Finished morph target implementation.

stripe2933 · Jan 25, 2025 · 09b9a88 · 09b9a88
1 parent 7744870
commit 09b9a88
Show file tree

Hide file tree

Showing 17 changed files with 194 additions and 46 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -120,6 +120,7 @@ target_sources(vk-gltf-viewer PRIVATE
         interface/gltf/AssetSceneGpuBuffers.cppm
         interface/gltf/AssetSceneHierarchy.cppm
         interface/gltf/MaterialVariantsMapping.cppm
+        interface/gltf/MeshWeights.cppm
         interface/helpers/AggregateHasher.cppm
         interface/helpers/concepts.cppm
         interface/helpers/fastgltf.cppm

diff --git a/impl/MainApp.cpp b/impl/MainApp.cpp
@@ -657,15 +657,15 @@ vk_gltf_viewer::MainApp::Gltf::Gltf(
     gpu { gpu },
     assetGpuBuffers { asset, gpu, threadPool, assetExternalBuffers },
     assetGpuTextures { asset, directory, gpu, threadPool, assetExternalBuffers },
-    sceneGpuBuffers { asset, scene, sceneHierarchy, gpu, assetExternalBuffers },
+    sceneGpuBuffers { asset, scene, sceneHierarchy, meshWeights, gpu, assetExternalBuffers },
     sceneMiniball { gltf::algorithm::getMiniball(asset, scene, [this](std::size_t nodeIndex, std::size_t instanceIndex) {
         return cast<double>(sceneGpuBuffers.getMeshNodeWorldTransform(nodeIndex, instanceIndex));
     }) } { }
 
 void vk_gltf_viewer::MainApp::Gltf::setScene(std::size_t sceneIndex) {
     scene = asset.scenes[sceneIndex];
     sceneHierarchy = { asset, scene };
-    sceneGpuBuffers = { asset, scene, sceneHierarchy, gpu, assetExternalBuffers };
+    sceneGpuBuffers = { asset, scene, sceneHierarchy, meshWeights, gpu, assetExternalBuffers };
     sceneMiniball = gltf::algorithm::getMiniball(asset, scene, [this](std::size_t nodeIndex, std::size_t instanceIndex) {
         return cast<double>(sceneGpuBuffers.getMeshNodeWorldTransform(nodeIndex, instanceIndex));
     });

diff --git a/impl/gltf/AssetGpuBuffers.cpp b/impl/gltf/AssetGpuBuffers.cpp
@@ -163,38 +163,20 @@ std::variant<vku::AllocatedBuffer, vku::MappedBuffer> vk_gltf_viewer::gltf::Asse
                 .pPositionMorphTargetAttributeMappingInfoBuffer = primitiveInfo.positionMorphTargetInfos.pMappingBuffer,
                 .pTexcoordAttributeMappingInfoBuffer = primitiveInfo.texcoordsInfo.pMappingBuffer,
                 .positionByteStride = primitiveInfo.positionInfo.byteStride,
-                .morphTargetCount = std::numeric_limits<std::uint32_t>::max(),
                 .materialIndex = primitiveInfo.materialIndex.transform(padMaterialIndex).value_or(0U),
             };
 
