common/end2end/GfxstreamEnd2EndVkTests.cpp

// Copyright (C) 2023 The Android Open Source Project
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include <log/log.h>

#include <atomic>
#include <thread>

#include "GfxstreamEnd2EndTestUtils.h"
#include "GfxstreamEnd2EndTests.h"

namespace gfxstream {
namespace tests {
namespace {

using namespace std::chrono_literals;
using testing::Eq;
using testing::Ge;
using testing::IsEmpty;
using testing::IsNull;
using testing::Ne;
using testing::Not;
using testing::NotNull;

template <typename DurationType>
constexpr uint64_t AsVkTimeout(DurationType duration) {
    return static_cast<uint64_t>(std::chrono::duration_cast<std::chrono::nanoseconds>(duration).count());
}

class GfxstreamEnd2EndVkTest : public GfxstreamEnd2EndTest {
   protected:
    // Gfxstream uses a vkQueueSubmit() to signal the VkFence and VkSemaphore used
    // in vkAcquireImageANDROID() calls. The guest is not aware of this and may try
    // to vkDestroyFence() and vkDestroySemaphore() (because the VkImage, VkFence,
    // and VkSemaphore may have been unused from the guest point of view) while the
    // host's command buffer is running. Gfxstream needs to ensure that it performs
    // the necessary tracking to not delete the VkFence and VkSemaphore while they
    // are in use on the host.
    void DoAcquireImageAndroidWithSync(bool withFence, bool withSemaphore) {
        auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
            VK_ASSERT(SetUpTypicalVkTestEnvironment());

        const uint32_t width = 32;
        const uint32_t height = 32;
        auto ahb = GL_ASSERT(ScopedAHardwareBuffer::Allocate(*mGralloc, width, height,
                                                             GFXSTREAM_AHB_FORMAT_R8G8B8A8_UNORM));

        const VkNativeBufferANDROID imageNativeBufferInfo = {
            .sType = VK_STRUCTURE_TYPE_NATIVE_BUFFER_ANDROID,
            .handle = mGralloc->getNativeHandle(ahb),
        };

        auto vkAcquireImageANDROID =
            PFN_vkAcquireImageANDROID(device->getProcAddr("vkAcquireImageANDROID"));
        ASSERT_THAT(vkAcquireImageANDROID, NotNull());

        const vkhpp::ImageCreateInfo imageCreateInfo = {
            .pNext = &imageNativeBufferInfo,
            .imageType = vkhpp::ImageType::e2D,
            .extent.width = width,
            .extent.height = height,
            .extent.depth = 1,
            .mipLevels = 1,
            .arrayLayers = 1,
            .format = vkhpp::Format::eR8G8B8A8Unorm,
            .tiling = vkhpp::ImageTiling::eOptimal,
            .initialLayout = vkhpp::ImageLayout::eUndefined,
            .usage = vkhpp::ImageUsageFlagBits::eSampled | vkhpp::ImageUsageFlagBits::eTransferDst |
                     vkhpp::ImageUsageFlagBits::eTransferSrc,
            .sharingMode = vkhpp::SharingMode::eExclusive,
            .samples = vkhpp::SampleCountFlagBits::e1,
        };
        auto image = device->createImageUnique(imageCreateInfo).value;

        vkhpp::MemoryRequirements imageMemoryRequirements{};
        device->getImageMemoryRequirements(*image, &imageMemoryRequirements);

        const uint32_t imageMemoryIndex = utils::getMemoryType(
            physicalDevice, imageMemoryRequirements, vkhpp::MemoryPropertyFlagBits::eDeviceLocal);
        ASSERT_THAT(imageMemoryIndex, Not(Eq(-1)));

        const vkhpp::MemoryAllocateInfo imageMemoryAllocateInfo = {
            .allocationSize = imageMemoryRequirements.size,
            .memoryTypeIndex = imageMemoryIndex,
        };

        auto imageMemory = device->allocateMemoryUnique(imageMemoryAllocateInfo).value;
        ASSERT_THAT(imageMemory, IsValidHandle());
        ASSERT_THAT(device->bindImageMemory(*image, *imageMemory, 0), IsVkSuccess());

        vkhpp::UniqueFence fence;
        if (withFence) {
            fence = device->createFenceUnique(vkhpp::FenceCreateInfo()).value;
        }

        vkhpp::UniqueSemaphore semaphore;
        if (withSemaphore) {
            semaphore = device->createSemaphoreUnique(vkhpp::SemaphoreCreateInfo()).value;
        }

        auto result = vkAcquireImageANDROID(*device, *image, -1, *semaphore, *fence);
        ASSERT_THAT(result, Eq(VK_SUCCESS));

        if (withFence) {
            fence.reset();
        }
        if (withSemaphore) {
            semaphore.reset();
        }
    }
};

TEST_P(GfxstreamEnd2EndVkTest, Basic) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());
}

TEST_P(GfxstreamEnd2EndVkTest, ImportAHB) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    const uint32_t width = 32;
    const uint32_t height = 32;
    auto ahb = GL_ASSERT(ScopedAHardwareBuffer::Allocate(*mGralloc, width, height,
                                                         GFXSTREAM_AHB_FORMAT_R8G8B8A8_UNORM));

    const VkNativeBufferANDROID imageNativeBufferInfo = {
        .sType = VK_STRUCTURE_TYPE_NATIVE_BUFFER_ANDROID,
        .handle = mGralloc->getNativeHandle(ahb),
    };

    auto vkQueueSignalReleaseImageANDROID = PFN_vkQueueSignalReleaseImageANDROID(
        device->getProcAddr("vkQueueSignalReleaseImageANDROID"));
    ASSERT_THAT(vkQueueSignalReleaseImageANDROID, NotNull());

