Add tests which cover CudaGraphTopologyRunner functionality

openvinotoolkit · Jun 26, 2023 · 3ed00bd · 3ed00bd
1 parent 4e0bfec
commit 3ed00bd
Showing 1 changed file with 127 additions and 0 deletions.
diff --git a/modules/nvidia_plugin/tests/unit/cuda_graph_topology_runner_test.cpp b/modules/nvidia_plugin/tests/unit/cuda_graph_topology_runner_test.cpp
@@ -0,0 +1,127 @@
+// Copyright (C) 2020-2023 Intel Corporation
+// SPDX-License-Identifier: Apache-2.0
+//
+
+#include <gtest/gtest.h>
+#include <gmock/gmock.h>
+#include <cuda_graph_topology_runner.hpp>
+#include <ops/parameter.hpp>
+#include <ops/result.hpp>
+#include <cuda_profiler.hpp>
+#include "test_networks.hpp"
+
+using namespace ov::nvidia_gpu;
+using namespace testing;
+
+class CudaGraphTopologyRunnerTest : public Test {
+protected:
+    static std::map<std::string, std::size_t> PopulateInputIndeces(std::shared_ptr<ov::Model> model) {
+        std::map<std::string, std::size_t> inputIndeces;
+        for (const auto& parameter : model->get_parameters()) {
+            const auto& parameter_index = model->get_parameter_index(parameter);
+            inputIndeces.emplace(ParameterOp::GetInputTensorName(*parameter), parameter_index);
+        }
+        return inputIndeces;
+    }
+
+    static std::map<std::string, std::size_t> PopulateOutputIndeces(std::shared_ptr<ov::Model> model) {
+        std::map<std::string, std::size_t> outputIndeces;
+        for (auto& result : model->get_results()) {
+            const auto& result_index = model->get_result_index(result->input_value(0));
+            for (const auto& outputName : ResultOp::GetOutputTensorName(*result)) {
+                outputIndeces.emplace(outputName, result_index);
+            }
+        }
+        return outputIndeces;
+    }
+
+    static std::vector<std::shared_ptr<ov::Tensor>> PopulateTensors(const std::vector<ov::Output<ov::Node>>& nodes) {
+        std::vector<std::shared_ptr<ov::Tensor>> ret;
+        for (const auto& node : nodes)
+            ret.push_back(std::make_shared<ov::Tensor>(node.get_element_type(), node.get_shape()));
+        return ret;
+    }
+
+    std::shared_ptr<ov::Model> model_ {CreateMatMulTestNetwork()};
+    CreationContext creationContext_ {{}, false};
+    ThreadContext threadContext_ {{}};
+    CancellationToken cancellationToken_ {};
+    CudaGraphContext cudaGraphContext_ {};
+    CudaGraphTopologyRunner runner_ {creationContext_, model_};
+    Profiler profiler_ {false, runner_.GetSubGraph()};
+    std::vector<std::shared_ptr<ov::Tensor>> inputTensors_ {PopulateTensors(model_->inputs())};
+    std::vector<std::shared_ptr<ov::Tensor>> outputTensors_ {PopulateTensors(model_->outputs())};
+    std::map<std::string, std::size_t> inputIndeces_ {PopulateInputIndeces(model_)};
+    std::map<std::string, std::size_t> outputIndeces_ {PopulateOutputIndeces(model_)};
+    InferenceRequestContext inferRequestContext_{inputTensors_,
+                                                inputIndeces_,
+                                                outputTensors_,
+                                                outputIndeces_,
+                                                threadContext_,
+                                                cancellationToken_,
+                                                profiler_,
+                                                cudaGraphContext_,
+                                                false};
+    DeviceMemBlock deviceMemBlock_ {runner_.GetSubGraph().memoryManager().mutableTensorsMemoryModel()};
+};
+
+TEST_F(CudaGraphTopologyRunnerTest, InstantiateGraphExec) {
+    runner_.UpdateContext(inferRequestContext_, deviceMemBlock_);
+    EXPECT_TRUE(inferRequestContext_.getCudaGraphContext().graphExec.has_value());
+}
+
+TEST_F(CudaGraphTopologyRunnerTest, BasicRun) {
+    EXPECT_NO_THROW(runner_.UpdateContext(inferRequestContext_, deviceMemBlock_));
+    EXPECT_NO_THROW(runner_.Run(inferRequestContext_, deviceMemBlock_));
+}
+
+TEST_F(CudaGraphTopologyRunnerTest, CheckGraphExecIsInstantiatedOnce) {
+    runner_.UpdateContext(inferRequestContext_, deviceMemBlock_);
+    CUDA::GraphExec* exec = &inferRequestContext_.getCudaGraphContext().graphExec.value();
+    runner_.UpdateContext(inferRequestContext_, deviceMemBlock_);
+    EXPECT_EQ(&inferRequestContext_.getCudaGraphContext().graphExec.value(), exec);
+}
+
+TEST_F(CudaGraphTopologyRunnerTest, CheckMemcpyNodesArePopulated) {
+    runner_.UpdateContext(inferRequestContext_, deviceMemBlock_);
+    EXPECT_GT(inferRequestContext_.getCudaGraphContext().parameterNodes.size(), 0);
+    EXPECT_GT(inferRequestContext_.getCudaGraphContext().resultNodes.size(), 0);
+}
+
+TEST_F(CudaGraphTopologyRunnerTest, CheckMemcpyNodesAreUpdated) {
+    runner_.UpdateContext(inferRequestContext_, deviceMemBlock_);
+    auto paramNodes = cudaGraphContext_.parameterNodes;
+    auto resultNodes = cudaGraphContext_.resultNodes;
+    std::vector<std::shared_ptr<ov::Tensor>> inputTensors {PopulateTensors(model_->inputs())};
+    std::vector<std::shared_ptr<ov::Tensor>> outputTensors {PopulateTensors(model_->outputs())};
+    InferenceRequestContext inferRequestContext{inputTensors,
+                                                    inputIndeces_,
+                                                    outputTensors,
+                                                    outputIndeces_,
+                                                    threadContext_,
+                                                    cancellationToken_,
+                                                    profiler_,
+                                                    cudaGraphContext_,
+                                                    false};
+    runner_.UpdateContext(inferRequestContext, deviceMemBlock_);
+    EXPECT_NE(cudaGraphContext_.parameterNodes, paramNodes);
+    EXPECT_NE(cudaGraphContext_.resultNodes, resultNodes);
+}
+
+TEST_F(CudaGraphTopologyRunnerTest, CheckMemcpyNodesAreNotUpdatedIfPointersUnchanged) {
+    runner_.UpdateContext(inferRequestContext_, deviceMemBlock_);
+    auto paramNodes = cudaGraphContext_.parameterNodes;
+    auto resultNodes = cudaGraphContext_.resultNodes;
+    InferenceRequestContext inferRequestContext{inputTensors_,
+                                                    inputIndeces_,
+                                                    outputTensors_,
+                                                    outputIndeces_,
+                                                    threadContext_,
+                                                    cancellationToken_,
+                                                    profiler_,
+                                                    cudaGraphContext_,
+                                                    false};
+    runner_.UpdateContext(inferRequestContext, deviceMemBlock_);
+    EXPECT_EQ(cudaGraphContext_.parameterNodes, paramNodes);
+    EXPECT_EQ(cudaGraphContext_.resultNodes, resultNodes);
+}