Commit Graph
210 Commits
Author SHA1 Message Date
Valiantsin DzerakhandClaude Opus 5.5 acd550e822 Merge the VRAM headroom: a background thread allocates the next VMA block of each memory type before the render thread needs it (vk_headroom.c, TOTK_VRAM_HEADROOM)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 16:09:38 +03:00
Valiantsin DzerakhandClaude Opus 5.5 0775796244 Merge the replay speedups: render target views live as long as their image, command buffer parts span at least a millisecond of recording (TOTK_SPLIT_US) and the shader memo skips its handle compare while nothing was bound; the fight replay on the desktop falls from 8.12 to 7.61 ms
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 16:09:38 +03:00
Valiantsin DzerakhandClaude Opus 5.5 d1149e5060 Allocate the next VMA block of each memory type on a background thread: a new block made by another thread wakes vk_headroom.c, which allocates half a block every 500 ms and keeps it as the anchor of a block it had to create, so the render thread finds room instead of calling vkAllocateMemory for 8-19 ms; no headroom above 85% of the VRAM budget (TOTK_VRAM_HEADROOM)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 16:05:24 +03:00
Valiantsin DzerakhandClaude Opus 5.5 bac4d6cf91 The shader memo skips its 1 KiB handle compare while no texture or image was bound
memo_match compared the vertex and fragment stages' 64 texture and 64 image handles on every draw. Every OP_BIND_TEX and now every OP_BIND_IMG bumps tex_serial; a key whose serial is current has the same handles, and a key that matched by the compare takes the current serial. memo_match went from 0.95% to 0.27% of the fight replay.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:54:00 +03:00
Valiantsin DzerakhandClaude Opus 5.5 c238c341c7 Replay command buffer parts span at least a millisecond of recording
A fight frame of about 2,000 draws made 15 parts on the 7900X at 64 draws a part, each split costing the render thread a submit, a new command buffer and a mirror prologue (vk_sync_split, 7.6% of the replay). maybe_split now also waits until TOTK_SPLIT_US (1000, 0 off) has passed since the part began, so a frame makes 7-9 parts there. A millisecond of recording is well under a millisecond of GPU work, so the GPU still starts early in the frame.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:53:59 +03:00
Valiantsin DzerakhandClaude Opus 5.5 c9c06f2501 Render target views of registered textures live as long as their image
vk_exec_pass_begin created every attachment view again in each submission and freed them at its fence: 2% of the fight replay on the 7900X, more on drivers with a slower vkCreateImageView. view_of now takes the view from vk_exec_target_view, which keeps it in the sampled-view tables under the texture lock, and vk_exec_image_gone releases it with the image. Images that are not registered textures keep the per-submission views.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:53:59 +03:00
Valiantsin DzerakhandClaude Opus 5.5 8e0d6b3119 Merge the overlay order: the settings menu draws over the performance overlay while it is open
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:26:00 +03:00
Valiantsin DzerakhandClaude Opus 5.5 403606bbcb The settings menu draws over the performance overlay while it is open
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:24:14 +03:00
Valiantsin DzerakhandClaude Opus 5.5 34b5716686 Merge the menu and overlay work: the left stick moves the menu cursor, the D-pad leaves the bottom buttons and the tabs for the page's rows, and the full performance overlay reads AMD sensors through ADL and shows the window, game image and 3D scene sizes, the scaling, frame limit and presentation mode
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:22:31 +03:00
Valiantsin DzerakhandClaude Opus 5.5 89cc960f09 Menu navigation with the left stick and back from the bottom buttons; AMD sensors and the picture lines in the performance overlay
The left stick now moves the menu cursor as the D-pad does (ImGui only scrolled with it). The page's rows are a
direct NavFlattened child of the menu window: behind a second child level ImGui did not clip them, so up from
Continue went to the tabs and down from the tabs to Continue.

