Commit Graph
48632 Commits
Author SHA1 Message Date
Henrik RydgårdandClaude Opus 5.5 80e346d802 Headless: Honor --nand
ApplyToConfig() set nandRootDirectory from it, but headless then reset it to
the memstick's PSP/NAND unconditionally, so --nand did nothing there.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 20:33:34 +02:00
Henrik RydgårdandClaude Opus 5.5 24a2f6110d VSH: Patch vsh_module so the XMB can launch unsigned homebrew
Picking homebrew in the XMB gave "The game cannot be started. The data
is corrupted." without ever calling LoadExec: vshmain.prx rejects a
PARAM.SFO without a valid DISC_ID, and a DATA.PSP that's a plain ELF
rather than ~PSP. NOP the three branches that do that, the same patches
ProCFW's vshctrl makes and JPCSP applies. They're for 6.61's build, and
are only applied if all three instructions are what we expect.

With it, Kosmodrones launches from the XMB, through the PSP boot logo.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 20:33:34 +02:00
Henrik Rydgård dcb1df6930 Merge pull request #22493 from hrydgard/ios-share-file
iOS: Add support for the internal "Share file" API just like Android
2026-10-10 19:55:48 +02:00
Henrik Rydgård 8e407deb90 Merge pull request #22492 from hrydgard/android-oboe-mic
Android: Hand microphone input over on the CPU thread, like the other hosts
2026-10-10 19:23:59 +02:00
Henrik RydgårdandClaude Opus 5.5 cfcf8cc17c iOS: Remove the iCade support
It has done nothing since 2017. iCade input came through iCadeReaderView,
a hidden UIKeyInput view that decoded the controller's keystrokes, and
f0ba66fcd6 ("add support for smart keyboard") commented out the code
that created it. Nothing has made one since, so the view controller's
iCadeEventDelegate methods and the ICadeTracker behind them were never
called.

iCade controllers present themselves to iOS as Bluetooth keyboards, so
their keys still arrive as hardware keyboard presses, which can be
mapped like any other keys. Modern controllers go through
GameController.framework anyway.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 19:10:37 +02:00
Henrik RydgårdandClaude Opus 5.5 c8c3ba8bb6 iOS: Give the soft keyboard a view of its own
The view controller was the keyboard's text input, and UIKit hands first
responder back to it whenever something it presented goes away: a share
sheet, a photo picker, the alert after an app icon change. Each time,
up came the keyboard, and hideKeyboard calls were sprinkled around to
put it away again, one on a four second timer.

Text input now goes to a small view that only becomes first responder
when a text field gets focus. The view controller is first responder
otherwise, for hardware keyboard presses, but takes no text, so it can't
bring up the keyboard. The workaround hideKeyboard calls are gone, and
so is the endEditing on rotation, which dropped the keyboard if you
turned the phone while typing. Showing and hiding are synchronous now,
since every caller is on the main thread.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 19:09:41 +02:00
Henrik RydgårdandClaude Opus 5.5 d80c0b9472 iOS: Support System_ShareFile
Offers the file through the share sheet, which works out the type from
the extension. Also gives NativeInit a cache directory, Library/Caches,
which ShareFilesAsZip needs for its zips. That moves the SPIR-V and
download caches there from Documents/PSP/SYSTEM/CACHE, out of backups
and the Files app, which is where Apple wants caches anyway.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 18:57:18 +02:00
Henrik RydgårdandClaude Opus 5.5 6ed178075c sceUsbMic: Fix QueueBuf::resize losing what was buffered
It allocated the new buffer, pointed buf_ at it, then popped from buf_,
so it copied the uninitialized new buffer onto itself. That happened
whenever a game asked for a larger block than before.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 18:55:13 +02:00
Henrik RydgårdandClaude Opus 5.5 8ef157ab21 iOS: Use the shared SampleRing for microphone input
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 18:53:08 +02:00
Henrik RydgårdandClaude Opus 5.5 6cf4872e78 Android: Hand microphone input over on the CPU thread, like the other hosts
The Oboe input callback used to call Microphone::addAudioData directly,
writing PSP memory from the audio thread. Now it only pushes samples
into a SampleRing, and pollRecording drains it on the CPU thread.
Starting and stopping still go through Java, which asks for permission.

Every host with a microphone polls now, so sceUsbMic drops what only the
push path needed: QueueBuf's recursive mutex, restarting the mic on a
device change, and the startMic parameter nobody read. Also removes the
unused System_AudioRecordingState and Microphone::isNeedInput.

