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

   1 // Copyright (C) 2018-2019 Intel Corporation
   2 // SPDX-License-Identifier: Apache-2.0
   3 //
   4
   5 #include <gtest/gtest.h>
   6 #include <gmock/gmock-spec-builders.h>
   7 #include "mkldnn_plugin/mkldnn_graph.h"
   8
   9 #include "test_graph.hpp"
  10
  11 #include "single_layer_common.hpp"
  12 #include <mkldnn_plugin/mkldnn_extension_utils.h>
  13 #include <inference_engine/cnn_network_impl.hpp>
  14 #include "tests_common.hpp"
  15
  16
  17 using namespace ::testing;
  18 using namespace std;
  19 using namespace mkldnn;
  20
  21
  22 struct lrn_test_params {
  23     struct {
  24         size_t n;
  25         size_t c;
  26         size_t h;
  27         size_t w;
  28     } in;
  29
  30     size_t local_size;
  31     float alpha;
  32     float beta;
  33     size_t k;
  34
  35     size_t num_prim_desc;
  36
  37     int selectedType;
  38
  39     std::vector<std::function<void(MKLDNNPlugin::PrimitiveDescInfo)>> comp;
  40 };
  41
  42 template <typename data_t>
  43 void ref_lrn(const InferenceEngine::TBlob<data_t> &src, InferenceEngine::TBlob<data_t> &dst, lrn_test_params prm)
  44 {
  45     size_t IW = prm.in.w;
  46     size_t IH = prm.in.h;
  47     size_t IC = prm.in.c;
  48
  49     const data_t *src_data = src.readOnly();
  50     data_t *dst_data = dst.data();
  51
  52     for (uint32_t c = 0; c < IC; c++) {
  53         for (uint32_t h = 0; h < IH; h++) {
  54             for (uint32_t w = 0; w < IW; w++) {
  55                 uint32_t oidx = c * IH * IW
  56                                 + h * IW + w;
  57
  58                 uint32_t sz = prm.local_size;
  59                 int32_t c_start = c - sz / 2;
  60                 int32_t c_end = c_start + sz;
  61                 if (c_start < 0) c_start = 0;
  62                 if (c_end > (int32_t)IC) c_end = IC;
  63                 data_t sum = 0.0;
  64                 for (int32_t c1 = c_start; c1 < c_end; c1++) {
  65                     uint32_t idx = c1 * IH * IW + h * IW + w;
  66                     data_t s = src_data[idx];
  67
  68                     sum += s * s;
  69                 }
  70
  71                 data_t norm_coef = powf(1. + prm.alpha * sum / sz, -prm.beta);
  72                 dst_data[oidx] = norm_coef * src_data[oidx];
  73             }
  74         }
  75     }
  76 }
  77
  78 class MKLDNNGraphLrnTests: public TestsCommon,
  79                                      public WithParamInterface<lrn_test_params> {
  80     std::string model_t = R"V0G0N(
  81 <Net Name="Lrn_Only" version="2" precision="FP32" batch="1">
  82     <layers>
  83         <layer name="in1" type="Input" precision="FP32" id="0">
  84             <output>
  85                 <port id="0">
  86                     <dim>_IN_</dim>
  87                     <dim>_IC_</dim>
  88                     <dim>_IH_</dim>
  89                     <dim>_IW_</dim>
  90                 </port>
  91             </output>
  92         </layer>
  93         <layer name="norm" id="1" type="LRN" precision="FP32">
  94             <lrn local_size="_LS_" alpha="_A_" beta="_B_" k="_K_" region="ACROSS" />
  95
  96             <input>
  97                 <port id="1">
  98                     <dim>_IN_</dim>
  99                     <dim>_IC_</dim>
 100                     <dim>_IH_</dim>
 101                     <dim>_IW_</dim>
 102                 </port>
 103             </input>
 104             <output>
 105                 <port id="2">
 106                     <dim>_IN_</dim>
 107                     <dim>_IC_</dim>
 108                     <dim>_IH_</dim>
 109                     <dim>_IW_</dim>
 110                 </port>
 111             </output>
 112         </layer>
 113     </layers>
 114     <edges>
 115         <edge from-layer="0" from-port="0" to-layer="1" to-port="1"/>
 116     </edges>
 117 </Net>
 118 )V0G0N";
 119
 120 protected:
 121     std::string getModel(lrn_test_params p) {
 122         std::string model = model_t;
 123
 124         REPLACE_WITH_NUM(model, "_IW_", p.in.w);
 125         REPLACE_WITH_NUM(model, "_IH_", p.in.h);
 126         REPLACE_WITH_NUM(model, "_IC_", p.in.c);
 127         REPLACE_WITH_NUM(model, "_IN_", p.in.n);
 128
 129         REPLACE_WITH_NUM(model, "_LS_", p.local_size);
 130         REPLACE_WITH_NUM(model, "_A_", p.alpha);
 131         REPLACE_WITH_NUM(model, "_B_", p.beta);
 132         REPLACE_WITH_NUM(model, "_K_", p.k);
 133
 134         return model;
 135     }
 136
 137     virtual void TearDown() {
 138     }
 139
