0
0
Fork 0

Add support for bindless textures from storage buffer on Vulkan (#6721)

* Halve primitive ID when converting quads to triangles

* Shader cache version bump

* Add support for bindless textures from storage buffer on Vulkan
This commit is contained in:
gdkchan 2024-05-14 11:47:16 -03:00 committed by GitHub
parent 44dbab3848
commit 3a3b51893e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
14 changed files with 76 additions and 17 deletions

View file

@ -36,6 +36,7 @@ namespace Ryujinx.Graphics.GAL
public readonly bool SupportsMismatchingViewFormat; public readonly bool SupportsMismatchingViewFormat;
public readonly bool SupportsCubemapView; public readonly bool SupportsCubemapView;
public readonly bool SupportsNonConstantTextureOffset; public readonly bool SupportsNonConstantTextureOffset;
public readonly bool SupportsQuads;
public readonly bool SupportsSeparateSampler; public readonly bool SupportsSeparateSampler;
public readonly bool SupportsShaderBallot; public readonly bool SupportsShaderBallot;
public readonly bool SupportsShaderBarrierDivergence; public readonly bool SupportsShaderBarrierDivergence;
@ -93,6 +94,7 @@ namespace Ryujinx.Graphics.GAL
bool supportsMismatchingViewFormat, bool supportsMismatchingViewFormat,
bool supportsCubemapView, bool supportsCubemapView,
bool supportsNonConstantTextureOffset, bool supportsNonConstantTextureOffset,
bool supportsQuads,
bool supportsSeparateSampler, bool supportsSeparateSampler,
bool supportsShaderBallot, bool supportsShaderBallot,
bool supportsShaderBarrierDivergence, bool supportsShaderBarrierDivergence,
@ -146,6 +148,7 @@ namespace Ryujinx.Graphics.GAL
SupportsMismatchingViewFormat = supportsMismatchingViewFormat; SupportsMismatchingViewFormat = supportsMismatchingViewFormat;
SupportsCubemapView = supportsCubemapView; SupportsCubemapView = supportsCubemapView;
SupportsNonConstantTextureOffset = supportsNonConstantTextureOffset; SupportsNonConstantTextureOffset = supportsNonConstantTextureOffset;
SupportsQuads = supportsQuads;
SupportsSeparateSampler = supportsSeparateSampler; SupportsSeparateSampler = supportsSeparateSampler;
SupportsShaderBallot = supportsShaderBallot; SupportsShaderBallot = supportsShaderBallot;
SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence; SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence;

