With shadow off the back end turns on guards, each with its own switch: the mirror syncs copy the synced pages into staging on the CPU at submit and wait for an earlier write-back into the same range first (TOTK_SYNC_AHEAD_MSYNC), a texture the CPU rewrote takes the CPU deswizzle (TOTK_SYNC_AHEAD_DESWZ), an invalidate of a range not read lately waits for the pending GPU write into it, from a bounded list pruned at completion (TOTK_SYNC_AHEAD_READS), and the frame wait waits up to the value of a one-off small target (Savage Wind's 16-frame, 64-entry rule, TOTK_SYNC_AHEAD_ONESHOT) or, off by default, of a samples-passed report (TOTK_SYNC_AHEAD_QUERIES). TOTK_SYNC_AHEAD_CHECK=1 runs the check-mode detector with the late wait itself. The shadow form and the default are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A binding the replay resolves again in the same submission hits a per-thread table of the last notes, and the byte
compare walks the memory state once per union range instead of once per written page: the check's game-thread cost on
the islands bench went from 5-9 to about 3 ms a frame. The log line gives the notes taken and the time of each step.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Whether the game writes what the GPU still reads does not depend on this machine's timing, so check mode now takes the
late wait's value at every frame wait (the frames where the GPU really was behind are counted apart). The set's bytes
are copied at the snap: a written page counts its changed bytes too, and the mirror sync copies are narrowed to the
bytes the frame binds from the mirror (class bound), since a page written only outside them races nothing it reads.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The replay notes each GPU-timeline access of guest memory with its submission's value (in-place bindings of imported
pools without a mirror, mirror sync copies, GPU deswizzle sources, mirror write-backs); at the frame wait the notes of
the frame the late wait would leave running form its read set, snapped with the write watch and checked again at the
next frame-wait entry. Invalidates and small-target write-backs of GPU results from that frame count as stale
readbacks. One summary line per second and a run summary with the largest offenders by pool (runtime/vk_ahead.c).
--probe-vkmem checks the detector with a synthetic write after the wait; tests/sync_ahead_check.py checks that check
mode keeps the wait exact.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
vk_objmv_wanted is FSR 3.1's latched upscaler half (DLSS with NGX up, or FSR 3.1 picked at start) or frame generation following the scene with its objects (vk_fg_objects_on), and vk_mvprev_frame turns the tables on for the running upscaler or for frame generation. vk_usnap.c keeps this branch's host regions: a region's host flag says whether this submission wrote its host copy, which covers frame generation going on and off live without fg's separate U.host pointer. The jitter fix in mv_pass stays a single copy.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Field still view, 28 s, TOTK_MVPREV_CHECK=1: mv_check 0 of 60 checks with over 0.5 % object pixels fit better by a variant; mv_prev intact 100 % in c5, c6, c7, c14, c15 and c16; 0 of 214238 last-frame copies differ from the neighbour copies.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Lifetime: a scene submission's arena (vk_usnap_scene, from the scene draws vk_mvprev notes) is kept at its vk_usnap_end as the last-frame copy instead of going to the free list at its fence. The next scene submission's end, or any end once the motion is off, hands the kept arena to the completion thread with that submission's fence, then keeps its own. The motion draws that bind the kept arena are listed and recorded within the next scene submission (vk_mvobj_execute drops a list of another fence), so the GPU has read it before that fence; the fences of one queue complete in order, so its own submission's reads are done too. One arena is held beyond those in flight, whatever the session length; vk_usnap_shutdown frees it with the others. A frame whose scene spans two submissions finds only the last one's blocks; the others keep the camera's motion.
