feat(render): Campaign V slice V6i-1 - one descriptor set per renderer scope

Plan section 5.5.8 recorded, and deliberately did not fix, that pointing one
binding at a second buffer within a frame silently corrupts the draws already
recorded against it: the backend rewrote the descriptor in place, and a
descriptor set's contents are read when the command buffer EXECUTES, not when it
was recorded. Nothing fired it while the Vulkan frame held only the retained UI.
Section 5.5.11 handed it forward as the first thing the world arm would hit,
because WbDrawDispatcher, EnvCellRenderer and TerrainModernRenderer each own
their own instance, batch and indirect buffers and all three bind set 0 in one
frame.

It is closed here, as its own commit and BEFORE the world arm, so that a blank or
corrupt first Vulkan world frame cannot be this defect wearing another face. That
sequencing is the point: sections 5.5.1 to 5.5.3 cost this campaign three days
because an instrument that was "usually right" sat underneath the thing being
measured.

What changed. There is no longer one (set 0, set 1) pair per flight slot; there
is an arena of them. VulkanBindingScopeArena - pure bookkeeping, no Vulkan
handles, nine unit tests - answers two questions per bind: which pair, and do its
descriptors need writing. VulkanFrameBindings keeps the Vulkan half: allocating
pairs from a growable pool list and writing the twelve descriptors when told to.

The scope key is the descriptor state itself - the ten storage buffer identities
and ranges, the plain bindings' offsets, and the two uniform buffer identities
and ranges. Deriving it is a decision, not an economy. The pinned contract has
nowhere to name a scope: BindStorageBuffer takes a buffer, an offset and a size,
and section 3.3 is frozen. Deriving also gives two properties a declared scope
would not: a renderer cannot forget to declare one, and two renderers that
genuinely share every buffer correctly share one pair rather than being told to
differ. A renderer's buffers are stable for its lifetime, so "distinct descriptor
state" is exactly "renderer scope".

Dynamic offsets stay free. A ring allocation moving between draws rides
vkCmdBindDescriptorSets's dynamic-offset array, so it costs neither a new pair
nor a descriptor write - section 4.4's "zero descriptor writes per frame"
property survives a frame having more than one binding state in it. Entries are
not invalidated at BeginFrame either, because the slot's previous submission has
retired and its descriptors still say what this frame is about to say; a steady
frame therefore rewrites nothing at all. An entry matched from the previous frame
is swapped below the live cursor so the rest of the frame cannot take it for a
different state - the ordering property the sixth test pins, where two renderers
swap submission order between frames.

What this does NOT do is draw a world. The captured Vulkan frame is still V6h's
retained UI over the fog clear, so the arena's multi-scope path is exercised by
its tests and not yet by a frame. That is recorded in the plan rather than
implied.

The plan's section 5.5.12 also records two blockers measured while scoping the
world arm and not fixed here: terrain_modern.vert declares TerrainClip without
ACDREAM_UBO_SET, so under the Vulkan dialect it lands at set 0 binding 2 where
the layout declares a storage buffer - the same class of gap 5.5.8 recorded for
UniformSkyParams, invisible until a terrain pipeline is created; and the offline
gate's scene takes the retail PView path rather than the flat safety path,
because ClipRoot falls back to Buildings.OutdoorNode, which puts
RetailPViewPassExecutor on the critical path to the first Vulkan Dereth frame and
makes the "terrain only" intermediate no cheaper than the whole arm.

Gates. Strict GL offline pixel gate against b9ab5890: 1.60e-05, 9 differing
pixels of 563,200, at the low end of the documented 9-31 px band and 62x under
the threshold - expected, since no GL file is touched. GL connected
run-repeat-connected-gate.ps1 -Runs 3: 3/3 RENDERED on the desktop witness and
3/3 on the client capture. One offline Vulkan run with VK_LAYER_KHRONOS_validation
proven inserted by the loader: zero errors, zero warnings, captured frame, no
[shutdown] diagnostic on either stream. App tests 4,086 / 3 skips (baseline 4,077
plus nine); complete Release suite 9,149 / 5. Issue #250's
SurfaceOverrideFingerprint_DictionaryHotPathAllocatesNothing failed once in a
whole-suite run and passed run alone, as that issue documents.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-07-28 13:17:25 +02:00
parent b9ab5890af
commit df6e2a7918
6 changed files with 759 additions and 100 deletions

