diff options
| author | riperiperi <rhy3756547@hotmail.com> | 2022-12-04 17:41:17 +0000 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2022-12-04 18:41:17 +0100 |
| commit | 9ac66336a216644a1b671e7949702d2e749838d2 (patch) | |
| tree | 549cb9cfd9f8f88f45a1064bafa57800eb241728 /Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs | |
| parent | 4965681e069eeedc5272030b131c2a45e6131e61 (diff) | |
GPU: Use lazy checks for specialization state (#4004)1.1.419
* GPU: Use lazy checks for specialization state
This PR adds a new class, the SpecializationStateUpdater, that allows elements of specialization state to be updated individually, and signal the state is checked when it changes between draws, instead of building and checking it on every draw. This also avoids building spec state when
Most state updates have been moved behind the shader state update, so that their specialization state updates make it in before shaders are fetched.
Downside: Fields in GpuChannelGraphicsState are no longer readonly. To counteract copies that might be caused this I pass it as `ref` when possible, though maybe `in` would be better? Not really sure about the quirks of `in` and the difference probably won't show on a benchmark.
The result is around 2 extra FPS on SMO in the usual spot. Not much right now, but it will remove costs when we're doing more expensive specialization checks, such as fragment output type specialization for macos. It may also help more on other games with more draws.
* Address Feedback
* Oops
Diffstat (limited to 'Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs')
| -rw-r--r-- | Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs | 138 |
1 files changed, 67 insertions, 71 deletions
diff --git a/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs b/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs index fe7e0d09..b611f4e7 100644 --- a/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs +++ b/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs @@ -1,6 +1,7 @@ using Ryujinx.Common.Logging; using Ryujinx.Common.Memory; using Ryujinx.Graphics.GAL; +using Ryujinx.Graphics.Gpu.Engine.GPFifo; using Ryujinx.Graphics.Gpu.Engine.Types; using Ryujinx.Graphics.Gpu.Image; using Ryujinx.Graphics.Gpu.Shader; @@ -16,9 +17,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed /// </summary> class StateUpdater { - public const int ShaderStateIndex = 16; + public const int ShaderStateIndex = 26; public const int RasterizerStateIndex = 15; - public const int ScissorStateIndex = 18; + public const int ScissorStateIndex = 16; public const int VertexBufferStateIndex = 0; public const int PrimitiveRestartStateIndex = 12; @@ -31,6 +32,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed private readonly ShaderProgramInfo[] _currentProgramInfo; private ShaderSpecializationState _shaderSpecState; + private SpecializationStateUpdater _currentSpecState; private ProgramPipelineState _pipeline; @@ -54,15 +56,17 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed /// <param name="channel">GPU channel</param> /// <param name="state">3D engine state</param> /// <param name="drawState">Draw state</param> - public StateUpdater(GpuContext context, GpuChannel channel, DeviceStateWithShadow<ThreedClassState> state, DrawState drawState) + /// <param name="spec">Specialization state updater</param> + public StateUpdater(GpuContext context, GpuChannel channel, DeviceStateWithShadow<ThreedClassState> state, DrawState drawState, SpecializationStateUpdater spec) { _context = context; _channel = channel; _state = state; _drawState = drawState; _currentProgramInfo = new ShaderProgramInfo[Constants.ShaderStages]; + _currentSpecState = spec; - // ShaderState must be updated after other state updates, as pipeline state is sent to the backend when compiling new shaders. + // ShaderState must be updated after other state updates, as specialization/pipeline state is used when fetching shaders. // Render target state must appear after shader state as it depends on information from the currently bound shader. // Rasterizer and scissor states are checked by render target clear, their indexes // must