• v3.0.0-rc.8 2f1c0c9c19

    Longshot v3.0.0-rc.8
    Some checks failed
    CI / Conventional Commit title (push) Has been skipped
    CI / Python source checks (ubuntu-24.04, part 1/2) (push) Has been skipped
    CI / Python source checks (ubuntu-24.04, part 2/2) (push) Has been skipped
    CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (push) Has been skipped
    ci/circleci: setup-main
    Third-party parts / The same as upstream (push) Has been skipped
    Guide screenshots / Guide images and local UI source review (push) Failing after 17s
    CI / Workflow lint (actionlint) (push) Successful in 22s
    ci/circleci: setup-release
    Third-party parts / As their notes say (push) Successful in 27s
    Release / appimage (push) Successful in 4m41s
    Release / The version (push) Successful in 19s
    Release / Linux builder (push) Successful in 0s
    Release / Longshot Builder zip (push) Successful in 1m44s
    Release / Windows Vulkan back end (llvm-mingw) (push) Successful in 42s
    Release / Windows builder (push) Successful in 0s
    Release / Sign and publish (push) Successful in 46s
    Pre-release

    HumanBean released this 2026-11-18 22:31:10 +00:00 | 149 commits to main since this release

    🎮 To play, get the Longshot Launcher

    🪟  Download the launcher for Windows

    🐧  Download the launcher for Linux (AppImage)

    Windows without installing: portable zip · the launcher's newest rc release

    Don't download the files below: they're the builder the launcher installs for you.


    🎉 126 pull requests were merged in this version!

    What's new in rc.8

    A faster game: up to 17 % more fps than rc.7 (Ryzen 9 7900X, RTX 3090, 1440p with DLSS, uncapped, Linux; one run each).

    • Castle Town's market: 64.1 → 75.0 fps.
    • Hyrule Field: 64.9 → 68.9 fps.
    • Lake Hylia: 61.9 → 72.3 fps, and its 1% low 32.1 → 37.3 fps.
    • The game's own work a frame is about 3 ms less in each of them (Market 13.1 → 9.6 ms): less work in Castle Town, cheaper draws and captures, and the game's hottest code compiled at -O2 in every build.
    • CPUs with 4 to 11 threads: Threaded rendering's Auto now picks by physical cores, the render thread from 4 cores and both threads from 6: +2.3, +4.8 and +6.1 fps at 4, 6 and 8 cores. On 4 cores, the game's job threads also stop oversleeping and yielding the whole processor: frames 3.2 % and about 2 % shorter in Castle Town's market.
    • The game's decompression runs as native code, which also ends the load freezes on an Intel Arc iGPU.
    • 4-6 GB cards give their VRAM to the memory the GPU reads most.
    • An update that changes no shader no longer rebuilds every pipeline, and a driver update warms the pipelines again before the game, not in play.

    Faster first builds: game files are unpacked while the game is translated, and the Builder runs at normal priority instead of below it.

    New:

    • NVIDIA DLAA before the game's bloom and tone map.
    • A Discord status that names where Link is, his age, hearts, Gold Skulltula tokens and pieces of heart; its text and picture are yours to set in longshot_options.ini.
    • NFC tags from your own dumps, with an NFC tab in the Insert menu.
    • Hand-drawn DualSense and Xbox button pictures.
    • The Longshot logo and version in the menu.

    Fixed:

    • Above 60 fps: Epona jumping fences, the Shadow Temple's barge carrying Link, the frozen Zoras, the water ripples, a song's notes, and Link's cap and other cloth.
    • Frame generation: made frames at their own times with V-Sync Off and On, a still HUD and moving frames under FSR 3.1 FG, and DLSS FG's multiplier capped at what the card supports (6x on RTX 50, 2x on RTX 40).
    • GTX 10 cards show the 3D at 100 % Render resolution.
    • The shader preparation and first start finish on Linux integrated GPUs (Mesa Intel, RADV APUs).
    • Mesh-only draws on Windows RDNA 2 cards use the mesh fallback.
    • A window behind another one no longer stops the GPU's work.
    • Bright lines along edges with Anisotropic filtering.
    • Ghosting with the camera inside Link.
    • Death Mountain Trail's khaki veil on Intel, and textures on the Radeon 860M.
    • Cutscene voices in a deep game folder, and builds in folders past 260 characters.
    • Closing the game waits for a save it is writing.
    • The saved logs leave out your user name.
    • The menu's language rows draw Русский, 日本語, 한국어 and 中文 in every menu language.
    • The mouse's side buttons in the game's menus.
    • A dump with many damaged files stops at once instead of running overnight.

    v3.0.0-rc.8 (2026-11-18), new since rc.7

    Features

    • prs: One ready-for-review list: tested on your own PC, or what can't be run by hand said; two new Everyone boxes (#1454)
    • vk: RX_VK_MEMORY_PRIORITY tells the driver which memory to keep in VRAM when it runs short (off by default) (#560)
    • nfc: NFC tags from the player's own dumps, with an NFC tab in the Insert menu (#280)
    • dev: LONGSHOT_TRACY=1 builds the game and the renderer with the Tracy profiler's zones, localhost only (#1543)
    • vk: A Present queue row on the Debug tab, for the presents' own GPU queue (#1554)
    • prompts: Hand-drawn DualSense and Xbox pictures for the System page, and the game's own tint (#1589)
    • menu: Longshot logo and name left of the tabs, version and a dev build name right of them (#1629)
    • perf: The game sends its own v2 summaries with replay phases (#1530)
    • vk: NVIDIA DLAA before the game's bloom and tone map (#1652)
    • discord: The status names where Link is, his age and stats; its text and picture are the player's to set (#1653)