View file

@ -1494,6 +1494,109 @@ instance and batch buffers and both bind bindings 0, 1, 3, 4 and 5 in one
frame; and the world path's own texture creation still has to reach
`IGpuTexture`, because a Vulkan draw cannot sample a GL handle.
#### 5.5.12 V6i-1 (2026-07-28): the descriptor-set hazard is closed before the world arm fires it
§5.5.8 recorded, and deliberately did not fix, that **a binding pointed at two
different buffers within one frame silently corrupts the earlier draws**: the
backend rewrote the descriptor in place, and a descriptor set's contents are read
when the command buffer EXECUTES, not when it was recorded. Nothing fired it
while the Vulkan frame held only the retained UI. §5.5.11 handed it forward as
the first thing the world arm would hit, because `WbDrawDispatcher`,
`EnvCellRenderer` and `TerrainModernRenderer` each own their own instance, batch
and indirect buffers and all three bind set 0 in one frame.
It is now closed, and it was closed *first* rather than alongside the world arm,
so that a blank or corrupt Vulkan world frame cannot be this defect wearing
another face.
**One set pair per renderer scope, derived rather than declared.** There is no
longer a single (set 0, set 1) pair per flight slot; there is an arena of them.
`VulkanBindingScopeArena` — pure bookkeeping, no Vulkan handles, unit-tested —
decides which pair a bind belongs to and whether its descriptors must be written.
The scope key is the descriptor state itself: the ten storage buffer identities
and ranges, the plain (non-dynamic) bindings' offsets, and the two uniform buffer
identities and ranges.
Deriving the scope is a decision, not an economy. **The pinned contract has no
place to name one** — `BindStorageBuffer` takes a buffer, an offset and a size,
and §3.3 is frozen. Deriving it from state also gives two properties a declared
scope would not: a renderer cannot forget to declare one, and two renderers that
genuinely share every buffer correctly share one pair instead of being told to
differ. A renderer's buffers are stable for its lifetime, so "distinct descriptor
state" *is* "renderer scope".
**Dynamic offsets stay free.** A ring allocation moving between draws rides
`vkCmdBindDescriptorSets`'s dynamic-offset array, so it costs neither a new pair
nor a descriptor write — §4.4's "zero descriptor writes per frame" property
survives a frame now having more than one binding state in it. A steady frame
rewrites nothing at all: entries are not invalidated at `BeginFrame`, because the
slot's previous submission has retired and its descriptors still say exactly what
this frame is about to say. An entry matched from the previous frame is swapped
below the live cursor, so the rest of the frame cannot take it for a different
state — the property nine tests pin, including the ordering case where two
renderers swap their submission order between frames.
**Two things the world arm still needs, unchanged from §5.5.11's handoff**, plus
one this slice measured:
1. **World texture CREATION must reach `IGpuTexture`.** V4t moved the table
ENTRY to the device and left creation with `TerrainAtlas`,
`CompositeTextureArrayCache`, `ManagedGLTextureArray` and `TextureCache`'s
array path. `TerrainAtlas` is the smallest of these and the only one terrain
needs; `TextureAtlasManager`/`ManagedGLTextureArray` is the largest, and it is
what statics, scenery and EnvCell shells sample through.
`ICompositeTextureArrayBackend` is already a seam and takes an RHI backend
directly; `TextureAtlasManager` is not, and reaches `OpenGLGraphicsDevice`
through `CreateTextureArrayInternal`. `BlockCompressionCodec` and
`BlockCompressionMipChain` (V6b) already supply the BC mip chains that path
needs, so the missing piece is an `ITextureArray` implementation over
`IGpuTexture`, not a codec.
2. **`terrain_modern.vert`'s `TerrainClip` block is in the wrong set for Vulkan,
and it is the same class of gap §5.5.8 recorded for `UniformSkyParams`.** It
is declared `layout(std140, binding = 2) uniform TerrainClip` with **no
`ACDREAM_UBO_SET`**, so under the Vulkan dialect it lands at set 0 binding 2 —
which set 0's layout declares as a STORAGE buffer. GL is unaffected (the macro
expands to nothing and the UBO namespace is separate), and nothing has caught
it because no terrain pipeline has ever been created on Vulkan. Whoever draws
terrain must add the macro, declare binding 2 in the uniform set layout, and
raise `DynamicUniformBindingCount` to 3 — still far under the guaranteed 8.
Worth auditing every remaining shader for the same omission in the same pass.
3. **The flat-versus-PView question decides how much of the world arm is one
slice.** `WorldSceneRenderer` is already backend-neutral — it takes
`IWorldScenePassExecutor` and `IWorldScenePViewRenderer` as interfaces — so
the Vulkan arm reuses it whole. But `WorldRenderFrame.ClipRoot` is
`Roots.ViewerRoot ?? Buildings.OutdoorNode`, and the offline gate's scene has
buildings, so the offline capture takes the **PView** path, not the flat
safety path. A Vulkan world arm that implements only `WorldScenePassExecutor`
therefore renders nothing in the very scene the pixel gate captures.
`RetailPViewPassExecutor` (685 lines) is on the critical path to the first
Vulkan Dereth PNG, and the "terrain only" intermediate §5.5.7 suggested is
consequently NOT cheaper than it looks.
**Gate results.** Strict GL offline pixel gate against `b9ab5890`: **1.60e-05**
(9 differing pixels of 563,200), at the low end of the documented 931 px band
and 62x under the 0.001 threshold — expected, since no GL file is touched.
GL connected `tools/run-repeat-connected-gate.ps1 -Runs 3`: **3/3 RENDERED** on
the desktop witness and 3/3 on the client capture. One offline Vulkan run with
`VK_LAYER_KHRONOS_validation` **proven inserted by the loader** (`Insert instance
layer "VK_LAYER_KHRONOS_validation"` from `VK_LOADER_DEBUG=layer`): **zero
validation errors and zero warnings**, a captured retained-UI frame, and no
`[shutdown]` diagnostic on either stream. App tests 4,086 / 3 skips (the 4,077
baseline plus nine); complete Release suite 9,149 / 5. `#250`'s
`SurfaceOverrideFingerprint_DictionaryHotPathAllocatesNothing` failed once in a
whole-suite run and passed run alone, exactly as that issue documents.
**What this slice deliberately does NOT do**: draw a world. The captured Vulkan
frame is V6h's — the retained UI over the atmosphere fog clear — because no world
renderer exists on the Vulkan arm yet. The arena's multi-scope path is therefore
exercised only by its unit tests; the live Vulkan frame binds one scope and
re-matches it every frame, which is the "rewrites nothing" case. That is the
honest state and it is why the hazard was closed as its own commit: when the
world arm lands and a frame first holds three renderers' buffers, a blank or
corrupt result cannot be this defect.
### 5.4 The null-target `BeginPass` divergence (V4c) — must be undone at V6
V4c had to stop GL's `BeginPass` from binding framebuffer 0 when a pass declares

