feat: add forward+ rendering

2026-10-11 00:12:23 +02:00
parent c96b2282fb
commit 4bf87bd91b
26 changed files with 1346 additions and 191 deletions
@@ -0,0 +1,126 @@
+#include <destrum/Components/LightComponent.h>
+
+#include <destrum/Graphics/RenderContext.h>
+#include <destrum/ObjectModel/Transform.h>
+
+#include <glm/gtc/constants.hpp>
+#include <glm/gtc/quaternion.hpp>
+#include <glm/gtx/quaternion.hpp>
+
+#include <algorithm>
+#include <array>
+
+#include <imgui.h>
+
+LightComponent::LightComponent(GameObject& parent, LightType type)
+    : Component(parent, "LightComponent")
+    , m_Type(type)
+{
+}
+
+void LightComponent::SetIntensity(float intensity)
+{
+    m_Intensity = std::max(intensity, 0.0f);
+}
+
+void LightComponent::SetRange(float range)
+{
+    m_Range = std::max(range, 0.01f);
+}
+
+void LightComponent::SetConeAngles(float innerConeAngle, float outerConeAngle)
+{
+    m_InnerConeAngle = std::clamp(innerConeAngle, 0.0f, glm::half_pi<float>());
+    m_OuterConeAngle = std::clamp(outerConeAngle, m_InnerConeAngle, glm::half_pi<float>());
+}
+
+void LightComponent::Render(const RenderContext& ctx)
+{
+    auto& transform = GetTransform();
+
+    ctx.renderer.submitLight(Light{
+        .type = m_Type,
+        .position = transform.GetWorldPosition(),
+        .direction = transform.GetWorldRotation() * glm::vec3{0.0f, 0.0f, -1.0f},
+        .color = m_Color,
+        .intensity = m_Intensity,
+        .range = m_Range,
+        .innerConeAngle = m_InnerConeAngle,
+        .outerConeAngle = m_OuterConeAngle,
+    });
+}
+
+nlohmann::json LightComponent::Serialize() const {
+    return {
+        {"type", static_cast<std::uint32_t>(m_Type)},
+        {"color", {m_Color.x, m_Color.y, m_Color.z}},
+        {"intensity", m_Intensity},
+        {"range", m_Range},
+        {"innerConeAngle", m_InnerConeAngle},
+        {"outerConeAngle", m_OuterConeAngle}
+    };
+}
+
+void LightComponent::Deserialize(const nlohmann::json& data) {
+    if (data.contains("type")) {
+        const auto type = data.at("type").get<std::uint32_t>();
+        if (type <= static_cast<std::uint32_t>(LightType::Spot)) {
+            m_Type = static_cast<LightType>(type);
+        }
+    }
+    if (data.contains("color")) {
+        const auto& value = data.at("color");
+        m_Color = {value.at(0).get<float>(), value.at(1).get<float>(), value.at(2).get<float>()};
+    }
+    if (data.contains("intensity")) {
+        SetIntensity(data.at("intensity").get<float>());
+    }
+    if (data.contains("range")) {
+        SetRange(data.at("range").get<float>());
+    }
+
+    float innerConeAngle = m_InnerConeAngle;
+    float outerConeAngle = m_OuterConeAngle;
+    if (data.contains("innerConeAngle")) {
+        innerConeAngle = data.at("innerConeAngle").get<float>();
+    }
+    if (data.contains("outerConeAngle")) {
+        outerConeAngle = data.at("outerConeAngle").get<float>();
+    }
+    SetConeAngles(innerConeAngle, outerConeAngle);
+}
+
+void LightComponent::ImGuiInspector() {
+    constexpr std::array typeNames{"Directional", "Point", "Spot"};
+
+    int type = static_cast<int>(m_Type);
+    if (ImGui::Combo("Type", &type, typeNames.data(), static_cast<int>(typeNames.size()))) {
+        SetType(static_cast<LightType>(type));
+    }
+
+    ImGui::ColorEdit3("Color", &m_Color.x);
+
+    float intensity = m_Intensity;
+    if (ImGui::DragFloat("Intensity", &intensity, 0.05f, 0.0f)) {
+        SetIntensity(intensity);
+    }
+
+    if (m_Type == LightType::Directional) {
+        return;
+    }
+
+    float range = m_Range;
+    if (ImGui::DragFloat("Range", &range, 0.05f, 0.01f)) {
+        SetRange(range);
+    }
+
+    if (m_Type == LightType::Spot) {
+        float innerConeAngle = m_InnerConeAngle;
+        float outerConeAngle = m_OuterConeAngle;
+        bool changed = ImGui::SliderAngle("Inner Cone", &innerConeAngle, 0.0f, 90.0f);
+        changed |= ImGui::SliderAngle("Outer Cone", &outerConeAngle, 0.0f, 90.0f);
+        if (changed) {
+            SetConeAngles(innerConeAngle, outerConeAngle);
+        }
+    }
+}
@@ -0,0 +1,271 @@
+#include <destrum/Graphics/Pipelines/LightCullingPipeline.h>
+
+#include "destrum/FS/AssetFS.h"
+#include "destrum/Graphics/GfxDevice.h"
+#include "destrum/Graphics/RenderResources.h"
+#include "destrum/Graphics/Util.h"
+
+#include <array>
+#include <cassert>
+#include <cmath>
+#include <cstdint>
+#include <stdexcept>
+
+#include "volk.h"
+#include "spdlog/spdlog.h"
+
+void LightCullingPipeline::init(GfxDevice& gfxDevice, RenderResources& resources, glm::ivec2 drawImageSize)
+{
+    try
+    {
+        const auto& device = gfxDevice.getDevice();
+
+        const auto pushConstant = VkPushConstantRange{
