Commit Graph
92 Commits
Author SHA1 Message Date
Valiantsin DzerakhandClaude Opus 5.5 1ab3856fdd TOTK_SYNC_AHEAD=check measures what a one-frame-late frame wait without shadow storage would race, with the wait left exact
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>
2026-11-21 09:19:54 +03:00
Valiantsin DzerakhandClaude Opus 5.5 b8edbdd873 The per-object motion reads last frame's object and bone blocks from last frame's uniform snapshot, kept one more frame, instead of having every snapshot copy both neighbours of each such block: the town's snapshot was 2836 KiB a submission with the neighbours and 1916 KiB without, and each byte is copied twice with the motion on (into host memory for the CPU's reads, then into the device copy). vk_mvprev_find still picks the address; the block's bytes and the motion draw's descriptor come from vk_usnap_prev. A block bound at the same address both frames keeps this frame's copy, as before.
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>
2026-11-20 19:49:47 +03:00
Valiantsin DzerakhandClaude Opus 5.5 82a1f526a9 The per-object motion keeps last frame's blocks only for the banks its listing looks up (c5, c6, c7, c14, c15 and c16 in the town and the field) instead of every bank a variant reads (c5 to c16 but c12): the scene draws stop probing the c8, c9, c10, c11 and c13 blocks into the table and the snapshot stops copying their neighbours, none of which a lookup ever read (c9 and c11 stay at this frame's block, the others belong to shaders that are not drawn again). The table entries of the banks looked up are the same as before, so the vectors are too. The listing's per-shader memo now comes before the shader's resource list and also keeps the filters' verdict, so a scene draw that is no candidate costs one memo probe.
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>
2026-11-20 19:44:10 +03:00
14ccef9a80 AMD FSR 3.1 upscales the 3D scene on the FidelityFX SDK 1.1.4 sources behind scene_upscaler = fsr3, off by default
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>
2026-11-20 14:39:57 +03:00
1ea8838da9 Frame generation inputs and dispatch with DLSS off: the replay follows the scene for it, records the camera motion vectors and the SDK's prepare at the scene's end, copies the picture before the HUD sRGB-encoded as the window texture holds it, and runs the optical flow and the interpolation after the frame's window texture, nothing shown yet
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>
2026-11-20 10:09:58 +03:00
itaskaev2andClaude Opus 5.5 4edf63a6dd Soft shadows keep their width on the ground across the sun cascades
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>
2026-11-19 11:41:00 +01:00
Valiantsin DzerakhandClaude Opus 5.5 0350d91b72 The water volume mask bias is decided once per state change and only for LEQUAL and GEQUAL draws (#1126)
prepare_draw sets ex->volume_bias where the draw's dynamic state changes, so record_fixed reads a flag instead of
checking the pipeline state on every draw. A census of the matching draws in the field, the town and rain found two more
states with the same blend and depth signature, both tested GREATER: the volume's back faces into the same channel and
another shader's front faces into red. GREATER's equal case fails, and the bias toward passing turned the game's tie
the other way for them; they are now left as the game sets them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-18 18:18:22 +03:00
ipavlin98 cc84958cee Give TotK's water volume mask draws a depth bias toward the viewer, so a puddle's surface, on which the volume's top face lies, counts as inside the volume in every frame instead of by rounding and the camera's sub-millimetre motion, which turned the puddle's reflection on and off in patches from frame to frame, more with DLAA 2026-11-18 05:41:05 +03:00
Valiantsin DzerakhandClaude Opus 5.5 af1329440d barrier_merge costs nothing per draw while off and guards the reads a merged Barrier cannot order
The storage check in prepare_draw ran on every draw with the setting off; it now runs only when the context's cached
flag is on, and after bind_resources, which may restart the pass (ub_written) and clear the note of a storage writer.
A pass whose target is a 3D BC stand-in no longer merges: its texture gets the writes only at the pass end. After a
merged Barrier, a draw with a storage image or a vertex texture that is an attachment of the pass ends the pass and
begins the next behind the full barrier (merge_guard), since the in-pass by-region barrier orders only fragment shader
sampling. The per-bit Barrier counters count only under TOTK_TRACE=gfx.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:03:42 +03:00
ripps818andClaude Sonnet 5.5 61a407b0e1 Add barrier_merge: keep a render pass open through the game's texture-invalidating barriers
The runtime ended the Vulkan pass at every game Barrier. A Barrier that asks only for fragment ordering and texture
invalidation (ORDER_FRAGMENTS | INVALIDATE_TEXTURE) now leaves the pass open when none of the pass's draws can write a
storage buffer or image, and the next draw sees the writes through a by-region barrier in the feedback layout. Off by
default (barrier_merge = 0, TOTK_BARRIER_MERGE). TOTK_TRACE=gfx also prints how many passes ended by a change of targets
and by the game's Barrier, how many Barrier calls were merged and which NVN barrier bits they carry.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-11-15 22:00:38 -06:00
Valiantsin DzerakhandClaude Opus 5.5 d6715029e5 A device loss logs the last descriptor pool events, and a pool with two owners is logged at once
The replay keeps a ring of the last 1024 descriptor pool events (a CbRes made, a pool handed to a CbRes from the spare
list or new, a pool reset or destroyed), a few per submission and nothing per draw. vk_device_fault_report logs the
newest 256 after a device loss. Each pool's owner is checked under spare_lock as the pool is handed out and as it is
reset: a pool handed out while another CbRes holds it, or reset by a CbRes that does not hold it, is logged at once
with the 64 events before it. On a GTX 1050 Ti with TOTK_BIND_INC on, the validation layer reported a descriptor pool
reset while a command buffer used it, then binds of its sets, before a device lost that slower diagnostic runs did
not bring back; the next one names the pool's owners.

