Commit Graph

696 Commits

Author SHA1 Message Date
chrg6
96d656ba1d Time the staging allocation's wait for an upload batch, and say what the batch limit and the staging pool do now.
A staging buffer that does not fit the staging pool's blocks waits for the oldest upload batch with the upload lock
held, and a replay submission that sends its uploads waits for that lock. The wait cannot leave the lock: the caller
holds the open batch's command buffer and the buffer it asks for belongs to that batch. What the wait costs on an area
load was not measured anywhere, so it is timed and logged now: the first four waits, then at every doubling of their
count, with the total and the longest.

Since staging stays inside the pool's blocks every batch is submitted at 64 MiB of staging, so upload_batch_mb = 0 is
that limit and not "no limit", and the values above it (the low-end mode's 256, the menu's 128 to 1024) change
nothing; a discrete card keeps 256 MiB of staging for the run, not 128. The settings table, the menu's text,
nx2navi.ini.example, the audit's table and the comments say so.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 22:27:25 -05:00
chrg6
345b393acc Give a texture handle asked for with an id out of range the reserved slot 0, inside the pool arrays.
The handle named slot 1,048,575, past the 32,768 entries of the pool arrays shaders index. A draw that binds the
handle resolved it to the stand-in, but a handle the game stores in a uniform buffer would have indexed those arrays
out of bounds on the GPU. Slot 0 is one of the 256 slots the device reports as reserved, which no game registers
into: it holds the stand-in in the arrays, and a bound handle of 0 is "no texture" already.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 22:25:01 -05:00
chrg6
4a6ba66115 Tell a start that was closed or stopped during the boot's builds from a driver crash, and limit the whole warm.
A marker or a journal left by a start that died in the boot's builds was always read as a driver crash. Since the
whole warm those builds run for minutes, so a start ended from outside in that time (the task manager, a program that
stops the game, the power) switched the pipeline prebuild off for that driver for good, or put up to 32 variants on
the skip list that the driver had merely been building. Closing the window was already an orderly end and still is.

The crash handler now leaves a note while the boot's builds run. With the note the next start does what it did: it
skips the prebuild, or the variants the journal names. Without it the end is taken for a stopped start: the pipelines
are built again and the journal's variants are blamed on nothing. Three such ends in a row still skip the prebuild,
for a driver that ends the process without a fault the handler sees. The boot's journal has a file of its own, so the
preparation's keeps its meaning.

The whole warm waits prewarm_full_s seconds at most (600): when they are up the game starts, the builders step down
to the background's thread counts and priority, and the file says "done", so the boot screen has one turn per driver.

tests\boot_warm_check.py covers what a leftover marker means with and without the note.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 22:24:22 -05:00
chrg6
e548180da2 Hand a pool's device address record to its new mirror by a store the GPU runs after the mirror's first copy.
The table still named a new mirror at record time, as the pages' twin: parts already handed to the queue wrote
through the table into the new block before its first copy, the same early access the delayed read was made to stop.
The reads then moved to the mirror some submissions later by a second change that took a second record, cleared its
note before it knew the record was there, and until it ran left the table's readers on guest memory, without what
other shaders had written into the mirror through a buffer binding.

Nothing is changed when the mirror is made now. The prologue that holds the mirror's first copy also holds a store
into the pool's record, after the copy: what was queued before it reads and writes guest memory through the table as
if no mirror existed, what comes after reads the mirror and writes both. No second record, no second change, no
window in which the table is behind. gpa_late = 0 still changes the table at once, as up to 0.9.2.

--probe-vkmem makes a mirror in a recorded submission and reads the table's own memory back: the page names the
mirror with guest memory as the twin, and the mirror holds the pool's bytes. tests\gpa_check.py runs the store as the
GPU would, before and after. A table with no record left now says so once in the log.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 22:18:14 -05:00
chrg6
a8639f1cb6 Keep the scene and the picture inside the shipped render-target heap when the grown heap table is not served.
The scene's share followed the settings alone. When the grown table could not be built or written (a table in a form
that is not rewritten, as a mod's may be, or a profile folder that cannot be written), the game kept the shipped heap
and the scene still went to the whole picture inside it, which is the crash the growth was made for.

The start now asks for the table before the first present and looks at the answer: without the grown table the scene
is three quarters of the picture at most and the picture stays the game's own 2560x1440, with a log line that says so.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 22:09:37 -05:00
chrg6
7ff1a21ef2 Grow the render-target heap for the scene's share only where the start's mode needs it, and never past the pool.
The heap grew by the full-picture factor whenever the upscaler setting was on, in every mode, and without a limit:
915 MiB at 3840x2160 where the modes below three quarters need 382, and at 10240x2880 an application pool of 13.5 GiB,
past the 12 GiB heap region, where the game's SetHeapSize is refused and nothing starts.

