using Ryujinx.Common.Logging; using Ryujinx.Graphics.GAL; using Ryujinx.Graphics.Gpu.Image; using Ryujinx.Graphics.Gpu.Memory; using Ryujinx.Graphics.Gpu.Shader; using Ryujinx.Graphics.Gpu.State; using Ryujinx.Graphics.Shader; using System; using System.Runtime.InteropServices; namespace Ryujinx.Graphics.Gpu.Engine { using Texture = Image.Texture; /// /// GPU method implementations. /// partial class Methods { private readonly GpuContext _context; private readonly ShaderProgramInfo[] _currentProgramInfo; /// /// In-memory shader cache. /// public ShaderCache ShaderCache { get; } /// /// GPU buffer manager. /// public BufferManager BufferManager { get; } /// /// GPU texture manager. /// public TextureManager TextureManager { get; } private bool _isAnyVbInstanced; private bool _vsUsesInstanceId; /// /// Creates a new instance of the GPU methods class. /// /// GPU context public Methods(GpuContext context) { _context = context; ShaderCache = new ShaderCache(_context); _currentProgramInfo = new ShaderProgramInfo[Constants.ShaderStages]; BufferManager = new BufferManager(context); TextureManager = new TextureManager(context); } /// /// Register callback for GPU method calls that triggers an action on the GPU. /// /// GPU state where the triggers will be registered public void RegisterCallbacks(GpuState state) { state.RegisterCallback(MethodOffset.LaunchDma, LaunchDma); state.RegisterCallback(MethodOffset.LoadInlineData, LoadInlineData); state.RegisterCallback(MethodOffset.Dispatch, Dispatch); state.RegisterCallback(MethodOffset.CopyBuffer, CopyBuffer); state.RegisterCallback(MethodOffset.CopyTexture, CopyTexture); state.RegisterCallback(MethodOffset.TextureBarrier, TextureBarrier); state.RegisterCallback(MethodOffset.InvalidateTextures, InvalidateTextures); state.RegisterCallback(MethodOffset.TextureBarrierTiled, TextureBarrierTiled); state.RegisterCallback(MethodOffset.ResetCounter, ResetCounter); state.RegisterCallback(MethodOffset.DrawEnd, DrawEnd); state.RegisterCallback(MethodOffset.DrawBegin, DrawBegin); state.RegisterCallback(MethodOffset.IndexBufferCount, SetIndexBufferCount); state.RegisterCallback(MethodOffset.Clear, Clear); state.RegisterCallback(MethodOffset.Report, Report); state.RegisterCallback(MethodOffset.FirmwareCall4, FirmwareCall4); state.RegisterCallback(MethodOffset.UniformBufferUpdateData, 16, UniformBufferUpdate); state.RegisterCallback(MethodOffset.UniformBufferBindVertex, UniformBufferBindVertex); state.RegisterCallback(MethodOffset.UniformBufferBindTessControl, UniformBufferBindTessControl); state.RegisterCallback(MethodOffset.UniformBufferBindTessEvaluation, UniformBufferBindTessEvaluation); state.RegisterCallback(MethodOffset.UniformBufferBindGeometry, UniformBufferBindGeometry); state.RegisterCallback(MethodOffset.UniformBufferBindFragment, UniformBufferBindFragment); } /// /// Updates host state based on the current guest GPU state. /// /// Guest GPU state private void UpdateState(GpuState state) { // Shaders must be the first one to be updated if modified, because // some of the other state depends on information from the currently // bound shaders. if (state.QueryModified(MethodOffset.ShaderBaseAddress, MethodOffset.ShaderState)) { UpdateShaderState(state); } if (state.QueryModified(MethodOffset.RtColorState, MethodOffset.RtDepthStencilState, MethodOffset.RtControl, MethodOffset.RtDepthStencilSize, MethodOffset.RtDepthStencilEnable)) { UpdateRenderTargetState(state, useControl: true); } if (state.QueryModified(MethodOffset.DepthTestEnable, MethodOffset.DepthWriteEnable, MethodOffset.DepthTestFunc)) { UpdateDepthTestState(state); } if (state.QueryModified(MethodOffset.DepthMode, MethodOffset.ViewportTransform, MethodOffset.ViewportExtents)) { UpdateViewportTransform(state); } if (state.QueryModified(MethodOffset.DepthBiasState, MethodOffset.DepthBiasFactor, MethodOffset.DepthBiasUnits, MethodOffset.DepthBiasClamp)) { UpdateDepthBiasState(state); } if (state.QueryModified(MethodOffset.StencilBackMasks, MethodOffset.StencilTestState, MethodOffset.StencilBackTestState)) { UpdateStencilTestState(state); } // Pools. if (state.QueryModified(MethodOffset.SamplerPoolState, MethodOffset.SamplerIndex)) { UpdateSamplerPoolState(state); } if (state.QueryModified(MethodOffset.TexturePoolState)) { UpdateTexturePoolState(state); } // Input assembler state. if (state.QueryModified(MethodOffset.VertexAttribState)) { UpdateVertexAttribState(state); } if (state.QueryModified(MethodOffset.PointSize)) { UpdatePointSizeState(state); } if (state.QueryModified(MethodOffset.PrimitiveRestartState)) { UpdatePrimitiveRestartState(state); } if (state.QueryModified(MethodOffset.IndexBufferState)) { UpdateIndexBufferState(state); } if (state.QueryModified(MethodOffset.VertexBufferDrawState, MethodOffset.VertexBufferInstanced, MethodOffset.VertexBufferState, MethodOffset.VertexBufferEndAddress)) { UpdateVertexBufferState(state); } if (state.QueryModified(MethodOffset.FaceState)) { UpdateFaceState(state); } if (state.QueryModified(MethodOffset.RtColorMaskShared, MethodOffset.RtColorMask)) { UpdateRtColorMask(state); } if (state.QueryModified(MethodOffset.BlendIndependent, MethodOffset.BlendStateCommon, MethodOffset.BlendEnableCommon, MethodOffset.BlendEnable, MethodOffset.BlendState)) { UpdateBlendState(state); } CommitBindings(); } /// /// Ensures that the bindings are visible to the host GPU. /// Note: this actually performs the binding using the host graphics API. /// private void CommitBindings() { UpdateStorageBuffers(); BufferManager.CommitBindings(); TextureManager.CommitGraphicsBindings(); } /// /// Updates storage buffer bindings. /// private void UpdateStorageBuffers() { for (int stage = 0; stage < _currentProgramInfo.Length; stage++) { ShaderProgramInfo info = _currentProgramInfo[stage]; if (info == null) { continue; } for (int index = 0; index < info.SBuffers.Count; index++) { BufferDescriptor sb = info.SBuffers[index]; ulong sbDescAddress = BufferManager.GetGraphicsUniformBufferAddress(stage, 0); int sbDescOffset = 0x110 + stage * 0x100 + sb.Slot * 0x10; sbDescAddress += (ulong)sbDescOffset; ReadOnlySpan sbDescriptorData = _context.PhysicalMemory.GetSpan(sbDescAddress, 0x10); SbDescriptor sbDescriptor = MemoryMarshal.Cast(sbDescriptorData)[0]; BufferManager.SetGraphicsStorageBuffer(stage, sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size); } } } /// /// Updates render targets (color and depth-stencil buffers) based on current render target state. /// /// Current GPU state /// Use draw buffers information from render target control register private void UpdateRenderTargetState(GpuState state, bool useControl) { var rtControl = state.Get(MethodOffset.RtControl); int count = useControl ? rtControl.UnpackCount() : Constants.TotalRenderTargets; var msaaMode = state.Get(MethodOffset.RtMsaaMode); int samplesInX = msaaMode.SamplesInX(); int samplesInY = msaaMode.SamplesInY(); for (int index = 0; index < Constants.TotalRenderTargets; index++) { int rtIndex = useControl ? rtControl.UnpackPermutationIndex(index) : index; var colorState = state.Get(MethodOffset.RtColorState, rtIndex); if (index >= count || !IsRtEnabled(colorState)) { TextureManager.SetRenderTargetColor(index, null); continue; } Texture color = TextureManager.FindOrCreateTexture(colorState, samplesInX, samplesInY); TextureManager.SetRenderTargetColor(index, color); if (color != null) { color.Modified = true; } } bool dsEnable = state.Get(MethodOffset.RtDepthStencilEnable); Texture depthStencil = null; if (dsEnable) { var dsState = state.Get(MethodOffset.RtDepthStencilState); var