+            .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+            .offset = 0,
+            .size = sizeof(PushConstants),
+        };
+
+        const auto pushConstants = std::array{pushConstant};
+
+        const auto layouts = std::array{
+            resources.getBindlessDescSetLayout()
+        };
+
+        VkPipelineLayoutCreateInfo pipelineLayoutInfo{};
+        pipelineLayoutInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
+        pipelineLayoutInfo.setLayoutCount = static_cast<std::uint32_t>(layouts.size());
+        pipelineLayoutInfo.pSetLayouts = layouts.data();
+        pipelineLayoutInfo.pushConstantRangeCount = static_cast<std::uint32_t>(pushConstants.size());
+        pipelineLayoutInfo.pPushConstantRanges = pushConstants.data();
+
+        if (vkCreatePipelineLayout(device, &pipelineLayoutInfo, nullptr, &m_pipelineLayout) != VK_SUCCESS)
+        {
+            throw std::runtime_error("Could not make light culling pipeline layout");
+        }
+
+        ComputePipelineConfigInfo pipelineConfig{};
+        pipelineConfig.name = "light culling compute pipeline";
+        pipelineConfig.pipelineLayout = m_pipelineLayout;
+
+        const auto lightCullingShaderPath = AssetFS::GetInstance().GetCookedPathForFile(
+            "engine://shaders/light_culling.comp");
+
+        lightCullingPipeline = std::make_unique<ComputePipeline>(
+            gfxDevice,
+            lightCullingShaderPath.generic_string(),
+            pipelineConfig
+        );
+
+        for (std::size_t i = 0; i < FRAMES_IN_FLIGHT; ++i)
+        {
+            auto& lightsBuffer = framesData[i].lightsBuffer;
+            lightsBuffer.capacity = MAX_LIGHTS;
+            lightsBuffer.buffer = gfxDevice.createBuffer(
+                MAX_LIGHTS * sizeof(GPULight),
+                VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT,
+                VMA_MEMORY_USAGE_CPU_TO_GPU);
+            vkutil::addDebugLabel(device, lightsBuffer.buffer.buffer, "lights");
+        }
+
+        createTileBuffer(gfxDevice, drawImageSize);
+    }
+    catch (...)
+    {
+        cleanup(gfxDevice);
+        throw;
+    }
+}
+
+void LightCullingPipeline::cleanup(GfxDevice& gfxDevice)
+{
+    for (auto& frame : framesData)
+    {
+        gfxDevice.destroyBuffer(frame.lightsBuffer.buffer);
+        frame.lightsBuffer.buffer = {};
+        frame.lightsBuffer.size = 0;
+    }
+
+    gfxDevice.destroyBuffer(tileLightIndicesBuffer);
+    tileLightIndicesBuffer = {};
+    tileCount = glm::uvec2{0u, 0u};
+
+    if (m_pipelineLayout != VK_NULL_HANDLE && gfxDevice.getDevice() != VK_NULL_HANDLE)
+    {
+        vkDestroyPipelineLayout(gfxDevice.getDevice(), m_pipelineLayout, nullptr);
+    }
+    m_pipelineLayout = VK_NULL_HANDLE;
+    lightCullingPipeline.reset();
+}
+
+void LightCullingPipeline::resize(GfxDevice& gfxDevice, glm::ivec2 drawImageSize)
+{
+    gfxDevice.waitIdle();
+
+    gfxDevice.destroyBuffer(tileLightIndicesBuffer);
+    tileLightIndicesBuffer = {};
+    createTileBuffer(gfxDevice, drawImageSize);
+}
+
+void LightCullingPipeline::createTileBuffer(GfxDevice& gfxDevice, glm::ivec2 drawImageSize)
+{
+    assert(drawImageSize.x > 0 && drawImageSize.y > 0);
+
+    tileCount = glm::uvec2{
+        (static_cast<std::uint32_t>(drawImageSize.x) + TILE_SIZE - 1) / TILE_SIZE,
+        (static_cast<std::uint32_t>(drawImageSize.y) + TILE_SIZE - 1) / TILE_SIZE,
+    };
+
+    const std::size_t numTiles = static_cast<std::size_t>(tileCount.x) * tileCount.y;
+    tileLightIndicesBuffer = gfxDevice.createBuffer(
+        numTiles * TILE_STRIDE * sizeof(std::uint32_t),
+        VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT);
+    vkutil::addDebugLabel(gfxDevice.getDevice(), tileLightIndicesBuffer.buffer, "tile light indices");
+
+    spdlog::debug("Forward+ light grid: {}x{} tiles", tileCount.x, tileCount.y);
+}
+
+void LightCullingPipeline::beginDrawing(std::size_t frameIndex)
+{
+    currentFrameIndex = frameIndex;
+    getCurrentFrameData().lightsBuffer.clear();
+}
+
+void LightCullingPipeline::appendLight(const Light& light)
+{
+    auto& lightsBuffer = getCurrentFrameData().lightsBuffer;
+    if (lightsBuffer.size >= lightsBuffer.capacity)
+    {
+        if (!warnedAboutLightLimit)
+        {
+            spdlog::warn("Light limit of {} reached, extra lights are ignored", MAX_LIGHTS);
+            warnedAboutLightLimit = true;
+        }
+        return;
+    }
+
+    const float directionLength = glm::length(light.direction);
+    const glm::vec3 direction = directionLength > 1e-6f
+                                    ? light.direction / directionLength
+                                    : glm::vec3{0.0f, -1.0f, 0.0f};
