A fixed 1024 map costs GPU time and texture heap on every install, also where the game would draw a smaller one; the
fixed sizes stay a choice in the menu and totk.ini.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reverts dafc35f: free_cores = 1 in totk.ini lifts the virtual core gate again, as in 0.2.8. The mode stays as an
experiment, and every place a player meets it says the game crashes: totk.ini.example, a WARNING line in the start log,
and a line under every crash report and FATAL while the gate is off (kern_free_cores_note), so a log from #212 names
the cause at once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
FSR only enlarges (vk_fsr_record returns 0 when the picture is not larger than the crop), so a 3840x2160 image in a 1920x1080 window is shrunk by the bilinear blit and the line read 'bilinear' next to a menu showing FSR (#219).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Issues #29 and #92 (0.1.1, 240 Hz monitors, fps_limit = display) crash in main+0x1292560 reading a Havok body's
shape. Forced steps between 1/252 and 1/230 s reproduce it in that call chain on the Depths load 3 of 3 times; steps
of 1/240 s and longer, or the same jitter through the floor, play. The comment now records it next to the #241 abort.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A GPU hang that takes the whole system down ends the log without an error (#255, #266, #270 end on a focus pair).
The time on the focus lines lets a report be lined up against the driver's entry in the Windows event log, to tell
whether an alt-tab or an overlay came before the hang or followed it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nvnTextureFlushTexels was a no-op stub. Call nvn_guest_written on CPU-visible texture storage so mirrored textures whose texels were flushed by the guest are stamped in the write-generation table for re-upload, matching nvnBufferFlushMappedRange.
ExitThread went back to the thread start with longjmp on a buffer from a non-unwinding setjmp, whose frame field is 0.
When Control Flow Guard is on for the process, ucrtbase's longjmp checks the buffer first and ends the process with
fast fail 13 (0xc0000409) on a frame of 0: offset 0xb373c of ucrtbase 10.0.26100.9711, the offset in the event logs of
#222, #176 and #75. Exploit protection can force CFG on any program, and the first guest thread to exit, right after the
game's first six-axis read, killed the process with no line in the log; the abort and invalid-parameter handlers never
run for a fast fail.
plat_jmp_save and plat_jmp_go keep and restore the Win64 callee-saved registers, MXCSR and the x87 control word
themselves and validate nothing; Linux keeps setjmp and longjmp. Started with CFG forced on, the build before this
change ends with 0xc0000409 at the same log line and in --probe-kernel; this one reaches gameplay and passes the probe.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
1.4.3 initializes and finalizes command buffers about 2000 times a second while a scene loads and several times a
frame in play. Finalize freed the recording buffer, so each new recording grew again from 64 KiB through up to five
reallocations; the two large per-frame recordings ended in fresh 1-2 MiB blocks, new commit twice a frame. On a
16 GB machine at render scale 200 one of those 1 MiB growths failed during a Depths load and the game ended with
"nvn: command recording buffer of 1048576 bytes" (#94).
Finalize now hands the buffer to a pool of 16 (largest kept, up to 4 MiB each) and the next recording takes the
largest one: a 60 s island run made 22 growths instead of about 110000. A failed recording buffer or command copy
allocation frees the pool and retries every 10 ms for up to 30 s before the fatal, so a short commit shortage (Windows
growing a system-managed page file) stalls the game instead of closing it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Earlier settings menus wrote free_cores = 1 under the label "Virtual core gate", which promised the gate. Without it a
thread that holds no virtual core reports its ideal core to GetCurrentProcessorNumber, so the game's job workers all
report core 0 and run its per-core code at once. On 1.2.1 two of them add the same per-core NVN control memory, which
retires the command handle the other just recorded, and its CallCommands fails at start (#212); on 1.4.3 the game
corrupts a heap in main+0x82e00 or hangs at the title menu. A player with such a totk.ini crashed at every start and
the menu no longer shows the row.
free_cores is no longer a setting: only TOTK_FREE_CORES in the environment lifts the gate, and a free_cores value in
totk.ini is logged as ignored. The kernel probe asks the kernel whether the gate is on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With a step under 1/240 s (ratio below 0.125) 1.4.3 aborts with Havok "Out of memory" from the physics pre-solve
buffer resize (main+0x1261630) within seconds of loading some saves, so a 280 Hz monitor with fps_limit = display
crashed on load (issue #241). frame_step_present floors the averaged step at 1/240 s, and the limit, including the
monitor's rate, stops at 240; a higher fps_limit now logs and gives 240 instead of 60. Limits of 240 and below step
exactly as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Each frame the engine sizes its shadow cascades to the area they cover (main+0xcebdf4 in 1.2.1): from 960 down to about
400 near the ground, so leaf shadows turn coarse, and every change of size moves the shadow texels, so they jump instead
of sliding. The Switch does the same; the community mods force a fixed size. The shadow patch set loads the size from
patch page +92 just before the engine rounds it (main+0xcec22c in 1.2.1, 0x683604 in 1.4.3), and the menu's Shadow map size
(shadow_res in totk.ini) chooses it: game, 512, 768, 1024 (default), 1536 or 2048, with the texture heap grown above 960. A write
watch on the size field found the one writer; with the setting, every shadow map in both versions is the fixed size.
recomp/patches.py lists the patch page words in use.