--probe-vkexec gains a set batch part (60 submissions of batched sets with splits and empty submissions beside them),
a pool owner part (one pool handed to two CbRes, then reset by the one that no longer holds it: two conflicts) and
calls the device loss report on the working device to check that it comes out and returns.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-15 18:55:39 +03:00
Valiantsin DzerakhandClaude Opus 5.5 2f32f21f5c The incremental draw preparation is on by default, and bind_inc and memx are listed in totk.ini.example (#381)
TOTK_BIND_INC had no settings row, so unset meant off. It now has one with default 1, like TOTK_MEMX: 0 or off turns
it off, check keeps the comparison mode. The check mode found 0 differences on 13 scenes (town, fight2, Kakariko,
Great Sky Island, Depths lit and dark, a shrine, rain, a cave, evening, a field, the menus, entering a building).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-14 20:03:24 +03:00
Valiantsin DzerakhandClaude Opus 5.5 d52c16e651 Merge dev into task/bindinc: the incremental binding keeps its logic and takes dev's inline compares and copies (TOTK_MEMX)
The ten conflicts in runtime/vk_exec_draw.c were dev's memcmp/memcpy -> MX_EQ/MX_NE/MX_CPY swap on lines task/bindinc
had rewritten. Each keeps the task/bindinc code with the swap applied to its hot-path compares and copies (push and set
signature compares, the shader memo's pipe compare, the pipeline memo and recorded-state compares, the vertex
attribute and stream lists). The check-mode compares that call bi_mismatch keep memcmp as the reference.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-14 18:47:31 +03:00
Valiantsin DzerakhandClaude Opus 5.5 354887a2a3 The replay's per-draw compares and copies run inline instead of calling ucrtbase memcmp and memcpy (TOTK_MEMX)
runtime/memx.h: MX_EQ and MX_CPY compare and copy with 32-byte loads in the caller; a constant size unrolls, a variable
one loops without a call. Used for the pipeline memo state and the recorded dynamic state (prepare_draw), the shader
answers (shader_for, which also compares the stored hash first), descriptor signatures and push sets (bind_resources),
c0 and support buffers, the uniform update copies, the vertex attribute and stream binds, pipe_find, the dynamic state
delta, nvn_hm_get and the recorder's tail_copy. TOTK_MEMX=1 (default) takes the inline forms, 0 keeps memcmp/memcpy,
check runs memcmp and logs any compare the inline form answers differently.