The share is now the start's own mode: all of the picture for dlaa, the game's own most for the modes below it. The
growth stops at what the pool can take (2,478 MiB): the scene's share is lowered to what fits, with a log line, and a
picture too large even for the game's own scene gets the most there is and a warning. A mode chosen in the menu that
renders more than the start's keeps the start's share until a restart, as the log and nx2navi.ini.example now say.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 22:08:44 -05:00
chrg6
cd0b1a387d Do the whole warm on the boot screen at the first start after a driver change.
A start that finds the pipeline cache written by another device or driver has nothing built, and everything was
compiled beside the game: an 8-core integrated GPU drew fewer than 600 frames in its first 11 minutes. Such a start
now waits on the "Preparing shaders" screen, with counts, for every recorded pipeline, for the compute pipelines on
every core at normal priority, and, when a sample says the driver's own cache lacks the prepared stage libraries, for
all of them on every core. The memory gate still holds for the builds, and the library builds write the preparation's
journal while the start waits, so a variant the driver's compiler dies on is skipped from the next start on.

boot-warm-owed.txt beside the caches names the GPU and driver: a start closed during the warm is followed by one that
goes on with it, three at most, and "done" keeps later starts ordinary while the cache file still carries the earlier
driver's header. When the boot stops waiting (prewarm_wait_ms, or a minute without progress) the builders step down
to the background's thread counts and idle priority. prewarm_full = 0 turns the warm off.

With nothing in the driver's cache the probe took 425 s on an RTX 5090 with 16 CPU threads: 87 s for 5,775 recorded
pipelines, 19 s for 4,075 compute pipelines and 317 s for 42,231 prepared variants.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 20:03:12 -05:00
chrg6
016c90fafb Wait at the start for every recorded pipeline when the pipeline cache was written by another device or driver.
After a driver update pipeline-cache.bin is dropped and every recorded pipeline is built from nothing. With the pipeline library the boot waited 20 s and left the rest to be built beside the game. In a player's log of an 8-core integrated Radeon that was 2,141 of 8,122 pipelines, finished 296 s after the start, with the 42,231 prepared variants being built behind them: the game drew fewer than 600 frames in its first 11 minutes, which reads as a freeze. A start that finds such a file now waits for all recorded pipelines, with their count on the boot screen, as a start without the library does; NX2NAVI_PREWARM_FULL=0 keeps the 20 s. The prepared variants still follow in the background. Checked with --probe-prewarm on a cache copy whose header names another device: the line is logged and all 5,775 pipelines are built before the stage ends.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:31:56 -05:00
chrg6
bc256496cb Build task and mesh shader pipelines without the driver's optimiser on AMD's driver for RDNA 1 and 2, unless vk_noopt says otherwise.
That driver loses the device in play ("No fault detected", a timeout) on RX 6600 to RX 6950 XT cards in every version from 0.6.0 to 0.9.2, with the pipeline library on or off and the lazy write-back on or off (#591, #602, #659, #683, #684, #689, #635, #719, #279, and fourteen of twenty sessions in a player's launcher log of an RX 6800). Where the pass marks name a pass the GPU was in, twelve logs of five reporters, its first draw is a mesh pipeline: mesh shader ba4302f1 with fragment shader f64a96dd in nine, the one-draw pass of 1bbdb17e with 6cc3a476 in three. The five logs of 0.9.1 and 0.9.2 stop between the compute dispatch before such a pass and the pass. vk_noopt = mesh was there as a test switch for this suspicion and no reporter had run with it.

It is now what an unset vk_noopt means on that driver branch; every other driver keeps 0, and vk_noopt = 0 gives the optimised pipelines back. The picture is the same; the mesh and task shaders run slower, by how much is not measured, and those pipelines are built once more because the flag is part of the driver's cache key. That the optimiser is the cause is not shown: no AMD card ran this.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:31:56 -05:00
chrg6
07557fe06c Give an empty slot's handle for a texture id out of range instead of ending the game.
nvnDeviceGetTextureHandle ended the game when the texture id was negative or past the table. The game passes -1 when it has no texture to give, as when its render-target heap is full; the handle of the last slot, which no game registers, draws the stand-in like any slot without a texture. Logged up to eight times.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:31:25 -05:00
chrg6
724a9d98ce Size the game's render-target heap for the scene's share of the picture, so a scene at the picture's full size fits it.
The heap the engine's render targets live in is made for a 2560x1440 picture whose scene is three quarters of it at most. It grew only with the picture. With DLAA, or FSR at native size, the scene is the whole picture: at the game's own 1440 lines that is 2560x1440 targets where the game never passes 1920x1080, in a heap that had not grown at all. The render-target manager's search for a free block then takes its out-of-memory path and reads through a null block: a player's game ended there (main+0x3ed00, reading address 0x48) each time a dialogue closed in Castle Town on 0.8.0 with dlss = dlaa (#605), the same function that failed at 5120x2160 before the heap followed the picture.

