Require callers of Config::AddConfigChangedCallback and
CPUThreadConfigCallback::AddConfigChangedCallback to handle the returned
ConfigChangedCallbackIDs to hopefully prevent future issues with
callbacks getting called after their associated objects have been
destroyed.
Use a single lambda as a callback which calls InitCustomPaths and
RefreshConfig instead of having separate callbacks for each of them.
This fixes the callback for InitCustomPaths not being removed on
shutdown; the callback for the lambda (previously for RefreshConfig) is
already removed in Shutdown().
Prevent SetHardcoreMode from being called after m_client is set to
nullptr. rc_client_set_hardcore_enabled() checks for nullptr so this
didn't cause any problems, but better not to rely on that.
Also prevents multiple SetHardcoreMode callbacks from piling up when
repeatedly toggling Config::RA_ENABLED.
Remove ConfigChangedCallback in MainWindow's destructor to prevent the
callback from accessing the destroyed MainWindow afterward.
After MainWindow is destroyed UICommon::Shutdown calls
LogManager::Shutdown which ultimately triggers any remaining callbacks.
This resulted in calling MainWindow::OnHardcoreChanged, which crashed in
debug builds and didn't have any obvious effect in release builds.
Fixing an oversight: this was causing the debugger to be disabled if achievements were disabled but hardcore mode was still enabled in the .ini. This fix properly checks for hardcore state via AchievementManager which takes both settings into account.
I've been playing Rock Band 3 recently and have experienced a bug where
sometimes if you disconnect and reconnect a USB microphone, the game
won't pick up on it connecting, not even it you disconnect and reconnect
it again. An investigation into what's going on inside Dolphin shows
that when the game triggers a call to OH0::DeviceOpen after the device
has been reinserted, Dolphin doesn't open the device because it's
already present in m_opened_devices.
Removing the device from m_opened_devices after calling OH0::TriggerHook
in OH0::OnDeviceChange resolves this specific issue in my testing. Doing
this matches us removing the device from m_opened_devices after calling
OH0::TriggerHook in OH0::DeviceClose, but I haven't looked at exactly
what real IOS does.
I have been able to reproduce a much rarer issue that has the same
symptoms on the surface but where OH0::DeviceOpen gets past its
m_opened_devices check. I'm currently not sure what the cause of this
remaining issue is.
If the build folder is created from an older commit and this flag is not set, the last value set for this flag will be used instead.
This is currently happening on our macOS build machine, causing macOS 10.15 to still be used as the deployment target.
71f654c added a new platform in the middle of the C++ platform enum
without updating the corresponding Android code, making the Android code
incorrectly treat Wii discs as WAD files, WAD files as DOL/ELF files,
and so on. This commit fixes the problem.
To be able to add the new Triforce entry into the Platform enum without
it leading to the UI getting an additional tab, I'm splitting the enum
into Platform and PlatformTab. Platform now exactly matches the C++
enum (previously it excluded ELFOrDOL), and PlatformTab has the same
content as the old Platform.
RetroAchievements disables pausing too frequently when running but there's no sense of doing this if RetroAchievements does not currently have a game running.
Some games open two USB interfaces, e.g. /dev/usb/oh0 and /dev/usb/hid.
This was causing us to run two scanning threads at once, using up more
CPU time for scanning than we need to.
We have identified that a failed RetroAchievements game load (most easily done when closing a game before the server can finish responding) can leave data behind that causes problems. As such, refactored CloseGame to always delete data even if there wasn't a game loaded when it was called, and call it on the failure paths of LoadGameCallback.
Instead of having USBScanner create "hooks" as it scans for devices,
let's have USBScanner present a list of devices to USBHost and have
USBHost diff the new device list with its old device list to create the
hook calls instead. This gets rid of some complex edge cases that the
next commit otherwise would have to deal with, in particular regarding
toggling determinism and adding new USBHosts to a USBScanner.