+
+    const auto gpuLight = GPULight{
+        .position = light.position,
+        .range = light.range,
+        .color = light.color,
+        .intensity = light.intensity,
+        .direction = direction,
+        .type = static_cast<std::uint32_t>(light.type),
+        .spotInnerCos = std::cos(light.innerConeAngle),
+        .spotOuterCos = std::cos(light.outerConeAngle),
+        .padding = glm::vec2{0.0f},
+    };
+    lightsBuffer.append(gpuLight);
+}
+
+void LightCullingPipeline::flushCurrentFrame(VkCommandBuffer cmd, GfxDevice& gfxDevice)
+{
+    auto& buffer = getCurrentFrameData().lightsBuffer.buffer;
+    gfxDevice.getMemoryManager().flushAllocation(buffer, 0, VK_WHOLE_SIZE);
+    vkutil::bufferHostWriteToShaderReadBarrier(
+        cmd,
+        buffer.buffer,
+        0,
+        static_cast<VkDeviceSize>(MAX_LIGHTS * sizeof(GPULight)));
+}
+
+void LightCullingPipeline::cullLights(
+    VkCommandBuffer cmd,
+    const RenderResources& resources,
+    const GPUBuffer& sceneDataBuffer,
+    ImageID depthImageId,
+    VkExtent2D renderExtent)
+{
+    if (!lightCullingPipeline)
+    {
+        return;
+    }
+
+    {
+        // previous frame's mesh pass might still be reading the tile lists
+        const VkBufferMemoryBarrier2 tileBufferBarrier{
+            .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER_2,
+            .srcStageMask = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT,
+            .srcAccessMask = VK_ACCESS_2_SHADER_STORAGE_READ_BIT,
+            .dstStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT,
+            .dstAccessMask = VK_ACCESS_2_SHADER_STORAGE_WRITE_BIT,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .buffer = tileLightIndicesBuffer.buffer,
+            .offset = 0,
+            .size = VK_WHOLE_SIZE,
+        };
+
+        const VkDependencyInfo dependencyInfo{
+            .sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO,
+            .bufferMemoryBarrierCount = 1,
+            .pBufferMemoryBarriers = &tileBufferBarrier,
+        };
+        vkCmdPipelineBarrier2(cmd, &dependencyInfo);
+    }
+
+    lightCullingPipeline->bind(cmd);
+    resources.bindBindlessDescSet(cmd, m_pipelineLayout, VK_PIPELINE_BIND_POINT_COMPUTE);
+
+    const auto cs = PushConstants{
+        .sceneDataBuffer = sceneDataBuffer.address,
+        .screenSize = glm::uvec2{renderExtent.width, renderExtent.height},
+        .depthTexId = static_cast<std::uint32_t>(depthImageId),
+        .padding = 0,
+    };
+
+    vkCmdPushConstants(cmd,
+                       m_pipelineLayout,
+                       VK_SHADER_STAGE_COMPUTE_BIT,
+                       0,
+                       sizeof(PushConstants),
+                       &cs);
+
+    // one workgroup per tile
+    vkCmdDispatch(cmd, tileCount.x, tileCount.y, 1);
+
+    const VkBufferMemoryBarrier2 tileBufferBarrier{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER_2,
+        .srcStageMask = VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT,
+        .srcAccessMask = VK_ACCESS_2_SHADER_STORAGE_WRITE_BIT,
+        .dstStageMask = VK_PIPELINE_STAGE_2_FRAGMENT_SHADER_BIT,
+        .dstAccessMask = VK_ACCESS_2_SHADER_STORAGE_READ_BIT,
+        .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+        .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+        .buffer = tileLightIndicesBuffer.buffer,
+        .offset = 0,
+        .size = VK_WHOLE_SIZE,
+    };
+
+    const VkDependencyInfo dependencyInfo{
+        .sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO,
+        .bufferMemoryBarrierCount = 1,
+        .pBufferMemoryBarriers = &tileBufferBarrier,
+    };
+    vkCmdPipelineBarrier2(cmd, &dependencyInfo);
+}
+
+VkDeviceAddress LightCullingPipeline::getLightsBufferAddress() const
+{
+    return getCurrentFrameData().lightsBuffer.buffer.address;
+}
+
+std::uint32_t LightCullingPipeline::getNumLights() const
+{
+    return static_cast<std::uint32_t>(getCurrentFrameData().lightsBuffer.size);
+}
+
+LightCullingPipeline::PerFrameData& LightCullingPipeline::getCurrentFrameData()
+{
+    return framesData[currentFrameIndex % FRAMES_IN_FLIGHT];
+}
+
+const LightCullingPipeline::PerFrameData& LightCullingPipeline::getCurrentFrameData() const
+{
+    return framesData[currentFrameIndex % FRAMES_IN_FLIGHT];
+}
@@ -23,66 +23,122 @@ void MeshPipeline::init(
    VkFormat drawImageFormat,
    VkFormat depthImageFormat)
 {
-    try {
-    const auto vertexShader =
-        AssetFS::GetInstance().GetCookedPathForFile("engine://shaders/mesh.vert");
+    try
+    {
+        const auto vertexShader =
+            AssetFS::GetInstance().GetCookedPathForFile("engine://shaders/mesh.vert");