Machine A, 4 cores, 1080p, bindinc on both sides: Lookout Landing replay 9.64 -> 9.16 ms over 4+4 runs (each ABBA block
-0.40 and -0.55 ms), Kakariko 9.04 -> 8.38 ms over 2+2. Check mode on dev: Lookout Landing pan, 0 mismatches.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-14 11:35:29 +03:00
Valiantsin DzerakhandClaude Opus 5.5 e7dbe74f44 Graphics descriptor sets are no longer shared within a submission, and the per-draw keep tests are made once per pipeline (TOTK_BIND_INC)
The set table probe cost about 54 ns a draw on Lookout Landing (a cache miss on the 4096-entry table plus two hashes of
every set) and shared 6.7% of the allocated sets. Batched allocation stays. bi_plan's walk over every binding is
replaced by per-binding tests (bi_tests, kept in the pipeline's RsList) that bind_resources evaluates in its own loop,
with the volatile flag kept per binding. A kept binding after a build rereads only the generation group it depends on.
Check mode: 0 mismatches in town, Kakariko, fight2, island and Depths, 60 s each.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-14 07:58:16 +03:00
Valiantsin DzerakhandClaude Opus 5.5 ecd0f004cc A kept descriptor is built again when a build earlier in the same draw moved the generations it depends on (TOTK_BIND_INC)
Check mode on the island save caught one kept texture descriptor whose view differed from the full build: a lookup
of an earlier binding of the same draw made a texture entry (vk_texture_slots_gen moved during the build), and the
full build resolved the later slot under the new generation. bi_plan now records per kept binding whether it depends
on G_tex or G_buf; after a binding is built, the loop reads both again and builds every later binding that depends
on one that moved. Island and Lookout Landing check, 60 s each: 0 mismatches.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-14 00:18:52 +03:00
Valiantsin DzerakhandClaude Opus 5.5 dba054110b Graphics descriptor sets are shared within a submission and allocated in batches (TOTK_BIND_INC)
Adapted from Savage Wind Recomp's set_finish and vk_alloc_set (MIT, row in THIRD_PARTY_NOTICES.md). A graphics set
that would be allocated and written is first looked up in a per-context table keyed by two independent 64-bit hashes
of its layout and packed descriptors and stamped with the submission's fence: an equal set of the same submission is
bound instead. New sets come from the submission's last pool in batches of one layout that grow 1, 2, 4, 8, 16 while
the layout repeats, one vkAllocateDescriptorSets per batch instead of per set. check stores each set's contents,
compares every hit with them and allocates as before. Lookout Landing: about 2,000 allocated graphics sets a frame,
6.7% shared; check 0 mismatches; on ran 60 s without errors.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 23:52:37 +03:00
Valiantsin DzerakhandClaude Opus 5.5 c26ed9ba7b A graphics draw with the previous draw's pipeline rebuilds only the descriptors whose inputs changed (TOTK_BIND_INC)
With TOTK_BIND_INC the graphics build writes into VkExecDraw.BI, which keeps the packed descriptors of the last build
and, per descriptor, what it counted for the traffic census and whether it copied guest bytes. bi_plan names the
bindings to build again: support always; c0 when a storage slot of the stage or the base instance changed; the
compiler constants when the program, its serial or G_buf changed; uniform and storage slots when marked or when G_buf
(pools and mirrors, snapshots, overlays, the submission) moved; textures and images when a handle was marked or G_tex
(slot registrations, samplers, the storage epoch, the feedback layout, DLSS's swap) moved; storage addresses held in
a uniform block, and any descriptor whose build copied guest bytes, every time. A pipeline with a written storage
binding is built in full. Kept descriptors count their census again; a set none of whose descriptors changed and that
is still pushed or bound records nothing, without the psig or sig compare. check builds everything as before and
compares each kept descriptor and its census with the build. Lookout Landing check, 60 s: about 17,000 kept bindings
compared a frame, 0 mismatches.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 23:49:21 +03:00
Valiantsin DzerakhandClaude Opus 5.5 c86feefadc A draw skips the vertex stream loop while none of the streams it reads changed (TOTK_BIND_INC)
bind_vertex_buffers resolves and compares the 17 streams the vertex input reads on every draw. When the last
prepared draw bound the same set of streams in this command buffer, no op changed one of their ranges, vk_mem_gen
stayed, and with static vertex input the strides did (pipe_gen), every stream would resolve to what is bound: the
loop is skipped. check runs the loop and counts a stream it would bind again. Lookout Landing check, 60 s: 62% of
draws skip it, 0 mismatches.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 23:47:24 +03:00
Valiantsin DzerakhandClaude Opus 5.5 772d9dea66 The feedback scan keeps its answer across texture binds that leave the fragment shader's slots alone (TOTK_BIND_INC)
feedback_scan's memo is keyed by tex_serial, which every texture or image bind of either stage moves. When the memo
holds the last prepared draw's scan (BiLast.fb_ok) of the same fragment shader in the same pass, no slot was
registered since, and no marked fragment texture slot is one the shader reads (BiMask.ftex, buffer textures
included), the answer stands. check scans and compares. Lookout Landing check, 60 s: kept on 68% of draws (the memo
alone 60%), 0 mismatches.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 23:45:44 +03:00
Valiantsin DzerakhandClaude Opus 5.5 dcd5dc3cf4 A draw keeps the previous draw's shaders while no texture or image slot their translation reads changed (TOTK_BIND_INC)
The shader memo already keeps them while the program, the bound state and the first 32 handles of both stages are
equal. With the marks it also keeps them when other slots were rebound: no program or serial change, the same
pipe_gen and slot generation, and no marked texture or image slot inside the program's slot mask (slot_mask_of,
~0u while unknown), which are all answers_for reads. BiLast.sm_ok says the memo holds the last prepared draw's
shaders, so a mark cleared by a later draw never hides a change. check looks the shaders up and compares. Lookout
Landing check, 60 s: shaders kept on 65% of draws (the memo alone 55%), 0 mismatches.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 23:44:01 +03:00
Valiantsin DzerakhandClaude Opus 5.5 ff3a95d0f1 A draw whose pipeline state no op changed takes the previous draw's pipeline and state without building and comparing them (TOTK_BIND_INC)
From task/fastrep 18eab597 (TOTK_PIPE_GEN), now under TOTK_BIND_INC: the pipeline memo keeps the pipe_gen its state
was built under; a draw with that generation and the same shaders, primitive, restart, feedback flags and depth mode
takes the memo's state and pipeline and skips the three 360-byte compares (memo, cache entry, recorded state). The
shader memo skips its compare of pipe the same way. check builds and compares as before and counts a state the
generation called unchanged that differs (bi_mismatch logs the first of each kind with its marks and generations).
The census runs only in check mode. Lookout Landing check, 60 s: 67% of draws keep the state, 0 mismatches.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 23:42:11 +03:00
Valiantsin DzerakhandClaude Opus 5.5 2abfa23a67 Bind ops mark the slots whose bytes they changed and a per-draw census counts what the next draw could keep (TOTK_BIND_INC)
vk_exec_bind compares before it writes: VkExec.dirt gets a bit per program, uniform, storage, texture, image and
vertex buffer slot that changed, pipe_gen moves with every change of the pipeline state (also depth clamp, raster
discard and the target formats), fbl_serial with the feedback-layout images, OV.gen with the uniform overlays.
vk_exec_draw_new_cb, copy_state and state_reset mark everything. With TOTK_BIND_INC=on or check, bi_census sorts each
prepared draw by what changed since the draw before (nothing, uniform slots only, other slots, another pipeline) and
counts the descriptors to rebuild, the generation moves and the shader, pipeline, vertex buffer and feedback keeps;
a bind_inc line every 300 frames. No draw is prepared differently yet.