The Java microphone_command branch didn't return true, so every command
logged "processCommand failed".

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 18:49:21 +02:00
Henrik RydgårdandClaude Opus 5.5 1e5a54e7b5 Move the microphone's lock-free sample ring to Common/Audio/SampleRing.h
Android is about to need the same one.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 18:49:21 +02:00
Henrik Rydgård b1ebb008ee Merge pull request #22491 from hrydgard/windows-wasapi-mic
Windows: Capture the microphone through WASAPI instead of Media Foundation
2026-10-10 18:49:14 +02:00
Henrik RydgårdandClaude Opus 5.5 3656fb46e2 Windows: Capture the microphone through WASAPI instead of Media Foundation
Windows now uses the same host interface as SDL and iOS: sceUsbMic sends
startRecording/stopRecording/deviceChanged, and drains buffered input on
the CPU thread with pollRecording. A capture thread asks WASAPI for mono
16-bit at the game's rate (it converts from the mix format, so no more
ffmpeg resampling), and copies packets into a lock-free ring buffer.
It reopens the device when the selection changes, and retries once a
second if the device is missing or gets unplugged.

The device list comes from the active capture endpoints. Their friendly
names are the ones Media Foundation reported, so existing MicDevice
settings still match. An empty setting, now selectable as Auto like on
SDL, means the default capture device.

The Media Foundation audio capture code is removed; the camera still
uses it.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 18:28:33 +02:00
Henrik Rydgård 5071ffa22d Merge pull request #22490 from hrydgard/ios-mic
Add iOS microphone support
2026-10-10 18:28:28 +02:00
Henrik RydgårdandClaude Opus 5.5 9dacdf8297 iOS: Add microphone support
sceUsbMic input comes from the input side of the RemoteIO unit that
already does the output. The input callback only copies samples into a
lock-free ring buffer, sized so it never allocates, and the CPU thread
drains it into Microphone::addAudioData when sceUsbMic polls, the way
the SDL backend works. Reconfiguring the unit happens on the main thread.

Whether the game is recording is kept apart from whether the unit's
input is on, so recording resumes after an interruption or a trip to
the background, which shut the unit down. The unit also runs, silently,
when sound is off but the game is recording.

While recording, the session is PlayAndRecord, with Bluetooth through
A2DP so headphones keep full quality and the built-in mic is the input.
Permission is asked for on iOS 17+ through AVAudioApplication.

Also moves the microphone recorder homebrew used to test this from
Tools/psp-microphone to pspautotests, as demos/microphone, and updates
the submodule.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 17:55:21 +02:00
Henrik Rydgård 7c6c666370 Merge pull request #22489 from hrydgard/fix-vsh-sceReg-overlay
sceReg: Fix the VSH reporting corrupted settings
2026-10-10 17:48:29 +02:00
Henrik RydgårdandClaude Opus 5.5 b4ec29111b Microphone: Let the host say whether it wants polling, rather than checking for SDL
The CPU-thread polling added for SDL was keyed off defined(SDL), which
headless and Core are built with too, hence the extra headless check.
Ask the host instead, with SYSPROP_MICROPHONE_NEEDS_POLLING, so other
hosts can hand their input over the same way.

startMic now sends the same startRecording command for every non-Win32
host, taking the rate from curSampleRate like the SDL path did. The
Android path read it from param, which leaked, and which is null when
coming from onMicDeviceChange.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 17:20:48 +02:00
Henrik Rydgård 7c780088ac Merge pull request #22488 from hrydgard/ios-cleanup
iOS: Platform layer cleanup
2026-10-10 17:18:48 +02:00
Henrik RydgårdandClaude Opus 5.5 55e9586897 sceReg: Refuse to create a key that already exists
sceRegCreateKey returns 0x80082728 for a key that already exists, whatever
its type, and leaves the key alone. We replaced the key with an empty one,
which wiped any value nobody wrote back afterwards.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 17:11:19 +02:00
Henrik Rydgård 7b716e6f33 Merge pull request #22459 from chriopter/sdl-microphone
SDL: Add microphone capture and device selection
2026-10-10 17:01:04 +02:00
Henrik Rydgård bd3e55a104 Merge pull request #22487 from hrydgard/android-lifecycle-fixes
Android lifecycle fixes, file sharing support (unused)
2026-10-10 16:43:58 +02:00
Henrik RydgårdandClaude Opus 5.5 6fab39ebb4 iOS: Keep camera frames upright in portrait
The camera delivers frames in the sensor's landscape orientation however
the phone is held, so in portrait the picture was 90 degrees off. Rotate
the capture connection: on iOS 17+ with a rotation coordinator, which
keeps the picture level with the horizon, and before that to match the
interface orientation.