    const vkhpp::ImageCreateInfo imageCreateInfo = {
        .pNext = &imageNativeBufferInfo,
        .imageType = vkhpp::ImageType::e2D,
        .extent.width = width,
        .extent.height = height,
        .extent.depth = 1,
        .mipLevels = 1,
        .arrayLayers = 1,
        .format = vkhpp::Format::eR8G8B8A8Unorm,
        .tiling = vkhpp::ImageTiling::eOptimal,
        .initialLayout = vkhpp::ImageLayout::eUndefined,
        .usage = vkhpp::ImageUsageFlagBits::eSampled |
                 vkhpp::ImageUsageFlagBits::eTransferDst |
                 vkhpp::ImageUsageFlagBits::eTransferSrc,
        .sharingMode = vkhpp::SharingMode::eExclusive,
        .samples = vkhpp::SampleCountFlagBits::e1,
    };
    auto image = device->createImageUnique(imageCreateInfo).value;

    vkhpp::MemoryRequirements imageMemoryRequirements{};
    device->getImageMemoryRequirements(*image, &imageMemoryRequirements);

    const uint32_t imageMemoryIndex = utils::getMemoryType(
        physicalDevice, imageMemoryRequirements, vkhpp::MemoryPropertyFlagBits::eDeviceLocal);
    ASSERT_THAT(imageMemoryIndex, Not(Eq(-1)));

    const vkhpp::MemoryAllocateInfo imageMemoryAllocateInfo = {
        .allocationSize = imageMemoryRequirements.size,
        .memoryTypeIndex = imageMemoryIndex,
    };

    auto imageMemory = device->allocateMemoryUnique(imageMemoryAllocateInfo).value;
    ASSERT_THAT(imageMemory, IsValidHandle());
    ASSERT_THAT(device->bindImageMemory(*image, *imageMemory, 0), IsVkSuccess());

    const vkhpp::BufferCreateInfo bufferCreateInfo = {
        .size = static_cast<VkDeviceSize>(12 * 1024 * 1024),
        .usage = vkhpp::BufferUsageFlagBits::eTransferDst |
                 vkhpp::BufferUsageFlagBits::eTransferSrc,
        .sharingMode = vkhpp::SharingMode::eExclusive,
    };
    auto stagingBuffer = device->createBufferUnique(bufferCreateInfo).value;
    ASSERT_THAT(stagingBuffer, IsValidHandle());

    vkhpp::MemoryRequirements stagingBufferMemoryRequirements{};
    device->getBufferMemoryRequirements(*stagingBuffer, &stagingBufferMemoryRequirements);

    const auto stagingBufferMemoryType = utils::getMemoryType(
        physicalDevice, stagingBufferMemoryRequirements,
        vkhpp::MemoryPropertyFlagBits::eHostVisible | vkhpp::MemoryPropertyFlagBits::eHostCoherent);

    const vkhpp::MemoryAllocateInfo stagingBufferMemoryAllocateInfo = {
        .allocationSize = stagingBufferMemoryRequirements.size,
        .memoryTypeIndex = stagingBufferMemoryType,
    };
    auto stagingBufferMemory = device->allocateMemoryUnique(stagingBufferMemoryAllocateInfo).value;
    ASSERT_THAT(stagingBufferMemory, IsValidHandle());
    ASSERT_THAT(device->bindBufferMemory(*stagingBuffer, *stagingBufferMemory, 0), IsVkSuccess());

    const vkhpp::CommandPoolCreateInfo commandPoolCreateInfo = {
        .queueFamilyIndex = queueFamilyIndex,
    };

    auto commandPool = device->createCommandPoolUnique(commandPoolCreateInfo).value;
    ASSERT_THAT(stagingBufferMemory, IsValidHandle());

    const vkhpp::CommandBufferAllocateInfo commandBufferAllocateInfo = {
        .level = vkhpp::CommandBufferLevel::ePrimary,
        .commandPool = *commandPool,
        .commandBufferCount = 1,
    };
    auto commandBuffers = device->allocateCommandBuffersUnique(commandBufferAllocateInfo).value;
    ASSERT_THAT(commandBuffers, Not(IsEmpty()));
    auto commandBuffer = std::move(commandBuffers[0]);
    ASSERT_THAT(commandBuffer, IsValidHandle());

    const vkhpp::CommandBufferBeginInfo commandBufferBeginInfo = {
        .flags = vkhpp::CommandBufferUsageFlagBits::eOneTimeSubmit,
    };
    commandBuffer->begin(commandBufferBeginInfo);
    commandBuffer->end();

    std::vector<vkhpp::CommandBuffer> commandBufferHandles;
    commandBufferHandles.push_back(*commandBuffer);

    auto transferFence = device->createFenceUnique(vkhpp::FenceCreateInfo()).value;
    ASSERT_THAT(commandBuffer, IsValidHandle());

    const vkhpp::SubmitInfo submitInfo = {
        .commandBufferCount = static_cast<uint32_t>(commandBufferHandles.size()),
        .pCommandBuffers = commandBufferHandles.data(),
    };
    queue.submit(submitInfo, *transferFence);

    auto waitResult = device->waitForFences(*transferFence, VK_TRUE, AsVkTimeout(3s));
    ASSERT_THAT(waitResult, IsVkSuccess());

    int fence;

    auto result = vkQueueSignalReleaseImageANDROID(queue, 0, nullptr, *image, &fence);
    ASSERT_THAT(result, Eq(VK_SUCCESS));
    ASSERT_THAT(fence, Not(Eq(-1)));