-            if (primitiveInfo.positionMorphTargetInfos.pMappingBuffer) {
-                gpuPrimitive.pPositionMorphTargetAttributeMappingInfoBuffer
-                    = primitiveInfo.positionMorphTargetInfos.pMappingBuffer;
-                gpuPrimitive.morphTargetCount = std::min<std::uint32_t>(
-                    gpuPrimitive.morphTargetCount,
-                    primitiveInfo.positionMorphTargetInfos.attributeInfos.size());
-            }
             if (primitiveInfo.normalInfo) {
                 gpuPrimitive.pNormalBuffer = primitiveInfo.normalInfo->address;
                 gpuPrimitive.normalByteStride = primitiveInfo.normalInfo->byteStride;
-                if (primitiveInfo.normalMorphTargetInfos.pMappingBuffer) {
-                    gpuPrimitive.pPositionMorphTargetAttributeMappingInfoBuffer
-                        = primitiveInfo.positionMorphTargetInfos.pMappingBuffer;
-                    gpuPrimitive.morphTargetCount = std::min<std::uint32_t>(
-                        gpuPrimitive.morphTargetCount,
-                        primitiveInfo.normalMorphTargetInfos.attributeInfos.size());
-                }
+                gpuPrimitive.pNormalMorphTargetAttributeMappingInfoBuffer
+                    = primitiveInfo.normalMorphTargetInfos.pMappingBuffer;
             }
             if (primitiveInfo.tangentInfo) {
                 gpuPrimitive.pTangentBuffer = primitiveInfo.tangentInfo->address;
                 gpuPrimitive.tangentByteStride = primitiveInfo.tangentInfo->byteStride;
-                if (primitiveInfo.tangentMorphTargetInfos.pMappingBuffer) {
-                    gpuPrimitive.pPositionMorphTargetAttributeMappingInfoBuffer
-                        = primitiveInfo.positionMorphTargetInfos.pMappingBuffer;
-                    gpuPrimitive.morphTargetCount = std::min<std::uint32_t>(
-                        gpuPrimitive.morphTargetCount,
-                        primitiveInfo.tangentMorphTargetInfos.attributeInfos.size());
-                }
+                gpuPrimitive.pTangentMorphTargetAttributeMappingInfoBuffer
+                    = primitiveInfo.tangentMorphTargetInfos.pMappingBuffer;
             }
             if (primitiveInfo.colorInfo) {
                 gpuPrimitive.pColorBuffer = primitiveInfo.colorInfo->address;

diff --git a/impl/gltf/AssetSceneGpuBuffers.cpp b/impl/gltf/AssetSceneGpuBuffers.cpp
@@ -44,13 +44,24 @@ std::vector<std::uint32_t> vk_gltf_viewer::gltf::AssetSceneGpuBuffers::createIns
     return result;
 }
 
-vku::AllocatedBuffer vk_gltf_viewer::gltf::AssetSceneGpuBuffers::createNodeBuffer(const vulkan::Gpu &gpu) const {
+vku::AllocatedBuffer vk_gltf_viewer::gltf::AssetSceneGpuBuffers::createNodeBuffer(
+    const fastgltf::Asset &asset,
+    const MeshWeights &meshWeights,
+    const vulkan::Gpu &gpu
+) const {
     const vk::DeviceAddress nodeTransformBufferStartAddress = gpu.device.getBufferAddress({ meshNodeWorldTransformBuffer });
 
     vku::AllocatedBuffer stagingBuffer = vku::MappedBuffer {
         gpu.allocator,
-        std::from_range, instanceOffsets | std::views::transform([=](std::uint32_t offset) {
-            return nodeTransformBufferStartAddress + sizeof(fastgltf::math::fmat4x4) * offset;
+        std::from_range, ranges::views::upto(asset.nodes.size()) | std::views::transform([&](std::size_t nodeIndex) {
+            return std::array {
+                to_optional(asset.nodes[nodeIndex].meshIndex)
+                    .transform([&](std::size_t meshIndex) {
+                        return meshWeights.segments[meshIndex].startAddress;
+                    })
+                    .value_or(vk::DeviceAddress { 0 }),
+                nodeTransformBufferStartAddress + sizeof(fastgltf::math::fmat4x4) * instanceOffsets[nodeIndex],
+            };
         }),
         gpu.isUmaDevice ? vk::BufferUsageFlagBits::eStorageBuffer : vk::BufferUsageFlagBits::eTransferSrc,
     }.unmap();

