The renderer's CharAnimState machine already drives player character
animations (Run=5, Walk=4, Jump, Swim, etc.) — remove the conflicting
camera controller code added in the previous commit.
Fix creature movement animations to use the correct WoW M2 IDs:
4=Walk, 5=Run. Both the per-frame sync loop and the SMSG_MONSTER_MOVE
spline callback now use Run (5) for NPC movement.
CameraController now transitions the player character to Run (anim 4)
on movement start and back to Stand (anim 0) on stop, guarded by a
prevPlayerMoving_ flag so animation time is not reset every frame.
Death animation (anim 1) is never overridden.
Application creature sync similarly switches creature models to Run (4)
when they move between server positions and Stand (0) when they stop,
with per-guid creatureWasMoving_ tracking to avoid per-frame resets.
Add GhostStateCallback to GameHandler, fired when PLAYER_FLAGS_GHOST
transitions on or off in UPDATE_OBJECT / login detection. Add
setInstanceOpacity() to CharacterRenderer to directly set opacity
without disturbing fade-in state. Application wires the callback to
set opacity 0.5 on ghost entry and 1.0 on resurrect.
Add CameraController::setSitting() and call it from the StandStateCallback
so the camera blocks movement when the server confirms the player is
sitting or kneeling (stand states 1-6, 8). This prevents the player
from sliding across the ground after sitting.
Death (state 7) deliberately leaves sitting=false so the player can
still respawn/move after death without input being blocked.
Previously disabled because the per-frame raycast caused erratic zoom
snapping at doorway transitions. Re-enable using an asymmetrically-
smoothed collision limit: pull-in reacts quickly (τ≈60 ms) to prevent
the camera from ever visibly clipping through walls, while recovery is
slow (τ≈400 ms) so walking through a doorway zooms back out gradually
instead of snapping.
Uses wmoRenderer->raycastBoundingBoxes() which already has strict wall
filters (|normal.z|<0.20, surface-alignment check, ±0.9 height band)
to ignore floors, ramps, and arch geometry.
The selection circle was positioned using the entity's game-logic
interpolator (entity->getX/Y/Z), while the actual M2 model is
positioned by CharacterRenderer's independent interpolator (moveInstanceTo).
These two systems can drift apart during movement, causing the circle
to appear under the wrong position relative to the visible model.
Fix: add CharacterRenderer::getInstancePosition / Application::getRenderPositionForGuid
and use the renderer's inst.position for XY (with footZ override for Z)
so the circle always tracks the rendered model exactly. Falls back to
the entity game-logic position when no CharacterRenderer instance exists.
Single shadow depth image shared across MAX_FRAMES=2 in-flight GPU frames
caused a race: frame N's main pass reads shadow map while frame N+1's
shadow pass clears and writes it, producing visible flashing standing
still and while moving.
Fix: give each in-flight frame its own VkImage, VmaAllocation, VkImageView,
and VkFramebuffer for the shadow depth attachment. renderShadowPass() now
indexes all shadow resources by getCurrentFrame(), and layout transitions
track per-frame state in shadowDepthLayout_[frame]. Cleanup loops over
MAX_FRAMES=2. Descriptor sets already written per-frame; updated shadow
image view binding to use the matching per-frame view.
Read the ambient color from the MOHD chunk (BGRA uint32) and store it
on WMOModel as a normalized RGB vec3. Pass it through ModelData into
the per-batch WMOMaterialUBO (replacing the unused pad[3] bytes, keeping
the struct at 64 bytes). The GLSL interior branch now floors vertex
colors against the WMO ambient instead of a hardcoded 0.5, so dungeon
interiors respect the artist-specified ambient tint from the WMO root
rather than always clamping to grey.
