crafterpika/MeloNX
0
0
Fork 0
mirror of https://git.743378673.xyz/MeloNX/MeloNX.git synced 2025-07-01 12:56:25 +02:00
Code Issues Projects Releases Packages Wiki Activity

Implement shader storage buffer operations using new Load/Store instructions (#4993)

Browse source 
* Implement storage buffer operations using new Load/Store instruction

* Extend GenerateMultiTargetStorageOp to also match access with constant offset, and log and comments

* Remove now unused code

* Catch more complex cases of global memory usage

* Shader cache version bump

* Extend global access elimination to work with more shared memory cases

* Change alignment requirement from 16 bytes to 8 bytes, handle cases where we need more than 16 storage buffers

* Tweak preferencing to catch more cases

* Enable CB0 elimination even when host storage buffer alignment is > 16 (for Intel)

* Fix storage buffer bindings

* Simplify some code

* Shader cache version bump

* Fix typo

* Extend global memory elimination to handle shared memory with multiple possible offsets and local memory
This commit is contained in:
gdkchan 2023-06-03 20:12:18 -03:00 • committed by GitHub
parent 81c9052847
commit 21c9ac6240
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
42 changed files with 1468 additions and 1259 deletions
Download patch file Download diff file Expand all files Collapse all files

151
src/Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs
View file

@ -110,12 +110,6 @@ namespace Ryujinx.Graphics.Shader.Translation
public UInt128 NextInputAttributesComponents { get; private set; }
public UInt128 ThisInputAttributesComponents { get; private set; }
public int AccessibleStorageBuffersMask { get; private set; }
public int AccessibleConstantBuffersMask { get; private set; }
private int _usedStorageBuffers;
private int _usedStorageBuffersWrite;
private readonly record struct TextureInfo(int CbufSlot, int Handle, bool Indexed, TextureFormat Format);
private struct TextureMeta
@ -127,18 +121,9 @@ namespace Ryujinx.Graphics.Shader.Translation
private readonly Dictionary<TextureInfo, TextureMeta> _usedTextures;
private readonly Dictionary<TextureInfo, TextureMeta> _usedImages;
private readonly Dictionary<int, int> _sbSlots;
private readonly Dictionary<int, int> _sbSlotsReverse;
private BufferDescriptor[] _cachedStorageBufferDescriptors;
private TextureDescriptor[] _cachedTextureDescriptors;
private TextureDescriptor[] _cachedImageDescriptors;
private int _firstStorageBufferBinding;
public int FirstStorageBufferBinding => _firstStorageBufferBinding;
public ShaderConfig(ShaderStage stage, IGpuAccessor gpuAccessor, TranslationOptions options)
{
Stage = stage;
@ -147,18 +132,12 @@ namespace Ryujinx.Graphics.Shader.Translation
_transformFeedbackDefinitions = new Dictionary<TransformFeedbackVariable, TransformFeedbackOutput>();
AccessibleStorageBuffersMask = (1 << GlobalMemory.StorageMaxCount) - 1;
AccessibleConstantBuffersMask = (1 << GlobalMemory.UbeMaxCount) - 1;
UsedInputAttributesPerPatch = new HashSet<int>();
UsedOutputAttributesPerPatch = new HashSet<int>();
_usedTextures = new Dictionary<TextureInfo, TextureMeta>();
_usedImages = new Dictionary<TextureInfo, TextureMeta>();
_sbSlots = new Dictionary<int, int>();
_sbSlotsReverse = new Dictionary<int, int>();
ResourceManager = new ResourceManager(stage, gpuAccessor, new ShaderProperties());
}
@ -173,11 +152,6 @@ namespace Ryujinx.Graphics.Shader.Translation
OutputTopology = outputTopology;
MaxOutputVertices = maxOutputVertices;
TransformFeedbackEnabled = gpuAccessor.QueryTransformFeedbackEnabled();
if (Stage != ShaderStage.Compute)
{
AccessibleConstantBuffersMask = 0;
