graph_optimization_test.cpp 15.7 KB
Newer Older
1
// Copyright (C) 2018-2019 Intel Corporation
openvino-pushbot's avatar
openvino-pushbot committed
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
// SPDX-License-Identifier: Apache-2.0
//

#include <gtest/gtest.h>
#include <gmock/gmock-spec-builders.h>
#include "mkldnn_plugin/mkldnn_graph.h"

#include "single_layer_common.hpp"
#include <mkldnn_plugin/mkldnn_extension_utils.h>
#include <mkldnn_plugin/mkldnn_extension_mngr.h>
#include "tests_common.hpp"
#include "../test_graph.hpp"


using namespace ::testing;
using namespace std;
using namespace mkldnn;

class MKLDNNGraphOptimizationTests: public TestsCommon {};

TEST_F(MKLDNNGraphOptimizationTests, TestNoFuseConvSumWithOneInput) {
    std::string model = R"V0G0N(
<net name="AlexNet" version="2" batch="1">
    <layers>
        <layer name="data" type="Input" precision="FP32" id="0">
            <output>
                <port id="0">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </output>
        </layer>
        <layer name="conv1" type="Convolution" precision="FP32" id="1">
            <convolution_data stride-x="1" stride-y="1" pad-x="0" pad-y="0" kernel-x="1" kernel-y="1" output="3" group="1"/>
            <input>
                <port id="1">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </input>
            <output>
                <port id="2">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </output>
            <weights offset="0" size="36"/>
            <biases offset="36" size="12"/>
        </layer>
        <layer name="res2a" type="Eltwise" precision="FP32" id="2">
            <elementwise_data operation="sum"/>
            <input>
                <port id="3">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
                <port id="4">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </input>
            <output>
                <port id="5">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </output>
        </layer>
    </layers>
    <edges>
        <edge from-layer="0" from-port="0" to-layer="1" to-port="1"/>
        <edge from-layer="0" from-port="0" to-layer="2" to-port="3"/>
        <edge from-layer="1" from-port="2" to-layer="2" to-port="4"/>
    </edges>
</net>

)V0G0N";

    InferenceEngine::CNNNetReader net_reader;
    ASSERT_NO_THROW(net_reader.ReadNetwork(model.data(), model.length()));

95
    InferenceEngine::TBlob<uint8_t> *weights = new InferenceEngine::TBlob<uint8_t>({ InferenceEngine::Precision::U8, {48}, InferenceEngine::C });
openvino-pushbot's avatar
openvino-pushbot committed
96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209
    weights->allocate();
    float * data = weights->buffer();

    fill_data((float *) weights->buffer(), weights->size() / sizeof(float));

    InferenceEngine::TBlob<uint8_t>::Ptr weights_ptr = InferenceEngine::TBlob<uint8_t>::Ptr(weights);

    net_reader.SetWeights(weights_ptr);

    MKLDNNGraphTestClass graph;
    ASSERT_NO_THROW(graph.CreateGraph(net_reader.getNetwork()));

    bool fused = true;
    auto& nodes = graph.getNodes();
    for (auto &node : nodes) {
        if (node->getType() == MKLDNNPlugin::Convolution) {
            fused = false;
        }
    }
    ASSERT_FALSE(fused);
}

TEST_F(MKLDNNGraphOptimizationTests, TestNoCrashForFuseConvSumAndInput) {
    std::string model = R"V0G0N(
<net name="AlexNet" version="2" batch="1">
    <layers>
        <layer name="data" type="Input" precision="FP32" id="0">
            <output>
                <port id="0">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </output>
        </layer>
        <layer name="conv1" type="Convolution" precision="FP32" id="1">
            <convolution_data stride-x="1" stride-y="1" pad-x="0" pad-y="0" kernel-x="1" kernel-y="1" output="3" group="1"/>
            <input>
                <port id="1">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </input>
            <output>
                <port id="2">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </output>
            <weights offset="0" size="36"/>
            <biases offset="36" size="12"/>
        </layer>
        <layer name="relu1" type="ReLU" precision="FP32" id="2">
            <input>
                <port id="1">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </input>
            <output>
                <port id="2">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </output>
        </layer>
        <layer name="res2a" type="Eltwise" precision="FP32" id="3">
            <elementwise_data operation="sum"/>
            <input>
                <port id="3">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
                <port id="4">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </input>
            <output>
                <port id="5">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>5</dim>
                    <dim>5</dim>
                </port>
            </output>
        </layer>
    </layers>
    <edges>
        <edge from-layer="0" from-port="0" to-layer="1" to-port="1"/>
        <edge from-layer="0" from-port="0" to-layer="2" to-port="1"/>
        <edge from-layer="1" from-port="2" to-layer="3" to-port="3"/>
        <edge from-layer="2" from-port="2" to-layer="3" to-port="4"/>
    </edges>
</net>

)V0G0N";

    InferenceEngine::CNNNetReader net_reader;
    ASSERT_NO_THROW(net_reader.ReadNetwork(model.data(), model.length()));