diff --git a/interface/MainApp.cppm b/interface/MainApp.cppm
@@ -10,6 +10,7 @@ import :gltf.AssetGpuFallbackTexture;
 import :gltf.AssetSceneGpuBuffers;
 import :gltf.AssetSceneHierarchy;
 import :gltf.MaterialVariantsMapping;
+import :gltf.MeshWeights;
 import :vulkan.dsl.Asset;
 import :vulkan.dsl.ImageBasedLighting;
 import :vulkan.dsl.Scene;
@@ -72,6 +73,8 @@ namespace vk_gltf_viewer {
             gltf::AssetGpuBuffers assetGpuBuffers;
             gltf::AssetGpuTextures assetGpuTextures;
 
+            gltf::MeshWeights meshWeights { asset, gpu };
+
             /**
              * @brief The glTF scene that is currently used by.
              *

diff --git a/interface/gltf/AssetGpuBuffers.cppm b/interface/gltf/AssetGpuBuffers.cppm
@@ -165,8 +165,8 @@ namespace vk_gltf_viewer::gltf {
             std::uint8_t colorComponentType;
             std::uint8_t colorComponentCount;
             char _padding0_[2];
-            std::uint32_t morphTargetCount;
             std::uint32_t materialIndex;
+            char _padding1_[4];
         };
         static_assert(sizeof(GpuPrimitive) == 80);
         static_assert(offsetof(GpuPrimitive, positionByteStride) == 64);
@@ -175,8 +175,7 @@ namespace vk_gltf_viewer::gltf {
         static_assert(offsetof(GpuPrimitive, colorByteStride) == 67);
         static_assert(offsetof(GpuPrimitive, colorComponentType) == 68);
         static_assert(offsetof(GpuPrimitive, colorComponentCount) == 69);
-        static_assert(offsetof(GpuPrimitive, morphTargetCount) == 72);
-        static_assert(offsetof(GpuPrimitive, materialIndex) == 76);
+        static_assert(offsetof(GpuPrimitive, materialIndex) == 72);
 
         std::unordered_map<const fastgltf::Primitive*, AssetPrimitiveInfo> primitiveInfos = createPrimitiveInfos();
 

diff --git a/interface/gltf/AssetSceneGpuBuffers.cppm b/interface/gltf/AssetSceneGpuBuffers.cppm
@@ -5,6 +5,7 @@ export import fastgltf;
 import :gltf.algorithm.traversal;
 export import :gltf.AssetPrimitiveInfo;
 export import :gltf.AssetSceneHierarchy;
+export import :gltf.MeshWeights;
 import :helpers.concepts;
 import :helpers.fastgltf;
 import :helpers.functional;
@@ -49,12 +50,13 @@ namespace vk_gltf_viewer::gltf {
             const fastgltf::Asset &asset [[clang::lifetimebound]],
             const fastgltf::Scene &scene [[clang::lifetimebound]],
             const AssetSceneHierarchy &sceneHierarchy,
+            const MeshWeights &meshWeights,
             const vulkan::Gpu &gpu [[clang::lifetimebound]],
             const BufferDataAdapter &adapter = {}
         ) : pAsset { &asset },
             instanceCounts { createInstanceCounts(scene) },
             meshNodeWorldTransformBuffer { createMeshNodeWorldTransformBuffer(scene, sceneHierarchy, gpu.allocator, adapter) },
-            nodeBuffer { createNodeBuffer(gpu) } { }
+            nodeBuffer { createNodeBuffer(asset, meshWeights, gpu) } { }
 
         /**
          * @brief Get world transform matrix of \p nodeIndex-th mesh node's \p instanceIndex-th instance in the scene.
@@ -215,6 +217,10 @@ namespace vk_gltf_viewer::gltf {
             };
         }
 
