/// <summary> /// Updates storage buffer bindings. /// </summary> 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; SbDescriptor sbDescriptor = _context.PhysicalMemory.Read <SbDescriptor>(sbDescAddress); BufferManager.SetGraphicsStorageBuffer(stage, sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size); } } }
/// <summary> /// Updates storage buffer bindings. /// </summary> private void UpdateStorageBuffers() { for (int stage = 0; stage < CurrentGpMeta.Info.Length; stage++) { ShaderProgramInfo info = CurrentGpMeta.Info[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 <byte> sbDescriptorData = _context.PhysicalMemory.GetSpan(sbDescAddress, 0x10); SbDescriptor sbDescriptor = MemoryMarshal.Cast <byte, SbDescriptor>(sbDescriptorData)[0]; BufferManager.SetGraphicsStorageBuffer(stage, sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size); } } }
/// <summary> /// Performs the compute dispatch operation. /// </summary> /// <param name="argument">Method call argument</param> private void SendSignalingPcasB(int argument) { var memoryManager = _channel.MemoryManager; _3dEngine.FlushUboDirty(); uint qmdAddress = _state.State.SendPcasA; var qmd = _channel.MemoryManager.Read <ComputeQmd>((ulong)qmdAddress << 8); ulong shaderGpuVa = ((ulong)_state.State.SetProgramRegionAAddressUpper << 32) | _state.State.SetProgramRegionB; shaderGpuVa += (uint)qmd.ProgramOffset; int localMemorySize = qmd.ShaderLocalMemoryLowSize + qmd.ShaderLocalMemoryHighSize; int sharedMemorySize = Math.Min(qmd.SharedMemorySize, _context.Capabilities.MaximumComputeSharedMemorySize); for (int index = 0; index < Constants.TotalCpUniformBuffers; index++) { if (!qmd.ConstantBufferValid(index)) { continue; } ulong gpuVa = (uint)qmd.ConstantBufferAddrLower(index) | (ulong)qmd.ConstantBufferAddrUpper(index) << 32; ulong size = (ulong)qmd.ConstantBufferSize(index); _channel.BufferManager.SetComputeUniformBuffer(index, gpuVa, size); } ulong samplerPoolGpuVa = ((ulong)_state.State.SetTexSamplerPoolAOffsetUpper << 32) | _state.State.SetTexSamplerPoolB; ulong texturePoolGpuVa = ((ulong)_state.State.SetTexHeaderPoolAOffsetUpper << 32) | _state.State.SetTexHeaderPoolB; GpuChannelPoolState poolState = new GpuChannelPoolState( texturePoolGpuVa, _state.State.SetTexHeaderPoolCMaximumIndex, _state.State.SetBindlessTextureConstantBufferSlotSelect); GpuChannelComputeState computeState = new GpuChannelComputeState( qmd.CtaThreadDimension0, qmd.CtaThreadDimension1, qmd.CtaThreadDimension2, localMemorySize, sharedMemorySize); CachedShaderProgram cs = memoryManager.Physical.ShaderCache.GetComputeShader(_channel, poolState, computeState, shaderGpuVa); _context.Renderer.Pipeline.SetProgram(cs.HostProgram); _channel.TextureManager.SetComputeSamplerPool(samplerPoolGpuVa, _state.State.SetTexSamplerPoolCMaximumIndex, qmd.SamplerIndex); _channel.TextureManager.SetComputeTexturePool(texturePoolGpuVa, _state.State.SetTexHeaderPoolCMaximumIndex); _channel.TextureManager.SetComputeTextureBufferIndex(_state.State.SetBindlessTextureConstantBufferSlotSelect); ShaderProgramInfo info = cs.Shaders[0].Info; for (int index = 0; index < info.CBuffers.Count; index++) { BufferDescriptor cb = info.CBuffers[index]; // NVN uses the "hardware" constant buffer for anything that is less than 8, // and those are already bound above. // Anything greater than or equal to 8 uses the emulated constant buffers. // They are emulated using global memory loads. if (cb.Slot < 8) { continue; } ulong cbDescAddress = _channel.BufferManager.GetComputeUniformBufferAddress(0); int cbDescOffset = 0x260 + (cb.Slot - 8) * 0x10; cbDescAddress += (ulong)cbDescOffset; SbDescriptor cbDescriptor = _channel.MemoryManager.Physical.Read <SbDescriptor>(cbDescAddress); _channel.BufferManager.SetComputeUniformBuffer(cb.Slot, cbDescriptor.PackAddress(), (uint)cbDescriptor.Size); } for (int index = 0; index < info.SBuffers.Count; index++) { BufferDescriptor sb = info.SBuffers[index]; ulong sbDescAddress = _channel.BufferManager.GetComputeUniformBufferAddress(0); int sbDescOffset = 0x310 + sb.Slot * 0x10; sbDescAddress += (ulong)sbDescOffset; SbDescriptor sbDescriptor = _channel.MemoryManager.Physical.Read <SbDescriptor>(sbDescAddress); _channel.BufferManager.SetComputeStorageBuffer(sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size, sb.Flags); } _channel.BufferManager.SetComputeStorageBufferBindings(info.SBuffers); _channel.BufferManager.SetComputeUniformBufferBindings(info.CBuffers); int maxTextureBinding = -1; int maxImageBinding = -1; TextureBindingInfo[] textureBindings = _channel.TextureManager.RentComputeTextureBindings(info.Textures.Count); for (int index = 0; index < info.Textures.Count; index++) { var descriptor = info.Textures[index]; Target target = ShaderTexture.GetTarget(descriptor.Type); textureBindings[index] = new TextureBindingInfo( target, descriptor.Binding, descriptor.CbufSlot, descriptor.HandleIndex, descriptor.Flags); if (descriptor.Binding > maxTextureBinding) { maxTextureBinding = descriptor.Binding; } } TextureBindingInfo[] imageBindings = _channel.TextureManager.RentComputeImageBindings(info.Images.Count); for (int index = 0; index < info.Images.Count; index++) { var descriptor = info.Images[index]; Target target = ShaderTexture.GetTarget(descriptor.Type); Format format = ShaderTexture.GetFormat(descriptor.Format); imageBindings[index] = new TextureBindingInfo( target, format, descriptor.Binding, descriptor.CbufSlot, descriptor.HandleIndex, descriptor.Flags); if (descriptor.Binding > maxImageBinding) { maxImageBinding = descriptor.Binding; } } _channel.TextureManager.SetComputeMaxBindings(maxTextureBinding, maxImageBinding); // Should never return false for mismatching spec state, since the shader was fetched above. _channel.TextureManager.CommitComputeBindings(cs.SpecializationState); _channel.BufferManager.CommitComputeBindings(); _context.Renderer.Pipeline.DispatchCompute(qmd.CtaRasterWidth, qmd.CtaRasterHeight, qmd.CtaRasterDepth); _3dEngine.ForceShaderUpdate(); }
/// <summary> /// Dispatches compute work. /// </summary> /// <param name="state">Current GPU state</param> /// <param name="argument">Method call argument</param> public void Dispatch(GpuState state, int argument) { FlushUboDirty(); uint qmdAddress = (uint)state.Get <int>(MethodOffset.DispatchParamsAddress); var qmd = _context.MemoryManager.Read <ComputeQmd>((ulong)qmdAddress << 8); GpuVa shaderBaseAddress = state.Get <GpuVa>(MethodOffset.ShaderBaseAddress); ulong shaderGpuVa = shaderBaseAddress.Pack() + (uint)qmd.ProgramOffset; int localMemorySize = qmd.ShaderLocalMemoryLowSize + qmd.ShaderLocalMemoryHighSize; int sharedMemorySize = Math.Min(qmd.SharedMemorySize, _context.Capabilities.MaximumComputeSharedMemorySize); for (int index = 0; index < Constants.TotalCpUniformBuffers; index++) { if (!qmd.ConstantBufferValid(index)) { continue; } ulong gpuVa = (uint)qmd.ConstantBufferAddrLower(index) | (ulong)qmd.ConstantBufferAddrUpper(index) << 32; ulong size = (ulong)qmd.ConstantBufferSize(index); BufferManager.SetComputeUniformBuffer(index, gpuVa, size); } ShaderBundle cs = ShaderCache.GetComputeShader( state, shaderGpuVa, qmd.CtaThreadDimension0, qmd.CtaThreadDimension1, qmd.CtaThreadDimension2, localMemorySize, sharedMemorySize); _context.Renderer.Pipeline.SetProgram(cs.HostProgram); var samplerPool = state.Get <PoolState>(MethodOffset.SamplerPoolState); var texturePool = state.Get <PoolState>(MethodOffset.TexturePoolState); TextureManager.SetComputeSamplerPool(samplerPool.Address.Pack(), samplerPool.MaximumId, qmd.SamplerIndex); TextureManager.SetComputeTexturePool(texturePool.Address.Pack(), texturePool.MaximumId); TextureManager.SetComputeTextureBufferIndex(state.Get <int>(MethodOffset.TextureBufferIndex)); ShaderProgramInfo info = cs.Shaders[0].Info; for (int index = 0; index < info.CBuffers.Count; index++) { BufferDescriptor cb = info.CBuffers[index]; // NVN uses the "hardware" constant buffer for anything that is less than 8, // and those are already bound above. // Anything greater than or equal to 8 uses the emulated constant buffers. // They are emulated using global memory loads. if (cb.Slot < 8) { continue; } ulong cbDescAddress = BufferManager.GetComputeUniformBufferAddress(0); int cbDescOffset = 0x260 + (cb.Slot - 8) * 0x10; cbDescAddress += (ulong)cbDescOffset; SbDescriptor cbDescriptor = _context.PhysicalMemory.Read <SbDescriptor>(cbDescAddress); BufferManager.SetComputeUniformBuffer(cb.Slot, cbDescriptor.PackAddress(), (uint)cbDescriptor.Size); } for (int index = 0; index < info.SBuffers.Count; index++) { BufferDescriptor sb = info.SBuffers[index]; ulong sbDescAddress = BufferManager.GetComputeUniformBufferAddress(0); int sbDescOffset = 0x310 + sb.Slot * 0x10; sbDescAddress += (ulong)sbDescOffset; SbDescriptor sbDescriptor = _context.PhysicalMemory.Read <SbDescriptor>(sbDescAddress); BufferManager.SetComputeStorageBuffer(sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size, sb.Flags); } BufferManager.SetComputeStorageBufferBindings(info.SBuffers); BufferManager.SetComputeUniformBufferBindings(info.CBuffers); var textureBindings = new TextureBindingInfo[info.Textures.Count]; for (int index = 0; index < info.Textures.Count; index++) { var descriptor = info.Textures[index]; Target target = ShaderTexture.GetTarget(descriptor.Type); textureBindings[index] = new TextureBindingInfo( target, descriptor.Binding, descriptor.CbufSlot, descriptor.HandleIndex, descriptor.Flags); } TextureManager.SetComputeTextures(textureBindings); var imageBindings = new TextureBindingInfo[info.Images.Count]; for (int index = 0; index < info.Images.Count; index++) { var descriptor = info.Images[index]; Target target = ShaderTexture.GetTarget(descriptor.Type); Format format = ShaderTexture.GetFormat(descriptor.Format); imageBindings[index] = new TextureBindingInfo( target, format, descriptor.Binding, descriptor.CbufSlot, descriptor.HandleIndex, descriptor.Flags); } TextureManager.SetComputeImages(imageBindings); TextureManager.CommitComputeBindings(); BufferManager.CommitComputeBindings(); _context.Renderer.Pipeline.DispatchCompute( qmd.CtaRasterWidth, qmd.CtaRasterHeight, qmd.CtaRasterDepth); _forceShaderUpdate = true; }
