Gambatte 0.4.1 released

[Robert]

GameBoy / GameBoy Color emulator

 

New release. It contains a few important bug fixes, such as MBC2 ROMs not getting any SRAM in 0.4.0. It also contains some major audio/video timing improvements, as well as improvements and additions to platform-specific audio/video back-ends (particularly on win32).

--------------------------------------------------------------------------------

Changes:

libgambatte:

- Fix HqXx filter pitch.

- Fix mbc2 not getting a rambank.

- Make sure to reset passed pointers when deleted. Fixes potential crash when loading ROM during OAM busy.

common:

- Substantially improved rate estimation averaging.

- RateEst: Add a convenient way of filtering measures that extend beyond a buffer time, and are as such probably invalid.

- RateEst: Allow using a custom timestamp in feed().

- RateEst: Keep a queue of the last ~100 msec worth of samples and duration, and filter out collective samples that give a pre-estimate that seems way off.

- Replace "Game Boy / Game Boy Color emulator" with "Game Boy Color emulator" for now to avoid misleading anyone on the current status.

gambatte_qt:

- Disable BlitterWidget updates (paintEvents) while not paused.

- Allow BlitterWidgets to opt in to get paintEvents while unpaused. Do so for QGLBlitter since it may need to clear buffers afterwards.

- QGLBlitter: Try to blit right after sync in the case of single buffering.

- Up default audio buffer latency to 100 ms (some common system audio servers require a lot of buffering to work well).

- Adaptively skip BlitterWidget syncs if audio buffer is low, in a manner that should minimize wasted skips in sync to vblank situation, and tries to be non-disturbing. This replaces frame time halving, and blitter specific rescueing.

- Clear display buffers in DirectDrawBlitter and Direct3DBlitter in exclusive mode, since blits don't necessarily cover the entire buffers.

- DirectDrawBlitter: Make sure that a minimum amount of time has passed between calls to WaitForVerticalBlank, since it can return in the same vblank period twice on a fast system.

- DirectDrawBlitter: Support vsync for refresh rate ~= 2x frame rate.

- DirectDrawBlitter: Refactor somewhat and get rid of a couple minor potential bugs.

- DirectDrawBlitter: Some tweaks to get updates closer to sync time in certain situations.

- DirectDrawBlitter: Some tweaks to better support DONOTWAIT.

- DirectDrawBlitter: Make only updating during vblank while page flipping optional.

- Direct3DBlitter: Some tweaks to get updates closer to sync time in certain situations.

- Filter out very short frame times in frame time estimation.

- Don't adjust frame time during turbo, but rather skip BlitterWidget syncs to speed up, which avoids vsync limits without disabling vsync.

- DirectDrawBlitter: Add triple buffering option.

- Direct3DBlitter: Use D3DSWAPEFFECT_DISCARD in non-exclusive mode.

- Direct3DBlitter: Allow triple buffering and vblank-only updates in non-excusive mode.

- Rename "Page flipping" in Direct3D and DirectDraw blitters to "Exclusive full screen".

- Pause audio on win32 titlebar clicks/drags to avoid looping audio due to underruns from blocked timerEvents.

- Use wildcards for platform detection to avoid being unnecessarily compiler/architecture specific. Fixes bug 2377772.

- Rewrite most of DirectSoundEngine, supporting primary buffer option, making it more robust, correct and hopefully cleaner. Only use part of the primary buffer if the desired buffer size is lower than the primary buffer size.

- Direct3DBlitter and DirectDrawBlitter: Force blocking updates when sync to vblank is enabled. Some updates only block if there's a prior unfinished update in progress. This screws up frame time estimation in turn screwing up vsync. To fix this we do a double update (and extra blit) if close to a frame time period has passed since the last update when sync to vblank is enabled. I really should have noticed this earlier as it pretty much breaks vsync adaption completely.

- Direct3DBlitter: Use the D3DCREATE_FPU_PRESERVE flag when creating device. Omitting this flag can screw up floating point calculations in other parts of the code. For instance WASAPI cursor timestamps get utterly screwed up here.

- Direct3DBlitter: It appears that managed textures are updated before they are unlocked, which screws up redraws, making things appear choppy in some situations. Use a default memory texture and a system memory texture and the UpdateTexure method instead.

- DirectSoundEngine: Make use of the sample period limit feature of RateEst, rather than duplicating the feature.

- Add polling WASAPI engine with exclusive mode support. Latency and rate estimation is generally better than DirectSound, and in exclusive mode there is less blocking as well as exclusive mode being better than shared mode in the other areas too.

- WasapiEngine: Add device selection.

- WasapiEngine: Add static isUsable() method. Only listed if isUsable(). Default engine if isUsable().

- WasapiEngine: Use default device if there's only one device available, since we don't show the combobox anyway.

- DirectSoundEngine: Provide the integrated read and status get write method optimization.

- XvBlitter: Set NosystemBackground attribute rather than OpaquePaintEvent. Reimplement paintEngine to return NULL as suggested by Qt docs.

- X11Blitter: Reimplement paintEngine to return NULL.

- AlsaEngine: Make use of sample period limit feature of RateEst. Don't increase estimated sample rate on underrun.

- OssEngine: Make use of sample period limit feature of RateEst. Don't increase estimated sample rate on underrun.

- Esc exits fullscreen on macx.

- Drop OpenAL from default macx binary.

- Add some useful but commented build flags for macx to .pro files.

>> Get it HERE.

 

If the link doesn't work, try it again. Sourceforge don't seem have any interest in fixing this bug.