diff --git a/StgSharp.Common/HighPerformance/MatrixKernel.cs b/StgSharp.Common/HighPerformance/MatrixKernel.cs
index d47d2b68..252c0970 100644
--- a/StgSharp.Common/HighPerformance/MatrixKernel.cs
+++ b/StgSharp.Common/HighPerformance/MatrixKernel.cs
@@ -41,20 +41,26 @@ using System.Threading.Tasks;
 
 namespace StgSharp.HighPerformance
 {
-    [StructLayout(LayoutKind.Explicit, Size = 64)]
+    [StructLayout(LayoutKind.Sequential)]
     public unsafe struct MatrixKernel<T> where T: unmanaged, INumber<T>
     {
 
-        [FieldOffset(0)] internal fixed byte Buffer[64];
-        [FieldOffset(0)] internal fixed double FP64Buffer[8];
-        [FieldOffset(0)] internal fixed float FP32Buffer[16];
-        [FieldOffset(0)] public M128 XMM0;
-        [FieldOffset(16)] public M128 XMM1;
-        [FieldOffset(32)] public M128 XMM2;
-        [FieldOffset(48)] public M128 XMM3;
-        [FieldOffset(0)] public M256 YMM0;
-        [FieldOffset(32)] public M256 YMM1;
-        [FieldOffset(64)] public M512 ZMM0;
+        private T c0r0;
+        private T c0r1;
+        private T c0r2;
+        private T c0r3;
+        private T c1r0;
+        private T c1r1;
+        private T c1r2;
+        private T c1r3;
+        private T c2r0;
+        private T c2r1;
+        private T c2r2;
+        private T c2r3;
+        private T c3r0;
+        private T c3r1;
+        private T c3r2;
+        private T c3r3;
 
         public MatrixKernel()
         {
@@ -68,24 +74,10 @@ namespace StgSharp.HighPerformance
             }
         }
 
-        public static MatrixKernel<T> Unit
+        public ref T this[int column, int row]
         {
-            get
-            {
-                MatrixKernel<T> kernel = new(true);
-                kernel.FP32Buffer[0] = 1f;
-                kernel.FP32Buffer[3] = 1f;
-                kernel.FP32Buffer[8] = 1f;
-                kernel.FP32Buffer[15] = 1f;
-                return kernel;
-            }
-        }
-
-        public ref T UnsafeIndex(int column, int row)
-        {
-            ref M128 col = ref Unsafe.As<byte, M128>(ref Buffer[0]);
-            col = ref Unsafe.Add(ref col, column);
-            return ref col.Member<T>(row);
+            [MethodImpl(MethodImplOptions.AggressiveInlining)]
+            get => ref Unsafe.Add(ref c0r0, (column * 4) + row);
         }
 
     }
diff --git a/StgSharp.Common/Mathematics/Linear/MatrixParallel.TaskPublic.cs b/StgSharp.Common/Mathematics/Linear/MatrixParallel.Publish.cs
similarity index 91%
rename from StgSharp.Common/Mathematics/Linear/MatrixParallel.TaskPublic.cs
rename to StgSharp.Common/Mathematics/Linear/MatrixParallel.Publish.cs
index b0b241d3..49d50125 100644
--- a/StgSharp.Common/Mathematics/Linear/MatrixParallel.TaskPublic.cs
+++ b/StgSharp.Common/Mathematics/Linear/MatrixParallel.Publish.cs
@@ -49,8 +49,11 @@ namespace StgSharp.Mathematics
                                   delegate*<MatrixKernel<T>*, MatrixKernel<T>*, MatrixKernel<T>*, void> Operation)
             where T: unmanaged, INumber<T>
         {
-            int columnCount = left.PrimSize;
-            int rowCount = right.PrimSize;
+            int length = _threads.Length,
+                columnCount = left.PrimSize,
+                rowCount = right.PrimSize,
+                cAverage = left.PrimSize / length, cRemain = left.PrimSize % length,
+                rAverage = right.PrimSize / length, rRemain = right.PrimSize % length;
 
