, op=nd.op, input=nd.input) 64 if nd.device: 65 snode.device = nd.device 66 return snode 67 68 def _StripGraph(self, gd): 69 return graph_pb2.GraphDef(node=[self._StripNode(nd) for nd in gd.node]) 70 71 def _VerifyBuildGraph(self, n, m, k, transpose_a, transpose_b, dtype): 72 graph = ops.Graph() 73 with graph.as_default(): 74 matmul_benchmark.build_graph(googletest.gpu_device_name(), n, m, k, 75 transpose_a, transpose_b, dtype) 76 gd = graph.as_graph_def() 77 dev = googletest.gpu_device_name() 78 proto_expected = """ 79 node { name: "random_uniform/shape" op: "Const" device: \"""" + dev + """\" } 80 node { name: "random_uniform/min" op: "Const" device: \"""" + dev + """\" } 81 node { name: "random_uniform/max" op: "Const" device: \"""" + dev + """\" } 82 node { name: "random_uniform/RandomUniform" op: "RandomUniform" input: "random_uniform/shape" device: \"""" + dev + """\" } 83 node { name: "random_uniform/sub" op: "Sub" input: "random_uniform/max" input: "random_uniform/min" device: \"""" + dev + """\" } 84 node { name: "random_uniform/mul" op: "Mul" input: "random_uniform/RandomUniform" input: "random_uniform/sub" device: \"""" + dev + """\" } 85 node { name: "random_uniform" op: "Add" input: "random_uniform/mul" input: "random_uniform/min" device: \"""" + dev + """\" } 86 node { name: "Variable" op: "VariableV2" device: \"""" + dev + """\" } 87 node { name: "Variable/Assign" op: "Assign" input: "Variable" input: "random_uniform" device: \"""" + dev + """\" } 88 node { name: "Variable/read" op: "Identity" input: "Variable" device: \"""" + dev + """\" } 89 node { name: "random_uniform_1/shape" op: "Const" device: \"""" + dev + """\" } 90 node { name: "random_uniform_1/min" op: "Const" device: \"""" + dev + """\" } 91 node { name: "random_uniform_1/max" op: "Const" device: \"""" + dev + """\" } 92 node { name: "random_uniform_1/RandomUniform" op: "RandomUniform" input: "random_uniform_1/shape" device: \"""" + dev + """\" } 93 node { name: "random_uniform_1/sub" op: "Sub" input: "random_uniform_1/max" input: "random_uniform_1/min" device: \"""" + dev + """\" } 94 node { name: "random_uniform_1/mul" op: "Mul" input: "random_uniform_1/RandomUniform" input: "random_uniform_1/sub" device: \"""" + dev + """\" } 95 node { name: "random_uniform_1" op: "Add" input: "random_uniform_1/mul" input: "random_uniform_1/min" device: \"""" + dev + """\" } 96 node { name: "Variable_1" op: "VariableV2" device: \"""" + dev + """\" } 97 node { name: "Variable_1/Assign" op: "Assign" input: "Variable_1" input: "random_uniform_1" device: \"""" + dev + """\" } 98 node { name: "Variable_1/read" op: "Identity" input: "Variable_1" device: \"""" + dev + """\" } 99 node { name: "MatMul" op: "MatMul" input: "Variable/read" input: "Variable_1/read" device: \"""" + dev + """\" } 100 node { name: "group_deps" op: "NoOp" input: "^MatMul" device: \"""" + dev + """\" } 101 """ 102 self.assertProtoEquals(str(proto_expected), self._StripGraph(gd)) 103 104 def _VerifyRunGraph(self, n, m, k, transpose_a, transpose_b, dtype): 105 benchmark_instance = matmul_benchmark.MatmulBenchmark() 106 duration = benchmark_instance.run_graph(googletest.gpu_device_name(), n, m, 107 k, transpose_a, transpose_b, 1, 108 dtype) 109 self.assertTrue(duration > 1e-6) 110 111 112 if __name__ == "__main__": 113 dtypes = [np.float32, np.float64] 114 index = 0 115 for _dtype in dtypes: 116 for _n, _m, (_transpose_a, _transpose_b) in itertools.product( 117 [512, 1024], [1, 8, 16, 128], [(False, False), (True, False), 118 (False, True)]): 119 _k = _n 120 setattr(MatmulBenchmarkTest, "testBuildGraph_" + str(index), 121 BuildGraphTest(_n, _m, _k, _transpose_a, _transpose_b, _dtype)) 122 setattr(MatmulBenchmarkTest, "testRunGraph_" + str(index), 123 RunGraphTest(_n, _m, _k, _transpose_a, _transpose_b, _dtype)) 124 index += 1 125 googletest.main() 126