    Fixes

    • build: The crash window copies log lines by a name the MSVC target links (no "undefined symbol: strdup") (#1379)
    • docs: One row a file for the tested runs of #1255 and #1256 (#1378)
    • perf: Fit perf summaries under the sender's limit, no stale or cutscene poses (#1383)
    • fs: A long path with a doubled separator opens: the cutscenes' voices in a deep game folder (#1346)
    • vk: The Anisotropic filtering row leaves images of one level alone, so no bright lines along edges (#910)
    • vk: Frame generation's made frames at their own times with V-Sync Off (#1362)
    • builder: A dump with many damaged files stops at once, not overnight, and a few find their stand-ins in one or two opens (#1453)
    • vk: FSR 3.1 frame generation keeps the HUD still: what the game drew over its picture after the upscale is marked, not guessed (#1464)
    • vk: The full-size network's stand-in draws at 100% Render resolution, so GTX 10 cards show the 3D there (#718)
    • vk: The swapchain's image is waited for on the CPU, so a window behind another one no longer stops the GPU's work (#1458)
    • vk: A load-at-create start whose mesh fallback builds stall in the driver goes on to the title, with progress on the way (#1450)
    • host: The saved logs leave out the player's user name (#1459)
    • vk: A finalized render target goes once the game has written over its storage (#1462)
    • vk: The shader preparation and first start finish on Linux iGPUs: Mesa's Intel driver and RADV APUs compile what the memory holds (#426)
    • vk: Frame generation's fixed multipliers on the game's timeline under V-Sync On (FIFO) too (#1442)
    • fps: Epona jumps fences above a 60 fps cap (#1434)
    • lift: SDIV of the most negative number by -1 gives the dividend, not a trap (#1423)
    • host: Closing the game waits for a save it is writing (#1422)
    • host: The samplers' stack walks stay on the walked thread's stack, so HOS_PROFILE=1 no longer crashes the game on Windows (#1343)
    • vk: Decode ASTC on the CPU for the Radeon 860M (#1296)
    • vk: Validate texture blits before vkCmdBlitImage (#1241)
    • vk: Keep mesh indirect argument ranges from overlapping (#1243)
    • vk: A sparse upload that runs out of staging binds what it queued (#1250)
    • vk: Cooperative matrix probe checks map and fence, keeps in-flight resources (#1247)
    • vk: Skip clears of textures kept in ASTC blocks (#1242)
    • build: The build runs from, in and into folders past 260 characters (#1132)
    • linux: A restart no longer hands the game its own log as stderr, so the new run moves it and the crash window finds it (#1059)
    • vk: An update that changes no shader no longer rebuilds every pipeline (#1418)
    • vk: Map the NVN 5-5-5 and BGRA8 formats by their bit layout (#924)
    • vk: Make the GPU's writes visible to the host's reads at each submission's end (#1031)
    • saves: The save script's Windows default for Longshot is the launcher's %APPDATA%\Longshot\save (#1397)
    • menu: The Save my pipeline list message is shown in the menu's language, and each language's count is its table's length (#1318)
    • perf: The performance sender reconnects when Windows aborts a kept connection (#1299)
    • vk: The shipped state list is made behind the preparing screen, and a half-written pipeline list is cut back (#1161)
    • fsr4: The tail is written before its reader also when the unrolled shader arrives between bind and dispatch (#1163)
    • vk: Leave out the depth draws of a mesh the game no longer shows: no ghosting with the camera inside Link (#1370) (#1447)
    • vk: The presents on their own compute queue again by default with NVIDIA's driver 621.24 and newer on Windows (#1476)
    • vk: DLSS FG's multiplier stops at its runtime's most + 1: 6x on an RTX 50, 2x on an RTX 40 (#1486)
    • vk: FSR 3.1 frame generation interpolates the picture from before the HUD and lays the frame's HUD over it (#1487)
    • vk: With V-Sync On, frame generation leaves out the made frames the refresh has no room for, so the game keeps its rate (#1496)
    • vk: The 2x hint names AMD FSR 3.1 only where FSR 3.1 makes the frame, not while DLSS FG makes it (#1497)
    • linux: The AppImage carries AMD's FidelityFX SDK licence beside the FSR 3.1 shaders built from it (#1500)
    • saves: The save script's Linux backups go beside the saves, where the launcher keeps its own (#1473)
    • vk: The overlays let go what a present's copy read only after every present still on the GPU (#1536)
    • vk: FSR 3.1 frame generation's made frames move again: its HUD-less colour is the presented frame's, sRGB-encoded under the HUD (#1538)
    • dev: A Tracy build captures at full rate again (fibres off unless asked for), and the game-code zones give a valid capture (#1546)
    • release: A read the forge cuts off or fails with a 5xx is made again, as in the launcher's copy (#1547)
    • fps: The Shadow Temple's barge carries Link and the enemies above 100 fps (#1528)
    • vk: The presents' own queue is opt-in again: RX_VK_PRESENT_QUEUE=1 (#1549)
    • input: The mouse's side buttons stay their controls' in the game's menus (#1329)
    • builder: Reject mixed runtime declarations in Windows packages (#690)
    • profiling: Reject incomplete or unusable chain captures (#682)
    • menu: The language rows draw Русский, 日本語, 한국어 and 中文 in any menu language, not ??? (#1562)
    • vk: The shader preparation's bar counts the compute programs at their cost, and leaves the prediction out on Most (#1618)
    • fps: Havok cloth keeps the clock above a 60 fps cap, solved at most 60 times a second (HOS_FPS_CLOTH_60=1, off by default) (#1485) (#1628)
    • vk: Raise the integrated GPU texture budget when ASTC is decoded on the CPU (#1559)
    • vk: Memory is added up and given back under pressure, host writes are marked, and late pages are found when a capture is taken (#1339)
    • fps: Link's cap and other cloth keep the game's clock below 60 fps (#1384)
    • fps: A song's notes show one at a time with its melody above a 60 fps cap (#1471)
    • vk: Let render targets go as newer ones together cover them, and pool their images at once (#943)
    • host: Native zstd keeps the virtual core during stream decodes (load freezes on an Arc iGPU) (#1632)
    • vk: A texture the GPU wrote, or whose upload failed, is uploaded again instead of kept stale (#1162)
    • vk: Say why HDR stays SDR when Windows has HDR on (#424)
    • fps: The water ripples step every frame above 60 fps instead of jumping (#1498)
    • vk: A driver update warms the noted pipelines again before the game, not in play (#789)
    • fps: The frozen Zoras stand still above a 60 fps cap (#1444)
    • vk: Use the mesh fallback for mesh-only draws on Windows RDNA 2 (#1650)
    • vk: No EQUAL depth margin for draws that blend (Death Mountain Trail's khaki veil on Intel) (#1631)
    • vk: Make a sparse BC image with the usage its sparse residency was checked for (#671)
    • vk: Draw quads with 8-bit indices on devices without index_uint8 (#595)
    • hle: Sizes and names the game passes are checked to lie in guest memory in more host functions (#1179)
    • linux: Follow deep slave graphs for the slow-disk warning (#655)
    • vk: A game ended while it closes, after a quit during the precompile, no longer counts as a crash (#498)
    • vk: The blend check of #1631 reads dynamic_state's program, so the renderer compiles again (#1659)
    • vk: Perf_report.cpp includes psapi.h again, so the Windows renderer compiles (#1665)