The menu and the software keyboard hold the game's input until everything they saw is released again, and that release
test counted buttons only. With the walk key a keyboard stick moves part way without setting a direction bit, so the
movement went through the moment the menu or the text entry closed; the four axes count as held now.
The walk key's conflict note reads the bindings from a snapshot rather than the live table that the polling thread
rebuilds, and it looks at every key of an action instead of the two its cells show, so a key bound third still names the
action it presses.
Co-authored-by: deepseek-v4.1-flash-1017 <noreply@deepseek.com>
A keyboard key is all or nothing, so the game only ever saw a full tilt from it and the keyboard could not walk. While
the key named by walk_key is held, the two keyboard stick samples are scaled to walk_scale, the same thing yuzu and eden
do with their modifier setting; the gamepad's axes and the mouse camera keep their own values.
0.51 rather than half: a sample of 32767 times 0.5 is 16383, one short of the 16384 the game's stick-direction bits
need, so half a tilt would leave the direction clear while the axis moved. No key is on by default, because Shift is B
in the default bindings. The Controls tab lists the keys players pick, steps the scale in hundredths and names the
action a key also presses.
Co-authored-by: deepseek-v4.1-flash-1017 <noreply@deepseek.com>
With TOTK_COLOR_* away from their defaults, present_image runs one compute pass (vk_color.c) that writes the game
picture's crop, adjusted, into an RGBA16F image, which FSR 1 or the blit then scales. At the defaults nothing is
recorded and the present takes the same source, format and rectangle as before. The Screen tab gets a Colour group
(profile and four sliders) after the field-of-view slider of #248; cfg.c and totk.ini.example keep both settings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
TOTK_FOV (70..140 percent, Screen tab, live) multiplies the camera's vertical view angle through the new patch set fov
at the two camera routines the aspect set patches; page +60 holds the factor. At the default 100 the set is off and the
camera runs the game's own instructions. Both fov sites and the rest of the 1.4.3 table pass patches.check against the
1.4.3 main. 1.2.1 and 1.1.1 have no fov sites and log that the setting does nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
1.1.1 (update 131072) gets its boot constants, expected hashes, the 60 fps
RomFS overlay, its own code patch table, the scene module slot, loader probe
values and shader census lookup, and the setup accepts it. 1.2.1 and 1.4.3
keep every value: the runtime branches on RT_V111, and an install without a
recorded version still falls back to 1.2.1.
1.1.1 has no aspect patch set; outputs other than 16:9 are scaled to the
window on that version.
Fixes#173
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
At start the game rewrites its save files with the same bytes and commits
twice; each of those commits made a new committed save and a backup, so one
start pushed every older backup out of save-backup. A Commit whose copy is
the newest save file for file (all hard links, same names) now only retries
a waiting swap.
Renaming the committed save away is retried for 200 ms instead of 2 s: a
failure no longer loses the commit, the next Commit or start retries, and the
game's next change to its save data waits for the commit thread.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three places still assumed the two older versions:
- loader.c compared a 1.1.1 load against 1.2.1's expectations, so --probe-loader failed every time:
main has 10,668 constructors against 1.2.1's 10,673, and its relocation kinds and totals differ.
1.1.1 has its own table; subsdk0 and sdk match 1.2.1 down to every count.
- gfx_shader_probe.c looked for build/v121 and build/v143 only, so a census built for 1.1.1 was never
found and --probe-shader failed on a checkout that has only that version. It looks for the running
version first, then the other two.
- layout.py put 1.1.1 first in SUPPORTED while steps.py took SUPPORTED[0] as the version of an install
made before versions were recorded, and such an install is 1.2.1: updating one picked 1.1.1's hashes
and invalidated its extraction. The tuple keeps 1.2.1 first and the fallback names 1.2.1 outright.
--probe-loader now reports 0 failures on a 1.1.1 install.
Co-authored-by: deepseek-v4.1-flash-1017 <noreply@deepseek.com>
frame_rate.c finds the player through a global pointer in main's data segment: 1.4.3 keeps it at
main+0x3add238, 1.2.1 at main+0x4728538, and 1.1.1 was borrowing 1.2.1's address, which points at
something else in that build.
