back to matrix, but trouble in combining L4 predict with matrixparallelqueue
This commit is contained in:
@@ -25,6 +25,7 @@
|
||||
//
|
||||
// -----------------------------------------------------------------------
|
||||
// -----------------------------------------------------------------------
|
||||
using StgSharp.HighPerformance.Memory;
|
||||
using StgSharp.Internal;
|
||||
using StgSharp.Mathematics;
|
||||
using System;
|
||||
@@ -37,7 +38,7 @@ using System.Threading.Tasks;
|
||||
|
||||
namespace StgSharp.Mathematics.Numeric
|
||||
{
|
||||
public partial class MatrixCompute
|
||||
public static partial class MatrixCompute
|
||||
{
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
@@ -45,8 +46,8 @@ namespace StgSharp.Mathematics.Numeric
|
||||
MatrixKernel<T>* buffer,
|
||||
int colLength,
|
||||
int x,
|
||||
int y)
|
||||
where T: unmanaged, INumber<T>
|
||||
int y
|
||||
) where T : unmanaged, INumber<T>
|
||||
{
|
||||
return (MatrixKernel*)(buffer + (colLength * x + y));
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user