be updated on the constants "RasterizerStateIndex" and "ScissorStateIndex" if modified. @@ -101,6 +105,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed nameof(ThreedClassState.DepthTestFunc)), new StateUpdateCallbackEntry(UpdateTessellationState, + nameof(ThreedClassState.TessMode), nameof(ThreedClassState.TessOuterLevel), nameof(ThreedClassState.TessInnerLevel), nameof(ThreedClassState.PatchVertices)), @@ -138,17 +143,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed new StateUpdateCallbackEntry(UpdateRasterizerState, nameof(ThreedClassState.RasterizeEnable)), - new StateUpdateCallbackEntry(UpdateShaderState, - nameof(ThreedClassState.ShaderBaseAddress), - nameof(ThreedClassState.ShaderState)), - - new StateUpdateCallbackEntry(UpdateRenderTargetState, - nameof(ThreedClassState.RtColorState), - nameof(ThreedClassState.RtDepthStencilState), - nameof(ThreedClassState.RtControl), - nameof(ThreedClassState.RtDepthStencilSize), - nameof(ThreedClassState.RtDepthStencilEnable)), - new StateUpdateCallbackEntry(UpdateScissorState, nameof(ThreedClassState.ScissorState), nameof(ThreedClassState.ScreenScissorState)), @@ -179,7 +173,21 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed new StateUpdateCallbackEntry(UpdateMultisampleState, nameof(ThreedClassState.AlphaToCoverageDitherEnable), - nameof(ThreedClassState.MultisampleControl)) + nameof(ThreedClassState.MultisampleControl)), + + new StateUpdateCallbackEntry(UpdateEarlyZState, + nameof(ThreedClassState.EarlyZForce)), + + new StateUpdateCallbackEntry(UpdateShaderState, + nameof(ThreedClassState.ShaderBaseAddress), + nameof(ThreedClassState.ShaderState)), + + new StateUpdateCallbackEntry(UpdateRenderTargetState, + nameof(ThreedClassState.RtColorState), + nameof(ThreedClassState.RtDepthStencilState), + nameof(ThreedClassState.RtControl), + nameof(ThreedClassState.RtDepthStencilSize), + nameof(ThreedClassState.RtDepthStencilEnable)), }); } @@ -209,17 +217,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed [MethodImpl(MethodImplOptions.AggressiveInlining)] public void Update() { - // If any state that the shader depends on changed, - // then we may need to compile/bind a different version - // of the shader for the new state. - if (_shaderSpecState != null) - { - if (!_shaderSpecState.MatchesGraphics(_channel, GetPoolState(), GetGraphicsState(), _vsUsesDrawParameters, false)) - { - ForceShaderUpdate(); - } - } - // The vertex buffer size is calculated using a different // method when doing indexed draws, so we need to make sure // to update the vertex buffers if we are doing a regular @@ -271,6 +268,18 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed _updateTracker.Update(ulong.MaxValue); + // If any state that the shader depends on changed, + // then we may need to compile/bind a different version + // of the shader for the new state. + if (_shaderSpecState != null && _currentSpecState.HasChanged()) + { + if (!_shaderSpecState.MatchesGraphics(_channel, ref _currentSpecState.GetPoolState(), ref _currentSpecState.GetGraphicsState(), _vsUsesDrawParameters, false)) + { + // Shader must be reloaded. _vtgWritesRtLayer should not change. + UpdateShaderState(); + } + } + CommitBindings(); if (tfEnable && !_prevTfEnable) @@ -302,7 +311,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed if (!_channel.TextureManager.CommitGraphicsBindings(_shaderSpecState) || (buffers.HasUnalignedStorageBuffers != hasUnaligned)) { - // Shader must be reloaded. + _currentSpecState.SetHasUnalignedStorageBuffer(buffers.HasUnalignedStorageBuffers); + // Shader must be reloaded. _vtgWritesRtLayer should not change. UpdateShaderState(); } @@ -351,6 +361,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed _state.State.PatchVertices, _state.State.TessOuterLevel.AsSpan(), _state.State.TessInnerLevel.AsSpan()); + + _currentSpecState.SetTessellationMode(_state.State.TessMode); } /// <summary> @@ -611,6 +623,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed _state.State.AlphaTestEnable, _state.State.AlphaTestRef, _state.State.AlphaTestFunc); + + _currentSpecState.SetAlphaTest( + _state.State.AlphaTestEnable, + _state.State.AlphaTestRef, + _state.State.AlphaTestFunc); } /// <summary> @@ -710,6 +727,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed _context.Renderer.Pipeline.SetDepthMode(GetDepthMode()); _context.Renderer.Pipeline.SetViewports(viewports, disableTransform); + + _currentSpecState.SetViewportTransformDisable(_state.State.ViewportTransformEnable == 0); + _currentSpecState.SetDepthMode(GetDepthMode() == DepthMode.MinusOneToOne); } /// <summary> @@ -847,6 +867,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed _channel.TextureManager.SetGraphicsTexturePool(texturePool.Address.Pack(), texturePool.MaximumId); _channel.TextureManager.SetGraphicsTextureBufferIndex((int)_state.State.TextureBufferIndex); + + _currentSpecState.SetPoolState(GetPoolState()); } /// <summary> @@ -887,6 +909,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed _pipeline.SetVertexAttribs(vertexAttribs); _context.Renderer.Pipeline.SetVertexAttribs(vertexAttribs); + _currentSpecState.SetAttributeTypes(ref _state.State.VertexAttribState); } /// <summary> @@ -914,6 +937,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed Origin origin = (_state.State.PointCoordReplace & 4) == 0 ? Origin.LowerLeft : Origin.UpperLeft; _context.Renderer.Pipeline.SetPointParameters(size, isProgramPointSize, enablePointSprite, origin); + + _currentSpecState.SetProgramPointSizeEnable(isProgramPointSize); + _currentSpecState.SetPointSize(size); } /// <summary> @@ -1212,6 +1238,16 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed alphaToCoverageEnable, _state.State.AlphaToCoverageDitherEnable, alphaToOneEnable)); + + _currentSpecState.SetAlphaToCoverageEnable(alphaToCoverageEnable, _state.State.AlphaToCoverageDitherEnable); + } + + /// <summary> + /// Updates the early z flag, based on guest state. + /// </summary> + private void UpdateEarlyZState() + { + _currentSpecState.SetEarlyZForce(_state.State.EarlyZForce); } /// <summary> @@ -1239,10 +1275,10 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed addressesSpan[index] = baseAddress + shader.Offset; } - GpuChannelPoolState poolState = GetPoolState(); - GpuChannelGraphicsState graphicsState = GetGraphicsState(); + CachedShaderProgram gs = shaderCache.GetGraphicsShader(ref _state.State, ref _pipeline, _channel, ref _currentSpecState.GetPoolState(), ref _currentSpecState.GetGraphicsState(), addresses); - CachedShaderProgram gs = shaderCache.GetGraphicsShader(ref _state.State, ref _pipeline, _channel, poolState, graphicsState, addresses); + // Consume the modified flag for spec state so that it isn't checked again. + _currentSpecState.SetShader(gs); _shaderSpecState = gs.SpecializationState; @@ -1290,46 +1326,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed } /// <summary> - /// Gets the current GPU channel state for shader creation or compatibility verification. - /// </summary> - /// <returns>Current GPU channel state</returns> - private GpuChannelGraphicsState GetGraphicsState() - { - ref var vertexAttribState = ref _state.State.VertexAttribState; - - Array32<AttributeType> attributeTypes = new Array32<AttributeType>(); - - for (int location = 0; location < attributeTypes.Length; location++) - { - VertexAttribType type = vertexAttribState[location].UnpackType(); - - attributeTypes[location] = type switch - { - VertexAttribType.Sint => AttributeType.Sint, - VertexAttribType.Uint => AttributeType.Uint, - _ => AttributeType.Float - }; - } - - return new GpuChannelGraphicsState( - _state.State.EarlyZForce, - _drawState.Topology, - _state.State.TessMode, - (_state.State.MultisampleControl & 1) != 0, - _state.State.AlphaToCoverageDitherEnable, - _state.State.ViewportTransformEnable == 0, - GetDepthMode() == DepthMode.MinusOneToOne, - _state.State.VertexProgramPointSize, - _state.State.PointSize, - _state.State.AlphaTestEnable, - _state.State.AlphaTestFunc, - _state.State.AlphaTestRef, - ref attributeTypes, - _drawState.HasConstantBufferDrawParameters, - _channel.BufferManager.HasUnalignedStorageBuffers); - } - - /// <summary> /// Gets the depth mode that is currently being used (zero to one or minus one to one). /// </summary> /// <returns>Current depth mode</returns> |