View file

@ -0,0 +1,257 @@
namespace AcDream.App.Rendering.Gpu.Vk;
/// <summary>
/// Campaign V slice V6i: which descriptor set a bind belongs to, and whether it
/// has to be written.
///
/// <para><b>The defect this closes.</b> Plan §5.5.8 recorded, and did not fix,
/// that pointing one binding at a second buffer within a frame silently corrupts
/// the draws recorded before it: <c>vkUpdateDescriptorSets</c> rewrites the set
/// in place, and a set's contents are read when the command buffer EXECUTES, not
/// when it was recorded. Nothing on the Vulkan arm did that while the frame held
/// only the retained UI. The world arm is what fires it —
/// <c>WbDrawDispatcher</c>, <c>EnvCellRenderer</c> and <c>TerrainModernRenderer</c>
/// each own their own instance, batch and indirect buffers and all three bind
/// set 0 in one frame.</para>
///
/// <para><b>Why the policy lives here rather than in the renderers.</b> The
/// pinned contract deliberately has no place to name a scope:
/// <see cref="IGpuPassEncoder.BindStorageBuffer"/> takes a buffer, an offset and
/// a size. So the backend DERIVES the scope from the descriptor state itself —
/// the buffer identities, plus offset and range for the plain (non-dynamic)
/// bindings whose descriptor carries them. A renderer's buffers are stable for
/// its lifetime, so "distinct descriptor state" is exactly "renderer scope",
/// with two properties a declared scope would not have: a renderer cannot forget
/// to declare one, and two renderers that genuinely share every buffer correctly
/// share one set instead of being told to differ.</para>
///
/// <para><b>Dynamic offsets are not part of the state.</b> A ring allocation
/// moving between draws rides <c>vkCmdBindDescriptorSets</c>'s dynamic-offset
/// array, so it costs neither a new entry nor a descriptor write. That is what
/// keeps the campaign's "zero descriptor writes per frame" property true now
/// that a frame has more than one binding state in it.</para>
///
/// <para><b>Lifetime.</b> Entries below <see cref="LiveCount"/> were claimed
/// earlier in the current frame and may have a recorded draw bound to them, so
/// they are never rewritten. Entries above it belong to the previous frame on
/// this flight slot, which has retired, so they are free to reclaim — and are
/// re-matched rather than rewritten when their state still holds, which is the
/// steady-state case. A match above the cursor is swapped down so the remainder
/// of the frame cannot take it for a different state.</para>
///
/// <para>This type is pure bookkeeping: no Vulkan handles, no allocation of
/// descriptor sets. <see cref="VulkanFrameBindings"/> owns those and asks this
/// class two questions — which index, and does it need writing.</para>
/// </summary>
internal sealed class VulkanBindingScopeArena
{
private readonly int _storageBindingCount;
private readonly int _uniformBindingCount;
private readonly Func<uint, bool> _isDynamicStorage;
private readonly ulong[] _storageBuffers;
private readonly uint[] _storageOffsets;
private readonly uint[] _storageRanges;
private readonly ulong[] _uniformBuffers;
private readonly uint[] _uniformOffsets;
private readonly uint[] _uniformRanges;
private readonly List<Entry> _entries = [];
private int _liveCount;
private int _active = -1;
private bool _dirty = true;
private sealed class Entry(int storageBindingCount, int uniformBindingCount)
{
public ulong[] StorageBuffers { get; } = new ulong[storageBindingCount];
public uint[] StorageOffsets { get; } = new uint[storageBindingCount];
public uint[] StorageRanges { get; } = new uint[storageBindingCount];
public ulong[] UniformBuffers { get; } = new ulong[uniformBindingCount];
public uint[] UniformRanges { get; } = new uint[uniformBindingCount];
public int Slot { get; set; } = -1;
}
internal VulkanBindingScopeArena(
int storageBindingCount,
int uniformBindingCount,
Func<uint, bool> isDynamicStorage)
{
ArgumentOutOfRangeException.ThrowIfNegativeOrZero(storageBindingCount);
ArgumentOutOfRangeException.ThrowIfNegativeOrZero(uniformBindingCount);
_storageBindingCount = storageBindingCount;
_uniformBindingCount = uniformBindingCount;
_isDynamicStorage = isDynamicStorage
?? throw new ArgumentNullException(nameof(isDynamicStorage));
_storageBuffers = new ulong[storageBindingCount];
_storageOffsets = new uint[storageBindingCount];
_storageRanges = new uint[storageBindingCount];
_uniformBuffers = new ulong[uniformBindingCount];
_uniformOffsets = new uint[uniformBindingCount];
_uniformRanges = new uint[uniformBindingCount];
}
/// <summary>Distinct descriptor states this slot has ever materialised.</summary>
internal int Count => _entries.Count;
/// <summary>Entries claimed so far in the current frame; none of these may be rewritten.</summary>
internal int LiveCount => _liveCount;
