* fix FastDebug QtFRED linking the debug CRT
FastDebug builds in the DLL-runtime tree (used whenever qtFRED is
enabled) compiled with /MDd and, since #7450, linked Qt's debug
libraries. Both import ucrtbased.dll, MSVCP140D.dll and
VCRUNTIME140D.dll, which Microsoft does not license for redistribution,
so end users could not run the FastDebug qtFRED shipped in nightlies
and releases. Static Qt is ruled out by the LGPL, so the only shippable
shape is the release CRT plus release Qt.
On MSVC, _DEBUG is the CRT's own contract: /MTd and /MDd define it
themselves, and the STL headers select msvcprtd.lib and
_ITERATOR_DEBUG_LEVEL=2 from it. Stop force-defining it for FastDebug
on MSVC; the engine keys its debug behaviour on NDEBUG, and the few
_DEBUG guards in code/ are switched to NDEBUG so they keep working.
The static tree is unchanged: /MTd still defines _DEBUG there.
Fixes#7915
Co-Authored-By: Claude Fable 5.1 <[email protected]>
* derive the MSVC runtime choice once
The "DLL runtime or not" condition was spelled out in the toolchain and
twice in fred2 with "keep in sync" comments, and gtest_force_shared_crt
and antlr4's WITH_STATIC_CRT had already drifted: they only looked at
MSVC_USE_RUNTIME_DLL, so a qtFRED tree with tests enabled would have
asked both for the static CRT. That is harmless today only because our
flag strings carry no literal /MD for them to rewrite.
Compute FSO_MSVC_RUNTIME_DLL once in toolchain-msvc.cmake and use it
everywhere. InstallRequiredSystemLibraries deliberately stays keyed on
MSVC_USE_RUNTIME_DLL: the qtFRED deploy ships no compiler runtime.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
* remove FastDebug runtime leftovers
The /MD -> /MDd substitution on the FastDebug flag strings has done
nothing since the runtime moved to CMAKE_MSVC_RUNTIME_LIBRARY: the
Release flags it copies contain no runtime switch. The commented-out
_ITERATOR_DEBUG_LEVEL=0 lines for gtest and the OpenXR loader belong to
the scheme #7450 removed from the toolchain.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
* use the release CRT for all FastDebug builds
Microsoft's documentation marks the static debug libraries (libcmtd,
libucrtd, libvcruntimed) "Not redistributable" and states that debug
versions of an application may only be copied to machines used for
testing, and the license grants distribution rights only for code on
the REDIST list, which contains no debug library in any form. So the
statically linked FastDebug game and FRED2 are no more shippable than
the /MDd qtFRED was.
Link every FastDebug build against the release runtime. The engine's
debug behaviour keys on NDEBUG, which FastDebug still leaves undefined,
so asserts, warnings and mprintf are unaffected; only the debug CRT's
heap checking goes, and that remains available in Debug builds. FRED2's
static MFC library follows the CRT: the debug variant requires the debug
runtime.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
* keep the debug icon, version flag and FRED2 asserts without _DEBUG
The resource compiler only ever sees the explicit definitions, so with
_DEBUG no longer defined for FastDebug the game lost its debug icon and
both editors lost the VS_FF_DEBUG version flag. Key those on NDEBUG,
which FastDebug still leaves undefined.
FRED2 also used MFC's ASSERT, VERIFY, ASSERT_VALID and TRACE0, which are
all compiled out without _DEBUG. Convert them to the FSO equivalents
FRED2 already uses nearly everywhere else: Assert, Verify, a null check
in place of ASSERT_VALID (AssertValid only exists in debug MFC), and
mprintf. The AssertValid/Dump overrides and DEBUG_NEW stay under
_DEBUG, as they must.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
---------
Co-authored-by: Claude Fable 5.1 <[email protected]>
* take a model reference when a ship changes class
ship_model_change() only loaded the model when the class had none, so a ship
switched to a class whose model had been loaded only by a select-screen
preview adopted that preview's single reference. Closing the screen then
freed the model from under the ship, and the first simulation frame asserted
in model_get(). This is what happened on a replayed mission with a restored
loadout: ship select was skipped, the weapon-select overhead loaded the new
class once, the commit switched the player to it, and the screen close
released it.
Load unconditionally, as ship_create() already does, so the ship holds its own
reference.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
* reject stale model ids in model_unload and release select-screen handles at level close
model_unload only used the slot part of the id it was given, so a handle kept
across missions would free whatever model had since been loaded into that slot.
Validate the id against the slot's occupant and warn instead.
Two holders could present such handles: the weapon-select icon models under a
scripted UI that overrides the state hooks (the native close never runs), and
the ship-select 3D model global, which was never reset between missions.
Release the select-screen handles from game_level_close, and unload the
ship-select model on screen close and on every selection change.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
* consolidate the select-screen resource release
The weapon side released its icons, anims, and ship class data from two
functions keyed on whether the screen was open, and the level-close path added
a third copy. Give the weapon side the same shape as the ship side: one
unconditional weapon_select_common_close(), with weapon_select_close() left as
pure UI teardown. common_select_level_close() now just calls the two common
closes, and common_select_close() calls it in turn. The ship-select 3D model
is released from ship_select_common_close() so both paths cover it.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
---------
Co-authored-by: Claude Fable 5.1 <[email protected]>
The failure checks after parsing (bad detail levels, debris objects or
parents, and too many textures) run after the TXTR and GLOW chunks have
already loaded bitmaps, but model_load only deleted the polymodel, so
those bitmap slots were never released. Page out the textures and free
the model through model_free, as the virtual POF cache already does for
models it loads.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Retail models (111 of them in sparky_fs2.vp) store -1 as their cross
section count, meaning none, which the original loader silently accepted.
