- Split core resource architecture into Core/Asset/ (IResource, ResourceManager, ResourceCache, etc.)
- Moved IO layer into Core/IO/ (IResourceLoader, ResourceFileSystem, etc.)
- Reorganized concrete resource types into Resources/{Texture,Mesh,Material,Shader,AudioClip}/
- Updated all include paths from relative to <XCEngine/...> format
- Fixed circular dependency in Material.h (removed unnecessary ResourceManager.h include)
- Fixed malformed include syntax in ResourceManager.h and AsyncLoader.h
- Fixed glad.h path issues in CMakeLists.txt
- Create unified RHIResourceView base interface with type-specific Initialize methods
- Implement D3D12ResourceView with RTV/DSV/SRV/UAV/CBV support
- Implement OpenGL ResourceView simulation layer using FBO, texture units, and UBO
- Add OpenGLTextureUnitAllocator for managing texture unit bindings
- Add OpenGLUniformBufferManager for UBO binding points
- Add OpenGLFramebuffer for FBO management
- Remove deprecated D3D12 view classes (RenderTargetView, DepthStencilView, etc.)
- Fix ExecuteCommandLists type confusion bug by adding GetNativeHandle to RHICommandList
- Fix test bug where Close() was called before ExecuteCommandLists
- Update all integration tests to use new D3D12ResourceView class
- All tests pass: 144 unit tests + 8 integration tests
GetCompletedValue() semantics: returns the highest confirmed completed value,
not the last Signal() target value. When Signal(5) -> Signal(3),
GetCompletedValue() should still return >= 5 since GPU already completed that.
- Remove IsSignaled() from RHIFence interface (semantic inconsistency)
- Remove Reset() from OpenGL implementation (no D3D12 counterpart)
- OpenGL Fence now uses single GLsync + CPU counters for timeline simulation
- OpenGL Fence Initialize() now accepts uint64_t initialValue (was bool)
- Add comprehensive timeline semantics tests for all backends:
- Signal increment/decrement scenarios
- Multiple signals
- Wait smaller than completed value
- GetCompletedValue stages verification
- Update documentation to reflect actual implementation
- Add m_boundRenderTargets and m_boundDepthStencil members to D3D12CommandList
- SetRenderTargetsHandle now saves bound RTV handles for later clear operations
- ClearRenderTargetView(ID3D12Resource*, ...) clears all bound render targets
- Reset() clears bound targets list
- Follows D3D12 best practice: RTV is bound via OMSetRenderTargets, then cleared
- CommandQueue::ExecuteCommandLists: fix type conversion from void** to ID3D12CommandList**, reset command list before execution, add null checks
- SwapChain::ShouldClose: add m_shouldClose member and implement getter/setter
- SwapChain::SetFullscreen: add m_fullscreen member to track state locally
- triangle/quad/sphere tests: Update to use new Initialize signature with OpenGLDevice*
- triangle/quad/sphere tests: Use GetGLContext() for RenderDoc SetDevice
- triangle/quad/sphere tests: Add wglMakeCurrent in render loop
- triangle/quad/sphere tests: Update capture timing to frame 25-35
- unit test fixture: Add GetDevice() method
- unit test swap_chain: Update to use new Initialize signature
- Add default value handling for sampleCount, mipLevels, depth in CreateTexture
- Add RTV descriptor heap to D3D12CommandList for future ClearRenderTarget implementation
- ClearRenderTarget remains stub (requires full D3D12 render pipeline state)
- OpenGLSwapChain now gets HDC from OpenGLDevice instead of creating its own
- Renamed OpenGLDevice::GetContext() to GetGLContext() for clarity
- Renamed OpenGLDevice::GetDC() to GetPresentationDC() for clarity
- OpenGLDevice now owns the HDC/HGLRC lifecycle
- OpenGLSwapChain::Initialize() now takes OpenGLDevice* parameter
- OpenGLSwapChain::Present() uses device's HDC for SwapBuffers
