1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
|
using Ryujinx.Graphics.Device;
using Ryujinx.Graphics.Gpu.Engine.InlineToMemory;
using Ryujinx.Graphics.Gpu.Engine.Threed;
using Ryujinx.Graphics.Gpu.Engine.Types;
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;
int samplerPoolMaximumId = _state.State.SetTexSamplerPoolCMaximumIndex;
GpuChannelPoolState poolState = new(
texturePoolGpuVa,
_state.State.SetTexHeaderPoolCMaximumIndex,
_state.State.SetBindlessTextureConstantBufferSlotSelect);
GpuChannelComputeState computeState = new(
qmd.CtaThreadDimension0,
qmd.CtaThreadDimension1,
qmd.CtaThreadDimension2,
localMemorySize,
sharedMemorySize,
_channel.BufferManager.HasUnalignedStorageBuffers);
CachedShaderProgram cs = memoryManager.Physical.ShaderCache.GetComputeShader(_channel, samplerPoolMaximumId, 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.SBuffers.Count; index++)
{
BufferDescriptor sb = info.SBuffers[index];
ulong sbDescAddress = _channel.BufferManager.GetComputeUniformBufferAddress(sb.SbCbSlot);
sbDescAddress += (ulong)sb.SbCbOffset * 4;
SbDescriptor sbDescriptor = _channel.MemoryManager.Physical.Read<SbDescriptor>(sbDescAddress);
uint size;
if (sb.SbCbSlot == Constants.DriverReservedUniformBuffer)
{
// Only trust the SbDescriptor size if it comes from slot 0.
size = (uint)sbDescriptor.Size;
}
else
{
// TODO: Use full mapped size and somehow speed up buffer sync.
size = (uint)_channel.MemoryManager.GetMappedSize(sbDescriptor.PackAddress(), Constants.MaxUnknownStorageSize);
}
_channel.BufferManager.SetComputeStorageBuffer(sb.Slot, sbDescriptor.PackAddress(), size, sb.Flags);
}
if (_channel.BufferManager.HasUnalignedStorageBuffers != computeState.HasUnalignedStorageBuffer)
{
// Refetch the shader, as assumptions about storage buffer alignment have changed.
computeState = new GpuChannelComputeState(
qmd.CtaThreadDimension0,
qmd.CtaThreadDimension1,
qmd.CtaThreadDimension2,
localMemorySize,
sharedMemorySize,
_channel.BufferManager.HasUnalignedStorageBuffers);
cs = memoryManager.Physical.ShaderCache.GetComputeShader(_channel, samplerPoolMaximumId, poolState, computeState, shaderGpuVa);
_context.Renderer.Pipeline.SetProgram(cs.HostProgram);
}
_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();
}
}
}
|