A portrait frame doesn't fit the landscape size the game asks for, so
scale frames to cover it and crop the rest, rather than stretching.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:42:25 +02:00
Henrik RydgårdandClaude Opus 5.5 9e2cadb685 iOS: Remove leftovers
An unused IAPManager startObserving (which would have registered the
observer twice), the obsolete UILaunchImageFile, debug logging of the
SceneDelegate class loading, and a commented-out JIT version check.
Also bring the file overview at the top of main.mm up to date.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 a2ba9c0550 iOS: Implement copying to the clipboard
It was commented out over doubts about the threading. Setting the
pasteboard on the main thread settles those.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 a4ca6af0dc iOS: Pass restart arguments on, and restart into GL when Vulkan fails
RESTART_APP's arguments, like --start-screen=touchscreentest, were only
logged. Worse, the block that got them captured param1 by reference, and
ran after the request's string was gone. Copy them, and parse them along
with the startup path when relaunching.

When Vulkan failed to initialize, the config switched to OpenGL, but
the Metal view controller stayed, so nothing rendered until the user
restarted the app by hand. Restart it automatically, unless OpenGL is
disabled, where config loading would pick Vulkan again and loop.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 81f77f5156 Move Android's command line splitter to CmdLine, as SplitCommandLine
So iOS can split its restart arguments the same way.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 aee93405f5 iOS: Scale camera frames to the requested size instead of cropping
The frame was drawn at its own size into a context of the size the game
asked for, so the game got one corner of the picture. Also drop the
preview layer, which was never shown.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 b4775cde07 iOS: Fix the safe area inset and touch scaling
The safe area insets were converted to pixels with UIScreen.scale, but
the display size with nativeScale. The two differ on the Plus models
and the minis, so the insets were off by up to 15% there. Both use the
window scene's nativeScale now.

Touch positions were truncated to integers before being multiplied by
the screen scale, putting them on a grid about 3 dp apart. Convert
straight from points to dp, by the view size.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 2e08b54402 iOS: Move off deprecated APIs
- Request an orientation update with setNeedsUpdateOfSupportedInterfaceOrientations
  on iOS 16+, where attemptRotationToDeviceOrientation is deprecated.
- Open the document picker with UTTypes on iOS 14+, and present it from
  sharedViewController rather than through the deprecated keyWindow.
- List cameras with a discovery session instead of devicesWithMediaType.
- IS_IPAD() instead of UI_USER_INTERFACE_IDIOM().

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 3375fbdbbd iOS: Replace the private keyboard and vibration APIs with public ones
Sideloaded builds read hardware keyboard events from the private
UIEvent._gsEvent, at offsets hardcoded for iOS 7 and 9, through a
UIApplication subclass. The UIPress path the App Store build uses works
from iOS 13.4, so use it in all builds, and delete the hack along with
PPSSPPUIApplication and SmartKeyboardMap.

System_Vibrate asked the private UIDevice feedbackSupportLevel whether
to use the feedback generator, in App Store builds too, and sideloaded
builds fell back to the private AudioServicesPlaySystemSoundWithVibration.
The feedback generator just does nothing on devices without a Taptic
Engine, so always use it. That leaves the iPhone 6s and first SE
without virtual button vibration in sideloaded builds.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 6fe0f420b1 iOS: Drop version checks for iOS 12 and older, use one deployment target
The minimum is iOS 13 (15 for the App Store build), so the iOS 10
display link fallbacks, the iOS 13 orientation check, the weak import
of GCController and the respondsToSelector checks for APIs from iOS 8
to 13 can go.

The keyboard presses checked for iOS 13.0 but used UIPress.key outside
the check, which needs 13.4, so it would crash on 13.0-13.3 if it ever
runs there. It only runs in App Store builds now. Folded the two copies
into one function with the right check.

The deployment target was 13.0 in the toolchain's compile flags but
12.0 in CMAKE_OSX_DEPLOYMENT_TARGET and 11.0 in the Xcode attribute for
non-App Store builds. All three now come from the toolchain file.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 895d54eb45 iOS: Make the emu thread's UIKit and Core Location calls on the main thread
System requests and System_Vibrate come from the emu thread. Camera,
GPS, share and keyboard requests called sharedViewController directly
from there, which the main thread can nil out during a restart, and the
share sheet, the location manager and the feedback generator were used
off the main thread. Dispatch them all to the main thread.