The full overlay reads temperature, board power, clock and GPU load from ADL's PM log on AMD GPUs (matched by PCI
bus), and shows the window, game image and 3D scene sizes, the scaling to the window (FSR and its sharpness), the
frame limit and the presentation mode in use.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:21:41 +03:00
Valiantsin DzerakhandClaude Opus 5.5 f8219ed793 Merge the game resolution work: the output and scene patch sets render 1.2.1 and 1.4.3 above 1080p with a grown texture heap (TOTK_OUTPUT_RES, TOTK_RENDER_SCALE, supersampling above 100 %), and AMD FSR 1 scales the game's output to a larger window (TOTK_UPSCALER, TOTK_SHARPNESS)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 15:02:55 +03:00
Valiantsin DzerakhandClaude Opus 5.5 eca89641fa Keep 1.2.1's scene at the wanted size, package zstd for the installer
1.2.1 renders the scene size it reads, while 1.4.3 scales it by output width / 1920: on 1.2.1 the scene set is on
whenever the output grows (2560x1440 gave a 1600x900 scene before, 2133x1199 now). An output at or below the shipped
size needs no grown HeapDef. The release deps carry the static libzstd and its header, totk.ini.example the four new
keys, THIRD_PARTY_NOTICES.md the MacNativeGames sites, FSR 1 and zstd.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:56:28 +03:00
Valiantsin DzerakhandClaude Opus 5.5 7abdbc90ac Render the game above 1080p and scale it to the window with FSR 1
The output and scene patch sets make the game read its output, swapchain and scene sizes from the patch page
(MacNativeGames' sites for 1.2.1 and 1.4.3); resolution.c picks them from TOTK_OUTPUT_RES and TOTK_RENDER_SCALE and
grows the HeapDef texture heap, its parent and the application pool to match. vk_fsr.c runs AMD FSR 1 EASU and RCAS
on the present when the window is larger than the game's output (TOTK_UPSCALER, TOTK_SHARPNESS).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:40:34 +03:00
Valiantsin DzerakhandClaude Opus 5.5 098f8a52da Merge the hactool build: tools/build_hactool.py and the two patches in tools/hactool/ make the extraction tool reproducible from the repository
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:18:32 +03:00
Valiantsin DzerakhandClaude Opus 5.5 ad70afaee0 Build hactool from the repository: tools/build_hactool.py fetches the pinned commit, applies tools/hactool/*.patch and links a static exe
The MacNativeGames BKTR and title-key changes and our UTF-8 manifest lived only in the maintainer's toolchain folder, so
nobody else could build the hactool that extraction and the installer need. The script's exe matches the one in use
byte for byte apart from its build time.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:18:23 +03:00
Valiantsin DzerakhandClaude Opus 5.5 49bd46440d Merge the repository cleanup: code comments without milestone and task labels, gfx_stubs.c removed, docs reduced to architecture, building, controls, tester guide, installer and the Maxwell reference, and the agent setup and working notes moved out of git into .claude/, notes/ and CLAUDE.local.md
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:15:11 +03:00
Valiantsin DzerakhandClaude Opus 5.5 e3e72dfb44 Keep the repository to what players and contributors use; move the agent setup and working notes out of git
docs/building.md merges the two toolchain docs and the README build steps, and lists the developer checks. The
Maxwell reference moves to docs/maxwell.md. architecture.md loses the milestone table, controls.md describes timed
script lines and input recording, and the tester guide includes shader preparation.