- Updated minimal test to use new API and capture from frame 25-35
- RenderDoc SetDevice now uses GetGLContext() for proper OpenGL hook
- Add RenderDocCapture include
- Initialize RenderDocCapture before D3D12 device creation
- SetDevice after D3D12 device initialization
- TriggerCapture on frame 29 before Present
- Add RenderDocCapture::Get().Shutdown() on exit
- Remove BeginCapture/EndCapture complexity for manual frame capture
- Use TriggerCapture for single frame capture (simpler API)
- Move TriggerCapture before Present to ensure frame is captured
- Add SetForegroundWindow/SetFocus before TriggerCapture
- Remove default capture path in Initialize to allow SetCaptureFilePath to work
- Add detailed RenderDocCapture API reference in Debug section
- Note OpenGL requires HGLRC (not HDC) as device pointer
- Update RenderDoc example description to mention both D3D12 and OpenGL support
- Add note about renderdoc.dll location
- Convert from MCP protocol layer to direct Python function calls
- 42 functions across 9 modules: session, event, pipeline, resource, data, shader, advanced, performance, diagnostic
- Requires Python 3.6 (renderdoc.pyd is compiled for Python 3.6)
- Fix renderdoc API calls: GetColorBlends, GetStencilFaces, GetViewport(i), GetScissor(i)
- Remove Python 3.10+ type annotations for Python 3.6 compatibility
- Add README.md with full API documentation
- Includes test.py for basic smoke testing
- OpenGL test: Use BeginCapture/EndCapture with bool return values
- D3D12 test: Use BeginCapture/EndCapture with bool return values
- Both tests: Use GetCapture() to log capture file info
- Added capture file path and size logging after capture
Critical fix: For OpenGL, RenderDoc requires HGLRC (OpenGL context) as
the device pointer, not HDC. Previously OpenGL test incorrectly passed
HDC which caused capture to silently fail (NumCaptures=0).
Changes:
- OpenGLDevice: Add wglCreateContextAttribsARB support for debug context
- OpenGLDevice: Create OpenGL 4.6 core profile with debug flag
- RenderDocCapture: Pass correct window to SetActiveWindow/StartFrameCapture
- OpenGL test: Pass HGLRC via SetDevice(), use BeginCapture/EndCapture
Fix root cause identified via RenderDoc docs analysis:
'For OpenGL it must be the HGLRC, GLXContext, or EGLContext'
- Add SetDevice() and GetNumCaptures() methods to RenderDocCapture
- Fix initialization order: RenderDoc before D3D12, then SetDevice after
- Use nullptr for window param in StartFrameCapture/EndFrameCapture
- Add BeginCapture at frame 29, EndCapture at frame 30
- Set capture file path template to .\minimal_frame30
- Copy renderdoc.dll (26MB redistributable) to test output
- Engine/third_party/renderdoc/ now contains working renderdoc.dll
- Add RenderDocCapture class for dynamic loading of renderdoc.dll
- Support BeginCapture/EndCapture/TriggerCapture APIs
- Add RenderDoc log category
- Add unit tests for RenderDocCapture in tests/debug
- Rename D3D12Enum.h to D3D12Enums.h for naming consistency
- Fix OpenGL unit test GLAD initialization by using gladLoadGL()
instead of gladLoadGLLoader(wglGetProcAddress) for fallback support
- Migrate remaining tests to use gtest_discover_tests for granular
test discovery (math, core, containers, memory, threading, debug,
components, scene, resources, input, opengl)
- Remove obsolete TEST_RESOURCES_DIR and copy_directory commands
from OpenGL unit test CMakeLists (minimal/Res doesn't exist)
- Update TEST_SPEC.md with performance metrics and per-module
build/test commands for faster development workflow
- Update CMake path references to use lowercase paths
Replace add_test() with gtest_discover_tests() to enable CTest to
discover all 55 individual Google Test cases instead of treating them
as a single test. This improves test reporting granularity from 1/1
to 55/55 tests.