View file

@ -18,6 +18,8 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
private readonly ShaderSpecializationState _newSpecState; private readonly ShaderSpecializationState _newSpecState;
private readonly int _stageIndex; private readonly int _stageIndex;
private readonly bool _isVulkan; private readonly bool _isVulkan;
private readonly bool _hasGeometryShader;
private readonly bool _supportsQuads;
/// <summary> /// <summary>
/// Creates a new instance of the cached GPU state accessor for shader translation. /// Creates a new instance of the cached GPU state accessor for shader translation.
@ -29,6 +31,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
/// <param name="newSpecState">Shader specialization state of the recompiled shader</param> /// <param name="newSpecState">Shader specialization state of the recompiled shader</param>
/// <param name="counts">Resource counts shared across all shader stages</param> /// <param name="counts">Resource counts shared across all shader stages</param>
/// <param name="stageIndex">Shader stage index</param> /// <param name="stageIndex">Shader stage index</param>
/// <param name="hasGeometryShader">Indicates if a geometry shader is present</param>
public DiskCacheGpuAccessor( public DiskCacheGpuAccessor(
GpuContext context, GpuContext context,
ReadOnlyMemory<byte> data, ReadOnlyMemory<byte> data,
@ -36,7 +39,8 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
ShaderSpecializationState oldSpecState, ShaderSpecializationState oldSpecState,
ShaderSpecializationState newSpecState, ShaderSpecializationState newSpecState,
ResourceCounts counts, ResourceCounts counts,
int stageIndex) : base(context, counts, stageIndex) int stageIndex,
bool hasGeometryShader) : base(context, counts, stageIndex)
{ {
_data = data; _data = data;
_cb1Data = cb1Data; _cb1Data = cb1Data;
@ -44,6 +48,8 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
_newSpecState = newSpecState; _newSpecState = newSpecState;
_stageIndex = stageIndex; _stageIndex = stageIndex;
_isVulkan = context.Capabilities.Api == TargetApi.Vulkan; _isVulkan = context.Capabilities.Api == TargetApi.Vulkan;
_hasGeometryShader = hasGeometryShader;
_supportsQuads = context.Capabilities.SupportsQuads;
if (stageIndex == (int)ShaderStage.Geometry - 1) if (stageIndex == (int)ShaderStage.Geometry - 1)
{ {
@ -100,7 +106,11 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
/// <inheritdoc/> /// <inheritdoc/>
public GpuGraphicsState QueryGraphicsState() public GpuGraphicsState QueryGraphicsState()
{ {
return _oldSpecState.GraphicsState.CreateShaderGraphicsState(!_isVulkan, _isVulkan || _oldSpecState.GraphicsState.YNegateEnabled); return _oldSpecState.GraphicsState.CreateShaderGraphicsState(
!_isVulkan,
_supportsQuads,
_hasGeometryShader,
_isVulkan || _oldSpecState.GraphicsState.YNegateEnabled);
} }
/// <inheritdoc/> /// <inheritdoc/>

View file

@ -22,7 +22,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
private const ushort FileFormatVersionMajor = 1; private const ushort FileFormatVersionMajor = 1;
private const ushort FileFormatVersionMinor = 2; private const ushort FileFormatVersionMinor = 2;
private const uint FileFormatVersionPacked = ((uint)FileFormatVersionMajor << 16) | FileFormatVersionMinor; private const uint FileFormatVersionPacked = ((uint)FileFormatVersionMajor << 16) | FileFormatVersionMinor;
private const uint CodeGenVersion = 6577; private const uint CodeGenVersion = 5936;
private const string SharedTocFileName = "shared.toc"; private const string SharedTocFileName = "shared.toc";
private const string SharedDataFileName = "shared.data"; private const string SharedDataFileName = "shared.data";

View file

@ -601,6 +601,8 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
TargetApi api = _context.Capabilities.Api; TargetApi api = _context.Capabilities.Api;
bool hasCachedGs = guestShaders[4].HasValue;
for (int stageIndex = Constants.ShaderStages - 1; stageIndex >= 0; stageIndex--) for (int stageIndex = Constants.ShaderStages - 1; stageIndex >= 0; stageIndex--)
{ {
if (guestShaders[stageIndex + 1].HasValue) if (guestShaders[stageIndex + 1].HasValue)
@ -610,7 +612,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
byte[] guestCode = shader.Code; byte[] guestCode = shader.Code;
byte[] cb1Data = shader.Cb1Data; byte[] cb1Data = shader.Cb1Data;
DiskCacheGpuAccessor gpuAccessor = new(_context, guestCode, cb1Data, specState, newSpecState, counts, stageIndex); DiskCacheGpuAccessor gpuAccessor = new(_context, guestCode, cb1Data, specState, newSpecState, counts, stageIndex, hasCachedGs);
TranslatorContext currentStage = DecodeGraphicsShader(gpuAccessor, api, DefaultFlags, 0); TranslatorContext currentStage = DecodeGraphicsShader(gpuAccessor, api, DefaultFlags, 0);
if (nextStage != null) if (nextStage != null)
@ -623,7 +625,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
byte[] guestCodeA = guestShaders[0].Value.Code; byte[] guestCodeA = guestShaders[0].Value.Code;
byte[] cb1DataA = guestShaders[0].Value.Cb1Data; byte[] cb1DataA = guestShaders[0].Value.Cb1Data;
DiskCacheGpuAccessor gpuAccessorA = new(_context, guestCodeA, cb1DataA, specState, newSpecState, counts, 0); DiskCacheGpuAccessor gpuAccessorA = new(_context, guestCodeA, cb1DataA, specState, newSpecState, counts, 0, hasCachedGs);
translatorContexts[0] = DecodeGraphicsShader(gpuAccessorA, api, DefaultFlags | TranslationFlags.VertexA, 0); translatorContexts[0] = DecodeGraphicsShader(gpuAccessorA, api, DefaultFlags | TranslationFlags.VertexA, 0);
} }
@ -711,7 +713,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
GuestCodeAndCbData shader = guestShaders[0].Value; GuestCodeAndCbData shader = guestShaders[0].Value;
ResourceCounts counts = new(); ResourceCounts counts = new();
ShaderSpecializationState newSpecState = new(ref specState.ComputeState); ShaderSpecializationState newSpecState = new(ref specState.ComputeState);
DiskCacheGpuAccessor gpuAccessor = new(_context, shader.Code, shader.Cb1Data, specState, newSpecState, counts, 0); DiskCacheGpuAccessor gpuAccessor = new(_context, shader.Code, shader.Cb1Data, specState, newSpecState, counts, 0, false);
gpuAccessor.InitializeReservedCounts(tfEnabled: false, vertexAsCompute: false); gpuAccessor.InitializeReservedCounts(tfEnabled: false, vertexAsCompute: false);
TranslatorContext translatorContext = DecodeComputeShader(gpuAccessor, _context.Capabilities.Api, 0); TranslatorContext translatorContext = DecodeComputeShader(gpuAccessor, _context.Capabilities.Api, 0);

