Lucas-blip-png/dkb-pc-recomp v0.4.0

DKB Recomp v0.4.0

on GitHub

DKB Recomp 0.4.0: a static recompilation of Donkey Kong Bananza (Switch 2) to a native Windows exe with a Vulkan renderer. It is still experimental (0.x). On recent NVIDIA cards players get well past the first level (Glacier, Snow World, Tempest), and it can crash.

The packages hold no game data, and no release ever will. You build the game on your own PC from your own dump of 1.0.0 or the 3.2.0 update, given as a .nup file or as ExeFS and RomFS folders. Other updates are not supported.

What is new since 0.3.0

• Crashes fixed or guarded. The Snow World descent device loss on NVIDIA is gone: one light quad (vertex stage ba1dd1e7ae8abb33) is skipped (#196, #197, #264). A crash while loading a save or the title screen at main+0x82fee0 is guarded (#157, #206, #173). The 3.2.0 game now has the layer-change guards 1.0.0 had (#218, as #126), a crash at a layer change when an id cache runs out of nodes is guarded on both versions (#218), and so is a record loop that reads through an unset table index (#229). A Linux crash report now gets as far as the game's function and frames (#148). On AMD's Windows driver the Vulkan recorder thread is off by default, which stopped the device loss for several players (#84, #124, #142, #155, #167, #176).
• Picture. The #9 light quads are drawn again, so their particles are back (#62), and the Tower light quads are drawn with a fold of their selector instead of faulting (#220). scene_res = full draws the 3D scene at the full output size instead of the 2/3 the game settles at (#210), on 1.0.0 and 3.2.0 (#223). Draws get depth twins like compute passes, so the validation layer is clean in the first level.
• Performance. Big render passes are recorded on 4 threads by default from 8 logical processors (not on AMD), dynamic uniform buffers are on by default (not on Intel), and the guest heap is lazy by default on Windows (commit about 9.3 GiB instead of 14.5 GiB). In the heavy -101 scene on an RTX 5060 Ti: about 57.5 fps with a 1% low of about 29 fps (0.3.0: 51.8 and 20.6). The end of a sliced pass no longer waits for its slices (#265), and small VRAM mirror write-backs are copied by one compute dispatch on NVIDIA (#269), and the destructible terrain's geometry-stage storage stays in VRAM instead of being copied back every frame (#268; Hilltop Layer on an RTX 5090: GPU time 16.1 to 12.4 ms a frame), with its bookkeeping kept off the mirror lock (#288). Command recordings reuse cached blocks and hand their buffer over instead of being copied, which removes almost all page faults in the Hilltop Layer (about 470,000 a second before, #289). Many lock and copy cuts in the sliced replay come from contributors (see Thanks).
• DLC and mods. Setup installs add-on content from a .nup into aoc-<n> folders (#183, #111, #258, #263). Loose RomFS mods load from a mods folder beside dkb.exe (#252).
• Sound. A Windows or macOS device that opens with a period that is not a multiple of the renderer's block is reopened with one block per period, which stopped the crackling on a 1024-frame device (#272).
• High resolutions. The game's GSys heap grows with the render-target pool, so scene_res = full at 3840x2160 and very large outputs no longer crash at start (#244, #279). A guide for 4 GB VRAM cards and laptops: docs/4gb-vram-optimization.md (#260).
• Settings and input. PlayStation button prompts when a DualShock or DualSense is connected at start (pad_prompts, #274). The vertical mouse camera no longer locks during automatic camera moves (#245). Big passes use 6 replay threads from 12 logical processors (#271). NVIDIA DLSS is listed in the upscaler row (#241), and its warm-up waits when another upscaler is chosen at start (#280). DLSS uses preset K except in Ultra Performance, which is cheaper than M on the GPU (dlss_preset, #287). console = 0 hides the console window for frontends such as Playnite (#238). Controller connection handling, DK moving on walls and ceilings above 60 fps (#201), the 3.2.0 version on the title screen.
• Installer. It refuses a RomFS folder that holds a RomFS image (#278), compiles only the sources a release ships (#203, #208), recompiles every runtime source when any of them changes (#275), and builds the ultrawide overlays. RecompLauncher can set up and start the Linux and macOS packages, and the release now has a self-contained Linux AppImage, DKB-Setup-0.4.0-x86_64.AppImage, beside the .tar.gz (#242, #174). .nx (NXFormat) dumps are accepted next to .nup (#224). CPUs without AVX2 can build the game: setup compiles it for x86-64-v2, which is slower (#231). The shader library prebuild of --warm-shaders resumes where it stopped after a restart (#3).
• Diagnostics. A crash outside the executable (a driver, a Vulkan layer) is reported with its module, and every crash writes a minidump to logs/ after the report (#219); keep it on your PC and send it only when asked, privately.
• Project. Issue forms ask for the build, game version and OS; a security policy and an accessibility page were added.