}
}
public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options) : this(header.Stage, gpuAccessor, options)
@ -433,8 +407,6 @@ namespace Ryujinx.Graphics.Shader.Translation
UsedInputAttributes |= other.UsedInputAttributes;
UsedOutputAttributes |= other.UsedOutputAttributes;
_usedStorageBuffers |= other._usedStorageBuffers;
_usedStorageBuffersWrite |= other._usedStorageBuffersWrite;
foreach (var kv in other._usedTextures)
{
@ -634,23 +606,6 @@ namespace Ryujinx.Graphics.Shader.Translation
UsedFeatures |= flags;
}
public void SetAccessibleBufferMasks(int sbMask, int ubeMask)
{
AccessibleStorageBuffersMask = sbMask;
AccessibleConstantBuffersMask = ubeMask;
}
public void SetUsedStorageBuffer(int slot, bool write)
{
int mask = 1 << slot;
_usedStorageBuffers |= mask;
if (write)
{
_usedStorageBuffersWrite |= mask;
}
}
public void SetUsedTexture(
Instruction inst,
SamplerType type,
@ -756,76 +711,6 @@ namespace Ryujinx.Graphics.Shader.Translation
return meta;
}
public BufferDescriptor[] GetStorageBufferDescriptors()
{
if (_cachedStorageBufferDescriptors != null)
{
return _cachedStorageBufferDescriptors;
}
return _cachedStorageBufferDescriptors = GetStorageBufferDescriptors(
_usedStorageBuffers,
_usedStorageBuffersWrite,
true,
out _firstStorageBufferBinding,
GpuAccessor.QueryBindingStorageBuffer);
}
private BufferDescriptor[] GetStorageBufferDescriptors(
int usedMask,
int writtenMask,
bool isArray,
out int firstBinding,
Func<int, int> getBindingCallback)
{
firstBinding = 0;
bool hasFirstBinding = false;
var descriptors = new BufferDescriptor[BitOperations.PopCount((uint)usedMask)];
int lastSlot = -1;
for (int i = 0; i < descriptors.Length; i++)
{
int slot = BitOperations.TrailingZeroCount(usedMask);
if (isArray)
{
// The next array entries also consumes bindings, even if they are unused.
for (int j = lastSlot + 1; j < slot; j++)
{
int binding = getBindingCallback(j);
if (!hasFirstBinding)
{
firstBinding = binding;
hasFirstBinding = true;
}
}
}
lastSlot = slot;
(int sbCbSlot, int sbCbOffset) = GetSbCbInfo(slot);
descriptors[i] = new BufferDescriptor(getBindingCallback(slot), slot, sbCbSlot, sbCbOffset);
if (!hasFirstBinding)
{
firstBinding = descriptors[i].Binding;
hasFirstBinding = true;
}
if ((writtenMask & (1 << slot)) != 0)
{
descriptors[i].SetFlag(BufferUsageFlags.Write);
}
usedMask &= ~(1 << slot);
}
return descriptors;
}
public TextureDescriptor[] GetTextureDescriptors()
{
return _cachedTextureDescriptors ??= GetTextureOrImageDescriptors(_usedTextures, GpuAccessor.QueryBindingTexture);
@ -922,45 +807,11 @@ namespace Ryujinx.Graphics.Shader.Translation
return FindDescriptorIndex(GetImageDescriptors(), texOp);
}
public int GetSbSlot(byte sbCbSlot, ushort sbCbOffset)
{
int key = PackSbCbInfo(sbCbSlot, sbCbOffset);
if (!_sbSlots.TryGetValue(key, out int slot))
{
slot = _sbSlots.Count;
_sbSlots.Add(key, slot);
_sbSlotsReverse.Add(slot, key);
}
return slot;
}
public (int, int) GetSbCbInfo(int slot)
{
if (_sbSlotsReverse.TryGetValue(slot, out int key))
{
return UnpackSbCbInfo(key);
}
throw new ArgumentException($"Invalid slot {slot}.", nameof(slot));
}
private static int PackSbCbInfo(int sbCbSlot, int sbCbOffset)
{
return sbCbOffset | ((int)sbCbSlot << 16);
}
private static (int, int) UnpackSbCbInfo(int key)
{
return ((byte)(key >> 16), (ushort)key);
}
public ShaderProgramInfo CreateProgramInfo(ShaderIdentification identification = ShaderIdentification.None)
{
return new ShaderProgramInfo(
ResourceManager.GetConstantBufferDescriptors(),
GetStorageBufferDescriptors(),
ResourceManager.GetStorageBufferDescriptors(),
GetTextureDescriptors(),
GetImageDescriptors(),
identification,