View file

@ -17,6 +17,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
private readonly int _stageIndex; private readonly int _stageIndex;
private readonly bool _compute; private readonly bool _compute;
private readonly bool _isVulkan; private readonly bool _isVulkan;
private readonly bool _hasGeometryShader;
private readonly bool _supportsQuads;
/// <summary> /// <summary>
/// Creates a new instance of the GPU state accessor for graphics shader translation. /// Creates a new instance of the GPU state accessor for graphics shader translation.
@ -25,12 +27,20 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="channel">GPU channel</param> /// <param name="channel">GPU channel</param>
/// <param name="state">Current GPU state</param> /// <param name="state">Current GPU state</param>
/// <param name="stageIndex">Graphics shader stage index (0 = Vertex, 4 = Fragment)</param> /// <param name="stageIndex">Graphics shader stage index (0 = Vertex, 4 = Fragment)</param>
public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state, int stageIndex) : base(context, state.ResourceCounts, stageIndex) /// <param name="hasGeometryShader">Indicates if a geometry shader is present</param>
public GpuAccessor(
GpuContext context,
GpuChannel channel,
GpuAccessorState state,
int stageIndex,
bool hasGeometryShader) : base(context, state.ResourceCounts, stageIndex)
{ {
_isVulkan = context.Capabilities.Api == TargetApi.Vulkan;
_channel = channel; _channel = channel;
_state = state; _state = state;
_stageIndex = stageIndex; _stageIndex = stageIndex;
_isVulkan = context.Capabilities.Api == TargetApi.Vulkan;
_hasGeometryShader = hasGeometryShader;
_supportsQuads = context.Capabilities.SupportsQuads;
if (stageIndex == (int)ShaderStage.Geometry - 1) if (stageIndex == (int)ShaderStage.Geometry - 1)
{ {
@ -105,7 +115,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <inheritdoc/> /// <inheritdoc/>
public GpuGraphicsState QueryGraphicsState() public GpuGraphicsState QueryGraphicsState()
{ {
return _state.GraphicsState.CreateShaderGraphicsState(!_isVulkan, _isVulkan || _state.GraphicsState.YNegateEnabled); return _state.GraphicsState.CreateShaderGraphicsState(
!_isVulkan,
_supportsQuads,
_hasGeometryShader,
_isVulkan || _state.GraphicsState.YNegateEnabled);
} }
/// <inheritdoc/> /// <inheritdoc/>

