avm1: Match Flash broadcaster prototype enumeration order#23892
Merged
SuchAFuriousDeath merged 1 commit intoJun 5, 2026
Merged
Conversation
Collaborator
Author
|
Oops, forgot some tests :( I'll try to see if I can solve the other order differences as well. Maybe the outputs can be more comparable with extra fixes. |
89c28e3 to
3e5bdc3
Compare
kjarosh
reviewed
Jun 5, 2026
kjarosh
approved these changes
Jun 5, 2026
3e5bdc3 to
cc96238
Compare
cc96238 to
a6f3de5
Compare
AVM1 enumerates own properties in reverse insertion order. The broadcaster methods on `MovieClipLoader`, `FileReference`, `FileReferenceList`, `System.IME`, `Stage`, `Key`, `Mouse` and `Selection` were defined in the order `_listeners, addListener, removeListener, broadcastMessage`, so `for-in` yielded the reverse — the opposite of Flash, which yields `_listeners, removeListener, addListener, broadcastMessage`. Define them in the matching reverse order so iteration lines up with Flash. Additionally, give `_listeners` the `DONT_DELETE` flag Flash assigns to it (previously `DONT_ENUM` only).
a6f3de5 to
4be22ca
Compare
Hancock33
added a commit
to Hancock33/batocera.piboy
that referenced
this pull request
Jun 7, 2026
----------------------------------------------------------------------------------- cemu.mk 326933b2485d87594c5e0d41638a1f0318510a3b # Version: Commits on Jun 05, 2026 ----------------------------------------------------------------------------------- nn_ac: Add local IP getter implementation for Linux (#1839), ---------------------------------------------------- clk.mk 2026-06-06 # Version: Commits on Jun 06, 2026 ---------------------------------------------------- This release improves parsing of Tandy CoCo CAS files. ------------------------------------------------------------------------------------------ duckstation.mk 0b700d1b812d2e6bad6de477980700a8eae83415 # Version: Commits on Jun 06, 2026 ------------------------------------------------------------------------------------------ GTE: Eliminate bounds check in dispatcher, ----------------------------------------------------------------------------------- eden.mk 5ba3ae32dc040ac813c93512b0d90a22c4ec278a # Version: Commits on Jun 06, 2026 ----------------------------------------------------------------------------------- [hle/sockets] fix crash when socket() isn't initialized but send()/recv() are called (#3729) -------------------------------------------------------------------------------------- flycast.mk 751295151eed3cae619b14bcee49bb07f4fdb59c # Version: Commits on Jun 05, 2026 -------------------------------------------------------------------------------------- Fetch translations & Recreate libretro_core_options_intl.h, -------------------------------------------------------------------------------------- openmsx.mk 8d8c45e445b6d5112ba69f6c96f939a926689b56 # Version: Commits on Jun 06, 2026 -------------------------------------------------------------------------------------- ROMDB cleanup and addition of ROM hashes, ----------------------------------------------------------------------------------- play.mk 3a904f67694ce6ce8f88fd97ebaf30240bd87dce # Version: Commits on Jun 05, 2026 ----------------------------------------------------------------------------------- Merge pull request #1589 from hector-tomazella/patch-5 Added idle skip patch for the game Virtua Fighter 4 - Evolution, ------------------------------------------------------------------------------------- ppsspp.mk ecc35b904a24a570201e4b6cf3f12473a97ed3ec # Version: Commits on Jun 05, 2026 ------------------------------------------------------------------------------------- Merge pull request #21792 from hrydgard/fix-another-depth-issue Fix menus in NBA2K12., ------------------------------------------------------------------------------------ rpcs3.mk 62d32ab45e80a0cded6d73e92d031b71f3de3fbc # Version: Commits on Jun 04, 2026 ------------------------------------------------------------------------------------ Update SDL to 3.4.10, --------------------------------------------------------------------------------------- touchhle.mk 54d0ebeb7e99f7833edec75bf2f3ee17f4261e18 # Version: Commits on Jun 05, 2026 --------------------------------------------------------------------------------------- Tweak SQLite3 license acknowledgement, ----------------------------------------------------------------------------------- ymir.mk 79bb23a7682123f57a5c6addd236ac59dce7af56 # Version: Commits on Jun 06, 2026 ----------------------------------------------------------------------------------- docs(build): Document another downside of FetchContent, ------------------------------------------------------------------------------------- ikemen.mk 41a5837d0faf9fe3600e9a8f7da7f9232e6fe4a3 # Version: Commits on Jun 06, 2026 ------------------------------------------------------------------------------------- style: fix code style issues with gofmt, ------------------------------------------------------------------------------------------------- moonlight-embedded.mk f32e415aea6797d261d6b470dcf8bf18727341c2 # Version: Commits on Jun 06, 2026 ------------------------------------------------------------------------------------------------- libgamestream: fix uniqueid.dat read check fread(unique_id, UNIQUEID_CHARS, 1, fd) returns the number of items read (1 on success), not bytes, so the != UNIQUEID_CHARS comparison is always true. Any existing uniqueid.dat is silently ignored and rewritten with the default 0123456789ABCDEF on every run, making a custom per-device uniqueid impossible., --------------------------------------------------------------- ruffle.mk nightly-2026-06-06 # Version: Commits on Jun 06, 2026 --------------------------------------------------------------- ## What's Changed * frontend-utils: Use native TLS instead of rustls by @kjarosh in ruffle-rs/ruffle#23894 * chore: Bump openssl from 0.10.73 to 0.10.80 by @dependabot[bot] in ruffle-rs/ruffle#23902 * chore: Additional Rust dependency version bumps by @torokati44 in ruffle-rs/ruffle#23903 * avm1: Match Flash broadcaster prototype enumeration order by @SuchAFuriousDeath in ruffle-rs/ruffle#23892 **Full Changelog**: ruffle-rs/ruffle@nightly-2026-06-05...nightly-2026-06-06, ---------------------------------------------------------------------------------------------------- gamenetworkingsockets.mk 858225c41dc72aa5f9e00ae64d4df9a15816867b # Version: Commits on Jun 06, 2026 ---------------------------------------------------------------------------------------------------- Fix SteamNetworkingUtils() interface for SteamAPI dedicated server The way we fallback to gameserver or user doesn't work anymore. SteamInternal_FindOrCreateUserInterface now crashes if called before steam is initted. P4:10716251 (cherry picked from commit f1edd32d2a6d71cfc0568112610752a28666f585), ------------------------------------------------------------------------------------------ fallout2-ce.mk 99a762b649d34b46afbcc338cee981fc964e9d67 # Version: Commits on Jun 06, 2026 ------------------------------------------------------------------------------------------ frm2png can convert to/from indexed pngs (#487) * frm2png can convert to/from indexed pngs, -------------------------------------------------------------------------------------- omf2097.mk c3d4b3fae327ce086eb8ffa3985bd0b04de22a5c # Version: Commits on Jun 05, 2026 -------------------------------------------------------------------------------------- Merge pull request #1313 from omf2097/adt/mod-support Mod support, --------------------------------------------------------------- xbox-xcloud-client.mk v2.4.2 # Version: Commits on Jun 05, 2026 --------------------------------------------------------------- - Fix frontend error message when token was expired #1581 - Added German language (credits to @crt0mega) #1567 ⚠� V2 is in maintenance mode. This means that only critical bugs will be fixed and no new features will be added. ℹ� V3 is being worked on that should include a lot of fixes and new features, no ETA yet!, --------------------------------------------------------------------------------------- mangohud.mk 92ed51ff2f1b824d65823a811eed2620dc78f956 # Version: Commits on Jun 05, 2026 --------------------------------------------------------------------------------------- vkroots: don't install header, ---------------------------------------------------- rclone.mk v1.74.3 # Version: Commits on Jun 05, 2026 ---------------------------------------------------- This is the v1.74.3 release of rclone. Full details of the changes can be found in [the changelog](https://rclone.org/changelog/#v1-74-3-2026-06-05). , ---------------------------------------------------------------------------------------------------- sdl2-gamecontrollerdb.mk 2eabb42cb8255367d6f6043661eb8ccc77f49639 # Version: Commits on Jun 06, 2026 ---------------------------------------------------------------------------------------------------- Added SteelSeries Nimbus Cloud (#955) * Added SteelSeries Nimbus Cloud So [SteelSeries Nimbus Cloud](https://steelseries.com/gaming-controllers/nimbus?model=cloud) is an Android and iOS controller that somehow works natively on that platforms in both extended shell wired mode and Bluetooth mode. The other platfrms are supported only in Bluetooth mode. There is a mention of PC among supported platforms, however in my experience that turned out to be false. I used [Gamepad Tool](https://generalarcade.com/gamepadtool/) to generate SDL string for all three desktop platforms (Windows, macOS, Linux). I have no idea how to do the same on Android and iOS, so no mappings for them. That software showed `mapping available` out of the box on Linux and macOS, but not on Windows. However, the controller didn't work until it was mapped manually. This gamepad has two back buttons, they are mapped as `paddle1` and `paddle2`. * Remove CRCs --------- Co-authored-by: Alynne <alviscera@gmail.com>, ----------------------------------------------------------------------------------------- xpad-noone.mk a52e32cf182435d608e66bef9a6c89ebac891999 # Version: Commits on Jun 05, 2026 ----------------------------------------------------------------------------------------- Publish repo-stats report via GitHub Pages and link from README, ---------------------------------------------------------------------------------------- retroarch.mk bdba046fa6766380bc2457532f38e589df769aaf # Version: Commits on Jun 06, 2026 ---------------------------------------------------------------------------------------- video_thread: set wrapper-active flag before underlying driver init so shader-backend detection resolves the wrapped driver ident (#19092), --------------------------------------------------------------------------------------- vpinball.mk 4c302cfdffacc4accb37f19d136b805aa470267c # Version: Commits on Jun 05, 2026 --------------------------------------------------------------------------------------- ci: support PRs (@francisdb), combine workflows, switch to sha short, ---------------------------------------------------------------------------------------- doomretro.mk 26a99671a5135b74596bc59a4a3b66fd7e08b697 # Version: Commits on Jun 06, 2026 ---------------------------------------------------------------------------------------- Fix wrong berserk power-up sprite in Freedoom, -------------------------------------------------------------------------------------- yquake2.mk 8c8c3117a00d896b473b4c24d4af1f226ff0ddcb # Version: Commits on Jun 05, 2026 -------------------------------------------------------------------------------------- game: fix SPAWNFLAG_MONSTER_DEAD usage without self->monsterinfo.currentmove, ------------------------------------------------------------------------------------------ xash3d-fwgs.mk 51d1bfbe08b11671e4961e7c71474869bacd59b3 # Version: Commits on Jun 05, 2026 ------------------------------------------------------------------------------------------ 3rdparty: mainui :update, --------------------------------------------------------------------------------------------------- libretro-dosbox-pure.mk b2825c2f0a705f93d3c4d1b4369c8728d76f2e07 # Version: Commits on Jun 05, 2026 --------------------------------------------------------------------------------------------------- Avoid long freezes and potential crashes when using a .conf file with an [autoexec] section that mounts a directory with an extreme amount of subdirectories (#744), ----------------------------------------------------------------------------------------------- libretro-flycast.mk 751295151eed3cae619b14bcee49bb07f4fdb59c # Version: Commits on Jun 05, 2026 ----------------------------------------------------------------------------------------------- Fetch translations & Recreate libretro_core_options_intl.h, -------------------------------------------------------------------------------------------------- libretro-gearcoleco.mk 8a9e230d6bba04f2f6b95f9420cdf509c767978d # Version: Commits on Jun 05, 2026 -------------------------------------------------------------------------------------------------- Initialize overscan state with safe defaults, ------------------------------------------------------------------------------------------------- libretro-geargrafx.mk 946196768260457f39cd1e23b4cbae66dfdb6ebc # Version: Commits on Jun 05, 2026 ------------------------------------------------------------------------------------------------- [libretro] Clear inputs when controllers are changed, -------------------------------------------------------------------------------------------------- libretro-gearsystem.mk d7b61da873c5e0fcc51c2c6a1f0c526a3823b96e # Version: Commits on Jun 05, 2026 -------------------------------------------------------------------------------------------------- [libretro] Clear inputs when controllers are changed, ---------------------------------------------------------------------------------------------------- libretro-parallel-n64.mk bec8bd830566f2d9fcc04d68291c2fe85f67f99d # Version: Commits on Jun 06, 2026 ---------------------------------------------------------------------------------------------------- glide64: clamp CopyFrameBuffer's destination extent to RDRAM CopyFrameBuffer() writes width*height pixels to RDRAM at the colour image address with no bounds check. The height comes from rdp.frame_buffers[]/rdp.ci_* bookkeeping, which is populated by the HLE display list pre-scan and is never maintained on the LLE path, so under a low-level RSP it can be garbage. GoldenEye 007 with the cxd4 RSP plugin hits this from rdp_setcolorimage: the copy runs far past the 8MB RDRAM array and tramples unrelated globals - in the Linux build it corrupts gfx_info's DPC register pointers and crashes inside the list processor; which global dies depends on link layout, so other configurations crash elsewhere from the same stomp. Clamp the copy height so the destination stays inside RDRAM, the same treatment the texture load paths already apply (glide64_gDP.c), with one element of slack for the 16bpp paths' (x + y * width) ^ 1 store. Copies that previously fit are unaffected: GoldenEye under HLE and SM64 under both HLE and cxd4 produce bit-identical RAM traces before and after. GoldenEye under cxd4 previously crashed around the boot movies and now runs. Verified: GoldenEye glide64+cxd4 1200 