path: root/test/src/unit/MultithreadingTests.cpp

// This code contains NVIDIA Confidential Information and is disclosed to you
// under a form of NVIDIA software license agreement provided separately to you.
//
// Notice
// NVIDIA Corporation and its licensors retain all intellectual property and
// proprietary rights in and to this software and related documentation and
// any modifications thereto. Any use, reproduction, disclosure, or
// distribution of this software and related documentation without an express
// license agreement from NVIDIA Corporation is strictly prohibited.
//
// ALL NVIDIA DESIGN SPECIFICATIONS, CODE ARE PROVIDED "AS IS.". NVIDIA MAKES
// NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO
// THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT,
// MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
//
// Information and code furnished is believed to be accurate and reliable.
// However, NVIDIA Corporation assumes no responsibility for the consequences of use of such
// information or for any infringement of patents or other rights of third parties that may
// result from its use. No license is granted by implication or otherwise under any patent
// or patent rights of NVIDIA Corporation. Details are subject to change without notice.
// This code supersedes and replaces all information previously supplied.
// NVIDIA Corporation products are not authorized for use as critical
// components in life support devices or systems without express written approval of
// NVIDIA Corporation.
//
// Copyright (c) 2016-2017 NVIDIA Corporation. All rights reserved.


#include "BlastBaseTest.h"
#include "AssetGenerator.h"

#include <iostream>
#include <memory>
#include "TaskDispatcher.h"

#include "NvBlastActor.h"
#include "NvBlastExtDamageShaders.h"

#if NV_XBOXONE
#undef min
#undef max
#endif

typedef std::function<void(const Nv::Blast::Actor&, NvBlastLog)> ActorTestFunction;
typedef std::function<void(std::vector<NvBlastActor*>&, NvBlastLog)> PostDamageTestFunction;


static void blast(std::set<NvBlastActor*>& actorsToDamage, GeneratorAsset* testAsset, GeneratorAsset::Vec3 localPos, float minRadius, float maxRadius, float compressiveDamage)
{
	std::vector<NvBlastChunkFractureData> chunkEvents; /* num lower-support chunks + bonds */
	std::vector<NvBlastBondFractureData> bondEvents; /* num lower-support chunks + bonds */
	chunkEvents.resize(testAsset->solverChunks.size());
	bondEvents.resize(testAsset->solverBonds.size());

	NvBlastFractureBuffers events = { static_cast<uint32_t>(bondEvents.size()), static_cast<uint32_t>(chunkEvents.size()), bondEvents.data(), chunkEvents.data() };

	std::vector<float> scratch(chunkEvents.size() + bondEvents.size(), 0.0f);

	std::vector<char> splitScratch;
	std::vector<NvBlastActor*> newActorsBuffer(testAsset->solverChunks.size());

	NvBlastExtRadialDamageDesc damage = {
		compressiveDamage,
		{ localPos.x, localPos.y, localPos.z },
		minRadius,
		maxRadius
	};

	NvBlastExtProgramParams programParams =
	{
		&damage,
		nullptr
	};

	NvBlastDamageProgram program = {
		NvBlastExtFalloffGraphShader,
		nullptr
	};

	size_t totalNewActorsCount = 0;
	for (std::set<NvBlastActor*>::iterator k = actorsToDamage.begin(); k != actorsToDamage.end();)
	{
		NvBlastActor* actor = *k;

		NvBlastActorGenerateFracture(&events, actor, program, &programParams, nullptr, nullptr);
		NvBlastActorApplyFracture(&events, actor, &events, nullptr, nullptr);

		bool removeActor = false;

		if (events.bondFractureCount + events.chunkFractureCount > 0)
		{
			NvBlastActorSplitEvent splitEvent;
			splitEvent.newActors = &newActorsBuffer.data()[totalNewActorsCount];
			uint32_t newActorSize = (uint32_t)(newActorsBuffer.size() - totalNewActorsCount);

			splitScratch.resize((size_t)NvBlastActorGetRequiredScratchForSplit(actor, nullptr));
			const size_t newActorsCount = NvBlastActorSplit(&splitEvent, actor, newActorSize, splitScratch.data(), nullptr, nullptr);
			totalNewActorsCount += newActorsCount;
			removeActor = splitEvent.deletedActor != NULL;
		}

		if (removeActor)
		{
			k = actorsToDamage.erase(k);
		}
		else
		{
			++k;
		}
	}
	
	for (size_t i = 0; i < totalNewActorsCount; ++i)
	{
		actorsToDamage.insert(newActorsBuffer[i]);
	}
}


template<int FailLevel, int Verbosity>
class MultithreadingTest : public BlastBaseTest<FailLevel, Verbosity>
{
public:
	MultithreadingTest()
	{