    ASSERT_THAT(mSync->wait(fence, 3000), Eq(0));
}

TEST_P(GfxstreamEnd2EndVkTest, DeferredImportAHB) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    const uint32_t width = 32;
    const uint32_t height = 32;
    auto ahb = GL_ASSERT(ScopedAHardwareBuffer::Allocate(*mGralloc, width, height,
                                                         GFXSTREAM_AHB_FORMAT_R8G8B8A8_UNORM));

    auto vkQueueSignalReleaseImageANDROID = PFN_vkQueueSignalReleaseImageANDROID(
        device->getProcAddr("vkQueueSignalReleaseImageANDROID"));
    ASSERT_THAT(vkQueueSignalReleaseImageANDROID, NotNull());

    const vkhpp::ImageCreateInfo imageCreateInfo = {
        .pNext = nullptr,
        .imageType = vkhpp::ImageType::e2D,
        .extent.width = width,
        .extent.height = height,
        .extent.depth = 1,
        .mipLevels = 1,
        .arrayLayers = 1,
        .format = vkhpp::Format::eR8G8B8A8Unorm,
        .tiling = vkhpp::ImageTiling::eOptimal,
        .initialLayout = vkhpp::ImageLayout::eUndefined,
        .usage = vkhpp::ImageUsageFlagBits::eSampled |
                 vkhpp::ImageUsageFlagBits::eTransferDst |
                 vkhpp::ImageUsageFlagBits::eTransferSrc,
        .sharingMode = vkhpp::SharingMode::eExclusive,
        .samples = vkhpp::SampleCountFlagBits::e1,
    };
    auto image = device->createImageUnique(imageCreateInfo).value;

    // NOTE: Binding the VkImage to the AHB happens after the VkImage is created.
    const VkNativeBufferANDROID imageNativeBufferInfo = {
        .sType = VK_STRUCTURE_TYPE_NATIVE_BUFFER_ANDROID,
        .handle = mGralloc->getNativeHandle(ahb),
    };

    const vkhpp::BindImageMemoryInfo imageBindMemoryInfo = {
        .pNext = &imageNativeBufferInfo,
        .image = *image,
        .memory = VK_NULL_HANDLE,
        .memoryOffset = 0,
    };
    ASSERT_THAT(device->bindImageMemory2({imageBindMemoryInfo}), IsVkSuccess());

    std::vector<vkhpp::Semaphore> semaphores;
    int fence;

    auto result = vkQueueSignalReleaseImageANDROID(queue, 0, nullptr, *image, &fence);
    ASSERT_THAT(result, Eq(VK_SUCCESS));
    ASSERT_THAT(fence, Not(Eq(-1)));

    ASSERT_THAT(mSync->wait(fence, 3000), Eq(0));
}

TEST_P(GfxstreamEnd2EndVkTest, BlobAHBIsNotMapable) {
    if (GetParam().with_gl) {
        GTEST_SKIP()
            << "Skipping test, data buffers are currently only supported in Vulkan only mode.";
    }
    if (GetParam().with_features.count("VulkanUseDedicatedAhbMemoryType") == 0) {
        GTEST_SKIP()
            << "Skipping test, AHB test only makes sense with VulkanUseDedicatedAhbMemoryType.";
    }

    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    const uint32_t width = 32;
    const uint32_t height = 1;
    auto ahb = GL_ASSERT(
        ScopedAHardwareBuffer::Allocate(*mGralloc, width, height, GFXSTREAM_AHB_FORMAT_BLOB));

    const vkhpp::ExternalMemoryBufferCreateInfo externalMemoryBufferCreateInfo = {
        .handleTypes = vkhpp::ExternalMemoryHandleTypeFlagBits::eAndroidHardwareBufferANDROID,
    };
    const vkhpp::BufferCreateInfo bufferCreateInfo = {
        .pNext = &externalMemoryBufferCreateInfo,
        .size = width,
        .usage = vkhpp::BufferUsageFlagBits::eTransferDst |
                 vkhpp::BufferUsageFlagBits::eTransferSrc |
                 vkhpp::BufferUsageFlagBits::eVertexBuffer,
        .sharingMode = vkhpp::SharingMode::eExclusive,
    };
    auto buffer = device->createBufferUnique(bufferCreateInfo).value;
    ASSERT_THAT(buffer, IsValidHandle());

    auto vkGetAndroidHardwareBufferPropertiesANDROID =
        reinterpret_cast<PFN_vkGetAndroidHardwareBufferPropertiesANDROID>(
            device->getProcAddr("vkGetAndroidHardwareBufferPropertiesANDROID"));
    ASSERT_THAT(vkGetAndroidHardwareBufferPropertiesANDROID, NotNull());

    VkAndroidHardwareBufferPropertiesANDROID bufferProperties = {
        .sType = VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_PROPERTIES_ANDROID,
        .pNext = nullptr,
    };
    ASSERT_THAT(vkGetAndroidHardwareBufferPropertiesANDROID(*device, ahb, &bufferProperties),
                Eq(VK_SUCCESS));

    const vkhpp::MemoryRequirements bufferMemoryRequirements{
        .size = bufferProperties.allocationSize,
        .alignment = 0,
        .memoryTypeBits = bufferProperties.memoryTypeBits,
    };

    const auto memoryProperties = physicalDevice.getMemoryProperties();
    for (uint32_t i = 0; i < memoryProperties.memoryTypeCount; i++) {
        if (!(bufferMemoryRequirements.memoryTypeBits & (1 << i))) {
            continue;
        }

        const auto memoryPropertyFlags = memoryProperties.memoryTypes[i].propertyFlags;
        EXPECT_THAT(memoryPropertyFlags & vkhpp::MemoryPropertyFlagBits::eHostVisible,
                    Ne(vkhpp::MemoryPropertyFlagBits::eHostVisible));
    }

    const auto bufferMemoryType = utils::getMemoryType(physicalDevice, bufferMemoryRequirements,
                                                       vkhpp::MemoryPropertyFlagBits::eDeviceLocal);
    ASSERT_THAT(bufferMemoryType, Ne(-1));

    const vkhpp::ImportAndroidHardwareBufferInfoANDROID importHardwareBufferInfo = {
        .buffer = ahb,
    };
    const vkhpp::MemoryAllocateInfo bufferMemoryAllocateInfo = {
        .pNext = &importHardwareBufferInfo,
        .allocationSize = bufferMemoryRequirements.size,
        .memoryTypeIndex = bufferMemoryType,
    };
    auto bufferMemory = device->allocateMemoryUnique(bufferMemoryAllocateInfo).value;
    ASSERT_THAT(bufferMemory, IsValidHandle());