The share sheet also had no popover anchor, which iPad requires.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 69d33ef8e9 iOS: Resize in place on rotation instead of restarting graphics
Both view controllers stopped the emu/render thread on every rotation
and started it again, tearing down all of graphics. Since d0e45b94d9
NativeFrame resizes the graphics context and posts GPU_DISPLAY_RESIZED
on iOS too, so a NativeResized() is all a size change needs, like
Android now does. GLKView resizes its drawable by itself, and the Vulkan
context recreates the swapchain, which MoltenVK sizes from the layer.

The resize happens in viewDidLayoutSubviews, which UIKit runs at the
final size as a rotation animation starts. Resizing when the rotation
completed instead left half the screen black while it animated. This
also covers window resizes, which matter more now that
UIRequiresFullScreen is deprecated (iPadOS 26).

NativeFrame sets the PSP_CoreParameter() size, so the main thread
doesn't have to.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 16:28:27 +02:00
Henrik RydgårdandClaude Opus 5.5 b9a706432a Add ShareFilesAsZip, for sharing several files as one zip
Zips the given files (content:// paths included) into a folder of their own
in the app's cache and offers the zip through System_ShareFile. The receiving
app reads it whenever it likes, so it can't be deleted right away: each share
deletes the folders of earlier shares once they're an hour old.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 16:06:55 +02:00
Henrik RydgårdandClaude Opus 5.5 77a6abfe6e Add System_ShareFile, for handing a file to another app
Offers a file through the system share sheet, for things like sending a
zipped savestate to a friend. Android for now (SYSPROP_SUPPORTS_SHARE_FILE):
content:// paths from a picked folder are shared as they are, and plain paths
through a FileProvider. Either way the receiving app can read just that file.
With no MIME type given, it's guessed from the extension.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:59:42 +02:00
Henrik RydgårdandClaude Opus 5.5 d35cdf8fcf Android: Track the activity that owns the native side by identity, and log changes
Replaces the instance counter with a weak reference to the current activity,
and says in the log when an activity takes over, and when a replaced one skips
something because the native side isn't its anymore.

The render loop state is static now, with the activity whose surface it runs
on: there's one render thread. An activity only stops the loop if it runs on
its own surface, and one that finds another activity's loop still running
logs an error and stops it before starting its own.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:59:16 +02:00
Henrik RydgårdandClaude Opus 5.5 bcb9e2e5ba sceReg: Keep a key's type when sceRegSetKeyValue writes it
The registry upgrade the VSH runs at boot recreates and rewrites BIN keys
like /CONFIG/NP/account_id (16 bytes). sceRegSetKeyValue guessed the type
from the size and stored them as STR, and the VSH then reported the
settings as corrupted. Use the type of the existing key, and fall back to
guessing only for keys that don't exist yet.

Fixes #22484

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:53:52 +02:00
Henrik RydgårdandClaude Opus 5.5 43bd05cd70 Headless: Say so when a game fails to boot, and exit with an error
Outside --compare, a boot failure printed nothing (only a GitHub Actions
annotation) and exited 0, which looks like a run that did nothing. Seen as
rare silent zero-output runs in headless_bench.py.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:53:36 +02:00
Henrik Rydgård 6779143827 Merge pull request #22486 from hrydgard/softgpu-x86-opt
SoftGPU: Do an optimization pass on x86-64 too
2026-10-10 15:48:04 +02:00
Henrik RydgårdandClaude Opus 5.5 34da53de5a Android GL: Support PresentMode::IMMEDIATE with eglSwapInterval
Now that EGL is ours rather than GLSurfaceView's, the swap interval can be set,
so turning vsync off (and fast-forward without skipping flips) works on Android
GL too. Not in VR, where OpenXR presents, and not with a config whose minimum
swap interval is 1.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:45:27 +02:00
Henrik RydgårdandClaude Opus 5.5 2123e7fdac Android: Tie emulation and audio to visibility, not focus
onPause used to join the render thread and stop the audio, so anything taking
focus while the game stayed on screen (a permission dialog, a share sheet, the
other app of a split screen before Android 10) tore down the whole graphics
context and silenced the game. Emulation, audio, sensors and input listeners
now follow onStart/onStop. The render loop itself already follows the surface;
onStop joins it too, for devices that keep a stopped activity's surface.
The camera stays on onPause/onResume, as it goes to the focused app.