The agent skills, hooks, playbook, backlog and the study notes on upstream projects are the maintainer's local setup:
.claude/, notes/ and CLAUDE.local.md are ignored now, and tools/worktree.py links them into task worktrees. AGENTS.md
keeps the project rules.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:12:50 +03:00
Valiantsin DzerakhandClaude Opus 5.5 97d3762ba1 Strip milestone and task labels from code comments and drop the dead gfx_stubs.c
Comments named the agent task (T1-T7, M2-M5, Task 2.x) that once owned each file or block, and pointed at local
notes the repository does not carry (backlog items, milestone plans, research notes). They now say what the code
does. gfx_stubs.c held weak stand-ins for functions that every build has defined for a long time.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:12:50 +03:00
Valiantsin DzerakhandClaude Opus 5.5 fb51973fa3 Collect what the Windows host costs into docs/research/windows-host.md: timer overshoot, counter conversion, small-file opens and flushes behind other writers, thread sampling pitfalls, the three virtual cores, VMA block allocation, run-to-run spread and bit flips; AGENTS.md points to it
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:01:22 +03:00
Valiantsin DzerakhandClaude Opus 5.5 42b73c4098 Note what each 2026-10 frame time fix needs on a Linux host: the SleepThread spin is for the Windows timer only, the file I/O moved off the game and render threads is portable with fsync in place of FlushFileBuffers, the slow-present profiler is Windows-only
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 13:01:22 +03:00
Valiantsin DzerakhandClaude Opus 5.5 b12bbe7705 Merge the overlay width fix: the full overlay's frame time plot and wait table fill the panel, and the wait numbers end at its right edge
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 12:17:31 +03:00
Valiantsin DzerakhandClaude Opus 5.5 d5d8497825 Stretch the full overlay's frame time plot and wait table to the panel width, with the numbers on the panel's right edge
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 12:17:27 +03:00
Valiantsin DzerakhandClaude Opus 5.5 c5a275126e Merge the Windows frame time work: SleepThread spins under 1 ms, recording copies indexed by 64 KiB block, the pipeline cache and the save Commit written on background threads, shader use masks read at start, a slow-present profiler (TOTK_PROFILE_SLOW) and input recorded on the world clock
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 12:11:32 +03:00
Valiantsin DzerakhandClaude Opus 5.5 277c8484e7 Read every shader use mask file on a background thread at start, so the render thread opens none in play
Each draw's first sight of a program opened its <hash>.use file on the render thread (use_find from gfx_shader_get_async). The open takes 0.1 ms alone but took 11-40 ms in play behind the translator's and the pipeline cache's writes and flushes on the same volume (slow-present dumps of a 1.4.3 fight session). The 45131 masks of a prepared 1.4.3 cache load in about 12 s below normal priority, before the first save load ends; until then lookups read the file as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 11:40:33 +03:00
Valiantsin DzerakhandClaude Opus 5.5 ecc90b5678 Record and replay input on a world clock: game seconds stepped since the player first stands away from the title screen's origin
With the real-time frame step, a frame advances the game by its own length, so a recording keyed by presents replayed at another frame rate pressed at other game times, and every load of another length shifted the rest. Lines after the first save load are now written as "@<ms> <what> <hold ms>" and replayed on that clock; menu input before it stays in presents. Two replays of one recording whose loads ended at presents 1887 and 2094 kept the player within 2 m for 15 game seconds, until the fight's own nondeterminism took over.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 11:40:33 +03:00
Valiantsin DzerakhandClaude Opus 5.5 adc4d7aa9c Dump every thread's call chains of a slow present (TOTK_PROFILE_SLOW=<ms>), and never copy a stack whose Rsp sits in its guard page
The sampler keeps the last half second of every thread's chains in a ring and writes the samples of a present interval over the threshold to <TOTK_PROFILE>.f<frame>, per thread and time (tstack), with the samples that took over 2 ms to stop their thread (tstall: the thread was in a long kernel call). A thread that ran no cycle since its last sample repeats its chain without being suspended. A thread stopped between its sub rsp and the stack probe has Rsp in the guard page; copying from there raised STATUS_GUARD_PAGE_VIOLATION and ended a run.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 11:40:32 +03:00