The heap and the application pool now grow by the scene's pixels over the game's most as well: 238 MiB at 1440 lines with the upscaler setting on, nothing without it, and nothing at 1080 lines, where a full scene is the game's own size. The share is decided at the start, from the settings; an upscaler first switched on in the menu of a run that began without one keeps the scene at three quarters until a restart, and the log says so. Compiled; not played.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:31:25 -05:00
chrg6
e3e4b2d650 Log what keeps the released textures above their size bound.
The bound on the released textures' memory only takes entries that are old enough and that no lookup reaches. The texture pool stamps every GPU-owned entry that a slot's last fill resolved to in each submission, a finalized slot's released entry included, so such an entry never counts as idle: in a session at 5120x2160 the released entries held 9.2 and then 10.1 GiB against a bound of 4,008 MiB, and its overwritten render targets were logged as "last used 1 presents ago". Nothing said which rule held how much. When the released entries stay a quarter over the bound after the drop, one line now splits them: released too lately, sampled lately, shown by a pool slot and not sampled (and of those, the ones another target was drawn over), reached by another lookup, next to go. No rule changes.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:21:43 -05:00
chrg6
e1b1691e28 Make a refused upload submission again once the batches in flight have finished and the kept buffers are given back.
vkQueueSubmit of an upload batch that the driver refused for memory ended the game at once ("upload submit failed (-2)": five logs of three reports, all on integrated GPUs, whose device heap is the machine's memory and fills before any commit check sees it). A refused submit leaves the command buffer and its fence as they were, so the batch is now submitted a second time after the batches in flight are waited for, the objects that waited for them are destroyed and the staging cache, the arena's free chunks, the snapshot pool and the spare blocks go back to the driver, as a refused staging buffer already does. A second refusal ends the run as before. Compiled; the refusal was not reproduced here.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:21:06 -05:00
chrg6
18aa16f2df Skip the menu's draw for a frame when its vertex or index buffer cannot be allocated, instead of copying through a null pointer.
The menu's Vulkan back end went on after a failed buffer allocation: it bound and mapped memory it did not have and copied the frame's vertices to the address the failed mapping left, null. Two reports of 0.8.0 end that way right after the settings menu opens with video memory used up, one on an integrated AMD GPU and one on an RTX 2050: two "got VkResult -2" lines, then an access violation in the C runtime's copy, writing address 0x20, not on a guest thread. The allocation now leaves no buffer and size 0 when it fails, the draw returns before anything is recorded, and the next frame asks for the buffer again. The change is marked in the vendored file and named in THIRD_PARTY_NOTICES.md. Compiled; the failing allocation was not reproduced here.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:19:59 -05:00
chrg6
222a025c52 Give back what a host thread keeps for itself when it ends.
Every guest thread is a host thread, and the game makes and ends them all through play. A thread that slept through plat_wait_ns kept a waitable timer, one that read or wrote a file an event for each, and one that decoded zstd through the host a decoder context of about 100 KiB; nothing closed or freed them when the thread ended, and a detached thread's own record stayed allocated. They are now given back when the thread's function returns: the handles in win32.c thread_main, the decoder context through plat_thread_end_hook, and the record by whichever of the thread and its detach comes last.

tests/thread_end_check.py runs 416 such threads and compares the process's handle count before and after. tests/sync_wake_check.c gets the two stand-ins kernel.c has needed since the service wait log, so that check builds and passes again. The Linux file only gains the hook call and was not compiled here.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:18:25 -05:00
chrg6
823e0cdcd4 Raise the pipeline cache's size limit to 1024 MB, and leave never-drawn list pipelines to their first draw on a machine short of memory.
The limit was 512 MB, and a file over it starts the run empty until a play file replaces it. With the pipeline list of 0.9.1 one run builds more than that: on an RTX 2080 Ti (#722) pipeline-cache.bin held 551 MB and the play file the run then wrote 541 MB, so both were dropped at every start, 7,247 of 8,531 pipelines were built again each time (17 s of the boot) and 541 MB were written in every session. Seven of the players' logs read here have a file over 512 MB, five of them between 517 and 691 MB. At 1024 MB those are read, the play file is folded in once and the loop ends; a start that still finds both files over the limit now says so.

pipe_seed_boot (auto, 1, 0) decides whether the boot builds the pipelines of the shared list that the install has never drawn. Each is a pipeline the driver holds for the whole run: the boot's pipeline stage went from about 1.6 GB of commit on 0.8.0 to 3.9-4.4 GB on NVIDIA cards with the list (1.5-1.8 GB on AMD's driver), and 0.3-0.6 GiB of video memory. auto, the default, builds them as before except on a machine that starts with less commit free than the game takes in play, the start's own memory warning: there they are made at their first draw from the pipeline cache and recorded as the install's own. Mesh pipelines are built at boot either way. Compiled; not run in the game.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:13:40 -05:00
chrg6
afe44e0e54 Let the device address table read a new VRAM mirror only once the mirror's first copy has run.
A mirror is made while a submission is recorded, and its first copy of the pool is in that submission's prologue. The table's pages moved to the mirror at once, and the table is memory the shaders read as they run: every part already handed to the queue read the mirror through it before the copy had run, memory nothing had written yet and, on a card over its budget, a block the driver had not made resident for those parts. A GTX 1650 4 GB lost its device that way at area loads in every session of 0.7.1 (#507): six reports of a GPU read fault at byte 0 of a mirror made in the faulting submission, each the first byte of a new 2 MiB-aligned block, with mirrors being made over the budget, and none with the table kept empty. The mirror code is the same in 0.9.2.