Note: After adding the missing locking of m_devices_mutex, I had to move
the locking of m_hooks_mutex to avoid a random deadlock between the CPU
thread and USB scanning thread. (Either that or I would have to lock
m_devices_mutex before m_hooks_mutex.)
This gets rid of the ugly direct access to USBScanner::m_devices that
was introduced by the previous commit.
This also fixes a potential thread safety issue.
USB_HIDv4::TriggerDeviceChangeReply loops through m_devices and calls
GetDeviceEntry for each device. If USB_HIDv4::TriggerDeviceChangeReply
is called after a new device is added to m_devices but before hooks are
dispatched, GetDeviceEntry crashes, because the hook that's supposed to
update m_device_ids hasn't run yet. With this commit, this issue can no
longer happen, because USBHost::m_devices_mutex doesn't get unlocked in
between updating m_devices and dispatching the hooks.
dolphin-start event was being generated twice for the normal
end-user case, as can be seen in analytics data for some years.
The problem occured when:
* Android reaped the process hosting the dolphin activity
(e.g. for power/memory saving).
and
* Dolphin activity was in "stopped" state for > 6 hours before
being switched back to.
Under above conditions, both calls to ReportStartToAnalytics
would be performed, as dolphin thought it was being launched anew,
and also thought it had been asleep for > 6 hours.
fixes https://bugs.dolphin-emu.org/issues/13675
Due to requests from RA Devs, updating the AchievementManager LoadGameCallback to still set MemoryPeeker (and set m_system) if the load game response is NO_GAME_LOADED, so that the memory inspector et al continue function properly on unidentified hashes. Without this, no memory is loaded and the memory inspector will show all zeroes.
If the development system is started for a game with an unrecognized hash, RA_Integration opens a dialog for connecting the hash with a title. That dialog is prepopulated by the results of GameTitleEstimateHandler.
Displays an additional message when an achievement unlocks that isn't on the site yet (either hasn't yet been uploaded or modified from remote) i.e. achievements the "player" is actively developing.
When you use TimePlayed, you have to provide a game ID either when
creating the object or when calling GetTimePlayed on it. If you don't
provide a game ID when creating the object, function calls that don't
take a game ID will silently fail, except for Reload. This isn't very
obvious, and there's no strong benefit to storing the game ID inside
TimePlayed anyway (it just lets TimePlayed skip calling EscapeFileName),
so this commit removes the TimePlayed constructor that takes a game ID
and instead makes the functions that need game IDs always take a game ID
argument.
It was being done manually, which a TODO comment advised against.
Using generic_string() from std::filesystem::path solves this.
Fix encoding issue using generic_wstring instead.
Fix overlays stacking on top of each other or not moving to the edge of
the screen when enabling or disabling overlays while emulation is
active.
This change only applies when Config::GFX_MOVABLE_PERFORMANCE_METRICS is
False.
This lets you use PS3 Rock Band controllers with Wii Rock Band and
Guitar Hero games.
A normal user will probably never have any reason to disable this
behavior, but I figured maybe there's some person out there who would
like to disable it. (For instance, I know there's a mod for RB3 that's
trying to implement the same kind of cross-console controller
compatibility, and that can only be tested if the behavior I'm adding is
disabled.) So the behavior is controlled by an INI-only setting.
Removed VolumeChanged signal, as ConfigChanged will trigger what is needed.
Only applies UpdateSoundStream to things that can change during emulation.
Settings::SetVolume might no longer be used, but left it in.
Gecko codes in Dolphin feature a dedicated field for the creator of the
cheat code. When saved into the INI file, the code name and the creator
name are concatenated, and then inserted in the `[Gecko]` section:
```ini
[Gecko]
$<cheat code name> [<creator>]
<code line 1>
<code line 2>
<code line 3>
<...>
$<other cheat code name> [<creator>]
<code line 1>
<code line 2>
<code line 3>
<...>
```
On the other hand, enabled codes are listed under the `[Gecko_Enabled]`
section, but in this case the creator name is omitted from the line:
```ini
[Gecko_Enabled]
$<cheat code name>
$<other cheat code name>
```
Having the creator name in the `[Gecko]` section but not in the
`[Gecko_Enabled]` section is arguably not ideal, but this is legacy
behavior in Dolphin.
