Mirrors/Ryujinx
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Vulkan: Defer guest barriers, and improve image barrier timings (#7012)

Browse source 
* More guarantees for buffer correct placement, defer guest requested buffers

* Split RP on indirect barrier rn

* Better handling for feedback loops.

* Qualcomm barriers suck too

* Fix condition

* Remove unused field

* Allow render pass barriers on turnip for now
This commit is contained in:
riperiperi 2024-07-18 00:21:32 +01:00 committed by GitHub
parent f77bebac80
commit 1a919e99b2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
18 changed files with 452 additions and 156 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
src/Ryujinx.Graphics.GAL/ResourceLayout.cs
View file

@ -74,13 +74,15 @@ namespace Ryujinx.Graphics.GAL
public int ArrayLength { get; }
public ResourceType Type { get; }
public ResourceStages Stages { get; }
public bool Write { get; }
public ResourceUsage(int binding, int arrayLength, ResourceType type, ResourceStages stages)
public ResourceUsage(int binding, int arrayLength, ResourceType type, ResourceStages stages, bool write)
{
Binding = binding;
ArrayLength = arrayLength;
Type = type;
Stages = stages;
Write = write;
}
public override int GetHashCode()

24
src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs
View file

@ -78,9 +78,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
ResourceStages stages = vertexAsCompute ? ResourceStages.Compute | ResourceStages.Vertex : VtgStages;
PopulateDescriptorAndUsages(stages, ResourceType.UniformBuffer, uniformSetIndex, 1, rrc.ReservedConstantBuffers - 1);
PopulateDescriptorAndUsages(stages, ResourceType.StorageBuffer, storageSetIndex, 0, rrc.ReservedStorageBuffers);
PopulateDescriptorAndUsages(stages, ResourceType.StorageBuffer, storageSetIndex, 0, rrc.ReservedStorageBuffers, true);
PopulateDescriptorAndUsages(stages, ResourceType.BufferTexture, textureSetIndex, 0, rrc.ReservedTextures);
PopulateDescriptorAndUsages(stages, ResourceType.BufferImage, imageSetIndex, 0, rrc.ReservedImages);
PopulateDescriptorAndUsages(stages, ResourceType.BufferImage, imageSetIndex, 0, rrc.ReservedImages, true);
}
/// <summary>
@ -91,10 +91,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="setIndex">Resource set index where the resources are used</param>
/// <param name="start">First binding number</param>
/// <param name="count">Amount of bindings</param>
private void PopulateDescriptorAndUsages(ResourceStages stages, ResourceType type, int setIndex, int start, int count)
/// <param name="write">True if the binding is written from the shader, false otherwise</param>
private void PopulateDescriptorAndUsages(ResourceStages stages, ResourceType type, int setIndex, int start, int count, bool write = false)
{
AddDescriptor(stages, type, setIndex, start, count);
AddUsage(stages, type, setIndex, start, count);
AddUsage(stages, type, setIndex, start, count, write);
}
/// <summary>
@ -216,11 +217,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="setIndex">Descriptor set number where the resource will be bound</param>
/// <param name="binding">Binding number where the resource will be bound</param>
/// <param name="count">Number of resources bound at the binding location</param>
private void AddUsage(ResourceStages stages, ResourceType type, int setIndex, int binding, int count)
/// <param name="write">True if the binding is written from the shader, false otherwise</param>
private void AddUsage(ResourceStages stages, ResourceType type, int setIndex, int binding, int count, bool write = false)
{
for (int index = 0; index < count; index++)
{
_resourceUsages[setIndex].Add(new ResourceUsage(binding + index, 1, type, stages));
_resourceUsages[setIndex].Add(new ResourceUsage(binding + index, 1, type, stages, write));
}
}
@ -238,7 +240,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
buffer.Binding,
1,
isStorage ? ResourceType.StorageBuffer : ResourceType.UniformBuffer,
stages));
stages,
buffer.Flags.HasFlag(BufferUsageFlags.Write)));
}
}
@ -254,7 +257,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
{
ResourceType type = GetTextureResourceType(texture, isImage);
GetUsages(texture.Set).Add(new ResourceUsage(texture.Binding, texture.ArrayLength, type, stages));
GetUsages(texture.Set).Add(new ResourceUsage(
texture.Binding,
texture.ArrayLength,
type,
stages,
texture.Flags.HasFlag(TextureUsageFlags.ImageStore)));
}
}