    ASSERT_THAT(device->bindBufferMemory(*buffer, *bufferMemory, 0), IsVkSuccess());
}

TEST_P(GfxstreamEnd2EndVkTest, HostMemory) {
    static constexpr const vkhpp::DeviceSize kSize = 16 * 1024;

    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    uint32_t hostMemoryTypeIndex = -1;
    const auto memoryProperties = physicalDevice.getMemoryProperties();
    for (uint32_t i = 0; i < memoryProperties.memoryTypeCount; i++) {
        const vkhpp::MemoryType& memoryType = memoryProperties.memoryTypes[i];
        if (memoryType.propertyFlags & vkhpp::MemoryPropertyFlagBits::eHostVisible) {
            hostMemoryTypeIndex = i;
        }
    }
    if (hostMemoryTypeIndex == -1) {
        GTEST_SKIP() << "Skipping test due to no host visible memory type.";
        return;
    }

    const vkhpp::MemoryAllocateInfo memoryAllocateInfo = {
        .allocationSize = kSize,
        .memoryTypeIndex = hostMemoryTypeIndex,
    };
    auto memory = device->allocateMemoryUnique(memoryAllocateInfo).value;
    ASSERT_THAT(memory, IsValidHandle());

    void* mapped = nullptr;

    auto mapResult = device->mapMemory(*memory, 0, VK_WHOLE_SIZE, vkhpp::MemoryMapFlags{}, &mapped);
    ASSERT_THAT(mapResult, IsVkSuccess());
    ASSERT_THAT(mapped, NotNull());

    auto* bytes = reinterpret_cast<uint8_t*>(mapped);
    std::memset(bytes, 0xFF, kSize);

    const vkhpp::MappedMemoryRange range = {
        .memory = *memory,
        .offset = 0,
        .size = kSize,
    };
    device->flushMappedMemoryRanges({range});
    device->invalidateMappedMemoryRanges({range});

    for (uint32_t i = 0; i < kSize; ++i) {
        EXPECT_THAT(bytes[i], Eq(0xFF));
    }
}

TEST_P(GfxstreamEnd2EndVkTest, GetPhysicalDeviceProperties2) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    auto props1 = physicalDevice.getProperties();
    auto props2 = physicalDevice.getProperties2();

    EXPECT_THAT(props1.vendorID, Eq(props2.properties.vendorID));
    EXPECT_THAT(props1.deviceID, Eq(props2.properties.deviceID));
}

TEST_P(GfxstreamEnd2EndVkTest, GetPhysicalDeviceFeatures2KHR) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    auto features1 = physicalDevice.getFeatures();
    auto features2 = physicalDevice.getFeatures2();
    EXPECT_THAT(features1.robustBufferAccess, Eq(features2.features.robustBufferAccess));
}

TEST_P(GfxstreamEnd2EndVkTest, GetPhysicalDeviceImageFormatProperties2KHR) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    const vkhpp::PhysicalDeviceImageFormatInfo2 imageFormatInfo = {
        .format = vkhpp::Format::eR8G8B8A8Unorm,
        .type = vkhpp::ImageType::e2D,
        .tiling = vkhpp::ImageTiling::eOptimal,
        .usage = vkhpp::ImageUsageFlagBits::eSampled,
    };
    const auto properties = VK_ASSERT_RV(physicalDevice.getImageFormatProperties2(imageFormatInfo));
    EXPECT_THAT(properties.imageFormatProperties.maxExtent.width, Ge(1));
    EXPECT_THAT(properties.imageFormatProperties.maxExtent.height, Ge(1));
    EXPECT_THAT(properties.imageFormatProperties.maxExtent.depth, Ge(1));
}

template <typename VkhppUniqueHandleType,
          typename VkhppHandleType = typename VkhppUniqueHandleType::element_type>
std::vector<VkhppHandleType> AsHandles(const std::vector<VkhppUniqueHandleType>& elements) {
    std::vector<VkhppHandleType> ret;
    ret.reserve(elements.size());
    for (const auto& e : elements) {
        ret.push_back(*e);
    }
    return ret;
}

struct DescriptorBundle {
    vkhpp::UniqueDescriptorPool descriptorPool;
    vkhpp::UniqueDescriptorSetLayout descriptorSetLayout;
    std::vector<vkhpp::UniqueDescriptorSet> descriptorSets;
};

vkhpp::Result ReallocateDescriptorBundleSets(vkhpp::Device device, uint32_t count, DescriptorBundle* bundle) {
    if (!bundle->descriptorSetLayout) {
        ALOGE("Invalid descriptor set layout.");
        return vkhpp::Result::eErrorUnknown;
    }

    const std::vector<vkhpp::DescriptorSetLayout> descriptorSetLayouts(count, *bundle->descriptorSetLayout);
    const vkhpp::DescriptorSetAllocateInfo descriptorSetAllocateInfo = {
        .descriptorPool = *bundle->descriptorPool,
        .descriptorSetCount = count,
        .pSetLayouts = descriptorSetLayouts.data(),
    };
    auto descriptorSets = VK_TRY_RV(device.allocateDescriptorSetsUnique(descriptorSetAllocateInfo));
    bundle->descriptorSets = std::move(descriptorSets);
    return vkhpp::Result::eSuccess;
}