Pull requests in this release

Thank you to everyone who sent code. Each one below is on bananza and in this release:

• #209 Slice passes under DLSS, and close the follow-ups of #195 (@GlebShekovtsov)
• #211 Take the texture and mirror locks out of the sliced draws, and reuse a draw's texture descriptors (@GlebShekovtsov)
• #213 Let RecompLauncher set up and start the Linux and macOS packages (@Felitendo)
• #217 Fail a store exclusive after another thread's change and change back (@thatbananaman08)
• #219 Report crashes outside the executable and write a minidump for every crash (@thatbananaman08)
• #220 Draw the faulting light quads with a fold of their selector instead of skipping them, and fix Tower's device loss (@thatbananaman08)
• #222 Run every test from one command, and fix three tests that no longer built or ran (@thatbananaman08)
• #223 Make scene_res draw the 3D scene at its ratio on 3.2.0 too (@wrothgaril)
• #225 Make the DLSS feature ahead of play and once for every input size (@wrothgaril)
• #226 Batch small GPU writebacks and reuse copy buffers (@oshoubber)
• #227 Keep the texture cache generation stable during replay (@oshoubber)
• #228 Batch terrain reset copy commands (@oshoubber)
• #234 Take the texture lock off a guest read's GPU wait, and sort only the new GPU write ranges of the uniform snapshot (@GlebShekovtsov)
• #235 docs: keep the 3.2.0 build status sentence in one tense (@lunar-me)
• #236 docs: fix the verb for the shader census step (@lunar-me)
• #237 docs: add the missing preposition in the cursor-release list (@lunar-me)
• #242 Add selfcontained AppImage build (@Osohow)
• #244 Grow Bananza's GSys heap with the render-target pool, so scene_res and large outputs no longer crash at start (@PorlocK-BR)
• #245 Fix intermittent vertical mouse camera lock in Bananza (@oshoubber)
• #246 Keep DK moving on walls and ceilings above 60 fps (@Zorkats)
• #250 Show the 3.2.0 version on the title screen (@Zorkats)
• #254 Improve controller connectivity, selection and hot-swapping (@shadowlink1174)
• #255 Retire a released pool's descriptors from an index instead of the whole table (@wrothgaril)
• #256 vk_exec_draw: add depth variant origin for mine voxel native GS on RDNA 1/2 (@MatyMeatBoy)
• #257 Take the texture and pipeline locks out of most sliced draws (@wrothgaril)
• #259 Wait for a texture write-back's GPU copy without the texture lock (@wrothgaril)
• #260 perf(docs): add 4GB VRAM optimization guide and config profile (@devadobe11-pixel)
• #261 Prepare Bananza ultrawide overlays during installation (@Marjaperse)
• #262 Find a resolved pool through a per-thread sorted list and a block map (@wrothgaril)
• #265 Let the end of a sliced pass go on without waiting for its slices, by default (@wrothgaril)
• #268 Keep the storage the geometry stage writes in VRAM (@wrothgaril)
• #269 Copy small VRAM mirror write-backs with the compute dispatch by default (@wrothgaril)
• #271 Record sliced passes on 6 threads by default from 12 logical processors (@wrothgaril)
• #273 Wait for an upload batch's fence without cmd_lock (@wrothgaril)
• #274 Show PlayStation button prompts in Bananza when a DualShock or DualSense pad is connected at start (@GlebShekovtsov)
• #275 Compile every runtime source of an install build again whenever its contents change (@GlebShekovtsov)
• #277 Document Steam Deck menu access when the fullscreen hold is enabled (@shadowlink1174)
• #280 Let the DLSS warm-up wait while another upscaler is chosen (@wrothgaril)
• #287 Ask DLSS for preset K outside Ultra Performance, with dlss_preset to choose (@wrothgaril)
• #288 Keep the kept-write bookkeeping off mirror_lock's hot path (@wrothgaril)
• #289 Keep large recording copies in cached blocks and hand the recording buffer over (@wrothgaril)

