Optimize the functions.

Optimize the MultiplyMe and DivMe functions when operate the scalar tensors.

source/tensor/core/arithmetic/Div.cpp

@@ -157,16 +157,9 @@ where i is the index of the item
 void DivMe(XTensor & a, const XTensor & b, DTYPE alpha, int leadingDim)
 {
     if (b.order == 0){
-        DTYPE scale = 1.0F / b.Get0D();
-        XTensor * tmp1 = NewTensorBufV2(&a, a.devID, a.mem);
-        XTensor * tmp2 = NewTensorBufV2(&a, a.devID, a.mem);
-
-        _ScaleAndShift(&a, tmp1, scale, 0.0F);
-        _ScaleAndShift(&a, tmp2, alpha, 0.0F);
-        _Sum(tmp2, tmp1, &a);
+        DTYPE scale = 1.0F / b.Get0D() + alpha;
 
-        DelTensorBuf(tmp1);
-        DelTensorBuf(tmp2);
+        _ScaleAndShift(&a, &a, scale, 0.0F);
     }
     else {
         int n = GetBroadcastDimIndex(a, b);

source/tensor/core/arithmetic/Multiply.cpp

@@ -158,16 +158,9 @@ where i is the index of the item
 void MultiplyMe(XTensor& a, const XTensor& b, DTYPE alpha, int leadingDim)
 {
     if (b.order == 0){
-        DTYPE scale = b.Get0D();
-        XTensor * tmp1 = NewTensorBufV2(&a, a.devID, a.mem);
-        XTensor * tmp2 = NewTensorBufV2(&a, a.devID, a.mem);
-
-        _ScaleAndShift(&a, tmp1, scale, 0.0F);
-        _ScaleAndShift(&a, tmp2, alpha, 0.0F);
-        _Sum(tmp2, tmp1, &a);
+        DTYPE scale = b.Get0D() + alpha;
 
-        DelTensorBuf(tmp1);
-        DelTensorBuf(tmp2);
+        _ScaleAndShift(&a, &a, scale, 0.0F);
     }
     else {
         int n = GetBroadcastDimIndex(a, b);