Lookout Landing, 4500 draws a frame: nothing changed 8.7%, uniforms only 55.5%, other slots 22.9%, another pipeline
12.8%; 3.4 of 11.2 descriptors a draw to rebuild; uniform overlays move on 2.3% of draws, snapshots 0.1%.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 23:40:01 +03:00
itaskaev2andClaude Opus 5.5 be5f4db3bf Skip 1.2.1's own scaling draw after DLSS instead of feeding it DLSS's picture (#1106)
On 1.2.1 the light target is scaled to the output by an upscaler of the game's own (twelve texelFetch taps,
fragment shader e6f7ae35b8479ef3), not by the FXAA draw vk_dlaa_skip_draw knows. It addresses its source in the
scene's pixels, so DLSS's output-sized picture swapped in for the light target (tex_info) showed only its top-left
part, two thirds at DLSS quality, stretched over the screen: the "floating camera" of #1106. #660 (169d057) made the
swap reach that draw: before it, DLSS ran on another target and its picture never reached the screen on 1.2.1.

DLSS already writes its picture into the scaling pass's target, so on 1.2.1 a draw in a pass larger than the light
target, without depth, that samples the light target is now skipped (vk_exec_draw_samples). The HUD drawn later into
the same target samples other textures and stays. 1.4.3 keeps the swap, where a tonemap can be the scaling draw.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-11 11:53:25 +01:00
itaskaev2andClaude Opus 5.5 0adcc96bae Bind the motion draws' set 1 before pushing set 0 under the validation layer, as the replay's draws do
vk_mvobj_execute's draws (mo_emit) push set 0 and then bind set 1. Under the validation layer (SDK 1.4.363) a push for
a new INDEPENDENT_SETS layout keeps the push set the layer made for an earlier layout (the replay's last draw) and drops
the writes to bindings that layout lacks: every run with DLSS and validation reported the last-frame banks as missing
(VUID-vkCmdDrawIndexed-None-08114, variable vp_c20) and then faulted inside VkLayer_khronos_validation.dll. The
replay's own draws avoid this by binding set 1 first (set1_first in bind_descriptors); the motion draws now do the
same, only while the layer is active.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-10 15:11:08 +01:00
Valiantsin DzerakhandClaude Opus 5.5 213bc6af6a Wait out a short out-of-memory spell instead of ending the game
Crash logs ended in fatal errors when the system's commit ran out for a moment: a staging buffer (-2), a guest commit
with the system at 24442 of 24447 MB, image creation (-2). Windows grows a system-managed page file and some drivers
charge device memory to the process commit, so the same request often succeeds a moment later.

oom.c holds the wait nvn_api_cmd.c realloc_wait already used: it requests a release, retries every 10 ms for up to
30 s, logs the first 8 with the commit numbers. rt_oom_request sets a flag that vk_texture_new_submit takes (one atomic
load per submit): it drops up to 256 released textures of the two kinds sweep_pressure drops (overwritten, clean
mirrored), never GPU-written content, and frees the headroom blocks. vk_oom_release does at once what needs no submit:
headroom blocks, finished upload batches and deferred objects.

- memory.c commit: waits and retries without frz_lock between tries; every caller waits, boot ones too.
- vk_staging and the draw-data chunks (vk_exec_draw.c scratch): release, then wait; the open upload batch is not
  submitted, its caller still records into it.
- create_tex: release and one retry; no wait under the texture lock, the texture is asked for again at its next use.
- the texture write-back buffer: release, then wait.
- the headroom thread keeps no spare block for 30 s after a failure.

TOTK_TEST_OOM=<site>:<n> (commit, staging, image, chunk) fails the first n attempts; probe-kernel forces 3 refused
commits, probe-vkexec 2 staging buffers, 1 image and 1 chunk, and both pass through them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-09 09:53:30 +03:00
ipavlin98 fd4427cb17 Run DLSS on the chunked replay
DLSS kept the replay on one thread: its decisions follow the passes in replay order, its jitter and camera are noted
draw by draw, and its evaluation reads the per-object motion list the scene's draws build. Now the scan takes DLSS's
decisions on its own state (vk_dlaa_plan, split out of vk_dlaa_pass) and notes the scene camera in replay order
(vk_dlaa_scan_draw), and the copy of the lit scene and the evaluation are records of the chunk that holds the op
(vk_dlaa_exec). A chunk's draws are jittered while they come before the op the scan ended the scene at. The
evaluation waits until the chunks before it are recorded, and the chunks after it wait for it (a hold), since they
sample its output. Chunks note their draws for the per-object motion one at a time (mo_lock in vk_exec_draw.c); the
uniform overrides of a motion draw and shader_for's last-frame variant flag are per thread, as the other chunks' draws
go on meanwhile; the FXAA shader table takes concurrent lookups. fxaa = 0 runs in chunks too.