260
src/Ryujinx.Graphics.Vulkan/BarrierBatch.cs
View file

@ -1,6 +1,7 @@
using Silk.NET.Vulkan;
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
namespace Ryujinx.Graphics.Vulkan
{
@ -8,22 +9,64 @@ namespace Ryujinx.Graphics.Vulkan
{
private const int MaxBarriersPerCall = 16;
private const AccessFlags BaseAccess = AccessFlags.ShaderReadBit | AccessFlags.ShaderWriteBit;
private const AccessFlags BufferAccess = AccessFlags.IndexReadBit | AccessFlags.VertexAttributeReadBit | AccessFlags.UniformReadBit;
private const AccessFlags CommandBufferAccess = AccessFlags.IndirectCommandReadBit;
private readonly VulkanRenderer _gd;
private readonly NativeArray<MemoryBarrier> _memoryBarrierBatch = new(MaxBarriersPerCall);
private readonly NativeArray<BufferMemoryBarrier> _bufferBarrierBatch = new(MaxBarriersPerCall);
private readonly NativeArray<ImageMemoryBarrier> _imageBarrierBatch = new(MaxBarriersPerCall);
private readonly List<BarrierWithStageFlags<MemoryBarrier>> _memoryBarriers = new();
private readonly List<BarrierWithStageFlags<BufferMemoryBarrier>> _bufferBarriers = new();
private readonly List<BarrierWithStageFlags<ImageMemoryBarrier>> _imageBarriers = new();
private readonly List<BarrierWithStageFlags<MemoryBarrier, int>> _memoryBarriers = new();
private readonly List<BarrierWithStageFlags<BufferMemoryBarrier, int>> _bufferBarriers = new();
private readonly List<BarrierWithStageFlags<ImageMemoryBarrier, TextureStorage>> _imageBarriers = new();
private int _queuedBarrierCount;
private enum IncoherentBarrierType
{
None,
Texture,
All,
CommandBuffer
}
private PipelineStageFlags _incoherentBufferWriteStages;
private PipelineStageFlags _incoherentTextureWriteStages;
private PipelineStageFlags _extraStages;
private IncoherentBarrierType _queuedIncoherentBarrier;
public BarrierBatch(VulkanRenderer gd)
{
_gd = gd;
}
public static (AccessFlags Access, PipelineStageFlags Stages) GetSubpassAccessSuperset(VulkanRenderer gd)
{
AccessFlags access = BufferAccess;
PipelineStageFlags stages = PipelineStageFlags.AllGraphicsBit;
if (gd.TransformFeedbackApi != null)
{
access |= AccessFlags.TransformFeedbackWriteBitExt;
stages |= PipelineStageFlags.TransformFeedbackBitExt;
}
if (!gd.IsTBDR)
{
// Desktop GPUs can transform image barriers into memory barriers.
access |= AccessFlags.DepthStencilAttachmentWriteBit | AccessFlags.ColorAttachmentWriteBit;
access |= AccessFlags.DepthStencilAttachmentReadBit | AccessFlags.ColorAttachmentReadBit;
stages |= PipelineStageFlags.EarlyFragmentTestsBit | PipelineStageFlags.LateFragmentTestsBit;
stages |= PipelineStageFlags.ColorAttachmentOutputBit;
}
return (access, stages);
}
private readonly record struct StageFlags : IEquatable<StageFlags>
{
public readonly PipelineStageFlags Source;
@ -36,47 +79,130 @@ namespace Ryujinx.Graphics.Vulkan
}
}
private readonly struct BarrierWithStageFlags<T> where T : unmanaged
private readonly struct BarrierWithStageFlags<T, T2> where T : unmanaged
{
public readonly StageFlags Flags;
public readonly T Barrier;
public readonly T2 Resource;
public BarrierWithStageFlags(StageFlags flags, T barrier)
{
Flags = flags;
Barrier = barrier;
Resource = default;
}
public BarrierWithStageFlags(PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags, T barrier)
public BarrierWithStageFlags(PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags, T barrier, T2 resource)
{
Flags = new StageFlags(srcStageFlags, dstStageFlags);
Barrier = barrier;
Resource = resource;
}
}
private void QueueBarrier<T>(List<BarrierWithStageFlags<T>> list, T barrier, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags) where T : unmanaged
private void QueueBarrier<T, T2>(List<BarrierWithStageFlags<T, T2>> list, T barrier, T2 resource, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags) where T : unmanaged
{
list.Add(new BarrierWithStageFlags<T>(srcStageFlags, dstStageFlags, barrier));
list.Add(new BarrierWithStageFlags<T, T2>(srcStageFlags, dstStageFlags, barrier, resource));
_queuedBarrierCount++;
}
public void QueueBarrier(MemoryBarrier barrier, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags)
{
QueueBarrier(_memoryBarriers, barrier, srcStageFlags, dstStageFlags);
QueueBarrier(_memoryBarriers, barrier, default, srcStageFlags, dstStageFlags);
}
public void QueueBarrier(BufferMemoryBarrier barrier, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags)
{
QueueBarrier(_bufferBarriers, barrier, srcStageFlags, dstStageFlags);
QueueBarrier(_bufferBarriers, barrier, default, srcStageFlags, dstStageFlags);
}
public void QueueBarrier(ImageMemoryBarrier barrier, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags)
public void QueueBarrier(ImageMemoryBarrier barrier, TextureStorage resource, PipelineStageFlags srcStageFlags, PipelineStageFlags dstStageFlags)
{
QueueBarrier(_imageBarriers, barrier, srcStageFlags, dstStageFlags);
QueueBarrier(_imageBarriers, barrier, resource, srcStageFlags, dstStageFlags);
}
public unsafe void Flush(CommandBuffer cb, bool insideRenderPass, Action endRenderPass)
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public unsafe void FlushMemoryBarrier(ShaderCollection program, bool inRenderPass)
{
if (_queuedIncoherentBarrier > IncoherentBarrierType.None)
{
// We should emit a memory barrier if there's a write access in the program (current program, or program since last barrier)
bool hasTextureWrite = _incoherentTextureWriteStages != PipelineStageFlags.None;
bool hasBufferWrite = _incoherentBufferWriteStages != PipelineStageFlags.None;
bool hasBufferBarrier = _queuedIncoherentBarrier > IncoherentBarrierType.Texture;
if (hasTextureWrite || (hasBufferBarrier && hasBufferWrite))
{
AccessFlags access = BaseAccess;
PipelineStageFlags stages = inRenderPass ? PipelineStageFlags.AllGraphicsBit : PipelineStageFlags.AllCommandsBit;
if (hasBufferBarrier && hasBufferWrite)
{
access |= BufferAccess;
if (_gd.TransformFeedbackApi != null)
{
access |= AccessFlags.TransformFeedbackWriteBitExt;
stages |= PipelineStageFlags.TransformFeedbackBitExt;
}
}
if (_queuedIncoherentBarrier == IncoherentBarrierType.CommandBuffer)
{