Valiantsin DzerakhandClaude Opus 5.5 6fca957670 Run the save Commit on a background thread: the copy, flushes and renames held the game thread 95-135 ms at every game save in the 1.4.3 fight save
The game's Commit returns at once. Every later change to the working copy, Unmount and the next Commit wait for the thread, so no write lands in the snapshot, and the window close, guest ExitProcess and --frames exits wait through fs_save_flush. A crash during the commit loses that commit only; the transaction and save_recover are unchanged. probe-fs waits for each commit and checks a write and an Unmount right after Commit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 11:01:54 +03:00
Valiantsin DzerakhandClaude Opus 5.5 650c69bcb5 Save the pipeline cache and append pipeline keys on an autosave thread, without P.lock: the save held P.lock through vkGetPipelineCacheData and the file write (about 60 ms) and ran on the present thread, so every 20 s the replay stopped at its next pipeline lookup or waited for the present thread (60-80 ms freezes on the 1.4.3 fight); shader record files are written outside RC.lock for the same reason
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 10:45:53 +03:00
Valiantsin DzerakhandClaude Opus 5.5 dc69797746 Index recording copies by 64 KiB block instead of 4 KiB page: AddControlMemory walked about 500 pages per call under the exclusive copy lock (53 calls a frame, 2 MiB ranges on the 1.4.3 field), and the main thread and job workers recording command buffers waited on it; field walk 72-77 fps to 78-80
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 10:17:14 +03:00
Valiantsin DzerakhandClaude Opus 5.5 95ae91687c Spin short SleepThread calls on the performance counter: the high-resolution waitable timer ended the SDK's 20 us polling sleeps after 460-515 us on the main thread and the job workers; the sched trace now reports wake-to-run and core-grant-to-run latency and asked versus taken sleep time per thread
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 10:17:13 +03:00
Valiantsin DzerakhandClaude Opus 5.5 639d00efb4 Merge the main thread wait tracing: TOTK_TRACE=sched reports the guest main and render threads' waits by object and by the thread whose signal ended them, and TOTK_PROFILE_MAIN=1 samples the main thread alone with call chains
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:54:28 +03:00
Valiantsin DzerakhandClaude Opus 5.5 a653f4cdd1 Merge the driver cache correction: a relink keeps the AMD driver's pipeline cache, so build_game no longer re-warms it after every relink; the cold libraries came from preparing another build dir, and --warm-shaders stays for that case
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:54:28 +03:00
Valiantsin DzerakhandClaude Opus 5.5 c79ed91d37 Merge the recording copy fix: a finished NVN recording is copied in one exact-size allocation and one memcpy, and a called recording is inlined as one block, instead of record by record with doubling reallocations on the game's main thread
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:54:27 +03:00
Valiantsin DzerakhandClaude Opus 5.5 ff75f32189 Copy a finished recording in one exact-size allocation and one memcpy, and inline a called recording as one block: record-by-record copies with doubling reallocations held the game's main thread about 1 ms a frame (MacNativeGames 5189ade does the same)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:44:53 +03:00
Valiantsin DzerakhandClaude Opus 5.5 bd9851f4f2 Trace the guest main thread's waits by object and by the thread whose signal ended them (TOTK_TRACE=sched), and sample the main thread alone with call chains (TOTK_PROFILE_MAIN=1)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:23:35 +03:00
Valiantsin DzerakhandClaude Opus 5.5 cac06a76d3 Merge the 1.4.3 install crash and window fixes: patches.c compiles per install, so a 1.4.3 install commits its own patch page instead of 1.2.1's (crash reading 0x7003adf000 soon after the save mount), and the game window opens behind the user's windows without taking focus
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:18:23 +03:00
Valiantsin DzerakhandClaude Opus 5.5 e9865f188c Open the game window behind the user's windows without taking focus
The window appeared over everything with the game's blank blue loading frames and took the focus. It now opens at the
bottom of the z-order, flashes its taskbar button until the player brings it up, and applies the fullscreen setting at
that first activation. The game counts itself in focus until the first WM_ACTIVATEAPP, so loading goes on behind;
keyboard input already waits for the foreground window.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:16:46 +03:00