/// <summary>
/// Seeds every binding's initial descriptor state — the shared dummy range
/// each unused binding points at. Called once, before the first frame.
/// </summary>
internal void SeedStorage(uint binding, ulong buffer, uint offsetBytes, uint rangeBytes)
{
_storageBuffers[binding] = buffer;
_storageOffsets[binding] = offsetBytes;
_storageRanges[binding] = rangeBytes;
}
/// <inheritdoc cref="SeedStorage"/>
internal void SeedUniform(uint binding, ulong buffer, uint rangeBytes)
{
_uniformBuffers[binding] = buffer;
_uniformRanges[binding] = rangeBytes;
}
internal void BeginFrame()
{
_liveCount = 0;
_active = -1;
_dirty = true;
}
/// <summary>
/// Records a storage bind. A DYNAMIC binding's descriptor depends on the
/// buffer and range alone; a PLAIN one also carries the offset, so moving it
/// is a descriptor-visible change.
/// </summary>
internal void SetStorage(uint binding, ulong buffer, uint offsetBytes, uint rangeBytes)
{
ArgumentOutOfRangeException.ThrowIfGreaterThanOrEqual(binding, (uint)_storageBindingCount);
bool dynamic = _isDynamicStorage(binding);
if (_storageBuffers[binding] != buffer
|| _storageRanges[binding] != rangeBytes
|| (!dynamic && _storageOffsets[binding] != offsetBytes))
{
_storageBuffers[binding] = buffer;
_storageRanges[binding] = rangeBytes;
_dirty = true;
}
_storageOffsets[binding] = offsetBytes;
}
internal void SetUniform(uint binding, ulong buffer, uint offsetBytes, uint rangeBytes)
{
ArgumentOutOfRangeException.ThrowIfGreaterThanOrEqual(binding, (uint)_uniformBindingCount);
if (_uniformBuffers[binding] != buffer || _uniformRanges[binding] != rangeBytes)
{
_uniformBuffers[binding] = buffer;
_uniformRanges[binding] = rangeBytes;
_dirty = true;
}
_uniformOffsets[binding] = offsetBytes;
}
internal uint StorageOffset(uint binding) => _storageOffsets[binding];
internal uint UniformOffset(uint binding) => _uniformOffsets[binding];
internal ulong StorageBuffer(uint binding) => _storageBuffers[binding];
internal uint StorageRange(uint binding) => _storageRanges[binding];
/// <summary>
/// The descriptor offset written INTO a binding's descriptor: zero for a
/// dynamic binding, which slides with the bind call, and the real offset for
/// a plain one.
/// </summary>
internal uint StorageDescriptorOffset(uint binding) =>
_isDynamicStorage(binding) ? 0u : _storageOffsets[binding];
internal ulong UniformBuffer(uint binding) => _uniformBuffers[binding];
internal uint UniformRange(uint binding) => _uniformRanges[binding];
/// <summary>
/// The arena entry the current state belongs to.
/// <paramref name="slot"/> is the caller's opaque handle for the entry — the
/// index of its (set 0, set 1) pair — or -1 when the entry is new and the
/// caller must allocate one and report it through <see cref="AssignSlot"/>.
/// <c>NeedsWrite</c> is true only when the entry's descriptors do not already
/// hold this state.
/// </summary>
internal (int Index, int Slot, bool NeedsWrite) Resolve()
{
if (!_dirty && _active >= 0)
return (_active, _entries[_active].Slot, false);
for (int i = 0; i < _entries.Count; i++)
{
if (!Matches(_entries[i]))
continue;
if (i >= _liveCount)
{
(_entries[i], _entries[_liveCount]) = (_entries[_liveCount], _entries[i]);
_active = _liveCount;
_liveCount++;
}
else
{
_active = i;
}
_dirty = false;
return (_active, _entries[_active].Slot, false);
}
while (_entries.Count <= _liveCount)
_entries.Add(new Entry(_storageBindingCount, _uniformBindingCount));
Entry target = _entries[_liveCount];
Adopt(target);
_active = _liveCount;
_liveCount++;
_dirty = false;
return (_active, target.Slot, true);
}
/// <summary>Records the caller's descriptor-set-pair handle for a newly materialised entry.</summary>
internal void AssignSlot(int index, int slot) => _entries[index].Slot = slot;
private bool Matches(Entry entry)
{
if (entry.Slot < 0)
return false;
for (uint binding = 0; binding < _storageBindingCount; binding++)
{
if (entry.StorageBuffers[binding] != _storageBuffers[binding])
return false;
if (entry.StorageRanges[binding] != _storageRanges[binding])
return false;
if (!_isDynamicStorage(binding) && entry.StorageOffsets[binding] != _storageOffsets[binding])
return false;
}
for (uint binding = 0; binding < _uniformBindingCount; binding++)
{
if (entry.UniformBuffers[binding] != _uniformBuffers[binding])
return false;
if (entry.UniformRanges[binding] != _uniformRanges[binding])
return false;
}
return true;
}
private void Adopt(Entry entry)
{
Array.Copy(_storageBuffers, entry.StorageBuffers, _storageBindingCount);
Array.Copy(_storageOffsets, entry.StorageOffsets, _storageBindingCount);
Array.Copy(_storageRanges, entry.StorageRanges, _storageBindingCount);
Array.Copy(_uniformBuffers, entry.UniformBuffers, _uniformBindingCount);
Array.Copy(_uniformRanges, entry.UniformRanges, _uniformBindingCount);
}
}