	}

	static void messageLog(int type, const char* msg, const char* file, int line)
	{
		BlastBaseTest<FailLevel, Verbosity>::messageLog(type, msg, file, line);
	}

	static void* alloc(size_t size)
	{
		return BlastBaseTest<FailLevel, Verbosity>::alignedZeroedAlloc(size);
	}

	static void free(void* mem)
	{
		BlastBaseTest<FailLevel, Verbosity>::alignedFree(mem);
	}

	static void testActorVisibleChunks(const Nv::Blast::Actor& actor, NvBlastLog)
	{
		const Nv::Blast::Asset& asset = *actor.getAsset();
		const NvBlastChunk* chunks = asset.getChunks();

		if (actor.isSubSupportChunk())
		{
			EXPECT_EQ(1, actor.getVisibleChunkCount());

			const uint32_t firstVisibleChunkIndex = (uint32_t)Nv::Blast::Actor::VisibleChunkIt(actor);

			EXPECT_EQ(actor.getIndex() - asset.m_graph.m_nodeCount, firstVisibleChunkIndex - asset.m_firstSubsupportChunkIndex);

			// Make sure the visible chunk is subsupport
			// Array of support flags
			std::vector<bool> isSupport(asset.m_chunkCount, false);
			for (uint32_t i = 0; i < asset.m_graph.m_nodeCount; ++i)
			{
				isSupport[asset.m_graph.getChunkIndices()[i]] = true;
			}

			// Climb hierarchy to find support chunk
			uint32_t chunkIndex = firstVisibleChunkIndex;
			while (chunkIndex != Nv::Blast::invalidIndex<uint32_t>())
			{
				if (isSupport[chunkIndex])
				{
					break;
				}
				chunkIndex = chunks[chunkIndex].parentChunkIndex;
			}

			EXPECT_FALSE(Nv::Blast::isInvalidIndex(chunkIndex));
		}
		else
		{
			// Array of visibility flags
			std::vector<bool> isVisible(asset.m_chunkCount, false);
			for (Nv::Blast::Actor::VisibleChunkIt i = actor; (bool)i; ++i)
			{
				isVisible[(uint32_t)i] = true;
			}

			// Mark visible nodes representing graph chunks
			std::vector<bool> visibleChunkFound(asset.m_chunkCount, false);

			// Make sure every graph chunk is represented by a visible chunk
			for (Nv::Blast::Actor::GraphNodeIt i = actor; (bool)i; ++i)
			{
				const uint32_t graphNodeIndex = (uint32_t)i;
				uint32_t chunkIndex = asset.m_graph.getChunkIndices()[graphNodeIndex];
				// Climb hierarchy to find visible chunk
				while (chunkIndex != Nv::Blast::invalidIndex<uint32_t>())
				{
					// Check that chunk owners are accurate
					EXPECT_EQ(actor.getIndex(), actor.getFamilyHeader()->getChunkActorIndices()[chunkIndex]);
					if (isVisible[chunkIndex])
					{
						visibleChunkFound[chunkIndex] = true;
						break;
					}
					chunkIndex = chunks[chunkIndex].parentChunkIndex;
				}
				EXPECT_FALSE(Nv::Blast::isInvalidIndex(chunkIndex));
			}

			// Check that all visible chunks are accounted for
			for (uint32_t i = 0; i < asset.m_chunkCount; ++i)
			{
				EXPECT_EQ(visibleChunkFound[i], isVisible[i]);
			}