 140     virtual void SetUp() {
 141         try {
 142             TestsCommon::SetUp();
 143             lrn_test_params p = ::testing::WithParamInterface<lrn_test_params>::GetParam();
 144             std::string model = getModel(p);
 145
 146             InferenceEngine::CNNNetReader net_reader;
 147             ASSERT_NO_THROW(net_reader.ReadNetwork(model.data(), model.length()));
 148
 149             MKLDNNGraphTestClass graph;
 150             graph.CreateGraph(net_reader.getNetwork());
 151             auto& nodes = graph.getNodes();
 152             for (int i = 0; i < nodes.size(); i++) {
 153                 if (nodes[i]->getType() == MKLDNNPlugin::Lrn) {
 154                     ASSERT_LE(p.num_prim_desc, nodes[i]->getSupportedPrimitiveDescriptors().size());
 155                     for (size_t j = 0; j < p.num_prim_desc && j < p.comp.size(); j++) {
 156                         p.comp.at(j)(nodes[i]->getSupportedPrimitiveDescriptors().at(j));
 157                     }
 158                     ASSERT_NE(nullptr, nodes[i]->getSelectedPrimitiveDescriptor());
 159                     ASSERT_EQ(p.selectedType,
 160                               nodes[i]->getSelectedPrimitiveDescriptor()->getImplementationType() & p.selectedType);
 161                 }
 162             }
 163             ASSERT_EQ(3, nodes.size());
 164
 165             InferenceEngine::SizeVector dims_src = {p.in.n, p.in.c, p.in.h, p.in.w};
 166
 167             InferenceEngine::Blob::Ptr src = InferenceEngine::make_shared_blob<float, const InferenceEngine::SizeVector>(InferenceEngine::Precision::FP32, InferenceEngine::NCHW, dims_src);
 168             src->allocate();
 169             fill_data(src->buffer(), src->size());
 170
 171             InferenceEngine::TBlob<float>* srcPtr = dynamic_cast<InferenceEngine::TBlob<float>*>(src.get());
 172
 173             if (srcPtr == nullptr)
 174                 FAIL() << "Cannot cast blob to TBlob<float>.";
 175
 176             InferenceEngine::BlobMap srcs;
 177             srcs.insert(std::pair<std::string, InferenceEngine::Blob::Ptr>("in1", src));
 178
 179             InferenceEngine::OutputsDataMap out;
 180             out = net_reader.getNetwork().getOutputsInfo();
 181             InferenceEngine::BlobMap outputBlobs;
 182
 183             std::pair<std::string, InferenceEngine::DataPtr> item = *out.begin();
 184
 185             InferenceEngine::TBlob<float>::Ptr output;
 186             output = InferenceEngine::make_shared_blob<float>(item.second->getTensorDesc());
 187             output->allocate();
 188             outputBlobs[item.first] = output;
 189
 190             graph.Infer(srcs, outputBlobs);
 191
 192             InferenceEngine::TBlob<float> dst_ref(item.second->getTensorDesc());
 193             dst_ref.allocate();
 194
 195             ref_lrn(*srcPtr, dst_ref, p);
 196
 197             compare(*output, dst_ref);
 198         } catch (const InferenceEngine::details::InferenceEngineException &e) {
 199             FAIL() << e.what();
 200         }
 201     }
 202 };
 203
 204 TEST_P(MKLDNNGraphLrnTests, TestsLrn) {}
 205
 206 INSTANTIATE_TEST_CASE_P(
 207         TestsLrn, MKLDNNGraphLrnTests,
 208         ::testing::Values(
 209                 lrn_test_params{
 210                         {1, 3, 228, 228},
 211                         5, 0.0001f, 0.75f, 1, 3, MKLDNNPlugin::impl_desc_type::ref_any, {
 212                                 [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 213                                     ASSERT_EQ(MKLDNNPlugin::impl_desc_type::ref_any, impl.getImplementationType());
 214                                     ASSERT_EQ(1, impl.getConfig().inConfs.size());
 215                                     ASSERT_EQ(1, impl.getConfig().outConfs.size());
 216                                     ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().inConfs.at(0).desc.getLayout());
 217                                     ASSERT_EQ(InferenceEngine::Layout::NCHW, impl.getConfig().outConfs.at(0).desc.getLayout());
 218                                 },
 219                                 [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 220                                     ASSERT_EQ(MKLDNNPlugin::impl_desc_type::ref_any, impl.getImplementationType());
 221                                     ASSERT_EQ(1, impl.getConfig().inConfs.size());
 222                                     ASSERT_EQ(1, impl.getConfig().outConfs.size());
 223                                     ASSERT_EQ(InferenceEngine::Layout::BLOCKED, impl.getConfig().inConfs.at(0).desc.getLayout());