TOTK_MVPREV_CHECK=1 still copies the neighbours and compares them with the kept copy. Field still view, 28 s: the mv_prev lines intact 100 % in c5, c6, c7, c14, c15 and c16; over whole blocks 87 of 218320 last-frame copies differ from the neighbour copy (the game changed the other copy after last frame's submit; the kept copy is what last frame drew); mv_check 0 of 60 checks with over 0.5 % object pixels fit better by a variant, object end-point error median 0.98 px against the camera's 1.27; the c7 lookups split as before (nearer to side 1.87, reference 1.83).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Noting only the candidates' blocks was tried and dropped: the table then misses the other draws' blocks, a sixth of the c7 "nearer" picks became "side" ones, the copies mo_similar refuses doubled and the field still check got 5 of 60 checks with over 0.5 % object pixels fit better by a variant (0 before).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Since the jitter went into the game's camera (#1284), cam_unjitter writes the scene camera's snapshot back without it from the scene's end, before the DLSS point where vk_mvobj_execute redraws the moving objects through that snapshot. The draws rasterised unjittered and their fragment shader took the jitter off a second time. TOTK_MV_CHECK on a still Link (FSR 3.1 path, task/fsr31) measured the object part equal to the frame's jitter in every check; with the jittered rows written back around the motion draws and taken off after them, object pixels on a still Link fell from 1.2-2.1 % to about 0, and the object vectors beat the camera's (median end-point error 0.68 px against 0.88, before 0.93 against 0.82).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Since the jitter went into the game's camera (#1284), cam_unjitter writes the scene camera's snapshot back without it from the scene's end, before the DLSS point where vk_mvobj_execute redraws the moving objects through that snapshot. The draws rasterised unjittered and their fragment shader took the jitter off a second time. TOTK_MV_CHECK on a still Link (FSR 3.1 path, task/fsr31) measured the object part equal to the frame's jitter in every check; with the jittered rows written back around the motion draws and taken off after them, object pixels on a still Link fell from 1.2-2.1 % to about 0, and the object vectors beat the camera's (median end-point error 0.68 px against 0.88, before 0.93 against 0.82).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
scene_upscaler (off | dlss | fsr3, read at start) replaces the dlaa toggle on the Screen tab; without the key dlaa = 1
still means DLSS, so nobody gets FSR 3.1 without picking it. The DLSS mode row is the upscaling mode of both engines,
the DLSS model, characters' mask, fixed exposure and effects-after rows show only with DLSS, and the performance
overlay names the upscaler that ran and its mode. resolution.c sizes the scene for the engine picked at start.
runtime/vk_fsr3.cpp drives the SDK's upscaler context on its Vulkan backend directly, as Savage Wind's fsr3.cpp does:
one context per output size with the output as its largest render size (the modes and the dynamic resolution need
no new context), HDR input (the light target before the bloom, where DLSS reads it), render-size motion vectors with
the per-object pass, the camera jitter, the camera's planes and field of view from the scene block (vk_fg_camera; a
cut starts a new history), FSR's sharpening off. vk_dlaa.c calls it at the DLSS point after the recorder's drain;
the NGX path is left as it was. The backend's per-frame rings hold four dispatches: a dispatch waits for the one three
back should it still be in flight. Frame generation stays off while FSR 3.1 runs (logged once).
The SDK subset gains the upscaler component, its shader headers, FSR 1's (its RCAS) and its GLSL, from Savage Wind's
copy unchanged; tools/ffx_shaders.py --effects fu builds its blobs (HDR, 32- and 16-bit floats) into
spirv/ffx_fsr3upscaler_vk.h, and vk_fg.cpp's blob lookup hands those out through vk_fsr3_blob. The luma history's
storage image is declared rgba16f, its images' format (rgba8 in the SDK; the validation layer reported it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Revivedx <82332536+Revivedx@users.noreply.github.com>
In the town the scene moved between 1600x896, 1432x808 and 1280x720 under a 1080p output; vk_fg_prepare compared the
scene size with the contexts' and made the contexts again each time, 58 times in 222 s of a user's session, each after
vkDeviceWaitIdle (89 ms measured). The contexts are now made for scenes up to the output's size (or a larger scene's,
once one comes) as the SDK's maxRenderSize, and each frame passes its own size as renderSize to the prepare and the
interpolation: the SDK sizes its images by maxRenderSize, dispatches over renderSize and divides the motion vector scale
by it. A frame at another size than the last starts the interpolation over (its dilated depth and vectors from last
frame are of the other size); the summary line counts those.
Checked: 60 s in the town with two scene size changes, no remake and the replay thread's FG time at most 6.4 ms; a run
at render_scale = 50 (960x536 scene, 1920x1080 contexts) generates Link running without a double image.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The motion image FG read held the camera's motion alone (fg_motion called mv_pass without its object pass). The camera
follows Link, so his pixels carried the world's motion and the interpolation drew him twice, with a smeared strip where
he had been; NPCs and enemies did the same. fg_motion now runs mv_pass with the object pass (vk_mvobj_execute), and its
gates follow vk_objmv_wanted: DLSS once NGX came up (vk_dlaa_available, as before) or FG following the scene with its
objects (vk_fg_objects_on). vk_mvprev_frame keeps last frame's block tables for either, vk_usnap_begin makes the host
copy for either and vk_usnap_bytes reads the copy that submission wrote (FG turns on and off live), and
vk_dlaa_prev_camera answers whenever the scene is followed. FG is off with DLSS, so every DLSS gate evaluates as before.
fg_objects = 0 keeps the camera's motion alone for comparisons.
