Adding view culling to non-depth culling variant

res/shaders/DepthCullingTask.slang

@@ -24,27 +24,8 @@ bool isBoxVisible(AABB bounding)
     // now we calculate what mip level we need to only sample up to 4 texels covering the entire meshlet
     uint2 screenCornerMin = mipDimensions;
     uint2 screenCornerMax = uint2(0, 0);
-    // we use reverse depth, so higher values are closer
+    // lower values are closer
-    float maxDepth = 0;
+    float maxDepth = bounding.projectScreenDepth(modelViewProjection, screenCornerMin, screenCornerMax);
-    {
-        float4 corners[8];
-        corners[0] = float4(bounding.min.x, bounding.min.y, bounding.min.z, 1.0f);
-        corners[1] = float4(bounding.min.x, bounding.min.y, bounding.max.z, 1.0f);
-        corners[2] = float4(bounding.min.x, bounding.max.y, bounding.min.z, 1.0f);
-        corners[3] = float4(bounding.min.x, bounding.max.y, bounding.max.z, 1.0f);
-        corners[4] = float4(bounding.max.x, bounding.min.y, bounding.min.z, 1.0f);
-        corners[5] = float4(bounding.max.x, bounding.min.y, bounding.max.z, 1.0f);
-        corners[6] = float4(bounding.max.x, bounding.max.y, bounding.min.z, 1.0f);
-        corners[7] = float4(bounding.max.x, bounding.max.y, bounding.max.z, 1.0f);
-        for(uint i = 0; i < 8; ++i)
-        {
-            float4 clipCorner = mul(modelViewProjection, corners[i]);
-            float4 screenCorner = clipToScreen(clipCorner);
-            screenCornerMin = uint2(min(screenCornerMin.x, uint(screenCorner.x)), min(screenCornerMin.y, uint(screenCorner.y)));
-            screenCornerMax = uint2(max(screenCornerMax.x, uint(screenCorner.x)), max(screenCornerMax.y, uint(screenCorner.y)));
-            maxDepth = max(maxDepth, screenCorner.z);
-        }
-    }
     uint mipOffset = 0;
     // in theory this wouldnt work if no corner was in screen, as min would be greater that max, however we verified that with view culling
     while(screenCornerMax.x - screenCornerMin.x > 1 || screenCornerMax.y - screenCornerMin.y > 1)
@@ -118,8 +99,10 @@ void taskMain(
             // if the meshlet is outside of the frustum, we skip it since we cant do depth culling anyways
 		    if(meshlet.bounding.insideFrustum(viewFrustum))
             {
+#ifdef DEPTH_CULLING
                 // if the meshlet bounding box is behind the cached depth buffer, we skip
                 if(isBoxVisible(meshlet.bounding))
+#endif
                 {
                     uint index;
                     InterlockedAdd(head, 1, index);

res/shaders/DepthMipGen.slang

@@ -62,5 +62,4 @@ void initialReduce(
     int2 texCoord = groupOffset + threadID.xy;
     float fDepth = pDepthAttachment.texture.Load(int3(texCoord, 0)).r;
     pDepthAttachment.buffer[texCoord.x + (texCoord.y * width)] = fDepth;
-    
 }

res/shaders/DrawListTask.slang

@@ -26,10 +26,7 @@ void taskMain(
         uint cull = p.cullingOffset + i;
         MeshletDescription meshlet = pScene.meshletInfos[m];
         MeshletCullingInfo culling = pScene.cullingInfos[cull];
-#ifdef DEPTH_CULLING
-        // if depth culling is disabled, we draw the whole scene as if cached
         if(culling.wasVisible())
-#endif
         {
             uint index;
             InterlockedAdd(head, 1, index);

res/shaders/lib/Bounding.slang

@@ -16,7 +16,7 @@ struct BoundingSphere
         bool result = true;
         for(int i = 0; i < 4 && result; ++i)
         {
-            if(dot(frustum.sides[i].n, centerRadius.xyz) - frustum.sides[i].d < -getRadius())
+            if(dot(frustum.sides[i].n, getCenter()) - frustum.sides[i].d < -getRadius())
             {
                 result = false;
             }
@@ -27,15 +27,15 @@ struct BoundingSphere
 