Hang/GPU device lost fix:
- M2_INSTANCES and WMO_INSTANCES finalization phases now create instances
incrementally (32 per step / 4 per step) instead of all at once, eliminating
the >1s main-thread stalls that caused GPU fence timeouts and device loss
M2 two-pass transparent rendering:
- Opaque/alpha-test batches render in pass 1, transparent/additive in pass 2
(back-to-front sorted) to fix wing transparency showing terrain instead of
trees — adds hasTransparentBatches flag to skip models with no transparency
Tile streaming improvements:
- Sort new load queue entries nearest-first so critical tiles load before
distant ones during fast taxi flight
- Increase taxi load radius 6→8 tiles, unload 9→12 for better coverage
Water refraction gated on FSR:
- Disable water refraction when FSR is not active (bugged without upscaling)
- Auto-disable refraction if FSR is turned off while refraction was on
Add initializeShadow() to TerrainRenderer that creates a depth-only
shadow pipeline reusing the existing shadow.vert/frag shaders (same
path as WMO/M2/character renderers). renderShadow() draws all terrain
chunks with sphere culling against the shadow coverage radius. Wire
both init and draw calls into Renderer so terrain now casts shadows
alongside buildings and NPCs.
The boneDescPool_ had MAX_BONE_SETS=2048 but sets were never freed when
instances were removed (only when clear() reset the whole pool on map load).
As tiles streamed in/out, each new animated instance consumed 2 pool slots
(one per frame index) permanently. After ~1024 animated instances created
total, vkAllocateDescriptorSets began failing silently and returning
VK_NULL_HANDLE. render() skips instances with null boneSet[frameIndex],
making them invisible — appearing as per-frame flicker as the culling pass
included them but the render pass excluded them.
Fix: destroyInstanceBones() now calls vkFreeDescriptorSets() for each
non-null boneSet before destroying the bone SSBO. The pool already had
VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT set for this purpose.
Also increased MAX_BONE_SETS from 2048 to 8192 for extra headroom.
texture.hpp / texture.cpp implemented an unfinished OpenGL texture loader
(loadFromFile was a TODO stub) that had no callers — the project's texture
loading is entirely handled by VkTexture (vk_texture.hpp/cpp) after the
Vulkan migration. Remove both files and their CMakeLists entries.
- AmdFsr3Runtime now probes both the legacy ffxFsr3* API and the newer
generic ffxCreateContext/ffxDispatch API; selects whichever the loaded
runtime library exports (GenericApi takes priority fallback)
- Generic API path implements full upscale + frame-generation context
creation, configure, dispatch, and destroy lifecycle
- dlopen error captured and surfaced in lastError_ on Linux so runtime
initialization failures are actionable
- FSR3 runtime init failure log now includes path kind, error string,
and loaded library path for easier debugging
- tools/generate_ffx_sdk_vk_permutations.sh added: auto-bootstraps
missing VK permutation headers; DXC auto-downloaded on Linux/Windows
MSYS2; macOS reads from PATH (CI installs via brew dxc)
- CMakeLists: add upscalers/include to probe include dirs, invoke
permutation script before SDK build, scope FFX pragma/ODR warning
suppressions to affected TUs, add runtime-copy dependency on wowee
- UI labels updated from "FSR2" → "FSR3" in settings, tuning panel,
performance HUD, and combo boxes
- CI macOS job now installs dxc via Homebrew for permutation codegen
- Motion vectors: single unjittered reprojection matrix (80 bytes) instead of
two jittered matrices (160 bytes), eliminating numerical instability from
jitter amplification through large world coordinates
- Sharpen pass: fix Y-flip for correct UV sampling, double-buffer descriptor
sets to avoid race with in-flight command buffers
- MSAA: auto-disable when FSR2 enabled, grey out AA setting in UI
- Accumulation: variance-based neighborhood clamping in YCoCg space,
correct history layout transitions
- Frame index: wrap at 256 for stable Halton sequence
Full FSR 2.2 pipeline with depth-based motion vector reprojection,
temporal accumulation with YCoCg neighborhood clamping, and RCAS
contrast-adaptive sharpening.
Architecture (designed for FSR 3.x frame generation readiness):
- Camera: Halton(2,3) sub-pixel jitter with unjittered projection
stored separately for motion vector computation
- Motion vectors: compute shader reconstructs world position from
depth + inverse VP, reprojects with previous frame's VP
- Temporal accumulation: compute shader blends 5-10% current frame
with 90-95% clamped history, adaptive blend for disocclusion
- History: ping-pong R16G16B16A16 buffers at display resolution
- Sharpening: RCAS fragment pass with contrast-adaptive weights
Integration:
- FSR2 replaces both FSR1 and MSAA when enabled
- Scene renders to internal resolution framebuffer (no MSAA)
- Compute passes run between scene and swapchain render passes
- Camera cut detection resets history on teleport
- Quality presets shared with FSR1 (0.50-0.77 scale factors)
- UI: "Upscaling" combo with Off/FSR 1.0/FSR 2.2 options
Root cause: bonesDirty was a single bool shared across both
double-buffered frame indices. When bones were copied to frame 0's
SSBO and bonesDirty cleared, frame 1's newly-allocated SSBO would
contain garbage/zeros and never get populated — causing animated
M2 instances to flash invisible on alternating frames.