Now the pages keep reading the imported buffer and take the mirror as their twin, so what shaders write through the table goes to both and what they read is guest memory, as before the mirror existed. Once the GPU has finished the submission that holds the copy, the pages move to the mirror with the imported buffer as the twin. gpa_late = 0 moves them at once, as before. tests/gpa_check.py covers the new table step; the mirror side was compiled and run through --probe-vkmem, not played.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 15:09:42 -05:00
Vasilis Manetas
eaf10677fa Import guest memory as dma-bufs through udmabuf on Linux instead of keeping shadow copies of it.
On Linux every CPU-visible pool had a host-visible shadow copy, because amdgpu treats a host-pointer import as a userptr object that it looks up again at every submission and invalidates whenever the write watch or a freeze re-protects its pages. Now the watched reservations are shared mappings of one memfd, /dev/udmabuf makes dma-bufs of page ranges of it, and the device imports them (VK_KHR_external_memory_fd, VK_EXT_external_memory_dma_buf). A dma-buf pins its pages once and is not tied to the process's page tables, so protection changes leave the GPU's mapping alone, and nothing is copied.

RADV puts every imported memory object into every submission, so pools share their imports: guest addresses are cut into 64 MiB spans (NX2NAVI_DMABUF_MB), and each pool is a buffer bound into its span's memory at its offset. A new span is cut back at the spans already there, so spans overlap only where a pool itself lies across an older span's edge, since amdgpu counts every span's pages in its GTT use. A decommit punches the pages out of the memfd, or zeroes them in place where a dma-buf pins them, and the pages a freed span pinned that nothing committed go back to the system.

It is on by default where /dev/udmabuf opens, udmabuf's size_limit_mb is 1024 or more, the device has both extensions and NVIDIA's kernel module is not loaded. A start-up probe imports one page as the pools will be; any refusal keeps the shadow copies, and the log says why. NX2NAVI_DMABUF=0 keeps the copies. tests/platform_dmabuf_test.c checks the shared mapping without a GPU, and docs/building.md ("Memory") describes the imports and what they need.

RX 6800 XT (RADV, Mesa 26.2.6), one build, interleaved pairs with NX2NAVI_DMABUF=1 and 0: Kokiri Forest 60.7 against 40.3 fps (3 pairs, +51%, standard error 0.4), 1% low 50.9 against 34.0, frames over 25 ms 5 against 768; Hyrule Field 55.7 against 42.4 fps (2 pairs, +31%, standard error 0.4). No pool was left without a span, no span was refused, and --probe-vkmem passes with and without the imports.
2026-11-12 15:00:27 -05:00
Vasilis Manetas
677eb5c242 Keep an area load's texture staging inside the staging pool's blocks made at boot, instead of making about 1.1 GiB of new host memory in the load frame.
The staging pool keeps 4 blocks of 64 MiB from its first use on a discrete GPU (2 on an integrated one, as before). A staging buffer or arena chunk that does not fit them takes the room of the kept staging buffers and idle arena chunks, then of the oldest upload batches in flight, which it waits for; an arena chunk a waited-for batch gives back is taken instead of a new one. Only when that makes no room (the open batch's own staging, staging buffers their callers hold, the blocks' fragmentation) does the pool grow by a block, and the log says so. The outermost vk_upload_end submits the open batch once it holds 64 MiB of staging, besides the NX2NAVI_UPLOAD_BATCH_MB limit; nothing is submitted or waited for once the device is lost. Entering Hyrule Field made 221 staging buffers (1,131 MiB) in one frame; on Linux each new 64 MiB block is a kernel allocation that tries 2-4 MiB pages first, and with the free 2 MiB blocks used up the kernel compacted memory inside those allocations: 30-360 ms each, and the load frame took up to 4.7 s instead of about 0.3 s (RX 6800 XT, amdgpu, 32 GB).
2026-11-12 14:59:47 -05:00
chrg6
cacf458b08 Say in a lost device's report what the GPU was about to do when it stopped between two marks, and add draw marks.
A pass mark names the pass or dispatch the GPU had started and the last it finished. When the two are the same mark the GPU stopped in between, and the report named nothing: in a player's log of 0.9.2 (RX 6700 XT) both were one compute dispatch, with 51 later marks made and none started. The report now adds the mark that never started and the kinds of commands the replay recorded between the two (copies, clears, the upscaler, write-backs), which the gap notes already name for the GPU pass timing and now keep for every mark.

