Separate GPU engines and make state follow official docs (part 1/2) (#2422)

* Use DeviceState for compute and i2m

* Migrate 2D class, more comments

* Migrate DMA copy engine

* Remove now unused code

* Replace GpuState by GpuAccessorState on GpuAcessor, since compute no longer has a GpuState

* More comments

* Add logging (disabled)

* Add back i2m on 3D engine
This commit is contained in:
gdkchan 2021-07-07 20:56:06 -03:00 committed by GitHub
parent 31cbd09a75
commit 8b44eb1c98
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
30 changed files with 2599 additions and 460 deletions

View file

@ -1,6 +1,5 @@
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.State;
using Ryujinx.Graphics.Shader;
namespace Ryujinx.Graphics.Gpu.Shader
@ -11,7 +10,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
class GpuAccessor : TextureDescriptorCapableGpuAccessor, IGpuAccessor
{
private readonly GpuContext _context;
private readonly GpuState _state;
private readonly GpuChannel _channel;
private readonly GpuAccessorState _state;
private readonly int _stageIndex;
private readonly bool _compute;
private readonly int _localSizeX;
@ -24,11 +24,13 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// Creates a new instance of the GPU state accessor for graphics shader translation.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="channel">GPU channel</param>
/// <param name="state">Current GPU state</param>
/// <param name="stageIndex">Graphics shader stage index (0 = Vertex, 4 = Fragment)</param>
public GpuAccessor(GpuContext context, GpuState state, int stageIndex)
public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state, int stageIndex)
{
_context = context;
_channel = channel;
_state = state;
_stageIndex = stageIndex;
}
@ -37,6 +39,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// Creates a new instance of the GPU state accessor for compute shader translation.
/// </summary>
/// <param name="context">GPU context</param>
/// <param name="channel">GPU channel</param>
/// <param name="state">Current GPU state</param>
/// <param name="localSizeX">Local group size X of the compute shader</param>
/// <param name="localSizeY">Local group size Y of the compute shader</param>
@ -45,7 +48,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="sharedMemorySize">Shared memory size of the compute shader</param>
public GpuAccessor(
GpuContext context,
GpuState state,
GpuChannel channel,
GpuAccessorState state,
int localSizeX,
int localSizeY,
int localSizeZ,
@ -53,6 +57,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
int sharedMemorySize)
{
_context = context;
_channel = channel;
_state = state;
_compute = true;
_localSizeX = localSizeX;
@ -79,7 +84,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <returns>Data at the memory location</returns>
public override T MemoryRead<T>(ulong address)
{
return _state.Channel.MemoryManager.Read<T>(address);
return _channel.MemoryManager.Read<T>(address);
}
/// <summary>
@ -89,7 +94,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <returns>True if the address is mapped, false otherwise</returns>
public bool MemoryMapped(ulong address)
{
return _state.Channel.MemoryManager.IsMapped(address);
return _channel.MemoryManager.IsMapped(address);
}
/// <summary>
@ -129,8 +134,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
public uint QueryConstantBufferUse()
{
return _compute
? _state.Channel.BufferManager.GetComputeUniformBufferUseMask()
: _state.Channel.BufferManager.GetGraphicsUniformBufferUseMask(_stageIndex);
? _channel.BufferManager.GetComputeUniformBufferUseMask()
: _channel.BufferManager.GetGraphicsUniformBufferUseMask(_stageIndex);
}
/// <summary>
@ -196,11 +201,22 @@ namespace Ryujinx.Graphics.Gpu.Shader
{
if (_compute)
{
return _state.Channel.TextureManager.GetComputeTextureDescriptor(_state, handle, cbufSlot);
return _channel.TextureManager.GetComputeTextureDescriptor(
_state.TexturePoolGpuVa,
_state.TextureBufferIndex,
_state.TexturePoolMaximumId,
handle,
cbufSlot);
}
else
{
return _state.Channel.TextureManager.GetGraphicsTextureDescriptor(_state, _stageIndex, handle, cbufSlot);
return _channel.TextureManager.GetGraphicsTextureDescriptor(
_state.TexturePoolGpuVa,
_state.TextureBufferIndex,
_state.TexturePoolMaximumId,
_stageIndex,
handle,
cbufSlot);
}
}
@ -210,7 +226,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <returns>True if early depth testing is forced</returns>
public bool QueryEarlyZForce()
{
return _state.Get<bool>(MethodOffset.EarlyZForce);
return _state.EarlyZForce;
}
}
}

View file

@ -0,0 +1,43 @@
namespace Ryujinx.Graphics.Gpu.Shader
{
/// <summary>
/// State used by the <see cref="GpuAccessor"/>.
/// </summary>
struct GpuAccessorState
{
/// <summary>
/// GPU virtual address of the texture pool.
/// </summary>
public ulong TexturePoolGpuVa { get; }
/// <summary>
/// Maximum ID of the texture pool.
/// </summary>
public int TexturePoolMaximumId { get; }
/// <summary>
/// Constant buffer slot where the texture handles are located.
/// </summary>
public int TextureBufferIndex { get; }
/// <summary>
/// Early Z force enable.
/// </summary>
public bool EarlyZForce { get; }
/// <summary>
/// Creates a new instance of the GPU accessor state.
/// </summary>
/// <param name="texturePoolGpuVa">GPU virtual address of the texture pool</param>
/// <param name="texturePoolMaximumId">Maximum ID of the texture pool</param>
/// <param name="textureBufferIndex">Constant buffer slot where the texture handles are located</param>
/// <param name="earlyZForce">Early Z force enable</param>
public GpuAccessorState(ulong texturePoolGpuVa, int texturePoolMaximumId, int textureBufferIndex, bool earlyZForce)
{
TexturePoolGpuVa = texturePoolGpuVa;
TexturePoolMaximumId = texturePoolMaximumId;
TextureBufferIndex = textureBufferIndex;
EarlyZForce = earlyZForce;
}
}
}