Fix: Make bonesDirty per-frame-index (bool[2]) so each buffer
independently tracks whether it needs bone data uploaded. When
bones are recomputed, both indices are marked dirty. When uploaded
during render, only the current frame index is cleared. New buffer
allocations in prepareRender force their frame index dirty.
Add 2-second cooldown timer before re-checking for unloaded tiles
when workers are idle, preventing excessive streamTiles() calls
that caused frame hitches right after world load.
- Overlap M2 and character animation updates via std::async (~2-5ms saved)
- Thread-local collision scratch buffers for concurrent floor queries
- Parallel terrain/WMO/M2 floor queries in camera controller
- Seed new M2 instance bones from existing siblings to eliminate pop-in flash
- Fix shadow flicker: snap center along stable light axes instead of in view space
- Increase shadow distance default to 300 units (slider max 500)
- Normal map CPU work (luminance→blur→Sobel) moved to background threads,
main thread only does GPU upload (~1-2ms vs 15-22ms per texture)
- Load screen warmup now waits until ALL spawn/equipment/gameobject queues
are drained before transitioning (prevents naked character, NPC pop-in)
- Exit condition: min 2s + 5 consecutive empty iterations, hard cap 15s
- Equipment queue processes 8 items per warmup iteration instead of 1
- Added LoadingScreen::renderOverlay() for future world-behind-loading use
Each loadTexture call was generating a normal/height map inline (3 full-image
passes: luminance + blur + Sobel). For models with 15-20 textures this added
30-40ms to the 70ms model upload. Now deferred to a per-frame budget (2/frame
in-game, 10/frame during load screen). Models render without POM until their
normal maps are ready.
Move CPU-heavy BLP texture decoding from main thread to background worker
threads for all hot paths: terrain M2 models, WMO doodad M2s, WMO textures,
creature models, and gameobject WMOs. Each renderer (M2, WMO, Character) now
accepts a pre-decoded BLP cache that loadTexture() checks before falling back
to synchronous decode.
Defer WMO normal/height map generation (3 per-pixel passes: luminance, box
blur, Sobel) during terrain streaming finalization — this was the dominant
remaining bottleneck after BLP pre-decoding.
Terrain streaming stalls: 1576ms → 124ms worst case.
- Replace per-frame VMA alloc/free of material UBOs with a ring buffer in
CharacterRenderer (~500 allocations/frame eliminated)
- Batch all ready terrain tiles into a single GPU upload during load screen
(processAllReadyTiles instead of one-at-a-time with individual fence waits)
- Lift per-frame creature/GO spawn budgets during load screen warmup phase
- Add background world preloader: saves last world position to disk, pre-warms
AssetManager file cache with ADT files starting at app init (login screen)
so terrain workers get instant cache hits when Enter World is clicked
- Distance-filter expensive collision guard to 8-unit melee range
- Merge 3 CharacterRenderer update loops into single pass
- Time-budget instrumentation for slow update stages (>3ms threshold)
- Count-based async creature model upload budget (max 3/frame in-game)
- 1-per-frame game object spawn + per-doodad time budget for transport loading
- Use deque for creature spawn queue to avoid O(n) front-erase
- Worker threads: use (cores - 1-2) instead of cores/2, minimum 4
- Outer upload batch in processReadyTiles: ALL model/texture uploads per
frame share a single command buffer submission + fence wait
- Upload multiple models per finalization step: 8 M2s, 4 WMOs, 16 doodads
per call instead of 1 each (all within same GPU batch)
- Terrain chunks: 64 per step instead of 16
- Skip redundant M2 file I/O: thread-safe uploadedM2Ids_ set lets
background workers skip re-reading+parsing models already on GPU
- processAllReadyTiles (loading screen) and processOneReadyTile also
wrapped in outer upload batches