			// Make sure that, if all siblings are intact, they are invisible
			for (uint32_t i = 0; i < asset.m_chunkCount; ++i)
			{
				bool allIntact = true;
				bool noneVisible = true;
				if (chunks[i].firstChildIndex < asset.getUpperSupportChunkCount()) // Do not check subsupport
				{
					for (uint32_t j = chunks[i].firstChildIndex; j < chunks[i].childIndexStop; ++j)
					{
						allIntact = allIntact && actor.getFamilyHeader()->getChunkActorIndices()[j] == actor.getIndex();
						noneVisible = noneVisible && !isVisible[j];
					}
					EXPECT_TRUE(!allIntact || noneVisible);
				}
			}
		}
	}

	class DamageActorTask : public TaskDispatcher::Task
	{
	public:
		DamageActorTask(NvBlastActor* actor, GeneratorAsset* asset, GeneratorAsset::Vec3 localPos, float minRadius, float maxRadius, float compressiveDamage, ActorTestFunction testFunction)
			: m_asset(asset)
			, m_localPos(localPos)
			, m_minRadius(minRadius)
			, m_maxRadius(maxRadius)
			, m_compressiveDamage(compressiveDamage)
			, m_testFunction(testFunction)
		{
			m_actors.insert(actor);
		}

		virtual void process()
		{
			blast(m_actors, m_asset, m_localPos, m_minRadius, m_maxRadius, m_compressiveDamage);

			// Test individual actors
			if (m_testFunction != nullptr)
			{
				for (std::set<NvBlastActor*>::iterator k = m_actors.begin(); k != m_actors.end(); ++k)
				{
					m_testFunction(*static_cast<Nv::Blast::Actor*>(*k), messageLog);
				}
			}
		}

		const std::set<NvBlastActor*>& getResult() const { return m_actors; }

	private:
		std::set<NvBlastActor*>			 m_actors;
		GeneratorAsset*                  m_asset;
		GeneratorAsset::Vec3             m_localPos;
		float                            m_minRadius;
		float                            m_maxRadius;
		float                            m_compressiveDamage;
		ActorTestFunction                m_testFunction;

		std::vector<NvBlastActor*> m_resultActors;
	};

	void damageLeafSupportActorsParallelized
	(
		uint32_t assetCount,
		uint32_t minChunkCount,
		uint32_t damageCount,
		uint32_t threadCount,
		ActorTestFunction actorTestFunction,
		PostDamageTestFunction postDamageTestFunction
	)
	{
		const float relativeDamageRadius = 0.05f;
		const float compressiveDamage = 1.0f;

		srand(0);

		std::cout << "Asset # (out of " << assetCount << "): ";
		for (uint32_t assetNum = 0; assetNum < assetCount; ++assetNum)
		{
			std::cout << assetNum + 1 << ".. ";
			CubeAssetGenerator::Settings settings;
			settings.extents = GeneratorAsset::Vec3(1, 1, 1);
			CubeAssetGenerator::DepthInfo depthInfo;
			depthInfo.slicesPerAxis = GeneratorAsset::Vec3(1, 1, 1);
			depthInfo.flag = NvBlastChunkDesc::Flags::NoFlags;
			settings.depths.push_back(depthInfo);
			uint32_t chunkCount = 1;
			while (chunkCount < minChunkCount)
			{
				uint32_t chunkMul;
				do
				{
					depthInfo.slicesPerAxis = GeneratorAsset::Vec3((float)(1 + rand() % 4), (float)(1 + rand() % 4), (float)(1 + rand() % 4));
					chunkMul = (uint32_t)(depthInfo.slicesPerAxis.x * depthInfo.slicesPerAxis.y * depthInfo.slicesPerAxis.z);
				} while (chunkMul == 1);
				chunkCount *= chunkMul;
				settings.depths.push_back(depthInfo);
				settings.extents = settings.extents * depthInfo.slicesPerAxis;
			}
			settings.depths.back().flag = NvBlastChunkDesc::SupportFlag;	// Leaves are support

			// Make largest direction unit size
			settings.extents = settings.extents * (1.0f / std::max(settings.extents.x, std::max(settings.extents.y, settings.extents.z)));

			// Create asset
			GeneratorAsset testAsset;
			CubeAssetGenerator::generate(testAsset, settings);

			NvBlastAssetDesc desc;
			desc.chunkDescs = &testAsset.solverChunks[0];
			desc.chunkCount = (uint32_t)testAsset.solverChunks.size();
			desc.bondDescs = testAsset.solverBonds.data();
			desc.bondCount = (uint32_t)testAsset.solverBonds.size();

			std::vector<char> scratch;
			scratch.resize((size_t)NvBlastGetRequiredScratchForCreateAsset(&desc, messageLog));
			void* mem = alloc(NvBlastGetAssetMemorySize(&desc, messageLog));
			NvBlastAsset* asset = NvBlastCreateAsset(mem, &desc, scratch.data(), messageLog);
			EXPECT_TRUE(asset != nullptr);