-        [[nodiscard]] vku::AllocatedBuffer createNodeBuffer(const vulkan::Gpu &gpu) const;
+        [[nodiscard]] vku::AllocatedBuffer createNodeBuffer(
+            const fastgltf::Asset &asset,
+            const MeshWeights &meshWeights,
+            const vulkan::Gpu &gpu
+        ) const;
     };
 }
diff --git a/interface/gltf/MeshWeights.cppm b/interface/gltf/MeshWeights.cppm
@@ -0,0 +1,107 @@
+export module vk_gltf_viewer:gltf.MeshWeights;
+
+import std;
+export import fastgltf;
+import :helpers.ranges;
+export import :vulkan.Gpu;
+
+#define FWD(...) static_cast<decltype(__VA_ARGS__)&&>(__VA_ARGS__)
+#define LIFT(...) [&](auto &&...xs) { return __VA_ARGS__(FWD(xs)...); }
+
+template <typename T>
+class writeonly {
+    T volatile *addr;
+
+public:
+    explicit writeonly(std::uintptr_t addr)
+        : addr { reinterpret_cast<T*>(addr) } { }
+
+    void operator=(const T &t) volatile { *addr = t; }
+
+    [[nodiscard]] std::uintptr_t address() const noexcept { return addr; }
+};
+
+/**
+ * @brief Create a combined buffer from given segments (a range of byte data) and return each segments' start offsets.
+ *
+ * Example: Two segments { 0xAA, 0xBB, 0xCC } and { 0xDD, 0xEE } will be combined to { 0xAA, 0xBB, 0xCC, 0xDD, 0xEE }, and their start offsets are { 0, 3 }.
+ *
+ * @tparam R Range type of data segments.
+ * @param allocator VMA allocator to allocate the buffer.
+ * @param segments Range of data segments. Each segment will be converted to <tt>std::span<const std::byte></tt>, therefore segment's elements must be trivially copyable.
+ * @param usage Usage flags of the result buffer.
+ * @return Pair of buffer and each segments' start offsets vector.
+ */
+template <std::ranges::forward_range R>
+    requires (
+        std::ranges::sized_range<std::ranges::range_value_t<R>>
+        && std::same_as<std::ranges::range_value_t<std::ranges::range_value_t<R>>, std::byte>)
+[[nodiscard]] std::pair<vku::MappedBuffer, std::vector<vk::DeviceSize>> createCombinedBuffer(
+    vma::Allocator allocator,
+    R &&segments,
+    vk::BufferUsageFlags usage
+) {
+    // Calculate each segments' copy destination offsets.
+    std::vector<vk::DeviceSize> copyOffsets { std::from_range, segments | std::views::transform(LIFT(std::size)) };
+    vk::DeviceSize sizeTotal = copyOffsets.back();
+    std::exclusive_scan(copyOffsets.begin(), copyOffsets.end(), copyOffsets.begin(), vk::DeviceSize { 0 });
+    sizeTotal += copyOffsets.back();
+
+    if (sizeTotal == 0) {
+        throw std::invalid_argument { "No data to write" };
+    }
+
+    // Create buffer.
+    vku::MappedBuffer buffer { allocator, vk::BufferCreateInfo { {}, sizeTotal, usage } };
+
+    // Copy segments to the buffer.
+    std::byte *mapped = static_cast<std::byte*>(buffer.data);
+    for (auto &&segment : FWD(segments)) {
+        mapped = std::ranges::copy(FWD(segment), mapped).out;
+    }
+
+    return { std::move(buffer), std::move(copyOffsets) };
+}
+
+namespace vk_gltf_viewer::gltf {
+    class MeshWeights {
+    public:
+        struct Segment {
+            vk::DeviceAddress startAddress;
+            writeonly<std::uint32_t> count;
+            std::span<writeonly<float>> weights;
+        };
+
+        std::vector<Segment> segments;
+
+        MeshWeights(const fastgltf::Asset &asset, const vulkan::Gpu &gpu [[clang::lifetimebound]])
+            : buffer { createBuffer(asset, gpu) } { }
+
+    private:
+        vku::MappedBuffer buffer;
+
+        [[nodiscard]] vku::MappedBuffer createBuffer(const fastgltf::Asset &asset, const vulkan::Gpu &gpu) {
+            auto [buffer, copyOffsets] = createCombinedBuffer(
+                gpu.allocator,
+                asset.meshes | std::views::transform([](const fastgltf::Mesh &mesh) {
+                    // [count, weight0, weight1, ..., weight(count-1)]
+                    return ranges::views::concat(
+                        std::bit_cast<std::array<std::byte, 4>>(static_cast<std::uint32_t>(mesh.weights.size())),
+                        as_bytes(std::span { mesh.weights }));
+                }),
+                vk::BufferUsageFlagBits::eStorageBuffer | vk::BufferUsageFlagBits::eShaderDeviceAddress);
+
+            segments.reserve(copyOffsets.size());
+            const vk::DeviceAddress bufferAddress = gpu.device.getBufferAddress({ buffer });
+            for (const auto &[mesh, copyOffset] : std::views::zip(asset.meshes, copyOffsets)) {
+                const std::uintptr_t segmentStart = reinterpret_cast<std::uintptr_t>(buffer.data) + copyOffset;
+                segments.emplace_back(
+                    bufferAddress + copyOffset,
+                    writeonly<std::uint32_t> { segmentStart },
+                    std::span { reinterpret_cast<writeonly<float>*>(segmentStart + sizeof(float)), mesh.weights.size() });
+            }
+
+            return std::move(buffer);
+        }
+    };
+}
diff --git a/shaders/depth.vert b/shaders/depth.vert
@@ -12,7 +12,13 @@
 #include "indexing.glsl"
 #include "types.glsl"
 