access |= CommandBufferAccess;
stages |= PipelineStageFlags.DrawIndirectBit;
}
MemoryBarrier barrier = new MemoryBarrier()
{
SType = StructureType.MemoryBarrier,
SrcAccessMask = access,
DstAccessMask = access
};
QueueBarrier(barrier, stages, stages);
_incoherentTextureWriteStages = program?.IncoherentTextureWriteStages ?? PipelineStageFlags.None;
if (_queuedIncoherentBarrier > IncoherentBarrierType.Texture)
{
if (program != null)
{
_incoherentBufferWriteStages = program.IncoherentBufferWriteStages | _extraStages;
}
else
{
_incoherentBufferWriteStages = PipelineStageFlags.None;
}
}
_queuedIncoherentBarrier = IncoherentBarrierType.None;
}
}
}
public unsafe void Flush(CommandBufferScoped cbs, bool inRenderPass, RenderPassHolder rpHolder, Action endRenderPass)
{
Flush(cbs, null, inRenderPass, rpHolder, endRenderPass);
}
public unsafe void Flush(CommandBufferScoped cbs, ShaderCollection program, bool inRenderPass, RenderPassHolder rpHolder, Action endRenderPass)
{
if (program != null)
{
_incoherentBufferWriteStages |= program.IncoherentBufferWriteStages | _extraStages;
_incoherentTextureWriteStages |= program.IncoherentTextureWriteStages;
}
FlushMemoryBarrier(program, inRenderPass);
if (!inRenderPass && rpHolder != null)
{
// Render pass is about to begin. Queue any fences that normally interrupt the pass.
rpHolder.InsertForcedFences(cbs);
}
while (_queuedBarrierCount > 0)
{
int memoryCount = 0;
@ -86,20 +212,20 @@ namespace Ryujinx.Graphics.Vulkan
bool hasBarrier = false;
StageFlags flags = default;
static void AddBarriers<T>(
static void AddBarriers<T, T2>(
Span<T> target,
ref int queuedBarrierCount,
ref bool hasBarrier,
ref StageFlags flags,
ref int count,
List<BarrierWithStageFlags<T>> list) where T : unmanaged
List<BarrierWithStageFlags<T, T2>> list) where T : unmanaged
{
int firstMatch = -1;
int end = list.Count;
for (int i = 0; i < list.Count; i++)
{
BarrierWithStageFlags<T> barrier = list[i];
BarrierWithStageFlags<T, T2> barrier = list[i];
if (!hasBarrier)
{
@ -162,21 +288,60 @@ namespace Ryujinx.Graphics.Vulkan
}
}
if (insideRenderPass)
if (inRenderPass && _imageBarriers.Count > 0)
{
// Image barriers queued in the batch are meant to be globally scoped,
// but inside a render pass they're scoped to just the range of the render pass.
// On MoltenVK, we just break the rules and always use image barrier.
// On desktop GPUs, all barriers are globally scoped, so we just replace it with a generic memory barrier.
// TODO: On certain GPUs, we need to split render pass so the barrier scope is global. When this is done,
// notify the resource that it should add a barrier as soon as a render pass ends to avoid this in future.
// Generally, we want to avoid this from happening in the future, so flag the texture to immediately
// emit a barrier whenever the current render pass is bound again.
if (!_gd.IsMoltenVk)
bool anyIsNonAttachment = false;
foreach (BarrierWithStageFlags<ImageMemoryBarrier, TextureStorage> barrier in _imageBarriers)
{
// If the binding is an attachment, don't add it as a forced fence.
bool isAttachment = rpHolder.ContainsAttachment(barrier.Resource);
if (!isAttachment)
{
rpHolder.AddForcedFence(barrier.Resource, barrier.Flags.Dest);
anyIsNonAttachment = true;
}
}
if (_gd.IsTBDR)
{
if (!_gd.IsMoltenVk)
{
if (!anyIsNonAttachment)
{
// This case is a feedback loop. To prevent this from causing an absolute performance disaster,
// remove the barriers entirely.
// If this is not here, there will be a lot of single draw render passes.
// TODO: explicit handling for feedback loops, likely outside this class.
_queuedBarrierCount -= _imageBarriers.Count;
_imageBarriers.Clear();
}
else
{
// TBDR GPUs are sensitive to barriers, so we need to end the pass to ensure the data is available.
// Metal already has hazard tracking so MVK doesn't need this.
endRenderPass();
inRenderPass = false;
}
}
}
else
{
// Generic pipeline memory barriers will work for desktop GPUs.
// They do require a few more access flags on the subpass dependency, though.
foreach (var barrier in _imageBarriers)
{
_memoryBarriers.Add(new BarrierWithStageFlags<MemoryBarrier>(
_memoryBarriers.Add(new BarrierWithStageFlags<MemoryBarrier, int>(
barrier.Flags,
new MemoryBarrier()
{
@ -190,6 +355,22 @@ namespace Ryujinx.Graphics.Vulkan
}
}
if (inRenderPass && _memoryBarriers.Count > 0)
{
PipelineStageFlags allFlags = PipelineStageFlags.None;
foreach (var barrier in _memoryBarriers)
{
allFlags |= barrier.Flags.Dest;
}
if (allFlags.HasFlag(PipelineStageFlags.DrawIndirectBit) || !_gd.SupportsRenderPassBarrier(allFlags))
{
endRenderPass();
inRenderPass = false;
}
}
AddBarriers(_memoryBarrierBatch.AsSpan(), ref _queuedBarrierCount, ref hasBarrier, ref flags, ref memoryCount, _memoryBarriers);
AddBarriers(_bufferBarrierBatch.AsSpan(), ref _queuedBarrierCount, ref hasBarrier, ref flags, ref bufferCount, _bufferBarriers);
AddBarriers(_imageBarrierBatch.AsSpan(), ref _queuedBarrierCount, ref hasBarrier, ref flags, ref imageCount, _imageBarriers);
@ -198,14 +379,14 @@ namespace Ryujinx.Graphics.Vulkan
{
PipelineStageFlags srcStageFlags = flags.Source;
if (insideRenderPass)
if (inRenderPass)
{
// Inside a render pass, barrier stages can only be from rasterization.
srcStageFlags &= ~PipelineStageFlags.ComputeShaderBit;
}
_gd.Api.CmdPipelineBarrier(
cb,
cbs.CommandBuffer,
srcStageFlags,
flags.Dest,
0,
@ -219,6 +400,41 @@ namespace Ryujinx.Graphics.Vulkan
}
}
private void QueueIncoherentBarrier(IncoherentBarrierType type)
{
if (type > _queuedIncoherentBarrier)
{
_queuedIncoherentBarrier = type;
}
}
public void QueueTextureBarrier()
{
QueueIncoherentBarrier(IncoherentBarrierType.Texture);
}
public void QueueMemoryBarrier()
{
QueueIncoherentBarrier(IncoherentBarrierType.All);
}
public void QueueCommandBufferBarrier()
{
QueueIncoherentBarrier(IncoherentBarrierType.CommandBuffer);
}
public void EnableTfbBarriers(bool enable)
{
if (enable)
{
_extraStages |= PipelineStageFlags.TransformFeedbackBitExt;
}
else
{
_extraStages &= ~PipelineStageFlags.TransformFeedbackBitExt;
}
}
public void Dispose()
{
_memoryBarrierBatch.Dispose();

