/// <summary>
/// Updates host shaders based on the guest GPU state.
/// </summary>
private void UpdateShaderState()
{
ShaderAddresses addresses = new ShaderAddresses();
Span <ShaderAddresses> addressesSpan = MemoryMarshal.CreateSpan(ref addresses, 1);
Span <ulong> addressesArray = MemoryMarshal.Cast <ShaderAddresses, ulong>(addressesSpan);
ulong baseAddress = _state.State.ShaderBaseAddress.Pack();
for (int index = 0; index < 6; index++)
{
var shader = _state.State.ShaderState[index];
if (!shader.UnpackEnable() && index != 1)
{
continue;
}
addressesArray[index] = baseAddress + shader.Offset;
}
GpuAccessorState gas = new GpuAccessorState(
_state.State.TexturePoolState.Address.Pack(),
_state.State.TexturePoolState.MaximumId,
(int)_state.State.TextureBufferIndex,
_state.State.EarlyZForce,
_drawState.Topology);
ShaderBundle gs = _channel.MemoryManager.Physical.ShaderCache.GetGraphicsShader(ref _state.State, _channel, gas, addresses);
byte oldVsClipDistancesWritten = _vsClipDistancesWritten;
_drawState.VsUsesInstanceId = gs.Shaders[0]?.Info.UsesInstanceId ?? false;
_vsClipDistancesWritten = gs.Shaders[0]?.Info.ClipDistancesWritten ?? 0;
if (oldVsClipDistancesWritten != _vsClipDistancesWritten)
{
UpdateUserClipState();
}
int storageBufferBindingsCount = 0;
int uniformBufferBindingsCount = 0;
for (int stage = 0; stage < Constants.ShaderStages; stage++)
{
ShaderProgramInfo info = gs.Shaders[stage]?.Info;
_currentProgramInfo[stage] = info;
if (info == null)
{
_channel.TextureManager.SetGraphicsTextures(stage, Array.Empty <TextureBindingInfo>());
_channel.TextureManager.SetGraphicsImages(stage, Array.Empty <TextureBindingInfo>());
_channel.BufferManager.SetGraphicsStorageBufferBindings(stage, null);
_channel.BufferManager.SetGraphicsUniformBufferBindings(stage, 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 = ShaderTexture.GetTarget(descriptor.Type);
textureBindings[index] = new TextureBindingInfo(
target,
descriptor.Binding,
descriptor.CbufSlot,
descriptor.HandleIndex,
descriptor.Flags);
}
_channel.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 = ShaderTexture.GetTarget(descriptor.Type);
Format format = ShaderTexture.GetFormat(descriptor.Format);
imageBindings[index] = new TextureBindingInfo(
target,
format,
descriptor.Binding,
descriptor.CbufSlot,
descriptor.HandleIndex,
descriptor.Flags);
}
_channel.TextureManager.SetGraphicsImages(stage, imageBindings);
_channel.BufferManager.SetGraphicsStorageBufferBindings(stage, info.SBuffers);
_channel.BufferManager.SetGraphicsUniformBufferBindings(stage, info.CBuffers);
if (info.SBuffers.Count != 0)
{
storageBufferBindingsCount = Math.Max(storageBufferBindingsCount, info.SBuffers.Max(x => x.Binding) + 1);
}
if (info.CBuffers.Count != 0)
{
uniformBufferBindingsCount = Math.Max(uniformBufferBindingsCount, info.CBuffers.Max(x => x.Binding) + 1);
}
}
_channel.BufferManager.SetGraphicsStorageBufferBindingsCount(storageBufferBindingsCount);
_channel.BufferManager.SetGraphicsUniformBufferBindingsCount(uniformBufferBindingsCount);
_context.Renderer.Pipeline.SetProgram(gs.HostProgram);
}
/// <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();
}