1
0
Fork 0
mirror of https://github.com/Ryujinx/Ryujinx.git synced 2024-11-18 11:36:41 +00:00
Ryujinx/Ryujinx.Graphics.Gpu/Engine/Compute/ComputeClass.cs
riperiperi 4965681e06
GPU: Swap bindings array instead of copying (#4003)
* GPU: Swap bindings array instead of copying

Reduces work on UpdateShaderState. Now the cost is a few reference moves for arrays, rather than copying data.

Downside: bindings arrays are no longer readonly.

* Micro optimisation

* Add missing docs

* Address Feedback
2022-12-04 18:18:40 +01:00

219 lines
8.9 KiB
C#

using Ryujinx.Graphics.Device;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Engine.InlineToMemory;
using Ryujinx.Graphics.Gpu.Engine.Threed;
using Ryujinx.Graphics.Gpu.Engine.Types;
using Ryujinx.Graphics.Gpu.Image;
using Ryujinx.Graphics.Gpu.Shader;
using Ryujinx.Graphics.Shader;
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
namespace Ryujinx.Graphics.Gpu.Engine.Compute
{
/// <summary>
/// Represents a compute engine class.
/// </summary>
class ComputeClass : IDeviceState
{
private readonly GpuContext _context;
private readonly GpuChannel _channel;
private readonly ThreedClass _3dEngine;
private readonly DeviceState<ComputeClassState> _state;
private readonly InlineToMemoryClass _i2mClass;
/// <summary>
/// Creates a new instance of the compute engine class.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="channel">GPU channel</param>
/// <param name="threedEngine">3D engine</param>
public ComputeClass(GpuContext context, GpuChannel channel, ThreedClass threedEngine)
{
_context = context;
_channel = channel;
_3dEngine = threedEngine;
_state = new DeviceState<ComputeClassState>(new Dictionary<string, RwCallback>
{
{ nameof(ComputeClassState.LaunchDma), new RwCallback(LaunchDma, null) },
{ nameof(ComputeClassState.LoadInlineData), new RwCallback(LoadInlineData, null) },
{ nameof(ComputeClassState.SendSignalingPcasB), new RwCallback(SendSignalingPcasB, null) }
});
_i2mClass = new InlineToMemoryClass(context, channel, initializeState: false);
}
/// <summary>
/// Reads data from the class registers.
/// </summary>
/// <param name="offset">Register byte offset</param>
/// <returns>Data at the specified offset</returns>
public int Read(int offset) => _state.Read(offset);
/// <summary>
/// Writes data to the class registers.
/// </summary>
/// <param name="offset">Register byte offset</param>
/// <param name="data">Data to be written</param>
public void Write(int offset, int data) => _state.Write(offset, data);
/// <summary>
/// Launches the Inline-to-Memory DMA copy operation.
/// </summary>
/// <param name="argument">Method call argument</param>
private void LaunchDma(int argument)
{
_i2mClass.LaunchDma(ref Unsafe.As<ComputeClassState, InlineToMemoryClassState>(ref _state.State), argument);
}
/// <summary>
/// Pushes a block of data to the Inline-to-Memory engine.
/// </summary>
/// <param name="data">Data to push</param>
public void LoadInlineData(ReadOnlySpan<int> data)
{
_i2mClass.LoadInlineData(data);
}
/// <summary>
/// Pushes a word of data to the Inline-to-Memory engine.
/// </summary>
/// <param name="argument">Method call argument</param>
private void LoadInlineData(int argument)
{
_i2mClass.LoadInlineData(argument);
}
/// <summary>
/// Performs the compute dispatch operation.
/// </summary>
/// <param name="argument">Method call argument</param>
private void SendSignalingPcasB(int argument)
{
var memoryManager = _channel.MemoryManager;
// Since we're going to change the state, make sure any pending instanced draws are done.
_3dEngine.PerformDeferredDraws();
// Make sure all pending uniform buffer data is written to memory.
_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,
_channel.BufferManager.HasUnalignedStorageBuffers);
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;
bool hasUnaligned = _channel.BufferManager.HasUnalignedStorageBuffers;
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);
}
if ((_channel.BufferManager.HasUnalignedStorageBuffers) != hasUnaligned)
{
// Refetch the shader, as assumptions about storage buffer alignment have changed.
cs = memoryManager.Physical.ShaderCache.GetComputeShader(_channel, poolState, computeState, shaderGpuVa);
_context.Renderer.Pipeline.SetProgram(cs.HostProgram);
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);
}
_channel.BufferManager.SetComputeBufferBindings(cs.Bindings);
_channel.TextureManager.SetComputeBindings(cs.Bindings);
// 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();
}
}
}