4
src/Ryujinx.Graphics.Vulkan/Effects/FsrScalingFilter.cs
View file

@ -59,14 +59,14 @@ namespace Ryujinx.Graphics.Vulkan.Effects
var scalingResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
.Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
.Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
var sharpeningResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 3)
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 4)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
.Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
.Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
_sampler = _renderer.CreateSampler(SamplerCreateInfo.Create(MinFilter.Linear, MagFilter.Linear));

2
src/Ryujinx.Graphics.Vulkan/Effects/FxaaPostProcessingEffect.cs
View file

@ -42,7 +42,7 @@ namespace Ryujinx.Graphics.Vulkan.Effects
var resourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
.Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
.Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
_samplerLinear = _renderer.CreateSampler(SamplerCreateInfo.Create(MinFilter.Linear, MagFilter.Linear));

6
src/Ryujinx.Graphics.Vulkan/Effects/SmaaPostProcessingEffect.cs
View file

@ -81,20 +81,20 @@ namespace Ryujinx.Graphics.Vulkan.Effects
var edgeResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
.Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
.Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
var blendResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 3)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 4)
.Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
.Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
var neighbourResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 2)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 1)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 3)
.Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
.Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
_samplerLinear = _renderer.CreateSampler(SamplerCreateInfo.Create(MinFilter.Linear, MagFilter.Linear));

17
src/Ryujinx.Graphics.Vulkan/FramebufferParams.cs
View file

@ -286,10 +286,23 @@ namespace Ryujinx.Graphics.Vulkan
_depthStencil?.Storage?.QueueLoadOpBarrier(cbs, true);
gd.Barriers.Flush(cbs.CommandBuffer, false, null);
gd.Barriers.Flush(cbs, false, null, null);
}
public (Auto<DisposableRenderPass> renderPass, Auto<DisposableFramebuffer> framebuffer) GetPassAndFramebuffer(
public void AddStoreOpUsage()
{
if (_colors != null)
{
foreach (var color in _colors)
{
color.Storage?.AddStoreOpUsage(false);
}
}
_depthStencil?.Storage?.AddStoreOpUsage(true);
}
public (RenderPassHolder rpHolder, Auto<DisposableFramebuffer> framebuffer) GetPassAndFramebuffer(
VulkanRenderer gd,
Device device,
CommandBufferScoped cbs)