Routing the count through pof_read_count made every such model warn on
load. Read it directly and floor it at zero, as before.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A virtual POF ignored the load status of its base POF, so a base that
failed to load (missing, bad version, or rejected after parsing) was
still built upon and reported as a successful load. Likewise, every
operation that copies from another POF used it even if that POF had
failed to load.
Return FAIL from read_virtual_model_file when the base POF fails, and
have the operations skip, with a warning, any source POF that failed.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
When appending submodels would take the combined POF past
MAX_MODEL_TEXTURES, the limit was only noticed partway through, after the
submodel array had been grown, subsystems copied, and n_textures
incremented past the limit, which later loops over maps[] then overran.
Work out the texture mapping first, and give up before touching the model.
Also skip the -1 texture id of untextured flat polygons, which was being
assigned a texture slot, and which replace_textures_used then used to
index Num_verts and Num_polies.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Clamping the TXTR count keeps maps[] safe, but texture indices are baked
into the BSP data, so a model declaring more than MAX_MODEL_TEXTURES
textures cannot be built correctly. Keep the clamp so that the chunk is
still read in full, then fail the load after parsing.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
process_tmap and process_tmap2 used the texture index from the BSP data
to index Num_polies and Num_verts without checking it, so a corrupt
model wrote past both arrays while its vertex buffers were built. The
collision tree parser already treats such an index as a fatal error, but
it runs afterward. Raise the same error here, and skip the polygon in
case the error is continued past.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The post-parse sanity checks and the bad-version check returned FAIL
without closing the POF. Nothing reads the file after the chunk loop, so
close it there, and close it before the bad-version return too.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
light_apply_rgb works in the model's local frame, but nothing rotated the
lights into that frame before insignias were drawn: light_rotate_all() was
only called for the briefing and tech room lights, and even then it skipped
Static_light, the separate copies of the sun lights that light_apply_rgb
reads. So the sun term used a stale, never-set direction and the point
lights were treated as if they sat at the model's origin.
Rotate the lights, including the sun copies, once the insignia's instance
is set up, and light each vertex at its actual position, which includes
the insignia's offset.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
In debug builds, subsystems found in a model but missing from ships.tbl
are meant to be dumped to data/tables/<model>.subsystems as stub table
entries. Since subsystem matching was deferred until after the model
file is read (#4590), the dump file was already closed by the time
do_new_subsystem ran, so it was always empty and deleted. Worse, ss_fp
was left pointing at a closed CFILE slot, which another file could reuse
before do_new_subsystem wrote to it, and the dump file leaked whenever a
load failed.
Open the dump file on the first unmatched subsystem instead, and close it
once the deferred subsystems have been matched. This also means no file
is created and deleted for models without unmatched subsystems. Reset
the once-per-model warning flags there as well, rather than only when
the dump file opened successfully.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The insignia normal array was declared with the vertex limit, filled in by
face index, and read by vertex index. An insignia with more than 81 faces
therefore wrote past the array, into the following insignia or, for the
sixth and final slot, past the insignia list into the rest of the
polymodel. And the renderer lit each vertex with the normal of whichever
face shared its index, or with a zero normal (ambient light only) when
there was no such face.
A normal belongs to a face, so size the array by the face limit and light
all three corners of each face with that face's normal.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A subobject that named itself, or one of its own descendants, as its parent
made the depth walk loop forever, and would have done the same to every
later walk up the hierarchy. Since each SOBJ chunk adds only one link, any
cycle must lead back to the subobject being read, so detect that during the
walk and treat the subobject as parentless.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Element counts were taken straight from the file, so a corrupt POF could
supply a negative count (stored alongside a null array) or one larger than
the fixed-size array it feeds. Add pof_read_count() to floor every count at
zero and clamp it to the destination's capacity, reading past the discarded
entries where more data follows in the same chunk.
The upper bounds that were checked at all were checked with debug-only
Asserts, or with an Error() that returns if the user picks Debug, so the
overflowing loop after it was reachable in both cases.
Also validate the submodel indices in SOBJ and TGUN/TMIS, the vertex indices
in insignia faces, and bounds-check the split plane array before writing to
it rather than after.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
A 3D shockwave's glow map already takes its frame from the shockwave's
elapsed time so the animation spans the shockwave's life. Do the same
for an animated base texture, and skip the radius fade for it, since the
artist's frames then control the brightness the way they do for the glow.
A static base texture still fades with the radius.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
3D shockwaves pass `1 - radius/outer_radius` as the warp alpha, which `model_render_determine_color` turns into the material colour for additive draws. Since a481999a2 (SamuelCho's OpenGL Code commit in August 2016), the model fragment shader has not multiplied the diffuse sample by that colour in the unlit branch, so the base texture stayed at full brightness and vanished abruptly. The same path carries the warp effect's radius fade and the alpha of unlit transparent weapons.
Clamp the alpha before converting it to a byte, as the pre-2016 code did; thrusters pass 1.2, and without the clamp they would now render dim.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
swap_bsp_sortnorm2 swapped the bounding box at offsets 8 and 20, which
re-swapped the front list offset and left the real box (at 16 and 28)
unswapped. swap_bsp_tmap2poly never swapped the chunk's bounding box
or normal. Both corrupted models using these chunks on big-endian
builds.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Report leaks after static destruction via _CRTDBG_LEAK_CHECK_DF instead of calling _CrtDumpMemoryLeaks() at the end of main(), which incorrectly reported every global container as a memory leak.