Measured on A (RX 9070, 7900X on 4 cores, 1080p, field, 60 s): 85.6 drawn with fg_objects = 0, 80.0 with 1 (about
0.8 ms a drawn frame on the replay thread, half of it the snapshot's host copy), object pass 0.035 ms GPU. Still grass
gets no false motion (SDK debug view: the game vector field stays within 0.2 px), and grass does not break up while the
camera turns. Validation: 0 errors.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The row and its description (for displays of 120 Hz and more, the game at half the limit, about half a frame more input delay, not with DLSS or HDR) are in the menu's catalog for English, Russian, Portuguese (Brazil) and Simplified Chinese. frame_gen is live: vk_fg.cpp vk_fg_engine reads it again after a settings change, the contexts are made at the first frame after it is turned on (a hitch of about 0.1 s once), turned off they stay unused until exit, and gfx_present.c builds the swapchain again when the extra image for the pair comes or goes. The full overlay gets a Shown row. totk.ini.example describes frame_gen.
The shown-frames counter and the overlay lines follow PR #1089 by xellow78-lgtm.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: xellow78-lgtm <336264245+xellow78-lgtm@users.noreply.github.com>
vk_backend.c present_pair: both blits are submitted at once (gfx_present.c vk_present_blit_pair, the flip queue of two), the generated picture is presented after a timeline wait on the frame's last submission and not before the pacer's display time, the game's frame after a wait on the value after both blits and half the smoothed interval between generated pictures (factor 0.1) after it, as AMD's frame interpolation swapchain and Savage Wind's present_wait_turn do: spaced by the CPU clock alone both pictures went on screen together under FIFO and G-SYNC. The swapchain gets one image more while frame_gen names a generator, for the two images held at once; the present slots stay at two (each blit waits on the fence of the blit two before, the previous frame's same picture, already shown). The generated picture's blit lays the menu and overlay out, the game's draws them again; colour and FSR 1 run per blit; the flight recorder keeps the game's frames; the generated present carries no present ID. frame_rate.c frame_step_cap_ns halves the limit, 30 drawn at least, while the last frame went out with a generated picture, so menus and fades keep the whole limit; the real-time step follows the game's presents and stays the drawn interval. With frame_gen = off the present path is unchanged. "fg present" log lines give shown and drawn fps and the present intervals (mean, sd, p99, each half of the pair) in the world.
Measured on the RX 9070 and the 7900X on 4 cores, field save, camera turning, 1080p output: limit 120 immediate 117 shown and 60 drawn, intervals 8.34 ms sd 0.16 from generated to game's, 8.38 sd 1.71 back; FIFO at 144 Hz the same at 120, and 141 shown and 72 drawn without a limit; immediate without a limit 169 shown and 86 drawn against 94 drawn with frame_gen = off. Replay-thread cost 0.59-0.65 ms a drawn frame.