With dlaa = 1 at 4K (DLSS Quality) on 1.2.1 and an RTX 5070, replay_workers = 4 takes the frame rate from 70 to 78
fps where the CPU is the limit, with no one-frame glitch in 241 consecutive frames.

(cherry picked from commit 04f8bb1b57)
2026-11-08 19:39:18 +03:00
ipavlin98 6b4d84ec50 Record large submissions in chunks on several threads
replay_workers = N (0, the default, keeps the sequential replay) records a submission of enough draws in chunks that start between passes, on N threads and in twice as many chunks, and submits them in order. In the town at 4K on an RTX 5070, 4 threads take the replay from 8.2 to 4.5 ms a frame and the frame rate from 71 to 79.

A sequential scan on the submitting thread applies only what is queue state (fixed-function and bind ops, render targets, render enable and conditions, samples-passed counting) and keeps a copy of it where each chunk starts, with how far its lists got: inline uniform updates, RENDER_COND results, contact shadow marches and the ops that drop uniform copies. A chunk takes those from the lists instead of from the chunks before it, starts with every image in GENERAL and ends with a full barrier, so its recording depends only on the copy and the lists.

Each thread records with a context of its own: the replay's memos, descriptor state, occlusion query segments and per-command-buffer state moved from file statics into VkExec and VkExecDraw, and the tables the contexts share (stand-ins, depth twins, views, mirrors, timestamps, completion callbacks, crumbs) take locks or atomics. The scan makes render targets live in replay order and notes the op (vk_texture_revive), since a chunk can sample a storage before the chunk that renders it got there or after the scan went past a later target there: a chunk's texture slots resolve as of its own op, and the chunks after a texture with no entry yet wait for the chunk that makes it. Uniform copies are dropped at the op that writes them (vk_usnap_drop_at).

Contact shadows plan their march in replay order (vk_contact_plan) and the chunk that holds its op records it; the HDR source keeps the copy that comes last in replay order. DLSS and fxaa = 0 act on the frame draw by draw and keep the replay sequential. TOTK_REPLAY_STATE_CHECK=1 compares the state after the last chunk with the scan's.

A draw or dispatch whose shaders write storage buffers drops the uniform copies they cover from its own op on, as the other GPU writes do (scan_writes). The scan still translates no shader: it reads the shaders the recording contexts translated from a table shared with them (vk_exec_draw_writes), and an op whose shader no context translated yet holds the chunks after it until it is recorded. Before, the chunk that recorded such a write dropped the copies for every position at that moment, so a later chunk recorded first bound the copy taken at the submit call: on an RX 7800 XT on RADV the grass near the camera came out dark or unshaded in some runs with 4 threads.

TOTK_TRACE=perf counts the chunks held per submission by reason and logs when each chunk of the last submission began. The settings menu has the thread count under Advanced, Scheduling, with the count this run uses under it.

The recorder thread begins, ends and submits the submission's open part (vk_rec.c rec_end_cb), so a chunk's mirror prologue stays off it: vk_sync_part waits until the recorder is idle, as the part pool allows one thread at a time, then records, ends and submits the prologue itself. Through the recorder the prologue could reach the queue before it was ended, and the GPU lost the device at the title screen.

(cherry picked from commit b94e2e9642)
2026-11-08 19:39:17 +03:00
Valiantsin DzerakhandClaude Opus 5.5 7c5f88d33c Rebind only the vertex streams a draw reads that changed, and leave the streams it does not read bound
With vb_used the streams a draw does not read got the zero buffer, so draws in a row that read different streams
bound all 17 slots again almost every draw. On an RTX 5060 Laptop that cost 16 % of the frame rate (issue #974). Now
only the read streams are compared, one bind covers the changed ones, and the others keep the buffer bound earlier
in the command buffer, which the deferred frees keep alive until the submission is done.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-08 13:50:32 +03:00
Valiantsin DzerakhandClaude Opus 5.5 36a83bd78f Keep pool memory and texture records until the submission being recorded is done
vk_pool_fini freed import, physical and virtual pools right after vkQueueWaitIdle, which covers submitted work only;
with the synchronous replay another game thread may be recording a submission that names the pool. Every pool kind
now retires through the completion queue at the later of the recording's value and the next one.

