fix(render): production packed classifier carries the foliage flags; deferred-alpha replay keeps them (Campaign VM VM6 review 2)

Narrow re-review of 43e3abed found every A1-A5/A7/A8 item resolved but
one new blocker in the production packed classifier.

F1 (BLOCKER): RetailPViewPassExecutor.DrawPackedProductionRoute — the
route production world geometry actually draws from — never computed
FoliageFlags at all. WbDrawDispatcher.PackedOracle.cs's
ClassifyPackedBatches built its GroupKey with the field defaulting to
0u, and GetOrCreatePackedGroup never copied it onto the created
InstanceGroup, so production BatchData.flags bits 1/2 were always zero
for every scenery entity: the world geometry never swayed even though
the independently-classified shadow caster did, so shadows visibly
swayed under rigid trees. Both classifier call sites now compute
FoliageFlags via the identical FoliageWindClassification.Classify call
and entity-scoped HasCutoutSubset OR the classic (non-packed) path
uses, and GetOrCreatePackedGroup copies it exactly like
GetOrCreateInstanceGroup always has. The G2/G3 classified-output
digest (AddOpaqueSubmissionGroup/BuildTransparentSubmissionDigest) now
also folds GroupKey.FoliageFlags into its hash — present in the key
since round 1 but never actually read by either digest function, so a
content-level (not just group-count-level) classic-vs-packed
divergence is now caught.

F2 (medium): the delayed-alpha replay path (PrepareDeferredAlphaDraws)
hardcoded Flags = 1, dropping bits 1/2 for any group replayed through
it — a trunk instance promoted into the alpha-blend group mid-fade
(the #188 translucency-promotion case) would stop swaying for the
duration of its fade. Now 1u | key.FoliageFlags.

F3 (nit): ComputeEntityHasCutoutSubset's three call sites (classic,
caster, and the newly-fixed packed classifier) each allocated a
closure over _meshAdapter per Setup entity per frame. A new
context-taking overload passes the mesh adapter as an explicit
argument to a static lambda instead, letting the compiler cache one
delegate for the method's lifetime rather than allocating fresh ones.

A3 test gap: WbDrawDispatcher.BindDirectionalShadowReceiver is now
internal so DirectionalShadowGpuTests can drive it directly with a
bare RecordingGpuDevice pass encoder, proving it emits
UniformAtmosphericFrame with the exact buffer/offset/size a
DirectionalShadowFrameBinding carries — paired with the existing test
proving that binding carries the caster's real bind forward untouched.

F1's missing test: PackedDispatcherOracleTests chains
FoliageWindClassification.Classify (called with the packed
classifier's exact argument shape) for a real 0x8... scenery entity id
through BuildIndirectArrays — the same shared, already-tested
production step both classic and packed group lists feed into BatchData —
proving the resulting flags word carries bit 0x2. Driving
ClassifyPackedBatches/GetOrCreatePackedGroup directly was not a "cheap
test": both are private instance methods reachable only through the
full RetailPViewPassExecutor route, which needs a real IGpuDevice,
world-pass scope, mesh manager, and compiled pipelines to construct —
no test anywhere in the App test project stands one up.

Nits: F4 corrects foliage_wind.glsl's header comment from "bit 31" to
the top-nibble test; F5 documents at the receiver bind site that the
caster's own AtmosphericFrameBufferBinding has its seven ABI v1
members zero/Identity by construction (only the two v2 wind members
are valid) — safe today because mesh_atmospheric.vert reads that
binding solely for wind displacement, flagged as a footgun for a
future v1-reading addition to that shader; F6 notes in the plan
(rather than fixes) that EntityCacheEntry does not proactively
invalidate when FoliageWindExclusions changes on a pack switch —
harmless with the pack off, self-heals on the entry's next natural
eviction.

foliage_wind.glsl's F4 comment-only change updated the SPIR-V
manifest's source hashes for the five includers (mesh_atmospheric.vert
+ four directional_shadow_world_* casters); the compiled .spv bytes
are byte-identical since comments do not affect bytecode.

Verify: Release build 0 warnings/0 errors. App hermetic-lane filter
6,043/0 failed. Core.Tests 4,695/0 failed. RenderPackValidator 30/30.
Full hermetic-filtered solution: 15,271/0 failed across 15 projects.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-08-23 02:28:17 +02:00
parent 43e3abed4d
commit a82959f1b7
10 changed files with 332 additions and 16 deletions