Also fix the three remaining real leaks which were discovered after all the noise was cleared up:
- Mc_point_list was never freed, because its atexit cleanup was only registered by code that no longer runs. But the list was only a temporary array of pointers used while building the collision tree, so remove it and have model_collide_parse_bsp_defpoints() copy the vertices directly into the tree's point list.
- model_batch_buffer's Mem_alloc copy was never freed. But the copy is actually not necessary, so remove it and pass Submodel_matrices directly.
- Goal_target_names strings were only freed when the next mission was parsed, so the last mission's names leaked at exit. Hold them in SCP_vm_unique_ptr so they are freed with the vector.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Every copied bank was written to the same array slot, so all but the last
were lost, and the slots left behind kept their zero-initialized contents.
A zeroed bitmap handle is slot 0 rather than -1, so the paging code would
then page out an unrelated bitmap.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Restore the original flag values that were consolidated in #7588. Since the flags are saved in the mission file as an integer, the original values must be preserved.
Since `MR_ATTACHED_MODEL` was removed in the original PR, that is renamed to `MR_EMPTY_SLOT6`.
Repro: create a prop, File -> New Mission & Discard, then create the same prop
again. Asserts in model_get with "Index collision between model <other>.pof and
requested model N".
model_unload() walks Ship_info and Weapon_info clearing any cached model_num
that pointed at the model being freed, but never did the same for Prop_info.
props_level_close() only clears the prop instances, so prop_info::model_num
survived the discard pointing at a freed slot. Placing that prop class again hit
EditorViewport::create_object_on_grid, which reads pip->model_num directly to
size the placement distance, and by then the slot had been reused by an
unrelated model.
Adds the matching Prop_info pass. Doing it in model_unload rather than
props_level_close mirrors how the two closest analogues are handled -- Ship_info
and Weapon_info are likewise class tables holding a cached handle for placeable
objects -- and covers any unload path, not just level close. (asteroid_level_close
solves the same problem for Asteroid_info from the other direction.)
This is not OFP-specific; SCP master has props and the same missing pass.
* Fix C4709 Warnings on latest VS
* more
* suppress false positive
* Switch to using any/none_of
* Address Feedback
* Fix Line endings?
* Fix commits and strip out flag_combinator
* more fixes
* nolint
* Shadow overhaul round 2: PCSS, dynamic cascades, cockpit/instanced shadows
Squashed snapshot of lafiel/shadow_overhaul_2's continuation work
(d3922aa87..010f9d2d9), which has no separate upstream PR of its own.
* Add Vulkan rendering backend
Squashed snapshot of the Vulkan backend as vendored into this branch
(ce2d1e85f..6ce539463, authored by Mara van der Laan with build/CI
fixes by Taylor Richards). This work has its own upstream PR #7553
(notimaginative:vulkan-pr-new) and is not otherwise original to true-hdr.
* Add true HDR output, RT shadows, and PCSS shadows for Vulkan
Squashed snapshot of true-hdr's own integration work: true HDR
framebuffer output, wiring the shadow overhaul into the Vulkan
backend (raytraced + cascaded shadow maps), logging refactors, log
filter UI, and CI/build fixes.
* refactor: update Vulkan logging calls to use `nprintf` for scoped logging
* fix: correct shadow cascade count in Vulkan logging calls
* fix: fall back to PATH-resolved ccache when CCACHE_PATH is stale
CI hardcodes /usr/local/bin/ccache, but macOS runners on Apple Silicon
hosts install Homebrew (and ccache) under /opt/homebrew instead,
causing all four Mac configs to fail with "No such file or directory".
* Fix vulkan texture readback to correctly handle transparent pipelines
* Put the pixel swizzles into a util file
* OOB safety check
* Fix fog handling
* Fix MSAA in Vulkan
* fix: enable SMAA for vulkan
* adapt the post-processing AA shaders to hdr
* fix clang-tidy warnings: concatenate nested namespaces, static methods
pixel_swizzle.{h,cpp}: collapse `namespace graphics { namespace util {`
into `namespace graphics::util {` (modernize-concat-nested-namespaces).
VulkanState.h/VulkanDraw.h: getClearColor()/setClearColor() only touch
gr_screen and other globals, no instance state, so make them static
(readability-convert-member-functions-to-static) and drop the
top-level const on getClearColor()'s return type
(readability-const-return-type).
* fix clang-tidy warning
* Extend raytraced shadows to all directional lights
RT shadows previously only shadowed the first directional light in the
scene, even though multi-sun missions push more than one directional
light into the engine. The TLAS raytraced shadows query against is
already light-agnostic, so this extends both the deferred lighting
pass and the forward MODEL shader to shadow every directional light
(capped by a new Max Raytraced Shadow Lights option, configurable via
the in-game Options menu or the Lab's Scene rendering options).
Cascaded shadow maps are untouched: they're built for a single light
only, so the deferred pass still limits itself to one shadowed
directional light whenever CSM (not raytracing) is the active method.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016DA6JHJ3twz2oHGQcZ1Udc
* refactor: improve shader flag handling and render target finalization in Vulkan backend
- Consolidate shader flag setup logic in VulkanPostProcessingLighting to simplify configuration.
- Fix dead field usage in VulkanDraw by ensuring correct image view selection for environment maps.
- Add mip chain regeneration in VulkanRenderer::endRenderTarget to ensure correct mip levels for cubemaps.
* fix: ensure correct mip chain initialization in cubemap rendering
- Add mip chain generation to VulkanDraw for cubemaps with multiple mip levels.
- Fix potential issue with uninitialized mip levels above level 0 in deferred-f.sdr sampling paths.