The 1.1.1 slot is main+0x4630a68. It was measured from the build itself: one function there walks
the whole chain this code walks (module+0x1e8 scene, the 14-component check at +0x50, component 13
at +0x58, the 0x70 link stride, the name compare) and 181 scene accesses read the same slot, which
makes it the scene module and not one of the single-instance pointers next to it.
Co-authored-by: deepseek-v4.1-flash-1017 <noreply@deepseek.com>
The build knows the player's update as v111: DISPLAY_VERSIONS maps it for the version line of the settings menu, and
boot.c carries its two checkpoint return addresses (0xeb8c14 for SetFocusHandlingMode, 0xebcdec for the first
nvnBootstrapLoader call) beside the 1.2.1 and 1.4.3 ones, with init_array counts of 10668, 45 and 116 for main, subsdk0
and sdk. The build guide lists v111 among the versions every step takes.
Co-authored-by: deepseek-v4.1-flash-1017 <noreply@deepseek.com>
Shader preparation reads MacNativeGames' whole shader list history,
release downloads resume from a .part file, appimagetool and the type-2
runtime come from release tags, and the preparation summary counts the
vertex and fragment variants it skips with the graphics pipeline library
off.
The gettid cache in runtime/platform/linux.c conflicted with the same fix
already on dev (8dd7e20, #236); dev's version is kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Windows refuses to rename a directory while any file under it is open, so a
virus scanner, Explorer's preview or a sync tool holding a file of
save/<uid> made every Commit of a session fail after its 2 s of retries.
The game had its Commit return 0 and showed the new slots from the working
copy, and the next start deleted that copy: hours of play were gone (#260,
#263, log line "cannot move the committed save (error 5)").
A Commit now renames its finished staging copy to save-tx/ready-<n> before it
touches the committed save. That rename is the commit point. When the swap
into save/<uid> fails, ready-<n> stays the newest save: later working copies
and the next start are made from it, and the next Commit or start retries the
swap. The failure log names the programs holding files in the committed save
(Restart Manager) or says the folder refuses the rename (no DELETE access).
--probe-fs holds a committed file open across two Commits and restarts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A player's game ended with "nvn: command recording buffer of 1048576 bytes" during a dive into the Depths (#94):
realloc refused 1 MiB with commit to spare. That leaves a buffer pointer that is no block of ours, from a command
buffer object the game overwrote or initialized again without Finalize. The fatal line now gives the object, its
magic, the old buffer and its size, the bytes used and asked, and Initialize logs (eight times at most) when it
replaces an object that was not finalized.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
abort() (a failed assert, std::terminate) and an invalid parameter to a C runtime function raise no exception: the
C runtime ends the process with 0xc0000409 and the log stops without a FATAL line. A SIGABRT handler and an
invalid-parameter handler now send both to the crash hook with the host return addresses in totk.exe, and the
report lists them as totk+offset frames after the guest registers. Checked with win32.c in a test program: assert,
abort and fclose(NULL) each give one report whose innermost frame is the failing call.
lifted_function no longer names the nearest lifted function for an address above the lifted code, so a host frame
of the runtime is not reported as game code.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The menu merges the Japanese, Korean and Chinese language names from a system font. Dear ImGui asserts on a font
stb_truetype cannot parse, and the assert ends the process with 0xc0000409 (exit code 127 in a shell) and no line in
the log, a few seconds after the window opens: the report of an AMD laptop on Windows 11 Home Single Language (#176)
matches that, and a test build of our ImGui with a damaged font exits the same way. A cut-off font passes the parse
and faults later, when the glyphs are made, because stb_truetype reads tables without bounds checks.
fonts_up now runs each system font through a private copy of stb_truetype whose asserts only mark the font bad, after
checking that every table of its directory lies inside the file, and leaves out a font that fails with a log line.
--probe-window checks that the test takes the embedded Roboto and refuses a cut-off copy, one without a character
map and one with an unknown tag.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The row was labelled "Virtual core gate" with the text "Lets only one thread run per virtual core, as on the
console", but switching it on wrote free_cores = 1, which lifts the gate. Without the gate guest threads run in
parallel on one virtual core, and TotK keeps its per-core data without locks: an Arc B580 player who turned it on
crashed at every start with "CallCommands: command handle ... is not a live recording" (#212). free_cores stays an
ini key for experiments; the log now says when it is set, and totk.ini.example warns that the game can crash.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The page write tracking dropped the render of any texture smaller than a page whenever the CPU wrote another
part of that page. The Purah Pad camera renders the ID, screen rect and depth of what is under the reticle into
4x4 targets in two slots and reads them a frame later. On 1.4.3 one slot's rect target shares its page with data
the CPU writes every frame, so the game read a zero rect every other frame and the target's name label blinked; on
1.2.1 the targets of both slots were dropped and nothing was recognised. A texture that fills no page now keeps a
copy of its storage bytes from the GPU write, and a page write counts only when those bytes changed.