Valiantsin DzerakhandClaude Opus 5.5 7fefa86fe3 Compile patches.c per install instead of into the prebuilt runtime library
The release built libtotkrt.a against generated/v121, so patches.c baked in 1.2.1's patch page (0x700472b000, the page
after 1.2.1's main). A 1.4.3 install lifts its code with the 1.4.3 patch table, whose patched sites read the page after
1.4.3's main (0x7003adf000): the runtime never committed it and the game faulted there soon after mounting the save, while
frame_rate.c wrote its frame step into sdk memory at the 1.2.1 address. patches.c now joins romfs.c in INSTALL_SIDE and
the release ships its source, so it compiles against the player's patches.h.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:12:15 +03:00
Valiantsin DzerakhandClaude Opus 5.5 5784084590 Drop the driver cache warm after every relink: a relink keeps the AMD cache (29,587 variants re-warmed in 1.8 s), the cold libraries in play came from preparing another build dir, which evicted this one's; --warm-shaders stays for that case
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:10:43 +03:00
Valiantsin DzerakhandClaude Opus 5.5 e1476494d5 Merge the save commit and driver cache fixes: a save writes only its changed files and hard-links the rest (300-380 ms freezes at each save), and build_game re-warms the GPU driver's cache after relinking a prepared build (40-220 ms freezes on every new effect after a rebuild)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:01:54 +03:00
Valiantsin DzerakhandClaude Opus 5.5 7daeffdc08 Warm the GPU driver's cache again when build_game relinks a prepared build: the cache is keyed to the exe, and after a relink every new stage library took 40-220 ms in play instead of about 2 ms; totk.exe --warm-shaders builds the prepared variants from the shader cache without translating
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 09:00:54 +03:00
Valiantsin DzerakhandClaude Opus 5.5 138a7d3a99 Commit a save by writing only the files that changed and hard-linking the unchanged ones to the committed save: the game rewrites 3-4 of 82 files per save, and flushing all 82 held the game thread for 300-380 ms
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 08:55:21 +03:00
Valiantsin DzerakhandClaude Opus 5.5 875a4af317 Ship tools/build_shaders.py in the release: build_game.py imports it since the ASTC compute decode, and the installer's compile step failed on the missing module
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 08:26:57 +03:00
Valiantsin DzerakhandClaude Opus 5.5 6d216840dd Merge the device-local memory census: a mem: vram line every 600 frames with TOTK_TRACE=perf, and backlog item 20 with the released texture entries as the grower and machine B at its budget
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 08:14:21 +03:00
Valiantsin DzerakhandClaude Opus 5.5 ad3fa470d1 Merge the replay CPU work: texture lookups and descriptors kept per slot, image views per image, texture types only for the slots a shader reads, the previous draw's pipeline reused, vertex buffer spans, plain-add traffic census, completion wakes only when due, storage checks per stamped 2 MiB region and shader stages per program serial; fight replay 13.8 -> 9.2 ms a frame on A, field 9.5 -> 7.2 ms
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 08:14:21 +03:00
Valiantsin DzerakhandClaude Opus 5.5 a3bbeac08a Record in backlog item 20 what holds device-local memory, the released texture entries as the grower, machine B at its budget and the plan for an age-ordered drop under pressure
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 07:51:12 +03:00
Valiantsin DzerakhandClaude Opus 5.5 9758b8365e Log what holds device-local memory every 600 frames with TOTK_TRACE=perf: VMA blocks and allocations, texture entries live and released with the bytes of their own memory, VRAM mirrors and pending releases
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 07:39:08 +03:00
Valiantsin DzerakhandClaude Opus 5.5 2bcba6bc3f Keep which shader of a program serves each stage per program serial instead of reading every stage's control block several times a draw
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 01:01:54 +03:00
Valiantsin DzerakhandClaude Opus 5.5 b92f113073 Check a texture's storage pages only in the 2 MiB regions stamped since its last check: an unchanged 4 MiB texture read 1,024 page generations, at about 1,100 checks a frame
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 01:01:54 +03:00
Valiantsin DzerakhandClaude Opus 5.5 895aa43ab9 Wake the completion thread for a new callback only when its value is already reached: the replay's callbacks wait for the submission it records, whose submit wakes the thread anyway
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-27 00:56:48 +03:00