    Performance

    • host: The game's zstd decodes run on the host's libzstd by default (HOS_NATIVE_ZSTD=0: as lifted code) (#1402)
    • vk: A freed sparse mirror's buffer is kept for the next mirror of its size, not destroyed (RX_VK_SPARSE_SPARE_MB) (#1359)
    • vk: Custom with no preset behind it trims the large textures where Auto would pick Minimum (#1388)
    • build: The chunks an all-O2 build compiles are cut into parts (#1431)
    • vk: A save's textures are made and uploaded during its loading screen, not at their first draw (RX_VK_TEX_AHEAD=1, off by default) (#1122)
    • runtime: Keep the virtual core for an invalidate with nothing to wait for (#1150)
    • vk: Six small costs of a draw and of a submission's parts removed (#391)
    • vk: Give the VRAM mirrors to the pools the GPU reads most on 4-6 GB cards (#1141)
    • vk: Scan pending uniform updates in place (#472)
    • host: The game's texture heap is compacted at most once a second, not at every frame (HOS_TEX_COMPACT_MS) (#1478)
    • vk: A mirror sync's pages from the captures go in one compute dispatch, not one copy region each (#326)
    • vk: Parts of 240 draws, uniform updates in one copy, a cheaper write-back forget, and the frame's GPU idle measured (#1484)
    • vi: The vsync ticker waits on a high-resolution timer instead of yielding (#1235)
    • hos: Splice the host's memcpy, memset and string hooks into the lifted SDK (#1209)
    • build: The Builder runs the build at normal priority, not below it (#779)
    • vk: Defer speculative pipeline probes when a draw needs its shader (#1564)
    • runtime: Reduce Market frame stalls and Vulkan capture overhead (#1553)
    • vk: Warm listed compute pipelines on Most (#1567)
    • build: Overlap package extraction and translation on cold installs (#1569)
    • vk: Convert complete texture tiles eight rows at a time (#1603)
    • vk: Fit mid-size captures to tight spare budgets (#1605)
    • vk: Stream guest page-table ranges into bounded transfers (#1594)
    • vk: Respect Linux memory limits when loading the NVIDIA pipeline cache (#1595)
    • vk: Combine contiguous sparse buffer mappings (#1624)
    • lift: Lifted stores mark only GPU pages, by a page bitmap (RX_MARK_BITMAP) (#1617)
    • Less game work a frame in Castle Town (registers as locals, vector rewrites) (#1576)
    • runtime: A correctly rounded FMA ahead of suyu's exact bodies on builds without FMA3 (#1195)
    • vk: Use temporary pipelines for required draws by default (#1643)
    • vk: Reuse matching live shader stages on first use (#1596)
    • build: Ship a hot list, so every build compiles the game's hottest functions at -O2 (#1635)
    • vk: Threaded rendering's Auto chooses by physical cores, in three tiers (#1664)
    • host: The game's nn::os::YieldThread gives up its virtual core only, not the processor (HOS_OS_YIELD=1 as before) (#1663)
    • host: The game's job barriers wait for their mask words instead of sleeping 20 us (HOS_OOT_BARRIER_WAIT) (#1662)

    Documentation

    • forge: Every pull request description starts with a plain TL;DR and a Breakdown; the check suggests them where they are missing (#1470)
    • guide: The Graphics guide image is retained after #1309, #1362 and #1464 (#1466)
    • contributing: A PR's switch carries its recommended value from the first push; measurements gate leaving WIP (#1634)

    Code

    • vk: 111 kill switches are deleted, the host's switches join the table, and the per-PC decisions get one place and one rule (#1309)
    • shaders: 15 kill switches the translation read are deleted, and its per-PC rules read the device policy (#1390)

    Build

    • windows: The builder no longer downloads emsdk, which the Windows build never uses (#1620)
    • Add a mise game build task (#860)

    Tests

    • profiling: The five checks the profiling changes broke pass again, on Windows too (#1323)