             MatrixParallelTaskPackage<T> package = new()
             {
diff --git a/StgSharp.Native/StgSharp.Native.vcxproj.filters b/StgSharp.Native/StgSharp.Native.vcxproj.filters
index 563ce522..e71f202f 100644
--- a/StgSharp.Native/StgSharp.Native.vcxproj.filters
+++ b/StgSharp.Native/StgSharp.Native.vcxproj.filters
@@ -13,9 +13,6 @@
       <UniqueIdentifier>{67DA6AB6-F800-4c08-8B7A-83BB121AAD01}</UniqueIdentifier>
       <Extensions>rc;ico;cur;bmp;dlg;rc2;rct;bin;rgs;gif;jpg;jpeg;jpe;resx;tiff;tif;png;wav;mfcribbon-ms</Extensions>
     </Filter>
-    <Filter Include="sdl-src">
-      <UniqueIdentifier>{bfdaad49-d6af-4b2e-aba9-2dcaa7f7d693}</UniqueIdentifier>
-    </Filter>
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="sn-src\dllmain.c">
diff --git a/StgSharp.Native/sn-head/sn_intrinsic.h b/StgSharp.Native/sn-head/sn_intrinsic.h
index 9ff16f3b..5071eac5 100644
--- a/StgSharp.Native/sn-head/sn_intrinsic.h
+++ b/StgSharp.Native/sn-head/sn_intrinsic.h
@@ -85,7 +85,6 @@ INTERNAL void SN_DECL f32_scalar_mul_avx(MAT_KERNEL(float) const *matrix, float
                                          MAT_KERNEL(float) *restrict ans);
 INTERNAL void SN_DECL f32_scalar_mul_512(MAT_KERNEL(float) const *matrix, float const scalar,
                                          MAT_KERNEL(float) *restrict ans);
-
 #pragma endregion
 
 #pragma region fma
@@ -104,9 +103,12 @@ INTERNAL uint64_t SN_DECL factorial_simd_sse(int n);
 
 INTERNAL void FORCEINLINE SN_DECL dot_41_sse(MAT_KERNEL(float) const *transpose,
                                              __m128 const *vector, __m128 *ans);
-INTERNAL void SN_DECL dot_41_avx(MAT_KERNEL(float) const *transpose, __m128 const *vector, __m128 *ans);
-INTERNAL void SN_DECL dot_42_sse(MAT_KERNEL(float) const *transpose, __m128 const *vector, __m128 *ans);
-INTERNAL void SN_DECL dot_42_avx(MAT_KERNEL(float) const *transpose, __m128 const *vector, __m128 *ans);
+INTERNAL void SN_DECL dot_41_avx(MAT_KERNEL(float) const *transpose, __m128 const *vector,
+                                 __m128 *ans);
+INTERNAL void SN_DECL dot_42_sse(MAT_KERNEL(float) const *transpose, __m128 const *vector,
+                                 __m128 *ans);
+INTERNAL void SN_DECL dot_42_avx(MAT_KERNEL(float) const *transpose, __m128 const *vector,
+                                 __m128 *ans);
 
 #pragma endregion
 
diff --git a/StgSharp.Native/sn-head/sn_matkernel.h b/StgSharp.Native/sn-head/sn_matkernel.h
index 8f7930b2..0bdafa7d 100644
--- a/StgSharp.Native/sn-head/sn_matkernel.h
+++ b/StgSharp.Native/sn-head/sn_matkernel.h
@@ -4,16 +4,16 @@
 #include <immintrin.h>
 
 #define ELEMENTSIZE(T) sizeof(T)
-#define XMMCOUNT(T) sizeof(T) / 16
-#define YMMCOUNT(T) sizeof(T) / 32
-#define ZMMCOUNT(T) sizeof(T) / 64
+#define XMMCOUNT(T) sizeof(T)
+#define YMMCOUNT(T) sizeof(T) / 2
+#define ZMMCOUNT(T) sizeof(T) / 4
 
 #define MAT_KERNEL(T)                    \
         union {                          \
                 __m128 xmm[XMMCOUNT(T)]; \
                 __m256 ymm[YMMCOUNT(T)]; \
                 __m512 zmm[ZMMCOUNT(T)]; \
-                T m[16];                 \
+                T m[4][4];               \
         }
 
 #endif