VkExpected<DescriptorBundle> AllocateDescriptorBundle(vkhpp::Device device, uint32_t count) {
    const vkhpp::DescriptorPoolSize descriptorPoolSize = {
        .type = vkhpp::DescriptorType::eUniformBuffer,
        .descriptorCount = 1 * count,
    };
    const vkhpp::DescriptorPoolCreateInfo descriptorPoolCreateInfo = {
        .flags = vkhpp::DescriptorPoolCreateFlagBits::eFreeDescriptorSet,
        .maxSets = count,
        .poolSizeCount = 1,
        .pPoolSizes = &descriptorPoolSize,
    };
    auto descriptorPool = VK_EXPECT_RV(device.createDescriptorPoolUnique(descriptorPoolCreateInfo));

    const vkhpp::DescriptorSetLayoutBinding descriptorSetBinding = {
        .binding = 0,
        .descriptorType = vkhpp::DescriptorType::eUniformBuffer,
        .descriptorCount = 1,
        .stageFlags = vkhpp::ShaderStageFlagBits::eVertex,
    };
    const vkhpp::DescriptorSetLayoutCreateInfo descriptorSetLayoutInfo = {
        .bindingCount = 1,
        .pBindings = &descriptorSetBinding,
    };
    auto descriptorSetLayout = VK_EXPECT_RV(device.createDescriptorSetLayoutUnique(descriptorSetLayoutInfo));

    DescriptorBundle bundle = {
        .descriptorPool = std::move(descriptorPool),
        .descriptorSetLayout = std::move(descriptorSetLayout),
    };
    VK_EXPECT_RESULT(ReallocateDescriptorBundleSets(device, count, &bundle));
    return std::move(bundle);
}

// Tests creating a bunch of descriptor sets and freeing them via vkFreeDescriptorSet.
// 1. Via vkFreeDescriptorSet directly
// 2. Via vkResetDescriptorPool
// 3. Via vkDestroyDescriptorPool
// 4. Via vkResetDescriptorPool and double frees in vkFreeDescriptorSet
// 5. Via vkResetDescriptorPool and double frees in vkFreeDescriptorSet
// 4. Via vkResetDescriptorPool, creating more, and freeing vai vkFreeDescriptorSet
// (because vkFree* APIs are expected to never fail)
// https://github.com/KhronosGroup/Vulkan-Docs/issues/1070
TEST_P(GfxstreamEnd2EndVkTest, DescriptorSetAllocFree) {
    constexpr const uint32_t kNumSets = 4;

    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    auto bundle = VK_ASSERT(AllocateDescriptorBundle(*device, kNumSets));

    auto descriptorSetHandles = AsHandles(bundle.descriptorSets);
    EXPECT_THAT(device->freeDescriptorSets(*bundle.descriptorPool, kNumSets, descriptorSetHandles.data()), IsVkSuccess());

    // The double free should also work
    EXPECT_THAT(device->freeDescriptorSets(*bundle.descriptorPool, kNumSets, descriptorSetHandles.data()), IsVkSuccess());

    // Alloc/free again should also work
    ASSERT_THAT(ReallocateDescriptorBundleSets(*device, kNumSets, &bundle), IsVkSuccess());

    descriptorSetHandles = AsHandles(bundle.descriptorSets);
    EXPECT_THAT(device->freeDescriptorSets(*bundle.descriptorPool, kNumSets, descriptorSetHandles.data()), IsVkSuccess());
}

TEST_P(GfxstreamEnd2EndVkTest, DescriptorSetAllocFreeReset) {
    constexpr const uint32_t kNumSets = 4;

    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    auto bundle = VK_ASSERT(AllocateDescriptorBundle(*device, kNumSets));

    device->resetDescriptorPool(*bundle.descriptorPool);

    // The double free should also work
    auto descriptorSetHandles = AsHandles(bundle.descriptorSets);
    EXPECT_THAT(device->freeDescriptorSets(*bundle.descriptorPool, kNumSets, descriptorSetHandles.data()), IsVkSuccess());

    // Alloc/reset/free again should also work
    ASSERT_THAT(ReallocateDescriptorBundleSets(*device, kNumSets, &bundle), IsVkSuccess());

    device->resetDescriptorPool(*bundle.descriptorPool);

    descriptorSetHandles = AsHandles(bundle.descriptorSets);
    EXPECT_THAT(device->freeDescriptorSets(*bundle.descriptorPool, kNumSets, descriptorSetHandles.data()), IsVkSuccess());
}

TEST_P(GfxstreamEnd2EndVkTest, DISABLED_DescriptorSetAllocFreeDestroy) {
    constexpr const uint32_t kNumSets = 4;

    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    auto bundle = VK_ASSERT(AllocateDescriptorBundle(*device, kNumSets));

    device->destroyDescriptorPool(*bundle.descriptorPool);

    // The double free should also work
    auto descriptorSetHandles = AsHandles(bundle.descriptorSets);
    EXPECT_THAT(device->freeDescriptorSets(*bundle.descriptorPool, kNumSets, descriptorSetHandles.data()), IsVkSuccess());
}

TEST_P(GfxstreamEnd2EndVkTest, MultiThreadedShutdown) {
    constexpr const int kNumIterations = 20;
    for (int i = 0; i < kNumIterations; i++) {
        auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
                VK_ASSERT(SetUpTypicalVkTestEnvironment());

        const vkhpp::BufferCreateInfo bufferCreateInfo = {
            .size = 1024,
            .usage = vkhpp::BufferUsageFlagBits::eTransferSrc,
        };

        // TODO: switch to std::barrier with arrive_and_wait().
        std::atomic_int threadsReady{0};
        std::vector<std::thread> threads;

        constexpr const int kNumThreads = 5;
        for (int t = 0; t < kNumThreads; t++) {
            threads.emplace_back([&, this](){
                // Perform some work to ensure host RenderThread started.
                auto buffer1 = device->createBufferUnique(bufferCreateInfo).value;

                ++threadsReady;
                while (threadsReady.load() != kNumThreads) {}

                // Sleep a little which is hopefully enough time to potentially get
                // the corresponding host ASG RenderThreads to go sleep waiting for
                // a WAKEUP via a GFXSTREAM_CONTEXT_PING.
                std::this_thread::sleep_for(std::chrono::milliseconds(100));

                auto buffer2 = device->createBufferUnique(bufferCreateInfo).value;