 struct AABB
 {
-	float3 min;
+	float3 minCorner;
     float pad0;
-	float3 max;
+	float3 maxCorner;
     float pad1;
     // modified version from https://learnopengl.com/Guest-Articles/2021/Scene/Frustum-Culling
 	bool insideFrustum(Frustum frustum)
     {
-        float3 center = (min + max) * 0.5;
+        float3 center = (minCorner + maxCorner) * 0.5;
-        float3 extents = float3(max.x - center.x, max.y - center.y, max.z - center.z);
+        float3 extents = float3(maxCorner.x - center.x, maxCorner.y - center.y, maxCorner.z - center.z);
         bool result = true;
         for(int i = 0; i < 4 && result; ++i)
         {
@@ -51,4 +51,26 @@ struct AABB
         }
         return result;
 	}
+    float projectScreenDepth(float4x4 mvp, inout uint2 screenCornerMin, inout uint2 screenCornerMax)
+    {
+        float maxDepth = 0;
+        float4 corners[8];
+        corners[0] = float4(minCorner.x, minCorner.y, minCorner.z, 1.0f);
+        corners[1] = float4(minCorner.x, minCorner.y, maxCorner.z, 1.0f);
+        corners[2] = float4(minCorner.x, maxCorner.y, minCorner.z, 1.0f);
+        corners[3] = float4(minCorner.x, maxCorner.y, maxCorner.z, 1.0f);
+        corners[4] = float4(maxCorner.x, minCorner.y, minCorner.z, 1.0f);
+        corners[5] = float4(maxCorner.x, minCorner.y, maxCorner.z, 1.0f);
+        corners[6] = float4(maxCorner.x, maxCorner.y, minCorner.z, 1.0f);
+        corners[7] = float4(maxCorner.x, maxCorner.y, maxCorner.z, 1.0f);
+        for(uint i = 0; i < 8; ++i)
+        {
+            float4 clipCorner = mul(mvp, corners[i]);
+            float4 screenCorner = clipToScreen(clipCorner);
+            screenCornerMin = uint2(min(screenCornerMin.x, uint(screenCorner.x)), min(screenCornerMin.y, uint(screenCorner.y)));
+            screenCornerMax = uint2(max(screenCornerMax.x, uint(screenCorner.x)), max(screenCornerMax.y, uint(screenCorner.y)));
+            maxDepth = max(maxDepth, screenCorner.z);
+        }
+        return maxDepth;
+    }
 };

src/Editor/Asset/MeshLoader.cpp

@@ -404,19 +404,19 @@ void findMeshRoots(aiNode* node, List<aiNode*>& meshNodes) {
 
 void MeshLoader::loadGlobalMeshes(const aiScene* scene, const Array<PMaterialInstanceAsset>& materials, Array<OMesh>& globalMeshes,
                                   Component::Collider& collider) {
-    //List<std::function<void()>> work;
+    List<std::function<void()>> work;
     for (int32 meshIndex = 0; meshIndex < scene->mNumMeshes; ++meshIndex) {
         aiMesh* mesh = scene->mMeshes[meshIndex];
         if (!(mesh->mPrimitiveTypes & aiPrimitiveType_TRIANGLE))
             continue;
-        globalMeshes.add(nullptr);
+        globalMeshes[meshIndex] = new Mesh();
 
         StaticMeshVertexData* vertexData = StaticMeshVertexData::getInstance();
         MeshId id = vertexData->allocateVertexData(mesh->mNumVertices);
         uint64 offset = vertexData->getMeshOffset(id);
         collider.boundingbox.adjust(Vector(mesh->mAABB.mMin.x, mesh->mAABB.mMin.y, mesh->mAABB.mMin.z));
         collider.boundingbox.adjust(Vector(mesh->mAABB.mMax.x, mesh->mAABB.mMax.y, mesh->mAABB.mMax.z));
-        //work.add([&]() {
+        work.add([=, &globalMeshes]() {
             // assume static mesh for now
             Array<Vector4> positions(mesh->mNumVertices);
             StaticArray<Array<Vector2>, MAX_TEXCOORDS> texCoords;
@@ -475,16 +475,15 @@ void MeshLoader::loadGlobalMeshes(const aiScene* scene, const Array<PMaterialIns
 