The **Cheat Code Editor** dialog is not acknowledging this subtle
behavior in Dolphin: the cheat code name and the creator name *can* be
both inserted in the name field. This issue manifests as an inconsistent
state where a Gecko code that *appears* to be enabled has no effect when
the game is launched.
As part of this fix, the creator name (if present) is now moved into the
dedicated creator field before the code is stored internally.
Test plan:
- Right-click on any game and open the **Properties** dialog.
- Switch to the **Gecko Codes** tab.
- Press the **Add New Code...** button.
- In the **Cheat Code Editor** dialog:
- Enter `This is a test [Jane Doe]` in the **Name:** field.
- Enter `01234567 00000000` in the **Code:** field.
- Press **Save**.
- Observe that the newly added code is now in the list, and *appears* to
be enabled.
- Close the **Properties** dialog.
- Right-click on the same game and open the **Properties** dialog again.
**Without** the fix, the newly added code, while still on the list, has
been inadvertently disabled (it was never really enabled!).
**With** the fix, the newly added code is the list and remains enabled.
This fixes https://bugs.dolphin-emu.org/issues/13695.
This is an Android continuation of bc95c00. We now call
InputDetector::Update immediately after receiving an input event from
Android instead of periodically calling it in a sleep loop. This
improves detection of very short inputs, which are especially likely to
occur for volume buttons on phones (or at least on my phone) if you
don't intentionally keep them held down.
Refactors the AR/Gecko/Patch code approval process to verify from every possible game ini, not just the base game ID. This fixes codes on specific revisions or codes general to any region.
Found via `codespell -q 3 -S "./Externals,./Data/Sys/wiitdb-??.txt,*.po,*.pot" -L andf,asnd,bootup,brocken,bufferin,clen,collet,datas,delt,diety,extint,fpr,inout,inport,interm,nd,nin,ontop,pixelx,re-use,re-used,sav,stateman,strat,transer,wil`
Found via `codespell -q 3 -S "./Externals,./Data/Sys/wiitdb-??.txt,*.po,*.pot" -L andf,asnd,bootup,brocken,bufferin,clen,collet,datas,delt,diety,extint,fpr,inout,inport,interm,nd,nin,ontop,pixelx,re-use,re-used,sav,stateman,strat,transer,wil`
As the PS2 version is the only one with proper 16:9 widescreen, this cheat brings widescreen to the Gamecube version. This cheat has been tested on the NTSC version, but is stated to be region free.
In PPCTables.cpp, the code is currently unused so I was unable to test it.
In CustomPipeline.cpp, a pointer to member function cannot be used due to 16.4.5.2.1 of the C++ Standard regarding "addressable functions". https://eel.is/c++draft/namespace.std#6
In Fs.cpp and DirectoryBlob.cpp, these examples used projections in a previous iteration of this commit, but no longer do. Still, they remain in this commit because the PR they would actually belong to is already merged.
In LabelMap.cpp, the code is currently unused so I was unable to test it.
In WiiUtils.cpp, the magic value `1u` was replaced by the constant value `DiscIO::PARTITION_UPDATE`.
Found via `codespell -q 3 -S "./Externals,./Data/Sys/wiitdb-??.txt,*.po,*.pot" -L andf,asnd,bootup,bufferin,clen,collet,datas,delt,fpr,inout,inport,interm,pixelx,re-use,re-used,sav,stateman,strat,wil`
Clamp overlays to the render window (with some padding), reset their
positions when the render window changes sizes, and add a setting to
enable moving the overlays (off by default, .ini only for now).
This adds the option to configure real Wiimotes by specifying their Bluetooth addresses in
the configuration file. This allows off-brand Wiimotes to work without using the
Bluetooth Passthrough option, if you know their Bluetooth addresses beforehand.
