youmuliu/stgsharp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

mat queue and mat predictor confuses. i need time to design and organize

Browse Source
This commit is contained in:
Youmu Liu
2026-05-06 18:07:31 +08:00
parent 1338bf2801
commit e636410ba6
5 changed files with 112 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Mathematics/Numeric/Runtime/MatrixParallelQueue.cs
+24 -26
View File
@@ -33,6 +33,7 @@ using System.Collections.Concurrent;
using System.Collections.Generic;
using System.Linq;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices.JavaScript;
using System.Text;
using System.Threading;
using System.Threading.Tasks;
@@ -42,12 +43,6 @@ namespace StgSharp.Mathematics.Numeric
internal unsafe interface IMatrixParallelQueue<TSelf> where TSelf : IMatrixParallelQueue<TSelf>
{
static abstract TSelf Create(
MatrixParallelTask* baseTask,
L4 buffer,
int numaCount
);
void Return(
MatrixParallelTask* task,
in L4CacheLine left,
@@ -64,32 +59,35 @@ namespace StgSharp.Mathematics.Numeric
}
internal sealed unsafe class BufferQueueLeftRightAns : IMatrixParallelQueue<BufferQueueLeftRightAns>
internal interface IBufferQueueLeftRightAns : IMatrixParallelQueue<IBufferQueueLeftRightAns> { }
internal sealed unsafe class BufferQueueLeftRightAns_FullNuma : IMatrixParallelQueue<IBufferQueueLeftRightAns>
{
private MatrixParallelTask* _baseTask;
private bool _needNuma;
private L4 _buffer;
private readonly nuint _byteSize;
private readonly nuint _count;
private volatile nuint _cursor;
private readonly nuint Count;
private BufferQueueLeftRightAns() { }
private volatile nuint _mat1;
private volatile nuint _mat2;
private volatile nuint _mat3;
private BufferQueueLeftRightAns(
MatrixParallelTask* baseTask
private BufferQueueLeftRightAns_FullNuma() { }
internal BufferQueueLeftRightAns_FullNuma(
MatrixParallelTask* baseTask
)
{
_baseTask = baseTask;
Count = ((nuint)_baseTask->CommonX) * ((nuint)_baseTask->CommonY);
}
_count = ((nuint)_baseTask->CommonX) * ((nuint)_baseTask->CommonY);
_byteSize = ((nuint)_baseTask->Mat1) + (((nuint)_baseTask->CommonX) * ((nuint)_baseTask->CommonY) * _baseTask->ElementType.Size);
public static unsafe BufferQueueLeftRightAns Create(
MatrixParallelTask* baseTask,
L4 buffer,
int numaCount
)
{
return new BufferQueueLeftRightAns(baseTask);
_mat1 = (nuint)_baseTask->Mat1;
_mat2 = (nuint)_baseTask->Mat2;
_mat3 = (nuint)_baseTask->Mat3;
}
public unsafe void Return(
@@ -121,11 +119,11 @@ namespace StgSharp.Mathematics.Numeric
}
private class Predictor(
BufferQueueLeftRightAns host
BufferQueueLeftRightAns_FullNuma host
) : IL4Predict
{
private readonly BufferQueueLeftRightAns _host;
private readonly BufferQueueLeftRightAns_FullNuma _host;
public PredictResult Predict(
Span<CacheLineDescription> node,
@@ -136,16 +134,16 @@ namespace StgSharp.Mathematics.Numeric
nuint mat1 = (nuint)_host._baseTask->Mat1;
nuint mat2 = (nuint)_host._baseTask->Mat2;
nuint mat3 = (nuint)_host._baseTask->Mat3;
nuint _cur = begin * _host._baseTask->ElementType.Size;
nuint _cur = 16 * begin * _host._baseTask->ElementType.Size;
nuint mat1_cur = ((nuint)_host._baseTask->Mat1) + _cur;
nuint mat2_cur = ((nuint)_host._baseTask->Mat2) + _cur;
nuint mat3_cur = ((nuint)_host._baseTask->Mat3) + _cur;
nuint end = ((nuint)_host._baseTask->Mat1) + (((nuint)_host._baseTask->CommonX) * ((nuint)_host._baseTask->CommonY) * _host._baseTask->ElementType.Size);
nuint end = _host._byteSize;
nuint mat1_end = ((nuint)_host._baseTask->Mat1) + end;
nuint mat2_end = ((nuint)_host._baseTask->Mat2) + end;
nuint mat3_end = ((nuint)_host._baseTask->Mat3) + end;
nuint size = 1024 * 16 * _host._baseTask->ElementType.Size;
nuint size = 1024 * 16 * ((nuint)_host._baseTask->ElementType.Size);
for (int i = 0; i < node.Length; i += 3)
{
int cycle = i / 3;