draw_crumbs = 1 (off by default, AMD buffer marks only) also writes a mark wherever the shaders or the pipeline state change inside a pass. The device losses reported on RX 6600 and RX 6800 class cards end inside one scene pass of 30 to 100 draws and the draw that hangs is not known; with the setting on, the report lists the shaders of the draws the GPU had not finished.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 14:57:21 -05:00
chrg6
d48a3d59f8 End a helper run with its own exit code when a library faults while the process is ending.
The shader preparation modes and the probes return from main, and the C runtime's exit then calls the detach of every library in the process, those other programs put there included. On a player's PC a post-processing injector's Vulkan layer read a null pointer there at the end of every --warm-shaders and --prepare-shaders, four runs over three versions: each printed "variants built into the driver cache", then a FATAL, and ended with code 71, so the launcher logged a finished preparation as one that did not finish. The crash handler now ends such a process with the code its mode returned and one "exit:" line; a fault while the mode still runs, or in the executable's own code on the way out, is reported as before. A crash inside that injector's library also gets the note that the library is not part of the game.

--probe-crash-exit makes the fault from an atexit handler, and tests/exit_fault_check.py checks both cases.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 14:57:13 -05:00
chrg6
26f3dc47ec Show the low-end GPU mode's settings under its switch in the Game tab, and read them only while the mode is on.
The settings had a tab of their own; they are rows of the Game tab now, listed under the switch once the mode is on after a restart, and the tab is gone. With the mode off every one of them reads as unset, whatever nx2navi.ini or the environment says, so nothing of the mode acts unless a player turned it on; the build reads guest_opt the same way. The notes describe that, and take the white edges in towns and the dropped textures from the known issues to the fixes.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 13:59:43 -05:00
chrg6
857d22fe4f Apply texture and sampler registrations between submissions, in the order the game made them.
A registration the game made while a submission was being replayed changed the slot table under that submission,
whose texture pool set is written as the replay goes and read when it runs. Such a registration now waits and takes
effect once the submission has gone to the queue; with the replay on its own thread it goes into the replay queue
ahead of the next item. NX2NAVI_REG_ORDER=0 restores the old behaviour. In a town session the white kerb, stair and
wall-top edges of single frames were gone with it; the log counts the registrations that waited.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 13:58:36 -05:00
chrg6
16b0b46155 Keep a released texture that shaders still sample out of the idle drops.
A finalized slot shows its released entry while that holds the texture, and the pool's check passes released entries by, so no lookup reached such an entry and it counted as idle while every frame sampled it. Past the released entries' size cap or count it was dropped, and the slot showed the stand-in until the game registered it again. An entry sampled within the stale time is not idle now, and the count sweep leaves one that was used.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 13:12:58 -05:00
chrg6
e3cf0710c5 Record a draw's own texture binding that got the stand-in, and a sampled pool slot that is not registered.
In towns, kerb, stair and wall-top edges are drawn untextured for one frame when the game registers a round of streamed textures, and no pool entry of those textures goes to the stand-in in such a frame. A binding written into a draw's own set, and a slot that is unregistered when its marks are read, left no event until now.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 13:08:16 -05:00
chrg6
d2fbaba1c3 Put the low-end GPU mode in the settings menu's Game tab, marked experimental.
The row sets low_end_gpu, read at start, and says what the mode turns on and that it was measured on one card; its text is there in the menu's four languages. The key is decided again once the settings file is read.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 13:06:02 -05:00
chrg6
2324a74602 Turn the settings for older graphics cards on with one key, low_end_gpu = 1, and leave them off without it.
Mesh shaders as compute, the texture that gets memory of its own, staged mirror uploads and the shorter wait for the GPU, widened write-watch queries, the video memory limits and reserve, early upload batches, the asynchronous replay, the hot-pool rule and the recording thread on AMD cards without mesh shaders were measured on a Radeon RX 580 and run on no other card. Each keeps what 0.9.1 did unless the key or its own setting says otherwise; the key gives the defaults the way the device profile does. The notes describe them as that option, hold main's two new entries in one Changed section, and list the white edges in towns as a known issue.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 13:00:13 -05:00
chrg6
052c60dc0d Merge remote-tracking branch 'nx2navi/main' into release-0.9.2 2026-11-12 12:56:51 -05:00
epicmeme546
385734dcc0 Sharpen FSR 3.1 more below a render scale of three quarters: its own sharpening towards the maximum, and a second, lighter RCAS pass over its picture.
FSR 3.1 was soft at Performance and Ultra Performance, most of all in motion. On a village save, while the camera turned at a steady rate (present screenshots every 0.4 s, the mean absolute Laplacian of the scene's luminance): FSR 3.1 Native AA 6.7, FSR 4 Performance 5.9 and Ultra Performance 5.4, FSR 3.1 Performance 4.0 and Ultra Performance 3.3. The library's own input checks found nothing wrong, and its tuning constants (velocity factor, accumulation per frame, shading change scale, behind the test switch NX2NAVI_FFX_KEYS) gave nothing or took detail away. Its own sharpening at 100 instead of the default 80 gave 5.0 and 4.1.