Issues resolved

Thank you to everyone who reported a problem, with logs, saves and recordings:

• Device loss on NVIDIA in the Snow World descent and the Glacier rock ride: #197 (@hotcoffeek), #264 (@dudesh34), #196 (@philneitz)
• Device loss or black screen on AMD's Windows driver, with the recorder thread now off by default: #84 (@philneitz), #123 (@BloodThirstyQ), #124 (@TeknaPT), #142 (@beijingpunk), #155 (@bubblesnout), #167 (@jesusen904), #176 (@BoySKAyt), #193 (@stuzukila82-bit), #215 (@thabans), #267 (@350Plays)
• DLC and the update from a .nup: #111 (@infinitytom), #183 (@Blinkk000), #258 (@ElBusAlv), #263 (@evylmen)
• Loose mods: #252 (@Caladbolg95)
• Render scale and 3D resolution: #149 (@formandfunctionlab), #210 (@dancaiman), #279 (@Dimiduh)
• DLSS: #152 (@Gavury), #241 (@thecharted135-sketch)
• Performance in heavy scenes: #154 (@patrickfuly), #156 (@supermahdi-enjoyer), #194 (@li-beihai), #207 (@solidaytug), #230 (@zachcristallguy-netizen), #284 (@wrothgaril)
• Above 60 fps, climbing walls: #201 (@XDM-Inc)
• Raw mouse camera: #177 (@pritomroyreal-cloud)
• Controllers: #253 (@shadowlink1174)
• Installer: #203 (@bika66), #208 (@Sapo2099), #278 (@sonic-un), #138 (@asluan287-gif)

Thanks

Pull requests: wrothgaril (#223, #225, #255, #257, #259, #262, #265, #269, #268, #271, #273, #280, #287, #288, #289), GlebShekovtsov (#209, #211, #234, #274, #275), oshoubber (#226, #227, #228, #245), PorlocK-BR (#244), devadobe11-pixel (#260), Osohow (#242), thatbananaman08 (#217, #219, #220, #222), shadowlink1174 (#254, #277), Zorkats (#246, #250), Marjaperse (#261), MatyMeatBoy (#256), Felitendo (#213), lunar-me (#235, #236, #237). Thank you.

Testing and reports: everyone who sent logs, saves and input recordings, in particular luis-7654, philneitz, Felitendo and Dimiduh.

Download

Requirements (Windows)

• NVIDIA GPU with a recent driver (tested on an RTX 5060 Ti and an RTX 3070 Ti Laptop). AMD and Intel are experimental: on AMD's Windows driver the device loss most players hit is fixed (recorder thread off), and RADV on Linux can still hang (#99, #122, #198, #204). Intel Arc: #3.
• 16 GB of RAM plus a large page file (the process commits about 9 GB), 32 GB is comfortable.
• A controller is recommended.

Known issues

• GPU hangs under RADV on Linux (#99, #122, #198, #204), and a driver fault on Intel Arc (#3).
• Crashes reported without a reproduction here: #126 (Lagoon to Hilltop tunnel, retest asked), #136 (Forest snow tunnel landing: needs a guard on a call, after this release), #148 (Linux, after the first boss: the next log will name the function).
• Black shards where salt or acid meets slime (#139, #266).
• Text is soft at 1440p and above: the game's fonts are drawn at a fixed size (#214).
• Heavy scenes can still fall under 60 fps, and the first visit to an area can stutter.
• Checkerboard textures (#5), #112 and unrecorded NVN 2 calls (#6).

Reporting a bug

Open an issue with the log (dkb.log), the build line at its top (dkb 0.4.0+<commit>), your GPU and driver, and what you did before it failed. Never share generated files, keys or screenshots that show game content.

Roadmap: #10.