* fix: don't crash on startup when environment locale is unavailable
std::locale("") pulls in the process's environment locale categories
(LC_CTYPE, LC_MONETARY, LC_PAPER, etc.), and throws std::runtime_error
if any of them names a locale not installed on the host (e.g.
mismatched regional-format overrides). Since SCP_default_locale is
constructed during static initialization, before main() and before any
of our own exception handling exists, an uncaught throw here aborted
the whole process before the game even started. Fall back to the
classic "C" locale instead, which is sufficient since this locale is
only ever used for ASCII case-folding (SCP_toupper/tolower/isdigit).
Co-Authored-By: Claude Sonnet 5 <[email protected]>
* `feat: add raytraced shadows for local lights and optimize TLAS management`
* Make the Commandline authoritative for -res and -vulkan / -opengl
* address BMagnu's comments on the original PR
* update prebuilt version hash to 62c89ef6
* refactor: standardize graphics API handling and document Vulkan limitations in qtFRED
- Updated `AboutDialogModel.cpp` to use `GraphicsAPI` enum for better clarity.
- Ensured qtFRED falls back to OpenGL when Vulkan is requested.
- Added detailed documentation in `qtfred/README.md` regarding Vulkan backend limitations and requirements.
* chore: include cmdline library in main and mission management modules
* docs: add known issue for blank render viewport under Wayland in README.md
* fix CI issues
* chore: add `GR_DEBUG_SCOPE` markers across Vulkan post-processing and rendering paths for improved debugging
* feat: improve Vulkan HDR pipeline with separate encode paths for SDR and HDR10
- Added an HDR10-specific output encode pass in `VulkanRenderer` and `VulkanPostProcessor` that applies PQ/BT.2020 transfer.
- Introduced an optimized SDR fallback path with direct blit support for devices that can skip shader draws.
- Updated `m_sceneLdrCompressed` to provide [0,1]-clamped proxies for HDR edge detection.
- Refactored `copyImageToImage` and added `blitImageToImage` to allow format conversions and optimized transfers.
- Replaced `hdr_mode` with explicit HDR10 flags and data structures for clearer state handling.
* refactor: encapsulate SMAA processing and refactor VulkanPostProcessor
- Moved SMAA-specific logic into `VulkanSMAA` class for better separation of concerns.
- Simplified `initSMAA`, `shutdownSMAA`, and `executeSMAA` logic by leveraging `VulkanLDR` and `PostProcessContext` abstractions.
- Improved initialization and cleanup code with clearer resource lifecycle management.
- Updated debug markers and messages to reference `VulkanSMAA`.
- Enhanced flexibility of fullscreen triangle rendering API with additional parameters (`sampleCount`, `bindGlobalSet`).
* Update Vulkan porting credits in AboutDialogModel
* fix CI issues again
* fix: correct Y-flip in viewport setup for Vulkan light volumes
- Adjusted viewport height to negative value to account for G-buffer render pass's negative Y-axis convention.
- Ensured proper alignment of point and tube light volumes with actual light positions.
* feat: add user-adjustable gamma correction to SDR and HDR10 output paths
- Added gamma correction to HDR10 path, applying user settings before PQ/BT.2020 encode.
- Replaced the SDR blit-only path with an encode pass that applies gamma correction, ensuring consistent behavior across devices.
- Refactored VulkanPostProcessor and associated shaders to support gamma slider adjustments for both SDR and HDR output.
* fix: ensure proper uniform and cull mode setup for Vulkan sphere rendering
- Added uniform buffer setup and default material uniforms in `vulkan_draw_sphere`.
- Disabled face culling in material definition for sphere rendering to align behavior with OpenGL.
* Ensure memory flushes after UBO updates and add explicit pipeline barrier for MoltenVK compatibility.
* fix: add clear pass for SMAA edge detection to prevent persistent artifacts
* feat: add Vulkan debug-utils messenger, opt-in sync validation, frame diagnostics
Groundwork for the renderer-review fix series (.agents/plans/
vulkan-renderer-review-fixes.md, Phase 0):
- Register a VK_EXT_debug_utils messenger (preferred) instead of only the
deprecated VK_EXT_debug_report callback, which is kept as a fallback for
loaders without debug-utils. The messenger config is also chained into
instance creation so vkCreateInstance-time messages are captured.
Validation errors/warnings now go to the main log via mprintf.
- New -gr_sync_validation cmdline flag (implies -gr_debug): chains
VkValidationFeaturesEXT with synchronization validation into instance
creation when VK_LAYER_KHRONOS_validation is available. This is the
regression net for the synchronization fixes in the following commits.
- Frame stats now report descriptor sets/writes allocated per frame and the
total pipeline count (descriptor-manager counters; DescriptorWriter::flush
moved out-of-line to report into the manager).
Co-Authored-By: Claude Fable 5 <[email protected]>
* refactor: unify Vulkan render-pass begins, clamp in-pass clears, add per-frame UBO ring
Phase 1 of the renderer-review fix series: shared helpers that the
correctness fixes build on.
- New VulkanRenderer::beginTrackedRenderPass(PassBeginDesc): every render-pass
begin on the frame command buffer now records vkCmdBeginRenderPass and
updates the state tracker (pass, color attachment count, sample count,
render area, viewport) in one place. Converted: setupFrame,
beginSceneRendering (both branches), resumeSceneRendering,
endSceneRendering, beginRenderTarget (viewport kept for gr_set_viewport),
resumeSwapChainPass, plus new resumeScenePassAfterCopy() shared by
copyEffectTexture/copySceneDepthForParticles and
rebeginSwapChainPassAfterReadback() shared by both readbackFramebuffer
exits (~200 lines of duplicated begin/clear/viewport blocks removed).
