The FP helpers compared c->fpcr with a cached copy before every operation to resync MXCSR lazily. MXCSR now
follows FPCR.FZ at once: MSR FPCR calls rt_fpcr_sync, and so do rt_call (thread, fiber and host entry into guest
code), fiber resume (FIBER_FLAG_FLOAT_SWITCH fibers keep their own MXCSR) and the differential test harness per
case. op and vop only test FPCR.RMode for the slow path, and their redo check is one ordered-not-equal compare of
|r| with the smallest normal, the same NaN-or-smallest-normal set as before.
Hot guest code shrinks by about a quarter (132 functions holding half the guest cycles, 1.4.3); fight frame time
moves 1-3%, inside the run-to-run spread. lift, simd, func, atomic and flags_check pass on 1.4.3; simd.py fails
without the harness sync.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With TOTK_TRACE=sched, TOTK_CRIT=<lo>-<hi> records every state change and wake of the guest threads and every
present over those frames (runtime/kernel.c crit_add from acct() and wake(), crit_frame at the present) and writes
them to TOTK_CRIT_OUT at the last frame. tools/crit_path.py walks back from each present through the waits to the
thread whose signal ended them and splits the frame interval by thread and state; the sums equal the interval.
A 1.4.3 fight: with the replay on the submitting thread the path is 7.3 ms of replay on the render thread and 4.3 ms
of guest code; with TOTK_REPLAY_ASYNC=1 it is 10.1 ms of guest code through the fork-join of the main thread and five
workers, at the same frame interval.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every flag-setting instruction packed N, Z, C and V into c->nzcv and every reader unpacked them. recomp/flags.py
now runs a reaching-definitions pass per function: a compare stores its operands in a local Flags F, and a reader
whose definitions are all of one kind computes its condition from them, so cmp and b.cond become one host compare.
CCMP, FCMP, MSR NZCV and ADCS write the packed form, which readers reached by several kinds also use. Calls and
returns publish nothing (AAPCS64 neither passes nor preserves NZCV); tail branches, falls into the next function and
br through rt_dispatch publish c->nzcv, since discovery can start a function inside another one. Functions with an
unresolved computed br keep flags in c->nzcv.
1.4.3 main: 492k direct and 89k packed flag reads, .text 342 -> 299 MiB. Fight bench (bench-fight2-2026-10-27b,
frames 5400-7500, uncapped) 13.15 and 13.63 ms vs 13.9 before. tests/flags_check.py checks the new helpers against
the packed ones; tests/diff/func.py no longer compares NZCV, which generated code leaves untouched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
TotK finalizes and initializes transient textures every frame: the global tex_gen and slots_gen moved about every 16
draws in the monsters fight, so every cache keyed on them (the slot cache in vk_texture_by_id_locked, the texture types
of answers_for, the texture descriptor memo) lived for a few draws. A slot resolves through the entries at its own
storage address only, so each address hash now has its own generation and each slot a token
(vk_texture_slot_token: registration serial and address generation, lock-free). The memos check the global generation
first and the slot's token only when it moved.
The descriptor memo no longer drops an entry at each submission (the first binding still makes the storage check; a
re-upload writes the same image) or pass (the feedback layout is set on each hit from the kept image), and is twice
the size. feedback_scan keeps each handle's sampled subresource across passes instead of looking up every fragment
texture after every texture bind. bind_vertex_buffers keeps each stream's resolved span while its range and
vk_mem_gen stay the same.
Monsters fight on the RX 9070: replay 12.0 -> 10.8 ms per frame; field 8.1 -> 7.4 ms. TOTK_TEX_MEMO=check and
TOTK_SHADER_MEMO=check runs found no difference.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tex_info resolved the slot, built a view key, probed the view table and looked up the sampler for every texture
binding of every draw. A direct-mapped memo keyed by handle, resource type and separate returns the descriptor while
vk_texture_slots_gen, vk_sampler_gen, the submission epoch and (with feedback-loop images) the pass are unchanged;
storage images are not kept. TOTK_TEX_MEMO=0 turns it off, =check recomputes every hit (a fight run with it and
TOTK_SHADER_MEMO=check stopped on nothing). Against main, profiler on, no bind stats: fight replay 13.9 to 11.9 ms per
frame, render thread busy 16.3 to 14.6 ms, CPU-bound windows 15 to 9; field replay 9.5 to 8.2 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Views made per command buffer were created anew every frame: vkCreateImageView under get_view took about 8% of the fight
replay on the 7900X. They now live in one table keyed as before, used under the texture lock, and destroy_tex hands an
image's views to the deferred release with the image (vk_exec_image_gone). The buffer range limits ubo_info and
ssbo_info read on every binding are read once. Fight replay 14.0 to 12.5 ms per frame (15.7 before the branch), field
9.5 to 9.2 ms; a fight run under the validation layer shows only the known cp_t_tcb_20 message.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
answers_for looked up the type of all 64 texture and image handles of a stage whenever one changed; the use mask of the
first shader of each program stage now limits it to the slots that select the variant, with a generation per slot and
the texture lock taken only on a miss (texture type recomputations per draw 0.39 to 0.05 in fight). prepare_draw keeps
the previous draw's pipeline info while the pipeline state is equal (TOTK_PIPE_MEMO), and reads the clock only for
draws that may wait. Fight on the 7900X: replay 15.7 to 14.0 ms per frame, 53.3 to 58.7 fps; field 11.1 to 9.5 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>