destroy_tex freed the texture record while other threads still held &e->pub (render targets for a pass, storage
bindings writing wr_fence through it): they read a stale VkImage and cached views by that handle. The record is now
freed with its image's stamp (vk_defer_free), and views of a dropped entry are deferred at once instead of cached.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 20:05:49 +03:00
Valiantsin DzerakhandClaude Opus 5.5 2eb5adebd6 Merge #898: record only the dynamic states that changed since the last pipeline state
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 11:07:22 +03:00
Claude 752e15bcc7 Record only the dynamic states that changed since the last pipeline state in the command buffer
Each change of pipeline state in prepare_draw re-recorded the whole dynamic state through vk_pipeline_set_dynamic:
about twenty commands, the vertex input packet rebuilt, and the blend enable, equation and write mask of every
attachment, however few fields differed. Every replay pipeline declares the same dynamic states, which a pipeline bind
leaves as they were, so vk_pipeline_set_dynamic_delta compares the values with what the command buffer already holds
and records only the commands that differ; the vertex input is rebuilt only when the vertex shader, attributes or
streams change. The record is per command buffer (B.dyn_valid, cleared with the rest of B on a new one or a split).
TOTK_DYN_DELTA=0 records the full set as before; ab alternates; TOTK_TRACE=perf counts calls, commands and calls that
changed nothing.

1.4.3, RTX 4070 Ti, 2560x1440 DLAA, standing still on Great Sky Island, TOTK_AB=10 (tools/ab_summary.py): 13.75 ms
a frame with the full sets, 12.59 ms with the deltas (-8.5%, 72.7 to 79.4 fps); replay 7.77 to 6.67 ms. An A/A run
with the same method moved the frame by 0.08 ms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BspNcezsiV8zT8wcnXrrME
2026-11-07 03:00:57 +00:00
Claude 324f85602f Clear the uniform overlays once their copies are recorded
After vk_exec_ub_flush the full barrier barrier_to records makes the copied bytes visible, so later bindings can read
the real buffer. The overlays stayed until the submission's end instead: every later binding inside their span
scanned them, and a block an overlay covered in part got a merged scratch copy that the descriptor build skip then
left alone. With the updates between passes overlaid too (Great Sky Island: 274 of 594 per submission) that was 290
merged copies and 1.1 ms more replay a frame, more than the barriers saved on the GPU (0.3 ms).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BspNcezsiV8zT8wcnXrrME
2026-11-07 00:18:52 +00:00
Claude 49b2314dd8 Overlay uniform updates between passes too, so a run of them before a pass costs one barrier
An update between passes was copied at once behind a full barrier of its own, and the copy it recorded made the next
update or pass begin need another: a run of k updates before a pass cost k + 1 full barriers, about 250 updates a
frame on Great Sky Island. They now become overlays like the updates inside a pass; every GPU reader of the real
buffer after them comes behind barrier_to, which records the copies first. TOTK_UB_OVERLAY_OUT: 1, 0, ab; the perf
line counts the overlays made between passes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BspNcezsiV8zT8wcnXrrME
2026-11-06 22:35:01 +00:00
Claude b317f1c37e Order overlapping uniform overlay copies with a transfer barrier
vk_exec_ub_flush recorded the copies of a pass's overlaid uniform updates back to back behind one barrier. Copies
with no barrier between them may run in any order, and ov_add only drops an older pending copy of exactly the same
range, so two updates that overlap in part could leave the older bytes in the buffer where they overlap. A copy that
overlaps one recorded since the last barrier now waits for it behind a transfer-to-transfer barrier; the perf line
counts those barriers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BspNcezsiV8zT8wcnXrrME
2026-11-06 22:33:24 +00:00
Valiantsin DzerakhandClaude Opus 5.5 da24dd4cba Bind only the vertex buffers a draw's vertex input reads
The game keeps about ten vertex streams bound while a draw reads a few of them, and bind_vertex_buffers resolved and
compared all 17 slots on every draw. vk_pipeline_set_dynamic now returns the bindings of the state's vertex input,
from the same vertex_input that builds the dynamic input or the pipeline's, and the other slots get the zero buffer.
A stream is taken from the previous bind only when it was resolved there. In the fort, bind_vertex_buffers drops from
9.0% to 7.2% of the replay thread's work (two profiles, recorder on). TOTK_VB_USED=0 binds every stream as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-06 20:43:33 +03:00
Claude 9069c9208f Keep uniform updates inside a pass as scratch overlays and copy them at the pass end
nvnCommandBufferUpdateUniformBuffer between draws was replayed as a buffer copy at that point, which needs a barrier
against the draws that read the old bytes, and a barrier ends the open pass. The terrain and the sky islands' meshes
update before most of their draws: about 200 updates inside passes in a field submission, each a pass end, a full
barrier and a pass begin.