The flip queue, the pair blit, the halving rule and the present without an ID follow PR #1089 by xellow78-lgtm.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: xellow78-lgtm <336264245+xellow78-lgtm@users.noreply.github.com>
vk_fg.cpp vk_fg_engine is the one place that reads the string and maps it to VkFgEngine; an unknown value is off and logged once. The setting was never released, so 0 and 1 are not kept as aliases.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
vk_dlaa.c follows the scene while frame_gen is live and DLSS is off (track: the arming and end of the DLSS path, without jitter, reactive mask or per-object motion), fg_motion runs mv_pass on the camera alone and vk_fg_prepare, vk_dlaa_hud takes the picture before the HUD by PR #1089's rule once the bloom chain began, vk_dlaa_sub_end runs the interpolation at the end of the submission that drew the window texture. vk_fg.cpp reads the camera's planes and depth convention from the scene block's projection (standard depth, near 0.1 m, far 25 km in TotK 1.4.3), makes the contexts again when sizes or the convention change, resets the history on a cut (10 m or 35 degrees, as Savage Wind), a frame without camera motion and after any frame without a generated picture, and stays off with DLSS, replay workers, sync_ahead, HDR and while a menu background or a fade is open. fg_test, fg_debug, fg_shot and fg_times are the checks; GPU timestamps and the replay thread's recording time go to the log.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Revivedx <82332536+Revivedx@users.noreply.github.com>
Co-Authored-By: xellow78-lgtm <336264245+xellow78-lgtm@users.noreply.github.com>
The SDK subset, its compat header and the SPIR-V shader header come from Savage Wind (MIT); tools/ffx_shaders.py regenerates the header with glslangValidator, build_game.py compiles the SDK sources with ffx_compat.h force-included, and runtime/vk_fg.cpp makes the backend, optical flow and frame interpolation contexts after the device exists when frame_gen = 1 (default 0, read at start).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Revivedx <82332536+Revivedx@users.noreply.github.com>
The firmware files keep the font XOR-encoded. The game's font loader reads the data at GetSharedFontAddress as a font only when it starts with an sfnt tag; otherwise it calls nn::fontll::ScalableFontEngineHelper::Decode, which returns NULL for the encoded payload, and the face is left without a name. font_data now decodes the payload in place after ov_bfttf_size accepts the file and checks for an sfnt tag. The services probe checks the unchanged header at Address - 8 and plain sfnt at Address. Same fix as MacNativeGames pull request #2145 (shared_font.c load).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With mouse_mode = direct the mouse reached only the cameras the direct hooks cover. Autobuild's placement camera runs
none of them on games 1.4.3 and 1.2.1 and turns by the right stick, so the mouse moved nothing there. mouse_direct.c
now notes each camera step that runs a hook; after two steps without one the mouse acts as the right stick
(input.c mouse_camera), as with mouse_mode = stick, until a hook runs again. A script's MOUSE lines move that stick
too, also under TOTK_INPUT=none, so the path can be tested with scripted input.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Patch 0017 adds two support buffer fields, which every translated shader declares, so the SPIR-V hash of the game's FXAA shader changed (0x5c43ae18948d52cb to 0xde20aef90fea4ce6, with zero_locals 0xaad42238801d31d8 to 0x755964e8dce7f34d). vk_dlaa_skip_draw then no longer skipped the scaling draw DLSS replaces, and on 1.4.3 that draw, sampling DLSS's output-sized picture in place of the scene, stretched the picture's top left over the whole screen in every DLSS mode but dlaa. FXAA_FS gets the two new hashes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The pcss radii and penumbra factor were in UV of whichever cascade map a pixel read, the same in every cascade. Each cascade's map spans 3.5 to 5 times the ground of the one before (measured on 1.4.3: the cascades end at about 4, 27, 120 and 450 m), so a soft shadow turned that much softer at every cascade step, a step that moves with the camera (issue #1291).
The runtime now measures each cascade from the camera block the lighting shaders read (c4 rows 42 to 57, a projective matrix per cascade, and row 58, the cascade ends): the UV and depth per metre where the cascade begins and ends. It passes two factors per cascade in two new support buffer fields, pcss_depth and pcss_scale, and translator patch 0017 applies them in the PCSS lookup. The values keep their meaning in the second cascade, which holds most shadows on screen, so the look at a given pcss value stays where most players see it. The narrowest radius stays in texels. The HDR field moves from byte 1440 to 1472. With pcss = 0 nothing changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A call adds no command memory since the previous commit. Control memory grew only at EndRecording, so a buffer queried
after calls and before its first EndRecording reported 0 command and 0 control bytes, which the game treats as an empty
buffer. The game's inline copy of the driver's call (1.4.3 main+0x2d33a4) writes one 24-byte control entry per segment
of the called handle; each called handle now adds that much. The NVN probe checks that a call-only buffer reports 0
command and some control memory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The real driver puts a call in control memory, so a call adds no command memory; a copy adds what the copied commands
take on hardware. We added our record bytes of the inlined recording for both. On 1.4.3 the game queries recordings
that hold calls every frame (field save, 20 s: 15,700 such queries from six call sites, up to 14,604 bytes reported
where the driver rule gives 3,204); copies were not seen. Each recording now keeps its hardware size from EndRecording
(copy_prefix.hw), and memory added while recording keeps that size whole. Same accounting as MacNativeGames 6909dbb7,
which measured 40 bytes on the original driver where their port reported 88 and traced a renderer heap crash near
Zora's Domain to the larger sizes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
input_rumble used to call SDL_RumbleGamepad under the input lock; on a Bluetooth pad the write can take tens of milliseconds and stalled SendVibrationValue and every other user of the lock. It now stores the newest request and wakes a rumble worker, which sends only the latest one. The worker takes its own handle on the open pad (SDL_OpenGamepad of its id) under the lock and writes without it, so closing the pad meanwhile cannot free it under the write. A pad going away drops the pending request, and exit sends a zero request. TOTK_TEST_PAD_RUMBLE_MS makes the test gamepad rumble with a slow write. Idea from Savage Wind Recomp commit a8ad674 (issue #730).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The previous commit read the combo from the merged buttons, where a key bound to ZL is the same bit as the left
trigger: with freecam_pad = zl+zr, the right mouse button and F switched the free camera in play. The keyboard has
freecam_key for that, so the gamepad combo now counts the gamepad and the input script, which stands in for one.
