Commit Graph

19 Commits

Author SHA1 Message Date
chrg6
06f4aeb694 Time what a submission spends on the GPU outside the game's passes and dispatches, with gpu_passes = 1.
A session with per-pass GPU timestamps showed the game's own render passes and compute dispatches at 1.6 to 2.3 ms of GPU time a frame while the perf line's GPU time was 7.7 to 13.5 ms: most of the frame's GPU time is in commands the replay records between the marks, and nothing said which.

The timed frames now also get a timestamp at the start and the end of each part of a submission, and every stretch between two marks (or a mark and the part's edge) is a gap. The replay names the kind of commands it records in a gap: buffer and texture copies, fills and clears, the temporal upscaler, write-back merges and copies, depth twin copies, the texture pool marks, the mesh draw counts and block-compressed target copies, each with a size where it has one. The report adds the gaps' time per frame summed by the kinds noted in them (a gap nobody noted holds barriers and state only), the parts' own time for comparison with the perf line, and each kind's count and size per frame. Nothing runs outside the timed frames.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
2026-11-09 18:49:12 -05:00
chrg6
8e656b85a4 Hold less video memory in empty space: smaller VMA blocks, spare blocks only where textures grow, a smaller ASTC scratch, and DLSS features freed once unused, with a VRAM breakdown in the perf log.
VMA's blocks on large heaps are 64 MiB instead of 256 (vma_block_mb): one session held 2.1 GiB free inside the
device blocks beside 8.2 GiB of allocations after a few area changes, and the driver counts a block whole until it
empties. The headroom thread keeps its spare block only for the device-local types textures and mirrors grow in; the
spare blocks it made for the host types and the PCIe BAR type at boot stayed empty for the whole run. The ASTC
decoder's scratch is 32 MiB, so it stays inside a block, and an area load that makes one per upload batch holds half
as much at once. A DLSS feature no frame used for 3600 evaluations is freed with its work images, and a model change
now frees all five sizes.

The perf log gets a "vram by use" line (textures, released textures, mirrors, physical pools, ASTC scratch, headroom
spare, free space inside the blocks, memory outside VMA and DLSS's own share), the free ranges per memory type, the
addresses whose duplicate texture entries hold the most, and a line when memory outside VMA moves by 256 MiB.
astc_bc_probe in nx2navi.ini turns on the ASTC opacity probe.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit 99cfc2fa14ae208b7b23e057ff9b87986cd6a5e0)
2026-11-09 14:56:17 -05:00
chrg6
dfc5b90b59 Fix what the 0.6.0 review found, and ship DLSS Frame Generation in the launcher release.
The commit census of the perf trace walked the address space on the presenting thread outside mem_census, which
brought back the hitch mem_census was added to remove; it now runs only with mem_census = 1. DLSS's output pass at
the mode's share fell back to no pass at all when the scene was not at that share (the first presents, a dynamic
resolution step): after such a frame the first pass wider than the scene is taken again for 120 frames. The launcher's
nx2navi.ini editor no longer rewrites a file it cannot read, keeps the byte order mark and bytes that are not UTF-8,
reads an empty line as the game does, and leaves no temporary file behind. The build kit carries the two modules
build_game.py imports on Linux, the Linux steps list the right packages and Python, and stale comments and notes are
corrected (Discord on by default, two prebuild threads, Enhanced HDR gone).