-layout (std430, buffer_reference, buffer_reference_align = 64) readonly buffer Node { mat4 transforms[]; };
+layout (buffer_reference, buffer_reference_align = 4) readonly buffer MeshWeights { uint count; float[] weights; };
+layout (buffer_reference, buffer_reference_align = 64) readonly buffer InstanceTransforms { mat4 data[]; };
+
+struct Node {
+    MeshWeights meshWeights;
+    InstanceTransforms instanceTransforms;
+};
 
 layout (location = 0) flat out uint outNodeIndex;
 

diff --git a/shaders/indexing.glsl b/shaders/indexing.glsl
@@ -10,7 +10,7 @@
 #define PRIMITIVE_INDEX gl_BaseInstance & 0xFFFFU
 #define PRIMITIVE primitives[PRIMITIVE_INDEX]
 #define NODE_INDEX gl_BaseInstance >> 16U
-#define TRANSFORM Node(nodes[NODE_INDEX]).transforms[gl_InstanceIndex - gl_BaseInstance]
+#define TRANSFORM nodes[NODE_INDEX].instanceTransforms.data[gl_InstanceIndex - gl_BaseInstance]
 #define MATERIAL_INDEX PRIMITIVE.materialIndex
 #define MATERIAL materials[MATERIAL_INDEX]
 

diff --git a/shaders/jump_flood_seed.vert b/shaders/jump_flood_seed.vert
@@ -12,7 +12,13 @@
 #include "indexing.glsl"
 #include "types.glsl"
 
-layout (std430, buffer_reference, buffer_reference_align = 64) readonly buffer Node { mat4 transforms[]; };
+layout (buffer_reference, buffer_reference_align = 4) readonly buffer MeshWeights { uint count; float[] weights; };
+layout (buffer_reference, buffer_reference_align = 64) readonly buffer InstanceTransforms { mat4 data[]; };
+
+struct Node {
+    MeshWeights meshWeights;
+    InstanceTransforms instanceTransforms;
+};
 
 layout (set = 0, binding = 0) readonly buffer PrimitiveBuffer {
     Primitive primitives[];

diff --git a/shaders/mask_depth.vert b/shaders/mask_depth.vert
@@ -14,7 +14,13 @@
 