View file

@ -475,7 +475,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <remarks>
/// This automatically translates, compiles and adds the code to the cache if not present.
/// </remarks>
/// <param name="state">Current GPU state</param>
/// <param name="channel">GPU channel</param>
/// <param name="gas">GPU accessor state</param>
/// <param name="gpuVa">GPU virtual address of the binary shader code</param>
/// <param name="localSizeX">Local group size X of the computer shader</param>
/// <param name="localSizeY">Local group size Y of the computer shader</param>
@ -484,7 +485,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="sharedMemorySize">Shared memory size of the compute shader</param>
/// <returns>Compiled compute shader code</returns>
public ShaderBundle GetComputeShader(
GpuState state,
GpuChannel channel,
GpuAccessorState gas,
ulong gpuVa,
int localSizeX,
int localSizeY,
@ -498,7 +500,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
{
foreach (ShaderBundle cachedCpShader in list)
{
if (IsShaderEqual(state.Channel.MemoryManager, cachedCpShader, gpuVa))
if (IsShaderEqual(channel.MemoryManager, cachedCpShader, gpuVa))
{
return cachedCpShader;
}
@ -508,7 +510,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
TranslatorContext[] shaderContexts = new TranslatorContext[1];
shaderContexts[0] = DecodeComputeShader(
state,
channel,
gas,
gpuVa,
localSizeX,
localSizeY,
@ -533,7 +536,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
isShaderCacheReadOnly = _cacheManager.IsReadOnly;
// Compute hash and prepare data for shader disk cache comparison.
shaderCacheEntries = CacheHelper.CreateShaderCacheEntries(state.Channel.MemoryManager, shaderContexts);
shaderCacheEntries = CacheHelper.CreateShaderCacheEntries(channel.MemoryManager, shaderContexts);
programCodeHash = CacheHelper.ComputeGuestHashFromCache(shaderCacheEntries);
}
@ -548,7 +551,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
}
// The shader isn't currently cached, translate it and compile it.
ShaderCodeHolder shader = TranslateShader(state.Channel.MemoryManager, shaderContexts[0]);
ShaderCodeHolder shader = TranslateShader(channel.MemoryManager, shaderContexts[0]);
shader.HostShader = _context.Renderer.CompileShader(ShaderStage.Compute, shader.Program.Code);
@ -832,7 +835,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <summary>
/// Decode the binary Maxwell shader code to a translator context.
/// </summary>
/// <param name="state">Current GPU state</param>
/// <param name="channel">GPU channel</param>
/// <param name="gas">GPU accessor state</param>
/// <param name="gpuVa">GPU virtual address of the binary shader code</param>
/// <param name="localSizeX">Local group size X of the computer shader</param>
/// <param name="localSizeY">Local group size Y of the computer shader</param>
@ -841,7 +845,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
/// <param name="sharedMemorySize">Shared memory size of the compute shader</param>
/// <returns>The generated translator context</returns>
private TranslatorContext DecodeComputeShader(
GpuState state,
GpuChannel channel,
GpuAccessorState gas,
ulong gpuVa,
int localSizeX,
int localSizeY,
@ -854,7 +859,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
return null;
}
GpuAccessor gpuAccessor = new GpuAccessor(_context, state, localSizeX, localSizeY, localSizeZ, localMemorySize, sharedMemorySize);
GpuAccessor gpuAccessor = new GpuAccessor(_context, channel, gas, localSizeX, localSizeY, localSizeZ, localMemorySize, sharedMemorySize);
var options = new TranslationOptions(TargetLanguage.Glsl, TargetApi.OpenGL, DefaultFlags | TranslationFlags.Compute);
return Translator.CreateContext(gpuVa, gpuAccessor, options);
@ -884,7 +889,13 @@ namespace Ryujinx.Graphics.Gpu.Shader
return null;
}
GpuAccessor gpuAccessor = new GpuAccessor(_context, state, (int)stage - 1);
GpuAccessorState gas = new GpuAccessorState(
state.Get<PoolState>(MethodOffset.TexturePoolState).Address.Pack(),
state.Get<PoolState>(MethodOffset.TexturePoolState).MaximumId,
state.Get<int>(MethodOffset.TextureBufferIndex),
state.Get<Boolean32>(MethodOffset.EarlyZForce));
GpuAccessor gpuAccessor = new GpuAccessor(_context, state.Channel, gas, (int)stage - 1);
var options = new TranslationOptions(TargetLanguage.Glsl, TargetApi.OpenGL, flags);
return Translator.CreateContext(gpuVa, gpuAccessor, options, counts);