#pragma once #include #include #include #include #include #include #include "RenderCommands3D.h" namespace ava2 { // 48-byte vertex; the GPU expands the packed fields to the same float/uint shader inputs. struct HardwareVertex { float position[4]; // FLOAT4 (raw-vertex batches may carry w != 1) uint8_t color[4]; // UBYTE4_NORM, RGBA float uv[2]; // FLOAT2 int16_t normal[4]{}; // SHORT4_NORM, unit xyz; w=1 when the vertex has a normal uint8_t bones[4]{}; // UBYTE4 (palette holds at most 128 bones) uint16_t weights[4]{}; // USHORT4_NORM }; static_assert(sizeof(HardwareVertex)==48); struct HardwareTransform { float world[16]{}, mvp[16]{}; float diffuse[4]{}, ambient[4]{}, emissive[4]{}; float lightDirection[4]{}, lightAmbient[4]{}, lightDiffuse[4]{}; float fogColor[4]{}, fogParams[4]{}; float modes[4]{}; // raw vertices, lighting, fog type, disable depth float skin[4]{}; // palette count }; static_assert(sizeof(HardwareTransform)==288); HardwareVertex MakeHardwareVertex(const Render3DDraw& mesh, size_t index); struct HardwareBatch { uint32_t firstIndex, indexCount; SDL_GPUTexture* texture; uint32_t cull; bool depth, depthWrite, blend, additive; float alphaRef; HardwareTransform transform{}; std::shared_ptr geometry; std::shared_ptr> palette; }; class HardwareScene { public: explicit HardwareScene(SDL_GPUDevice* device): device_(device) {} ~HardwareScene(); SDL_GPUTexture* FindTexture(const std::string& key); void SetTextureBudget(size_t bytes) { textureBudget_=bytes; } size_t GetTextureBytes() const { return textureBytes_; } size_t GetTextureUploadCount() const { return textureUploads_; } SDL_GPUTexture* Texture(const std::string& key, int width, int height, const void* rgba); bool Render(SDL_Renderer* renderer, int width, int height, const std::vector& vertices, const std::vector& indices, const std::vector& batches); size_t GetResidentUploadCount() const { return residentUploads_; } size_t GetLastDrawCallCount() const { return lastDrawCalls_; } size_t GetLastDynamicUploadBytes() const { return lastDynamicBytes_; } private: struct ResidentMesh { std::shared_ptr source; SDL_GPUBuffer *vertices=nullptr, *indices=nullptr; uint32_t count=0, bytes=0; uint64_t lastUsed=0; }; ResidentMesh* Resident(const std::shared_ptr& geometry); std::unordered_map resident_; uint64_t frame_=0; size_t residentBytes_=0, residentUploads_=0, lastDrawCalls_=0, lastDynamicBytes_=0; SDL_GPUBuffer* instances_=nullptr; SDL_GPUTransferBuffer* instanceTransfer_=nullptr; uint32_t instanceBytes_=0; bool Initialize(SDL_Renderer* renderer, int width, int height); SDL_GPUGraphicsPipeline* Pipeline(const HardwareBatch& batch); SDL_GPUDevice* device_; SDL_GPUShader *vertex_=nullptr, *fragment_=nullptr; SDL_GPUSampler* sampler_=nullptr; SDL_GPUTexture *color_=nullptr, *depth_=nullptr; SDL_Texture* output_=nullptr; SDL_GPUTexture* white_=nullptr; struct TextureEntry {SDL_GPUTexture* texture;size_t bytes;uint64_t lastUsed;}; std::unordered_map textures_; size_t textureBytes_=0,textureUploads_=0,textureBudget_=256u*1024u*1024u; SDL_GPUBuffer *vertices_=nullptr, *indices_=nullptr; SDL_GPUTransferBuffer* transfer_=nullptr; uint32_t vertexBytes_=0, indexBytes_=0; int width_=0,height_=0; std::unordered_map pipelines_; }; }