210
    InferenceEngine::TBlob<uint8_t> *weights = new InferenceEngine::TBlob<uint8_t>({ InferenceEngine::Precision::U8, {48}, InferenceEngine::C });
openvino-pushbot's avatar
openvino-pushbot committed
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
    weights->allocate();
    float * data = weights->buffer();

    fill_data((float *) weights->buffer(), weights->size() / sizeof(float));

    InferenceEngine::TBlob<uint8_t>::Ptr weights_ptr = InferenceEngine::TBlob<uint8_t>::Ptr(weights);

    net_reader.SetWeights(weights_ptr);

    MKLDNNGraphTestClass graph;
    ASSERT_NO_THROW(graph.CreateGraph(net_reader.getNetwork()));

    bool fused = false;
    auto& nodes = graph.getNodes();
    for (auto &node : nodes) {
226
        if (node->isFusedWith(MKLDNNPlugin::Eltwise)) {
openvino-pushbot's avatar
openvino-pushbot committed
227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423
            fused = true;
        }
    }
    ASSERT_TRUE(fused);
}

namespace GraphOptimizationUtils {

using fake_ext_factory = std::function<InferenceEngine::ILayerImplFactory*(const InferenceEngine::CNNLayer *)>;

class FakeReLUImpl : public InferenceEngine::ILayerExecImpl {
public:
    FakeReLUImpl(const InferenceEngine::CNNLayer *layer) {
        cnnLayer = const_cast<InferenceEngine::CNNLayer *>(layer);
    }
    InferenceEngine::StatusCode getSupportedConfigurations(std::vector<InferenceEngine::LayerConfig>& conf, InferenceEngine::ResponseDesc *resp) noexcept override {
        InferenceEngine::LayerConfig config;
        config.dynBatchSupport = 0;
        if (cnnLayer->outData.size() != 1 && cnnLayer->insData.size() != 1)
            return InferenceEngine::GENERAL_ERROR;
        InferenceEngine::DataConfig cfg;
        cfg.constant = false;
        cfg.inPlace = 0;
        InferenceEngine::SizeVector order;
        for(size_t i = 0; i < cnnLayer->outData[0]->getTensorDesc().getDims().size(); i++) {
            order.push_back(i);
        }
        cfg.desc = InferenceEngine::TensorDesc(cnnLayer->outData[0]->getTensorDesc().getPrecision(),
                                               cnnLayer->outData[0]->getTensorDesc().getDims(),
                                               {cnnLayer->outData[0]->getTensorDesc().getDims(), order});
        config.outConfs.push_back(cfg);
        config.inConfs.push_back(cfg);
        conf.push_back(config);
        return InferenceEngine::OK;
    }
    InferenceEngine::StatusCode init(InferenceEngine::LayerConfig& config, InferenceEngine::ResponseDesc *resp) noexcept override {
        if (config.dynBatchSupport)
            return InferenceEngine::NOT_IMPLEMENTED;
        for(auto input : config.inConfs) {
            if (input.constant)
                return InferenceEngine::GENERAL_ERROR;
        }
        for(auto output : config.outConfs) {
            if (output.constant)
                return InferenceEngine::GENERAL_ERROR;
        }
        return InferenceEngine::OK;
    }
    InferenceEngine::StatusCode execute(std::vector<InferenceEngine::Blob::Ptr>& inputs, std::vector<InferenceEngine::Blob::Ptr>& outputs, InferenceEngine::ResponseDesc *resp) noexcept override {
        const float *src_data = inputs[0]->buffer();
        float *dst_data = outputs[0]->buffer();
        if (src_data != dst_data)
            return InferenceEngine::GENERAL_ERROR;
        return InferenceEngine::OK;
    }

private:
    InferenceEngine::CNNLayer* cnnLayer;
};

class FakeReLUFactory : public InferenceEngine::ILayerImplFactory {
public:
    FakeReLUFactory(const InferenceEngine::CNNLayer *layer) {
        cnnLayer = const_cast<InferenceEngine::CNNLayer *>(layer);
    }
    // set output shapes by input shapes.
    InferenceEngine::StatusCode getShapes(const std::vector<InferenceEngine::TensorDesc>& inShapes, std::vector<InferenceEngine::TensorDesc>& outShapes, InferenceEngine::ResponseDesc *resp) noexcept override {
        outShapes.push_back(inShapes[0]);
        return InferenceEngine::OK;
    }
    // First implementation has more priority than next
    InferenceEngine::StatusCode getImplementations(std::vector<InferenceEngine::ILayerImpl::Ptr>& impls, InferenceEngine::ResponseDesc *resp) noexcept override {
        impls.push_back(InferenceEngine::ILayerImpl::Ptr(new FakeReLUImpl(cnnLayer)));
        return InferenceEngine::OK;
    }

private:
    InferenceEngine::CNNLayer * cnnLayer;
};

class FakeFabric : public InferenceEngine::IExtension {
public:
    FakeFabric() {
        factories["ReLU"] = [](const InferenceEngine::CNNLayer * cnnLayer) -> InferenceEngine::ILayerImplFactory* { return new FakeReLUFactory(cnnLayer); };
    }