                // 2 vkDestroyBuffer() calls happen here with the destruction of `buffer1`
                // and `buffer2`. vkDestroy*() calls are async (return `void`) and the
                // guest thread continues execution without waiting for the command to
                // complete on the host.
                //
                // The guest ASG and corresponding virtio gpu resource will also be
                // destructed here as a part of the thread_local HostConnection being
                // destructed.
                //
                // Note: Vulkan commands are given a sequence number in order to ensure that
                // commands from multi-threaded guest Vulkan apps are executed in order on the
                // host. Gfxstream's host Vulkan decoders will spin loop waiting for their turn to
                // process their next command.
                //
                // With all of the above, a deadlock would previouly occur with the following
                // sequence:
                //
                // T1: Host-RenderThread-1: <sleeping waiting for wakeup>
                //
                // T2: Host-RenderThread-2: <sleeping waiting for wakeup>
                //
                // T3: Guest-Thread-1: vkDestroyBuffer() called,
                //                     VkEncoder grabs sequence-number-10,
                //                     writes sequence-number-10 into ASG-1 via resource-1
                //
                // T4: Guest-Thread-2: vkDestroyBuffer() called,
                //                     VkEncoder grabs sequence-number-11,
                //                     writes into ASG-2 via resource-2
                //
                // T5: Guest-Thread-2: ASG-2 sends a VIRTIO_GPU_CMD_SUBMIT_3D with
                //                     GFXSTREAM_CONTEXT_PING on ASG-resource-2
                //
                // T6: Guest-Thread-2: guest thread finishes,
                //                     ASG-2 destructor destroys the virtio-gpu resource used,
                //                     destruction sends VIRTIO_GPU_CMD_RESOURCE_UNREF on
                //                     resource-2
                //
                // T7: Guest-Thread-1: ASG-1 sends VIRTIO_GPU_CMD_SUBMIT_3D with
                //                     GFXSTREAM_CONTEXT_PING on ASG-resource-1
                //
                // T8: Host-Virtio-Gpu-Thread: performs VIRTIO_GPU_CMD_SUBMIT_3D from T5,
                //                             pings ASG-2 which wakes up Host-RenderThread-2
                //
                // T9: Host-RenderThread-2: woken from T8,
                //                          reads sequence-number-11 from ASG-2,
                //                          spin looping waiting for sequence-number-10 to execute
                //
                // T10: Host-Virtio-Gpu-Thread: performs VIRTIO_GPU_CMD_RESOURCE_UNREF for
                //                              resource-2 from T6,
                //                              resource-2 is used by ASG-2 / Host-RenderThread-2
                //                              waits for Host-RenderThread-2 to finish
                //
                // DEADLOCKED HERE:
                //
                //   *  Host-Virtio-GpuThread is waiting for Host-RenderThread-2 to finish before
                //      it can finish destroying resource-2
                //
                //   *  Host-RenderThread-2 is waiting for Host-RenderThread-1 to execute
                //      sequence-number-10
                //
                //   *  Host-RenderThread-1 is asleep waiting for a GFXSTREAM_CONTEXT_PING
                //      from Host-Virtio-GpuThread
            });
        }

        for (auto& thread : threads) {
            thread.join();
        }
    }
}

TEST_P(GfxstreamEnd2EndVkTest, AcquireImageAndroidWithFence) {
    DoAcquireImageAndroidWithSync(/*withFence=*/true, /*withSemaphore=*/false);
}

TEST_P(GfxstreamEnd2EndVkTest, AcquireImageAndroidWithSemaphore) {
    DoAcquireImageAndroidWithSync(/*withFence=*/false, /*withSemaphore=*/true);
}

TEST_P(GfxstreamEnd2EndVkTest, AcquireImageAndroidWithFenceAndSemaphore) {
    DoAcquireImageAndroidWithSync(/*withFence=*/true, /*withSemaphore=*/true);
}

VKAPI_ATTR void VKAPI_CALL MemoryReportCallback(const VkDeviceMemoryReportCallbackDataEXT*, void*) {
    // Unused
}

TEST_P(GfxstreamEnd2EndVkTest, DeviceMemoryReport) {
    int userdata = 1;
    vkhpp::DeviceDeviceMemoryReportCreateInfoEXT deviceDeviceMemoryReportInfo = {
        .pfnUserCallback = &MemoryReportCallback,
        .pUserData = &userdata,
    };

    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment({
            .deviceExtensions = {{
                VK_EXT_DEVICE_MEMORY_REPORT_EXTENSION_NAME,
            }},
            .deviceCreateInfoPNext = &deviceDeviceMemoryReportInfo,
        }));

    const vkhpp::MemoryAllocateInfo memoryAllocateInfo = {
        .allocationSize = 1024,
        .memoryTypeIndex = 0,
    };
    auto memory = device->allocateMemoryUnique(memoryAllocateInfo).value;
    ASSERT_THAT(memory, IsValidHandle());
}