dsSize = state.Get (MethodOffset.RtDepthStencilSize); depthStencil = TextureManager.FindOrCreateTexture(dsState, dsSize, samplesInX, samplesInY); } TextureManager.SetRenderTargetDepthStencil(depthStencil); if (depthStencil != null) { depthStencil.Modified = true; } } /// /// Checks if a render target color buffer is used. /// /// Color buffer information /// True if the specified buffer is enabled/used, false otherwise private static bool IsRtEnabled(RtColorState colorState) { // Colors are disabled by writing 0 to the format. return colorState.Format != 0 && colorState.WidthOrStride != 0; } /// /// Updates host depth test state based on current GPU state. /// /// Current GPU state private void UpdateDepthTestState(GpuState state) { _context.Renderer.Pipeline.SetDepthTest(new DepthTestDescriptor( state.Get(MethodOffset.DepthTestEnable), state.Get(MethodOffset.DepthWriteEnable), state.Get(MethodOffset.DepthTestFunc))); } /// /// Updates host viewport transform and clipping state based on current GPU state. /// /// Current GPU state private void UpdateViewportTransform(GpuState state) { DepthMode depthMode = state.Get(MethodOffset.DepthMode); _context.Renderer.Pipeline.SetDepthMode(depthMode); bool flipY = (state.Get(MethodOffset.YControl) & 1) != 0; float yFlip = flipY ? -1 : 1; Viewport[] viewports = new Viewport[Constants.TotalViewports]; for (int index = 0; index < Constants.TotalViewports; index++) { var transform = state.Get(MethodOffset.ViewportTransform, index); var extents = state.Get (MethodOffset.ViewportExtents, index); float x = transform.TranslateX - MathF.Abs(transform.ScaleX); float y = transform.TranslateY - MathF.Abs(transform.ScaleY); float width = transform.ScaleX * 2; float height = transform.ScaleY * 2 * yFlip; RectangleF region = new RectangleF(x, y, width, height); viewports[index] = new Viewport( region, transform.UnpackSwizzleX(), transform.UnpackSwizzleY(), transform.UnpackSwizzleZ(), transform.UnpackSwizzleW(), extents.DepthNear, extents.DepthFar); } _context.Renderer.Pipeline.SetViewports(0, viewports); } /// /// Updates host depth bias (also called polygon offset) state based on current GPU state. /// /// Current GPU state private void UpdateDepthBiasState(GpuState state) { var depthBias = state.Get(MethodOffset.DepthBiasState); float factor = state.Get(MethodOffset.DepthBiasFactor); float units = state.Get(MethodOffset.DepthBiasUnits); float clamp = state.Get(MethodOffset.DepthBiasClamp); PolygonModeMask enables; enables = (depthBias.PointEnable ? PolygonModeMask.Point : 0); enables |= (depthBias.LineEnable ? PolygonModeMask.Line : 0); enables |= (depthBias.FillEnable ? PolygonModeMask.Fill : 0); _context.Renderer.Pipeline.SetDepthBias(enables, factor, units, clamp); } /// /// Updates host stencil test state based on current GPU state. /// /// Current GPU state private void UpdateStencilTestState(GpuState state) { var backMasks = state.Get (MethodOffset.StencilBackMasks); var test = state.Get (MethodOffset.StencilTestState); var backTest = state.Get(MethodOffset.StencilBackTestState); CompareOp backFunc; StencilOp backSFail; StencilOp backDpPass; StencilOp backDpFail; int backFuncRef; int backFuncMask; int backMask; if (backTest.TwoSided) { backFunc = backTest.BackFunc; backSFail = backTest.BackSFail; backDpPass = backTest.BackDpPass; backDpFail = backTest.BackDpFail; backFuncRef = backMasks.FuncRef; backFuncMask = backMasks.FuncMask; backMask = backMasks.Mask; } else { backFunc = test.FrontFunc; backSFail = test.FrontSFail; backDpPass = test.FrontDpPass; backDpFail = test.FrontDpFail; backFuncRef = test.FrontFuncRef; backFuncMask = test.FrontFuncMask; backMask = test.FrontMask; } _context.Renderer.Pipeline.SetStencilTest(new