[IE][VPU][GT] Add pass to reshape convolution by parameter from IR (#3038)

inference-engine/src/readers/ir_reader/ie_ir_parser.cpp

@@ -547,6 +547,10 @@ std::shared_ptr<ngraph::Node> V10Parser::createNode(const std::vector<ngraph::Ou
         if (pr_data) {
             rtInfo["PrimitivesPriority"] = std::make_shared<::ngraph::VariantWrapper<std::string> >(pr_data.value());
         }
+        const auto aw_data = dn.attribute("alt_width");
+        if (aw_data) {
+            rtInfo["alt_width"] = std::make_shared<::ngraph::VariantWrapper<std::string> >(aw_data.value());
+        }
     }
 
     ngraphNode->set_friendly_name(params.name);

inference-engine/src/vpu/graph_transformer/include/vpu/graph_transformer.hpp

@@ -112,6 +112,7 @@ struct CompilationConfig final {
     bool forcePureTensorIterator = false;
     bool enableMemoryTypesAnnotation = false;
     bool enableWeightsAnalysis = true;
+    bool enableCustomReshapeParam = false;
 
     //
     // Deprecated options

inference-engine/src/vpu/graph_transformer/include/vpu/middleend/pass_manager.hpp

@@ -249,6 +249,8 @@ public:
 
     Pass::Ptr annotateMemoryTypes();
 
+    Pass::Ptr reshapeBeforeConvTiling();
+
 protected:
     StageBuilder::Ptr _stageBuilder;
     BackEnd::Ptr _backEnd;

inference-engine/src/vpu/graph_transformer/include/vpu/private_plugin_config.hpp

@@ -44,6 +44,13 @@ DECLARE_VPU_CONFIG(MYRIAD_ENABLE_EARLY_ELTWISE_RELU_FUSION);
  */
 DECLARE_VPU_CONFIG(MYRIAD_ENABLE_WEIGHTS_ANALYSIS);
 
+/**
+ * @brief Used to enable reshapeBeforeConvTiling pass in cases where
+ * user have reshape parameter "alt_width" in IR.
+ * Default is "NO".
+ */
+DECLARE_VPU_CONFIG(MYRIAD_ENABLE_CUSTOM_RESHAPE_PARAM);
+
 //
 // Debug options
 //

inference-engine/src/vpu/graph_transformer/src/middleend/pass_manager.cpp

@@ -166,6 +166,18 @@ PassSet::Ptr PassManager::buildMiddleEnd() {
             ADD_DUMP_PASS("reshapeDilationConv");
         }
 
+        //
+        // "reshapeBeforeConvTiling" pass changes geometry of convolution stages in order
+        // to get more efficient HW tiling (pass "hwConvTiling") using reshape stages.
+        //
+        // Pass should be located before "adjustDataBatch" because "adjustDataBatch" specifies "origConvOutput" attribute
+        // for convolution in order to provide that information to "hwConvTiling" pass.
+        // Otherwise, "hwConvTiling" will see incorrect values in "origConvOutput" attribute.
+        if (env.config.enableCustomReshapeParam) {
+            ADD_PASS(reshapeBeforeConvTiling);
+            ADD_DUMP_PASS("reshapeBeforeConvTiling");
+        }
+
         ADD_PASS(upliftActivationStages);
         ADD_DUMP_PASS("upliftActivationStages");
 