View file

@ -106,8 +106,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// Creates a new graphics state from this state that can be used for shader generation. /// Creates a new graphics state from this state that can be used for shader generation.
/// </summary> /// </summary>
/// <param name="hostSupportsAlphaTest">Indicates if the host API supports alpha test operations</param> /// <param name="hostSupportsAlphaTest">Indicates if the host API supports alpha test operations</param>
/// <param name="hostSupportsQuads">Indicates if the host API supports quad primitives</param>
/// <param name="hasGeometryShader">Indicates if a geometry shader is used</param>
/// <param name="originUpperLeft">If true, indicates that the fragment origin is the upper left corner of the viewport, otherwise it is the lower left corner</param>
/// <returns>GPU graphics state that can be used for shader translation</returns> /// <returns>GPU graphics state that can be used for shader translation</returns>
public readonly GpuGraphicsState CreateShaderGraphicsState(bool hostSupportsAlphaTest, bool originUpperLeft) public readonly GpuGraphicsState CreateShaderGraphicsState(bool hostSupportsAlphaTest, bool hostSupportsQuads, bool hasGeometryShader, bool originUpperLeft)
{ {
AlphaTestOp alphaTestOp; AlphaTestOp alphaTestOp;
@ -130,6 +133,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
}; };
} }
bool isQuad = Topology == PrimitiveTopology.Quads || Topology == PrimitiveTopology.QuadStrip;
bool halvePrimitiveId = !hostSupportsQuads && !hasGeometryShader && isQuad;
return new GpuGraphicsState( return new GpuGraphicsState(
EarlyZForce, EarlyZForce,
ConvertToInputTopology(Topology, TessellationMode), ConvertToInputTopology(Topology, TessellationMode),
@ -149,7 +155,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
in FragmentOutputTypes, in FragmentOutputTypes,
DualSourceBlendEnable, DualSourceBlendEnable,
YNegateEnabled, YNegateEnabled,
originUpperLeft); originUpperLeft,
halvePrimitiveId);
} }
/// <summary> /// <summary>

View file

@ -339,7 +339,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
if (gpuVa != 0) if (gpuVa != 0)
{ {
GpuAccessor gpuAccessor = new(_context, channel, gpuAccessorState, stageIndex); GpuAccessor gpuAccessor = new(_context, channel, gpuAccessorState, stageIndex, addresses.Geometry != 0);
TranslatorContext currentStage = DecodeGraphicsShader(gpuAccessor, api, DefaultFlags, gpuVa); TranslatorContext currentStage = DecodeGraphicsShader(gpuAccessor, api, DefaultFlags, gpuVa);
if (nextStage != null) if (nextStage != null)

View file

@ -161,6 +161,7 @@ namespace Ryujinx.Graphics.OpenGL
supportsBgraFormat: false, supportsBgraFormat: false,
supportsR4G4Format: false, supportsR4G4Format: false,
supportsR4G4B4A4Format: true, supportsR4G4B4A4Format: true,
supportsScaledVertexFormats: true,
supportsSnormBufferTextureFormat: false, supportsSnormBufferTextureFormat: false,
supports5BitComponentFormat: true, supports5BitComponentFormat: true,
supportsSparseBuffer: false, supportsSparseBuffer: false,
@ -175,7 +176,7 @@ namespace Ryujinx.Graphics.OpenGL
supportsMismatchingViewFormat: HwCapabilities.SupportsMismatchingViewFormat, supportsMismatchingViewFormat: HwCapabilities.SupportsMismatchingViewFormat,
supportsCubemapView: true, supportsCubemapView: true,
supportsNonConstantTextureOffset: HwCapabilities.SupportsNonConstantTextureOffset, supportsNonConstantTextureOffset: HwCapabilities.SupportsNonConstantTextureOffset,
supportsScaledVertexFormats: true, supportsQuads: HwCapabilities.SupportsQuads,
supportsSeparateSampler: false, supportsSeparateSampler: false,
supportsShaderBallot: HwCapabilities.SupportsShaderBallot, supportsShaderBallot: HwCapabilities.SupportsShaderBallot,
supportsShaderBarrierDivergence: !(intelWindows || intelUnix), supportsShaderBarrierDivergence: !(intelWindows || intelUnix),