frames clean (previously segv at ~335 VIs, also present before the frame-boundary refactor at the identical emulated state); GoldenEye glide64+HLE 1200-frame trace bit-identical to an unpatched build; SM64 glide64 HLE and cxd4 400-frame traces bit-identical to baselines., --------------------------------------------------------------------------------------------- libretro-pd777.mk 57bf25430c94239964e3fae3fd912678c9a63a10 # Version: Commits on Jun 05, 2026 --------------------------------------------------------------------------------------------- Fetch translations & Recreate libretro_core_options_intl.h, ---------------------------------------------------------------------------------------------- libretro-ppsspp.mk ecc35b904a24a570201e4b6cf3f12473a97ed3ec # Version: Commits on Jun 05, 2026 ---------------------------------------------------------------------------------------------- Merge pull request #21792 from hrydgard/fix-another-depth-issue Fix menus in NBA2K12., ---------------------------------------------------------------------------------------------- libretro-prboom.mk 6767e41d275c52a9309680966f8d759f437f5c5d # Version: Commits on Jun 06, 2026 ---------------------------------------------------------------------------------------------- r_segs: restore the dcvars fields the rounded and dithered drawers read Regression from 4f5ac4f (r_segs: skip filter-only per-column work when wall filtering is off). That commit keyed all the skipped work on filterwall == LINEAR, but the optional dcvars fields have two different selectors: - prevsource/nextsource are read by the RoundedUV wall drawers as well as LinearUV (both chosen by filterwall). With the wall filter set to ounded\ in General -> Video, the drawers dereferenced the NULL defaults: instant crash on the first rendered frame. - nextcolormap is read by the *_LinearZ dither drawers, which are chosen by filterz independently of filterwall. Z filtering with a point wall filter read a NULL nextcolormap: same crash. Split the predicate: prevsource/nextsource fill when filterwall is anything but POINT, nextcolormap when filterz is LINEAR -- the same split the span setup in r_plane.c already uses. The half-texel texturecolumn centering keeps its LINEAR-only condition from the original filter code. The default point/point path skips exactly the same work as before and its output is bit-identical (400-frame checksum run at 2560x1600); all nine filterwall x filterz combinations now render instead of crashing., ---------------------------------------------------------------------------------------------- libretro-stella.mk 9b2ef7f4407299f11b6099e867fd67fdb80e5c41 # Version: Commits on Jun 05, 2026 ---------------------------------------------------------------------------------------------- For macOS, determine home directory properly, instead of using a hardcoded path.,
Hancock33
added a commit
to Hancock33/batocera.piboy
that referenced
this pull request
Jun 14, 2026
----------------------------------------------------------------------------------- cemu.mk fe70fdcfe5ced5a4032b7c4561f7cad6f011314b # Version: Commits on Jun 13, 2026 ----------------------------------------------------------------------------------- debugger: Don't clear register views when continuing to run code, ------------------------------------------------------------------------------------------ duckstation.mk cfde5bf5c3418d3ae4be566dc21dca467da1c5b5 # Version: Commits on Jun 07, 2026 ------------------------------------------------------------------------------------------ Achievements: Move achievement pinning to database, -------------------------------------------------------------------------------------- openmsx.mk 3a40a511c97cbf42f86aa90394a4fe07ad455cd3 # Version: Commits on Jun 13, 2026 -------------------------------------------------------------------------------------- corrections and additions to the romdb, ---------------------------------------------------- pcsx2.mk v2.7.416 # Version: Commits on Jun 13, 2026 ---------------------------------------------------- - [Misc: clean up COM code](PCSX2/pcsx2#14562) , ------------------------------------------------------------------------------------- ppsspp.mk 14dc5a51560628581c023235e9d35ec28fb14b6a # Version: Commits on Jun 14, 2026 ------------------------------------------------------------------------------------- Merge pull request #21825 from hrydgard/ridge-racer-sprite Fix the ugly line on Ridge Racer lens flares, ------------------------------------------------------------------------------------ rpcs3.mk 4fed5e7b88d78c03359009377b740a3c826505a7 # Version: Commits on Jun 14, 2026 ------------------------------------------------------------------------------------ Fix iso integrity log message, ----------------------------------------------------------------------------------- ymir.mk bca07409fdd89d5e136356a063da71d9874d3424 # Version: Commits on Jun 13, 2026 ----------------------------------------------------------------------------------- feat(sys): Use GCD to ensure SH2 overclocking doesnt introduce slight clock errors to other components, ------------------------------------------------------------------------------------- ikemen.mk c42967d13e5b229f7dcf9450f471fd2a15d5858f # Version: Commits on Jun 13, 2026 ------------------------------------------------------------------------------------- Merge pull request #3705 from SuperFromND/bgdef-modelrotate feat: `modelrotate` for 3D BGDefs, --------------------------------------------------- ruffle.mk v0.3.0 # Version: Commits on Jun 14, 2026 --------------------------------------------------- ## What's Changed * avm2: Fixes to Error's prototype by @Lord-McSweeney in ruffle-rs/ruffle#23719 * aur: Update maintainers for ruffle-nightly-bin by @kjarosh in ruffle-rs/ruffle#23727 * Add testutils CLI by @Dinnerbone in ruffle-rs/ruffle#23677 * ci: Use correct npm dist-tag for nightly and stable releases by @kjarosh in ruffle-rs/ruffle#23728 * desktop: Add Rust stack trace to panic reports by @kjarosh in ruffle-rs/ruffle#23714 * avm2: Fix new Video(0, 0) to use default size by @1bugproof in ruffle-rs/ruffle#23708 * avm2: Stop assuming Object is an enum by @SuchAFuriousDeath in ruffle-rs/ruffle#23725 * Enable jpegxr on Linux Nightlies by @Fancy2209 in ruffle-rs/ruffle#23741 * avm1: Fix behavior of `Sound` ownership by @ChrisCPI in ruffle-rs/ruffle#23733 * chore: Bump the npm-minor group in /web with 7 updates by @dependabot[bot] in ruffle-rs/ruffle#23748 * chore: Bump enumset to 1.1.13 by @torokati44 in ruffle-rs/ruffle#23753 * chore: Bump cssnano to 8.0.1 and cssnano-preset-advanced to 8.0.1 in /web by @torokati44 in ruffle-rs/ruffle#23752 * chore: Run `npm audit fix` in web/ by @torokati44 in ruffle-rs/ruffle#23742 * video/vp6: Init nihav decoder with encoded size, not display size by @SuchAFuriousDeath in ruffle-rs/ruffle#23665 * web/audio: Pin AudioContext to 44.1 kHz by @SuchAFuriousDeath in ruffle-rs/ruffle#23701 * web: Fix MVP wasm build by explicitly disabling post-MVP features by @SuchAFuriousDeath in ruffle-rs/ruffle#23756 * avm2: Set dobj as transformed_by_script on transform `colorTransform` and `matrix` setters by @ChrisCPI in ruffle-rs/ruffle#23759 * avm2: Return ArgumentError instead of panicking on invalid ShaderJob target by @SuchAFuriousDeath in ruffle-rs/ruffle#23763 * chore: Migrate to current Node.js 26 from 25 for testing by @torokati44 in ruffle-rs/ruffle#23661 * avm2: Print full error details when logging caught error by @Lord-McSweeney in ruffle-rs/ruffle#23726 * avm2: Remove mostly unused `crate::avm2::string` by @Lord-McSweeney in ruffle-rs/ruffle#23773 * wstr: Remove unsafe usage from str_eq by @kjarosh in ruffle-rs/ruffle#23774 * desktop: Remove fontconfig from Windows and add fontconfig-dlopen by @kjarosh in ruffle-rs/ruffle#23768 * chore: Bump nihav-vp6 to latest git master by @torokati44 in ruffle-rs/ruffle#23765 * chore: Bump eslint-plugin-jsdoc from 62.9.0 to 63.0.0 in /web by @dependabot[bot] in ruffle-rs/ruffle#23781 * chore: Bump the npm-minor group across 1 directory with 7 updates by @dependabot[bot] in ruffle-rs/ruffle#23779 * chore: Fix a bunch of typos (mostly in comments) by @torokati44 in ruffle-rs/ruffle#23782 * chore: Bump qs and express in /web by @dependabot[bot] in ruffle-rs/ruffle#23784 * avm2: Fix stack trace for context3DCreated event handler by @Lord-McSweeney in ruffle-rs/ruffle#23745 * avm2: Verify that lazy namespace is actually a Namespace in the verifier by @Lord-McSweeney in ruffle-rs/ruffle#23777 * tests: Begin adding test coverage for signatures of flash.events classes by @Lord-McSweeney in ruffle-rs/ruffle#23786 * ci: Generate AVM2 report only for nightlies by @kjarosh in ruffle-rs/ruffle#23787 * avm1: Clear stack only between frames by @kjarosh in ruffle-rs/ruffle#23795 * avm1: Introduce AVM1 playerglobal by @Dinnerbone in ruffle-rs/ruffle#23730 * core: Fix movie clip looping by @kjarosh in ruffle-rs/ruffle#23802 * tests: Add --skip-no-player to testutils execute by @kjarosh in ruffle-rs/ruffle#23805 * tests: Trace warnings if enabled when executing tests by @kjarosh in ruffle-rs/ruffle#23804 * avm1: Do not trace an error on with(undefined)/with(null) by @kjarosh in ruffle-rs/ruffle#23809 * core: Normalize \blank\ target to \_blank\ in getURL/navigateToURL by @SuchAFuriousDeath in ruffle-rs/ruffle#23792 * avm1: Remove magic properties and this from SWF4 by @kjarosh in ruffle-rs/ruffle#23806 * avm2: Implement a number of missing `flash.events` classes by @Lord-McSweeney in ruffle-rs/ruffle#23812 * tests: Use FP's output in avm2/graphics_draw_triangles by @kjarosh in ruffle-rs/ruffle#23818 * chore: Bump josStorer/get-current-time from 2.1.2 to 2.1.3 by @dependabot[bot] in ruffle-rs/ruffle#23820 * chore: Bump the npm-minor group in /web with 2 updates by @dependabot[bot] in ruffle-rs/ruffle#23822 * chore: Bump the cargo-minor group with 3 updates by @dependabot[bot] in ruffle-rs/ruffle#23823 * chore: Bump webdriverio package family to 9.27.2 in web/ by @torokati44 in ruffle-rs/ruffle#23828 * avm2: More fixes to signatures of `flash.events` classes by @Lord-McSweeney in ruffle-rs/ruffle#23833 * avm1: Support property syntax in delete2 by @kjarosh in ruffle-rs/ruffle#23834 * 🤖 core: Move flash.text.engine TextLine onto its own DisplayObject by @doctorpangloss in ruffle-rs/ruffle#23817 * core: Extract run_file_dialog helper to reduce boilerplate in file dialog loaders by @SuchAFuriousDeath in ruffle-rs/ruffle#23831 * core/render: Add support for loading JPEGXR images by @ChrisCPI in ruffle-rs/ruffle#23789 * tests: Add asc compiler for compiling AVM2 tests by @kjarosh in ruffle-rs/ruffle#23794 * chore: Bump tmp from 0.2.5 to 0.2.7 in /web by @dependabot[bot] in ruffle-rs/ruffle#23843 * refactor: Enum-map related improvements by @SuchAFuriousDeath in ruffle-rs/ruffle#23846 * wgpu: Yeet the copy_srgb shader and pipeline by @SuchAFuriousDeath in ruffle-rs/ruffle#23827 * tests: Add a failing visual test for scale_rotation_cache by @torokati44 in ruffle-rs/ruffle#23796 * avm2: Minor performance improvements by @Lord-McSweeney in ruffle-rs/ruffle#23840 * core: Include `ruffle_render/jpegxr` in `jpegxr` feature by @ChrisCPI in ruffle-rs/ruffle#23852 * chore: Clippy beta fixes by @danielhjacobs in ruffle-rs/ruffle#23762 * ci: Enable fontconfig-dlopen in the macOS Rust tester job by @torokati44 in ruffle-rs/ruffle#23855 * core: Make loaded image MovieClips work the way SWF >= 5 movies do wrt. scale and rotation caching by @torokati44 in ruffle-rs/ruffle#23838 * avm2: Refactor draw_triangles_internal by @SuchAFuriousDeath in ruffle-rs/ruffle#23761 * core: Rework infrastructure around the FileDialogResult trait by @SuchAFuriousDeath in ruffle-rs/ruffle#23844 * text: Prevent measuring text twice during layout by @kjarosh in ruffle-rs/ruffle#23862 * core: Simplify code in MovieClip.event_dispatch by @kjarosh in ruffle-rs/ruffle#23867 * swf: Add support for PlaceObject/allEventFlags by @kjarosh in ruffle-rs/ruffle#23868 * wstr: Implement Pattern for owned [u8; N] and [u16; N] arrays by @SuchAFuriousDeath in ruffle-rs/ruffle#23858 * ci: Don't enable fontconfig-dlopen in the macOS Rust tester job by @torokati44 in ruffle-rs/ruffle#23873 * ci: Enable full LTO for releases by @kjarosh in ruffle-rs/ruffle#23863 * web: Polyfill document.embeds to return ruffle-embeds too by @danielhjacobs in ruffle-rs/ruffle#22449 * web: Support named access on document.embeds by @danielhjacobs in ruffle-rs/ruffle#23878 * core: Add MultiFileDialogResult and display_file_open_dialog_multiple by @SuchAFuriousDeath in ruffle-rs/ruffle#23866 * avm2: Add type enum for error 2006 by @ChrisCPI in ruffle-rs/ruffle#23883 * chore: Bump the npm-minor group in /web with 11 updates by @dependabot[bot] in ruffle-rs/ruffle#23885 * chore: Bump zbus, zvariant, toml crates by @torokati44 in ruffle-rs/ruffle#23887 * chore: Bump the cargo-minor group across 1 directory with 6 updates by @dependabot[bot] in ruffle-rs/ruffle#23886 * chore: Cargo.lock maintenance by @torokati44 in ruffle-rs/ruffle#23888 * core: Do not dispatch clip events not present in allEventFlags by @kjarosh in ruffle-rs/ruffle#23869 * Fix panic on writeback from empty CPU dirty region (prevalent in Earn to Die) by @n-raine in ruffle-rs/ruffle#23898 * avm2: Implement most of the version gating for `flash.events` classes by @Lord-McSweeney in ruffle-rs/ruffle#23836 * avm2: More minor crossbridge performance improvements by @Lord-McSweeney in ruffle-rs/ruffle#23849 * tests: add `known_failure` support to audio assertions by @moulins in ruffle-rs/ruffle#23861 * avm2: Fix setProgramConstantsFromByteArray implementation by @SuchAFuriousDeath in ruffle-rs/ruffle#23901 * frontend-utils: Use native TLS instead of rustls by @kjarosh in ruffle-rs/ruffle#23894 * chore: Bump openssl from 0.10.73 to 0.10.80 by @dependabot[bot] in ruffle-rs/ruffle#23902 * chore: Additional Rust dependency version bumps by @torokati44 in ruffle-rs/ruffle#23903 * avm1: Match Flash broadcaster prototype enumeration order by @SuchAFuriousDeath in ruffle-rs/ruffle#23892 * avm1: Stub several classes and properties by @SuchAFuriousDeath in ruffle-rs/ruffle#23899 * avm1: Define `constructor` as own property on system class objects by @SuchAFuriousDeath in ruffle-rs/ruffle#23893 * avm2: Add a field to ops instead of adding separate integral variants by @Lord-McSweeney in ruffle-rs/ruffle#23923 * chore: Bump the npm-minor group in /web with 6 updates by @dependabot[bot] in ruffle-rs/ruffle#23925 * avm1: Xmlsocket connect null fix by @SuchAFuriousDeath in ruffle-rs/ruffle#23931 * Slice logic cleanup, take 2 by @moulins in ruffle-rs/ruffle#23676 * core: Make use of `istr!