 224                                     ASSERT_EQ(InferenceEngine::Layout::BLOCKED, impl.getConfig().outConfs.at(0).desc.getLayout());
 225                                 },
 226                                 [](MKLDNNPlugin::PrimitiveDescInfo impl) {
 227                                     ASSERT_EQ(MKLDNNPlugin::impl_desc_type::ref_any, impl.getImplementationType());
 228                                     ASSERT_EQ(1, impl.getConfig().inConfs.size());
 229                                     ASSERT_EQ(1, impl.getConfig().outConfs.size());
 230                                     ASSERT_EQ(InferenceEngine::Layout::BLOCKED, impl.getConfig().inConfs.at(0).desc.getLayout());
 231                                     ASSERT_EQ(InferenceEngine::Layout::BLOCKED, impl.getConfig().outConfs.at(0).desc.getLayout());
 232                                 }
 233                         }},
 234                 lrn_test_params{{1, 16, 228, 228}, 5, 0.0001f, 0.75f, 1, 3, MKLDNNPlugin::impl_desc_type::jit}));
 235
 236 class MKLDNNGraphDynBatchLrnTests: public MKLDNNGraphLrnTests {
 237 protected:
 238     virtual void SetUp() {
 239         try {
 240             TestsCommon::SetUp();
 241             lrn_test_params p = ::testing::WithParamInterface<lrn_test_params>::GetParam();
 242             std::string model = getModel(p);
 243             size_t MB = p.in.n;
 244             if (MB < 2)
 245                 MB = 2;
 246
 247             InferenceEngine::CNNNetReader net_reader;
 248             ASSERT_NO_THROW(net_reader.ReadNetwork(model.data(), model.length()));
 249             InferenceEngine::CNNNetwork network = net_reader.getNetwork();
 250             auto implNet = dynamic_cast<InferenceEngine::details::CNNNetworkImpl *>(&((InferenceEngine::ICNNNetwork&)network));
 251             ASSERT_NE(nullptr, implNet) << "Failed to cast ICNNNetwork to CNNNetworkImpl";
 252             InferenceEngine::ResponseDesc resp;
 253             InferenceEngine::StatusCode sts  = implNet->setBatchSizeReshape(MB, &resp);
 254             ASSERT_EQ((int)InferenceEngine::StatusCode::OK, sts) << resp.msg;
 255
 256             MKLDNNGraphTestClass graph;
 257             graph.setProperty({{InferenceEngine::PluginConfigParams::KEY_DYN_BATCH_ENABLED, InferenceEngine::PluginConfigParams::YES}});
 258             graph.CreateGraph(net_reader.getNetwork());
 259
 260             InferenceEngine::SizeVector dims_src = {MB, p.in.c, p.in.h, p.in.w};
 261
 262             InferenceEngine::Blob::Ptr src = InferenceEngine::make_shared_blob<float, const InferenceEngine::SizeVector>(InferenceEngine::Precision::FP32, InferenceEngine::NCHW, dims_src);
 263             src->allocate();
 264             fill_data(src->buffer(), src->size());
 265
 266             InferenceEngine::TBlob<float>* srcPtr = dynamic_cast<InferenceEngine::TBlob<float>*>(src.get());
 267
 268             if (srcPtr == nullptr)
 269                 FAIL() << "Cannot cast blob to TBlob<float>.";
 270
 271             InferenceEngine::BlobMap srcs;
 272             srcs.insert(std::pair<std::string, InferenceEngine::Blob::Ptr>("in1", src));
 273
 274             InferenceEngine::OutputsDataMap out;
 275             out = net_reader.getNetwork().getOutputsInfo();
 276             InferenceEngine::BlobMap outputBlobs;
 277
 278             std::pair<std::string, InferenceEngine::DataPtr> item = *out.begin();
 279
 280             InferenceEngine::TBlob<float>::Ptr output;
 281             output = InferenceEngine::make_shared_blob<float>(item.second->getTensorDesc());
 282             output->allocate();
 283             outputBlobs[item.first] = output;
 284
 285             auto checkLRN = [](const MKLDNNPlugin::MKLDNNNodePtr& node) {
 286                 return node->getType() == MKLDNNPlugin::Lrn;
 287             };
 288             graph.checkDynBatch(srcs, outputBlobs, MB, MB, checkLRN);
 289             graph.checkDynBatch(srcs, outputBlobs, 1, MB, checkLRN);
 290         } catch (const InferenceEngine::details::InferenceEngineException &e) {
 291             FAIL() << e.what();
 292         }
 293     }
 294 };
 295
 296 TEST_P(MKLDNNGraphDynBatchLrnTests, TestsDynBatchLrn) {}
 297
 298 INSTANTIATE_TEST_CASE_P(
 299         TestsDynBatchLrn, MKLDNNGraphDynBatchLrnTests,
 300         ::testing::Values(
 301                 lrn_test_params{{1, 3, 228, 228}, 5, 0.0001f, 0.75f, 1, 3, MKLDNNPlugin::impl_desc_type::ref_any},
 302                 lrn_test_params{{1, 16, 228, 228}, 5, 0.0001f, 0.75f, 1, 3, MKLDNNPlugin::impl_desc_type::jit}));