View file

@ -30,24 +30,33 @@ namespace AcDream.App.Rendering.Gpu.Vk;
/// descriptor set layout and one pipeline layout rather than a permutation per
/// renderer — plan §4.4's requirement, and the thing that makes switching
/// pipelines mid-pass free.</para>
///
/// <para><b>Slice V6i: one set pair per renderer scope.</b> There is no longer a
/// single (set 0, set 1) pair per flight slot; there is an arena of them, and
/// <see cref="VulkanBindingScopeArena"/> decides which pair a bind belongs to and
/// whether its descriptors have to be written. That is what closes plan §5.5.8's
/// recorded one-binding-two-buffers hazard before the world arm fires it — see
/// the arena's own documentation for why the scope is derived from the
/// descriptor state rather than declared by the renderer.</para>
/// </summary>
internal sealed unsafe class VulkanFrameBindings : IDisposable
{
private readonly Silk.NET.Vulkan.Vk _vk;
private readonly Device _device;
private readonly DescriptorPool _pool;
private readonly DescriptorSet _storageSet;
private readonly DescriptorSet _uniformSet;
private readonly uint[] _storageOffsets = new uint[GpuBindingModel.StorageBindingCount];
private readonly uint[] _uniformOffsets = new uint[UniformBindingCount];
private readonly Silk.NET.Vulkan.Buffer[] _storageBuffers =
new Silk.NET.Vulkan.Buffer[GpuBindingModel.StorageBindingCount];
private readonly Silk.NET.Vulkan.Buffer[] _uniformBuffers =
new Silk.NET.Vulkan.Buffer[UniformBindingCount];
private readonly VulkanPipelineLayouts.Created _layouts;
private readonly VulkanBindingScopeArena _arena;
private readonly List<DescriptorPool> _pools = [];
private readonly List<(DescriptorSet Storage, DescriptorSet Uniform)> _sets = [];
private bool _disposed;
/// <summary>
/// Set pairs one descriptor pool serves. Distinct descriptor states in a
/// frame are the world renderers plus the retained UI, so this is generous;
/// exceeding it allocates another pool rather than failing.
/// </summary>
private const int PairsPerPool = 16;
/// <summary>
/// Set 0's dynamic-offset slots, in binding order — the order
/// <c>vkCmdBindDescriptorSets</c> requires. A plain binding has no slot.
@ -92,61 +101,24 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
{
_vk = vk ?? throw new ArgumentNullException(nameof(vk));
_device = device;
ArgumentNullException.ThrowIfNull(layouts);
_layouts = layouts ?? throw new ArgumentNullException(nameof(layouts));
ArgumentNullException.ThrowIfNull(ring);
ArgumentNullException.ThrowIfNull(dummy);
DescriptorPoolSize* sizes = stackalloc DescriptorPoolSize[3];
sizes[0] = new DescriptorPoolSize
{
Type = DescriptorType.StorageBufferDynamic,
DescriptorCount = VulkanPipelineLayouts.DynamicStorageBindingCount,
};
sizes[1] = new DescriptorPoolSize
{
Type = DescriptorType.StorageBuffer,
DescriptorCount =
GpuBindingModel.StorageBindingCount - VulkanPipelineLayouts.DynamicStorageBindingCount,
};
sizes[2] = new DescriptorPoolSize
{
Type = DescriptorType.UniformBufferDynamic,
DescriptorCount = UniformBindingCount,
};
var poolCreate = new DescriptorPoolCreateInfo
{
SType = StructureType.DescriptorPoolCreateInfo,
MaxSets = 2,
PoolSizeCount = 3,
PPoolSizes = sizes,
};
VulkanInterop.Check(
_vk.CreateDescriptorPool(_device, &poolCreate, null, out _pool),
"vkCreateDescriptorPool (frame bindings)");
_storageSet = Allocate(layouts.Storage);
_uniformSet = Allocate(layouts.Uniform);
for (uint binding = 0; binding < GpuBindingModel.StorageBindingCount; binding++)
{
_storageBuffers[binding] = dummy.Handle;
WriteStorage(
binding,
dummy.Handle,
offsetBytes: 0,
(uint)Math.Min(dummy.SizeBytes, MaxBindingRangeBytes),
VulkanPipelineLayouts.IsDynamicStorageBinding(binding));
}
// Only the two bindings the layout declares exist; the rest of the
// array is bookkeeping so the offsets stay index-aligned.
WriteUniform(GpuBindingModel.UniformSceneLighting, dummy.Handle, (uint)Math.Min(dummy.SizeBytes, 65536));
WriteUniform(GpuBindingModel.UniformTerrainTiling, dummy.Handle, (uint)Math.Min(dummy.SizeBytes, 65536));
_uniformBuffers[GpuBindingModel.UniformSceneLighting] = dummy.Handle;
_uniformBuffers[GpuBindingModel.UniformTerrainTiling] = dummy.Handle;
Ring = ring;
Dummy = dummy;
_arena = new VulkanBindingScopeArena(
(int)GpuBindingModel.StorageBindingCount,
UniformBindingCount,
VulkanPipelineLayouts.IsDynamicStorageBinding);
uint dummyStorageRange = (uint)Math.Min(dummy.SizeBytes, MaxBindingRangeBytes);
for (uint binding = 0; binding < GpuBindingModel.StorageBindingCount; binding++)
_arena.SeedStorage(binding, dummy.Handle.Handle, offsetBytes: 0, dummyStorageRange);
uint dummyUniformRange = (uint)Math.Min(dummy.SizeBytes, 65536);
for (uint binding = 0; binding < UniformBindingCount; binding++)
_arena.SeedUniform(binding, dummy.Handle.Handle, dummyUniformRange);
}
internal VulkanGpuBuffer Ring { get; }
@ -154,51 +126,57 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
internal VulkanGpuBuffer Dummy { get; }
/// <summary>
/// Points a storage binding at a range.
///
/// <para>A DYNAMIC binding re-writes its descriptor only when the BUFFER
/// changes; the offset rides the bind call. A PLAIN binding has no such
/// channel, so the descriptor itself carries the offset and is re-written
/// when either moves.</para>
/// Diagnostic: distinct descriptor states this flight slot materialised. One
/// per renderer scope in a steady frame, so a number that keeps climbing is
/// a renderer pointing a binding at a fresh buffer every draw.
/// </summary>
internal int ScopeCount => _arena.Count;
/// <summary>
/// Recycles the arena for a new frame on this slot. Safe because the slot's
/// previous submission has retired before <c>BeginFrame</c> returns, which is
/// the same guarantee that lets the ring rewind.
/// </summary>
internal void BeginFrame() => _arena.BeginFrame();
internal void SetStorage(uint binding, VulkanGpuBuffer buffer, uint offsetBytes, uint sizeBytes)
{
ArgumentOutOfRangeException.ThrowIfGreaterThanOrEqual(binding, GpuBindingModel.StorageBindingCount);
bool dynamic = VulkanPipelineLayouts.IsDynamicStorageBinding(binding);
bool bufferChanged = _storageBuffers[binding].Handle != buffer.Handle.Handle;
bool offsetChanged = _storageOffsets[binding] != offsetBytes;
if (bufferChanged || (!dynamic && offsetChanged))
{
_storageBuffers[binding] = buffer.Handle;
WriteStorage(
binding,
buffer.Handle,
dynamic ? 0 : offsetBytes,
ClampRange(buffer, sizeBytes, dynamic ? 0 : offsetBytes),
dynamic);
}
_storageOffsets[binding] = offsetBytes;