` macro in several places by @SuchAFuriousDeath in ruffle-rs/ruffle#23935 * core: Remove unnecessary `dyn` usage by @SuchAFuriousDeath in ruffle-rs/ruffle#23936 * core: Use __jcvt() intrinsic on AArch64 for f64→i32 conversion by @SuchAFuriousDeath in ruffle-rs/ruffle#23937 * chore: Bump shell-quote from 1.8.3 to 1.8.4 in /web by @dependabot[bot] in ruffle-rs/ruffle#23941 * frontend-utils: Allow switching reqwest back to rustls by @torokati44 in ruffle-rs/ruffle#23947 * avm1: Remove \Target not found\ trace by @kjarosh in ruffle-rs/ruffle#23946 * avm2: Fix serializing invalid Date objects by @Lord-McSweeney in ruffle-rs/ruffle#23951 * swf: Seek to glyph shape offsets when reading DefineFont2/3 by @kjarosh in ruffle-rs/ruffle#23949 * + libssl-dev, openssl-devel by @sombraguerrero in ruffle-rs/ruffle#23953 * avm1: Support SWF4 target paths properly by @kjarosh in ruffle-rs/ruffle#23948 * chore: Bump WebdriverIO to 9.28.0 in web/ by @torokati44 in ruffle-rs/ruffle#23954 * swf: Fix underflow on keyPress clip action without key code by @kjarosh in ruffle-rs/ruffle#23958 * swf: Fix overflow while reading convolution filter by @kjarosh in ruffle-rs/ruffle#23961 * swf: Fix underflow/OOM errors when decompressing by @kjarosh in ruffle-rs/ruffle#23962 * core: Add AvmString::split_dependent by @SuchAFuriousDeath in ruffle-rs/ruffle#23955 * swf: Fix potential OOM in DefineSceneAndFrameLabelData by @kjarosh in ruffle-rs/ruffle#23960 * avm1: Fix SetTarget2 movie clip selection by @kjarosh in ruffle-rs/ruffle#23964 * chore: Update translations from Crowdin by @kjarosh in ruffle-rs/ruffle#23967 * avm1: Handle edge values in setProperty() by @kjarosh in ruffle-rs/ruffle#23966 * avm2: Simplify AMF serialization logic by @Lord-McSweeney in ruffle-rs/ruffle#23956 * avm2: Remove redundant method `PropertyMap::insert_with_namespace` by @Lord-McSweeney in ruffle-rs/ruffle#23969 * chore: Bump esbuild from 0.28.0 to 0.28.1 in /web by @dependabot[bot] in ruffle-rs/ruffle#23968 * chore: Update translations from Crowdin by @kjarosh in ruffle-rs/ruffle#23970 * refactor: Use api-report as dependency and move stub-report to tools/ by @kjarosh in ruffle-rs/ruffle#23922 * avm1/avm2: Add AMF NetConnection Array serialization known failures by @danielhjacobs in ruffle-rs/ruffle#23957 * tests: Add avm1/set_target_2_swf5 test by @kjarosh in ruffle-rs/ruffle#23965 ## New Contributors * @doctorpangloss made their first contribution in ruffle-rs/ruffle#23817 **Full Changelog**: ruffle-rs/ruffle@v0.2.0...v0.3.0, --------------------------------------------------------------------------------------- thextech.mk db3a01db31435fc4524bb42ab81d99b3f6aa0807 # Version: Commits on Jun 14, 2026 --------------------------------------------------------------------------------------- changelog.txt: move MDX to 1.3.7.3 changelog, ------------------------------------------------------------------------------------------- jazz2-native.mk a09fd6b6e54c42c9e5c622704b14a6ae32e06d1c # Version: Commits on Jun 13, 2026 ------------------------------------------------------------------------------------------- Tilemap draw call aggregation, improved player color customization, ----------------------------------------------------- ledspicer.mk 0.7.6 # Version: Commits on Jun 14, 2026 ----------------------------------------------------- LEDSpicer v0.7.6 - Release Notes Release Date June 14, 2026 Overview This release focuses on critical stability improvements and bug fixes for crafted profile handling and memory management. � Bug Fixes Crafted Profile System Fixed use-after-free crash when loading the same crafted profile twice Improved profile caching - Crafted profiles are now properly cached by game name instead of being rebuilt on every request Fixed profile stack history - FinishLastProfile now correctly steps back one profile at a time through the history stack Flag Handling Corrected FORCE_RELOAD flag - Profile reload now properly ignores cache and rebuilds in place Corrected REPLACE flag - Profile replacement logic now works correctly on the craft/load path � Changes Configuration Updates Configuration values now use proper boolean capitalization (True/False instead of true/false) Enhanced configuration defaults in ledspicer.conf Code Quality Refactored profile management to eliminate redundant cache operations Improved memory management with proper reference counting for replaced profiles Enhanced logging for better debugging of profile lifecycle Documentation & Branding Updated README with improved project description and branding Enhanced feature list and typical use cases documentation Updated copyright year to 2026 CI/CD Updated GitHub Actions checkout action from v4 to v6 Package Updates Updated version across all package managers (Debian, RPM, Arch) 🔧 Technical Details Changed default platform parameter handling in craftProfile() Introduced replaceProfileReferences() for cleaner memory management Optimized cache key handling for platform templates vs. crafted profiles ---------------------------------------------------------------------------------------- retroarch.mk 2669eaad7b4318c415028e3d2feaf9d7f092885e # Version: Commits on Jun 14, 2026 ---------------------------------------------------------------------------------------- Fetch translations from Crowdin, ---------------------------------------------------------------------------------------- doomretro.mk e011e4abf085fbbf5d7c9ba862bab40f48543bed # Version: Commits on Jun 14, 2026 ---------------------------------------------------------------------------------------- Fix double use of uppercase(), -------------------------------------------------------------------------------------- yquake2.mk dcda9144d34a5d76ff462675ff05d8acde30ecfe # Version: Commits on Jun 13, 2026 -------------------------------------------------------------------------------------- models: fix MDR uncompressed-frame heap overflow from byte/float count mix-up, ------------------------------------------------------------------------------------------ xash3d-fwgs.mk 1e31fbcd513bdabd557a098fcdd06fa2037cd99e # Version: Commits on Jun 14, 2026 ------------------------------------------------------------------------------------------ ref: shut up DecalShoot about decals must hit mod_brush This is an annoying message that's caused by the game code. Decals, without custom renderer, cannot be rendered on non-brush entities, and everybody knows that anyway., ------------------------------------------------------------------------------------------------------- libretro-beetle-pce-fast.mk 718dd0abb9ffcd95308ae79594e86d15148d7a2f # Version: Commits on Jun 14, 2026 ------------------------------------------------------------------------------------------------------- libretro: add zero-copy software-framebuffer direct render path Use RETRO_ENVIRONMENT_GET_CURRENT_SOFTWARE_FRAMEBUFFER so the VDC can render straight into the frontend's (often GPU-mapped) framebuffer, removing the surface->frontend copy that video_cb otherwise performs. The renderer already writes through a single MDFN_Surface target whose pixels/pitch it dereferences at render time (target = surface->pixels + (frame_counter-14)*surface->pitch, all Mix*/Draw* honour surface->pitch), so the direct path needs no renderer changes: in retro_run, before Emulate, request a framebuffer; if granted, repoint surf->pixels/pitch at it for the frame, render, ship video_cb from there with the framebuffer's pitch, then restore the persistent fallback surface. Why this is safe: - The VDC renderer is write-only on the target surface (verified: every reference is a store), so an uncached framebuffer (no RETRO_MEMORY_TYPE_CACHED) is fine; access_flags is WRITE only. - We do NOT rely on persistent buffer contents: for every scanline in the shipped rect [DisplayRect.y, +h) the renderer writes the whole line (Mix + DrawOverscan sides, or a full DrawOverscan in burst/blank), so a frontend that rotates buffers per frame yields no garbage. - Only used when the granted buffer is RGB565 (our native output), large enough (>= FB_WIDTH x FB_HEIGHT), and has an even byte pitch; otherwise the calloc'd fallback surface is used unchanged. - Frameskip frames don't request a buffer (video_cb(NULL,...) as