Despite correctly setting the LAP to `0x9e8b00` in `WiimoteScannerLinux::FindWiimotes`
while scanning, which is indeed enough to make off-brand / knock-off Wiimotes respond to a
Bluetooth Inquiry, some (several? all?) bluetooth adapters seem to override and ignore
this given LAP value when performing the Inquiry, and actually use the `0x9e8b33` value as
if a null pointer have been given to `hci_inquiry`, as inspection of USB/Bluetooth packets
by Wireshark indicate. Off-brand Wiimotes don't respond to inquiries with this LAP.
If one happens to know the Bluetooth address of their Wiimote (for example, by checking
`BluetoothPassthrough.LinkKeys` after using Bluetooth Passthrough, or other means such as
directly using `libusb` to force the adapter to use the correct LAP in the Inquiry), then
it's enough to add those addresses to the vector of found Wiimotes.
Since this a niche use case and I only happen to know and have tested in Linux, this
change only affects the `WiimoteScannerLinux` backend. It's likely that it could be added
to other backends, but I'm unfamiliar with these.
If no addresses are given or this config section does not exist, behavior is completely
unchanged.
Introduce a new "Enable Time Tracking" checkbox in the InterfacePane UI. The checkbox is dynamically enabled or disabled based on the emulation state, preventing changes while emulation is active.
Creates TimePlayed class and implemented constructors, AddTime, GetTimePlayed, and Reload methods. Updates CMakeLists.txt and DolphinLib.props as appropriate.
I think someone confused these with the actual token and bounding box
registers in PE, which were added later. In CP they never did anything
and it's suspicious that they have the same addresses as their PE
counterparts. On real hardware they always read as zero.
Move ImGui::End() calls out of if(ImGui::Begin()) blocks.
Quoting from ImGui::Begin's function comment in imgui.cpp:
"You always need to call ImGui::End() even if false is returned."
In practice this didn't cause problems because the windows don't have
title bars and thus can't be collapsed, and so the block containing
::End would always run, but let's do it the right way.
Instead of having UserDataImportWarningDialog set an
`onResultDismiss` callback that examines `mustRestartApp`, and having
UserDataActivity set `mustRestartApp`, just have UserDataActivity set
the callback directly.
This approach is no more data-race-y than the previous approach, and it
simplifies the code. (The behavior of restarting the app when the task
finishes is specific to the user data import flow, and there is no
reason for TaskViewModel to be directly aware of it.)
You can encode a shifted 12-bit immediate in a SUB instruction on ARM64.
We exploit this to avoid materializing the immediate.
This approach saves an instruction if it does not need to be
materialized in a register afterwards. Otherwise, we just materialize
it later and the total number of instructions stays the same.
Before:
0x52a00218 mov w24, #0x100000 ; =1048576
0xcb180379 sub x25, x27, x24
After:
0xd1440379 sub x25, x27, #0x100, lsl #12 ; =0x100000
You can encode a 12-bit immediate in a SUB instruction on ARM64. We can
exploit this to avoid materializing the immediate.
This approach saves an instruction if it does not need to be
materialized in a register afterwards. Otherwise, we just materialize
it later and the total number of instructions stays the same.
Before:
0x5280003a mov w26, #0x1 ; =1
0xcb1a033b sub x27, x25, x26
After:
0xd100073b sub x27, x25, #0x1
While we cannot always avoid materializing immediates, we can still
inspect the most significant bit and potentially skip sign extension.
This can sometimes save an instruction.
Before:
0x5280003a mov w26, #0x1 ; =1
0x93407f5b sxtw x27, w26
0xcb38c37b sub x27, x27, w24, sxtw
After:
0x5280003a mov w26, #0x1 ; =1
0xcb38c35b sub x27, x26, w24, sxtw
Before:
0x52a20018 mov w24, #0x10000000 ; =268435456
0x93407f79 sxtw x25, w27
0xcb38c339 sub x25, x25, w24, sxtw
After:
0x52a20018 mov w24, #0x10000000 ; =268435456
0x93407f79 sxtw x25, w27
0xcb180339 sub x25, x25, x24
You can encode a shifted 12-bit immediate in an ADD instruction on
ARM64. If the negated constant fits in this range, we can exploit this
to avoid materializing the immediate.