uint descriptorOffset = VulkanPipelineLayouts.IsDynamicStorageBinding(binding) ? 0 : offsetBytes;
_arena.SetStorage(
binding,
buffer.Handle.Handle,
offsetBytes,
ClampRange(buffer, sizeBytes, descriptorOffset));
}
internal void SetUniform(uint binding, VulkanGpuBuffer buffer, uint offsetBytes, uint sizeBytes)
{
ArgumentOutOfRangeException.ThrowIfGreaterThanOrEqual(binding, (uint)UniformBindingCount);
if (_uniformBuffers[binding].Handle != buffer.Handle.Handle)
{
_uniformBuffers[binding] = buffer.Handle;
WriteUniform(binding, buffer.Handle, Math.Min(ClampRange(buffer, sizeBytes, offsetBytes: 0), 65536));
}
_uniformOffsets[binding] = offsetBytes;
_arena.SetUniform(
binding,
buffer.Handle.Handle,
offsetBytes,
Math.Min(ClampRange(buffer, sizeBytes, offsetBytes: 0), 65536));
}
/// <summary>Binds all three sets with the current dynamic offsets.</summary>
internal void Bind(CommandBuffer commands, VulkanGpuDevice device)
{
(int index, int slot, bool needsWrite) = _arena.Resolve();
if (slot < 0)
{
slot = _sets.Count;
_sets.Add(AllocatePair());
_arena.AssignSlot(index, slot);
}
if (needsWrite)
WritePair(_sets[slot]);
DescriptorSet* sets = stackalloc DescriptorSet[3];
sets[0] = _storageSet;
sets[1] = _uniformSet;
sets[0] = _sets[slot].Storage;
sets[1] = _sets[slot].Uniform;
sets[2] = device.TextureTable.Set;
int dynamicCount = DynamicStorageBindings.Length + 2;
@ -206,9 +184,9 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
// Dynamic offsets are ordered by set, then by binding number, and only
// the DYNAMIC descriptors have a slot at all.
for (int i = 0; i < DynamicStorageBindings.Length; i++)
offsets[i] = _storageOffsets[DynamicStorageBindings[i]];
offsets[DynamicStorageBindings.Length + 0] = _uniformOffsets[GpuBindingModel.UniformSceneLighting];
offsets[DynamicStorageBindings.Length + 1] = _uniformOffsets[GpuBindingModel.UniformTerrainTiling];
offsets[i] = _arena.StorageOffset(DynamicStorageBindings[i]);
offsets[DynamicStorageBindings.Length + 0] = _arena.UniformOffset(GpuBindingModel.UniformSceneLighting);
offsets[DynamicStorageBindings.Length + 1] = _arena.UniformOffset(GpuBindingModel.UniformTerrainTiling);
_vk.CmdBindDescriptorSets(
commands,
@ -221,6 +199,74 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
offsets);
}
private void WritePair((DescriptorSet Storage, DescriptorSet Uniform) pair)
{
for (uint binding = 0; binding < GpuBindingModel.StorageBindingCount; binding++)
{
WriteStorage(
pair.Storage,
binding,
new Silk.NET.Vulkan.Buffer(_arena.StorageBuffer(binding)),
_arena.StorageDescriptorOffset(binding),
_arena.StorageRange(binding),
VulkanPipelineLayouts.IsDynamicStorageBinding(binding));
}
// Only the two bindings the layout declares exist; the rest of the array
// is bookkeeping so the offsets stay index-aligned.
WriteUniform(
pair.Uniform,
GpuBindingModel.UniformSceneLighting,
new Silk.NET.Vulkan.Buffer(_arena.UniformBuffer(GpuBindingModel.UniformSceneLighting)),
_arena.UniformRange(GpuBindingModel.UniformSceneLighting));
WriteUniform(
pair.Uniform,
GpuBindingModel.UniformTerrainTiling,
new Silk.NET.Vulkan.Buffer(_arena.UniformBuffer(GpuBindingModel.UniformTerrainTiling)),
_arena.UniformRange(GpuBindingModel.UniformTerrainTiling));
}
private (DescriptorSet Storage, DescriptorSet Uniform) AllocatePair()
{
if (_sets.Count % PairsPerPool == 0)
_pools.Add(CreatePool());
DescriptorPool pool = _pools[^1];
return (Allocate(pool, _layouts.Storage), Allocate(pool, _layouts.Uniform));
}
private DescriptorPool CreatePool()
{
DescriptorPoolSize* sizes = stackalloc DescriptorPoolSize[3];
sizes[0] = new DescriptorPoolSize
{
Type = DescriptorType.StorageBufferDynamic,
DescriptorCount = VulkanPipelineLayouts.DynamicStorageBindingCount * PairsPerPool,
};
sizes[1] = new DescriptorPoolSize
{
Type = DescriptorType.StorageBuffer,
DescriptorCount =
(GpuBindingModel.StorageBindingCount - VulkanPipelineLayouts.DynamicStorageBindingCount)
* PairsPerPool,
};
sizes[2] = new DescriptorPoolSize
{
Type = DescriptorType.UniformBufferDynamic,
DescriptorCount = DynamicUniformBindingCount * PairsPerPool,
};
var poolCreate = new DescriptorPoolCreateInfo
{
SType = StructureType.DescriptorPoolCreateInfo,
MaxSets = 2 * PairsPerPool,
PoolSizeCount = 3,
PPoolSizes = sizes,
};
VulkanInterop.Check(
_vk.CreateDescriptorPool(_device, &poolCreate, null, out DescriptorPool pool),
"vkCreateDescriptorPool (frame bindings)");
return pool;
}
private static uint ClampRange(VulkanGpuBuffer buffer, uint requested, uint offsetBytes)
{
long remaining = buffer.SizeBytes - offsetBytes;
@ -237,13 +283,13 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
return requested == 0 ? available : Math.Min(Math.Max(requested, 16), available);
}
private DescriptorSet Allocate(DescriptorSetLayout layout)
private DescriptorSet Allocate(DescriptorPool pool, DescriptorSetLayout layout)
{
DescriptorSetLayout handle = layout;
var allocate = new DescriptorSetAllocateInfo
{
SType = StructureType.DescriptorSetAllocateInfo,
DescriptorPool = _pool,
DescriptorPool = pool,
DescriptorSetCount = 1,
PSetLayouts = &handle,
};
@ -254,6 +300,7 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
}
private void WriteStorage(
DescriptorSet set,
uint binding,
Silk.NET.Vulkan.Buffer buffer,
uint offsetBytes,
@ -269,7 +316,7 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
var write = new WriteDescriptorSet
{
SType = StructureType.WriteDescriptorSet,
DstSet = _storageSet,
DstSet = set,
DstBinding = binding,
DescriptorCount = 1,
DescriptorType = dynamic
@ -280,7 +327,11 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
_vk.UpdateDescriptorSets(_device, 1, &write, 0, null);
}
private void WriteUniform(uint binding, Silk.NET.Vulkan.Buffer buffer, uint rangeBytes)
private void WriteUniform(
DescriptorSet set,
uint binding,
Silk.NET.Vulkan.Buffer buffer,
uint rangeBytes)
{
var info = new DescriptorBufferInfo
{
@ -291,7 +342,7 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
var write = new WriteDescriptorSet
{
SType = StructureType.WriteDescriptorSet,
DstSet = _uniformSet,
DstSet = set,
DstBinding = binding,
DescriptorCount = 1,
DescriptorType = DescriptorType.UniformBufferDynamic,
@ -305,7 +356,13 @@ internal sealed unsafe class VulkanFrameBindings : IDisposable
if (_disposed)
return;
_disposed = true;
if (_pool.Handle != 0)
_vk.DestroyDescriptorPool(_device, _pool, null);
foreach (DescriptorPool pool in _pools)
{
if (pool.Handle != 0)
_vk.DestroyDescriptorPool(_device, pool, null);
}
_pools.Clear();
_sets.Clear();
}
}