    Earlier release candidates: rc.1, rc.2, rc.3, rc.4, rc.5, rc.6, rc.7

    Features

    • fps: A Dev level of the FPS counter, past Breakdown, with the renderer's own numbers (#8)
    • menu: The Game frame rate cap, Descriptor reuse and Sampler cache apply in play (#7)
    • saves: Import and export Hookshot saves; a cleaner saves script and guide (#20)
    • forge: Pull requests open as WIP until measured, with pr ready, wip, edit, update, status and a description check (#48)
    • forge: A Testing section in the PR template, and the whole checklist in every PR (#72)
    • release: Breaking: Releases on Forgejo, signed by one key; tools.lock; Forgejo Actions only (#13)
    • forge: Flags and settings is a note, never a gate; If applicable keeps only the boxes that apply; the PR template linked up front (#76)
    • forge: Pr review posts a review with inline comments (#121)
    • release: Big launcher download links in the README and on every release page, through magicbean.dev's permanent links (#148)
    • prepare: Compile shaders All / Most / None, Most skipping the predicted pipelines (#158)
    • website: Add Longshot player site and team pages (#104)
    • release: A launcher release page links its files through git-cdn.magicbean.dev (#166)
    • website: Pick the launcher's channel on the install page (#168)
    • input: Mouse camera Off, for playing without the mouse (#178)
    • log: A log line can be debug or trace, and the movie networks a card leaves out are trace lines (#177)
    • dev: Test a change against a base, as a pack of what it changes (#147)
    • forge: Pr comment --attach puts pictures in the comment (#254)
    • vk: Quick pipelines for draws whose full build is not done (#234)
    • gameplay: A Damage taken row (Half to Quadruple) (#138)
    • input: Hold-to-toggle for any bound button (#137)
    • audio: A Mono choice on the Audio output row (#136)
    • display: Hide the cursor in play once the pads or keys are used (Windows) (#245)
    • control: Stream sampled game state through the bridge (#58)
    • game: The shader caches and the button prompts in the game folder's cache, the logs in its logs (#321)
    • menu: Skip the title screen and load the newest save (#399)
    • input: Add a held or toggled Walk key with adjustable speed (#66)
    • menu: Type and save a custom output size in the Insert menu (#51)
    • runtime: Keep local crash and unclean-end records (#75)
    • control: Keep durable command-line test sessions (#47)
    • profile: HOS_HLE_STATS=1 counts HLE, host-call and NVN export calls (#153)
    • saves: Scrubbed saves in the repository, and oot_saves.py is now longshot_saves.py (#543)
    • host: A window right after a crash says what happened, with a Report button that copies a redacted report and opens the issue form (#593)
    • scripts: The pipeline tour, driven through the control bridge (#111)
    • vk: Enable PsychoV35 upscaler compatibility by default (#608)
    • ui: The HUD laid out for every screen shape at its 16:9 size, anchored to the screen's edges (#95)
    • saves: Kokiri-start, a save right after the opening (#610)
    • options: Settings in the launcher's user folder, so an update keeps them (#729)
    • website: A save scrubber page, longshot_saves.py scrub in the browser (#747)
    • builder: The game's own name and icon for its exe and window, from the .nup (#857)
    • release: An rc's notes list what's new since the rc before, then the earlier release candidates (#745)
    • vk: Add selectable CRT monitor and TV filters (#109)
    • forge: Pr create allows edits from maintainers (#905)
    • vk: Intel XeSS Super Resolution as a temporal upscaler (Vulkan, the player's libxess.dll) (#459)
    • vk: FPS counter Breakdown in the graph: the frame's parts, the render thread's lane, the game's waits by cause (#928)
    • scripts: Memtour.py, a memory tour through the control bridge (#570)
    • tests: Run_all.py --changed runs the checks that read what changed (#32)
    • release: A nightly's notes list what is new since the nightly before (#972)
    • debug: Add transition autosave control and quick menu shortcut (#614)
    • control: Relaunch, the game started again over the bridge as the Insert menu's Restart now does (#97)
    • log: The recording streams' line counts the descriptor pools made when a recording's own ran out (#387)
    • build: Translate shows its progress as each of the lift's steps ends (#359)
    • vk: Log the guest pools the page table cannot reach (#474)
    • vk: Built-in shaders from a folder with RX_VK_SHADER_DIR (#33)
    • texts: The ocarina microphone hint says your microphone, not the console's (#604)
    • options: A Shadow resolution setting keeps thin shadows near Link smooth (#773)
    • vk: FPS counter overhaul 2: size row, details page, steadier graph, Dev hardware footer (#998)
    • audio: Select the playback output device (#1001)
    • scripts: The PC's job queue (job_queue.py) and dev_build.py, whose builds share one translation and one set of objects (#1085)
    • dev: Dev_build.py on Windows too, with tools.lock's Windows build tools (#1095)
    • build: The lift key hashes only the environment that changes the lifted C (#22)
    • audio: Add microphone testing, level meter and gain (#1102)
    • vk: RX_VK_TEST_VRAM_MB takes the card's VRAM as at most n MiB wherever the renderer sizes something by it (#1107)
    • build: A new version's game files hard-linked from another game folder of the same .nup (#360)
    • audio: Support Linux playback output selection (#1205)
    • menu: A HUD layout row, the HUD at the screen's edges or in the middle 16:9 (#1152)
    • scripts: Memtour.py --repo-save, a tour from a repository save (#1221)
    • vk: FSR 3.1 frame generation for everyone: FSR 3.1 FG on Frame generation by, any multiplier and Auto, the HUD kept still, actor motion (#1103)
    • telemetry: Crash reports from the game's own window, and performance statistics (#983)
    • profiling: Find what a stutter is: the slow-frame recorder (HOS_SLOW_FRAMES), the main thread's time in the frame trace, and the cause of a long wait for a virtual core (#1218)
    • profiling: HOS_PROFILE=1, one switch for every profiling tool, into a folder of its own, and scripts/profile_report.py to read it (#1279)
    • scripts: Crash_groups.py lists the crash reports in named groups, the largest first (#1288)
    • graphics: Choose the monitor used for fullscreen (#1277)
    • vk: The renderer's switches in one table (gpu/vk/switches.def), and a start that says which are set (#1300)