10
src/Ryujinx.Graphics.Vulkan/HelperShader.cs
View file

@ -115,7 +115,7 @@ namespace Ryujinx.Graphics.Vulkan
var strideChangeResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2).Build();
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2, true).Build();
_programStrideChange = gd.CreateProgramWithMinimalLayout(new[]
{
@ -125,7 +125,7 @@ namespace Ryujinx.Graphics.Vulkan
var colorCopyResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
.Add(ResourceStages.Compute, ResourceType.TextureAndSampler, 0)
.Add(ResourceStages.Compute, ResourceType.Image, 0).Build();
.Add(ResourceStages.Compute, ResourceType.Image, 0, true).Build();
_programColorCopyShortening = gd.CreateProgramWithMinimalLayout(new[]
{
@ -155,7 +155,7 @@ namespace Ryujinx.Graphics.Vulkan
var convertD32S8ToD24S8ResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2).Build();
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2, true).Build();
_programConvertD32S8ToD24S8 = gd.CreateProgramWithMinimalLayout(new[]
{
@ -165,7 +165,7 @@ namespace Ryujinx.Graphics.Vulkan
var convertIndexBufferResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2).Build();
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2, true).Build();
_programConvertIndexBuffer = gd.CreateProgramWithMinimalLayout(new[]
{
@ -175,7 +175,7 @@ namespace Ryujinx.Graphics.Vulkan
var convertIndirectDataResourceLayout = new ResourceLayoutBuilder()
.Add(ResourceStages.Compute, ResourceType.UniformBuffer, 0)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 1)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 2, true)
.Add(ResourceStages.Compute, ResourceType.StorageBuffer, 3).Build();
_programConvertIndirectData = gd.CreateProgramWithMinimalLayout(new[]

101
src/Ryujinx.Graphics.Vulkan/PipelineBase.cs
View file

@ -55,6 +55,7 @@ namespace Ryujinx.Graphics.Vulkan
protected FramebufferParams FramebufferParams;
private Auto<DisposableFramebuffer> _framebuffer;
private RenderPassHolder _rpHolder;
private Auto<DisposableRenderPass> _renderPass;
private RenderPassHolder _nullRenderPass;
private int _writtenAttachmentCount;
@ -85,8 +86,6 @@ namespace Ryujinx.Graphics.Vulkan
private bool _tfActive;
private readonly PipelineColorBlendAttachmentState[] _storedBlend;
private ulong _drawCountSinceBarrier;
public ulong DrawCount { get; private set; }
public bool RenderPassActive { get; private set; }
@ -135,48 +134,7 @@ namespace Ryujinx.Graphics.Vulkan
public unsafe void Barrier()
{
if (_drawCountSinceBarrier != DrawCount)
{
_drawCountSinceBarrier = DrawCount;
// Barriers are not supported inside a render pass on Apple GPUs.
// As a workaround, end the render pass.
if (Gd.Vendor == Vendor.Apple)
{
EndRenderPass();
}
}
MemoryBarrier memoryBarrier = new()
{
SType = StructureType.MemoryBarrier,
SrcAccessMask = AccessFlags.MemoryReadBit | AccessFlags.MemoryWriteBit,
DstAccessMask = AccessFlags.MemoryReadBit | AccessFlags.MemoryWriteBit,
};
PipelineStageFlags pipelineStageFlags = PipelineStageFlags.VertexShaderBit | PipelineStageFlags.FragmentShaderBit;
if (Gd.Capabilities.SupportsGeometryShader)
{
pipelineStageFlags |= PipelineStageFlags.GeometryShaderBit;
}
if (Gd.Capabilities.SupportsTessellationShader)
{
pipelineStageFlags |= PipelineStageFlags.TessellationControlShaderBit | PipelineStageFlags.TessellationEvaluationShaderBit;
}
Gd.Api.CmdPipelineBarrier(
CommandBuffer,
pipelineStageFlags,
pipelineStageFlags,
0,
1,
memoryBarrier,
0,
null,
0,
null);
Gd.Barriers.QueueMemoryBarrier();
}
public void ComputeBarrier()
@ -203,6 +161,7 @@ namespace Ryujinx.Graphics.Vulkan
public void BeginTransformFeedback(PrimitiveTopology topology)
{
Gd.Barriers.EnableTfbBarriers(true);
_tfEnabled = true;
}
@ -249,7 +208,7 @@ namespace Ryujinx.Graphics.Vulkan
CreateRenderPass();
}
Gd.Barriers.Flush(Cbs.CommandBuffer, RenderPassActive, EndRenderPassDelegate);
Gd.Barriers.Flush(Cbs, RenderPassActive, _rpHolder, EndRenderPassDelegate);
BeginRenderPass();
@ -287,7 +246,7 @@ namespace Ryujinx.Graphics.Vulkan
CreateRenderPass();
}
Gd.Barriers.Flush(Cbs.CommandBuffer, RenderPassActive, EndRenderPassDelegate);
Gd.Barriers.Flush(Cbs, RenderPassActive, _rpHolder, EndRenderPassDelegate);
BeginRenderPass();
@ -299,24 +258,7 @@ namespace Ryujinx.Graphics.Vulkan
public unsafe void CommandBufferBarrier()
{
MemoryBarrier memoryBarrier = new()
{
SType = StructureType.MemoryBarrier,
SrcAccessMask = BufferHolder.DefaultAccessFlags,
DstAccessMask = AccessFlags.IndirectCommandReadBit,
};
Gd.Api.CmdPipelineBarrier(
CommandBuffer,
PipelineStageFlags.AllCommandsBit,
PipelineStageFlags.DrawIndirectBit,
0,
1,
memoryBarrier,
0,
null,
0,
null);
Gd.Barriers.QueueCommandBufferBarrier();
}
public void CopyBuffer(BufferHandle source, BufferHandle destination, int srcOffset, int dstOffset, int size)
@ -722,6 +664,7 @@ namespace Ryujinx.Graphics.Vulkan
public void EndTransformFeedback()
{
Gd.Barriers.EnableTfbBarriers(false);
PauseTransformFeedbackInternal();
_tfEnabled = false;
}
@ -1408,24 +1351,7 @@ namespace Ryujinx.Graphics.Vulkan
public unsafe void TextureBarrier()
{
MemoryBarrier memoryBarrier = new()
{
SType = StructureType.MemoryBarrier,
SrcAccessMask = AccessFlags.MemoryReadBit | AccessFlags.MemoryWriteBit,
DstAccessMask = AccessFlags.MemoryReadBit | AccessFlags.MemoryWriteBit,
};
Gd.Api.CmdPipelineBarrier(
CommandBuffer,
PipelineStageFlags.FragmentShaderBit,
PipelineStageFlags.FragmentShaderBit,
0,
1,
memoryBarrier,
0,
null,
0,
null);
Gd.Barriers.QueueTextureBarrier();
}
public void TextureBarrierTiled()
@ -1532,12 +1458,15 @@ namespace Ryujinx.Graphics.Vulkan
// Use the null framebuffer.
_nullRenderPass ??= new RenderPassHolder(Gd, Device, new RenderPassCacheKey(), FramebufferParams);
_rpHolder = _nullRenderPass;
_renderPass = _nullRenderPass.GetRenderPass();
_framebuffer = _nullRenderPass.GetFramebuffer(Gd, Cbs, FramebufferParams);
}
else
{
(_renderPass, _framebuffer) = FramebufferParams.GetPassAndFramebuffer(Gd, Device, Cbs);
(_rpHolder, _framebuffer) = FramebufferParams.GetPassAndFramebuffer(Gd, Device, Cbs);
_renderPass = _rpHolder.GetRenderPass();
}
}
@ -1564,7 +1493,7 @@ namespace Ryujinx.Graphics.Vulkan
}
}
Gd.Barriers.Flush(Cbs.CommandBuffer, RenderPassActive, EndRenderPassDelegate);
Gd.Barriers.Flush(Cbs, _program, RenderPassActive, _rpHolder, EndRenderPassDelegate);
_descriptorSetUpdater.UpdateAndBindDescriptorSets(Cbs, PipelineBindPoint.Compute);
}
@ -1629,7 +1558,7 @@ namespace Ryujinx.Graphics.Vulkan
}
}
Gd.Barriers.Flush(Cbs.CommandBuffer, RenderPassActive, EndRenderPassDelegate);
Gd.Barriers.Flush(Cbs, _program, RenderPassActive, _rpHolder, EndRenderPassDelegate);
_descriptorSetUpdater.UpdateAndBindDescriptorSets(Cbs, PipelineBindPoint.Graphics);
@ -1708,6 +1637,8 @@ namespace Ryujinx.Graphics.Vulkan
{
if (RenderPassActive)
{
FramebufferParams.AddStoreOpUsage();
PauseTransformFeedbackInternal();
Gd.Api.CmdEndRenderPass(CommandBuffer);
SignalRenderPassEnd();