This approach saves an instruction if it does not need to be
materialized in a register afterwards. Otherwise, we just materialize
it later and the total number of instructions stays the same.
Before:
0x52bff01a mov w26, #-0x800000 ; =-8388608
0x93407f1b sxtw x27, w24
0xcb3ac37b sub x27, x27, w26, sxtw
After:
0x93407f1b sxtw x27, w24
0x9160037b add x27, x27, #0x800, lsl #12 ; =0x800000
You can encode a 12-bit immediate in an ADD instruction on ARM64. If the
negated constant fits in this range, we can exploit this to avoid
materializing the immediate.
This approach saves an instruction if it does not need to be
materialized in a register afterwards. Otherwise, we just materialize
it later and the total number of instructions stays the same.
Before:
0x12800019 mov w25, #-0x1 ; =-1
0x93407f5b sxtw x27, w26
0xcb39c37b sub x27, x27, w25, sxtw
After:
0x93407f5b sxtw x27, w26
0x9100077b add x27, x27, #0x1
You can encode a shifted 12-bit immediate in a SUB instruction on ARM64.
Constants in this range do not need to be sign extended, so we can
exploit this to avoid materializing the immediate.
This approach saves an instruction if it does not need to be
materialized in a register afterwards. Otherwise, we just materialize
it later and the total number of instructions stays the same.
Before:
0x52a00099 mov w25, #0x40000 ; =262144
0x93407f7a sxtw x26, w27
0xcb39c35a sub x26, x26, w25, sxtw
After:
0x93407f7a sxtw x26, w27
0xd141035a sub x26, x26, #0x40, lsl #12 ; =0x40000
You can encode a 12-bit immediate in a SUB instruction on ARM64.
Constants in this range do not need to be sign extended, so we can
exploit this to avoid materializing the immediate.
This approach saves an instruction if it does not need to be
materialized in a register afterwards. Otherwise, we just materialize
it later and the total number of instructions stays the same.
Before:
0x52800416 mov w22, #0x20 ; =32
0x93407f78 sxtw x24, w27
0xcb36c318 sub x24, x24, w22, sxtw
After:
0x93407f78 sxtw x24, w27
0xd1008318 sub x24, x24, #0x20
A number of settings in the `debugger` group were wrongly using a newly
constructed `QSettings` object instead of the singleton object that
`GetQSettings()` provides.
This made the application create a spurious, extra configuration file in
the user directory:
```
~/.config/Dolphin Emulator/dolphin-emu.conf
```
Notice that, by default, the application configuration files are stored
in `~/.config/dolphin-emu`; not in `~/.config/Dolphin Emulator`.
Adding a community-requested list of Gecko and Action Replay codes to the allowlist. Many of these codes were from the wiki and are being added to Dolphin's repo for the first time.
Change the displayed controls in the TAS Input window when the
controller's extension (including MotionPlus) is changed.
This previously required restarting Dolphin after the attachment was
changed, as the controls were never updated after the WiiTASInputWindow
was created at Dolphin startup.
When I wrote 71e9766519, there was an interaction I didn't take into
account: When setting eq, SetCRFieldBit assumes that all bits in the
passed-in host register except the least significant bit are 0. But if
we use EON or ORN, all bits except the least significant bit get set to
1. This can cause eq to end up unset when it should be set.
This commit fixes the issue.
crandc is unaffected by the issue because the "1" bits get ANDed with
"0" bits from the first operand.
Note that in practice, we never have both bits_1_to_31_are_set and
negate at once, so while it looks like this commit adds an extra AND
instruction in some cases, those cases don't happen in practice, meaning
this fix shouldn't affect performance.
QCheckBox::toggled and other similar signals are used to save changes and to update widget status (such as enabled).. OnConfigChanged needs to load new values and trigger widget updates, but the new value shouldn't trigger a save. A save is unnecessary (the config has the correct values and the UI is being updated to those values) and it'd trigger another ConfigChanged signal. This commit blocks the save without blocking the signal entirely.