    Fixes

    • linux: The AppImage carries zbic's zstd sources, so a build with NVIDIA DLSS compiles (#4)
    • Discord status as Longshot's own Discord application (#10)
    • Issue forms and a pull request template for Forgejo, and a bug report link that fills the form (#18)
    • A "Looking for the game or a ROM" link on the new-issue page (#21)
    • dlss: The DLSS model row's preset is taken once it holds it, not a feature for each one stepped past (#39)
    • release: The upload address is the organization variable RELEASE_UPLOAD_API (#79)
    • memory: Give commit back near Windows' commit limit instead of crashing (#77)
    • video: A decoder refuses submissions while 16 wait, so a movie's first seconds are not skipped (#110)
    • The setup merges the third-party tool into the builder's third_party, and the Hookshot item is back (#117)
    • vk: One slow load no longer takes an install back to loading before the title (#94)
    • forge: Pr status doesn't tell a work-in-progress PR to rebase (#132)
    • lift: A function with code below its entry starts at its entry (#131)
    • issues: The templates use the new labels kind/bug and kind/feature (#133)
    • forge: One process renews the git sign-in at a time, so renewals at once no longer lose it (#140)
    • release: A release file's download leaves the token behind when Forgejo redirects it to R2 (#154)
    • vk: On Most the Builder's progress bar leaves the prediction's part out; its check runs All and Most (#164)
    • builder: MainWindow.xaml is valid XML again, so the Windows builder compiles (#167)
    • release: The Windows zip carries tools/version.txt, so a release's game and builder know their version (#149)
    • menu: The version shows as players know it: v3.0.0-rc.1, Nightly 46 · 434712bb (#170)
    • menu: A new install starts at its screen's size where that is smaller than the game's own, and no longer asks to restart (#172)
    • linux: A Linux build with AMD FSR 4 compiles its own Vulkan back end, from AMD's FSR API headers the AppImage now carries (#174)
    • builder: Build times measured on nightly-56: the docs, the Builders' estimates and when to measure again (#175)
    • vk: Auto (this PC) reads the primary monitor when there's no window, so an update no longer lowers the resolution (#176)
    • website: Remove game branding from public pages (#193)
    • discord: The Discord status says Longshot, not the game's title (#194)
    • vk: Give the game's occlusion reports the GPU's counts, so it culls what is hidden (#197)
    • oot: Preserve Scarecrow's Song recording and playback timing (#78)
    • linux: Warn about low physical RAM even with ample swap (#54)
    • vk: An opt-in RGBA16F route for PsychoV35 HDR with DLSS and NVIDIA frame generation (#108)
    • vk: RX_VK_JOB_STEAL=true no longer turns off a draw's pipeline build where it waits (#233)
    • runtime: Let the floor web react above 60 fps (#49)
    • vk: OoT's clustered lights sample their own shadow maps (handle tables as descriptor arrays, from Lichii) (#107)
    • menu: Keep render scale when switching upscalers (#55)
    • website: The download page's logo and favicon load, it looks like the site, and it links the docs (#257)
    • website: Recover progress collection and reject broken deployments (#258)
    • vk: Opt-in RGBA16F route for PsychoV35 HDR with AMD FSR 4 (#247)
    • vk: Occlusion counts stay visible unless known: one query numbering across queues, no count from a draw that didn't run (#260)
    • input: A mouse button released outside the game is not a click on return (#265)
    • menu: A layout preset is applied whole after a face button was cleared (#266)
    • setup: The setup copies the Builder from a zip unpacked deeper than MAX_PATH (#151)
    • saves: A copy that fails part way leaves the target's save as it was (#277)
    • menu: Remove stale Debug tools help translation keys (#273)
    • forge: Pr view and the other commands print UTF-8 into a pipe, so Windows doesn't stop on an emoji (#120)
    • scripts: Check_third_party reads git check-ignore in bytes, so Windows skips ignored files (#80)
    • linux: The log says why the renderer did not load (#34)
    • oot: The scarecrow's recall takes the note remaps too (kind 11) (#281)
    • vk: Order overlapping inline buffer updates (#329)
    • menu: Stop reporting guessed defaults in Bug reports (#103)
    • vk: HUD labels no longer turn to noise when a render target over their memory is evicted (#323)
    • tests: The game-free checks pass on Windows (#255)
    • vk: HUD labels the GPU drew no longer go blank after a zone change (#325)
    • vk: On Intel a pipeline cache's data is fetched with no build under way (#190)
    • vk: Intel's Windows driver no longer loses the device on float adds to storage buffers: they are the GPU's own atomic (#191)
    • vk: Task + mesh compute fallback by default for Navi 23 on Windows, one rule for the IDs (#219)
    • gameplay: Half damage uses the full hit before life is clamped (#269)
    • vk: On Intel iGPUs a split submission's uniform updates are written by its last part (#189)
    • hos: Report the game's mapped pages as used memory; commit the heap lazily on Windows (#324)
    • vk: A picture larger than the window is filtered over all its pixels, not sampled (#307)
    • runtime: A new install draws the game's 2560x1440, not its reduced 2048x1152 (#292)
    • docs: Keep new test runs out of the frozen table (#96)
    • vk: A slow movie shows its pictures, and a fault in Intel's Windows compiler leaves that shader out (#252)
    • vk: Apply PsychoV35 to the adapted scene permutation (#400)
    • vk: Actor motion pairs a model's copies by place, so the Gossip Stones stay whole under frame generation (#59)
    • vk: Report GPU resource waits beyond recording-slot reuse (#60)
    • builder: --fetch-ngx-sdk works through --cli without LONGSHOT_NGX_SDK (#465)
    • host: The core kept for the replay is one the process may run on (#522)
    • vk: Arc B-series no longer loses the device in Kokiri Forest: float adds are the GPU's own atomic (#444)
    • vk: On weak Intel laptops the shaders are prepared and the scene's passes draw within seconds, the pipeline cache is saved, and a Debug button sets the lightest rows (#404)
    • control: A launched game gets the supervisor's hidden console, not a console window of its own (#589)
    • scripts: Dev_packs builds a pack for a profile switch, and a renderer pack with the profile its base had (#602)
    • release: An upload the forge answers with a 5xx is made again, four times in all (#598)
    • control: The bridge's pipe lets a sandboxed game's agent in (AppContainer, a farm user) (#544)
    • build: One build at a time in a work folder, so a second build doesn't compile again what the first is compiling (#630)
    • builder: A game dump Windows refuses to read says why and what to do, not "Permission denied" (#641)
    • vk: The start-up screen counts the compute pipelines and says when it writes the pipeline cache (#721)
    • vi: Pace at 240 while a movie plays, so its player stops losing pictures (#488)
    • vk: Honor cache-off for rolling pipeline files (#415)
    • menu: The GPU device row keeps its card, not its place in the list (#736)
    • build: A romfs folder without the game's files is refused before the build (#534)
    • vk: The late pages' fix no longer puts older words over the game's newer ones (#526)
    • pipelines: Run All prediction after preparing on Most (#713)
    • vk: The graphics in AMD's desktop CPUs decode ASTC on the CPU, so Kokiri Forest no longer resets the GPU (#489)
    • vk: A recording without a draw stream is submitted instead of crashing (#552)
    • menu: Longshot_options.ini is replaced whole or not at all (#550)
    • vk: The page files' room to grow counts before the commit rule cuts the textures (#346)
    • vk: Arc B580 no longer draws white shards while moving: a part ends only right before a plain draw (#443)
    • forge: An expired sign-in isn't renewed where the new token can't be saved, so a sandboxed run no longer loses it (#855)
    • vk: Shader programs over 1 MiB, or with a stray branch after their end, are no longer left out (#427)
    • build: A romfs without the game's heap definition is refused too, also from a .nup and for an update (#863)
    • vk: Give memory back under Linux memory pressure and after the precompile (#223)
