feature was originally added via PR #764 to the gtk3 frontend.
this makes it possible to run the scaling on the GPU, avoiding the
incredibly slow software scaling that's otherwise done via cairo
when view->window size is set to anything > 1.0.
note that the window size "scale" needs to be identical to the
chosen GPU scale factor, otherwise software scaling kicks in again.
unlike the scale setting in the CLI port, which simply upscales the
native NDS framebuffer in hardware, this setting scales up even
the actual 3D textures, resulting in a sharper image, at the cost
of higher CPU/GPU usage. a game using demanding 3D scenes, like
zelda phantom hourglass' intro scene, may be able to still trash
the FPS.
the original PR also reported issues when setting the GPU scale to
a fraction, therefore the increments are currently locked to 1.0.
until now, the boost key was hardcoded to 'o' even when the config
said otherwise, and not treated in case of a joypad at all.
when triggered with the o key, it even behaved differently than
the gtk ui - the boost wasn't released together with the key,
but only when pressed again, so it was more like a shortcut for
"disable fps limiter".
this change implements the desired outcome of the second part of PR #822,
but without introducing more hacks and relying on magic values.
closes#822
because the GTK frontends use GDK keysyms, not SDL ones, - the former
are being stored in the config file and used by the GTK ui) - a temporary
workaround was put into place 14 years ago: the loaded config values
were simply being overwritten with the hardcoded defaults.
this commit removes the overriding of the config, and introduces
a cli frontend specific section "SDLKEYS", which is written by the
GTK2 frontend upon a configuration change.
it tries to convert the GDK keycodes into SDL2 ones while doing so.
an alternative solution (involving less code changes) would have been
to do the conversion in the cli frontend, but that would require having
the gdk header available for compilation, which may not be the case
if the user only wants the cli frontend. such a user could now create
the config file on another machine with the GTK frontend, or simply
manually take the desired values from the SDL_keycode.h header.
this change is instigated by one of the changes in PR #822, which simply
removed the workaround and kept parsing on error, which mitigated the
problem for some keys, but not all.
[GTK*] Some gamepad input rework
- Unbind gamepad keys by default. The default bindings may fit one gamepad model but work weirdly with another causing issues like [LINUX/GTK3] Whitescreen freeze when pressing square on ds4 #834
- Use non-blocking method to obtain gamepad keys/axes during configuration to avoid visible emulator freeze and possible deadlock (see [linux] editing controlls sometimes (often) freezes the emulator #843)
closes#834closes#843
* GTK cheats UI inputs/displays hex for address (offset)
* Added range bound to keep internal cheat data within specified size
The decision to change the value column to store G_TYPE_UINT is purely pragmatic--having all data be treated and displayed as unsized is simpler than writing a custom display to handle signedness and cleaner than 1-3 byte ints appearing unsigned and 4 byte int appearing signed.
* Added index and cheat type data to ListStore
My implementation plan is to use a GtkTreeModelFilter to create separate section for internal and ActionReplay cheats, but I was not convinced the indices in the tree path would correspond to the indices in , hence the additional column.
* Filter raw and AR cheats, display both filters in UI, patch up raw update/delete
* Action Replay UI elements [GTK]
* Memory leak fixes
+ some additional clean-up comments and small bug patches
* Backport to GTK2
- Using the new OpenGLGeometryResource and OpenGLRenderStatesResource classes, the 3.2 Core Profile and ES renderers now use triple-buffering for all geometry rendering resources and framebuffer constants.
- Delete the InitFinalRenderStates() method, which has been obsoleted over the years. Its functionality has been rolled into the InitExtensions() and _RenderGeometryLoopBegin() methods.
- In the 3.2 Core Profile and ES renderers, synchronization of vertex info uploads now occurs at a more reasonable time.
- Geometry rendering is now Y-flipped by default, eliminating the need to Y-flip the final output framebuffer under all conditions.
- Legacy OpenGL no longer needs to perform any color conversion of the final output framebuffer if FBOs are supported.
- Rename some functions and variables to better describe what things are doing now.
- 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.
Max Fedotov