             // collider.physicsMesh.addCollider(positions, indices, Matrix4(1.0f));
 
-            globalMeshes[meshIndex] = new Mesh();
             globalMeshes[meshIndex]->vertexData = vertexData;
             globalMeshes[meshIndex]->id = id;
             globalMeshes[meshIndex]->referencedMaterial = materials[mesh->mMaterialIndex];
             globalMeshes[meshIndex]->meshlets = std::move(meshlets);
             globalMeshes[meshIndex]->indices = std::move(indices);
             globalMeshes[meshIndex]->vertexCount = mesh->mNumVertices;
-            //});
+            });
     }
-    //getThreadPool().runAndWait(std::move(work));
+    getThreadPool().runAndWait(std::move(work));
 }
 
 Matrix4 convertMatrix(aiMatrix4x4 matrix) {

src/Engine/Graphics/RenderPass/BasePass.cpp

@@ -143,10 +143,6 @@ void BasePass::render() {
                         {
                             .cullMode = Gfx::SeCullModeFlags(twoSided ? Gfx::SE_CULL_MODE_NONE : Gfx::SE_CULL_MODE_BACK_BIT),
                         },
-                    .depthStencilState =
-                        {
-                            .depthCompareOp = Gfx::SE_COMPARE_OP_GREATER_OR_EQUAL,
-                        },
                     .colorBlend =
                         {
                             .attachmentCount = 1,
@@ -168,10 +164,6 @@ void BasePass::render() {
                         {
                             .cullMode = Gfx::SeCullModeFlags(twoSided ? Gfx::SE_CULL_MODE_NONE : Gfx::SE_CULL_MODE_BACK_BIT),
                         },
-                    .depthStencilState =
-                        {
-                            .depthCompareOp = Gfx::SE_COMPARE_OP_GREATER_OR_EQUAL,
-                        },
                     .colorBlend =
                         {
                             .attachmentCount = 1,

src/Engine/Graphics/RenderPass/CachedDepthPass.cpp

@@ -81,10 +81,6 @@ void CachedDepthPass::render() {
                     {
                         .samples = viewport->getSamples(),
                     },
-                .depthStencilState =
-                    {
-                        .depthCompareOp = Gfx::SE_COMPARE_OP_GREATER,
-                    },
                 .colorBlend =
                     {
                         .attachmentCount = 1,
@@ -102,10 +98,6 @@ void CachedDepthPass::render() {
                     {
                         .samples = viewport->getSamples(),
                     },
-                .depthStencilState =
-                    {
-                        .depthCompareOp = Gfx::SE_COMPARE_OP_GREATER,
-                    },
                 .colorBlend =
                     {
                         .attachmentCount = 1,

src/Engine/Graphics/RenderPass/DepthCullingPass.cpp

@@ -118,8 +118,6 @@ void DepthCullingPass::render() {
 
     query->beginQuery();
     graphics->beginRenderPass(renderPass);
-    if (useDepthCulling) {
-
     Array<Gfx::ORenderCommand> commands;
 
     Gfx::ShaderPermutation permutation = graphics->getShaderCompiler()->getTemplate("DepthPass");
@@ -152,10 +150,6 @@ void DepthCullingPass::render() {
                     {
                         .samples = viewport->getSamples(),
                     },
-                    .depthStencilState =
-                        {
-                            .depthCompareOp = Gfx::SE_COMPARE_OP_GREATER,
-                        },
                 .colorBlend =
                     {
                         .attachmentCount = 1,
@@ -173,10 +167,6 @@ void DepthCullingPass::render() {
                     {
                         .samples = viewport->getSamples(),
                     },
-                    .depthStencilState =
-                        {
-                            .depthCompareOp = Gfx::SE_COMPARE_OP_GREATER,
-                        },
                 .colorBlend =
                     {
                         .attachmentCount = 1,
@@ -187,8 +177,8 @@ void DepthCullingPass::render() {
         }
         command->bindDescriptor({viewParamsSet, vertexData->getVertexDataSet(), vertexData->getInstanceDataSet(), set});
         uint32 offset = 0;
-            command->pushConstants(Gfx::SE_SHADER_STAGE_TASK_BIT_EXT | Gfx::SE_SHADER_STAGE_VERTEX_BIT, 0,
+        command->pushConstants(Gfx::SE_SHADER_STAGE_TASK_BIT_EXT | Gfx::SE_SHADER_STAGE_VERTEX_BIT, 0, sizeof(VertexData::DrawCallOffsets),
-                                   sizeof(VertexData::DrawCallOffsets), &offset);
+                               &offset);
         if (graphics->supportMeshShading()) {
             command->drawMesh(vertexData->getNumInstances(), 1, 1);
         } else {
@@ -212,7 +202,6 @@ void DepthCullingPass::render() {
     }
 