View file

@ -114,6 +114,9 @@ internal sealed unsafe partial class VulkanGpuDevice
private void BeginFrameResources(int slotIndex) => _timerPool?.BeginSlot(slotIndex);
/// <summary>Slice V6i: the descriptor-set arena for one flight slot.</summary>
private VulkanFrameBindings FrameBindingsAt(int slotIndex) => _frameBindings[slotIndex];
private void EndFrameResources(int slotIndex, CommandBuffer commands)
{
_ = slotIndex;

View file

@ -322,6 +322,10 @@ internal sealed unsafe partial class VulkanGpuDevice : IGpuDevice
_uploads.ReleaseCompleted(CompletedSerial());
_ringStates[slot].Reset();
// Campaign V slice V6i: the descriptor-set arena rewinds with the ring,
// and for the same reason — this slot's previous submission has retired,
// so no recorded draw can still be reading the sets it hands out.
FrameBindingsAt(slot).BeginFrame();
_acquiredImageIndex = null;
if (_backbuffer is not null)

View file

@ -0,0 +1,235 @@
using AcDream.App.Rendering.Gpu;
using AcDream.App.Rendering.Gpu.Vk;
namespace AcDream.App.Tests.Rendering.Gpu.Vk;
/// <summary>
/// Campaign V slice V6i. Plan §5.5.8's second recorded gap was that pointing one
/// binding at a second buffer within a frame silently corrupts the draws already
/// recorded against it, because a descriptor set is read when the command buffer
/// executes and not when it was recorded. These tests pin the property that
/// closes it: within one frame, an entry a draw is already bound to is never
/// rewritten.
/// </summary>
public sealed class VulkanBindingScopeArenaTests
{
private const ulong RingBuffer = 0x1000;
private const ulong DummyBuffer = 0x2000;
private const ulong DispatcherBuffer = 0x3000;
private const ulong EnvCellBuffer = 0x4000;
private static VulkanBindingScopeArena CreateArena()
{
var arena = new VulkanBindingScopeArena(
(int)GpuBindingModel.StorageBindingCount,
VulkanFrameBindings.UniformBindingCount,
VulkanPipelineLayouts.IsDynamicStorageBinding);
for (uint binding = 0; binding < GpuBindingModel.StorageBindingCount; binding++)
arena.SeedStorage(binding, DummyBuffer, offsetBytes: 0, rangeBytes: 65536);
for (uint binding = 0; binding < VulkanFrameBindings.UniformBindingCount; binding++)
arena.SeedUniform(binding, DummyBuffer, rangeBytes: 65536);
return arena;
}
[Fact]
public void TwoRenderersBindingTheSameBindingToDifferentBuffersGetDifferentEntries()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
arena.SetStorage(GpuBindingModel.StorageInstances, DispatcherBuffer, 0, 4096);
(int firstIndex, _, bool firstWrite) = arena.Resolve();
arena.AssignSlot(firstIndex, slot: 0);
Assert.True(firstWrite);
arena.SetStorage(GpuBindingModel.StorageInstances, EnvCellBuffer, 0, 4096);
(int secondIndex, _, bool secondWrite) = arena.Resolve();
arena.AssignSlot(secondIndex, slot: 1);
Assert.True(secondWrite);
Assert.NotEqual(firstIndex, secondIndex);
Assert.Equal(2, arena.Count);
Assert.Equal(2, arena.LiveCount);
}
[Fact]
public void ReturningToAnEarlierRenderersBuffersReusesItsEntryWithoutRewriting()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
arena.SetStorage(GpuBindingModel.StorageInstances, DispatcherBuffer, 0, 4096);
(int dispatcher, _, _) = arena.Resolve();
arena.AssignSlot(dispatcher, slot: 0);
arena.SetStorage(GpuBindingModel.StorageInstances, EnvCellBuffer, 0, 4096);
(int envCell, _, _) = arena.Resolve();
arena.AssignSlot(envCell, slot: 1);
arena.SetStorage(GpuBindingModel.StorageInstances, DispatcherBuffer, 0, 4096);
(int again, int slot, bool needsWrite) = arena.Resolve();
Assert.Equal(dispatcher, again);
Assert.Equal(0, slot);
Assert.False(needsWrite);
Assert.Equal(2, arena.Count);
}
[Fact]
public void ADynamicBindingsMovingOffsetCostsNoNewEntryAndNoWrite()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
arena.SetStorage(GpuBindingModel.StorageInstances, RingBuffer, 0, 4096);
(int index, _, _) = arena.Resolve();
arena.AssignSlot(index, slot: 0);
arena.SetStorage(GpuBindingModel.StorageInstances, RingBuffer, 8192, 4096);
(int second, _, bool needsWrite) = arena.Resolve();
Assert.Equal(index, second);
Assert.False(needsWrite);
Assert.Equal(1, arena.Count);
Assert.Equal(8192u, arena.StorageOffset(GpuBindingModel.StorageInstances));
// A dynamic descriptor addresses the whole range and slides with the
// bind call, so its descriptor offset stays zero.
Assert.Equal(0u, arena.StorageDescriptorOffset(GpuBindingModel.StorageInstances));
}
[Fact]