The computed value is only used when the register is equal to zero, so
we can fully precompute it and materialize the constant instead. In
other words, we change from
```
return reg == 0 ? (reg | 1ULL << 63) : reg;
```
to
```
return reg == 0 ? 1ULL << 63 : reg;
```
The number of instructions remains the same, but we eliminate an
unnecessary dependency on the register value.
Before:
0xb241037a orr x26, x27, #0x8000000000000000
0xeb1f037f cmp x27, xzr
0x9a9a137b csel x27, x27, x26, ne
After:
0xd2f0001a mov x26, #-0x8000000000000000 ; =-9223372036854775808
0xeb1f037f cmp x27, xzr
0x9a9a137b csel x27, x27, x26, ne
In NandPaths.cpp, the `std::initializer_list<char>` of illegal characters has been turned into a `char[]` (similar to the one in GameList.cpp).
The reverse iteration in ResourcePack.cpp seemed to provide no benefits, and doing without it it seemed to have no ill effects.
The new `Common::Contains` and `Common::ContainsSubrange` function objects mirror C++23's `std::ranges::contains` and `std::ranges::contains_subrange`, respectively.
Recently there was some issues in TASVideos trying to sync a Donkey Kong Country Returns TAS. It eventually was synced by directly using the config from the TAS author. The exact setting which caused the desync was narrowed down to being in SYSCONF, with the country code. The TAS author lives in the US, so the country code matched the US country code, while the person attempting to sync the TAS did not live in the US.
Adding SYSCONF country code to the DTM should avoid this being an issue for future Dolphin versions.
When the input register and carry flags are known, we can always
precompute the result.
We still materialize the immediate when the condition register
needs to be updated, but this seems to be a general problem. I might
look into that one day, but for now this'll do.
- ConstantFalse
Before:
0x52800119 mov w25, #0x8 ; =8
0x2a1903fa mov w26, w25
After:
N/A
- ConstantTrue
Before:
0x52800119 mov w25, #0x8 ; =8
0x1100073a add w26, w25, #0x1
After:
N/A
Same optimization we did for subfex. Skip loading the carry flag into a
temporary register first when we're dealing with zero.
Before:
0x394bd3b8 ldrb w24, [x29, #0x2f4]
0x2a1803f9 mov w25, w24
After:
0x394bd3b9 ldrb w25, [x29, #0x2f4]
When both the input register and the carry flag are constants, the
result can be precomputed.
Before:
0x52800016 mov w22, #0x0 ; =0
0x2a3603f6 mvn w22, w22
After:
The result is either -1 or 0 depending on the state of the carry flag.
This can be done with a csetm instruction.
Before:
0x1280001a mov w26, #-0x1 ; =-1
0x1a1f035a adc w26, w26, wzr
After:
0x5a9f23fa csetm w26, lo
When the immediate is zero, we can load the carry flag from memory
directly to the destination register.
Before:
0x394bd3b8 ldrb w24, [x29, #0x2f4]
0x2a1803f9 mov w25, w24
After:
0x394bd3b9 ldrb w25, [x29, #0x2f4]
To my knowledge, all of the GameCube versions of *Metroid Prime* and *Metroid Prime 2: [Dark] Echoes* only support 4:3, not 16:9 .
Currently, Dolphin's widescreen heuristic will fail to detect this and will erratically switch between 4:3 and 16:9 when Aspect Ratio is set to Auto.
These changes prevent the erratic aspect ratio switching by manually declaring that the game is solely 4:3.
Resolves duplicate OSD messages for Loading and Found custom textures.
VideoBackend initialization results in HiresTexture::Init being called.
We already call HiresTexture::Update when OnNewTitleLoad is called.
Thus we can remove HiresTextures::Init completely as it is redundant.
No games seem to use this, so this isn't useful as a performance
optimization, but it's required for correctness because the (sh == 0)
case of our implementation doesn't handle zero masks.
