inference-engine/tests/unit/engines/mkldnn/graph/layers/internal/graph_reshape_test.cpp

   1 // Copyright (C) 2018 Intel Corporation
   2 //
   3 // SPDX-License-Identifier: Apache-2.0
   4 //
   5
   6 #include <gtest/gtest.h>
   7 #include <gmock/gmock-spec-builders.h>
   8 #include "mkldnn_plugin/mkldnn_graph.h"
   9 #include "mock_mkldnn_primitive.hpp"
  10 #include "single_layer_common.hpp"
  11
  12 #include "test_graph.hpp"
  13
  14 #include <mkldnn_plugin/mkldnn_extension_utils.h>
  15 #include "tests_common.hpp"
  16
  17
  18 using namespace ::testing;
  19 using namespace std;
  20 using namespace mkldnn;
  21
  22
  23 struct reshape_test_params {
  24     InferenceEngine::SizeVector in;
  25     InferenceEngine::SizeVector out;
  26     std::vector<size_t> shape;
  27
  28     int axis;
  29     int num_axes;
  30
  31     size_t num_prim_desc;
  32
  33     MKLDNNPlugin::impl_desc_type selectedType;
  34
  35     std::vector<std::function<void(MKLDNNPlugin::PrimitiveDescInfo)>> comp;
  36 };
  37
  38 template <typename data_t>
  39 void ref_reshape(const InferenceEngine::TBlob<data_t> &src, InferenceEngine::TBlob<data_t> &dst) {
  40     const data_t *src_data = src.readOnly();
  41     data_t *dst_data = dst.data();
  42
  43 #pragma omp parallel for
  44     for (int i=0; i < src.size(); i++)
  45         dst_data[i] = src_data[i];
  46 }
  47
  48 class MKLDNNGraphReshapeTests: public TestsCommon,
  49                                      public WithParamInterface<reshape_test_params> {
  50     std::string model_t = R"V0G0N(
  51 <Net Name="Reshape_Only" version="2" precision="FP32" batch="1">
  52     <layers>
  53         <layer name="in1" type="Input" precision="FP32" id="0">
  54             <output>
  55                 <port id="0">
  56 __SRC_DIMS__
  57                 </port>
  58             </output>
  59         </layer>
  60         <layer name="norm" id="1" type="Reshape" precision="FP32">
  61             <data dim="_SHAPE_" axis="_AX_" num_axes="_NAX_"/>
  62
  63             <input>
  64                 <port id="1">
  65 __SRC_DIMS__
  66                 </port>
  67             </input>
  68             <output>
  69                 <port id="2">
  70 __DST_DIMS__
  71                 </port>
  72             </output>
  73         </layer>
  74     </layers>
  75     <edges>
  76         <edge from-layer="0" from-port="0" to-layer="1" to-port="1"/>
  77     </edges>
  78 </Net>
  79 )V0G0N";
  80
  81     std::string getModel(reshape_test_params p) {
  82         std::string model = model_t;
  83
  84                 std::string src_dims;
  85                 for (auto& dim : p.in) {
  86                         src_dims += "<dim>";
  87                         src_dims += std::to_string(dim) + "</dim>\n";
  88                 }
  89                 REPLACE_WITH_STR(model, "__SRC_DIMS__", src_dims);
  90
  91                 std::string dst_dims;
  92                 for (auto& dim : p.out) {
  93                         dst_dims += "<dim>";
  94                         dst_dims += std::to_string(dim) + "</dim>\n";
  95                 }
  96                 REPLACE_WITH_STR(model, "__DST_DIMS__", dst_dims);
  97
  98         REPLACE_WITH_NUM(model, "_AX_", p.axis);
  99         REPLACE_WITH_NUM(model, "_NAX_", p.num_axes);
 100
 101         std::string shape_str;
 102         for (auto& dim : p.shape) {
 103             if (!shape_str.empty())
 104                 shape_str += ",";
 105             shape_str += std::to_string(dim);
 106         }
 107         REPLACE_WITH_STR(model, "_SHAPE_", shape_str);
 108         return model;
 109     }
 110
 111 protected:
 112     virtual void TearDown() {
 113     }
 114
 115     virtual void SetUp() {
 116         try {
 117             TestsCommon::SetUp();
 118             reshape_test_params p = ::testing::WithParamInterface<reshape_test_params>::GetParam();
 119             std::string model = getModel(p);
 120
 121             InferenceEngine::CNNNetReader net_reader;
 122             ASSERT_NO_THROW(net_reader.ReadNetwork(model.data(), model.length()));
 123
 124             MKLDNNGraphTestClass graph;
 125             graph.CreateGraph(net_reader.getNetwork());
 126             auto& nodes = graph.getNodes();
 127             for (int i = 0; i < nodes.size(); i++) {
 128                 if (nodes[i]->getType() == MKLDNNPlugin::Reshape) {
 129                     ASSERT_EQ(p.num_prim_desc, nodes[i]->getSupportedPrimitiveDescriptors().size());
 130                     for (size_t j = 0; j < p.num_prim_desc && j < p.comp.size(); j++) {
 131                         p.comp.at(j)(nodes[i]->getSupportedPrimitiveDescriptors().at(j));
 132                     }
 133                     ASSERT_NE(nullptr, nodes[i]->getSelectedPrimitiveDescriptor());
 134                     ASSERT_EQ(p.selectedType, nodes[i]->getSelectedPrimitiveDescriptor()->getImplementationType());
 135                 }
 136             }
 137
 138             InferenceEngine::Blob::Ptr src = InferenceEngine::make_shared_blob<float, const InferenceEngine::SizeVector>(InferenceEngine::Precision::FP32, InferenceEngine::ANY, p.in);