DLSS Frame Generation needs nx2navi_dlssg.dll, NVIDIA Streamline's runtime and nvngx_dlssg.dll beside the game, which
no release carried, so the menu offered it and the game turned it off. tools/make_launcher.py --dlssg and --streamline
put them with their licences and a notice in the release's bin/dlssg/, and tools/make_game.py place_dlssg copies the
whole set beside the game. THIRD_PARTY_NOTICES.md has the row; the notes and README say RTX 40 and 50.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-09 09:10:36 -05:00
spider
7d35eada1d Upscale to the picture's size at Ultra Performance, not to the first pass wider than the scene.
At Ultra Performance (an 854x480 scene) the Ocarina frame shape took the first single-target pass wider than the scene
as the output pass: a 960x540 post pass comes before the picture's, so DLSS upscaled to 960x540 and that was stretched
to the output. The output pass must now be at least the scene over the mode's share (widescreen.c dlss_share, exposed
as wide_dlss_share), which skips it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-09 08:16:53 -05:00
epicmeme546
778c55777d Say a restart is needed only when one is, and drop the line under Preparing shaders.
The Upscaling row no longer carries "DLSS on or off: after restart". It shows "after restart" only while the upscaler picked cannot start in this run. Frame generation says so only for a change to or from NVIDIA's, since AMD's applies at once, and the HDR row no longer says it twice.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-09 04:29:01 -05:00
chrg6
1f7e9c975b Speed up the replay and fix what the perf pass found, and add a frame capture key for chasing rendering faults.
Performance: draws reuse the bindings that did not change and record only the dynamic states that differ from what
the command buffer holds; small uploads come from a staging arena; finished submissions are held briefly and handed to
the queue in batches; the recorder thread can keep a CPU core of its own; ASTC block-linear textures are deswizzled on
the GPU; dynamic resolution steps the upscaler mode to hold a frame-time target. Each has a setting in nx2navi.ini
(bind_reuse, dyn_delta, stg_arena, submit_batch, recorder_core, recorder_prio, astc_gpu_deswizzle, dyn_res) and the
staging cache now frees what it holds at shutdown.

Capture: NX2NAVI_DUMP=key saves the frame after each press of Insert, with mesh draws listed by task, mesh and fragment
shader, depth test and face culling, and mesh_prims.txt giving each mesh draw's primitive, clipping and fragment counts.
All of it is off unless NX2NAVI_DUMP is set.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-09 03:34:46 -05:00
chrg6
e32fd15952 Take Ocarina's DLSS camera from the bytes the scene draws read, and log why a frame has none.
In Kokiri Forest dialogue (Saria, dlss = performance) about 1 frame in 20 had no camera, and DLSS started a new history twice for each: 24 to 26 camera resets per 300 frames. The camera block was looked up in the uniform copy of the submission DLSS ran in, which on those frames had none: the frame's work had been held past its present by the event order and ran in the present thread's empty submission. vk_dlaa_draw now keeps the camera rows as the first scene draw read them (vk_usnap_read), and mv_pass uses those. A frame with no scene-context draw at all keeps DLSS's history with zero motion, for up to 6 frames in a row.

Measured at the Saria spot: camera resets went from 24 to 26 per 300 frames to none in 3600 frames (the first frame only). The whole-picture flashes stayed: they come from the presents that go out with that work held, handled separately.

The log now explains frames without a camera, frames without the DLAA point (with the held submissions at that present), light target changes and a second scene after DLSS ran, plus a line every 300 DLSS frames with resets. dlaa_cam_keep = 0 (NX2NAVI_DLAA_CAM_KEEP=0) restores the old camera handling.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-09 03:34:46 -05:00
epicmeme546
754ded99e0 Raise the frame rate in Kokiri Forest and apply more settings during a run.
The foliage's sun shadows took most of the GPU time: one mesh draw shaded about 1,000 million fragments in the far cascade. The far cascade now shades them at a 2x2 rate, and the two far cascades drop one side of the double-sided cards. At the forest entrance the frame rate went from 18 to about 43 fps on an RX 7900 XT.

FSR 3.1, its frame generation and the console mode now take effect without a restart. The full performance overlay lists the settings shown in the options menu.