An update inside a pass now goes to the scratch buffer only. Later bindings of a range it covers bind the scratch
bytes (ub_bind, from ubo_info and bind_range); a binding it covers only in part, or at an offset the alignment
refuses, gets a merged copy of the snapshot or guest memory with the submission's updates on top. The copies into the
real buffers are recorded behind one barrier of their own at the next full barrier (vk_exec_ub_flush from
barrier_to). A new update of a range replaces the older overlay and pending copy of it, so the lists stay as short as
the blocks. Copies, fills and storage writes of the same bytes retire the overlays they touch, and a storage binding
over a pending copy brings the copies forward with a pass restart. bind_sig hashes the overlay each of the draw's
uniform and storage bindings reads, so a draw of an updated block binds the newest bytes and the others still skip
their descriptor build. TOTK_UB_OVERLAY=0 copies at once as before; ab alternates; TOTK_TRACE=perf prints the
counts.

1.4.3, RTX 4070 Ti, 2560x1440 DLAA, standing still on Great Sky Island, TOTK_AB=10 (tools/ab_summary.py):
17.03 ms a frame with the copies at once, 14.39 ms with the overlays (-15.5%, 58.7 to 69.5 fps); replay time on the
CPU 8.81 to 8.30 ms.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BspNcezsiV8zT8wcnXrrME
2026-11-04 23:54:02 +00:00
Valiantsin DzerakhandClaude Opus 5.5 8d349fa50d Merge branch 'dev' into task/wbmerge
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-04 21:55:37 +03:00
Valiantsin DzerakhandClaude Opus 5.5 a248cc9f54 Wait for the pipelines of item icon captures instead of skipping their draws
The icon capture draws into a colour target rendered for the first time, with the per-frame depth buffer. Under
pipe_skip any and fullscreen the recent depth made the pass look per-frame, so draws with a pipeline still building
were dropped and the icon kept missing parts (a fused weapon without its blade) until the item was captured again.
vk_exec_targets_recent now ignores a recent depth target while a colour target is rendered for the first time.

On fight2 with a warm driver cache no draw waited. With a cold cache the first world frame after loading waits for
the scene's new HDR target (2.9 s without prepared shaders), and one 64x64 capture waited 29 ms in the fight.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-04 15:33:07 +03:00
Valiantsin Dzerakh d5e5a454ce Merge branch 'dev' into task/wbmerge
# Conflicts:
#	runtime/cfg.c
2026-11-04 14:09:50 +03:00
Valiantsin DzerakhandClaude Opus 5.5 80349df1a6 Merge pull request #594: per-object motion vectors for DLSS, Ultra Performance, live DLSS modes and per-mode presets (DizzyMan1337)
Merged on top of #566. The camera motion is #594's: its blocks in camb also feed the per-object pass, and
dlaa_mv.comp stays as on main, fed the identity and the reprojection made on the CPU in double. #566's
vk_usnap_host goes, vk_usnap_bytes serves the same. #566's scaling-draw swap, camera rune guard and output-size
scaling pass stay in vk_dlaa.c. Translator patch 0012 becomes 0015 after dev's 0012-0014.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 19:27:04 +03:00
DizzyMan1337andClaude Opus 5.5 ad3ecd8f3c Per-object motion vectors for DLSS, camera motion in double precision, Ultra Performance and DLSS presets.
Solid scene draws whose position reads an object's own uniform blocks (bones c5, place c7, c6) are drawn a second time
with a vertex shader variant (translator patch 0012) that computes last frame's position from last frame's blocks; the
game writes those into two copies that take turns, and the uniform snapshot keeps both (vk_mvprev.c). The draws are
listed during the replay and recorded through the recorder thread with repeated state skipped. Blocks whose last-frame
copies measured wrong (c9 and c11 of walls and cliffs) keep this frame's; positions without an object block (terrain,
grass) keep the camera's motion, which measured as good.

The camera motion is reprojected in double precision on the CPU: the float inverse made noise on far surfaces and a
still camera that DLSS followed into a wobble. DLSS gets Ultra Performance (a third of the output, also in the F1 menu)
and NVIDIA's recommended preset per mode (K, M for Performance, L for Ultra Performance; TOTK_DLSS_PRESET overrides).

Test switches: TOTK_MV_CHECK measures the motion vectors and DLSS's output against the picture, TOTK_DLAA_TIMES times
DLSS and the motion passes on the GPU. A reactive mask pass for transparents is included but off (TOTK_DLSS_REACTIVE):
its mask covers sky and fog too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 17:31:58 +08:00
Michael Anthony Bitoon 5b23d6671b Make DLSS's picture reach the screen and stop the whole image wobbling with DLSS and DLAA
With a DLSS mode the game's own scaling draw covered DLSS's picture with the jittered scene. At a 1280x720 scene and
1920x1080 output the scaling is done by the game's tonemap, not by the FXAA shader vk_dlaa_skip_draw looks for, so no
hash list could catch it. From DLSS's evaluation to the frame's end, passes larger than the light target now sample
DLSS's output in its place (vk_dlaa_swap_img, applied in vk_exec_draw.c tex_info), whichever shader scales.

The motion-vector shader inverted the view-projection in 32-bit floats. With a still camera (identical matrices each
frame) the rounding still gave a few hundredths of a pixel of motion that changed with every jitter step; DLSS
followed it and the whole picture wobbled. mv_pass now makes last frame's view-projection times the inverse of this
frame's in double precision from the uniform snapshot (vk_usnap_host), and the shader only applies it.

