diff --git a/StgSharp.Mathematics/Numeric/MatrixCompute.Basic.cs b/StgSharp.Mathematics/Numeric/MatrixCompute.Basic.cs
index ffa2731e..c70b7711 100644
--- a/StgSharp.Mathematics/Numeric/MatrixCompute.Basic.cs
+++ b/StgSharp.Mathematics/Numeric/MatrixCompute.Basic.cs
@@ -110,9 +110,93 @@ namespace StgSharp.Mathematics.Numeric
             package->Scalar = scalar;
             MatrixParallelHandle taskGroup = MatrixParallel.ScheduleTask(package);
             MatrixParallel.LaunchParallel(taskGroup, SleepMode.DeepSleep);
-
+            MatrixParallelFactory.Release(scalar);
             taskGroup.Dispose();
         }
 
+        public static unsafe Matrix<float> Mul(this Matrix<float> left, float right)
+        {
+            long count = (long)left.KernelColumnLength * left.KernelRowLength;
+            Matrix<float> ans = Matrix.FromDefault<float>(left.ColumnLength, left.RowLength);
+            ScalarPacket* scalar = MatrixParallelFactory.CreateScalarPacket();
+            scalar->Data<float>(0) = right;
+            if (count <= 256)
+            {
+                MatrixKernel<float>* leftPtr = left.Buffer;
+                MatrixKernel<float>* ansPtr = ans.Buffer;
+                NativeIntrinsic.Intrinsic.f32_scalar_mul(leftPtr, ansPtr, scalar, count);
+
+                return ans;
+            } else if (count <= 1024)
+            {
+                MatrixParallelThread[] pool = MatrixParallel.LeadParallel(1);
+
+                MatrixKernel<float>* leftPtr = left.Buffer;
+                MatrixKernel<float>* ansPtr = ans.Buffer;
+                NativeIntrinsic.Intrinsic.f32_scalar_mul(leftPtr, ansPtr, scalar, count);
+
+                MatrixParallel.ReturnParallelResources(ref pool);
+                return ans;
+            }
+            MatrixParallelTaskPackage* package = MatrixParallelFactory.CreateBaseTask<float>();
+            package->Right = (nint)left.Buffer;
+            package->Result = (nint)ans.Buffer;
+            package->ElementSize = sizeof(float);
+            package->PrimCount = left.KernelColumnLength;
+            package->SecCount = left.KernelRowLength;
+            package->Scalar = scalar;
+            package->ComputeMode = new MatrixOpMode(MatrixOperationStyle.BUFFER_OP, MatrixOperationParam.RIGHT_ANS_SCALAR_PARAM);
+            package->ComputeHandle = (IntPtr)NativeIntrinsic.Intrinsic.f32_scalar_mul;
+            MatrixParallelHandle taskGroup = MatrixParallel.ScheduleTask(package);
+
+            MatrixParallel.LaunchParallel(taskGroup, SleepMode.DeepSleep);
+            taskGroup.Dispose();
+            MatrixParallelFactory.Release(scalar);
+            return ans;
+        }
+
+        public static unsafe Matrix<float> Sub(this Matrix<float> left, Matrix<float> right)
+        {
+            if (left.ColumnLength != right.ColumnLength || left.RowLength != right.RowLength) {
+                throw new IndexOutOfRangeException("Matrix dimensions must agree.");
+            }
+            long count = (long)left.KernelColumnLength * left.KernelRowLength;
+            Matrix<float> ans = Matrix.FromDefault<float>(left.ColumnLength, left.RowLength);
+            if (count <= 256)
+            {
+                MatrixKernel<float>* leftPtr = left.Buffer;
+                MatrixKernel<float>* rightPtr = right.Buffer;
+                MatrixKernel<float>* ansPtr = ans.Buffer;
+                NativeIntrinsic.Intrinsic.f32_sub(leftPtr, rightPtr, ansPtr, count);
+
+                return ans;
+            } else if (count <= 1024)
+            {
+                MatrixParallelThread[] pool = MatrixParallel.LeadParallel(1);
+
+                MatrixKernel<float>* leftPtr = left.Buffer;
+                MatrixKernel<float>* rightPtr = right.Buffer;
+                MatrixKernel<float>* ansPtr = ans.Buffer;
+                NativeIntrinsic.Intrinsic.f32_sub(leftPtr, rightPtr, ansPtr, count);
+
+                MatrixParallel.ReturnParallelResources(ref pool);
+                return ans;
+            }
+            MatrixParallelTaskPackage* package = MatrixParallelFactory.CreateBaseTask<float>();
+            package->Left = (nint)left.Buffer;
+            package->Right = (nint)right.Buffer;
+            package->Result = (nint)ans.Buffer;
+            package->ElementSize = sizeof(float);
+            package->PrimCount = left.KernelColumnLength;
+            package->SecCount = left.KernelRowLength;
+            package->ComputeMode = new MatrixOpMode(MatrixOperationStyle.BUFFER_OP, MatrixOperationParam.LEFT_RIGHT_ANS_PARAM);
+            package->ComputeHandle = (IntPtr)NativeIntrinsic.Intrinsic.f32_sub;
+            MatrixParallelHandle taskGroup = MatrixParallel.ScheduleTask(package);
+
+            MatrixParallel.LaunchParallel(taskGroup, SleepMode.DeepSleep);
+            taskGroup.Dispose();
+            return ans;
+        }
+
     }
 }
diff --git a/StgSharp.Native/sn-src/k_bufop.c b/StgSharp.Native/sn-src/k_bufop.c
index 62a881b4..ce3d29ad 100644
--- a/StgSharp.Native/sn-src/k_bufop.c
+++ b/StgSharp.Native/sn-src/k_bufop.c
@@ -97,12 +97,12 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, sub)
 
 DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, sse, scalar_mul)
 {
-        register __m128 scalar_vec = ((__m128 *)(scalar->data))[0];
+        register __m128 vec = _mm_set_ps1(((float *)(scalar->data))[0]);
         for (size_t i = 0; i < count; i++) {
-                register __m128 c0 = _mm_mul_ps(right[i].xmm[0], scalar_vec);
-                register __m128 c1 = _mm_mul_ps(right[i].xmm[1], scalar_vec);
-                register __m128 c2 = _mm_mul_ps(right[i].xmm[2], scalar_vec);
-                register __m128 c3 = _mm_mul_ps(right[i].xmm[3], scalar_vec);
+                register __m128 c0 = _mm_mul_ps(right[i].xmm[0], vec);
+                register __m128 c1 = _mm_mul_ps(right[i].xmm[1], vec);
+                register __m128 c2 = _mm_mul_ps(right[i].xmm[2], vec);
+                register __m128 c3 = _mm_mul_ps(right[i].xmm[3], vec);
 
                 ans[i].xmm[0] = c0;
                 ans[i].xmm[1] = c1;
@@ -113,10 +113,10 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, sse, scalar_mul)
 
 DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, avx, scalar_mul)
 {
-        register __m256 scalar_vec = ((__m256 *)(scalar->data))[0];
+        register __m256 vec = _mm256_broadcast_ss(((float *)(scalar->data)));
         for (size_t i = 0; i < count; i++) {
-                register __m256 c0 = _mm256_mul_ps(right[i].ymm[0], scalar_vec);
-                register __m256 c1 = _mm256_mul_ps(right[i].ymm[1], scalar_vec);
+                register __m256 c0 = _mm256_mul_ps(right[i].ymm[0], vec);
+                register __m256 c1 = _mm256_mul_ps(right[i].ymm[1], vec);
 
                 ans[i].ymm[0] = c0;
                 ans[i].ymm[1] = c1;
@@ -125,9 +125,9 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, avx, scalar_mul)
 
 DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, 512, scalar_mul)
 {
-        register __m512 scalar_vec = ((__m512 *)(scalar->data))[0];
+        register __m512 vec = _mm512_broadcastss_ps(((__m128 *)(scalar->data))[0]);
         for (size_t i = 0; i < count; i++) {
-                register __m512 c0 = _mm512_mul_ps(right[i].zmm[0], scalar_vec);
+                register __m512 c0 = _mm512_mul_ps(right[i].zmm[0], vec);
 
                 ans[i].zmm[0] = c0;
         }
diff --git a/StgSharpDebug/Program.cs b/StgSharpDebug/Program.cs
index 0da9b457..38534e38 100644
--- a/StgSharpDebug/Program.cs
+++ b/StgSharpDebug/Program.cs
@@ -85,6 +85,7 @@ namespace StgSharp
                     }
                     GC.Collect();
                 }
+                n = n.Mul(12);
                 for (int i = 0; i < 20; i++)
                 {
                     for (int j = 0; j < 20; j++) {