View file

@ -2,8 +2,10 @@
#define ACDREAM_FOLIAGE_WIND_GLSL
// Campaign VM VM6 (2026-08-22): weather-driven foliage sway for procedural
// scenery (trees/bushes — see ProceduralSceneryIdAllocator, bit 31 of the
// entity id). Shared verbatim between mesh_atmospheric.vert (the world
// scenery (trees/bushes — see ProceduralSceneryIdAllocator.IsInNamespace,
// the top nibble 0x8 of the entity id — NOT bit 31 alone, which also
// matches LandblockStaticEntityIdAllocator's 0xC... namespace). Shared
// verbatim between mesh_atmospheric.vert (the world
// receiver) and the four directional_shadow_world_*.vert casters so a
// displaced leaf's shadow moves with it by construction — the caster and
// receiver read the identical uAtmosphereClockWind/uAtmosphereWindAmplitude

View file

@ -151,7 +151,7 @@
"stages": [
{
"stage": "vert",
"sourceSha256": "ae6026d4bc04bc07e5c67c1dca6c3dfd16ad377104153808a2e11b60bbe492f7",
"sourceSha256": "a30acec1fd60fcbc8a0cfa377726a65077a82836cbcf730cbcc756ce1b9299c5",
"compiled": true
},
{
@ -167,7 +167,7 @@
"stages": [
{
"stage": "vert",
"sourceSha256": "c26158346a1db03e305192a23b595d288ed27a3622b8c2b49cc2fe6b27550946",
"sourceSha256": "7651a9622227c0322f75c8e3af2685bceb9bc28b692b53df305ec73161e589ff",
"compiled": true
},
{
@ -183,7 +183,7 @@
"stages": [
{
"stage": "vert",
"sourceSha256": "c752d8f85e0aa4a5bd2b7cb8679276ba46850844362eef0efce885401021433b",
"sourceSha256": "9513c5baa4f269a9ddc265e9fa49c71a2834a004aca36939b99f9b5067d25ed1",
"compiled": true
},
{
@ -199,7 +199,7 @@
"stages": [
{
"stage": "vert",
"sourceSha256": "e483aae19e35ad9bdd7226cb1604db33268f8ae4474d774dfa08d5ca681717ee",
"sourceSha256": "edf89ab8a8c02bf8fc3099b091ca3508a5b9c18e2ebbfe5bb5a7d6c50a82c745",
"compiled": true
},
{
@ -215,7 +215,7 @@
"stages": [
{
"stage": "vert",
"sourceSha256": "949a50149a6dd7f8f2b44fa035ab80a4227a71a1cbd4ec812bf451d1c2971db3",
"sourceSha256": "a00aa83a8bd8ab4791b60bfb967a80b43bafcb0bf758f5e2e056b4272f3c8b1b",
"compiled": true
},
{

View file

@ -81,6 +81,33 @@ internal static class FoliageWindClassification
return false;
}
/// <summary>
/// Campaign VM VM6 review fix round 2 (F3): zero-alloc production
/// overload of <see cref="ComputeEntityHasCutoutSubset{T}"/>. The
/// single-generic overload above is exactly what a caller needs to
/// predicate on an instance field (e.g. the mesh adapter) — but doing so
/// with a lambda that closes over <c>this</c> allocates a fresh closure
/// object AND a fresh delegate on every call, and this is called once
/// per Setup entity per frame from the world receiver, the shadow
/// caster, and the packed production classifier. Passing the would-be-
/// captured value as an explicit <typeparamref name="TContext"/>
/// argument to a <see langword="static"/> lambda (no captures at all)
/// lets the C# compiler cache a single delegate instance for the
/// method's lifetime instead of allocating one per call.
/// </summary>
internal static bool ComputeEntityHasCutoutSubset<T, TContext>(
IReadOnlyList<T> setupParts,
TContext context,
Func<TContext, T, bool> hasCutoutSubset)
{
for (int i = 0; i < setupParts.Count; i++)
{
if (hasCutoutSubset(context, setupParts[i]))
return true;
}
return false;
}
/// <summary>
/// The full top-nibble <c>0x8...</c> test via
/// <see cref="ProceduralSceneryIdAllocator.IsInNamespace"/> — NOT bit 31