/// <summary> /// Dispatches compute work. /// </summary> /// <param name="state">Current GPU state</param> /// <param name="argument">Method call argument</param> public void Dispatch(GpuState state, int argument) { uint qmdAddress = (uint)state.Get <int>(MethodOffset.DispatchParamsAddress); var qmd = _context.MemoryAccessor.Read <ComputeQmd>((ulong)qmdAddress << 8); GpuVa shaderBaseAddress = state.Get <GpuVa>(MethodOffset.ShaderBaseAddress); ulong shaderGpuVa = shaderBaseAddress.Pack() + (uint)qmd.ProgramOffset; int localMemorySize = qmd.ShaderLocalMemoryLowSize + qmd.ShaderLocalMemoryHighSize; int sharedMemorySize = Math.Min(qmd.SharedMemorySize, _context.Capabilities.MaximumComputeSharedMemorySize); Shader.Shader cs = ShaderCache.GetComputeShader( shaderGpuVa, qmd.CtaThreadDimension0, qmd.CtaThreadDimension1, qmd.CtaThreadDimension2, localMemorySize, sharedMemorySize); CurrentCpMeta = cs.Meta; _context.Renderer.Pipeline.SetProgram(cs.HostProgram); var samplerPool = state.Get <PoolState>(MethodOffset.SamplerPoolState); TextureManager.SetComputeSamplerPool(samplerPool.Address.Pack(), samplerPool.MaximumId, qmd.SamplerIndex); var texturePool = state.Get <PoolState>(MethodOffset.TexturePoolState); TextureManager.SetComputeTexturePool(texturePool.Address.Pack(), texturePool.MaximumId); TextureManager.SetComputeTextureBufferIndex(state.Get <int>(MethodOffset.TextureBufferIndex)); ShaderProgramInfo info = cs.Meta.Info[0]; uint sbEnableMask = 0; uint ubEnableMask = 0; for (int index = 0; index < Constants.TotalCpUniformBuffers; index++) { if (!qmd.ConstantBufferValid(index)) { continue; } ubEnableMask |= 1u << index; ulong gpuVa = (uint)qmd.ConstantBufferAddrLower(index) | (ulong)qmd.ConstantBufferAddrUpper(index) << 32; ulong size = (ulong)qmd.ConstantBufferSize(index); BufferManager.SetComputeUniformBuffer(index, gpuVa, size); } for (int index = 0; index < info.CBuffers.Count; index++) { BufferDescriptor cb = info.CBuffers[index]; // NVN uses the "hardware" constant buffer for anything that is less than 8, // and those are already bound above. // Anything greater than or equal to 8 uses the emulated constant buffers. // They are emulated using global memory loads. if (cb.Slot < 8) { continue; } ubEnableMask |= 1u << cb.Slot; ulong cbDescAddress = BufferManager.GetComputeUniformBufferAddress(0); int cbDescOffset = 0x260 + cb.Slot * 0x10; cbDescAddress += (ulong)cbDescOffset; ReadOnlySpan <byte> cbDescriptorData = _context.PhysicalMemory.GetSpan(cbDescAddress, 0x10); SbDescriptor cbDescriptor = MemoryMarshal.Cast <byte, SbDescriptor>(cbDescriptorData)[0]; BufferManager.SetComputeUniformBuffer(cb.Slot, cbDescriptor.PackAddress(), (uint)cbDescriptor.Size); } for (int index = 0; index < info.SBuffers.Count; index++) { BufferDescriptor sb = info.SBuffers[index]; sbEnableMask |= 1u << sb.Slot; ulong sbDescAddress = BufferManager.GetComputeUniformBufferAddress(0); int sbDescOffset = 0x310 + sb.Slot * 0x10; sbDescAddress += (ulong)sbDescOffset; ReadOnlySpan <byte> sbDescriptorData = _context.PhysicalMemory.GetSpan(sbDescAddress, 0x10); SbDescriptor sbDescriptor = MemoryMarshal.Cast <byte, SbDescriptor>(sbDescriptorData)[0]; BufferManager.SetComputeStorageBuffer(sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size); } ubEnableMask = 0; for (int index = 0; index < info.CBuffers.Count; index++) { ubEnableMask |= 1u << info.CBuffers[index].Slot; } BufferManager.SetComputeStorageBufferEnableMask(sbEnableMask); BufferManager.SetComputeUniformBufferEnableMask(ubEnableMask); 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.CbufOffset, descriptor.CbufSlot); } else { textureBindings[index] = new TextureBindingInfo(target, descriptor.HandleIndex); } } TextureManager.SetComputeTextures(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.SetComputeImages(imageBindings); BufferManager.CommitComputeBindings(); TextureManager.CommitComputeBindings(); _context.Renderer.Pipeline.DispatchCompute( qmd.CtaRasterWidth, qmd.CtaRasterHeight, qmd.CtaRasterDepth); UpdateShaderState(state); }