- The state tracker records the active render area; all four
vkCmdClearAttachments sites clamp their rects against it and skip empty
results. Previously gr_screen-sized clear rects could exceed a smaller
off-screen render target's render area, which is invalid API usage.
- New PerFrameUboRing utility (VulkanPerFrameUbo.h/.cpp): a persistently
mapped uniform buffer with one slot region per frame in flight and a
self-flushing alloc(). Used by the next commit to fix the cross-frame
UBO races.
Deliberate behavior deltas (all safe-direction): the helper always sets
colorAttachmentCount/sampleCount (some sites previously inherited stale
values), and resumeSwapChainPass/resumeSceneRendering now also reset the
standard flipped viewport instead of leaving the previous one.
Co-Authored-By: Claude Fable 5 <[email protected]>
* fix: eliminate Vulkan cross-frame UBO/buffer races, make gr_sync honest
Phase 2 of the renderer-review fix series: with two frames in flight, several
host-visible resources were rewritten by the CPU while the previous frame's
GPU work could still be reading them.
- The post-processing scratch UBO is now a PerFrameUboRing (per-frame-in-
flight slot regions, persistently mapped, self-flushing alloc). All the
per-pass map/unmap/cursor bookkeeping in bloom/LDR/fog/SMAA is deleted;
the cursor resets exactly once per frame via the new
VulkanPostProcessor::beginFrame() hook in setupFrame.
This also fixes two latent bugs: (a) fog consumed scratch slots mid-scene
and bloom's later same-frame cursor reset overwrote them before submission
whenever fog+bloom were both active; (b) volumetric_fog_data (288 bytes)
overflowed the 256-byte slots - the memcpy spilled into the neighboring
slot and the shader's bound range ended before the aspect/fov/noise
fields. Slots are now 512 bytes (static_assert at the fog call site) and
the per-frame budget is 32 slots.
- The dedicated tonemap, output-encode, and MSAA-resolve UBOs (each a single
instance rewritten per frame, i.e. racy; the MSAA one also never flushed)
are deleted entirely - those passes build their structs on the stack and
allocate from the scratch ring.
- Static-buffer rewrites orphan instead of racing: getVkBuffer() stamps a
last-used frame number, and a full-content updateBufferData() on a Static
buffer used within the last MAX_FRAMES_IN_FLIGHT frames defers the old
VkBuffer to the deletion queue and writes a fresh one. Offset writes
(GPU-heap appends to fresh regions) and PersistentMapping buffers (the
engine holds their mapped pointer and syncs via gr_sync) are exempt.
- gr_sync_wait now honors its contract: VulkanRenderFrame::waitForFinish
takes a timeout and reports expiry without consuming the fence;
VulkanRenderer::waitForFrame returns false for fences taken during the
still-recording frame (previously it waited on a stale or idle slot and
reported success) and resolves the frame->slot mapping correctly. This
makes UniformBufferManager's segment-fence throttling actually protective
on Vulkan for the first time.
Co-Authored-By: Claude Fable 5 <[email protected]>
* Refactor framebuffer and target lifecycle: centralized creation/resizing logic, improved resource management across Vulkan post-processing subsystems.
* Improve Vulkan cross-frame sync with explicit access masks and validation updates
- Unified Vulkan validation flag handling in setup, separating engine debug output from synchronization validation.
- Specified access masks and stage flags to address cross-frame WRITE_AFTER_WRITE vulnerabilities, notably in G-buffer, post-processing, and MSAA resolve paths.
- Adjusted subpass dependencies and image layout transitions to prevent read-after-write and write-after-write hazards, detected using `-gr_sync_validation`.
- Fixed uniform buffer and barrier transitions to ensure proper ordering in render passes across frames.
* Refactor Vulkan render-pass dependencies and sync stages for clarity and future testing
- Clarified dependency logic for cross-pass and cross-frame synchronization, ensuring explicit stage and access flag comments match observed behavior.
- Adjusted subpass dependencies to decouple render-pass compatibility definitions from dependency assumptions.
- Highlighted MoltenVK-specific issues and retained workarounds with notes for re-testing on macOS hardware.
- Updated synchronization barriers and validation notes to address historical inconsistencies across platforms.
* Remove transfer queue support; consolidate all operations into the graphics queue
- Removed transfer queue handling, simplifying buffer and queue logic to use the graphics queue for all transfer operations.
- Updated buffer sharing modes and removed related code paths and parameters for transfer queue families.
- Introduced exclusive buffer access, noting async transfer and queue ownership transfers are future work.
- Limited enabled device features to only those actively used by the renderer, reducing unnecessary feature requests (e.g., robustBufferAccess).
* Refactor frame advancement and descriptor management in VulkanRenderer
- Removed `endFrame()` from `VulkanDescriptorManager`, replacing it with `setCurrentFrame(frameIndex)` for explicit frame synchronization.
- Centralized frame index advancement within `VulkanRenderer` to ensure consistency across components.
- Invalidated cached descriptor set bindings at render-pass boundaries to handle raw recorder staleness correctly.
- Optimized pipeline descriptor rebind logic by removing redundant clears during pipeline changes.
* Refactor `PipelineConfig::hash` and Global descriptor set management for improved hashing and memoization efficiency
- Replaced inefficient shift-based hashing in `PipelineConfig::hash` with a robust `hashCombine` strategy to minimize bucket collisions.
- Introduced `maskBits` helper to streamline bitmask creation for boolean fields.