-    const auto fragShader =
-        AssetFS::GetInstance().GetCookedPathForFile("engine://shaders/mesh.frag");
+        const auto fragShader =
+            AssetFS::GetInstance().GetCookedPathForFile("engine://shaders/mesh.frag");

-    const auto bufferRange = VkPushConstantRange{
-        .stageFlags = VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_FRAGMENT_BIT,
-        .offset = 0,
-        .size = sizeof(PushConstants),
-    };
+        const auto bufferRange = VkPushConstantRange{
+            .stageFlags = VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_FRAGMENT_BIT,
+            .offset = 0,
+            .size = sizeof(PushConstants),
+        };

-    const auto pushConstantRanges = std::array{bufferRange};
+        const auto pushConstantRanges = std::array{bufferRange};

-    const auto layouts = std::array{
-        resources.getBindlessDescSetLayout()
-    };
+        const auto layouts = std::array{
+            resources.getBindlessDescSetLayout()
+        };

-    VkPipelineLayoutCreateInfo pipelineLayoutInfo{};
-    pipelineLayoutInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
-    pipelineLayoutInfo.setLayoutCount = static_cast<std::uint32_t>(layouts.size());
-    pipelineLayoutInfo.pSetLayouts = layouts.data();
-    pipelineLayoutInfo.pushConstantRangeCount =
-        static_cast<std::uint32_t>(pushConstantRanges.size());
-    pipelineLayoutInfo.pPushConstantRanges = pushConstantRanges.data();
+        VkPipelineLayoutCreateInfo pipelineLayoutInfo{};
+        pipelineLayoutInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
+        pipelineLayoutInfo.setLayoutCount = static_cast<std::uint32_t>(layouts.size());
+        pipelineLayoutInfo.pSetLayouts = layouts.data();
+        pipelineLayoutInfo.pushConstantRangeCount =
+            static_cast<std::uint32_t>(pushConstantRanges.size());
+        pipelineLayoutInfo.pPushConstantRanges = pushConstantRanges.data();