    • website: Neutral wording for every game name in players' issue titles, so they can't block deployment (#731)
    • vk: With the render thread off, a frame is no longer drawn before the data the game writes after its submit (Link dark for a frame) (#769)
    • vk: Give image descriptors a dummy view when nullDescriptor is off (#403)
    • vk: Auto settings take SteamDeck=1 as a handheld only on an integrated GPU (#559)
    • scripts: Hardware_report.py counts a Steam Deck as one, by whole words (#531)
    • host: A thread lowered right after it is created runs lowered on Linux (#496)
    • audio: Support up to 64 microphone devices (#751)
    • vk: Keep a copy into BC3 to whole blocks or the level's edge (#672)
    • vk: Read RX_VK_SHADOW_BIAS once, safely across recording threads (#661)
    • vk: Read and publish a compute program's pipeline atomically (#621)
    • vk: FSR 3.1 frees its images once the chain lets it go, not at its next upscale (#556)
    • vk: A timed-out read-back's source images are kept until the GPU is past the copy (#554)
    • vk: A read-back after a timed-out one waits for the GPU, and its textures are kept (#553)
    • linux: A saves move that fails part way brings back what it moved (#557)
    • A log file that cannot be opened no longer closes stderr (#558)
    • vk: On Linux the texture helpers started inside a replay run on every core again (#499)
    • vk: Image, image-buffer and texture-buffer arrays are arrays in the SPIR-V (#450)
    • vk: Storage image and texel buffer arrays are indexed nonuniformly where the GPU can (#502)
    • forge: Login --token, so signing in on another device doesn't sign this one out (#887)
    • vk: A mesh fallback program with bigger payloads reuses the record arena (#551)
    • input: Walk mode set to Hold in the Insert menu drops a toggled walk (#537)
    • prompts: The Keyboard prompts show the movement keys as they are bound (#529)
    • runtime: Handle windowless Windows stop requests (#102)
    • oot: The Lost Woods skull kids' songs take the note remaps too (kind 7) (#483)
    • vk: Release pages ending at a GPU import boundary (#469)
    • vk: Retain guest pages during import handoff and retirement (#468)
    • audio: A frame mixed for the old device is dropped when the new one has another channel count (#448)
    • menu: Show learned songs correctly in the inventory editor (#429)
    • build: A game image past 2 GB fails the link instead of running broken (#356)
    • input: An input still held after the Insert menu no longer blocks the pads' other buttons (#340)
    • vk: Samplers past the driver's allocation limit are reused, not lost (#442)
    • vk: The replay's own buffer writes stay in command order when unaligned or split (#440)
    • input: A Toggle button no longer latches the game menus' mouse clicks (#339)
    • hid: HOS_INPUT_SCRIPT_BACKGROUND no longer lets the player's keyboard into a test run (#319)
    • host: Longshot_env.txt can't set HOS_CONTROL_PEER_SID (#911)
    • control: Wait for busy Windows pipes using the native error code (#494)
    • control: Reject non-finite CLI numbers before connecting (#633)
    • builder: A compile that raises stops build.py at once, not after every queued compile (#540)
    • control: A command's reply wait ends at its deadline while events keep arriving (#538)
    • build: The object store's clean-up skips a folder it cannot read (#487)
    • prompts: A failed rebuild of the button pictures keeps the last good file (#486)
    • website: Improve mobile browsing and launcher downloads (#611)
    • vk: Seed_keys is filled only for RX_VK_SEED_CHECK, its one reader (#393)
    • scripts: Dev_packs.py makes a folder only when it isn't there, for the Windows farm sandbox (#948)
    • voice: A Voice Language saved in option.sav is applied after a restart (#970)
    • build: The hash record is written under its build's start, and keeps only the keys used (#35)
    • vk: Closing the game while its shaders are prepared no longer counts as a crash (#428)
    • vcores: A thread starved of its core gets it after 1 s, not an escape after 2 s (#262) (#594)
    • vk: Scan the import table as far back as its largest live import (#431)
    • vk: A timeline wait that comes back device-lost is said as a lost device, not as a timeout (#670)
    • linux: A start from the Steam Deck's Game Mode no longer compiles every pipeline the preparation built (#900)
    • vk: GPUs on the compute fallback load programs when the game makes them, not all 32,645 pipelines before the title (#975)
    • vk: Free on the GPU's progress, not on a frame count: the upscalers' and the headless overlays' frees (#788)
    • Honor WIP titles in local PR description checks (#600)
    • vk: The readback line also counts the waits for a whole submission whose pages were written in place (#700)
    • vk: Count every program not translated, past the 30 lines that name them (#423)
    • vk: The render thread may leave its core when another program takes it: no 30-40 ms stalls in Castle Town's market (#838)
    • docs: The Steam Deck page names places neutrally, and docs changes run the site's checks (#1016)
    • input: Invert mouse horizontal swaps the mouse's left and right, in native look and as the stick (#1028)
    • builder: RecompLauncher builds the game in its own install folder, not ~/Games (#1000)
    • website: Tell the launcher's version from the game's (#1055)
    • vi: Pace the 30 FPS cap in the host (#609)
    • build: A failure after the game linked keeps the translation and the game's objects; only the renderer's go (#1080)
    • builder: The update's running-game check counts only longshot.exe, not Longshot Launcher kept in the game folder (#1081)
    • vk: An allowance AMD's driver gives as the count itself is taken as past it (#1091)
    • vk: A submit's capture no longer misses the pages the replay's collection is marking again (#1083)
    • dev: A job cancelled on Windows ends with everything it started (#1100)
    • vk: Frame generation by offers DLSS FG, not Auto (#1041)
    • vk: A start that compiled its noted pipelines writes the pipeline cache before the title (#903)
    • wide: NPC speech bubbles stay over their speaker at every picture shape (#1191)
    • vk: Arc discrete cards no longer lose Kokiri Forest to the sky: uniforms snapshotted on every Intel GPU (#446)
    • vk: Frame generation by and CRT lines shown only when they apply; Frame pacing and Threaded rendering on the Debug tab (#1106)
    • vk: Float adds native on Panther Lake iGPUs, where the loop lost the exposure's adds (#816)
    • vk: A revisited area's ASTC textures take the finalized copies' images, not a decode (#1222)
    • vk: The write-back diff pass has a checkpoint of its own in device-lost reports (#1220)
    • ui: The text box's choices stay on a 16:10 or narrower screen (#1233)
    • linux: Guest memory as dma-bufs under Steam and Flatpak, where udmabuf's size_limit_mb can't be read (#1155)
    • setup: A rename blocked by a scanner is tried again, and the message carries Windows' error (#1138)
    • settings: A crash or a kill while a setting is written no longer leaves its file empty or cut short (#976)
    • oot: Let Gohma reach her ceiling position above 60 fps (#1098)
    • vk: A read-back pool whose command buffer or fence failed is made again, not used half made (#1251)
    • vk: Failed sparse binds leave their blocks unbound for the retry (#1248)
    • vk: Skip texel buffer views of formats the device cannot use (#1244)
    • audio: Auto counts 5.1 speakers, not channels (#1167)
    • hos: Wait at the commit limit for the host's own allocations, then say out of memory (#341)
    • build: The compiler check names a folder Windows lets no program list, and never says "no message" (#1238)
    • fps: The battle music stops after a fight above a 60 fps cap (#1224)
    • vk: The Intel shader compiles row's hint names the launcher when RX_VK_NO_INTEL_SERIAL holds it (#781)
    • build: A clang that can't read the play profile says which clang it is and where the one that reads it is (#1267)
    • vk: At Windows' commit limit the renderer's C++ allocations wait for memory, then say out of memory (#1291)
    • vk: Six rows of the switch table say what the code does (#1302)
    • vk: Leave the Vulkan drivers to the loader, no VK_ICD_FILENAMES from DriverStore (#1332)