TOTK_WB_EXACT=0 restores the old decision.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The AMD device losses with the recorder (#5, #40) stopped after objects
were kept until their submission completed and the recorder began ending
its own parts; RDNA 1 and 2 cards in #5 ran with vk_recorder on and off.
vk_recorder = 0 still turns it off.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
plat_thread_id made a gettid syscall on every call, and plat_rmutex_lock
calls it on every acquire: about 685,000 syscalls a second in the open
world on 1.4.3. A thread's id never changes and the runtime starts child
processes with posix_spawn only, so the id now comes from a __thread cache
after the first call. Windows is unchanged: GetCurrentThreadId reads the TEB.
Found, measured and fixed by H.A. Forster in issue #236 (RTX 2060 laptop:
36.3 to 46.0 fps with replay_async). SchulzCode confirmed it on an RTX 2070
Super with a 7800X3D: 98 to 120 fps at game resolution, 76 to 90 fps at
render_scale 150.
Co-Authored-By: H.A. Forster <154697759+h-a-forster@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The camera update and setup routines that the aspect set patches load the camera's vertical view angle a few words
before the ratio and pass it to sead's perspective projection. The new set fov loads the angle as before and multiplies
it by a factor the runtime writes to patch page +60, so TOTK_FOV (70..140 percent, menu Screen tab, live) widens or
narrows the view. The game's own zooms (bow, scope) scale with it; 100 switches the set off.
Checked on 1.4.3 at 1920x1080 at 70, 100, 130 and 140: the view narrows and widens around Link, the HUD is unchanged,
no objects appear or vanish at the edges, no validation errors. 1.2.1 has no fov sites yet (its original words are
unverified here); the setting logs that and does nothing there.
The perf line's "bound" said cpu at 640x360 on a Ryzen 5 4600G (Vega 7, RADV) and also at 1600x900, where lowering the
resolution showed the GPU held the frame back; the GPU busy time alone cannot tell the two apart. With
VK_KHR/EXT_calibrated_timestamps the GPU timestamps of each part go on the CPU clock (QPC on Windows, CLOCK_MONOTONIC
elsewhere), each part records when it was handed to vkQueueSubmit2, and a third perf line splits the gap before each
timed part: submitted after the previous part ended (the GPU waited for the replay or the game) or already queued
(untimed uploads and presents, or a wait). It also gives the gaps over 0.5 ms, the longest one and the mean time from
submit to start.
On the 4600G at 640x360 (44 fps) it reads 11.3 ms idle a frame, 10.3 ms of it waiting for the next submission, in 3-4
gaps of up to 8 ms, with the replay at 16 ms a frame: the GPU starves behind the replay. Without the extension, or when
the driver refuses the domain pair, the line is left out. Only TOTK_TRACE=perf and the full overlay time the GPU.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Issue #136. With TOTK_COLOR_* away from their defaults, present_image runs one compute pass (vk_color.c, shaders/color_grade.comp) that writes the game picture's crop, adjusted, into an RGBA16F image, and FSR 1 or the blit scale that one; at the defaults nothing runs. The Screen tab gets a Colour group: a profile (Game's own, Vivid, Clarity, Custom) that sets the four values, and the four sliders. The adjustment, its order, ranges and presets are STS78721's patch from the issue, reworked onto dev: it now applies with FSR 1 scaling as well as the blit, its menu texts come from the catalog in English and Russian, and the patch's unused accessors and build_game.py change are left out.
Co-authored-by: STS78721 <134802927+STS78721@users.noreply.github.com>
patches-v121.txt takes the aspect set at upstream's 1.2.1 sites (game_patches.c aspect_install): the same two camera
loads and two viewport divisions as 1.4.3, with the same original words and the same replacement from page +56. The
installer now builds the romfs-patches-<w>x<h> overlays for 1.2.1 too; ultrawide_layout.py passes on its
Common.Product.110 at all ten sizes. Checked in game at 3440x1440 (camera 2.3889) and 1920x1200 (1.6000): wider or
taller view, HUD at the corners or edges, no validation errors.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
astc_probe.c came from dev with _putenv_s, which the Linux build of the merged runtime rejects. check_platform.py now
flags _putenv and _putenv_s in shared files.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>