* CMake in MSYS2 now depends on Ninja and uses it by default, use it
instead of Make as it has much easier to read output when doing
parallel builds and doesn't need an extra program
* Ninja uses the maximum number of cores by default, so we don't have to
run nproc --all
* We don't need mesa for its headers anymore
* Report correct Windows architecture for the build in the manifest
* Output processed .rc/.manifest in the CMake binary dir to prevent
potential issues with the files conflicting if cross compiling melonDS
on the same Windows machine
* Make the original file name actually make sense
This may not work, as I'm not sure why the code originally did as it
did. But this at least lets it compile, we'll need someone with a real
Windows ARM device to test this as I only have a VM with no GPU
acceleration.
* WIP: use Duckstation's context code to directly render into QT Widget from separate thread without two OpenGL contexts
currently only works on Windows
* reenable gay OSD
* add back vsync
* make it atleast a little more thread safe
* linux support
* don't segfault on closing
* reorganise and cleanup build system
it's still not good, but better than before
* macos?
* try to get it working on Ubuntu CI
also update instructions
* let's try this
* ok how about this
* try creating an OGL 4.3 context first
(https://i.kym-cdn.com/photos/images/original/001/264/842/220.png)
* fix Ubuntu
* hm
* try again for Windows
* let's try this
* make the OpenGL renderer work again
that was stupid
* do OGL surface resizing from the mainthread
* Fix small mistake in GL context creation on macOS causing version 3.2 to
be considered invalid
* C stupidness
* cleanup
* don't let the emuthread deinit OGL if there's no OGL
* reset lastScreenWidth/Height when deiniting OpenGL
* disable stencil test while drawing framebuffers
* macOS: Link Cocoa framework explicitly when not building with Qt6
Seems to be needed for the classes used by DuckStation's GL context
code.
* Set ScreenPanelGL's minimum size immediately
Fixes GL context creation for OpenGL display on macOS using the wrong
size as the underlying window was not resized to the correct size by Qt
yet.
* don't emit window updates when OGL display is used
* stuff Arisotura said
Co-authored-by: Nadia Holmquist Pedersen <nadia@nhp.sh>
* The way -O3 was set for release builds was accidentally removing
-DNDEBUG
* -Og seems to mess with debugging with lldb, even though the GCC manual
page says to use it for debug builds, so remove it
According to the Qt documentation we should have this for proper
high-DPI support on macOS. Whether or not it's still relevant I'm not
sure, but if it isn't it might at least help on older macOS or Qt.
* attempt at betterer wifi
* add preliminary sync mechanism
* fix gaps in wifi implementation
* move local-MP comm to its own module instead of cramping Platform.cpp
* remove some stupid cruft
* as you wish, Sorer
(starting work on shared-memory system)
* shared-memory IPC that actually works (albeit Windows-only for now)
* shut up logging from NULL writes on ARM7 (ffs Nintendo learn to code)
* get this somewhat good
* leave client sync mode when host deauths. makes download play actually work.
* start implementing MP-comm error handling
* * add MP-reply error counters
* feeble attempt at fixing slowdown/desync/etc problems
* somewhat better exchange/sync method
* * when entering power-saving mode, be sure to finish transferring the current frame first
* fix misc bug due to old cruft leftover
makes for a more stable connection
* remove a bunch of cruft
* set wifi time interval to 34 cycles instead of 33. games seem sensitive to the general timing of wifi vs the rest of the system, and this seems to make things run better, atleast until I rewrite this to use a proper scheduler.
* more graceful handling of disconnects
* deal with FIFO overflow more gracefully
* BAHAHAHAHAHAHAHAHHHH
THE SNEAKY BASTARDS
so, when the DS receives a beacon with the right BSSID
that beacon's timestamp is copied to USCOUNTER
* attempt at making the connection process smoother for weird games
* * begin adding POWCNT2, only applies to wifi for now
* begin work on wifi scheduler
* implement the shitty timers
* add the RF wakeup thing
* begin work on receiving frames. for now it can just receive melonAP beacons, but hey, it's a start.
* add enough TX functionality that online wifi is a possibility again.
* there are problems with this scheduler thing. committing it anyway
* kind of a rollback... we're gonna work out a compromise on this, I guess
* don't transmit shit if RXCNT.bit15 isn't set
* move RX-finish to its own function. more accurate filtering. implement RXFILTER.
* remove some cruft
* fix some of the shittiness when trying to connect more than two players
* fix some more shittiness
* fix more wifi shittiness (mainly don't try to receive shit while sending a frame)
* run wifi every 8µs. improves performance.
* fix IRQ14/IRQ15
* make this work under Linux
* Make it work on macOS, for now using a custom sem_timedwait
implementation.
If anyone knows anything about mach ports and have an idea for how to
make this work using mach IPC, please do let me know.
* 25ms seems like a good timeout
* begin work on proper multiplayer UI shito.
for now, determine a global instance ID, and derivate the system MAC from it. remove 'randomize MAC' option.
* finish removing RandomizeMAC
* lay groundwork for instance-unique config
* work some on the UI... make it not labelled Fart
* more UI work: make it explicit that some things are instance-unique
* separate firmware files for multiplayer instances
* make instances save to different save files, too
* more UI work, make things somewhat less shitty
* lay base for the multiplayer settings dialog
* actually hook up most of that dialog
* actually implement the fun audio settings
* ensure all the wifi shit is properly savestated and reset. properly update timings for the wifi region when wifi is disabled.
* add more fun labels
* * ignore WEP frames if WEP is off
* implement RX_LEN_CROP
* fake enough of WEP processing to make Inazuma Eleven work
* * do not copy more ROM banner data than actually needed
* avoid trying to read out of bounds if the banner offset is bad
* Fix oversight with the preferences action causing the build to fail on macOS
Co-authored-by: Nadia Holmquist Pedersen <nadia@nhp.sh>
Title strings should be null-terminated.
Read the string up until this point instead of reading the full 128 characters.
(Also fix the .ui file of ROMInfoDialog to prevent it from being too wide.)