- Implemented memoization for Global (Set 0) descriptor sets, reducing redundant rebuilds across frames and improving performance.
- Added dirty state tracking (`m_globalSetDirty`) and invalidation logic for Global set updates following relevant input changes (e.g., TLAS or UBO modifications).
* Refactor Vulkan post-processing output logic: extracted shared `encodeToSwapChainPass` method to streamline HDR and SDR output paths.
* Optimize Vulkan descriptor set management with per-frame memoization
- Added memoization for Material (Set 1) and PerDraw (Set 2) descriptor sets to reuse cached sets across frames if inputs remain unchanged.
- Implemented precise input tracking for Material and PerDraw descriptor sets to detect and rebuild only when necessary.
- Reduced redundant Vulkan descriptor allocations and writes, improving performance on repeated or batched draw calls.
* Fix Vulkan access masks for G-buffer and scene pass transitions
- Added `eColorAttachmentRead` to access masks in multiple barriers to ensure proper synchronization for load operations (e.g., `loadOp=eLoad`) across render passes.
- Prevented `READ_AFTER_WRITE` synchronization validation issues in deferred and multi-pass rendering scenarios.
- Updated comments to clarify the reasoning behind the access mask changes and load dependencies.
* Refactor Vulkan resource management and indexed draw logic
- Consolidated destruction queue methods in `VulkanDeletionQueue` using a shared `queueHandle` template to reduce redundancy.
- Introduced `ModelDrawParams` structure and `computeModelDrawParams` function to unify indexed draw parameters for models and shadows.
- Simplified draw logic by eliminating repeated calculation of indexed draw parameters and improving code reuse.
* Add `releaseAnimationSlotRef` for ref-count handling of shared animation texture arrays
- Introduced `releaseAnimationSlotRef` to manage reference counting for shared animation texture-array slots, preventing premature destruction of shared resources.
- Updated `flushTextures` and related methods to integrate `releaseAnimationSlotRef` for shared slot cleanup.
- Refactored staging buffer creation into `createStagingBuffer` for 3D texture uploads and shared path consistency.
- Enhanced texture upload logic with improved tracking and logging for re-upload cases, aiding preloading gap analysis.
- Added instrumentation for on-demand mid-frame texture uploads in `FrameStats`.
* Refactor Vulkan resource and texture management, extend debug logging
- Updated smart pointer usage for consistency across Vulkan resource managers (`std::make_unique`).
- Removed redundant `cubeImageView` member and related cleanup logic; unified sampling view management.
- Added per-instance debug log counters (avoiding static state) for Vulkan texture uploads and bindings, improving visibility during initialization and run-time.
- Optimized `VulkanRenderer` property/feature getters by caching physical device properties and features during device creation.
- Improved memory region flushing logic for out-of-date offsets, aligning with Vulkan Memory Allocator behavior.
- Refactored render pass state tracking to reapply dynamic state reliably after mid-frame render passes.
- Enhanced validation and error handling for mismatched update texture formats.
* fix CI build warning
* Fix on-demand YUV texture locking and set alpha uniform for movie rendering in Vulkan
- Corrected bitmap lock type from `BMP_TEX_OTHER` to `BMP_AABITMAP` for YUV textures during on-demand loading.
- Updated movie rendering to bind alpha from the `MovieData` UBO (set 2, binding 4) for Vulkan, aligning behavior with OpenGL implementation.
* revert inadvertent changes introduced by a squash commit
* address ci issues
* remove wip stuff from code comments that make no sense to people who aren't me
* Remove unused LD_LIBRARY_PATH export and fix shader shadow map sampling.
* Refactor batched rendering uniform setup in Vulkan
- Removed unused fields from `genericData` uniform block, simplifying shader definitions.
- Updated Vulkan material uniform handling to align with OpenGL's batched rendering logic.
- Introduced `batched_data` struct for streamlined UBO updates, improving maintainability and clarity.
* Add `-hdr` command-line option for HDR nits configuration
- Introduced `-hdr` parameter to set paper white and peak luminance in nits, overriding in-game HDR settings.
- Added input validation with error feedback for incorrect format.
* Remove unused `MODEL_SDR_FLAG_SHADOW_MAP` macro from shader flags
* clean up unrelated changes
* Invalidate Vulkan state tracker bindings after external command buffer modifications
- Added `invalidateExternalBindings()` to `VulkanStateTracker` to handle stale pipelines, descriptors, and dynamic states caused by external Vulkan commands (e.g., ImGui renders mid-pass).
- Updated ImGui rendering flow to call `invalidateExternalBindings()` post-draw, ensuring consistent pipeline state for subsequent engine renders.
- Adjusted image usage flags to account for `eTransferSrc` in texture readbacks.
- Fixed barrier `dstAccessMask` and pipeline stage flags to ensure proper synchronization for image transitions.
* Add a few more generically useful lines to the log
* Simplify Qt deployment logic and improve handling for system-installed Qt on Linux
* add cmdline headers to make qtFred compile
* use VK_KHR_synchronization2 for sync and barrier purposes
* Simplify and clean up Qt deployment logic in CMake for qtfred
* update get_bitmap_from_texture to match the new API
* Enhance `-hdr` command-line option to force HDR output and support optional nits configuration with improved input validation and overrides.
* Improve Vulkan robustness, add device feature checks, and refine memory/logging.
- Added depth clamping support gated by device feature checks, with fallback where unsupported.
- Improved deferred lighting UBO overflow handling with user-accessible warnings and buffer size adjustments.
- Enhanced Vulkan logging for validation, allocation, and debugging:
- Automatic logging for descriptor pool exhaustion/fragmentation.
