- Remove some methods in OpenGLRenderer_2_0 and OpenGLRenderer_2_1 that have negligible contribution to either performance or code simplicity.
- OpenGLRenderer_1_2, OpenGLRenderer_2_0, and OpenGLRenderer_2_1 are now instantiated with their specific variant IDs.
- Calls to malloc_alignedCacheLine() have been replaced with the more appropriate malloc_alignedPage().
- Do some misc. code cleanup.
- Also fixes build errors related to explicitly building SIMD files for colorspacehandler. Don't do it! The proper way is to simply include "colorspacehandler.cpp" alone and let the compiler's preprocessor macros determine which SIMD file to use.
- Release builds now use -Ofast optimization instead of -O3.
- Release builds no longer strip their symbols. This will be needed for profiling and debugging.
- These changes now presume that standard OpenGL and OpenGL ES are mutually exclusive. We will NOT support running a standard OpenGL context and an OpenGL ES context in the same process.
- Clients must explicitly request supporting the OpenGL 3D renderer in their build configuration. Build configurations must define ENABLE_OPENGL_STANDARD (replacing the HAVE_OPENGL macro) or ENABLE_OPENGL_ES. If neither macro is defined, then the OpenGL 3D renderer will be assumed unavailable.
- Meson and Autotools now use better header/library checks for OpenGL functionality with their associated context type.
- Add a new Code::Blocks project file that can make builds for CLI, GTK, and GTK2.
- GTK and GTK2 ports now have the option to run a legacy OpenGL context, a 3.2 Core Profile context, or simply choosing the best context automatically like before.
- GTK and GTK2 ports have GLX and EGL as new context types. OSMesa and SDL have been updated to the latest design pattern.
- GTK and GTK2 ports can now be configured (via Meson or Autotools) to use a GLX, OSMesa, EGL, or SDL context.
- OSMesa contexts are now marked as deprecated. I don't know of anyone who still uses them, and I've never been able to get it to work correctly for years. Now that we have GLX contexts for compatibility purposes, OSMesa contexts are now completely redundant.
- Fix a bug with the GTK port where ancient GPUs without FBO support can still run framebuffers at custom sizes.
- For POSIX ports, move all "avout" and context creation files to the "shared" directory for better file organization.
- Fix a compiling bug for ES due to missing tokens.
- Fix Fog and Edge Mark feature availability when running legacy OpenGL. (Regressions from commit 0c7cb99 and commit 8b5ac56.)
- glDrawBuffer() now determines its own algorithm at runtime instead of at compile time. This change is being done to be consistent with all of the other Standard vs ES changes.
- Respect the draw buffers ordering rules that are unique to ES.
- Report the actual OpenGL variant being requested when initializing the renderer.
- Add oglrender_deinit() function pointer so that clients can properly handle the destruction of their associated context resources at the correct time.
- Certain internal OpenGL info is now assigned at run time instead of at compile time. This change now allows any of the OpenGL renderers to run side-by-side.
- Fix a potential unaligned access crashing bug in ES when clear images are to be rendered.
- Fix an ES bug where clear images would fail to render when MSAA is enabled.
- Fix a legacy OpenGL bug where toon table colors were not ignoring their alpha bit, according to GBATEK.
- These graphical glitches are resolved only when running 3.2 Core Profile with the GL_ARB_sample_shading extension, available on all modern GPUs.
- Do some minor optimizations to the Edge Mark and Fog shaders.
- Also fix an FBO attachment bug in legacy OpenGL that was introduced in commit 8b5ac56.
- The output framebuffers now bind their own FBOs rather than changing draw targets with glDrawBuffer().
- Rework the general FBO management.
- Legacy OpenGL now outputs native RGBA color if FBOs are supported. This should give a minor performance increase on older GPUs.
- The fixed-function pipeline can now flip the framebuffer on GPU. This greatly reduces the CPU usage when doing the final color conversion and gives a significant performance increase on ancient GPUs.
- There is no more need to switch to the working texture as the destination for the fog output. This change will be essential for future commits.
- The dual-source blending method has been obsoleted and removed.
- FBOs are no longer required for the fog feature, easing requirements for ancient GPUs.
- Ancient GPUs may see a small performance benefit due to shader simplification.
- Centralize all header includes into OGLDisplayOutput.h.
- Update all extension versions of OpenGL functions/tokens that would be core in OpenGL v1.5.
- Change the remaining FBO attachments that I missed the last time to use RGBA8 internal format.
- Fix picky ancient drivers that won't accept GL_RED as a texture internal format. It has been changed to GL_LUMINANCE.
- Fix picky ES drivers that demand that the external format used in both glTexImage2D() and glTexSubImage2D() are exactly the same.
- OpenGL ES doesn't support GL_UNSIGNED_SHORT_1_5_5_5_REV for texture data, and GL_UNSIGNED_SHORT_5_5_5_1 is incompatible with our data. So instead, all 16-bit data will be converted to 32-bit LE before uploading it via the textures, and such ES textures will now take GL_UNSIGNED_BYTE format.
- Remove the #include for EGL/egl.h in OGLRender.h, since EGL shouldn't exist at this level of the code stack.
- ColorspaceConvert5551To8888()
- ColorspaceConvert5551To6665()
- ColorspaceConvertBuffer5551To8888()
- ColorspaceConvertBuffer5551To6665()
- Also rename the existing 16-bit color conversion functions to help further distinguish the functions from one another.
- glTexParameteri() with GL_TEXTURE_2D_MULTISAMPLE throws INVALID_ENUM when trying to assign a sampler-related state, so these calls have been removed.
- glMapBufferRange() throws GL_INVALID_OPERATION if the buffer size is 0, so check for this condition first.
- Any ES-specific error messages are now reported as "OpenGL ES" instead of "OpenGL".
- Explicitly declare the internal format of all textures and RBOs used as FBO attachments as "GL_RGBA8" instead of the generic "GL_RGBA". This may fix some extremely picky ES drivers that would throw GL_INVALID_ENUM if we don't use a sized internal format.
- Hard code the read pixels format as GL_RGBA and the read data type as GL_UNSIGNED_BYTE. It is meaningless to rely on GL_IMPLEMENTATION_COLOR_READ_FORMAT and GL_IMPLEMENTATION_COLOR_READ_TYPE since we are reading from our own FBOs that are hard coded to GL_RGBA8 format.
- Require that OpenGL ES contexts support the GL_OES_surfaceless_context extension, since we are doing offscreen rendering strictly on FBOs only. This extension has been around for a very long time, and so this serves as a kind of compatibility check.
rogerman