Now the share of extra sharpening grows from none at three quarters of the output's width to all of it at half (vk_dlaa_fsr3_boost). FSR 3.1's own sharpness goes that share of the way to its maximum, and its picture gets a second RCAS pass (shaders/dlaa_sharpen.comp, AMD's FSR 1 header with the linear light taken through c / (1 + max c) and back), before the HUD is drawn. Both follow the sharpness slider; 0 turns them off, and NX2NAVI_FSR3_SHARPEN=0 leaves the second pass out. At RCAS's full strength the pass took Performance to 8.2, past Native AA, with grain on every surface, so it runs 1.5 stops weaker: Performance 5.7, Ultra Performance 4.7, Quality 5.2. 63.4 fps against 65.1 without the pass. The HDR picture is clean. FSR 4 is untouched.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:50:43 -05:00
epicmeme546
7581bb887d Let the free camera go anywhere, and make ground avoidance a switch that is off unless asked for.
The camera kept its height (only ZR and ZL changed it) and rose whenever the closest surface in the lower part of the view came within about three metres, so flying up to a character, a rock or a wall sent it upwards. It now moves where it looks and nothing stops it, the ground included.

Ground avoidance is a checkbox in the debug menu's free camera section. With it on, the nine view distances read back from the lower view become points of the scene (from the view's angles and the camera's pitch), and only a column whose lowest surface rises by less than about 30 degrees counts as ground: the camera goes no lower than a metre above it. Steeper things, objects and walls, are left alone. The read-back runs only while the switch is on.

Tested on a village save, 4 m/s forward for 5 s and then ZL for 4 s: with avoidance off the camera ended 17.4 m lower, through the floor; with it on it ended 0.7 m lower and held there. Flown at Link with the first version of the ground test, its height did not change.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:50:43 -05:00
chrg6
f2b6f39c5f Give a natively run network program this frame's input picture: the RGB9E5 words the post pass stored in the scene target are copied into the RGB9E5 texture the network samples.
The game's scene upscaler samples the post pass's picture through an RGB9E5 texture at the scene target's address. On the console both are one memory; here they are two images, and the RGB9E5 one kept what it held when it was made. With NX2NAVI_NATIVE_TENSOR=1 the network therefore read an old picture or the target's R11G11B10F words as shared-exponent texels, the source of the infinities and noise it wrote. The stand-in already read the target's image for this reason. A network program's output image without storage use is now logged.

Only dispatches of tensor-core programs take the new path, and those run only with NX2NAVI_NATIVE_TENSOR=1, which stays 0 by default. The file compiles without warnings; no game run has checked the picture.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:39:54 -05:00
FIocker
5e71cee08c Record the depth bias again at each pass, as the viewport and scissor are, so a pass into a scaled shadow map never keeps an unscaled slope term.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 12:39:53 -05:00
FIocker
58efccd883 Scale the slope-scaled depth bias with the shadow map scale, so the larger maps keep the bias the game asked for.
The slope term is a bias per unit of depth slope across one pixel of the target. A shadow map made k times larger on
each side (shadow_scale 2 to 4) has 1/k of the slope across each texel, so the game's factor alone gave k times less
bias on sloped surfaces than at the game's own size, which is what shows as acne. vk_exec_draw.c slope_bias
multiplies the factor by the pass's scale for both the direct draws (record_fixed) and the queued mesh draws; the
constant term counts the depth format's smallest steps and keeps its value. Nothing changes at shadow_scale 1.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 12:39:53 -05:00
chrg6
81e7da91aa Bring --probe-vkmem's texture cases up to the rules the runtime has now.
The probe ended in FAIL on an unchanged tree, so none of its cases after the texture slots ran. Three of its expectations were older than the rule they test; the runtime does what its comments say in each.

A finalized slot of a CPU-written texture shows its own image while that image still holds the texture (stale_kept). The probe expected such a slot to resolve to nothing after its texture's release although another texture object kept the image current. It now expects the image with tex_stale_kept on and nothing with it off, and a case of its own covers the rule: after the finalize the slot shows the same image, bytes written to the storage afterwards do not reach it and nothing is uploaded; registered again, in a later submission, it holds the storage's bytes of then.

A slot follows another descriptor's live texture at its storage only when that texture has the slot's size (follow_size_ok). The probe put a 32x32 texture under a 64x64 slot and expected the slot to follow it. It now checks both: a 64x64 texture of another format is followed, and under a 32x32 one the slot gets an image of its own descriptor.