What's Changed

• Slice passes under DLSS, and close the follow-ups of #195 by @GlebShekovtsov in #209
• Batch terrain reset copy commands by @oshoubber in #228
• Let RecompLauncher set up and start the Linux and macOS packages by @Felitendo in #213
• Run every test from one command, and fix three tests that no longer built or ran by @thatbananaman08 in #222
• Fail a store exclusive after another thread's change and change back by @thatbananaman08 in #217
• Batch small GPU writebacks and reuse copy buffers by @oshoubber in #226
• Keep the texture cache generation stable during replay by @oshoubber in #227
• Take the texture and mirror locks out of the sliced draws, and reuse a draw's texture descriptors by @GlebShekovtsov in #211
• docs: keep the 3.2.0 build status sentence in one tense by @lunar-me in #235
• docs: fix the verb for the shader census step by @lunar-me in #236
• docs: add the missing preposition in the cursor-release list by @lunar-me in #237
• Show the 3.2.0 version on the title screen by @Zorkats in #250
• vk_exec_draw: add depth variant origin for mine voxel native GS on RDNA 1/2 by @MatyMeatBoy in #256
• Make scene_res draw the 3D scene at its ratio on 3.2.0 too by @wrothgaril in #223
• Keep DK moving on walls and ceilings above 60 fps by @Zorkats in #246
• Retire a released pool's descriptors from an index instead of the whole table by @wrothgaril in #255
• Take the texture and pipeline locks out of most sliced draws by @wrothgaril in #257
• Find a resolved pool through a per-thread sorted list and a block map by @wrothgaril in #262
• Wait for a texture write-back's GPU copy without the texture lock by @wrothgaril in #259
• Prepare Bananza ultrawide overlays during installation by @Marjaperse in #261
• Make the DLSS feature ahead of play and once for every input size by @wrothgaril in #225
• Improve controller connectivity, selection and hot-swapping by @shadowlink1174 in #254
• Document Steam Deck menu access when the fullscreen hold is enabled by @shadowlink1174 in #277
• Wait for an upload batch's fence without cmd_lock by @wrothgaril in #273
• Take the texture lock off a guest read's GPU wait, and sort only the new GPU write ranges of the uniform snapshot by @GlebShekovtsov in #234
• Grow Bananza's GSys heap with the render-target pool, so scene_res and large outputs no longer crash at start by @PorlocK-BR in #244
• Fix intermittent vertical mouse camera lock in Bananza by @oshoubber in #245
• Show PlayStation button prompts in Bananza when a DualShock or DualSense pad is connected at start by @GlebShekovtsov in #274
• Compile every runtime source of an install build again whenever its contents change by @GlebShekovtsov in #275
• perf(docs): add 4GB VRAM optimization guide and config profile by @devadobe11-pixel in #260
• Record sliced passes on 6 threads by default from 12 logical processors by @wrothgaril in #271
• Copy small VRAM mirror write-backs with the compute dispatch by default by @wrothgaril in #269
• Add selfcontained AppImage build by @Osohow in #242
• Let the end of a sliced pass go on without waiting for its slices, by default by @wrothgaril in #265
• Report crashes outside the executable and write a minidump for every crash by @thatbananaman08 in #219
• Draw the faulting light quads with a fold of their selector instead of skipping them, and fix Tower's device loss by @thatbananaman08 in #220
• Let the DLSS warm-up wait while another upscaler is chosen by @wrothgaril in #280
• Ask DLSS for preset K outside Ultra Performance, with dlss_preset to choose by @wrothgaril in #287
• Keep the kept-write bookkeeping off mirror_lock's hot path by @wrothgaril in #288
• Keep large recording copies in cached blocks and hand the recording buffer over by @wrothgaril in #289

New Contributors

• @Felitendo made their first contribution in #213
• @thatbananaman08 made their first contribution in #222
• @lunar-me made their first contribution in #235
• @MatyMeatBoy made their first contribution in #256
• @Marjaperse made their first contribution in #261
• @shadowlink1174 made their first contribution in #254
• @PorlocK-BR made their first contribution in #244
• @devadobe11-pixel made their first contribution in #260
• @Osohow made their first contribution in #242

Full Changelog: v0.3.0...v0.4.0