-    if (vkCreatePipelineLayout(
+        if (vkCreatePipelineLayout(
            gfxDevice.getDevice(),
            &pipelineLayoutInfo,
            nullptr,
            &m_pipelineLayout) != VK_SUCCESS)
-    {
-        throw std::runtime_error("Could not create mesh pipeline layout");
+        {
+            throw std::runtime_error("Could not create mesh pipeline layout");
+        }
+
+        {
+            PipelineConfigInfo pipelineConfig{};
+            Pipeline::DefaultPipelineConfigInfo(pipelineConfig);
+
+            pipelineConfig.name = "Mesh Depth Prepass Pipeline";
+            pipelineConfig.pipelineLayout = m_pipelineLayout;
+
+            pipelineConfig.vertexAttributeDescriptions = {};
+            pipelineConfig.vertexBindingDescriptions = {};
+
+            pipelineConfig.colorAttachments = {};
+            pipelineConfig.colorBlendInfo.attachmentCount = 0;
+            pipelineConfig.depthAttachment = depthImageFormat;
+
+            m_depthPrepassPipeline = std::make_unique<Pipeline>(
+                gfxDevice,
+                vertexShader.string(),
+                "",
+                pipelineConfig
+            );
+        }
+
+        {
+            PipelineConfigInfo pipelineConfig{};
+            Pipeline::DefaultPipelineConfigInfo(pipelineConfig);
+
+            pipelineConfig.name = "Mesh Pipeline";
+            pipelineConfig.pipelineLayout = m_pipelineLayout;
+
+            pipelineConfig.vertexAttributeDescriptions = {};
+            pipelineConfig.vertexBindingDescriptions = {};
+
+            pipelineConfig.colorAttachments = {drawImageFormat};
+            pipelineConfig.depthAttachment = depthImageFormat;
+
+            // depth was already written by the pre-pass, only shade the visible surface
+            pipelineConfig.depthStencilInfo.depthWriteEnable = VK_FALSE;
+            pipelineConfig.depthStencilInfo.depthCompareOp = VK_COMPARE_OP_LESS_OR_EQUAL;
+
+            m_pipeline = std::make_unique<Pipeline>(
+                gfxDevice,
+                vertexShader.string(),
+                fragShader.string(),
+                pipelineConfig
+            );
+        }
    }
-
-    PipelineConfigInfo pipelineConfig{};
-    Pipeline::DefaultPipelineConfigInfo(pipelineConfig);
-
-    pipelineConfig.name = "Mesh Pipeline";
-    pipelineConfig.pipelineLayout = m_pipelineLayout;
-
-    pipelineConfig.vertexAttributeDescriptions = {};
-    pipelineConfig.vertexBindingDescriptions = {};
-
-    pipelineConfig.colorAttachments = {drawImageFormat};
-    pipelineConfig.depthAttachment = depthImageFormat;
-
-        m_pipeline = std::make_unique<Pipeline>(
-        gfxDevice,
-        vertexShader.string(),
-        fragShader.string(),
-        pipelineConfig
-        );
-    } catch (...) {
+    catch (...)
+    {
        cleanup(gfxDevice.getDevice());
        throw;
    }
 }

+void MeshPipeline::drawDepthPrepass(
+    VkCommandBuffer cmd,
+    VkExtent2D renderExtent,
+    const RenderResources& resources,
+    const Camera& camera,
+    const GPUBuffer& sceneDataBuffer,
+    const std::vector<MeshDrawCommand>& drawCommands,
+    const std::vector<std::size_t>& sortedDrawCommands)
+{
+    if (!m_depthPrepassPipeline)
+    {
+        return;
+    }
+
+    drawMeshes(
+        cmd,
+        *m_depthPrepassPipeline,
+        renderExtent,
+        resources,
+        camera,
+        sceneDataBuffer,
+        drawCommands,
+        sortedDrawCommands);
+}
+
 void MeshPipeline::draw(
    VkCommandBuffer cmd,
    VkExtent2D renderExtent,
@@ -92,13 +148,35 @@ void MeshPipeline::draw(
    const std::vector<MeshDrawCommand>& drawCommands,
    const std::vector<std::size_t>& sortedDrawCommands)
 {
-    if (!m_pipeline) {
+    if (!m_pipeline)
+    {
        return;
    }

+    drawMeshes(
+        cmd,
+        *m_pipeline,
+        renderExtent,
+        resources,
+        camera,
+        sceneDataBuffer,
+        drawCommands,
+        sortedDrawCommands);
+}
+
+void MeshPipeline::drawMeshes(
+    VkCommandBuffer cmd,
+    const Pipeline& pipeline,
+    VkExtent2D renderExtent,
+    const RenderResources& resources,
+    const Camera& camera,
+    const GPUBuffer& sceneDataBuffer,
+    const std::vector<MeshDrawCommand>& drawCommands,
+    const std::vector<std::size_t>& sortedDrawCommands)
+{
    const MeshCache& meshCache = resources.meshes();