The texture detail cases and the case of an ASTC decode without scratch memory returned without a check whenever the image was neither RGBA8 nor BC3, which was meant for a device that samples ASTC itself. Since opaque ASTC textures are made as BC1 that held for every image of those cases: the probe printed their line and checked nothing. They read BC1 images now.

The probe passes with the default settings (188 GPU checks; 164 when it last passed), with tex_stale_kept = 0, tex_same_big = 1, pool_presnap_runs = 0, shadow storage, ASTC images as BC3 and as RGBA8, and texture detail reduced and low.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:33:06 -05:00
chrg6
0533b3028e Keep the asynchronous replay, the hot-pool rule and the -O2 game code off unless a player turns them on.
The three change what every card does: the asynchronous replay can show a dark or flickering frame now and then, the hot-pool rule moves pools into system memory on every discrete card, and -O2 makes each rebuild of the game several times longer. Their rows stay in the settings tab. The USB joystick mapping is described by the device's own name, and the tab's introduction says what its settings are for.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:17:55 -05:00
chrg6
a464ffafbc Merge Tobim6's work for older graphics cards: mesh shaders emulated with compute, video memory kept within the heaps' budgets, staged mirror uploads and the submit wait that goes with them.
Conflicts: the settings key lists take both sides; frame generation keeps the camera-cut handling and logs its counters every 600 frames with the cuts among them; the pass-time mark moves into draw_viewport; mo_desc keeps the pose depth and notes the reader.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:14:58 -05:00
chrg6
88b74fc8f2 Merge branch 'claude/pr-intake-6-43af31' into release-0.9.2 2026-11-12 12:10:44 -05:00
FIocker
ac4af77973 Tell FSR 3.1 frame generation of camera cuts, so a generated frame is never blended from two unrelated pictures.
vk_dlaa_fg_inputs hands out DLSS's new history with the frame's depth and motion, and FSR 4's and NVIDIA's frame
generation pass it on, but vk_fg_frame dropped it: FSR 3.1 takes a cut only in the generation callback, which its
swapchain calls later on its own thread. vk_fg_frame now notes each frame id's cut in a small ring
(runtime/vk_fg_reset.h), and generate sets the callback's reset from it. A frame after one that was not prepared (the
menu open, no inputs) counts as a cut too, and so does an id the ring no longer holds.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 12:09:37 -05:00
Vasilis Manetas
bfa486c2aa Hand every queue submission to the driver's own submit thread on Linux, so RADV's vkQueueSubmit2 leaves the replaying thread.
Mesa's Vulkan drivers keep timeline semaphores as kernel sync objects, and their queues submit on the calling thread unless a submission waits on a value not submitted yet. The replay hands a part of the frame to the queue each time it splits one off (vk_exec.c maybe_split), so every such vkQueueSubmit2, with the command stream's checks and the kernel's ioctl, ran inside the replay, the thread that limits the frame rate on an RX 6800 XT under RADV.

MESA_VK_ENABLE_SUBMIT_THREAD=1 sends every submission to a thread of the queue's own. vk_device.c mesa_submit_thread sets it before any thread can load the driver, unless the environment sets it already. submit_thread in nx2navi.ini (NX2NAVI_SUBMIT_THREAD) is 1 by default on Linux and 0 on Windows; 0 leaves the driver's own choice. On a Mesa driver the log says which one applies.