StencilTestDescriptor( test.Enable, test.FrontFunc, test.FrontSFail, test.FrontDpPass, test.FrontDpFail, test.FrontFuncRef, test.FrontFuncMask, test.FrontMask, backFunc, backSFail, backDpPass, backDpFail, backFuncRef, backFuncMask, backMask)); } /// /// Updates current sampler pool address and size based on guest GPU state. /// /// Current GPU state private void UpdateSamplerPoolState(GpuState state) { var texturePool = state.Get(MethodOffset.TexturePoolState); var samplerPool = state.Get(MethodOffset.SamplerPoolState); var samplerIndex = state.Get(MethodOffset.SamplerIndex); int maximumId = samplerIndex == SamplerIndex.ViaHeaderIndex ? texturePool.MaximumId : samplerPool.MaximumId; TextureManager.SetGraphicsSamplerPool(samplerPool.Address.Pack(), maximumId, samplerIndex); } /// /// Updates current texture pool address and size based on guest GPU state. /// /// Current GPU state private void UpdateTexturePoolState(GpuState state) { var texturePool = state.Get(MethodOffset.TexturePoolState); TextureManager.SetGraphicsTexturePool(texturePool.Address.Pack(), texturePool.MaximumId); TextureManager.SetGraphicsTextureBufferIndex(state.Get(MethodOffset.TextureBufferIndex)); } /// /// Updates host vertex attributes based on guest GPU state. /// /// Current GPU state private void UpdateVertexAttribState(GpuState state) { VertexAttribDescriptor[] vertexAttribs = new VertexAttribDescriptor[16]; for (int index = 0; index < 16; index++) { var vertexAttrib = state.Get(MethodOffset.VertexAttribState, index); if (!FormatTable.TryGetAttribFormat(vertexAttrib.UnpackFormat(), out Format format)) { Logger.PrintDebug(LogClass.Gpu, $"Invalid attribute format 0x{vertexAttrib.UnpackFormat():X}."); format = Format.R32G32B32A32Float; } vertexAttribs[index] = new VertexAttribDescriptor( vertexAttrib.UnpackBufferIndex(), vertexAttrib.UnpackOffset(), format); } _context.Renderer.Pipeline.SetVertexAttribs(vertexAttribs); } /// /// Updates host point size based on guest GPU state. /// /// Current GPU state private void UpdatePointSizeState(GpuState state) { float size = state.Get(MethodOffset.PointSize); _context.Renderer.Pipeline.SetPointSize(size); } /// /// Updates host primitive restart based on guest GPU state. /// /// Current GPU state private void UpdatePrimitiveRestartState(GpuState state) { PrimitiveRestartState primitiveRestart = state.Get(MethodOffset.PrimitiveRestartState); _context.Renderer.Pipeline.SetPrimitiveRestart( primitiveRestart.Enable, primitiveRestart.Index); } /// /// Updates host index buffer binding based on guest GPU state. /// /// Current GPU state private void UpdateIndexBufferState(GpuState state) { var indexBuffer = state.Get(MethodOffset.IndexBufferState); _firstIndex = indexBuffer.First; _indexCount = indexBuffer.Count; if (_indexCount == 0) { return; } ulong gpuVa = indexBuffer.Address.Pack(); // Do not use the end address to calculate the size, because // the result may be much larger than the real size of the index buffer. ulong size = (ulong)(_firstIndex + _indexCount); switch (indexBuffer.Type) { case IndexType.UShort: size *= 2; break; case IndexType.UInt: size *= 4; break; } BufferManager.SetIndexBuffer(gpuVa, size, indexBuffer.Type); // The index buffer affects the vertex buffer size calculation, we // need to ensure that they are updated. UpdateVertexBufferState(state); } /// /// Updates host vertex buffer bindings based on guest GPU state. /// /// Current GPU state private void UpdateVertexBufferState(GpuState state) { _isAnyVbInstanced = false; for (int index = 0; index < 16; index++) { var vertexBuffer = state.Get(MethodOffset.VertexBufferState, index); if (!vertexBuffer.UnpackEnable()) { BufferManager.SetVertexBuffer(index, 0, 0, 0, 0); continue; } GpuVa endAddress = state.Get(MethodOffset.VertexBufferEndAddress, index); ulong address = vertexBuffer.Address.Pack(); int stride = vertexBuffer.UnpackStride(); bool instanced = state.Get(MethodOffset.VertexBufferInstanced + index); int divisor = instanced ? vertexBuffer.Divisor : 0; _isAnyVbInstanced |= divisor != 0; ulong size; if (_drawIndexed || stride == 0 || instanced) { // This size may be (much) larger than the real vertex buffer size. // Avoid calculating it this way, unless we don't have any other option. size = endAddress.Pack() - address + 1; } else { // For non-indexed draws, we can guess the size from the vertex count // and stride. int firstInstance = state.Get(MethodOffset.FirstInstance); var drawState = state.Get(MethodOffset.VertexBufferDrawState); size = (ulong)((firstInstance + drawState.First + drawState.Count) * stride); } BufferManager.SetVertexBuffer(index, address, size, stride, divisor); } } /// /// Updates host face culling and orientation based on guest GPU state. /// /// Current GPU state private void UpdateFaceState(GpuState state) { var face = state.Get(MethodOffset.FaceState); _context.Renderer.Pipeline.SetFaceCulling(face.CullEnable, face.CullFace); _context.Renderer.Pipeline.SetFrontFace(face.FrontFace); } /// /// Updates host render target color masks, based on guest GPU state. /// This defines which color channels are written to each color buffer. /// /// Current GPU state private void UpdateRtColorMask(GpuState state) { bool rtColorMaskShared = state.Get(MethodOffset.RtColorMaskShared); uint[] componentMasks = new uint[Constants.TotalRenderTargets]; for (int index = 0; index < Constants.TotalRenderTargets; index++) { var colorMask = state.Get(MethodOffset.RtColorMask, rtColorMaskShared ? 0 : index); uint componentMask; componentMask = (colorMask.UnpackRed() ? 1u : 0u); componentMask |= (colorMask.UnpackGreen() ? 2u : 0u); componentMask |= (colorMask.UnpackBlue() ? 4u : 0u); componentMask |= (colorMask.UnpackAlpha() ? 8u : 0u); componentMasks[index] = componentMask; } _context.Renderer.Pipeline.SetRenderTargetColorMasks(componentMasks); } /// /// Updates host render target color buffer blending state, based on guest state. /// /// Current GPU state private void UpdateBlendState(GpuState state) { bool blendIndependent = state.Get(MethodOffset.BlendIndependent); for (int index = 0; index < 8; index++) { BlendDescriptor descriptor; if (blendIndependent) { bool enable = state.Get (MethodOffset.BlendEnable, index); var blend = state.Get(MethodOffset.BlendState, index); descriptor = new BlendDescriptor( enable, blend.ColorOp, blend.ColorSrcFactor, blend.ColorDstFactor, blend.AlphaOp, blend.AlphaSrcFactor, blend.AlphaDstFactor); } else { bool enable = state.Get (MethodOffset.BlendEnable, 0); var blend = state.Get(MethodOffset.BlendStateCommon); descriptor = new BlendDescriptor( enable, blend.ColorOp, blend.ColorSrcFactor, blend.ColorDstFactor, blend.AlphaOp, blend.AlphaSrcFactor, blend.AlphaDstFactor); } _context.Renderer.Pipeline.SetBlendState(index, descriptor); } } /// /// Storage buffer address and size information. /// private struct SbDescriptor { public uint AddressLow; public uint AddressHigh; public int Size; public int Padding; public ulong PackAddress() { return AddressLow | ((ulong)AddressHigh << 32); } } /// /// Updates host shaders based on the guest GPU state. /// /// Current GPU state private void UpdateShaderState(GpuState state) { ShaderAddresses addresses = new ShaderAddresses(); Span addressesSpan = MemoryMarshal.CreateSpan(ref addresses, 1); Span addressesArray = MemoryMarshal.Cast(addressesSpan); ulong baseAddress = state.Get(MethodOffset.ShaderBaseAddress).Pack(); for (int index = 0; index < 6; index++) { var shader = state.Get(MethodOffset.ShaderState, index); if (!shader.UnpackEnable() && index != 1) { continue; } addressesArray[index] = baseAddress + shader.Offset; } GraphicsShader gs = ShaderCache.GetGraphicsShader(state, addresses); _vsUsesInstanceId = gs.Shaders[0]?.Program.Info.UsesInstanceId ?? false; for (int stage = 0; stage < Constants.ShaderStages; stage++) { ShaderProgramInfo info = gs.Shaders[stage]?.Program.Info; _currentProgramInfo[stage] = info; if (info == null) { continue; } var textureBindings = new TextureBindingInfo[info.Textures.Count]; for (int index = 0; index < info.Textures.Count; index++) { var descriptor = info.Textures[index]; Target target = GetTarget(descriptor.Type); if (descriptor.IsBindless) { textureBindings[index] = new TextureBindingInfo(target, descriptor.CbufSlot, descriptor.CbufOffset); } else { textureBindings[index] = new TextureBindingInfo(target, descriptor.HandleIndex); } } TextureManager.SetGraphicsTextures(stage, textureBindings); var imageBindings = new TextureBindingInfo[info.Images.Count]; for (int index = 0; index < info.Images.Count; index++) { var descriptor = info.Images[index]; Target target = GetTarget(descriptor.Type); imageBindings[index] = new TextureBindingInfo(target, descriptor.HandleIndex); } TextureManager.SetGraphicsImages(stage, imageBindings); uint sbEnableMask = 0; uint ubEnableMask = 0; for (int index = 0; index < info.SBuffers.Count; index++) { sbEnableMask |= 1u << info.SBuffers[index].Slot; } for (int index = 0; index < info.CBuffers.Count; index++) { ubEnableMask |= 1u << info.CBuffers[index].Slot; } BufferManager.SetGraphicsStorageBufferEnableMask(stage, sbEnableMask); BufferManager.SetGraphicsUniformBufferEnableMask(stage, ubEnableMask); } _context.Renderer.Pipeline.SetProgram(gs.HostProgram); } /// /// Gets texture target from a sampler type. /// /// Sampler type /// Texture target value private static Target GetTarget(SamplerType type) { type &= ~(SamplerType.Indexed | SamplerType.Shadow); switch (type) { case SamplerType.Texture1D: return Target.Texture1D; case SamplerType.TextureBuffer: return Target.TextureBuffer; case SamplerType.Texture1D | SamplerType.Array: return Target.Texture1DArray; case SamplerType.Texture2D: return Target.Texture2D; case SamplerType.Texture2D | SamplerType.Array: return Target.Texture2DArray; case SamplerType.Texture2D | SamplerType.Multisample: return Target.Texture2DMultisample; case SamplerType.Texture2D | SamplerType.Multisample | SamplerType.Array: return Target.Texture2DMultisampleArray; case SamplerType.Texture3D: return Target.Texture3D; case SamplerType.TextureCube: return Target.Cubemap; case SamplerType.TextureCube | SamplerType.Array: return Target.CubemapArray; } Logger.PrintWarning(LogClass.Gpu, $"Invalid sampler type \"{type}\"."); return Target.Texture2D; } /// /// Issues a texture barrier. /// This waits until previous texture writes from the GPU to finish, before /// performing new operations with said textures. /// /// Current GPU state (unused) /// Method call argument (unused) private void TextureBarrier(GpuState state, int argument) { _context.Renderer.Pipeline.TextureBarrier(); } /// /// Invalidates all modified textures on the cache. /// /// Current GPU state (unused) /// Method call argument (unused) private void InvalidateTextures(GpuState state, int argument) { TextureManager.Flush(); } /// /// Issues a texture barrier. /// This waits until previous texture writes from the GPU to finish, before /// performing new operations with said textures. /// This performs a per-tile wait, it is only valid if both the previous write /// and current access has the same access patterns. /// This may be faster than the regular barrier on tile-based rasterizers. /// /// Current GPU state (unused) /// Method call argument (unused) private void TextureBarrierTiled(GpuState state, int argument) { _context.Renderer.Pipeline.TextureBarrierTiled(); } } }