 #define HAS_VARIADIC_OUT HAS_BASE_COLOR_TEXTURE || HAS_COLOR_ALPHA_ATTRIBUTE
 
-layout (std430, buffer_reference, buffer_reference_align = 64) readonly buffer Node { mat4 transforms[]; };
+layout (buffer_reference, buffer_reference_align = 4) readonly buffer MeshWeights { uint count; float[] weights; };
+layout (buffer_reference, buffer_reference_align = 64) readonly buffer InstanceTransforms { mat4 data[]; };
+
+struct Node {
+    MeshWeights meshWeights;
+    InstanceTransforms instanceTransforms;
+};
 
 layout (location = 0) flat out uint outNodeIndex;
 layout (location = 1) flat out uint outMaterialIndex;

diff --git a/shaders/mask_jump_flood_seed.vert b/shaders/mask_jump_flood_seed.vert
@@ -14,7 +14,13 @@
 
 #define HAS_VARIADIC_OUT HAS_BASE_COLOR_TEXTURE || HAS_COLOR_ALPHA_ATTRIBUTE
 
-layout (std430, buffer_reference, buffer_reference_align = 64) readonly buffer Node { mat4 transforms[]; };
+layout (buffer_reference, buffer_reference_align = 4) readonly buffer MeshWeights { uint count; float[] weights; };
+layout (buffer_reference, buffer_reference_align = 64) readonly buffer InstanceTransforms { mat4 data[]; };
+
+struct Node {
+    MeshWeights meshWeights;
+    InstanceTransforms instanceTransforms;
+};
 
 layout (location = 0) flat out uint outMaterialIndex;
 #if HAS_VARIADIC_OUT

diff --git a/shaders/primitive.vert b/shaders/primitive.vert
@@ -14,7 +14,13 @@
 
 #define HAS_VARIADIC_OUT !FRAGMENT_SHADER_GENERATED_TBN || TEXCOORD_COUNT >= 1 || HAS_COLOR_ATTRIBUTE
 
-layout (std430, buffer_reference, buffer_reference_align = 64) readonly buffer Node { mat4 transforms[]; };
+layout (buffer_reference, buffer_reference_align = 4) readonly buffer MeshWeights { uint count; float[] weights; };
+layout (buffer_reference, buffer_reference_align = 64) readonly buffer InstanceTransforms { mat4 data[]; };
+
+struct Node {
+    MeshWeights meshWeights;
+    InstanceTransforms instanceTransforms;
+};
 
 layout (location = 0) out vec3 outPosition;
 layout (location = 1) flat out uint outMaterialIndex;

diff --git a/shaders/types.glsl b/shaders/types.glsl
@@ -63,8 +63,8 @@ struct Primitive {
     uint8_t colorComponentType;
     uint8_t colorComponentCount;
     uint8_t _padding0_[2];
-    uint morphTargetCount;
     uint materialIndex;
+    float _padding1_;
 };
 
 #endif

diff --git a/shaders/unlit_primitive.vert b/shaders/unlit_primitive.vert
@@ -14,7 +14,13 @@
 
 #define HAS_VARIADIC_OUT HAS_BASE_COLOR_TEXTURE || HAS_COLOR_ATTRIBUTE
 
-layout (std430, buffer_reference, buffer_reference_align = 64) readonly buffer Node { mat4 transforms[]; };
+layout (buffer_reference, buffer_reference_align = 4) readonly buffer MeshWeights { uint count; float[] weights; };
+layout (buffer_reference, buffer_reference_align = 64) readonly buffer InstanceTransforms { mat4 data[]; };
+
+struct Node {
+    MeshWeights meshWeights;
+    InstanceTransforms instanceTransforms;
+};
 
 layout (location = 0) flat out uint outMaterialIndex;
 #if HAS_VARIADIC_OUT