Kokiri Forest, 0.9.1, RX 6800 XT, Mesa 26.2.6, 3 pairs of 45 s runs of one build with the setting on and off: the replay took 13.9 ms a frame against 15.0, lower in every pair; frames over 25 ms fell from 910 to 406; the frame rate went from 40.0 to 40.8 fps, within the runs' noise.
2026-11-12 12:09:37 -05:00
chrg6
0985e80466 Leave Link's own motion vectors off until they have been played, and take their settings through nx2navi.ini.
The pose motion vectors for Link and his attached equipment were on by default for game 1.1.1 and had not been run in
the game on this version. selective_motion = 1 turns them on. Its settings are listed among the known ones, so
nx2navi.ini takes them without an unknown key line, and the rotation threshold is read through the settings instead
of from the environment alone. nx2navi.ini.example describes the switch.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:09:37 -05:00
automata.
b29cbbe6a6 Add shared pose motion vectors for Link and attached equipment 2026-11-12 12:09:37 -05:00
DeepDaddyTTV
38afed16d5 Record the draws rejected during preparation in a captured frame.
A draw can be left out before it opens a render pass, and the pass list of a frame capture then showed nothing of it.
passes.txt now ends with the rejected draws of the captured frame by mesh or vertex draw and reason, each with the
first program it met, and says so when there were none. A draw left out by NX2NAVI_SKIP_PROG or NX2NAVI_SKIP_SHADERS
is counted under a reason of its own. The frame to capture and its files are set together under the recorder's lock.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:09:37 -05:00
Leonardo Chappuis
d991c04284 Fix a deadlock between the replay workers in vk_exec_pass_end.
vk_exec_pass_end took tfr_lock and then, in vk_texture_gpu_written, the texture lock. A chunk recorded by another
thread can hold the texture lock in the pool sync and end its pass there (a lookup's alias copy through
vk_exec_inline_begin), and then wait for tfr_lock. The two chunks waited for each other and the other chunks spun in
their turns, so loading after Continue never finished with replay_workers = 3. The targets are now marked written
before tfr_lock is taken, which guards only the target history.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 12:09:37 -05:00
chrg6
a654d079e1 Keep a released texture's image while a registered slot still names it.
The game finalizes a texture object while another one it initialized at the same storage with the same parameters stays registered in the pool, and the entry the two share is released all the same. Two frames later its image went to the next texture of its parameters and the entry was dropped; the slot then had nothing to show, because the pool makes no entry for storage the CPU cannot read, and kept its stand-in until the game drew into the texture again. Ocarina's 384x384 shadow maps are such textures: a player's recording has four slots of them sampled as stand-ins in the frame after three of their images had gone to other shadow maps, and a house front without its shadow in that frame. The reuse now passes over a released entry that a slot which is not finalized names; tex_reuse_named = 1 restores the old choice.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:09:37 -05:00
chrg6
b93219c394 Make a slot's new texture at once when the slot is showing another one.
Ocarina streams its textures by detail level: while the player moves it registers about 32 pool slots with their next level every four or five frames. A pool refresh made 16 of them and wrote the stand-in into the set for the rest until its next round; the set is read as it is when its submission goes to the queue, so an object whose slot was among the rest could be drawn grey for a frame. A player's recording shows 1,525 such slots in 600 frames, each with its texture made later in the same frame. A slot whose entry in the set is a texture's view when it is looked up now has its texture made by that lookup, beside the count budget and the frame's creation time, 64 of them a present; pool_shown_first = 0 leaves them to the budget. The flight recorder also notes a slot that still had no picture when its submission ended.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:09:37 -05:00
chrg6
76c040d35b Record a pool slot losing its picture in the flight recorder.
When the pool's entry of a slot goes from a texture's view to the stand-in, events.tsv gets a line with the slot and what its own texture was at that moment: finalized, released, superseded, overwritten, uploaded, its storage readable, held out. The set is written with update-after-bind, so every draw of the submission that samples the slot reads the stand-in from then on; the check at the marks' read-back comes a frame later and finds the slot resolved again.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:09:37 -05:00
chrg6
67a2146948 Record what an upload read, slot registrations and sampled stand-ins in the flight recorder.
Three more texture events for a dump's events.tsv: after each upload a hash of 64 probes of the storage and how many of them were all zero, so a second upload a frame later tells new bytes from the same ones and a first upload of unfilled storage shows; a slot registered with a CPU-written texture, with what the slot named before; and a slot a shader sampled that has no image of its own and follows none, so its stand-in was on screen. They are for telling why a texture shows wrong for one frame while the game streams textures in.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:09:37 -05:00
chrg6
57a15e288d Count only the making of textures against a frame's texture creation time.
The limit on what a frame's pool refreshes may spend creating textures was charged with the whole refresh, including its check of every entry the pools hold. In Kakariko that check takes 0.8 ms a submission, six submissions a frame, so most of the 6 ms was gone before a texture was made and the frame's later refreshes made none: 87 refreshes in 600 frames, with 1,090 slots left between them. The game moves about thirty textures to new storage every four or five frames there, and the ones a refresh left were on screen as their grey stand-ins for a frame; a player's recording shows three lamp posts turn grey for one frame that way. The limit now counts the time the lookups spend making entries and uploading them the first time.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:09:36 -05:00
chrg6
bd7d140a54 Draw lines as wide as the game sets them.
The game's line width was dropped at the call and every line drawn 1 pixel wide; the game draws two lines a frame at 4.5 pixels into a 144x144 mask. The width is now recorded with the draw state and set as dynamic state for draws of the line topology class, held to the device's range. Only pipelines of that class list the new dynamic state, so the others keep their entries in the pipeline caches and the driver's. line_width = 0 draws every line 1 pixel wide as before.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-12 12:09:36 -05:00
FIocker
f0d0fae13b Tell FSR 3.1 frame generation of camera cuts, so a generated frame is never blended from two unrelated pictures.
vk_dlaa_fg_inputs hands out DLSS's new history with the frame's depth and motion, and FSR 4's and NVIDIA's frame
generation pass it on, but vk_fg_frame dropped it: FSR 3.1 takes a cut only in the generation callback, which its
swapchain calls later on its own thread. vk_fg_frame now notes each frame id's cut in a small ring
(runtime/vk_fg_reset.h), and generate sets the callback's reset from it. A frame after one that was not prepared (the
menu open, no inputs) counts as a cut too, and so does an id the ring no longer holds.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 12:07:19 -05:00