A finishing activity lets go of the render thread in onPause, since a quick
relaunch creates the next activity before this one's onStop, and that onStop
then leaves the audio and location updates alone.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:45:27 +02:00
Henrik RydgårdandClaude Opus 5.5 87353cfd01 Android: Fix activity state lost on recreate, and a destroyed activity kept alive
- Initialize() runs for the first activity only, but set up per-activity
  fields: after any recreate, audioManager was null, so audio focus was
  never taken or given up again, and sustained performance mode was off.
- The static camera helper kept the first activity's display. It now gets
  the application context.
- An activity backed out of (or destroyed by the system) with the native
  side left running stayed referenced by it until the next one registered,
  and native code could still post commands to it. It now lets go, and a
  command arriving at a destroyed activity is dropped (an input box request
  answered as cancelled).
- An open input box is dismissed in onDestroy, instead of its window
  leaking and its request never being answered.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:31:24 +02:00
Henrik Rydgård e312af9b04 Merge pull request #22483 from hrydgard/reinterpret-subpixel
Further improve the Tantalus games
2026-10-10 15:27:21 +02:00
Henrik Rydgård 32fc6c0462 Support the "FastEmulatedGPU" compat flag in the software renderer too. 2026-10-10 15:25:20 +02:00
Henrik RydgårdandClaude Opus 5.5 b98a757157 headless_bench.py: Report CPU time on Windows too
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:15:04 +02:00
Henrik RydgårdandClaude Opus 5.5 6245285ca7 SoftGPU: Don't linger in drawing threads on Windows, and less in transform helpers
On native Windows (Ryzen 9 9950X3D), lingering only paid in the transform
helpers. With no drawing-thread linger and 100 us for the helpers, wall
time stayed within 1% of 200/300 us in Daxter, GTA VCS, Wipeout Pure,
Tekken 6, Burnout Dominator, LocoRoco and Ridge Racer, with 9-27% less
CPU time. No linger at all cost up to 7% (Daxter), almost all of it from
the transform helpers.

The linger times are now per platform; Linux keeps the WSL2 values.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:15:03 +02:00
Henrik RydgårdandClaude Opus 5.5 c53448dd6b SoftGPU: Pick the mip level once for staged rows of a triangle with one q
Where the mip level comes from q (slope mode, or auto with UV planes) and q is
the same over the whole triangle, its pixels all have one level. The staged
spans went through ApplyTexturing for each four pixels anyway, to work out that
level and call the quad sampler; a variant of DrawStagedSpans now works it out
once and calls the quad sampler directly. It's a separate variant, as the extra
code in the shared one cost other games about 1% of instructions on an M1.

LocoRoco takes 7.5% fewer instructions and 5% less wall time, Wipeout 2% fewer
instructions; the others are unchanged. The frame dump set renders the same.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 15:01:49 +02:00
Henrik RydgårdandClaude Opus 5.5 e415e667d0 SoftGPU: Draw 1:1 sprites as rectangles unless DrawSprite has a fast path for them
DrawSprite is only fast in its single pixel paths; for any other pixel state
(fog, depth test, other blending) it sampled and drew a pixel at a time, while
DrawRectangle draws four in vector lanes. Such sprites now go to DrawRectangle.
On an M1: Tekken 6 and LocoRoco take 4-5% fewer instructions, Burnout 1%. The
frame dump set renders the same.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-10 15:01:49 +02:00
Henrik RydgårdandClaude Opus 5.5 c32a01965b SoftGPU: Multiply light colors with 16-bit multiplies
The light color products went through Vec4<int>'s multiply, a 32-bit
lane multiply that SSE2 emulates, though their factors are at most 10
bits. MulSmall uses CrossSIMD's Mul16 where Vec4<int> has a register.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:01:49 +02:00
Henrik RydgårdandClaude Opus 5.5 3540e4ef64 SoftGPU: Scale by powers of two through the exponent bits
GEDot3 and GERowSum called ldexpf for each result, and the UV plane
setup ldexp and ilogb for every triangle and tile, about 2% of the CPU
time between them. Adding to the exponent field gives the same values
while they stay normal (ScaleByPow2Exact, which calls ldexpf otherwise),
and the planes' values and shifts always are. 2-4% fewer instructions
in every game measured.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-10 15:01:49 +02:00