-    m_pipeline->bind(cmd);
+    pipeline.bind(cmd);

    resources.bindBindlessDescSet(cmd, m_pipelineLayout);

@@ -130,8 +208,10 @@ void MeshPipeline::draw(

    const auto frustum = edge::createFrustumFromCamera(camera);

-    const auto drawOne = [&](const MeshDrawCommand& dc) {
-        if (!edge::isInFrustum(frustum, dc.worldBoundingSphere)) {
+    const auto drawOne = [&](const MeshDrawCommand& dc)
+    {
+        if (!edge::isInFrustum(frustum, dc.worldBoundingSphere))
+        {
            return;
        }

@@ -139,7 +219,8 @@ void MeshPipeline::draw(

        const auto& mesh = meshCache.getMesh(dc.meshId);

-        if (dc.meshId != prevMeshId) {
+        if (dc.meshId != prevMeshId)
+        {
            prevMeshId = dc.meshId;

            vkCmdBindIndexBuffer(
@@ -156,9 +237,9 @@ void MeshPipeline::draw(
            .transform = dc.transformMatrix,
            .sceneDataBuffer = sceneDataBuffer.address,
            .vertexBuffer =
-                dc.skinnedMesh != nullptr
-                    ? dc.skinnedMesh->skinnedVertexBuffer.address
-                    : mesh.vertexBuffer.address,
+            dc.skinnedMesh != nullptr
+                ? dc.skinnedMesh->skinnedVertexBuffer.address
+                : mesh.vertexBuffer.address,
            .materialId = dc.materialId,
            .padding = 0,
        };
@@ -175,26 +256,30 @@ void MeshPipeline::draw(
        vkCmdDrawIndexed(cmd, mesh.numIndices, 1, 0, 0, 0);
    };

-    if (!sortedDrawCommands.empty()) {
-        for (std::size_t drawIndex : sortedDrawCommands) {
+    if (!sortedDrawCommands.empty())
+    {
+        for (std::size_t drawIndex : sortedDrawCommands)
+        {
            assert(drawIndex < drawCommands.size());
            drawOne(drawCommands[drawIndex]);
        }
-    } else {
-        for (const MeshDrawCommand& dc : drawCommands) {
+    }
+    else
+    {
+        for (const MeshDrawCommand& dc : drawCommands)
+        {
            drawOne(dc);
        }
    }
-
-    // Optional:
-    // spdlog::debug("Actual mesh draw calls: {}", actualDrawCalls);
 }

 void MeshPipeline::cleanup(VkDevice device)
 {
    m_pipeline.reset();
+    m_depthPrepassPipeline.reset();

-    if (m_pipelineLayout != VK_NULL_HANDLE && device != VK_NULL_HANDLE) {
+    if (m_pipelineLayout != VK_NULL_HANDLE && device != VK_NULL_HANDLE)
+    {
        vkDestroyPipelineLayout(device, m_pipelineLayout, nullptr);
    }
    m_pipelineLayout = VK_NULL_HANDLE;
@@ -119,14 +119,16 @@ void SkyboxPipeline::draw(

    const glm::mat3 r = glm::mat3(skyboxRotation);

+    // rotation only, translation in the inverse makes the sky jitter far from the origin
+    const glm::mat4 viewRotation = glm::mat4(glm::mat3(camera.GetViewMatrix()));
+
    const auto pcs = SkyboxPushConstants{
-        .invViewProj = glm::inverse(camera.GetViewProjectionMatrix()),
+        .invViewProj = glm::inverse(camera.GetProjectionMatrix() * viewRotation),
        .skyboxRot = {
            glm::vec4(r[0], 0.f),
            glm::vec4(r[1], 0.f),
            glm::vec4(r[2], 0.f)
        },
-        .cameraPos = camera.GetPosition(),
        .skyboxTextureId = static_cast<std::uint32_t>(skyboxTextureId),
    };

@@ -266,10 +266,13 @@ const VkDescriptorSet& RenderResources::getBindlessDescSet() const {
    return imageCache->bindlessSetManager.getDescSet();
 }

-void RenderResources::bindBindlessDescSet(VkCommandBuffer cmd, VkPipelineLayout layout) const {
+void RenderResources::bindBindlessDescSet(
+    VkCommandBuffer cmd,
+    VkPipelineLayout layout,
+    VkPipelineBindPoint bindPoint) const {
    vkCmdBindDescriptorSets(
        cmd,
-        VK_PIPELINE_BIND_POINT_GRAPHICS,
+        bindPoint,
        layout,
        0,
        1,
@@ -21,7 +21,8 @@ GameRenderer::GameRenderer()

 void GameRenderer::init(GfxDevice& gfxDevice, RenderResources& _resources, glm::ivec2 drawImageSize)
 {
-    try {
+    try
+    {
        resources = &_resources;
        sceneDataBuffer.init(
            gfxDevice,
@@ -34,12 +35,15 @@ void GameRenderer::init(GfxDevice& gfxDevice, RenderResources& _resources, glm::
        meshPipeline = std::make_unique<MeshPipeline>();
        meshPipeline->init(gfxDevice, _resources, drawImageFormat, depthImageFormat);