runtime/input.c poll_merged: script_apply writes its buttons into script_seen, which is then merged into the buttons as
before. freecam_gate counts a position of the combo when pad_seen has it or script_seen has the Switch buttons it stands
for; Home stays gamepad only. docs/controls.md ("Free camera") says so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
freecam_gate read the combo from the physical gamepad only, so a tester's input script that pressed ZL+ZR+L3
sent it to the game instead (#1213). Each position of the combo now also counts when the Switch buttons it stands
for are down in the merged buttons, as the Minus+Plus menu combo already does; Home has no Switch button and stays
gamepad only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
freecam_key and freecam_pad were F6 and ZL+ZR+L3 for everyone, potentially causing unwanted toggles. F6 is also the
quick save key of #1188. Both are none now.
runtime/cfg.c: the FREECAM_PAD and FREECAM_KEY rows of the settings table default to none. runtime/input.c
freecam_pad_mask reads the combo with cfg_value instead of falling back to its own zl+zr+lstick, so an unset combo is
the table's none like the key. The comments in runtime/freecam.c, runtime/input.c and runtime/ui/menu.cpp
rows_freecam, docs/controls.md ("Free camera") and totk.ini.example say so and keep F6 and ZL+ZR+L3 as examples.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
input.c freecam_gate also switches the free camera on a press of TOTK_FREECAM_KEY (f6 by default, none for no key).
While it is on, ZL, ZR, L and R come from the merged Switch buttons, so the keys bound to them (Z, X, Q and R by
default) go down, up and change the speed like the gamepad's; the keys' sticks already flew and turned it. The mouse's
counts are taken in the poll and turn it at the game camera's rate (0.08 degrees a count times mouse_sens, through
freecam_mouse), and input_direct_mouse_sample gives the game's mouse camera none meanwhile; in the stick emulation mode
the mouse arrives in the right stick as before. The Controls tab gets a Free camera key row (off, F2 to F7, F9), with
text in all four menu languages; docs/controls.md lists the keyboard column.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A row under the walk rows offers the trigger pair or the shoulder pair with either stick click, or off, named as the
open pad labels its buttons (binds_page.cpp binds_pad_label, the Controls tab's labels). TOTK_FREECAM_PAD gets a live
row in the cfg table with its default zl+zr+lstick; totk.ini still takes any combo of button positions, shown as
written. On other game versions than 1.4.3 the row is off and says so. Text in all four menu languages.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The camera component update (main+0x628460) copies each component's eye, target and up into its sead camera. The site
freecam (main+0x6286d4, the load every path to that copy runs first) calls runtime/freecam.c freecam_hook, which
writes the component's one-shot eye and target overrides (+0x180, +0x198) while the free camera is on; the update
takes them in place of the pose and clears them, so the game camera's own state is never changed and resumes the
frame the free camera ends. The up vector is levelled through the component's up offset (+0x200), which is kept and
put back. Of the three components a frame updates, the hook drives the two that share the player camera's eye, leaves
the one at the default pose and picks again after a load; with the free camera off it does only the displaced load.
input.c freecam_gate latches TOTK_FREECAM_PAD (zl+zr+lstick when unset, none for no combo) like Minus+Plus and, while
the free camera is on, gives it the sticks, the trigger axes and L and R presses and the game nothing. The left stick
flies along the view and sideways, the triggers straight up and down, the right stick turns, each L or R press halves
or doubles the speed (0.25 to 512 m/s, 8 at the start); the flight follows the wall clock. TOTK_FREECAM_PATCH=0
removes the hook.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>