			NvBlastActorDesc actorDesc;
			actorDesc.initialBondHealths = actorDesc.initialSupportChunkHealths = nullptr;
			actorDesc.uniformInitialBondHealth = actorDesc.uniformInitialLowerSupportChunkHealth = 1.0f;
			void* fmem = alloc(NvBlastAssetGetFamilyMemorySize(asset, messageLog));
			NvBlastFamily* family = NvBlastAssetCreateFamily(fmem, asset, nullptr);	// Using zeroingAlloc in case actorTest compares memory blocks
			scratch.resize((size_t)NvBlastFamilyGetRequiredScratchForCreateFirstActor(family, messageLog));
			NvBlastActor* actor = NvBlastFamilyCreateFirstActor(family, &actorDesc, scratch.data(), messageLog);
			EXPECT_TRUE(actor != nullptr);

			// Run parallelized damage through TaskDispatcher
			std::set<NvBlastActor*> resultActors;
			{
				uint32_t damageNum = 0;

				// create DamageActorTask and it to dispatcher helper function
				auto addDamageTaskFunction = [&](TaskDispatcher& dispatcher, NvBlastActor* actor)
				{
					GeneratorAsset::Vec3 localPos = settings.extents*GeneratorAsset::Vec3((float)rand() / RAND_MAX - 0.5f, (float)rand() / RAND_MAX - 0.5f, (float)rand() / RAND_MAX - 0.5f);
					auto newTask = std::unique_ptr<DamageActorTask>(new DamageActorTask(actor, &testAsset, localPos, relativeDamageRadius, relativeDamageRadius*1.2f, compressiveDamage, actorTestFunction));
					dispatcher.addTask(std::move(newTask));
				};

				// on task finished function for dispatcher (main thread)
				TaskDispatcher::OnTaskFinishedFunction onTaskFinishedFunction = [&](TaskDispatcher& dispatcher, std::unique_ptr<TaskDispatcher::Task> task) {
					const DamageActorTask* damageTask = static_cast<const DamageActorTask*>(task.get());
					const std::set<NvBlastActor*>& actors = damageTask->getResult();
					for (NvBlastActor* actor : actors)
					{
						if (damageNum >= damageCount)
						{
							resultActors.insert(actor);
						}
						else
						{
							damageNum++;
							addDamageTaskFunction(dispatcher, actor);
						}
					}
				};

				// create dispatcher, add first task and run 
				TaskDispatcher dispatcher(threadCount, onTaskFinishedFunction);
				addDamageTaskFunction(dispatcher, actor);
				dispatcher.process();
			}

			// Test fractured actor set
			if (postDamageTestFunction)
			{
				std::vector<NvBlastActor*> actorArray(resultActors.begin(), resultActors.end());
				postDamageTestFunction(actorArray, messageLog);
			}

			// Release remaining actors
			for (std::set<NvBlastActor*>::iterator k = resultActors.begin(); k != resultActors.end(); ++k)
			{
				NvBlastActorDeactivate(*k, messageLog);
			}
			resultActors.clear();

			const uint32_t actorCount = NvBlastFamilyGetActorCount(family, messageLog);
			EXPECT_TRUE(actorCount == 0);

			free(family);

			// Release asset data
			free(asset);
		}
		std::cout << "done.\n";
	}
};


// Specializations
typedef MultithreadingTest<NvBlastMessage::Error, 1> MultithreadingTestAllowWarnings;
typedef MultithreadingTest<NvBlastMessage::Error, 1> MultithreadingTestStrict;


TEST_F(MultithreadingTestStrict, MultithreadingTestDamageLeafSupportActorsTestVisibility)
{
	damageLeafSupportActorsParallelized(1, 1000, 50, 4, testActorVisibleChunks, nullptr);
}

TEST_F(MultithreadingTestStrict, MultithreadingTestDamageLeafSupportActors)
{
	damageLeafSupportActorsParallelized(1, 3000, 1000, 4, nullptr, nullptr);
}