change TensorCopy to ShareDataWith in matmul_grad op (#33755)

Author: pangyoki

paddle/fluid/operators/matmul_v2_op_npu.cc

@@ -141,17 +141,13 @@ class MatMulV2GradNPUKernel : public framework::OpKernel<T> {
           if ((x->dims().size() == 3) && (dout->dims().size() == 3) &&
               (dy->dims().size() == 2)) {
             framework::Tensor dout_;
-            TensorCopy(*dout, ctx.GetPlace(), &dout_);
-            ctx.template device_context<paddle::platform::NPUDeviceContext>()
-                .Wait();
+            dout_.ShareDataWith(*dout);
             std::vector<int> vec_dim = framework::vectorize<int>(dout_.dims());
             std::vector<int> vec_dim_v{vec_dim[0] * vec_dim[1], vec_dim[2]};
             dout_.Resize(framework::make_ddim(vec_dim_v));
 
             framework::Tensor x_;
-            TensorCopy(*x, ctx.GetPlace(), &x_);
-            ctx.template device_context<paddle::platform::NPUDeviceContext>()
-                .Wait();
+            x_.ShareDataWith(*x);
             std::vector<int> vec_dim_x = framework::vectorize<int>(x_.dims());
             std::vector<int> vec_dim_x_v{vec_dim_x[0] * vec_dim_x[1],
                                          vec_dim_x[2]};