-skyboxPipeline = std::make_unique<SkyboxPipeline>();
+        skyboxPipeline = std::make_unique<SkyboxPipeline>();
        skyboxPipeline->init(gfxDevice, _resources, drawImageFormat, depthImageFormat);

        skinningPipeline = std::make_unique<SkinningPipeline>();
        skinningPipeline->init(gfxDevice);

+        lightCullingPipeline = std::make_unique<LightCullingPipeline>();
+        lightCullingPipeline->init(gfxDevice, _resources, drawImageSize);
+
        lineRenderingPass = std::make_unique<LineRenderingPass>();
        lineRenderingPass->init(gfxDevice, drawImageFormat);

@@ -48,7 +52,9 @@ skyboxPipeline = std::make_unique<SkyboxPipeline>();

        GameState::GetInstance().SetRenderer(this);
        initialized = true;
-    } catch (...) {
+    }
+    catch (...)
+    {
        cleanup(gfxDevice);
        throw;
    }
@@ -59,6 +65,7 @@ void GameRenderer::beginDrawing(GfxDevice& gfxDevice)
    flushMaterialUpdates(gfxDevice);
    meshDrawCommands.clear();
    skinningPipeline->beginDrawing(gfxDevice.getCurrentFrameIndex());
+    lightCullingPipeline->beginDrawing(gfxDevice.getCurrentFrameIndex());
    QuadRenderingManager::GetInstance().ClearQuads();
 }

@@ -70,13 +77,16 @@ void GameRenderer::endDrawing()
    std::stable_sort(
        sortedMeshDrawCommands.begin(),
        sortedMeshDrawCommands.end(),
-        [this](std::size_t left, std::size_t right) {
+        [this](std::size_t left, std::size_t right)
+        {
            const auto& lhs = meshDrawCommands[left];
            const auto& rhs = meshDrawCommands[right];
-            if (lhs.meshId != rhs.meshId) {
+            if (lhs.meshId != rhs.meshId)
+            {
                return lhs.meshId < rhs.meshId;
            }
-            if (lhs.materialId != rhs.materialId) {
+            if (lhs.materialId != rhs.materialId)
+            {
                return lhs.materialId < rhs.materialId;
            }
            return (lhs.skinnedMesh != nullptr) < (rhs.skinnedMesh != nullptr);
@@ -106,9 +116,17 @@ void GameRenderer::draw(VkCommandBuffer cmd, GfxDevice& gfxDevice, const Camera&
        }
    }

+    {
+        TracyVkZone(gfxDevice.getTracyVkCtx(), cmd, "Lights Buffer Flush");
+
+        lightCullingPipeline->flushCurrentFrame(cmd, gfxDevice);
+    }
+
+    // TODO: there has to be a better way to do this
    const auto gpuSceneData = GPUSceneData{
        .view = sceneData.camera.GetViewMatrix(),
        .proj = sceneData.camera.GetProjectionMatrix(),
+        .invProj = glm::inverse(sceneData.camera.GetProjectionMatrix()),
        .viewProj = sceneData.camera.GetViewProjectionMatrix(),
        .cameraPos = glm::vec4{sceneData.camera.m_position, 1.f},
        .ambientColor = {sceneData.ambientColor},
@@ -116,6 +134,12 @@ void GameRenderer::draw(VkCommandBuffer cmd, GfxDevice& gfxDevice, const Camera&
        .fogColor = {sceneData.fogColor},
        .fogDensity = sceneData.fogDensity,
        .materialsBuffer = resources->materials().getMaterialDataBufferAddress(),
+        .lightsBuffer = lightCullingPipeline->getLightsBufferAddress(),
+        .tileLightIndicesBuffer = lightCullingPipeline->getTileLightIndicesBufferAddress(),
+        .numLights = lightCullingPipeline->getNumLights(),
+        .tileCountX = lightCullingPipeline->getTileCount().x,
+        .debugLightHeatmap = renderLightHeatmap ? 1u : 0u,
+        .padding = 0,
    };

    {
@@ -165,12 +189,60 @@ void GameRenderer::draw(VkCommandBuffer cmd, GfxDevice& gfxDevice, const Camera&
        depthImage.setLayout(VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_OPTIMAL);
    }