View file

@ -102,6 +102,11 @@ namespace Ryujinx.Graphics.Shader
/// </summary> /// </summary>
public readonly bool OriginUpperLeft; public readonly bool OriginUpperLeft;
/// <summary>
/// Indicates that the primitive ID values on the shader should be halved due to quad to triangles conversion.
/// </summary>
public readonly bool HalvePrimitiveId;
/// <summary> /// <summary>
/// Creates a new GPU graphics state. /// Creates a new GPU graphics state.
/// </summary> /// </summary>
@ -124,6 +129,7 @@ namespace Ryujinx.Graphics.Shader
/// <param name="dualSourceBlendEnable">Indicates whether dual source blend is enabled</param> /// <param name="dualSourceBlendEnable">Indicates whether dual source blend is enabled</param>
/// <param name="yNegateEnabled">Indicates if negation of the viewport Y axis is enabled</param> /// <param name="yNegateEnabled">Indicates if negation of the viewport Y axis is enabled</param>
/// <param name="originUpperLeft">If true, indicates that the fragment origin is the upper left corner of the viewport, otherwise it is the lower left corner</param> /// <param name="originUpperLeft">If true, indicates that the fragment origin is the upper left corner of the viewport, otherwise it is the lower left corner</param>
/// <param name="halvePrimitiveId">Indicates that the primitive ID values on the shader should be halved due to quad to triangles conversion</param>
public GpuGraphicsState( public GpuGraphicsState(
bool earlyZForce, bool earlyZForce,
InputTopology topology, InputTopology topology,
@ -143,7 +149,8 @@ namespace Ryujinx.Graphics.Shader
in Array8<AttributeType> fragmentOutputTypes, in Array8<AttributeType> fragmentOutputTypes,
bool dualSourceBlendEnable, bool dualSourceBlendEnable,
bool yNegateEnabled, bool yNegateEnabled,
bool originUpperLeft) bool originUpperLeft,
bool halvePrimitiveId)
{ {
EarlyZForce = earlyZForce; EarlyZForce = earlyZForce;
Topology = topology; Topology = topology;
@ -164,6 +171,7 @@ namespace Ryujinx.Graphics.Shader
DualSourceBlendEnable = dualSourceBlendEnable; DualSourceBlendEnable = dualSourceBlendEnable;
YNegateEnabled = yNegateEnabled; YNegateEnabled = yNegateEnabled;
OriginUpperLeft = originUpperLeft; OriginUpperLeft = originUpperLeft;
HalvePrimitiveId = halvePrimitiveId;
} }
} }
} }

View file

@ -135,6 +135,7 @@ namespace Ryujinx.Graphics.Shader
default, default,
false, false,
false, false,
false,
false); false);
} }

View file

@ -84,6 +84,10 @@ namespace Ryujinx.Graphics.Shader.Instructions
value = context.IConvertU32ToFP32(value); value = context.IConvertU32ToFP32(value);
} }
} }
else if (offset == AttributeConsts.PrimitiveId && context.TranslatorContext.Definitions.HalvePrimitiveId)
{
value = context.ShiftRightS32(value, Const(1));
}
context.Copy(Register(rd), value); context.Copy(Register(rd), value);
} }
@ -187,6 +191,12 @@ namespace Ryujinx.Graphics.Shader.Instructions
} }
} }
} }
else if (op.Imm10 == AttributeConsts.PrimitiveId && context.TranslatorContext.Definitions.HalvePrimitiveId)
{
// If quads are used, but the host does not support them, they need to be converted to triangles.
// Since each quad becomes 2 triangles, we need to compensate here and divide primitive ID by 2.
res = context.ShiftRightS32(res, Const(1));
}
else if (op.Imm10 == AttributeConsts.FrontFacing && context.TranslatorContext.GpuAccessor.QueryHostHasFrontFacingBug()) else if (op.Imm10 == AttributeConsts.FrontFacing && context.TranslatorContext.GpuAccessor.QueryHostHasFrontFacingBug())
{ {
// gl_FrontFacing sometimes has incorrect (flipped) values depending how it is accessed on Intel GPUs. // gl_FrontFacing sometimes has incorrect (flipped) values depending how it is accessed on Intel GPUs.