View file

@ -108,7 +108,12 @@ public sealed partial class WbDrawDispatcher
return PipelinesFor(encoder);
}
private static void BindDirectionalShadowReceiver(
// Campaign VM VM6 review fix round 2 (A3 test gap): internal, not
// private, so DirectionalShadowReceiverTests can drive it directly with
// a bare RecordingGpuDevice pass encoder instead of standing up the
// full WbDrawDispatcher/mesh-manager/world-pass-scope dependency chain
// just to prove this one bind call is correct.
internal static void BindDirectionalShadowReceiver(
IGpuPassEncoder encoder,
in DirectionalShadowFrameBinding binding)
{
@ -127,6 +132,20 @@ public sealed partial class WbDrawDispatcher
// depending on that earlier bind surviving un-reset until this pass
// runs. Unbound (the declared-pack / no-caster-this-frame case) is a
// no-op here exactly like the caster side.
//
// Review fix round 2 (F5): the buffer bound here is the CASTER's own
// independent allocation (AtmosphericPostProcessGraph
// .BuildShadowAtmosphericFrameBinding), NOT the receiver's/post-
// process's fuller frame block. Its seven ABI v1 members (sun
// screen/color, viewport, weather, sun direction, policy, inverse-
// view-projection) are ALL ZERO/Identity by construction — only the
// two appended v2 members (clock/wind, wind amplitude) carry real
// content. That is safe today because mesh_atmospheric.vert reads
// this binding SOLELY for foliage_wind.glsl's wind displacement; if
// a future change makes mesh_atmospheric.vert or .frag read any v1
// member (sun/weather/policy/etc.) during this same world-mesh
// pass, it will read zeros here and must get its own real binding
// instead of assuming this one is fully populated.
if (binding.AtmosphericFrame.IsBound)
{
encoder.BindUniformBuffer(

View file

@ -1028,11 +1028,16 @@ public sealed partial class WbDrawDispatcher
// FoliageWindClassification.ComputeEntityHasCutoutSubset's
// doc comment. Same shared helper the world-receiver
// Setup-part loop uses, so the caster's trunk-part
// classification agrees with the receiver's.
// classification agrees with the receiver's. Review
// fix round 2 (F3): the context-taking overload
// passes _meshAdapter explicitly to a static lambda
// instead of closing over it — zero allocation per
// entity per frame.
bool entityHasCutoutSubset = FoliageWindClassification
.ComputeEntityHasCutoutSubset(
renderData.SetupParts,
part => _meshAdapter.TryGetRenderData(part.GfxObjId)
_meshAdapter,
static (adapter, part) => adapter.TryGetRenderData(part.GfxObjId)
is { HasCutoutSubset: true });
for (int setupPartIndex = 0;

View file

@ -437,6 +437,25 @@ public sealed unsafe partial class WbDrawDispatcher
if (renderData.IsSetup
&& renderData.SetupParts.Count > 0)
{
// Campaign VM VM6 review fix round 2 (F1): same entity-
// scoped HasCutoutSubset OR the classic world-receiver loop
// computes (WbDrawDispatcher.cs) and the caster loop
// (DirectionalShadows.cs) — see
// FoliageWindClassification.ComputeEntityHasCutoutSubset's
// doc comment. A Setup composite's parts are separate
// GfxObjs with independently-cached HasCutoutSubset; without
// this bounded per-entity scan, the opaque trunk part of a
// production-classified tree would never get the trunk
// flag. The context-taking overload passes _meshAdapter
// explicitly to a static lambda (F3) — zero allocation per
// entity per frame.
bool entityHasCutoutSubset = FoliageWindClassification
.ComputeEntityHasCutoutSubset(
renderData.SetupParts,
_meshAdapter,
static (adapter, part) => adapter.TryGetRenderData(part.GfxObjId)
is { HasCutoutSubset: true });
for (int setupPartIndex = 0;
setupPartIndex < renderData.SetupParts.Count;
setupPartIndex++)
@ -477,7 +496,8 @@ public sealed unsafe partial class WbDrawDispatcher
indoor,
selectionLighting,
opacity,
cacheEntry))
cacheEntry,
entityHasCutoutSubset))
{
reusableAcrossFrames = false;
}
@ -593,8 +613,17 @@ public sealed unsafe partial class WbDrawDispatcher
bool indoor,
Vector2 selectionLighting,
float opacity,
PackedProjectionClassificationEntry? cacheEntry)
PackedProjectionClassificationEntry? cacheEntry,
// Campaign VM VM6 review fix round 2 (F1): mirrors the classic
// dispatcher's ClassifyBatches override exactly. HasCutoutSubset is
// a per-PART (per-GfxObj) fact; the caller passes the ENTITY/Setup-
// scoped OR across every resolved part for a Setup composite. null
// (the default) means "use renderData.HasCutoutSubset directly",
// which is already correct for a non-Setup single-mesh entity.
bool? entityHasCutoutSubsetOverride = null)
{
bool entityHasCutoutSubset =
entityHasCutoutSubsetOverride ?? renderData.HasCutoutSubset;
bool reusableAcrossFrames = true;
for (int batchIndex = 0;
batchIndex < renderData.Batches.Count;
@ -617,6 +646,22 @@ public sealed unsafe partial class WbDrawDispatcher
if (!texture.Slot.IsAssigned)
continue;
// Campaign VM VM6 review fix round 2 (F1 BLOCKER): the packed
// production classifier never computed FoliageFlags, so the
// production BatchData.flags word was always 0 for every
// scenery entity — the world geometry never swayed even though
// the independently-classified shadow caster did. Classify
// BEFORE constructing the key, from the RAW (pre-#188-
// promotion) batch.Translucency, exactly as the classic
// ClassifyBatches does — see that method's own comment for why
// raw translucency is used for classification but the (possibly
// promoted) local `translucency` is still what the key/group
// partitions draws by.
uint foliageFlags = FoliageWindClassification.Classify(
entity.LocalEntityId,
FoliageWindExclusions.Contains(meshRef.GfxObjId),
batch.Translucency,
entityHasCutoutSubset);
var key = new GroupKey(
batch.FirstIndex,
(int)batch.BaseVertex,
@ -624,7 +669,8 @@ public sealed unsafe partial class WbDrawDispatcher
texture.Slot,
texture.Layer,
translucency,
batch.CullMode);
batch.CullMode,
foliageFlags);
var classified = new PackedClassifiedBatch(
key,
restPose,
@ -759,6 +805,17 @@ public sealed unsafe partial class WbDrawDispatcher
TextureLayer = key.TextureLayer,
Translucency = key.Translucency,
CullMode = key.CullMode,
// Campaign VM VM6 review fix round 2 (F1 BLOCKER): this was the
// missing copy. GetOrCreateInstanceGroup (the classic route) has
// set FoliageFlags from key.FoliageFlags since the round-1
// fix; this packed-route sibling never did, so every packed
// production InstanceGroup carried FoliageFlags == 0 regardless
// of what ClassifyPackedBatches computed into the key —
// BatchData.flags bits 1/2 were always 0 for production world
// geometry even though the independently-classified shadow
// caster set them correctly, so casters swayed while the
// meshes they shadowed did not.
FoliageFlags = key.FoliageFlags,
Registration = _nextPackedGroupRegistration++,
LastUsedFrame = _packedGroupFrame,
};