     graphics->executeCommands(std::move(commands));
-    }
     graphics->endRenderPass();
     query->endQuery();
     // Sync depth read/write with compute read

src/Engine/Graphics/StaticMeshVertexData.h

@@ -9,16 +9,6 @@
 namespace Seele {
 class StaticMeshVertexData : public VertexData {
   public:
-    struct StaticMatInstance {
-        PMaterialInstance instance;
-        Array<uint32> meshletIds;
-        Gfx::OShaderBuffer culledMeshletBuffer;
-        // Gfx::OShaderBuffer indirectDrawBuffer;
-    };
-    struct StaticMatData {
-        PMaterial material;
-        Array<StaticMatInstance> staticInstance;
-    };
     StaticMeshVertexData();
     virtual ~StaticMeshVertexData();
     static StaticMeshVertexData* getInstance();
@@ -37,12 +27,10 @@ class StaticMeshVertexData : public VertexData {
     virtual Gfx::PDescriptorSet getVertexDataSet() override;
     virtual std::string getTypeName() const override { return "StaticMeshVertexData"; }
     virtual Gfx::PShaderBuffer getPositionBuffer() const override { return positions; }
-    constexpr const Array<StaticMatData>& getStaticMeshes() const { return staticData; }
 
   private:
     virtual void resizeBuffers() override;
     virtual void updateBuffers() override;
-    Array<StaticMatData> staticData;
 
     Gfx::OShaderBuffer positions;
     Array<Vector4> positionData;

src/Engine/Graphics/Vulkan/Buffer.cpp

@@ -184,10 +184,14 @@ Buffer::~Buffer() {
 }
 
 void Buffer::updateContents(uint64 regionOffset, uint64 regionSize, void* buffer) {
+    if (buffers.size() == 0)
+        return;
     getAlloc()->updateContents(regionOffset, regionSize, buffer);
 }
 
 void Buffer::readContents(uint64 regionOffset, uint64 regionSize, void* buffer) {
+    if (buffers.size() == 0)
+        return;
     getAlloc()->readContents(regionOffset, regionSize, buffer);
 }
 
@@ -218,7 +222,6 @@ void Buffer::rotateBuffer(uint64 size, bool preserveContents) {
 }
 
 void Buffer::createBuffer(uint64 size, uint32 destIndex) {
-    if (size > 0) {
     uint32 family = graphics->getFamilyMapping().getQueueTypeFamilyIndex(initialOwner);
     VkBufferCreateInfo info = {
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
@@ -242,7 +245,6 @@ void Buffer::createBuffer(uint64 size, uint32 destIndex) {
                         VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
     }
 }
-}
 
 void Buffer::copyBuffer(uint64 src, uint64 dst) {
     if (src == dst) {
@@ -283,10 +285,16 @@ void Buffer::copyBuffer(uint64 src, uint64 dst) {
                          &dstBarrier, 0, nullptr);
 }
 
-void Buffer::transferOwnership(Gfx::QueueType newOwner) { getAlloc()->transferOwnership(newOwner); }
+void Buffer::transferOwnership(Gfx::QueueType newOwner) {
+    if (buffers.size() == 0)
+        return;
+    getAlloc()->transferOwnership(newOwner);
+}
 
 void Buffer::pipelineBarrier(VkAccessFlags srcAccess, VkPipelineStageFlags srcStage, VkAccessFlags dstAccess,
                              VkPipelineStageFlags dstStage) {
+    if (buffers.size() == 0)
+        return;
     getAlloc()->pipelineBarrier(srcAccess, srcStage, dstAccess, dstStage);
 }