 139             src->allocate();
 140             fill_data(src->buffer(), src->size());
 141
 142             InferenceEngine::TBlob<float>* srcPtr = dynamic_cast<InferenceEngine::TBlob<float>*>(src.get());
 143
 144             if (srcPtr == nullptr)
 145                 FAIL() << "Cannot cast blob to TBlob<float>.";
 146
 147             InferenceEngine::BlobMap srcs;
 148             srcs.insert(std::pair<std::string, InferenceEngine::Blob::Ptr>("in1", src));
 149
 150             InferenceEngine::OutputsDataMap out;
 151             out = net_reader.getNetwork().getOutputsInfo();
 152             InferenceEngine::BlobMap outputBlobs;
 153
 154             std::pair<std::string, InferenceEngine::DataPtr> item = *out.begin();
 155
 156             InferenceEngine::TBlob<float>::Ptr output;
 157             output = InferenceEngine::make_shared_blob<float>(item.second->getTensorDesc());
 158             output->allocate();
 159             outputBlobs[item.first] = output;
 160
 161             graph.Infer(srcs, outputBlobs);
 162
 163             InferenceEngine::TBlob<float> dst_ref(item.second->getTensorDesc());
 164             dst_ref.allocate();
 165
 166             ref_reshape(*srcPtr, dst_ref);
 167
 168             compare(*output, dst_ref);
 169         } catch (const InferenceEngine::details::InferenceEngineException &e) {
 170             FAIL() << e.what();
 171         }
 172     }
 173 };
 174
 175 TEST_P(MKLDNNGraphReshapeTests, TestsReshape) {}
 176
 177
 178 INSTANTIATE_TEST_CASE_P(
 179         TestsReshape, MKLDNNGraphReshapeTests,
 180         ::testing::Values(
 181         reshape_test_params{ {1, 3, 228, 228}, {1, 24, 2, 3249}, {1, 24, 2, 3249}, 0, -1, 2,
 182             MKLDNNPlugin::impl_desc_type::unknown, { [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 183                 ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 184                 ASSERT_EQ(1, impl.getConfig().inConfs.size());
 185                 ASSERT_EQ(1, impl.getConfig().outConfs.size());
 186                 ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 187                 ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 188         } } },
 189         reshape_test_params{ { 4 },{ 2, 2 },{ 2, 2 }, 0, -1, 2,
 190             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 191             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 192             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 193             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 194             ASSERT_EQ(InferenceEngine::Layout::C, impl.getConfig().inConfs.at(0).desc.getLayout());
 195             ASSERT_EQ(InferenceEngine::Layout::NC, impl.getConfig().outConfs.at(0).desc.getLayout());
 196         } } },
 197         reshape_test_params{ { 4 },{ 1, 2, 2 },{ 1, 2, 2 }, 0, -1, 2,
 198             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 199             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 200             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 201             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 202             ASSERT_EQ(InferenceEngine::Layout::C, impl.getConfig().inConfs.at(0).desc.getLayout());
 203             ASSERT_EQ(InferenceEngine::Layout::CHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 204         } } },
 205         reshape_test_params{ { 4 },{ 1, 4, 1, 1 },{ 1, 4, 1, 1 }, 0, -1, 2,
 206             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 207             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 208             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 209             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 210             ASSERT_EQ(InferenceEngine::Layout::C, impl.getConfig().inConfs.at(0).desc.getLayout());
 211             ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 212         } } },
 213         reshape_test_params{ { 4, 4 },{ 1, 4, 4 },{ 1, 4, 4 }, 0, -1, 2,
 214             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 215             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 216             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 217             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 218             ASSERT_EQ(InferenceEngine::Layout::NC, impl.getConfig().inConfs.at(0).desc.getLayout());
 219             ASSERT_EQ(InferenceEngine::Layout::CHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 220         } } },
 221         reshape_test_params{ { 4, 4 },{ 1, 4, 2, 2 },{ 1, 4, 2, 2 }, 0, -1, 2,
 222             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 223             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 224             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 225             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 226             ASSERT_EQ(InferenceEngine::Layout::NC, impl.getConfig().inConfs.at(0).desc.getLayout());
 227             ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 228         } } },
 229         reshape_test_params{ { 4, 2, 2 },{ 1, 4, 2, 2 },{ 1, 4, 2, 2 }, 0, -1, 2,
