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/MatrixCompute.cs
-62
View File
@@ -40,68 +40,6 @@ namespace StgSharp.Mathematics.Numeric
public partial class MatrixCompute
{
internal static unsafe void ComputeFmaPanelF32(
MatrixParallelTaskPackage* p
)
{
int f32 = (int)MatrixElementType.F32;
MatrixKernel* leftBuffer = (MatrixKernel*)p->Left;
MatrixKernel* rightBuffer = (MatrixKernel*)p->Right;
MatrixKernel* ansBuffer = (MatrixKernel*)p->Answer;
ref long offsetRef = ref Unsafe.As<int, long>(ref p->PrimTileOffset);
ref long lengthRef = ref Unsafe.As<int, long>(ref p->PrimCount);
int ansWidth = p->AnsPrimOffset;
int ansHeight = p->AnsSecOffset;
int commonK = p->LeftPrimOffset;
M128* enumeration = stackalloc M128[1];
enumeration->Member<int>(2) = commonK;
enumeration->Member<int>(3) = ansHeight;
// in this occasion size must be 1
for (long c = offsetRef; c < offsetRef + lengthRef; c++)
{
int i = (int)(c / ansHeight);
int j = (int)(c % ansHeight);
enumeration->Member<int>(0) = i;
enumeration->Member<int>(1) = j;
NativeIntrinsic.Context.mat[f32].kernel_tile_fma(leftBuffer, rightBuffer, ansBuffer,
enumeration);
}
}
internal static unsafe void ComputeFmaPanelF64(
MatrixParallelTaskPackage* p
)
{
int f64 = (int)MatrixElementType.F64;
// int size = 1;
MatrixKernel* leftBuffer = (MatrixKernel*)p->Left;
MatrixKernel* rightBuffer = (MatrixKernel*)p->Right;
MatrixKernel* ansBuffer = (MatrixKernel*)p->Answer;
ref long offsetRef = ref Unsafe.As<int, long>(ref p->PrimTileOffset);
ref long lengthRef = ref Unsafe.As<int, long>(ref p->PrimCount);
int ansWidth = p->AnsPrimOffset;
int ansHeight = p->AnsSecOffset;
int commonK = p->LeftPrimOffset;
M128* enumeration = stackalloc M128[1];
enumeration->Member<int>(2) = commonK;
enumeration->Member<int>(3) = ansHeight;
// in this occasion size must be 1
for (long c = offsetRef; c < offsetRef + lengthRef; c++)
{
int i = (int)(c / ansHeight);
int j = (int)(c % ansHeight);
enumeration->Member<int>(0) = i;
enumeration->Member<int>(1) = j;
NativeIntrinsic.Context.mat[f64].kernel_tile_fma(leftBuffer, rightBuffer, ansBuffer,
enumeration);
}
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
internal static unsafe MatrixKernel* GetKernelAddressUnsafe<T>(
MatrixKernel<T>* buffer,