In JitRegCache.cpp, the lambda predicate were replaced by a pointer to member function because ranges algorithms are able to invoke those.
In ConvertDialog.cpp, the `std::mem_fn` helper was removed because ranges algorithms are able to handle pointers to member functions as predicates.
In BoundingBox.cpp, the lambda predicate was returning the bool element unchanged, so `std::identity` was a better fit.
In WiimoteReal.cpp, JitRegCache.cpp, lambda predicates were replaced by pointers to member functions because ranges algorithms are able invoke those.
In ConvertDialog.cpp, the `std::mem_fn` helper was removed because ranges algorithms are able to handle pointers to member functions as predicates.
In DITSpecification.cpp, MaterialAsset.cpp, and ShaderAsset.cpp, lambda predicates were replaced by pointers to member functions because ranges algorithms are able invoke those.
In NetPlayClient.cpp, the non-trivial `NetPlay::Player` elements were being passed by value in `NetPlayClient::DoAllPlayersHaveGame()`. This has been fixed.
In WIABlob.cpp, the second example's predicate was returning the `std::optional` by value instead of implicitly converting it to a bool. This has been fixed.
Creates a layer outside the game config layer system and passes it to the created gfx widows, so as to not interfere with the global config system.
Supports multiple game properties being open at once.
Supports editing while a game is playing, but the options only save and update the active game when the window is closed.
Right-clicking will remove a property from the game ini.
New code adds a test failure if there's a Patches/Gecko/AR_Retroachievements_Verified code that doesn't appear to actually exist in the file. This will catch if the allowed patch is formatted wrong, which I found happening several times already due to not realizing that the patch author's name would need to be omitted.
Prefer BLENDVPD over VBLENDVPD if the latter doesn't save any
instructions.
VBLENDVPD allows separate source and destination registers, which can
eliminate a MOVAPD/MOVSD. However, on Intel since Skylake, VBLENDVPD
takes additional uops to execute compared to BLENDVPD (according to
https://uops.info). On AMD and older Intel microarchitectures there is no
difference.
Some generators (like Unix Makefiles and Xcode) copy an app's Info.plist at configure time.
This causes a problem when we need to generate the Info.plist at build time, like how we
currently do it with ScmRevGen. Instead of generating the Info.plist directly in ScmRevGen,
provide an Info.plist without any version information to CMake at configure time, have
ScmRevGen generate a separate plist file with the version information at build time, and
then merge the two together to create the final Info.plist.
Now that patches and codes are enabled on a case by case basis, remove patcher code blocking codes entirely in hardcore mode, and reword the warning to be more accurate.
This apparently didn't compile on macOS six years ago before c++20, but
it should be fine by now.
While I'm at it, make the constants upper case per convention.
The value being stored must be loaded into a register. In the case of an
immediate value, this means it must be materialized. The value is
eventually byteswapped before performing the store.
This can be simplified for the value 0 for two reasons:
- ARM64 has a dedicated zero register, so does not need to be
materialized.
- Byteswapping zero is still zero, so we can skip this step.
We could skip byteswapping for other values by immediately materializing
the byteswapped value in a register, but the benefits are not so clear
there (if the value needs to be materialized anyway, it is better to do
it up front).
Before:
0x5280001b mov w27, #0x0 ; =0
0xb9404fba ldr w26, [x29, #0x4c]
0x12881862 mov w2, #-0x40c4 ; =-16580
0x0b020342 add w2, w26, w2
0x5ac00b61 rev w1, w27
0xb8226b81 str w1, [x28, x2]
After:
0xb9404fbb ldr w27, [x29, #0x4c]
0x12881862 mov w2, #-0x40c4 ; =-16580
0x0b020362 add w2, w27, w2
0xb8226b9f str wzr, [x28, x2]
Unlike on x64, inverting EQ or GT in SetCRFieldBit saves us one
instruction. Also unlike on x64, inverting SO or LT in GetCRFieldBit
requires an extra instruction (just like in SetCRFieldBit). Due to this,
replacing an invert in GetCRFieldBit with an invert in SetCRFieldBit
when possible is either equally good or better - never worse.