inference-engine/src/vpu/graph_transformer/src/middleend/passes/reshape_conv.cpp

@@ -0,0 +1,107 @@
+// Copyright (C) 2020 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+// This pass changes geometry of convolution stages in order
+// to get more efficient HW tiling (pass "hwConvTiling") using reshape stages.
+
+#include <vpu/middleend/pass_manager.hpp>
+
+namespace vpu {
+
+namespace {
+
+class PassImpl final : public Pass {
+public:
+    explicit PassImpl(const StageBuilder::Ptr& stageBuilder) : _stageBuilder(stageBuilder) {}
+
+    void run(const Model& model) override;
+
+private:
+    StageBuilder::Ptr _stageBuilder;
+};
+
+void PassImpl::run(const Model& model) {
+    VPU_PROFILE(reshapeBeforeConvTiling);
+
+    for (const auto& stage : model->getStages()) {
+        if (stage->type() != StageType::StubConv) {
+            continue;
+        }
+
+        const auto tryHW = stage->attrs().getOrDefault<bool>("tryHW", false);
+        if (!tryHW) {
+            continue;
+        }
+
+        const auto input = stage->input(0);
+        const auto output = stage->output(0);
+
+        const auto& inputDesc = input->desc();
+        const auto& outputDesc = output->desc();
+
+        if ((inputDesc.dimsOrder() != DimsOrder::NCHW) ||
+            (outputDesc.dimsOrder() != DimsOrder::NCHW)) {
+            continue;
+        }
+
+        if (stage->attrs().get<int>("kernelSizeX") != 1 ||
+            stage->attrs().get<int>("kernelSizeY") != 1)
+            continue;
+
+        int dimH = inputDesc.dim(Dim::H);
+        int dimW = inputDesc.dim(Dim::W);
+        int resultH = 0;
+        int resultW = 0;
+
+        if (stage->origLayer()->params.count("alt_width")) {
+            const auto alt_width = stage->origLayer()->params.at("alt_width");
+            if (!alt_width.empty() &&
+                std::find_if(alt_width.begin(), alt_width.end(),
+                             [](unsigned char c) { return !std::isdigit(c); }) == alt_width.end()) {
+                resultW = std::stoul(alt_width);
+            }
+        }
+
+        if (resultW == 0) {
+            continue;
+        }
+
+        resultH = dimH * dimW / resultW;
+
+        IE_ASSERT(dimH * dimW == resultH * resultW);
+
+        auto inputNewDesc = inputDesc;
+        inputNewDesc.setDim(Dim::W, resultW);
+        inputNewDesc.setDim(Dim::H, resultH);
+
+        auto outputNewDesc = outputDesc;
+        outputNewDesc.setDim(Dim::W, resultW);
+        outputNewDesc.setDim(Dim::H, resultH);
+
+        auto newInput = model->duplicateData(input, "@input-data-after-reshape",
+                inputNewDesc);
+
+        auto newOutput = model->duplicateData(output, "@output-data-before-reshape",
+                outputNewDesc);
+
+        model->replaceStageInput(stage->inputEdge(0), newInput);
+        model->replaceStageOutput(stage->outputEdge(0), newOutput);
+
+        _stageBuilder->addReshapeStage(model,
+                stage->name() + "@copy-reinterpret-input-data",
+                nullptr, input, newInput);
+
+        _stageBuilder->addReshapeStage(model,
+                stage->name() + "@copy-reinterpret-output-data",
+                nullptr, newOutput, output);
+    }
+}
+
+}  // namespace
+
+Pass::Ptr PassManager::reshapeBeforeConvTiling() {
+    return std::make_shared<PassImpl>(_stageBuilder);
+}
+
+}  // namespace vpu

inference-engine/src/vpu/graph_transformer/src/parsed_config.cpp

@@ -69,6 +69,7 @@ IE_SUPPRESS_DEPRECATED_START
         ie::MYRIAD_DISABLE_CONVERT_STAGES,
         ie::MYRIAD_ENABLE_WEIGHTS_ANALYSIS,
         ie::MYRIAD_ENABLE_EARLY_ELTWISE_RELU_FUSION,
+        ie::MYRIAD_ENABLE_CUSTOM_RESHAPE_PARAM,
 
         //
         // Debug options
@@ -186,6 +187,7 @@ void ParsedConfig::parse(const std::map<std::string, std::string>& config) {
     setOption(_compileConfig.disableConvertStages,           switches, config, ie::MYRIAD_DISABLE_CONVERT_STAGES);
     setOption(_compileConfig.enableWeightsAnalysis,          switches, config, ie::MYRIAD_ENABLE_WEIGHTS_ANALYSIS);
     setOption(_compileConfig.enableEarlyEltwiseReLUFusion,   switches, config, ie::MYRIAD_ENABLE_EARLY_ELTWISE_RELU_FUSION);
+    setOption(_compileConfig.enableCustomReshapeParam,       switches, config, ie::MYRIAD_ENABLE_CUSTOM_RESHAPE_PARAM);
 
     setOption(_compileConfig.irWithVpuScalesDir,                       config, ie::MYRIAD_IR_WITH_SCALES_DIRECTORY);
     setOption(_compileConfig.noneLayers,                               config, ie::MYRIAD_NONE_LAYERS, parseStringSet);