33
src/Ryujinx.Graphics.Vulkan/PipelineConverter.cs
View file

@ -9,13 +9,6 @@ namespace Ryujinx.Graphics.Vulkan
{
static class PipelineConverter
{
private const AccessFlags SubpassAccessMask =
AccessFlags.MemoryReadBit |
AccessFlags.MemoryWriteBit |
AccessFlags.ShaderReadBit |
AccessFlags.ColorAttachmentWriteBit |
AccessFlags.DepthStencilAttachmentWriteBit;
public static unsafe DisposableRenderPass ToRenderPass(this ProgramPipelineState state, VulkanRenderer gd, Device device)
{
const int MaxAttachments = Constants.MaxRenderTargets + 1;
@ -108,7 +101,7 @@ namespace Ryujinx.Graphics.Vulkan
}
}
var subpassDependency = CreateSubpassDependency();
var subpassDependency = CreateSubpassDependency(gd);
fixed (AttachmentDescription* pAttachmentDescs = attachmentDescs)
{
@ -129,29 +122,33 @@ namespace Ryujinx.Graphics.Vulkan
}
}
public static SubpassDependency CreateSubpassDependency()
public static SubpassDependency CreateSubpassDependency(VulkanRenderer gd)
{
var (access, stages) = BarrierBatch.GetSubpassAccessSuperset(gd);
return new SubpassDependency(
0,
0,
PipelineStageFlags.AllGraphicsBit,
PipelineStageFlags.AllGraphicsBit,
SubpassAccessMask,
SubpassAccessMask,
stages,
stages,
access,
access,
0);
}
public unsafe static SubpassDependency2 CreateSubpassDependency2()
public unsafe static SubpassDependency2 CreateSubpassDependency2(VulkanRenderer gd)
{
var (access, stages) = BarrierBatch.GetSubpassAccessSuperset(gd);
return new SubpassDependency2(
StructureType.SubpassDependency2,
null,
0,
0,
PipelineStageFlags.AllGraphicsBit,
PipelineStageFlags.AllGraphicsBit,
SubpassAccessMask,
SubpassAccessMask,
stages,
stages,
access,
access,
0);
}