View file

@ -1855,9 +1855,14 @@ public sealed partial class WbDrawDispatcher : IDisposable
// each part's own. A part missing this frame contributes
// nothing; the entity is already marked incomplete by the
// loop below and reclassifies once every part has loaded.
// Review fix round 2 (F3): the context-taking overload
// passes _meshAdapter as an explicit argument to a static
// lambda instead of closing over it, so this call allocates
// nothing per entity per frame.
bool entityHasCutoutSubset = FoliageWindClassification.ComputeEntityHasCutoutSubset(
renderData.SetupParts,
part => _meshAdapter.TryGetRenderData(part.GfxObjId) is { HasCutoutSubset: true });
_meshAdapter,
static (adapter, part) => adapter.TryGetRenderData(part.GfxObjId) is { HasCutoutSubset: true });
// #188: setupPartIndex is the SAME index space
// TransparentPartHook.PartIndex addresses — retail's CPartArray
@ -2576,6 +2581,12 @@ public sealed partial class WbDrawDispatcher : IDisposable
hash.Add(key.TextureLayer);
hash.Add((int)key.Translucency);
hash.Add((int)key.CullMode);
// Review fix round 2 (F1): fold FoliageFlags into the G2/G3
// classified-output digest so a scenery/non-scenery group-
// partition divergence between the classic and packed
// dispatchers is caught by content, not only by an incidental
// group-count difference.
hash.Add(key.FoliageFlags);
hash.Add(entry.ViewerDistance);
AddSubmissionInstance(
ref hash,
@ -2626,6 +2637,10 @@ public sealed partial class WbDrawDispatcher : IDisposable
hash.Add(key.TextureLayer);
hash.Add((int)key.Translucency);
hash.Add((int)key.CullMode);
// Review fix round 2 (F1): see the matching comment in
// BuildTransparentSubmissionDigest — folds FoliageFlags into the
// G2/G3 classified-output digest.
hash.Add(key.FoliageFlags);
hash.Add(group.Matrices.Count);
ulong xorLow = 0;
@ -2786,8 +2801,17 @@ public sealed partial class WbDrawDispatcher : IDisposable
// DrawMesh invokes RenderMeshSubset with detail enabled for
// every built-mesh material subset while curr_detail_surface
// is installed. The per-instance category still rejects
// ordinary objects in mesh_detail.
Flags = 1,
// ordinary objects in mesh_detail. Review fix round 2 (F2):
// bit 0 (the #226 built-mesh marker) is unconditional here —
// every delayed-alpha entry IS a built-mesh alpha submission
// — but the previous hardcoded `1` silently dropped bits 1/2
// (Campaign VM VM6 foliage classification). A trunk instance
// promoted into the alpha-blend group mid-fade (translucency
// < 1 forces AlphaBlend even for an otherwise-Opaque trunk —
// see ClassifyBatches's #188 promotion) must keep swaying
// through the deferred-alpha replay instead of going rigid
// for the duration of its fade.
Flags = 1u | key.FoliageFlags,
};
_indirectCommands[i] = new DrawElementsIndirectCommand
{