View file

@ -66,9 +66,9 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
if (nvHandle.AsgOp is not Operation handleOp || if (nvHandle.AsgOp is not Operation handleOp ||
handleOp.Inst != Instruction.Load || handleOp.Inst != Instruction.Load ||
handleOp.StorageKind != StorageKind.Input) (handleOp.StorageKind != StorageKind.Input && handleOp.StorageKind != StorageKind.StorageBuffer))
{ {
// Right now, we only allow bindless access when the handle comes from a shader input. // Right now, we only allow bindless access when the handle comes from a shader input or storage buffer.
// This is an artificial limitation to prevent it from being used in cases where it // This is an artificial limitation to prevent it from being used in cases where it
// would have a large performance impact of loading all textures in the pool. // would have a large performance impact of loading all textures in the pool.
// It might be removed in the future, if we can mitigate the performance impact. // It might be removed in the future, if we can mitigate the performance impact.

View file

@ -45,6 +45,8 @@ namespace Ryujinx.Graphics.Shader.Translation
public bool YNegateEnabled => _graphicsState.YNegateEnabled; public bool YNegateEnabled => _graphicsState.YNegateEnabled;
public bool OriginUpperLeft => _graphicsState.OriginUpperLeft; public bool OriginUpperLeft => _graphicsState.OriginUpperLeft;
public bool HalvePrimitiveId => _graphicsState.HalvePrimitiveId;
public ImapPixelType[] ImapTypes { get; } public ImapPixelType[] ImapTypes { get; }
public bool IaIndexing { get; private set; } public bool IaIndexing { get; private set; }
public bool OaIndexing { get; private set; } public bool OaIndexing { get; private set; }

View file

@ -691,6 +691,7 @@ namespace Ryujinx.Graphics.Vulkan
supportsBgraFormat: true, supportsBgraFormat: true,
supportsR4G4Format: false, supportsR4G4Format: false,
supportsR4G4B4A4Format: supportsR4G4B4A4Format, supportsR4G4B4A4Format: supportsR4G4B4A4Format,
supportsScaledVertexFormats: FormatCapabilities.SupportsScaledVertexFormats(),
supportsSnormBufferTextureFormat: true, supportsSnormBufferTextureFormat: true,
supports5BitComponentFormat: supports5BitComponentFormat, supports5BitComponentFormat: supports5BitComponentFormat,
supportsSparseBuffer: features2.Features.SparseBinding && mainQueueProperties.QueueFlags.HasFlag(QueueFlags.SparseBindingBit), supportsSparseBuffer: features2.Features.SparseBinding && mainQueueProperties.QueueFlags.HasFlag(QueueFlags.SparseBindingBit),
@ -705,7 +706,7 @@ namespace Ryujinx.Graphics.Vulkan
supportsMismatchingViewFormat: true, supportsMismatchingViewFormat: true,
supportsCubemapView: !IsAmdGcn, supportsCubemapView: !IsAmdGcn,
supportsNonConstantTextureOffset: false, supportsNonConstantTextureOffset: false,
supportsScaledVertexFormats: FormatCapabilities.SupportsScaledVertexFormats(), supportsQuads: false,
supportsSeparateSampler: true, supportsSeparateSampler: true,
supportsShaderBallot: false, supportsShaderBallot: false,
supportsShaderBarrierDivergence: Vendor != Vendor.Intel, supportsShaderBarrierDivergence: Vendor != Vendor.Intel,