2
src/Ryujinx.Graphics.Vulkan/PipelineFull.cs
View file

@ -257,7 +257,7 @@ namespace Ryujinx.Graphics.Vulkan
PreloadCbs = null;
}
Gd.Barriers.Flush(Cbs.CommandBuffer, false, null);
Gd.Barriers.Flush(Cbs, false, null, null);
CommandBuffer = (Cbs = Gd.CommandBufferPool.ReturnAndRent(Cbs)).CommandBuffer;
Gd.RegisterFlush();

40
src/Ryujinx.Graphics.Vulkan/RenderPassHolder.cs
View file

@ -1,5 +1,7 @@
using Silk.NET.Vulkan;
using System;
using System.Collections.Generic;
using System.Linq;
namespace Ryujinx.Graphics.Vulkan
{
@ -29,10 +31,13 @@ namespace Ryujinx.Graphics.Vulkan
}
}
private readonly record struct ForcedFence(TextureStorage Texture, PipelineStageFlags StageFlags);
private readonly TextureView[] _textures;
private readonly Auto<DisposableRenderPass> _renderPass;
private readonly HashTableSlim<FramebufferCacheKey, Auto<DisposableFramebuffer>> _framebuffers;
private readonly RenderPassCacheKey _key;
private readonly List<ForcedFence> _forcedFences;
public unsafe RenderPassHolder(VulkanRenderer gd, Device device, RenderPassCacheKey key, FramebufferParams fb)
{
@ -105,7 +110,7 @@ namespace Ryujinx.Graphics.Vulkan
}
}
var subpassDependency = PipelineConverter.CreateSubpassDependency();
var subpassDependency = PipelineConverter.CreateSubpassDependency(gd);
fixed (AttachmentDescription* pAttachmentDescs = attachmentDescs)
{
@ -138,6 +143,8 @@ namespace Ryujinx.Graphics.Vulkan
_textures = textures;
_key = key;
_forcedFences = new List<ForcedFence>();
}
public Auto<DisposableFramebuffer> GetFramebuffer(VulkanRenderer gd, CommandBufferScoped cbs, FramebufferParams fb)
@ -159,6 +166,37 @@ namespace Ryujinx.Graphics.Vulkan
return _renderPass;
}
public void AddForcedFence(TextureStorage storage, PipelineStageFlags stageFlags)
{
if (!_forcedFences.Any(fence => fence.Texture == storage))
{
_forcedFences.Add(new ForcedFence(storage, stageFlags));
}
}
public void InsertForcedFences(CommandBufferScoped cbs)
{
if (_forcedFences.Count > 0)
{
_forcedFences.RemoveAll((entry) =>
{
if (entry.Texture.Disposed)
{
return true;
}
entry.Texture.QueueWriteToReadBarrier(cbs, AccessFlags.ShaderReadBit, entry.StageFlags);
return false;
});
}
}
public bool ContainsAttachment(TextureStorage storage)
{
return _textures.Any(view => view.Storage == storage);
}
public void Dispose()
{
// Dispose all framebuffers.

4
src/Ryujinx.Graphics.Vulkan/ResourceLayoutBuilder.cs
View file

@ -23,7 +23,7 @@ namespace Ryujinx.Graphics.Vulkan
}
}
public ResourceLayoutBuilder Add(ResourceStages stages, ResourceType type, int binding)
public ResourceLayoutBuilder Add(ResourceStages stages, ResourceType type, int binding, bool write = false)
{
int setIndex = type switch
{
@ -35,7 +35,7 @@ namespace Ryujinx.Graphics.Vulkan
};
_resourceDescriptors[setIndex].Add(new ResourceDescriptor(binding, 1, type, stages));
_resourceUsages[setIndex].Add(new ResourceUsage(binding, 1, type, stages));
_resourceUsages[setIndex].Add(new ResourceUsage(binding, 1, type, stages, write));
return this;
}

71
src/Ryujinx.Graphics.Vulkan/ShaderCollection.cs
View file

@ -27,6 +27,9 @@ namespace Ryujinx.Graphics.Vulkan
public uint Stages { get; }
public PipelineStageFlags IncoherentBufferWriteStages { get; }
public PipelineStageFlags IncoherentTextureWriteStages { get; }
public ResourceBindingSegment[][] ClearSegments { get; }
public ResourceBindingSegment[][] BindingSegments { get; }
public DescriptorSetTemplate[] Templates { get; }
@ -131,6 +134,7 @@ namespace Ryujinx.Graphics.Vulkan
ClearSegments = BuildClearSegments(sets);
BindingSegments = BuildBindingSegments(resourceLayout.SetUsages, out bool usesBufferTextures);
Templates = BuildTemplates(usePushDescriptors);
(IncoherentBufferWriteStages, IncoherentTextureWriteStages) = BuildIncoherentStages(resourceLayout.SetUsages);
// Updating buffer texture bindings using template updates crashes the Adreno driver on Windows.
UpdateTexturesWithoutTemplate = gd.IsQualcommProprietary && usesBufferTextures;
@ -377,6 +381,73 @@ namespace Ryujinx.Graphics.Vulkan
return templates;
}
private PipelineStageFlags GetPipelineStages(ResourceStages stages)
{
PipelineStageFlags result = 0;
if ((stages & ResourceStages.Compute) != 0)
{
result |= PipelineStageFlags.ComputeShaderBit;
}
if ((stages & ResourceStages.Vertex) != 0)
{
result |= PipelineStageFlags.VertexShaderBit;
}
if ((stages & ResourceStages.Fragment) != 0)
{
result |= PipelineStageFlags.FragmentShaderBit;
}
if ((stages & ResourceStages.Geometry) != 0)
{
result |= PipelineStageFlags.GeometryShaderBit;
}
if ((stages & ResourceStages.TessellationControl) != 0)
{
result |= PipelineStageFlags.TessellationControlShaderBit;
}
if ((stages & ResourceStages.TessellationEvaluation) != 0)
{
result |= PipelineStageFlags.TessellationEvaluationShaderBit;
}
return result;
}
private (PipelineStageFlags Buffer, PipelineStageFlags Texture) BuildIncoherentStages(ReadOnlyCollection<ResourceUsageCollection> setUsages)
{
PipelineStageFlags buffer = PipelineStageFlags.None;
PipelineStageFlags texture = PipelineStageFlags.None;
foreach (var set in setUsages)
{
foreach (var range in set.Usages)
{
if (range.Write)
{
PipelineStageFlags stages = GetPipelineStages(range.Stages);
switch (range.Type)
{
case ResourceType.Image:
texture |= stages;
break;
case ResourceType.StorageBuffer:
case ResourceType.BufferImage:
buffer |= stages;
break;
}
}
}
}
return (buffer, texture);
}
private async Task BackgroundCompilation()
{
await Task.WhenAll(_shaders.Select(shader => shader.CompileTask));