TEST_P(GfxstreamEnd2EndVkTest, DescriptorUpdateTemplateWithWrapping) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    const vkhpp::BufferCreateInfo bufferCreateInfo = {
        .size = 1024,
        .usage = vkhpp::BufferUsageFlagBits::eUniformBuffer,
    };
    auto buffer = VK_ASSERT_RV(device->createBufferUnique(bufferCreateInfo));

    const std::vector<VkDescriptorBufferInfo> descriptorInfo = {
        VkDescriptorBufferInfo{
            .buffer = *buffer,
            .offset = 0,
            .range = 1024,
        },
        VkDescriptorBufferInfo{
            .buffer = *buffer,
            .offset = 0,
            .range = 1024,
        },
        VkDescriptorBufferInfo{
            .buffer = *buffer,
            .offset = 0,
            .range = 1024,
        },
        VkDescriptorBufferInfo{
            .buffer = *buffer,
            .offset = 0,
            .range = 1024,
        },
    };

    const std::vector<vkhpp::DescriptorPoolSize> descriptorPoolSizes = {
        {
            .type = vkhpp::DescriptorType::eUniformBuffer,
            .descriptorCount = 4,
        },
    };
    const vkhpp::DescriptorPoolCreateInfo descriptorPoolCreateInfo = {
        .flags = vkhpp::DescriptorPoolCreateFlagBits::eFreeDescriptorSet,
        .maxSets = 1,
        .poolSizeCount = static_cast<uint32_t>(descriptorPoolSizes.size()),
        .pPoolSizes = descriptorPoolSizes.data(),
    };
    auto descriptorPool =
        VK_ASSERT_RV(device->createDescriptorPoolUnique(descriptorPoolCreateInfo));

    const std::vector<vkhpp::DescriptorSetLayoutBinding> descriptorSetBindings = {
        {
            .binding = 0,
            .descriptorType = vkhpp::DescriptorType::eUniformBuffer,
            .descriptorCount = 1,
            .stageFlags = vkhpp::ShaderStageFlagBits::eVertex,
        },
        {
            .binding = 1,
            .descriptorType = vkhpp::DescriptorType::eUniformBuffer,
            .descriptorCount = 1,
            .stageFlags = vkhpp::ShaderStageFlagBits::eVertex,
        },
        {
            .binding = 2,
            .descriptorType = vkhpp::DescriptorType::eUniformBuffer,
            .descriptorCount = 1,
            .stageFlags = vkhpp::ShaderStageFlagBits::eVertex,
        },
        {
            .binding = 3,
            .descriptorType = vkhpp::DescriptorType::eUniformBuffer,
            .descriptorCount = 1,
            .stageFlags = vkhpp::ShaderStageFlagBits::eVertex,
        },
    };
    const vkhpp::DescriptorSetLayoutCreateInfo descriptorSetLayoutInfo = {
        .bindingCount = static_cast<uint32_t>(descriptorSetBindings.size()),
        .pBindings = descriptorSetBindings.data(),
    };
    auto descriptorSetLayout =
        VK_ASSERT_RV(device->createDescriptorSetLayoutUnique(descriptorSetLayoutInfo));

    const std::vector<vkhpp::DescriptorSetLayout> descriptorSetLayouts = {*descriptorSetLayout};
    const vkhpp::DescriptorSetAllocateInfo descriptorSetAllocateInfo = {
        .descriptorPool = *descriptorPool,
        .descriptorSetCount = static_cast<uint32_t>(descriptorSetLayouts.size()),
        .pSetLayouts = descriptorSetLayouts.data(),
    };
    auto descriptorSets =
        VK_ASSERT_RV(device->allocateDescriptorSetsUnique(descriptorSetAllocateInfo));
    auto descriptorSet = std::move(descriptorSets[0]);

    const vkhpp::PipelineLayoutCreateInfo pipelineLayoutCreateInfo = {
        .setLayoutCount = static_cast<uint32_t>(descriptorSetLayouts.size()),
        .pSetLayouts = descriptorSetLayouts.data(),
    };
    auto pipelineLayout =
        VK_ASSERT_RV(device->createPipelineLayoutUnique(pipelineLayoutCreateInfo));

    const std::vector<vkhpp::DescriptorUpdateTemplateEntry> descriptorUpdateEntries = {
        {
            .dstBinding = 0,
            .dstArrayElement = 0,
            .descriptorCount = 4,
            .descriptorType = vkhpp::DescriptorType::eUniformBuffer,
            .offset = 0,
            .stride = sizeof(VkDescriptorBufferInfo),
        },
    };
    const vkhpp::DescriptorUpdateTemplateCreateInfo descriptorUpdateTemplateCreateInfo = {
        .descriptorUpdateEntryCount = static_cast<uint32_t>(descriptorUpdateEntries.size()),
        .pDescriptorUpdateEntries = descriptorUpdateEntries.data(),
        .descriptorSetLayout = *descriptorSetLayout,
        .pipelineBindPoint = vkhpp::PipelineBindPoint::eGraphics,
        .pipelineLayout = *pipelineLayout,
        .set = 0,
    };
    auto descriptorUpdateTemplate = VK_ASSERT_RV(
        device->createDescriptorUpdateTemplateUnique(descriptorUpdateTemplateCreateInfo));

    device->updateDescriptorSetWithTemplate(*descriptorSet, *descriptorUpdateTemplate,
                                            descriptorInfo.data());
}

TEST_P(GfxstreamEnd2EndVkTest, MultiThreadedVkMapMemory) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    static constexpr const vkhpp::DeviceSize kSize = 1024;
    const vkhpp::BufferCreateInfo bufferCreateInfo = {
        .size = kSize,
        .usage = vkhpp::BufferUsageFlagBits::eTransferSrc,
    };
    auto buffer = device->createBufferUnique(bufferCreateInfo).value;

    vkhpp::MemoryRequirements bufferMemoryRequirements{};
    device->getBufferMemoryRequirements(*buffer, &bufferMemoryRequirements);

    const uint32_t bufferMemoryIndex = utils::getMemoryType(
        physicalDevice, bufferMemoryRequirements,
        vkhpp::MemoryPropertyFlagBits::eHostVisible | vkhpp::MemoryPropertyFlagBits::eHostCoherent);
    if (bufferMemoryIndex == -1) {
        GTEST_SKIP() << "Skipping test due to no memory type with HOST_VISIBLE | HOST_COHERENT.";
    }

    std::vector<std::thread> threads;
    std::atomic_int threadsReady{0};

    constexpr const int kNumThreads = 2;
    for (int t = 0; t < kNumThreads; t++) {
        threads.emplace_back([&, this]() {
            // Perform some work to ensure host RenderThread started.
            auto buffer2 = device->createBufferUnique(bufferCreateInfo).value;
            ASSERT_THAT(buffer2, IsValidHandle());