before). - If the frontend doesn't implement the call, the result is latched off (fb_direct_supported) so we probe only once. surf->pixels/pitch are always restored after the frame, so the fallback surface is intact whenever a frame declines the framebuffer. Pitch is taken from the framebuffer (bytes -> uint16), not assumed to be FB_WIDTH. (The old psp_hw_render branch rendered its SW path into its own surface and used the PSP GU for HW; it predates this env call, so there was no code to port - only the confirmation that the renderer funnels through one retargetable surface pointer, which makes this drop-in.) Verified: full clean build (make -j4, exit 0, .so, 0 warnings); libretro.c passes -std=gnu89 -Werror=declaration-after-statement., -------------------------------------------------------------------------------------------------- libretro-beetle-psx.mk d460f8342060526678e7fd8222048324c2a80d86 # Version: Commits on Jun 14, 2026 -------------------------------------------------------------------------------------------------- gpu: SSE2-direct Gouraud RGB lane build in DrawSpanVec_NT (no stack roundtrip) Replaces the scalar-loop-then-_mm_loadu_si128 Gouraud channel build with direct SSE2 lane construction. The original for (l = 0; l < step; l++) { R8[l] = (uint16_t)CHAN(cr, dr, l); G8[l] = (uint16_t)CHAN(cg, dg, l); B8[l] = (uint16_t)CHAN(cb, db, l); } R = _mm_loadu_si128((const __m128i*)R8); /* + G, B */ lowered under gcc 13 -O3 to per-lane leal (cr + dr*l) into GPRs, store to stack at R8[l], movd back to xmm, punpck-chains. This shape was the dominant spill source in the GOURAUD-on (G1) non-textured (T0) hot loops -- in the unpatched gpu.s, the .L31928 family contained loops with 97-116 xmm operations and 38-39 stack writes each, across every T0 + (BMopaque | BMavg | BMaddq | BMadd) blend specialization. The contributing mechanism: the per-lane GPR live ranges produced by the scalar loop straddle the dense SIMD blend body that follows, so the register allocator chooses to materialize each lane's GPR result on the stack before reloading into xmm. Replacing the loop with straight SSE2 ops (vbase broadcast + step vectors + paddd + psrld + packssdw) keeps the channel intermediates xmm-resident across the entire blend body. Approach: lane l holds ((base + d*l) >> 24). d2..d7 are produced via paired adds (cheaper than the imuls gcc would pick for d * l, and the 32-bit lanes have plenty of headroom -- worst-case dr*7 has fewer than 25 bits given the Gouraud range). After the >> 24 shift each lane is <= 255, so the final packs_epi32 down to u16x8 cannot saturate. setr_epi32 calls fold to a single mov-from-rodata when dr/dg/db are constant per span (the common case at the polygon level), and to a small insert chain otherwise. Spill counts (gcc 13.3, -O3, x86_64): function before after G_Command_DrawPolygon_NV4_G1_T0_BMopaque* spills=39 spills=0 G_Command_DrawPolygon_NV4_G1_T0_BMavg* spills=39 spills=0 G_Command_DrawPolygon_NV4_G1_T0_BMaddq* spills=39 spills=0 G_Command_DrawPolygon_NV4_G1_T0_BMadd* spills=39 spills=0 G_Command_DrawPolygon_NV4_G1_T0_BMsub* spills=39 spills=0 The remaining spill leaders shift to the T1 (textured) variants at xmm=16 spills=28-30; those follow the same scalar-then-pack pattern on UV coordinates and are a separate follow-up. End-to-end bench (x86_64 Linux, gcc 13.3, full DrawSpanVec_NT BMopaque path -- Gouraud build + dither-off + pack RGB565 + bg load + write-back, 100000 spans x 320 pixels): -O3: 21.0 -> 14.8 ms (1.42x) 1520 -> 2165 Mpix/s -O2: 21.9 -> 15.8 ms (1.39x) 1455 -> 2028 Mpix/s Bit-exact across 500000 random Gouraud lane verifications (dr range 10..24 bits, both step=4 and step=8, plus edge cases at dr=0 and dr=0xFFFFFFFF) and 200 random spans rendered into VRAM. The NEON / scalar path is unchanged. On NEON, the original `vld1q_u16(R8)` plus per-lane access via vget_lane_u8 lowers to direct register extracts without the stack roundtrip the SSE2 path saw, so the same gcc lowering bug doesn't apply. ARMv7 NEON specifically may need its own audit but the codegen on AArch64 doesn't show comparable spill density., ---------------------------------------------------------------------------------------------- libretro-fake08.mk 814991a2571ad3970e386cef48f3b148aa1c27b9 # Version: Commits on Jun 13, 2026 ---------------------------------------------------------------------------------------------- Merge pull request #267 from ysamlan/fix-vibrato-modulation Fix vibrato effect: correct sawtooth to triangle wave modulation, ------------------------------------------------------------------------------------------------ libretro-gambatte.mk e99e1bd9b91de67ac12c77c3679c85447c26e8c8 # Version: Commits on Jun 14, 2026 ------------------------------------------------------------------------------------------------ ppu: extend BG palette expansion to the CGB fast path Follow-up to a991736 (\ppu: precomputed BG palette expansion for DMG fast path\). That commit treated only doFullTilesUnrolledDmg; the CGB renderer at doFullTilesUnrolledCgb had the same eight-dependent- load gather shape but with per-tile palette selection driven by the attribute byte (bgPalette + (nattrib & 7) * 4) and was deferred pending an L1-pressure measurement against the [8][256][4] table. Microbench result: the working set was a non-issue. Even with all 8 palettes cycling per tile (worst case -- every attribute byte different), the 32 KiB total table is L1-friendly because the access pattern (16 bytes per palette per tile) hits the same line repeatedly within a tile. Typical CGB scenes use only a few palettes per frame, making the hot working set much smaller than the full table. Microbench (full do-while incl. ntileword and nattrib recompute, 5000 frames x 144 lines x 20 tiles, x86_64 gcc 13.3 -O3): Scenario A: 1 palette used (typical): old: 53.0 ms (272 Mtiles/s) -> new: 27.5 ms (523 Mtiles/s) 1.93x Scenario B: 8 palettes cycling (worst-case for L1): old: 53.0 ms (272 Mtiles/s) -> new: 27.2 ms (529 Mtiles/s) 1.95x Bit-exact across 50000 random trials. Implementation: - bgPaletteExpanded changes shape from [256][4] to [8][256][4]. Size goes from 4 KiB to 32 KiB (u32 build) / from 2 KiB to 16 KiB (VIDEO_RGB565 / VIDEO_ABGR1555 build). The DMG renderer indexes the [0] slot; the CGB renderer indexes [nattrib & 7]. - refreshBgPaletteExpansion gains a unsigned palette_id argument. DMG-mode callers pass 0; the CGB hooks pass the slot affected by the BCPD write (index >> 3). Refreshing one palette costs 256 iterations of the 4-lane inner expansion -- about 1024 stores -- so per-palette refresh keeps writes proportional to the work the palette change actually invalidated. - refreshPalettes (the bulk-reload path used by loadState, swapToDMG, the colour-correction setters) rebuilds all 8 slots in CGB mode and just slot 0 in DMG mode. doCgbBgColorChange (the one-entry-at-a-time BCPD path) rebuilds the affected slot. dmgBgPaletteChange (BGP register) rebuilds slot 0. These are the only paths that write ppu.bgPalette in the codebase. Codegen (gcc 13.3 -O3, x86_64): both inner loops now have the same shape -- the do-while body lowers to four movdqu pairs per unrolled iteration (gcc unrolls by 2, processing 16 pixels per loop body), plus the nattrib/ntileword recompute scalar tail. The CGB path's 32 KiB worst case is still smaller than every modern L1d (x86 32-48 KiB, ARM Cortex-A53/A72 32 KiB), and the typical hot working set is one or two 4 KiB slots., -------------------------------------------------------------------------------------------------- libretro-gearcoleco.mk e6c987725eb2ac6dc0edc85b0dea30fdc7e938d7 # Version: Commits on Jun 14, 2026 -------------------------------------------------------------------------------------------------- Update readme, ------------------------------------------------------------------------------------------------- libretro-geargrafx.mk 31e8fb7d94e4e4b723790fbd19a02760df5ac816 # Version: Commits on Jun 14, 2026 ------------------------------------------------------------------------------------------------- Update readme, -------------------------------------------------------------------------------------------------- libretro-gearsystem.mk a4f2cf7fdcb946ca3d150206c1b3e426fabd59c1 # Version: Commits on Jun 14, 2026 -------------------------------------------------------------------------------------------------- Update readme, --------------------------------------------------------------------------------------------- libretro-opera.mk 64f35570aad94992f1122a810af6cc932862505a # Version: Commits on Jun 14, 2026 --------------------------------------------------------------------------------------------- MADAM: ignore bottom 2 bits in CCB processing The bottom two bits of the address bus aren't even wired up to MADAM. Most titles use properly aligned addresses but not all. This patch ensures the pointers read from the CCB and the CCB address itself are masked to align them to word boundry. Fixes Ballz and possibly other titles that had unaligned addresses., ---------------------------------------------------------------------------------------------------- libretro-parallel-n64.mk 2eff7ef440c8411d6b8670ea32d9b5a40492db12 # Version: Commits on Jun 13, 2026 ---------------------------------------------------------------------------------------------------- libretro: only auto-force HLE graphics when the RSP plugin is on auto The core forced the GFX task through the HLE emitter whenever the active video plugin was not the one paired with the build's native low-level path (angrylion for the threaded software build, parallel for the Vulkan build). That override ran unconditionally, so selecting the cxd4 RSP explicitly while using a different video plugin silently executed the HLE display-list emitter instead of the requested low-level microcode, contradicting the user's choice and making the two paths impossible to compare. Gate the override on the RSP plugin being left on \auto\: an explicit cxd4 or parallel selection is now honoured as low-level, while auto keeps the previous convenience fallback., ---------------------------------------------------------------------------------------------- libretro-ppsspp.mk 14dc5a51560628581c023235e9d35ec28fb14b6a # Version: Commits on Jun 14, 2026 ---------------------------------------------------------------------------------------------- Merge pull request #21825 from hrydgard/ridge-racer-sprite Fix the ugly line on Ridge Racer lens flares, ---------------------------------------------------------------------------------------------- libretro-prboom.mk 84cd28bd2d58bbfc8a2e48997862fefcac132c6e # Version: Commits on Jun 14, 2026 ---------------------------------------------------------------------------------------------- Revert the above-eye 3D-floor water chain (translucent ceiling approach) This reverts the four-commit chain that drew a swimmable water surface seen across an opening from below it: 1ae278f draw above-eye 3D-floor water as a translucent ceiling plane bb17de3 tint the submerged volume for above-eye 3D-floor water 554a06b blend 3D-floor water at the map's alpha instead of fixed 50/50 4327cc2 give 3D-floor water a solid colour when its flat is a bare marker The approach is structurally wrong and cannot be salvaged by tuning. It draws the water as a single flat screen-space translucent span. But the far wall of the flooded room spans from its floor (~z -16) to its ceiling (z 448) within one screen column, and only the part below the surface (z 138) is underwater. A flat span cannot express 'underwater up to the waterline, air above it' when the far geometry crosses the surface mid-column, so the tint covers far above the real water line regardless of clamp direction or alpha. The result reads as a blue sheet hung over the top of the opening, not a water surface -- visibly worse than drawing nothing. Correct 3D-floor water needs the surface clipped as a true horizontal plane against the per-column far geometry (a real waterline), which is a substantial renderer feature, not a span tweak. Reverting until that exists rather than shipping a broken effect. Restores the exact pre-chain state (bb842a3); vanilla demo hash unchanged (ab8eb93a72c8a66b)., ------------------------------------------------------------------------------------------- libretro-ps2.mk 9b74364f0111b2159a47b8a3b12adbb2e255604f # Version: Commits on Jun 14, 2026 ------------------------------------------------------------------------------------------- common/GS: load and run on Windows 8 by resolving placeholder APIs at runtime The fastmem area (HostSys.cpp) and the GS local-memory mirror (GS.cpp) called VirtualAlloc2 / MapViewOfFile3 / UnmapViewOfFile2 directly. These are Windows 10 1803+ exports, so as ordinary imports they land in the DLL's import table and the loader cannot resolve the module on Windows 8 / 8.1 - the whole core fails to load before any code runs, even though both subsystems can operate without those APIs. Resolve the three placeholder APIs once via GetProcAddress from kernelbase.dll (added as PCSX2_HasPlaceholderAPIs in RedtapeWindows.h) so they stay out of the import table, and branch on availability: - HostSys fastmem: if the APIs are missing, Create() returns null and vtlb_Core_Alloc falls back to no-fastmem mode (the same path already taken when a 4 GB region can't be placed). Map()/Unmap() call through the resolved pointers, which are guaranteed present because the area only exists when Create() found all three. - GS wrapped memory: GSLocalMemory cannot tolerate a null mapping (it memsets the buffer immediately), so a Win8 fallback is provided that reserves a contiguous repeat*size hole, releases it, and maps the file view into each slot with MapViewOfFileEx (available since Windows 2000). A flag records which path was used so the matching teardown runs in GSFreeWrappedMemory. On Windows 10+ the resolver finds all three APIs on first use and the existing placeholder workflow runs exactly as before - only the otherwise-unreachable absence path is new - so behavior there is unchanged. No effect on non-Windows builds. NOTE: unverified on a real Windows toolchain. The added code was compile-checked against a Windows target with MinGW (typedefs, call shapes and control flow), but the full core was not built and this has not been run on Windows 8 or Windows 10. Needs a Win10 SDK build and a runtime check on both before merging., --------------------------------------------------------------------------------------------------- libretro-snes9x-next.mk 305ab6f1ab2c8dd03a70659a43001ec39bf9ac23 # Version: Commits on Jun 14, 2026 --------------------------------------------------------------------------------------------------- tile: register-only palette gather in TILE_ROW_PRE_SSE2 (no stack roundtrip) Replaces the storel-to-stack + 8 byte-loads + align-load palette gather with pextrw to GPR + pinsrw straight back into xmm. Touches the central TILE_ROW_PRE_SSE2 macro (six callers across the math-variant n1x1 helpers) and the open-coded SSE2 path in tile_draw_row_nomath_n1x1. Found via autovec-aware codegen audit: gcc 13 -O3 lowers _mm_storel_epi64((__m128i *)pix_buf, pix); col_buf[0] = palette[pix_buf[0]]; ... [eight byte indices] ... col_buf[7] = palette[pix_buf[7]]; into