- Vertex format and resource destruction checks.
- Notifications for generic layout transitions and volumetric texture uploads.
- Consolidated shader logic to ensure Vulkan and OpenGL parity.
* Remove HDR10 encode shader and consolidate gamma correction logic for SDR and HDR10. Replace `Verify` assertions with `Assert`
* Add braces to conditional in object lighting logic for clarity
* Refine Vulkan memory allocation checks with `Verification` for improved error handling and messaging.
* Refine VulkanBuffer resize logic to preserve contents with CPU-side copy and add safeguards for non-coherent memory scenarios.
* Add detailed comments explaining descriptor array capacity constants in VulkanDescriptorManager
* Gate shadow support on depth clamp feature availability and clarify its necessity with improved comments.
---------
Co-authored-by: BMagnu <[email protected]>
Co-authored-by: Mara van der Laan <[email protected]>
Co-authored-by: Claude <[email protected]>
Ship's implicit move operations compiled but were semantically broken in
two ways: moving copied the subsys_list sentinel's links verbatim,
leaving the bookend nodes of the intrusive subsystem list pointing at
the old sentinel address; and the raw owning warpin_effect /
warpout_effect pointers were copied without nulling the source, so a
vacated Ships[] slot aliased live WarpEffects that ship::clear() (on
slot reuse) or ship_close() (which walks every slot) would delete out
from under the destination ship.
Fix both at the member level so the compiler-generated moves are
correct by composition:
- Add ship_subsys_sentinel, derived from ship_subsys, whose move
operations re-link the bookend nodes (handling the null-linked,
self-referential-empty, and populated source states) and leave the
moved-from sentinel as an initialized empty list. Copying is
deleted. Public inheritance keeps &ship::subsys_list convertible to
ship_subsys*, so list_init, GET_FIRST, and every other call site is
unchanged.
- Convert warpin_effect and warpout_effect to std::unique_ptr. The
delete-and-null blocks in ship::clear(), ship_close(), and
ship_set_warp_effects() collapse into reset(); call sites that borrow
the pointer into a raw local now use get().
With the members handling themselves, declare ship's special member
functions: defaulted default constructor and moves, deleted copies
(previously implicit via the unique_ptr members, now stated intent).
The moves are deliberately not declared noexcept because members like
SCP_map and std::deque do not have standard-guaranteed nothrow moves;
the computed exception specification is used instead.
Co-Authored-By: Claude Fable 5 <[email protected]>
glowpoint_bank_override_map stored raw pointers into the growable global glowpoint_bank_overrides vector. Store indices instead, which survive vector reallocation. The readers also lose a redundant second map lookup (they previously called find() and then operator[] on the same key), and prop_info's parallel map gets the same conversion.
Co-Authored-By: Claude Fable 5 <[email protected]>
The hand-written ~350-line memberwise move existed only to manage the
one raw owning pointer, and silently left any member it didn't mention
uninitialized in the destination. With subsystems as a unique_ptr, the
defaulted moves and destructor are correct by composition. Copies
become explicitly deleted (use clone()). n_subsystems stays a plain
int because it is passed to varargs functions. The moves are not
declared noexcept because libstdc++ computes a potentially-throwing
implicit specification; NOLINT suppresses clang-tidy accordingly.
Co-Authored-By: Claude Fable 5 <[email protected]>
The draw-distance cull for attached external weapon models checked
MR_ATTACHED_MODEL, but nothing has set that flag since commit
1ef7688fa1, so $Weapon Model Draw Distance: has been dead for years.
The check also needed the ship resolved from the render params' object
number, which is deliberately not set while weapon models render
(setting it would make model_render_queue clobber the explicitly
passed gatling model instance), and it compared against a depth
computed as if the weapon model's ship-local position were in world
space.
Carry the distance in model_render_params instead
(set_attached_model_draw_distance, negative = not an attached model),
set by ship_render_weapon_models from the ship's tabled value. When
set, model_render_queue measures depth from the eye position expressed
in the parent's frame (model_draw_list::get_view_position), which is
correct under the pushed parent transform, culls against the tabled
distance, and uses that same corrected depth for detail level
selection -- previously attached weapon models picked their LOD from a
garbage depth (roughly the eye's distance from the world origin).
MR_ATTACHED_MODEL is removed; nothing else referenced it. The unused
flags are also removed and the remaining flags are consolidated.
Co-Authored-By: Claude Fable 5 <[email protected]>
For standard translation offsets, the wrong axis was being used for translation. Fix that. Also fix a string offset when parsing a custom axis.
Follow-up to #4582 and #3056.
w_bank: add a move constructor. object_copy_including_array_member
returns by value, and without a move constructor the return fell back
to the shallow copy constructor whenever NRVO was not applied (e.g.
MSVC debug builds), after which the local's destructor deleted the
freshly allocated arrays -- leaving the stored bank dangling and
double-deleting at model unload. Also add noexcept to the move
operations, replace the destructor call in move-assignment with
explicit delete[]s (using an object after its lifetime has ended is
undefined behavior), guard against self-move, and document that the
defaulted copy constructor is intentionally shallow for
object_copy_including_array_member's copy-then-replace pattern.
LuaThread: declare the defaulted move operations noexcept. This
exposed that LuaTable was not movable at all (its user-declared
destructor suppressed the implicit moves), so LuaThread's implicit
move spec was formally throwing -- a hard error on gcc 9, which
still requires the spec on a defaulted redeclaration to match the
implicit one. Restore LuaTable's rule of five and declare the
LuaValue and LuaFunction copy operations noexcept (they only copy a
shared_ptr, a raw pointer, and an enum), making the noexcept claims
hold throughout the hierarchy.