            ++threadsReady;
            while (threadsReady.load() != kNumThreads) {
            }

            constexpr const int kNumIterations = 100;
            for (int i = 0; i < kNumIterations; i++) {
                auto buffer3 = device->createBufferUnique(bufferCreateInfo).value;
                ASSERT_THAT(buffer3, IsValidHandle());

                const vkhpp::MemoryAllocateInfo buffer3MemoryAllocateInfo = {
                    .allocationSize = bufferMemoryRequirements.size,
                    .memoryTypeIndex = bufferMemoryIndex,
                };
                auto buffer3Memory = device->allocateMemoryUnique(buffer3MemoryAllocateInfo).value;
                ASSERT_THAT(buffer3Memory, IsValidHandle());

                ASSERT_THAT(device->bindBufferMemory(*buffer3, *buffer3Memory, 0), IsVkSuccess());

                void* mapped = nullptr;
                ASSERT_THAT(device->mapMemory(*buffer3Memory, 0, VK_WHOLE_SIZE,
                                              vkhpp::MemoryMapFlags{}, &mapped),
                            IsVkSuccess());
                ASSERT_THAT(mapped, NotNull());

                device->unmapMemory(*buffer3Memory);
            }
        });
    }

    for (auto& thread : threads) {
        thread.join();
    }
}

TEST_P(GfxstreamEnd2EndVkTest, MultiThreadedResetCommandBuffer) {
    auto [instance, physicalDevice, device, queue, queueFamilyIndex] =
        VK_ASSERT(SetUpTypicalVkTestEnvironment());

    static constexpr const vkhpp::DeviceSize kSize = 1024;
    const vkhpp::BufferCreateInfo bufferCreateInfo = {
        .size = kSize,
        .usage = vkhpp::BufferUsageFlagBits::eTransferSrc,
    };

    static std::mutex queue_mutex;
    std::vector<std::thread> threads;
    std::atomic_int threadsReady{0};

    constexpr const int kNumThreads = 10;
    for (int t = 0; t < kNumThreads; t++) {
        threads.emplace_back([&, this]() {
            // Perform some work to ensure host RenderThread started.
            auto buffer2 = device->createBufferUnique(bufferCreateInfo).value;
            ASSERT_THAT(buffer2, IsValidHandle());

            ++threadsReady;
            while (threadsReady.load() != kNumThreads) {
            }

            const vkhpp::CommandPoolCreateInfo commandPoolCreateInfo = {
                .queueFamilyIndex = queueFamilyIndex,
            };
            auto commandPool = device->createCommandPoolUnique(commandPoolCreateInfo).value;

            const vkhpp::CommandBufferAllocateInfo commandBufferAllocateInfo = {
                .level = vkhpp::CommandBufferLevel::ePrimary,
                .commandPool = *commandPool,
                .commandBufferCount = 1,
            };
            auto commandBuffers = device->allocateCommandBuffersUnique(commandBufferAllocateInfo).value;
            ASSERT_THAT(commandBuffers, Not(IsEmpty()));
            auto commandBuffer = std::move(commandBuffers[0]);
            ASSERT_THAT(commandBuffer, IsValidHandle());

            auto transferFence = device->createFenceUnique(vkhpp::FenceCreateInfo()).value;
            ASSERT_THAT(commandBuffer, IsValidHandle());

            constexpr const int kNumIterations = 1000;
            for (int i = 0; i < kNumIterations; i++) {
                commandBuffer->reset();
                const vkhpp::CommandBufferBeginInfo commandBufferBeginInfo = {
                    .flags = vkhpp::CommandBufferUsageFlagBits::eOneTimeSubmit,
                };
                commandBuffer->begin(commandBufferBeginInfo);

                commandBuffer->end();

                std::vector<vkhpp::CommandBuffer> commandBufferHandles;
                commandBufferHandles.push_back(*commandBuffer);

                const vkhpp::SubmitInfo submitInfo = {
                    .commandBufferCount = static_cast<uint32_t>(commandBufferHandles.size()),
                    .pCommandBuffers = commandBufferHandles.data(),
                };
                {
                    std::lock_guard<std::mutex> qm(queue_mutex);
                    queue.submit(submitInfo, *transferFence);
                }
                auto waitResult = device->waitForFences(*transferFence, VK_TRUE, AsVkTimeout(3s));
                ASSERT_THAT(waitResult, IsVkSuccess());
            }
        });
    }

    for (auto& thread : threads) {
        thread.join();
    }
}

std::vector<TestParams> GenerateTestCases() {
    std::vector<TestParams> cases = {TestParams{
                                         .with_gl = false,
                                         .with_vk = true,
                                         .with_transport = GfxstreamTransport::kVirtioGpuAsg,
                                     },
                                     TestParams{
                                         .with_gl = true,
                                         .with_vk = true,
                                         .with_transport = GfxstreamTransport::kVirtioGpuAsg,
                                     },
                                     TestParams{
                                         .with_gl = false,
                                         .with_vk = true,
                                         .with_transport = GfxstreamTransport::kVirtioGpuPipe,
                                     },
                                     TestParams{
                                         .with_gl = true,
                                         .with_vk = true,
                                         .with_transport = GfxstreamTransport::kVirtioGpuPipe,
                                     }};
    cases = WithAndWithoutFeatures(cases, {"VulkanSnapshots"});
    cases = WithAndWithoutFeatures(cases, {"VulkanUseDedicatedAhbMemoryType"});
    return cases;
}

INSTANTIATE_TEST_CASE_P(GfxstreamEnd2EndTests, GfxstreamEnd2EndVkTest,
                        ::testing::ValuesIn(GenerateTestCases()), &GetTestName);

}  // namespace
}  // namespace tests
}  // namespace gfxstream