SEVEN redundant 16-byte stack stores of `pix` plus eight movzbl loads at distinct offsets. SLP analysis treats each byte-load as needing its own non-aliasing copy of the source vector, so the spill multiplier is one-store-per-byte-load. Reading the indices through pextrw goes register-only, and pinsrw places the gathered colors directly in xmm so the subsequent mask/blend uses no scratchpad either. Spill counts in the generated assembly (gcc 13.3 -O3 -msse2): function before after DrawTile16_Normal1x1 spills=14 spills=0 DrawTile16Add_Normal1x1 spills=15 spills=0 DrawTile16Sub_Normal1x1 spills=14 spills=0 DrawTile16SubF1_2_Normal1x1 spills=14 spills=0 The max-spill loop in tile.c (.L20167, xmm=94 spills=15) is gone; the worst residual loop now has 7 spills, mean dropped from ~14 to ~3. Microbench (x86_64 Linux, gcc 13.3, full per-row helper -- mask, db, gather, store): -O3: 150.3 -> 137.4 ms (1.094x speedup) 213 -> 233 Mrows/s -O2: 149.6 -> 137.1 ms (1.091x speedup) 214 -> 234 Mrows/s Bit-exact against the scalar reference over 200000 random trials, and across 4000 row calls in the bench harness. PIX_BUF / COL_BUF stay on the macro parameter list so the six callers compile unchanged; the unused stack arrays are DCE'd at -O2/-O3. The parallel NEON path (TILE_ROW_PRE_NEON, NEON branch of tile_draw_row_nomath_n1x1) is unchanged -- on NEON, vget_lane_u8 extracts directly into a GPR without a stack roundtrip, so the same gcc lowering bug doesn't apply; a NEON-side audit can come separately., ---------------------------------------------------------------------------------------------- libretro-stella.mk 6fcd49280bf923e171cfe87183d97ec20ce2fc99 # Version: Commits on Jun 13, 2026 ---------------------------------------------------------------------------------------------- added cassette icons for MP3/WAV files, --------------------------------------------------------------------------------------------------- libretro-swanstation.mk 93b213d805591c4f1488339c4a16f0b4cb68d44a # Version: Commits on Jun 14, 2026 --------------------------------------------------------------------------------------------------- gpu_sw: SSE2/NEON kernels for CopyOutRow16<RGBA8> and <BGRA8> The two desktop-output specialisations of CopyOutRow16 -- RGBA8 and BGRA8, used whenever the host display path is a 32bpp framebuffer (the common case on Windows/Linux/macOS GL/Vulkan backends) -- were the only ones left without a SIMD kernel. The RGBA5551 and RGB565 variants for mobile/GLES targets already had SSE2 and NEON paths. Each 16-bit MASB1555 source pixel expands to a 32-bit ARGB pixel with the same per-channel scaling pattern as the existing kernels (<<3 on each 5-bit channel; bit 15 expanded to 0x00/0xFF for RGBA, hard-coded to 0xFF for BGRA, matching the scalar VRAM16ToOutput specialisations bit for bit). The kernels process 8 pixels per iteration: r8 = (v & 0x1F) << 3 g8 = ((v >> 5) & 0x1F) << 3 b8 = ((v >> 10) & 0x1F) << 3 a_hi = (v >>arith 15) << 8 // 0xFF00 if STP bit set else 0 pair_lo = r8 | (g8 << 8) // R in low byte, G in high pair_hi = b8 | a_hi // B in low byte, A in high store unpacklo(pair_lo, pair_hi) store unpackhi(pair_lo, pair_hi) The 16-lane interleave on SSE2 (punpcklwd/punpckhwd) produces u32 lanes in little-endian R|G|B|A order, which is the RGBA8 byte layout. BGRA8 just swaps the channel pairing (B|G in pair_lo, R|A in pair_hi) and uses a constant alpha vector since BGRA8's VRAM16ToOutput hard-codes alpha = 0xFF. The NEON path uses vst2q_u16 to do the same interleave: storing {r|g, b|a} produces the [r|g, b|a, r|g, b|a, ...] u16 sequence which reads as [R|G|B|A, R|G|B|A, ...] in u32. Tail pixels (width % 8) keep the scalar VRAM16ToOutput call exactly as the RGBA5551 and RGB565 kernels do. Microbench (x86_64 Linux, gcc 13.3, full 640-pixel row, 500K iters): RGBA8: -O3: 148 -> 75 ms (1.97x) 2161 -> 4247 Mpix/s -O2: 351 -> 74 ms (4.74x) 911 -> 4317 Mpix/s BGRA8: -O3: 119 -> 66 ms (1.80x) 2683 -> 4827 Mpix/s -O2: 299 -> 64 ms (4.68x) 1070 -> 5004 Mpix/s Bit-exact across 33000 trials over widths 0..32 (exercising the scalar tail at every remainder), input pixels drawn from the full 65536 u16 range., ------------------------------------------------------------------------------------------------ libretro-tyrquake.mk 5e10328faf30161923fa0f918792fbeb7772ef5b # Version: Commits on Jun 14, 2026 ------------------------------------------------------------------------------------------------ r_aclip: batch clipped fan into a single D_PolysetDraw dispatch R_AliasClipTriangle clips an alias triangle against the frustum planes via Sutherland-Hodgman, producing a polygon of up to 7 vertices, then triangulates it as a fan around vertex 0. The old code emitted one D_PolysetDraw call per fan triangle, with the source carrying a 'FIXME: do all at once as trifan?' note above the loop. All fan triangles already share the same pfinalverts buffer (R_AliasClipTriangle's stack-local fv[pingpong]) and the same facesfront flag (inherited from the input triangle), so batching is mechanical: build a mtriangle_t array on the stack with vertindex[0..2] = (0, t+1, t+2) for t in [0, k-3], set numtriangles = k - 2, and dispatch once. mtri_fan[5] covers the worst-case fan size: a 3-vertex triangle clipped against 4 frustum planes (left, right, top, bottom) grows to at most 3 + 4 = 7 vertices, yielding a 5-triangle fan. Z-clip is handled separately upstream and never inflates the polygon past the per-plane growth of the XY clips. Performance impact on the SW path is modest (per-call D_PolysetDraw overhead is small after the static-spans change in the previous commit), but the structural change matters for two reasons. First, the GPU compute path in D_PolysetDraw (Phase 5b-06: the g_rhi->dispatch_3d_alias short-circuit) is dispatch-heavy -- one GPU submit per fan triangle on a partially-clipped entity is exactly what the unclipped batching in R_AliasPreparePoints (commit message at r_alias.c:477) was written to avoid. Bringing the clipped path under the same batching treatment closes the remaining per-triangle-dispatch hole. Second, the loop is now straight-line code with no global-state mutation inside the body (no per-iter writes to r_affinetridesc.ptriangles or to a shared mtri scratch), which simplifies reasoning about the rasterizer's preconditions when reading this code. Bit-identical output: the fan triangles are generated in the same order ((0,1,2), (0,2,3), (0,3,4), ...) and consumed by D_DrawNonSubdiv / D_DrawSubdiv in the same order (their inner loops iterate 0..numtriangles-1 with the same per-tri math), so the resulting pixel emissions are unchanged. Builds clean on x86_64, aarch64, and armv7 NEON crosses.,
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Noticed during development of #23889.
AVM1 enumerates own properties in reverse insertion order. The broadcaster methods on
MovieClipLoader,FileReference,FileReferenceList,System.IME,Stage,Key,MouseandSelectionwere defined in the order_listeners, addListener, removeListener, broadcastMessage, sofor-inyielded the reverse — the opposite of Flash, which yields_listeners, removeListener, addListener, broadcastMessage.Define them in the matching reverse order so iteration lines up with Flash. Additionally, give
_listenerstheDONT_DELETEflag Flash assigns to it (previouslyDONT_ENUMonly).Testing
Good luck reviewing the diff :(
It seems to be true that _listeners comes prior to broadcastMessage everywhere I looked though (which is not every changed line), so it seems this change is correct. There are other order differences with Flash, so comparing it visually is difficult.
Checklist