Co-Authored-By: Claude Fable 5 <[email protected]>
* Extract shadows into own shader and render pass
# Conflicts:
# code/def_files/data/effects/main-g.sdr
* Consolidate render queue code
* More cleanup
* Fix missing clear of shadow buffer inter frame
* Request capability for viewport layer rendering
* instanced shadows
* Fix issues
* Switch to PCF+PCSS and proper textures for it
* Smooth PCSS toggle
* Fix for windows
* Fix OGL error
* Cleanup of old flag
* Expose shadow smoothness
* Fix missing uniform
* Make number of shadow cascades dynamic
* Progress on cockpit shadow integration
* Wrangle Cockpit and show-ship shadows into correct behaviour
* Cleanup
* make player ship shadow table options more intuitive
* Cleanup
* Only read from correct set of shadow maps, also stretch first main-scene map to 0
* Remove now non-necessary validation
* Make shadow cascade buffer more typesafe and less wasteful
* Don't allocate cockpit cascades if not used
* Cleanup
* Final fixes
* Fix Warnings
* clang tidy
* Fix missing fallback for shader precaching
* That weird FOV multiplier that always ruins your day
* Avoid shadow geometry artifacts by using hardware depth clamping
* Correctly handle shadow cascade overrides with hardcoded length
* GCC9 and its sad state of C++17 support.
* Address Feedback
# Conflicts:
# code/graphics/shadows.cpp
Replaces the copy-then-reset idiom with std::exchange in five
steal-and-reset move constructors and assignment operators.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
Set model instances to -1 when objects are deleted, and guard against null model instances in `modelinstance_replace_active_texture`. This may fix a crash in Event Horizon.
Fixes a startup assertion failure in Star Fox: Event Horizon, where a Lua
call to gr.freeAllModels() wiped Polygon_model_instances but left the UI
render instance cache holding stale indices. The next state transition
called model_clear_cached_ui_render_instances() which asserted in
model_delete_instance().
Closes#7513
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* change to new fog system
* Fix fog conversion issues
* remove unused old neb values
* Update scripting API
* temporary old-value round trip
* Rewrite nebula fog with four explicit parameters
Replace the legacy `+Fog Near Mult` / `+Fog Far Mult` model — which
depended on fixed engine constants — with four parameters that map
directly to renderer behavior:
+Fog 1000m Visibility fraction of light surviving 1000m of fog
+Fog Near Distance meters from camera where fog begins
+Fog Skybox Clip Distance max render distance for the skybox in fog
+Fog Clip Distance max render distance for the scene in fog
The deferred-lighting fog shader, material system, model renderer,
object sort, and the Lua mission API all consume the new parameters
directly. The legacy `Neb2_fog_near_mult` / `Neb2_fog_far_mult`
globals are removed.
Mission parsing accepts either the legacy keys or the modern keys.
Legacy values are converted to the modern set at parse time, and a
per-mission flag (`Neb2_fog_use_legacy_values`) records which keys to
write back on save. The flag is cleared only when something actually
changes, so opening and OK'ing an unmodified legacy mission preserves
its on-disk format. Saving uses `optional_string_fred` to preserve
user comments through edit cycles.
FRED2 and QtFRED background editors both expose the four parameters
in place of the old multipliers, with friendly labels and tooltips.
Both editors validate input (visibility clamped to [0, 1]; near
distance must be > 0; skybox and scene clip distances may be 0 to
disable fog at that range) and use a snapshot-and-compare at close
time so that a no-op open-then-OK never spuriously rewrites a legacy
mission in the modern format.
FRED2's background dialog is also reorganized so that "Display bg
bitmaps in nebula", "Override nebula fog palette", the RGB row, and
the four fog parameters all sit inside the Nebula group box; the
"Old Nebula" group box is restored below it with its Pattern, Color,
Pitch, Bank, and Heading controls back inside the frame. The dialog
now overrides PreTranslateMessage so that the new fog tooltips —
and the previously broken "Save sun/bitmap angles in correct format"
tooltip — actually fire.
QtFRED's BackgroundEditor.ui adds four QDoubleSpinBox controls in
the existing nebula form layout, with per-field decimals chosen to
avoid scientific notation, and per-field min/max declaring the
validation rules. The QtFRED help page is updated to describe the
new fields.
* fix tooltips
* restore forward-renderer fog
* Clang Tidy
* address feedback
* Refine nebula fog rewrite
- Rename Neb2_fog_use_legacy_values to Neb2_fog_save_legacy_values
to reflect that the flag governs save format, not runtime
behavior.
- Decide both the save flag and the legacy-to-modern conversion
purely from whether the file contains modern fog keys, dropping
the prior heuristic based on required_fso_version. A mission
without modern keys (legacy- or retail-format) converts on load
and saves back in legacy format; missions with modern keys are
saved as modern.
- Snapshot the four fog values as floats at editor open and use
fl_equal for the no-change comparison in both FRED2 and QtFRED,
replacing the FRED2 CString comparison and the QtFRED strict
inequality.
- Document the on-disk invariant in missionsave: when not saving
legacy keys, the four modern keys must always be written (never
gated on "only if non-default"), since the parser uses their
presence to decide whether to run the conversion on next load.
---------
Co-authored-by: Goober5000 <[email protected]>
Fix#4089, part 2:
- properly link special-point subsystems with -destroyed submodels, if such a relationship exists and a game_settings.tbl flag is set
- add some helper functions for -destroyed variants of submodels (and now subsystems)