NX2NAVI_GPU_PASSES times each pass and counts its fragments, and NX2NAVI_SET_AT changes settings at a given frame for test runs. Both are off by default.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-09 02:40:51 -05:00
chrg6
3731155efc Run DLAA with native HDR too.
With hdr = native the game's output pass writes a 10-bit PQ target (A2B10G10R10), not the 8-bit sRGB one vk_dlaa_pass looked for, so DLAA never ran and frame generation had no depth or motion vectors for most frames. The native HDR picture's format is now named once (vk_hdr_native_picture) and taken by both DLAA and the HDR present. From a tester's fixes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-08 15:51:48 -05:00
chrg6
d9fb0fe1ba Give characters in Ocarina's frame their own motion for DLSS and frame generation, so a running Link no longer trails.
Ocarina's frame has neither the per-object motion pass nor the class buffer the characters' mask reads, so DLSS and
FSR 3.1 frame generation moved every pixel with the camera: the character the camera follows dragged a faint copy of
itself through DLSS's history and doubled in generated frames. The motion pass (dlaa_mv.comp, vk_dlaa.c mv_pass) now
keeps each pixel's view distance for the next frame and checks the camera's vector against it. Where last frame's
depth at the place the camera's motion points to disagrees, the surface moved on its own; when last frame's depth at
the same pixel agrees, it kept its place on screen and its vector is zero. Such pixels also get the characters' bias
mask (dlss_char_mask), so DLSS takes this frame's colour where its history holds something else. The check takes the
best of 3x3 pixels, so silhouettes reprojected a pixel off are not taken for motion, and leaves the screen border to
DLSS. NX2NAVI_DLAA_ZFIT=0 turns it off; NX2NAVI_DLAA_MASKDEBUG=1 shows the mask.

Frame dumps list a border sampler's colour.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-08 04:10:35 -05:00
chrg6
501d5bd084 Release pass: fix crashes, freezes and races, cut VRAM use, and translate more shaders.
Crashes: the render-target heap (HeapDef DynamicTextureAllocatorMem and its parent) grows with the widescreen
picture, so the pause menu no longer reads a null block at 5120x2160; texture uploads keep the game's memory mapped
while they copy it, so UnmapPhysicalMemory no longer pulls it away mid-copy; DLSS Frame Generation no longer crashes
in Streamline's set-up (physical devices and the window surface go through its proxies, its swapchain images are
counted first, resources are tagged per frame); a stale core-wait target is cleared before its thread is freed.

Freezes and races: a resize during a swapchain rebuild is kept, a failed frame generation swapchain is retried
instead of replaced by a plain one, three acquire timeouts in a row rebuild the swapchain, device waits take the
queue lock, a translator child that stops answering times out, an in-flight compute pipeline builds at normal
priority, a save Commit started before the window closes finishes and none starts after, a settings change just
before closing is saved, and a movie sample taken before a flush is dropped.

HDR: Enhanced HDR extends the game's own tonemap so highlights go above the picture's white (the hue slider,
which no longer applies, is gone from the menu); frame generation gets a HUD-less picture, the HDR flag and the HDR10
metadata under HDR too.

Memory: ASTC textures are encoded to BC3 unless the PC has 20 GiB of VRAM and 48 GiB of RAM, as Windows charges video
memory to the commit charge (a long session held 17 GiB of RGBA8 textures and 45.7 GiB of commit on a 32 GiB PC).

Shaders: the translator lowers spill-mode programs with more than eight uniform register groups in a wider spill
layout, and local loads and stores with a uniform-register address (36029 of the game's 36031 variants translate,
20 failed before).

The launcher release ships the shader tools and AMD's FidelityFX library (make_launcher.py --ffx); windres there gets
the toolchain on PATH.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 16:02:30 -05:00
chrg6
b9a042969e Keep the HUD out of frame generation, show the frame generation rate, and start DLSS Frame Generation.
Frame generation now gets the picture without the HUD each frame: the upscaler's output before the HUD is drawn over
it, encoded to sRGB as the game's last pass encodes its picture and put through the same colour, scaling and borders
as the presented frame, so it tells the HUD apart and leaves it unwarped in the generated frames
(NX2NAVI_FG_HUDLESS_SHOW=1 shows that picture, a test). The performance overlay gains an FG row and the fps line the
display's rate, generated frames included. The device enables VK_KHR_get_memory_requirements2 whenever AMD's frame
generation is on, which crashed at start with DLSS upscaling.

