- Add new malloc_alignedN() functions for easier dynamic allocation of aligned memory blocks.
- Rework buffer allocations using the new malloc_alignedN() functions.
- To enable SSSE3, also require ENABLE_SSE2 and ENABLE_SSE3.
- Add some more SSE2/SSSE3 optimizations.
- CACHE_ALIGN and malloc_alignedCacheLine() now set 64 byte alignment on 64-bit systems.
- Do a bunch more code cleanup.
- Fix compiling for CLI port. (Regression from r5198.)
- Fix issue with GTK port where the video output framebuffer wasn't getting cleared on reset. (Regression from r5198.)
- Fix bug where the depth LUT wasn’t being generated correctly, causing the clear image depth buffer to malfunction. (Regression from r5187.)
- In SoftRasterizer, obsolete GFX3D_Zelda_Shadow_Depth_Hack for depth-equals tests. We’re now using a fixed tolerance of +/-0x200, according to GBATEK.
- In SoftRasterizer, z-depth is now calculated using the depth LUT instead of with << 9. This spreads the depth value more evenly across the range of [0 - 0x00FFFFFF]. This change will need additional testing.
- Do some small optimizations to SoftRasterizer.
- Do more code cleanup.
- In SoftRasterizer, do multithreading optimization for the fog and edge mark pass. This involved a change to the edge marking algorithm, so this will need additional testing.
- Fix bug where a SoftRasterizer renderer object wouldn’t get destroyed properly. (Regression from r5187.)
- Fix bug where the user wasn’t able to switch between different threaded versions of SoftRasterizer. (Regression from r5187.)
- Fix a potential bug that might occur if an OpenGL renderer object failed to create. (Regression from r5188.)
- Upload the toon table through the render state UBO instead of through a 1D texture. (OpenGL 3.2 only.)
- Small optimization in the edge mark fragment shader.
- Fix compiling issues. (Regression from r5162. Fixes bug #1468.)
- Fix crashing issue when selecting the OpenGL 3.2 renderer. Failure to init should now fallback properly. (Regression from r5180. Fixes bug #1470.)
- In the OpenGL renderer, do better handling of the geometry index buffer, and also load its data in OpenGLRenderer::BeginRender().
- In the OpenGL renderer, remove a bunch of extraneous binds.
- Fix bug in SoftRasterizer where clear-image depth wasn’t being written correctly. (Regression from r5176.)
- Reduce the buffer sizes in the core 3D engine.
- Do even more refactoring.
- In the OpenGL renderer, optimize framebuffer clearing (OpenGL v3.2 only).
- In SoftRasterizer, multithread the rendering state setup (requires at least 4 threads).
- Do more code refactoring.
- In the OpenGL renderer, optimize edge marking performance.
- In SoftRasterizer, fix a bug where edge marking and fog weren’t being drawn if multithreading was off.
- In the 3D Rendering Settings panel in the DeSmuME Preferences view, move the Enable Edge Marking and Enable Fog checkboxes to the General Settings section.
- Update tooltips to reflect the new behavior.
- Bring back vertex draw batching from r4522, fixing the bug that caused it to fail on Metroid Prime Hunters. This gives a small performance improvement for users with older drivers.
rogerman