/// <summary> /// Dispatches compute work. /// </summary> /// <param name="state">Current GPU state</param> /// <param name="argument">Method call argument</param> public void Dispatch(GpuState state, int argument) { uint dispatchParamsAddress = (uint)state.Get <int>(MethodOffset.DispatchParamsAddress); var dispatchParams = _context.MemoryAccessor.Read <ComputeParams>((ulong)dispatchParamsAddress << 8); GpuVa shaderBaseAddress = state.Get <GpuVa>(MethodOffset.ShaderBaseAddress); ulong shaderGpuVa = shaderBaseAddress.Pack() + (uint)dispatchParams.ShaderOffset; // Note: A size of 0 is also invalid, the size must be at least 1. int sharedMemorySize = Math.Clamp(dispatchParams.SharedMemorySize & 0xffff, 1, _context.Capabilities.MaximumComputeSharedMemorySize); ComputeShader cs = ShaderCache.GetComputeShader( shaderGpuVa, sharedMemorySize, dispatchParams.UnpackBlockSizeX(), dispatchParams.UnpackBlockSizeY(), dispatchParams.UnpackBlockSizeZ()); _context.Renderer.Pipeline.SetProgram(cs.HostProgram); var samplerPool = state.Get <PoolState>(MethodOffset.SamplerPoolState); TextureManager.SetComputeSamplerPool(samplerPool.Address.Pack(), samplerPool.MaximumId, dispatchParams.SamplerIndex); var texturePool = state.Get <PoolState>(MethodOffset.TexturePoolState); TextureManager.SetComputeTexturePool(texturePool.Address.Pack(), texturePool.MaximumId); TextureManager.SetComputeTextureBufferIndex(state.Get <int>(MethodOffset.TextureBufferIndex)); ShaderProgramInfo info = cs.Shader.Program.Info; uint sbEnableMask = 0; uint ubEnableMask = dispatchParams.UnpackUniformBuffersEnableMask(); for (int index = 0; index < dispatchParams.UniformBuffers.Length; index++) { if ((ubEnableMask & (1 << index)) == 0) { continue; } ulong gpuVa = dispatchParams.UniformBuffers[index].PackAddress(); ulong size = dispatchParams.UniformBuffers[index].UnpackSize(); BufferManager.SetComputeUniformBuffer(index, gpuVa, size); } for (int index = 0; index < info.SBuffers.Count; index++) { BufferDescriptor sb = info.SBuffers[index]; sbEnableMask |= 1u << sb.Slot; ulong sbDescAddress = BufferManager.GetComputeUniformBufferAddress(0); int sbDescOffset = 0x310 + sb.Slot * 0x10; sbDescAddress += (ulong)sbDescOffset; ReadOnlySpan <byte> sbDescriptorData = _context.PhysicalMemory.GetSpan(sbDescAddress, 0x10); SbDescriptor sbDescriptor = MemoryMarshal.Cast <byte, SbDescriptor>(sbDescriptorData)[0]; BufferManager.SetComputeStorageBuffer(sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size); } ubEnableMask = 0; for (int index = 0; index < info.CBuffers.Count; index++) { ubEnableMask |= 1u << info.CBuffers[index].Slot; } BufferManager.SetComputeStorageBufferEnableMask(sbEnableMask); BufferManager.SetComputeUniformBufferEnableMask(ubEnableMask); 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.CbufOffset, descriptor.CbufSlot); } else { textureBindings[index] = new TextureBindingInfo(target, descriptor.HandleIndex); } } TextureManager.SetComputeTextures(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.SetComputeImages(imageBindings); BufferManager.CommitComputeBindings(); TextureManager.CommitComputeBindings(); _context.Renderer.Pipeline.DispatchCompute( dispatchParams.UnpackGridSizeX(), dispatchParams.UnpackGridSizeY(), dispatchParams.UnpackGridSizeZ()); UpdateShaderState(state); }