DLSS Frame Generation through NVIDIA Streamline (vk_dlssg.c, and nx2navi_dlssg.dll from runtime/dlssg, built with
MSVC by tools/build_dlssg.py against the SDK kept outside the repository) creates the instance and device through
Streamline's proxies and calls its swapchain hooks; Streamline still fails in its plugin start-up here, so the menu
lists it only while it is set.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 13:31:07 -05:00
chrg6
7c11ff6816 Add AMD FSR 3.1 frame generation: a generated frame between every two the game draws.
With Frame generation set to AMD FSR 3.1 (after a restart), AMD's frame generation swapchain takes the place of the
present path's own: the device creates two more graphics queues for its presents and image acquires, the present
path acquires, gets images and presents through its replacement functions, and every frame configures generation
and records its preparation from the depth and camera motion vectors the DLSS or FSR 3.1 upscaling pass made (one of
them must be on). Generation pauses while the menu is open. On an RTX 5090 at a 60 fps game, about one frame was
generated for every frame drawn in play.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 13:31:06 -05:00
chrg6
ca5e5eb5a4 Add AMD FSR 3.1 upscaling, and run DLSS's smaller modes on a 16:9 picture too.
The Upscaling choice gains FSR 3.1 Native AA, Quality, Balanced, Performance and Ultra Performance. FSR 3.1 runs
through AMD's signed amd_fidelityfx_vk.dll (loaded at run time; the build kit ships it) in DLSS's place, on the same
scene light, depth and camera motion vectors and at the same render scales. Its Vulkan back end needs
VK_KHR_get_memory_requirements2, which the device now enables when FSR 3.1 is chosen.

A half-size effect pass with three targets and a half-size depth was taken for a G-buffer, which switched every later
frame to the deferred frame shape: there the output pass of a smaller scene was never found, so DLSS Quality,
Balanced, Performance and Ultra Performance did nothing on a 16:9 picture. Such a pass smaller than the scene is now
left alone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 13:31:06 -05:00
chrg6
ee6fbaf0bb Zero shader locals with DLSS and DLAA on too, which removes the speckled floor tiles there.
With DLAA on, function-local variables stayed undefined, and NVIDIA's driver filled the game's clustered-light reads
with other warp lanes' values: floors around Link lit in 4x8-pixel tiles. The setting was also part of every shader
cache key, so shaders prepared with DLSS off missed with it on. Zeroing is now on for every setting; DLSS still finds
the game's FXAA by its zeroed hash and keeps its camera motion. A log line now says when DLSS replaces the game's
scaling draw.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 13:31:06 -05:00
chrg6
43a8096f37 Give DLSS Ocarina's camera, so it keeps its history along the camera's motion.
Ocarina's opaque scene draws bind a 0x3500-byte scene context at c3, and their vertex shaders take the world position
through its rows 0-2 (the view) and then rows 7-10 (the projection: an infinite-far projection whose z row is its w row
less twice the 0.1 near plane). The camera motion pass now uses that block's view-projection, from the block most of
the frame's light-target draws bind, and DLSS keeps its history instead of starting anew every frame. Measured with
NX2NAVI_MV_CHECK while turning: 176 to 227 px of camera motion a frame, a warp error of 0.006 to 0.015 with the vectors
against about 0.025 without.

There is no jitter on Ocarina's frame yet: with it the picture went white within seconds, with or without motion
vectors, while the same frames without it look right.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 10:16:34 -05:00
chrg6
3c0377abf8 Run DLSS on Ocarina's frame, which has no G-buffer and tonemaps straight into its 8-bit output.
Without a scene-size G-buffer the DLSS point arms on the first scene-size light-target pass (B10G11R11 with the D32F
scene depth) and runs at the pass that tonemaps it: into the output when that is larger (DLSS scales and the tonemap
samples its picture), or back into the light target for an sRGB output of the scene size (DLAA). The sizes come from
the frame, since Ocarina's builds have no scene or output patch sets for res_sizes. The deferred engine's shape is
kept as it was.

Ocarina's camera for the motion vectors is not found yet: the c3 view block its light-target draws bind gives a
reprojection that moves a still picture by tens of pixels (NX2NAVI_MV_CHECK), so there is no camera motion and no
jitter there, and DLSS works on each frame alone. dlaa stays off by default.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 09:26:51 -05:00
chrg6
1b39aabd37 Describe what the code does in comments and docs.
Comments say what the code does and why, docs/maxwell.md keeps the Maxwell facts, and fetch_refs.sh clones only the checkout the build reads. Code is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:40 -05:00
Valiantsin Dzerakh
0478c13e45 Import the base tree (MIT, see LICENSE) 2026-11-07 05:04:38 -05:00