public void APlainBindingsMovingOffsetIsDescriptorVisibleAndTakesANewEntry()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
// Binding 4 (global lights) is plain: V6g's rule keeps only ring-fed
// bindings dynamic, so a plain binding's offset lives in its descriptor.
Assert.False(VulkanPipelineLayouts.IsDynamicStorageBinding(GpuBindingModel.StorageGlobalLights));
arena.SetStorage(GpuBindingModel.StorageGlobalLights, RingBuffer, 0, 256);
(int first, _, _) = arena.Resolve();
arena.AssignSlot(first, slot: 0);
arena.SetStorage(GpuBindingModel.StorageGlobalLights, RingBuffer, 512, 256);
(int second, _, bool needsWrite) = arena.Resolve();
arena.AssignSlot(second, slot: 1);
Assert.NotEqual(first, second);
Assert.True(needsWrite);
Assert.Equal(512u, arena.StorageDescriptorOffset(GpuBindingModel.StorageGlobalLights));
}
[Fact]
public void ASteadyFrameRewritesNothing()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
arena.SetStorage(GpuBindingModel.StorageInstances, DispatcherBuffer, 0, 4096);
(int index, _, bool firstWrite) = arena.Resolve();
arena.AssignSlot(index, slot: 0);
Assert.True(firstWrite);
// The next frame on the same flight slot binds the same buffers. The
// previous frame has retired, so its entry is free to reclaim — and its
// descriptors already say exactly this, so nothing is written.
arena.BeginFrame();
arena.SetStorage(GpuBindingModel.StorageInstances, DispatcherBuffer, 0, 4096);
(int reused, int slot, bool needsWrite) = arena.Resolve();
Assert.Equal(0, reused);
Assert.Equal(0, slot);
Assert.False(needsWrite);
Assert.Equal(1, arena.Count);
}
[Fact]
public void AnEntryReclaimedFromTheLastFrameIsProtectedForTheRestOfThisOne()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
arena.SetStorage(GpuBindingModel.StorageInstances, DispatcherBuffer, 0, 4096);
(int dispatcher, _, _) = arena.Resolve();
arena.AssignSlot(dispatcher, slot: 0);
arena.SetStorage(GpuBindingModel.StorageInstances, EnvCellBuffer, 0, 4096);
(int envCell, _, _) = arena.Resolve();
arena.AssignSlot(envCell, slot: 1);
// Next frame, in the opposite order. The EnvCell entry is matched from
// above the cursor and swapped down; the dispatcher must then get the
// OTHER entry rather than overwriting the one EnvCell's draw is bound to.
arena.BeginFrame();
arena.SetStorage(GpuBindingModel.StorageInstances, EnvCellBuffer, 0, 4096);
(int reusedEnvCell, int envCellSlot, bool envCellWrite) = arena.Resolve();
arena.SetStorage(GpuBindingModel.StorageInstances, DispatcherBuffer, 0, 4096);
(int reusedDispatcher, int dispatcherSlot, bool dispatcherWrite) = arena.Resolve();
Assert.False(envCellWrite);
Assert.False(dispatcherWrite);
Assert.NotEqual(reusedEnvCell, reusedDispatcher);
Assert.NotEqual(envCellSlot, dispatcherSlot);
Assert.Equal(2, arena.Count);
Assert.Equal(2, arena.LiveCount);
}
[Fact]
public void AThirdDistinctStateInOneFrameMaterialisesAThirdEntry()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
Bind(arena, DispatcherBuffer, slot: 0);
Bind(arena, EnvCellBuffer, slot: 1);
Bind(arena, RingBuffer, slot: 2);
Assert.Equal(3, arena.Count);
Assert.Equal(3, arena.LiveCount);
static void Bind(VulkanBindingScopeArena arena, ulong buffer, int slot)
{
arena.SetStorage(GpuBindingModel.StorageInstances, buffer, 0, 4096);
(int index, int existing, bool needsWrite) = arena.Resolve();
Assert.True(needsWrite);
Assert.Equal(-1, existing);
arena.AssignSlot(index, slot);
}
}
[Fact]
public void AUniformBufferChangeAlsoSeparatesScopes()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
arena.SetUniform(GpuBindingModel.UniformSceneLighting, DispatcherBuffer, 0, 256);
(int first, _, _) = arena.Resolve();
arena.AssignSlot(first, slot: 0);
arena.SetUniform(GpuBindingModel.UniformTerrainTiling, EnvCellBuffer, 0, 576);
(int second, _, bool needsWrite) = arena.Resolve();
arena.AssignSlot(second, slot: 1);
Assert.NotEqual(first, second);
Assert.True(needsWrite);
}
[Fact]
public void AUniformBuffersMovingOffsetCostsNoNewEntry()
{
VulkanBindingScopeArena arena = CreateArena();
arena.BeginFrame();
arena.SetUniform(GpuBindingModel.UniformSceneLighting, RingBuffer, 0, 256);
(int index, _, _) = arena.Resolve();
arena.AssignSlot(index, slot: 0);
arena.SetUniform(GpuBindingModel.UniformSceneLighting, RingBuffer, 1024, 256);
(int second, _, bool needsWrite) = arena.Resolve();
Assert.Equal(index, second);
Assert.False(needsWrite);
Assert.Equal(1024u, arena.UniformOffset(GpuBindingModel.UniformSceneLighting));
}
}