 230             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 231             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 232             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 233             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 234             ASSERT_EQ(InferenceEngine::Layout::CHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 235             ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 236         } } },
 237         reshape_test_params{ { 2, 2 },{ 4 },{ 4 }, 0, -1, 2,
 238             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 239             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 240             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 241             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 242             ASSERT_EQ(InferenceEngine::Layout::NC, impl.getConfig().inConfs.at(0).desc.getLayout());
 243             ASSERT_EQ(InferenceEngine::Layout::C, impl.getConfig().outConfs.at(0).desc.getLayout());
 244         } } },
 245             reshape_test_params{ { 1, 2, 2 },{ 4 },{ 4 }, 0, -1, 2,
 246             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 247             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 248             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 249             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 250             ASSERT_EQ(InferenceEngine::Layout::CHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 251             ASSERT_EQ(InferenceEngine::Layout::C, impl.getConfig().outConfs.at(0).desc.getLayout());
 252         } } },
 253         reshape_test_params{ { 1, 1, 2, 2 },{ 4 },{ 4 }, 0, -1, 2,
 254             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 255             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 256             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 257             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 258             ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 259             ASSERT_EQ(InferenceEngine::Layout::C, impl.getConfig().outConfs.at(0).desc.getLayout());
 260         } } },
 261         reshape_test_params{ { 4, 2, 2 },{ 4, 4 },{ 4, 4 }, 0, -1, 2,
 262             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 263             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 264             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 265             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 266             ASSERT_EQ(InferenceEngine::Layout::CHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 267             ASSERT_EQ(InferenceEngine::Layout::NC, impl.getConfig().outConfs.at(0).desc.getLayout());
 268         } } },
 269         reshape_test_params{ { 1, 4, 2, 2 },{ 4, 4 },{ 4, 4 }, 0, -1, 2,
 270             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 271             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 272             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 273             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 274             ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 275             ASSERT_EQ(InferenceEngine::Layout::NC, impl.getConfig().outConfs.at(0).desc.getLayout());
 276         } } },
 277         reshape_test_params{ { 1, 4, 2, 2 },{ 4, 2, 2 },{ 4, 2, 2 }, 0, -1, 2,
 278             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 279             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 280             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 281             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 282             ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 283             ASSERT_EQ(InferenceEngine::Layout::CHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 284         } } },
 285         reshape_test_params{ { 1, 4, 2, 2 },{ 4, 2, 2, 1, 1 },{ 4, 2, 2, 1, 1 }, 0, -1, 2,
 286             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 287             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 288             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 289             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 290             ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 291             ASSERT_EQ(InferenceEngine::Layout::BLOCKED, impl.getConfig().outConfs.at(0).desc.getLayout());
 292         } } },
 293         reshape_test_params{ { 4, 2, 2, 1, 1 },{ 1, 4, 2, 2 },{ 1, 4, 2, 2 }, 0, -1, 2,
 294             MKLDNNPlugin::impl_desc_type::unknown,{ [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 295             ASSERT_EQ(MKLDNNPlugin::impl_desc_type::unknown, impl.getImplementationType());
 296             ASSERT_EQ(1, impl.getConfig().inConfs.size());
 297             ASSERT_EQ(1, impl.getConfig().outConfs.size());
 298             ASSERT_EQ(InferenceEngine::Layout::BLOCKED, impl.getConfig().inConfs.at(0).desc.getLayout());
 299             ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 300         } } }
 301 ));