Measured on 1.4.3 indoors, still camera, mean per-pixel change between consecutive frames on static areas:
DLSS quality 0.18 -> 0.06, DLAA 0.14 -> 0.05. A camera pan at 10 px per frame shows no ghosting.
2026-11-03 08:40:13 +08:00
Michael Anthony Bitoon e2fdb5b226 Merge a shader's storage-buffer writes back into guest memory word by word
In mirror mode the replay copied every range a shader wrote from the VRAM mirror back to guest memory at the end of the
submission. TotK's particle emitters write a new particle's whole record into a free slot of a buffer the simulation
shaders update, so the copy put the mirror's stale words over fresh spawns: the waterfall splashes and mist showed about
once a second and the falling rain drifted. Now each written range gets a snapshot in the prologue of the part that
first writes it, after the CPU-to-mirror copies, and at the end of the submission shaders/wb_merge.comp writes only the
32-bit words that differ from the snapshot. Words the GPU left alone keep what the CPU wrote. About 0.1 MiB and three or
four dispatches a frame in rain. TOTK_WB_MERGE=0 restores the copies. The merge's push descriptors go through rec_push
so they keep their place in the recorder's call order.
2026-11-03 01:19:10 +08:00
Valiantsin DzerakhandClaude Opus 5.5 710b7f5dd0 The device-lost report names the pass or dispatch the GPU was running
The replay marks the start and the end of every render pass and dispatch
(vk_exec.c vk_exec_pass_begin and vk_exec_pass_end, vk_exec_draw.c
op_dispatch and op_dispatch_ind), outside the pass so no pass is broken.
With VK_AMD_buffer_marker the GPU writes the last started and finished mark
into words 2 and 3 of the breadcrumb buffer; with
VK_NV_device_diagnostic_checkpoints the same points are checkpoints, read
with vkGetQueueCheckpointData2NV after the loss. A ring of 4096 records maps
a mark to its submission, index, target formats and size, draw count and the
input ids of the first and last draw's shaders. vk_device_fault_report logs
both marks with their records. The calls go through the recorder thread.

TOTK_PASS_CRUMBS (menu: GPU pass marks): unset, AMD buffer marks only; 1,
also NVIDIA checkpoints, whose extension is enabled only then; 0, none.
The design follows Savage Wind Recomp 3fae40c; the code is ours.

--probe-vkexec gets a pass marks part: pass, dispatch, pass in one
submission; the marks read back after the fence name the last pass and the
report text says so.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-02 09:50:30 +03:00
MierzwaMatheus f4f9d6dd53 Keep the replay's transient buffers and the zero buffer in host-coherent memory
The CPU fills these chunks and the zero buffer and never flushes them. On Intel integrated GPUs (ANV on Meteor
Lake) the host-cached memory type that VMA picks for random host access is not coherent, so the GPU read what
an earlier submission had left there. A lighting fragment shader got another draw's UpdateUniformBuffer payload
as its step count (0x3f800000 instead of 96 or 128), looped about a billion times and lost the device; other
stale payloads drew the black frames. Requiring HOST_COHERENT keeps the same memory types on GPUs where host
memory is already coherent.
2026-11-02 01:22:13 -03:00
Valiantsin DzerakhandClaude Opus 5.5 f700c28cc4 Merge pull request #113: Add NVIDIA DLSS/DLAA and a toggle for the game's FXAA (DizzyMan1337)
The sampler keeps both the DLSS mip bias and pull request #321's anisotropic level; DLAA's pass hook runs after the
contact shadows'; the extension arrays grow for NGX's extensions beside HDR's and the calibrated timestamps.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 11:40:27 +03:00
Valiantsin DzerakhandClaude Opus 5.5 4ab1dc80f5 Merge task/shadows: pull request #89, shadow options (itaskaev2), with the follow-ups
Keeps pull request #321's anisotropic filtering, which #89 added a second time, and #267's shadow_res beside #89's
shadow_scale. The HDR parameters move from byte 1408 of the support buffer to 1440: translator patches 0010 and 0011
read pcss at 1408 and shadow_spread at 1424.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-01 11:39:38 +03:00
ProSepandClaude Opus 5.5 007786d487 Add HDR10 output
With hdr = 1 the game's tonemap shader is replaced by one that keeps highlights up to the display's peak, the present
reads the float picture the game copies into its 8-bit window texture, and a compute pass encodes it as PQ in Rec. 2020
for an HDR10 swapchain. Peak, paper white and HUD brightness are live settings in the menu and totk.ini. The approach
follows the RenoDX TotK mod by souperman9 and Musa; no code was copied.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-31 21:17:12 +01:00
SchulzCode 9e079622dc Merge main (0.2.1) into the DLSS branch
Conflicts: the known-key and section lists in cfg.c keep both sides; vk_backend.c, vk_device.c and vk_exec_draw.c include platform.h and vk_dlaa.h; create_instance takes the window's instance extensions from gfx_window_vk_extensions as on main, in a 12-entry array for NGX's.
2026-10-31 16:19:27 +01:00