    virtual ~FakeFabric() {
        factories.clear();
    }

    void GetVersion(const InferenceEngine::Version *&versionInfo) const noexcept override {}
    void SetLogCallback(InferenceEngine::IErrorListener &listener) noexcept override {}
    void Unload() noexcept override {}
    void Release() noexcept override {
        delete this;
    }
    InferenceEngine::StatusCode getPrimitiveTypes(char**& types, unsigned int& size, InferenceEngine::ResponseDesc* resp) noexcept override {
        types = new char *[factories.size()];
        size_t count = 0;
        for (auto it = factories.begin(); it != factories.end(); it++, count ++) {
            types[count] = new char[it->first.size() + 1];
            std::copy(it->first.begin(), it->first.end(), types[count]);
            types[count][it->first.size() ] = '\0';
        }
        return InferenceEngine::OK;
    };
    InferenceEngine::StatusCode getFactoryFor(InferenceEngine::ILayerImplFactory *&factory,
                                              const InferenceEngine::CNNLayer *cnnLayer,
                                              InferenceEngine::ResponseDesc *resp) noexcept override {
        if (factories.find(cnnLayer->type) == factories.end()) {
            std::string errorMsg = std::string("Factory for ") + cnnLayer->type + " wasn't found!";
            errorMsg.copy(resp->msg, sizeof(resp->msg) - 1);
            return InferenceEngine::NOT_FOUND;
        }
        factory = factories[cnnLayer->type](cnnLayer);
        return InferenceEngine::OK;
    }

    InferenceEngine::StatusCode getShapeInferImpl(InferenceEngine::IShapeInferImpl::Ptr& impl, const char* type,
                                                  InferenceEngine::ResponseDesc* resp) noexcept override {
        return InferenceEngine::NOT_IMPLEMENTED;
    }

private:
    std::map<std::string, fake_ext_factory> factories;
};
}

TEST_F(MKLDNNGraphOptimizationTests, TestNoFuseCustomActivation) {
    std::string model = R"V0G0N(
<net name="AlexNet" version="2" batch="1">
    <layers>
        <layer name="data" type="Input" precision="FP32" id="0">
            <output>
                <port id="0">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>227</dim>
                    <dim>227</dim>
                </port>
            </output>
        </layer>
        <layer name="conv1" type="Convolution" precision="FP32" id="1">
            <convolution_data stride-x="4" stride-y="4" pad-x="0" pad-y="0" kernel-x="11" kernel-y="11" output="96" group="1"/>
            <input>
                <port id="1">
                    <dim>1</dim>
                    <dim>3</dim>
                    <dim>227</dim>
                    <dim>227</dim>
                </port>
            </input>
            <output>
                <port id="2">
                    <dim>1</dim>
                    <dim>96</dim>
                    <dim>55</dim>
                    <dim>55</dim>
                </port>
            </output>
            <weights offset="0" size="139392"/>
            <biases offset="139392" size="384"/>
        </layer>
        <layer name="relu1" type="ReLU" precision="FP32" id="2">
            <input>
                <port id="3">
                    <dim>1</dim>
                    <dim>96</dim>
                    <dim>55</dim>
                    <dim>55</dim>
                </port>
            </input>
            <output>
                <port id="4">
                    <dim>1</dim>
                    <dim>96</dim>
                    <dim>55</dim>
                    <dim>55</dim>
                </port>
            </output>
        </layer>
    </layers>
    <edges>
        <edge from-layer="0" from-port="0" to-layer="1" to-port="1"/>
        <edge from-layer="1" from-port="2" to-layer="2" to-port="3"/>
    </edges>
</net>
)V0G0N";

    std::shared_ptr<InferenceEngine::IExtension> extension;
    extension.reset(new GraphOptimizationUtils::FakeFabric());
    MKLDNNPlugin::MKLDNNExtensionManager::Ptr extMgr(new MKLDNNPlugin::MKLDNNExtensionManager());
    extMgr->AddExtension(extension);

    InferenceEngine::CNNNetReader net_reader;
    ASSERT_NO_THROW(net_reader.ReadNetwork(model.data(), model.length()));

424
    InferenceEngine::TBlob<uint8_t> *weights = new InferenceEngine::TBlob<uint8_t>({ InferenceEngine::Precision::U8, {139776}, InferenceEngine::C });
openvino-pushbot's avatar
openvino-pushbot committed
425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445
    weights->allocate();
    float * data = weights->buffer();

    fill_data((float *) weights->buffer(), weights->size() / sizeof(float));

    InferenceEngine::TBlob<uint8_t>::Ptr weights_ptr = InferenceEngine::TBlob<uint8_t>::Ptr(weights);

    net_reader.SetWeights(weights_ptr);

    MKLDNNGraphTestClass graph;
    ASSERT_NO_THROW(graph.CreateGraph(net_reader.getNetwork(), extMgr));

    bool fused = true;
    auto& nodes = graph.getNodes();
    for (auto &node : nodes) {
        if (node->getType() == MKLDNNPlugin::Convolution) {
            fused = false;
        }
    }
    ASSERT_FALSE(fused);
}