2
src/Ryujinx.Graphics.Vulkan/TextureCopy.cs
View file

@ -407,7 +407,7 @@ namespace Ryujinx.Graphics.Vulkan
ImageLayout.General,
ImageLayout.General);
var subpassDependency = PipelineConverter.CreateSubpassDependency2();
var subpassDependency = PipelineConverter.CreateSubpassDependency2(gd);
fixed (AttachmentDescription2* pAttachmentDescs = attachmentDescs)
{

19
src/Ryujinx.Graphics.Vulkan/TextureStorage.cs
View file

@ -38,6 +38,8 @@ namespace Ryujinx.Graphics.Vulkan
public TextureCreateInfo Info => _info;
public bool Disposed { get; private set; }
private readonly Image _image;
private readonly Auto<DisposableImage> _imageAuto;
private readonly Auto<MemoryAllocation> _allocationAuto;
@ -433,6 +435,17 @@ namespace Ryujinx.Graphics.Vulkan
return FormatCapabilities.IsD24S8(Info.Format) && VkFormat == VkFormat.D32SfloatS8Uint;
}
public void AddStoreOpUsage(bool depthStencil)
{
_lastModificationStage = depthStencil ?
PipelineStageFlags.LateFragmentTestsBit :
PipelineStageFlags.ColorAttachmentOutputBit;
_lastModificationAccess = depthStencil ?
AccessFlags.DepthStencilAttachmentWriteBit :
AccessFlags.ColorAttachmentWriteBit;
}
public void QueueLoadOpBarrier(CommandBufferScoped cbs, bool depthStencil)
{
PipelineStageFlags srcStageFlags = _lastReadStage | _lastModificationStage;
@ -458,7 +471,7 @@ namespace Ryujinx.Graphics.Vulkan
_info.GetLayers(),
_info.Levels);
_gd.Barriers.QueueBarrier(barrier, srcStageFlags, dstStageFlags);
_gd.Barriers.QueueBarrier(barrier, this, srcStageFlags, dstStageFlags);
_lastReadStage = PipelineStageFlags.None;
_lastReadAccess = AccessFlags.None;
@ -491,7 +504,7 @@ namespace Ryujinx.Graphics.Vulkan
_info.GetLayers(),
_info.Levels);
_gd.Barriers.QueueBarrier(barrier, _lastModificationStage, dstStageFlags);
_gd.Barriers.QueueBarrier(barrier, this, _lastModificationStage, dstStageFlags);
_lastModificationAccess = AccessFlags.None;
}
@ -514,6 +527,8 @@ namespace Ryujinx.Graphics.Vulkan
public void Dispose()
{
Disposed = true;
if (_aliasedStorages != null)
{
foreach (var storage in _aliasedStorages.Values)

4
src/Ryujinx.Graphics.Vulkan/TextureView.cs
View file

@ -993,7 +993,7 @@ namespace Ryujinx.Graphics.Vulkan
throw new NotImplementedException();
}
public (Auto<DisposableRenderPass> renderPass, Auto<DisposableFramebuffer> framebuffer) GetPassAndFramebuffer(
public (RenderPassHolder rpHolder, Auto<DisposableFramebuffer> framebuffer) GetPassAndFramebuffer(
VulkanRenderer gd,
Device device,
CommandBufferScoped cbs,
@ -1006,7 +1006,7 @@ namespace Ryujinx.Graphics.Vulkan
rpHolder = new RenderPassHolder(gd, device, key, fb);
}
return (rpHolder.GetRenderPass(), rpHolder.GetFramebuffer(gd, cbs, fb));
return (rpHolder, rpHolder.GetFramebuffer(gd, cbs, fb));
}
public void AddRenderPass(RenderPassCacheKey key, RenderPassHolder renderPass)

5
src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs
View file

@ -939,6 +939,11 @@ namespace Ryujinx.Graphics.Vulkan
ScreenCaptured?.Invoke(this, bitmap);
}
public bool SupportsRenderPassBarrier(PipelineStageFlags flags)
{
return !(IsMoltenVk || IsQualcommProprietary);
}
public unsafe void Dispose()
{
if (!_initialized)