+    {
+        TracyVkZone(gfxDevice.getTracyVkCtx(), cmd, "Depth Prepass");
+
+        auto depthPrepassInfo = vkutil::createRenderingInfo({
+            .renderExtent = drawImage.getExtent2D(),
+            .depthImageView = depthImage.imageView,
+            .depthImageClearValue = 1.f,
+        });
+        depthPrepassInfo.renderingInfo.pDepthAttachment = &depthPrepassInfo.depthAttachment;
+
+        vkCmdBeginRendering(cmd, &depthPrepassInfo.renderingInfo);
+
+        meshPipeline->drawDepthPrepass(
+            cmd,
+            drawImage.getExtent2D(),
+            *resources,
+            camera,
+            sceneDataBuffer.getBuffer(),
+            meshDrawCommands,
+            sortedMeshDrawCommands);
+
+        vkCmdEndRendering(cmd);
+    }
+
+    {
+        TracyVkZone(gfxDevice.getTracyVkCtx(), cmd, "Light Culling");
+
+        vkutil::transitionImage(
+            cmd,
+            depthImage.image,
+            depthImage.layout,
+            VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
+        depthImage.setLayout(VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
+
+        lightCullingPipeline->cullLights(
+            cmd,
+            *resources,
+            sceneDataBuffer.getBuffer(),
+            depthImageId,
+            drawImage.getExtent2D());
+
+        vkutil::transitionImage(
+            cmd,
+            depthImage.image,
+            depthImage.layout,
+            VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_OPTIMAL);
+        depthImage.setLayout(VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_OPTIMAL);
+    }
+
    auto renderInfo = vkutil::createRenderingInfo({
        .renderExtent = drawImage.getExtent2D(),
        .colorImageView = drawImage.imageView,
        .colorImageClearValue = glm::vec4{0.f, 0.f, 0.f, 1.f},
        .depthImageView = depthImage.imageView,
-        .depthImageClearValue = 1.f,
    });
    renderInfo.renderingInfo.pColorAttachments = &renderInfo.colorAttachment;
    renderInfo.renderingInfo.pDepthAttachment = &renderInfo.depthAttachment;
@@ -226,11 +298,11 @@ void GameRenderer::draw(VkCommandBuffer cmd, GfxDevice& gfxDevice, const Camera&
            *resources,
            drawImage,
            glm::ortho(0.0f,
-                static_cast<float>(drawImage.extent.width),
-                static_cast<float>(drawImage.extent.height),
-                0.0f, -1.0f, 1.0f));
+                       static_cast<float>(drawImage.extent.width),
+                       static_cast<float>(drawImage.extent.height),
+                       0.0f, -1.0f, 1.0f));
    }
-// vkutil::cmdEndLabel(cmd);
+    // vkutil::cmdEndLabel(cmd);
 }

 void GameRenderer::drawQuads(
@@ -246,11 +318,12 @@ void GameRenderer::cleanup(GfxDevice& gfxDevice)
 {
    VkDevice device = gfxDevice.getDevice();

-    if (device != VK_NULL_HANDLE) {
+    if (device != VK_NULL_HANDLE)
+    {
        vkDeviceWaitIdle(device);
    }

-if (skinningPipeline)
+    if (skinningPipeline)
        skinningPipeline->cleanup(gfxDevice);

    if (lineRenderingPass)
@@ -259,6 +332,9 @@ if (skinningPipeline)
    if (quadRendererPass)
        quadRendererPass->cleanup(gfxDevice);

+    if (lightCullingPipeline)
+        lightCullingPipeline->cleanup(gfxDevice);
+
    if (skyboxPipeline)
        skyboxPipeline->cleanup(device);

@@ -281,6 +357,7 @@ if (skinningPipeline)
    skinningPipeline.reset();
    lineRenderingPass.reset();
    quadRendererPass.reset();
+    lightCullingPipeline.reset();
    pendingMaterialUploads.clear();
    meshDrawCommands.clear();
    sortedMeshDrawCommands.clear();
@@ -323,6 +400,11 @@ void GameRenderer::drawSkinnedMesh(MeshID id,
    });
 }

+void GameRenderer::submitLight(const Light& light)
+{
+    lightCullingPipeline->appendLight(light);
+}
+
 const GPUImage& GameRenderer::getDrawImage() const
 {
    assert(resources);
@@ -349,7 +431,8 @@ void GameRenderer::setSkyboxTexture(ImageID skyboxImageId)

 void GameRenderer::flushMaterialUpdates(GfxDevice& gfxDevice)
 {
-    if (!pendingMaterialUploads.empty()) {
+    if (!pendingMaterialUploads.empty())
+    {
        // Material data is currently shared by all frame contexts. Wait before
        // mutating it so an older in-flight frame cannot read the overwritten
        // range. This can later be replaced with per-frame material buffers.