    Performance

    • prepare: The segment heap, so the driver's compiles run side by side on Windows (#15)
    • vk: A program's descriptor counts are kept with it, and a build asks for the thread's writer once (#41)
    • host: The write marks' collection steps over 256 clear marks at a time (#63)
    • vk: A draw takes its thread's pass and descriptor writer once, not at every use (#198)
    • vk: A 6 GB card mirrors a quarter of its VRAM, not an eighth or 512 MB (#237)
    • vk: RX_VK_READBACK_LATE, the readbacks a rule names are read without their wait (opt-in) (#238)
    • host: Three per-call costs on hot paths removed (own-thread lookups, a Signal with no waiter, switches read once) (#156)
    • vk: Walk the captured ranges once a scan, not a binary search a page (#216)
    • vk: Fix the late pages on the GPU, word by word, instead of waiting for their write-back (#248)
    • vk: The recorder thread's slot resets without the wait and an idle that waits on its ring (RX_VK_RECORDER_OWN, RX_VK_RECORDER_IDLE=wait; opt-in) (#162)
    • vk: Sparse VRAM mirrors, memory only where a game pool is (on by default) (#250)
    • lift: A Windows lift runs one worker per 0.75 GB free, and makes the runtime tables beside the edits (#357)
    • prompts: The button prompts in ~20 s, not ~160 s, with the same files (#358)
    • vk: The replay's reserved core on Linux too (#445)
    • build: One object store for every work folder, so a new one compiles only what no other built (#371)
    • vk: The shader caches' stamp leaves out vk_api.h, so a menu, input or audio change keeps every player's translated programs (#370)
    • build: Compile giant lifted functions in pieces cut at their labels, on by default (LONGSHOT_SPLIT_GIANTS) (#385)
    • build: The game is built for the player's own CPU (-march=native), keyed by the CPU clang resolves it to (#524)
    • vk: The renderer is built with a committed play profile, about 3-5 % less frame time (#569)
    • build: A lift reused across work folders, so a new one skips translating the same game (#372)
    • vk: A render thread for the replay, behind a Threaded rendering setting (Auto: on from 12 logical processors) (#588)
    • vk: The renderer's play profile covers the compute fallback, the overworld and the dungeons (#599)
    • pipelines: 22,987 more pipelines prepared before the title, from a tour of every map, actor and cutscene (#563)
    • build: The release ships the third-party code compiled, so a player's build links it instead of compiling it (#629)
    • build: Keep unchanged button prompts in packaged builds too (#314)
    • vk: The VRAM budgets set by the card, not a menu row or a preset; 8 GB cards' mirrors 2 GB (#735)
    • vk: Arc cards read the occlusion counts from a GPU copy: Castle Town's market from 34.8 to 45.8 fps on a B580 (#761)
    • nvn: Recording threads wait less on the recordings' copy lock (#766)
    • vk: Flat per-draw caches, inline op payload copies and a flat c1 map in the replay (#231)
    • vk: The Arc B-series compiles shaders side by side by default: a new area no longer freezes the picture for minutes (#500)
    • vk: Below 60 fps the eviction's waits are seconds, not presents: the compaction's hold and the GPU-drawn textures' keep (#771)
    • build: An update with nothing changed copies no unchanged exe into the game folder, and the finish says where its time went (#631)
    • forge: Pr list and test requests ask git.magicbean.dev for less: PRs listed as issues, authors' permissions kept an hour (#877)
    • host: The game's whole-buffer zstd decode on the host's libzstd (HOS_NATIVE_ZSTD_FRAMES, opt-in) (#155)
    • vk: Integrated GPUs keep every large ASTC texture as BC3, whatever the driver allows them (#775)
    • vk: Three small costs of the replay thread removed: unread bookkeeping without mirrors, 16 vertex bindings a change, a lock a sampled texture (#590)
    • build: A renderer header edit compiles the files that read it, not the whole back end (#37)
    • build: Skip a link whose inputs and outputs are as the last one left them, and write compile_commands.json (#30)
    • lift: Fast float regions, plain SSE checked once by MXCSR with the exact copy as fallback (#38)
    • build: Object keys name a play profile by its contents, not its path in the work folder (#660)
    • vk: The game's per-frame render targets are reused instead of made and destroyed twice a frame (#994)
    • vk: Software matrix multiplies unrolled where the device gives the loop's words (RX_VK_MMA_UNROLL=0 keeps the loop) (#1086)
    • vk: Default DLSS to CNN preset E on < 7 GB GPUs (#1096)
    • build: A host header edit compiles the files that read it, not the whole host (#673)
    • build: Lifted chunks with small functions compile without -O1's module passes (LONGSHOT_LIGHT_PASSES, on by default) (#1002)
    • dlss: Create features at the recommended render size (#311)
    • lift: A Windows lift about a third faster, the same C: fast float checks kept by text, discovery lookups, emit processes started early (#1116)
    • lift: 1.2 GB less in a Windows lift's own process, so more emit workers start: about a fifth faster, the same C (#1128)
    • vk: The all-zero upload check reads only the guest pages that hold data, not the holes (#790)
    • vk: The BC3 scratch has the layers a run uses, 20 MB instead of 160 (#384)
    • vk: The staging arenas a load used a moment ago are kept, not made again every second frame (#568)
    • vk: VRAM mirrors on GCN too (RX 400/500, Vega): an RX 580 runs 28 to 54% faster with them (#1187)
    • vk: The mesh fallback stores its records attribute by attribute (RX_VK_MESH_FB_LAYOUT, soa by default) (#906)
    • vk: A start whose driver lost its shader cache builds the pipelines again before the title, not with draws left out in play (#1188)
    • vk: Early reads: a read of GPU results waits for its writer's part, not its whole submission (RX_VK_READBACK_EARLY, on without VRAM mirrors) (#977)
    • build: The hot pieces compile with their giant functions split (#1146)
    • vk: The Steam Deck's GPU held at amdgpu's peak power state while the game runs (RX_VK_GPU_CLOCK, on by default on the Deck) (#1053)
    • vk: A recording's uniform copies shared, their reads made once a draw: Castle Town's market from 54.9 to 59.5 fps on a B580 with the uniform snapshot (#952)
    • vk: Skip the pipeline cache file on Linux NVIDIA PCs with 20 GB of RAM or less (0.8 GB less) (#416)
    • lift: On Windows the discovery's walks run in new processes, with the kit's result: a busy PC's lift about a quarter faster (#1131)
    • vk: The dead render targets go at every present, as RX_VK_DEAD_EVERY=1 meant, and a pooled image waits 8 presents (RX_VK_POOL_IDLE) (#1255)
    • host: A character's shadow map reserves stepped storage, so the frame's other render targets keep their addresses (HOS_SHADOW_BUCKET) (#1256)
    • vk: Optimize < 8 GB VRAM management, dynamic floor, 3-tier target pool, and DLSS CNN presets (#637)