MUL and SUB can be combined in one instruction.
Before:
0x1b1a7c01 mul w1, w0, w26
0x4b010318 sub w24, w24, w1
After:
0x1b1ae018 msub w24, w0, w26, w24
`std::erase` is a replacement for the remove-erase idiom.
Changes to `OpenModeToAndroid` inadvertently revealed that the prior implementation had UB (potentially deleting the end iterator). This is now fixed.
When we boot the core, it needs to have a valid surface to draw graphics
to. Our Kotlin code does wait for a valid surface to exist before it
calls NativeLibrary.Run, but there's a chance for the surface to be
deleted before Run locks s_surface_lock. If that happens, the core boots
without a valid surface, which presumably would cause a crash. (I
haven't been able to reproduce the problem myself.)
Storing the log type names in a map results in them getting re-sorted by
their keys, which doesn't quite give us the sorting we want. In
particular, the Achievements category ended up being sorted at R (for
RetroAchivements) instead of at A. Every use of the map is just
iterating through it, so there's no real reason why it has to be a map
anyway.
Normally, the asserts added in 34b0a6ea90 are only triggered when
something actually went wrong in Dolphin. But there is one exception:
In FallBackToInterpreter, we flush all registers regardless of whether
they're discarded. This is fine as long as none of the discarded
registers are inputs to the instruction that the interpreter will run.
To avoid false positive asserts, this change adds a parameter to Flush
that controls whether to skip the asserts for discarded registers.
Additionally, an assert for discarded registers is added to
Arm64FPRCache::Flush. (Previously JitArm64 asserted for GPRs (and CRs)
only, whereas Jit64 asserted both for GPRs and FPRs. I most likely
didn't think of FPRs when writing 34b0a6ea90.)
2024-04-01 16:24:26 +02:00
1038 changed files with 100771 additions and 93428 deletions
set(USE_SYSTEM_${upperlib}""CACHESTRING"Use system ${library} instead of bundled. ON - Always use system and fail if unavailable, OFF - Always use bundled, AUTO - Use system if available, otherwise use bundled, blank - Delegate to USE_SYSTEM_LIBS. Default is blank.")
message(FATAL_ERROR "Couldn't find libudev. Can't build hardware database.\nDisable ENABLE_HWDB if you wish to build without hardware database support")
endif()
endif()
if(ENABLE_EVDEV)
find_package(LIBUDEV REQUIRED)
find_package(LIBEVDEV REQUIRED)
@ -695,12 +712,10 @@ if (APPLE OR WIN32)
include_directories(Externals/ed25519)
endif()
# Using static soundtouch from Externals
# Unable to use system soundtouch library: We require shorts, not floats.
This guide is for developers who wish to contribute to the Dolphin codebase. It will detail how to properly style and format code to fit this project. This guide also offers suggestions on specific functions and other varia that may be used in code.
This guide is for developers who wish to contribute to the Dolphin codebase. It details how to properly style and format code for this project. This guide also offers suggestions on specific functions and other elements that may be used in code.
Following this guide and formatting your code as detailed will likely get your pull request merged much faster than if you don't (assuming the written code has no mistakes in itself).
Following this guide and formatting your code as detailed will likely get your pull request merged much faster than if you don't (assuming the code itself has no mistakes).
This project uses clang-format (stable branch) to check for common style issues. In case of conflicts between this guide and clang-format rules, the latter should be followed instead of this guide.
This project uses clang-format 13.0 to check for common style issues. In case of conflicts between this guide and clang-format rules, the latter should be followed instead of this guide.
## <aname="intro-formatting-issues"></a>Checking and fixing formatting issues
@ -49,7 +49,7 @@ In most cases, clang-format can and **should** be used to automatically reformat
- Formatting issues can be checked for before committing with a lint script that is included with the codebase. To enable it as a pre-commit hook (assuming you are in the repository root):
