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

many things

Browse Source
This commit is contained in:
Youmu Liu
2026-02-26 18:05:35 +08:00
parent 1b04196314
commit 15434ed053
15 changed files with 530 additions and 591 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Mathematics/Numeric/MatrixParallelThread.Model.cs
+19 -78
View File
@@ -34,100 +34,41 @@ namespace StgSharp.Mathematics.Numeric
public partial class MatrixParallelThread
{
internal static unsafe void BufferComputeBinary(
MatrixParallelTaskPackage* p
)
{
int size = p->ElementSize * 16;
long count = Unsafe.As<int, long>(ref p->PrimCount);
long offset = Unsafe.As<int, long>(ref p->PrimTileOffset) * size;
private const int KernelStealCount = 64;
((delegate* unmanaged[Cdecl]<void*, void*, void*, long, void>)p->ComputeHandle)((void*)(p->Left + offset),
(void*)(p->Right + offset),
(void*)(p->Answer + offset),
count);
MatrixParallelFactory.Release(p);
}
internal static unsafe void BufferComputeBinaryScalar(
MatrixParallelTaskPackage* p
internal unsafe void BufferCompute(
MatrixParallelTaskPackage* p
)
{
// while () { }
int size = p->ElementSize * 16;
long count = Unsafe.As<int, long>(ref p->PrimCount);
long offset = Unsafe.As<int, long>(ref p->PrimTileOffset) * size;
ScalarPacket* scalar = p->Scalar;
((delegate* unmanaged[Cdecl]<void*, void*, void*, ScalarPacket*, long, void>)p->ComputeHandle)((void*)(p->Left + offset),
(void*)(p->Right + offset),
(void*)(p->Answer + offset),
scalar,
count);
MatrixParallelFactory.Release(p);
IntPtr computeHandle = _context[p->ComputeHandle];
((delegate* unmanaged[Cdecl]<void*, void*, void*, void*, long, void>)computeHandle)((void*)(p->Left + offset),
(void*)(p->Right + offset),
(void*)(p->Answer + offset),
scalar,
count);
}
internal static unsafe void BufferComputeNoOperatorScalar(
MatrixParallelTaskPackage* p
[MethodImpl(MethodImplOptions.AggressiveInlining)]
internal unsafe void SpecialBufferCompute(
MatrixParallelTaskPackage* p
)
{
int size = p->ElementSize * 16;
long count = Unsafe.As<int, long>(ref p->PrimCount);
long offset = Unsafe.As<int, long>(ref p->PrimTileOffset) * size;
((delegate* unmanaged[Cdecl]<void*, ScalarPacket*, long, void>)p->ComputeHandle)((void*)(p->Answer + offset),
p->Scalar, count);
MatrixParallelFactory.Release(p);
}
internal static unsafe void BufferComputeUnary(
MatrixParallelTaskPackage* p
)
{
int size = p->ElementSize * 16;
long count = Unsafe.As<int, long>(ref p->PrimCount);
long offset = Unsafe.As<int, long>(ref p->PrimTileOffset) * size;
((delegate* unmanaged[Cdecl]<void*, void*, long, void>)p->ComputeHandle)((void*)(p->Right + offset),
(void*)(p->Answer + offset),
count);
MatrixParallelFactory.Release(p);
}
internal static unsafe void BufferComputeUnaryScalar(
MatrixParallelTaskPackage* p
)
{
int size = p->ElementSize * 16;
long count = Unsafe.As<int, long>(ref p->PrimCount);
long offset = Unsafe.As<int, long>(ref p->PrimTileOffset) * size;
((delegate* unmanaged[Cdecl]<void*, void*, ScalarPacket*, long, void>)p->ComputeHandle)((void*)(p->Right + offset),
(void*)(p->Answer + offset),
p->Scalar,
count);
MatrixParallelFactory.Release(p);
}
internal static unsafe void SpecialBufferCompute(
MatrixParallelTaskPackage* p
)
{
switch ((MatrixOperationSpecial)p->ComputeMode.ParameterStyle)
switch ((MatrixElementType)p->ElementSize)
{
case MatrixOperationSpecial.SEQ_FMA:
switch ((MatrixElementType)p->ElementSize)
{
case MatrixElementType.F64:
MatrixCompute.ComputeFmaPanelF64(p);
return;
case MatrixElementType.F32:
MatrixCompute.ComputeFmaPanelF32(p);
break;
default:
break;
}
case MatrixElementType.F64:
ComputeFmaPanelF64(p);
return;
case MatrixElementType.F32:
ComputeFmaPanelF32(p);
break;
default:
break;
}
MatrixParallelFactory.Release(p);
}
}