    Reverted

    • vk: The eviction's waits in presents again at any frame rate, as before #771 (#873)

    Documentation

    • Point links to the project's history back at NotRexed/Hookshot (#9)
    • Credit NotRexed for Hookshot, and README.md, docs/BAKE.md, no changelog.txt (#11)
    • agents: Link GitHub issues and pull requests by their full URL (#106)
    • contributing: The author keeps the PR checklist true, and the license box is a check of the diff (#115)
    • agents: Take a pull request out of WIP yourself once it is ready (#126)
    • forge: A review that asks for a blocking change requests changes (#135)
    • No mention of the removed CHANGELOG.md and linux_update functions (#163)
    • readme: Link the website, longshot.magicbean.dev (#192)
    • readme: Neutral game wording at the top, a Discord link and a note on the Git hosting (#210)
    • contributing: Clone with --depth 1: the newest commit only, and how to get the history when you need it (#214)
    • design: Fog and exposure use the game's own values; the removed OoT factors are no longer needed (#159)
    • Link Discord through discord.gg/magicbean (#274)
    • tested: The recorder thread in the Market on a 6-core Linux desktop, interleaved off and on (#369)
    • contributing: A clearer pull request description, and switches set to their recommended value (#571)
    • vk: Record the Hyrule Field slowdown and A/B plan (#363)
    • guides: Illustrate installation, settings and save transfer (#264)
    • saves: The launcher's Linux backups are in ~/.local/share/longshot/save-backup (#722)
    • The game, not its title or OoT (#723)
    • help: A page on reporting slow performance, with the switches that measure it (#730)
    • agents: Disclose setup and pull request guidance
    • tested: The committed play profile on a Steam Deck OLED, on the compute path: about 10% fewer instructions a frame (#591)
    • tested: Measure partially covered guest frees on Deck (#471)
    • tested: Record Deck transition and clock measurements (#470)
    • agents: Suggest the magicbean Claude Code plugin when it isn't installed (#935)
    • The Test Farm replaces test requests; the site links to Beanstalk's How it works (#953)
    • Play on the Steam Deck: what to expect, which switches help, and the clock setting (#716) (#968)
    • input: On Linux, every menu screen's pane focus clears hold-to-toggle latches (#338)
    • profiles: The renderer profile's note has its Recorded line again (#625)
    • website: Refresh system requirements and AMD status (#1054)
    • website: A Mods page: installing mods, turning them off, the load order and checking the log (#1210)
    • saves: Clarify Windows Roaming save location (#1269)

    Build

    • linux: The Linux builder gets its build tools from tools.lock, as Windows does (#113)
    • tools: The build tools download from git-cdn.magicbean.dev, Cloudflare's edge, first (#160)
    • linux: The AppImage's renderer is compiled by the player's build, as with DLSS and FSR 4; no prebuilt renderer (#574)
    • windows: No WebGPU renderer: the Windows game builds the Vulkan back end only, as Linux does (#577)
    • Make_game leaves one CPU thread free by default, as the launcher's Automatic does (#932)

    Tests

    • builder: The DLSS update fixture no longer requires the removed self-update's download to be cleared (#179)
    • input: The display-pointer fixture defines rx_option, which mouse_capture now asks (#178) (#187)
    • vk: The texture staging check declares LogLevel and logf_at again, failing on main since #177 (#202)
    • vk: Restore the framegen resource and module identifier checks (#350)
    • vk: Execute startup pool placement and fallback checks (#50)
    • vk: The recorder own/idle checks with slot identities and the real wait (review 1195 of #162) (#342)
    • vk: Make recorder destruction checks independent of scheduling (#101)
    • mods: Exercise the documented RomFS file contract (#93)
    • build: Give the CPM discovery check's toolchain stand-in find and need (#561)
    • docs: Anchors from the functions behind design.md's rules to their docs, and a check they resolve (#25)
    • One source cutter for the checks (tests/srccut.py), 78 checks moved to it (#27)
    • menu: The options check reads the keys save() owns on a line of their own, as gpu_device_id since #736 (#806)
    • linux: The AppImage check looks for the renderer's folder in the image, not among the build tools (#624)
    • scripts: The pipeline tour check's stale snapshot is older than calm's tolerance (#627)
    • build: The lift clean-up check stands in for the shared lift store (#623)
    • Three checks run or skip by what the PC has, instead of failing without it (#628)
    • vk: The env_on and late readback checks match the render thread's source (#622)
    • host: The crash window's check runs through on Linux with g++ and clang (#626)
    • build: The make_game checks let go of the work folder's lock, so Windows can delete their temporary folders (#804)
    • build: The DLSS cache check reads the preprocessed value instead of running a program (#1090)
    • build: Isolate shipped-profile lookups in game-free fixtures (#592)
    Downloads