diff --git a/paddle/phi/kernels/funcs/elementwise_functor.h b/paddle/phi/kernels/funcs/elementwise_functor.h
index 508240a99164480069de0b74995803cb8c6dcebf..84938597e82b32e4f1967f618d8b217c75c079c2 100644
--- a/paddle/phi/kernels/funcs/elementwise_functor.h
+++ b/paddle/phi/kernels/funcs/elementwise_functor.h
@@ -117,7 +117,7 @@ struct DivGradXYFunctor {
     // dy = - dout * out / y
     phi::Array<OutT, 2> outs;
     outs[0] = a / c;
-    outs[1] = -a * b / c;
+    outs[1] = -a * ((b / c) / c);
     return outs;
   }
 };
@@ -130,7 +130,7 @@ struct DivGradXYFunctor<ComplexType<InT>, ComplexType<OutT>> {
       const ComplexType<InT> c) {
     phi::Array<ComplexType<OutT>, 2> outs;
     ComplexType<InT> c_conj(c.real, -c.imag);
-    ComplexType<InT> out_div_c_conj((b / c).real, -(b / c).imag);
+    ComplexType<InT> out_div_c_conj(((b / c) / c).real, -((b / c) / c).imag);
     outs[0] = a / c_conj;
     outs[1] = -a * out_div_c_conj;
     return outs;
@@ -157,7 +157,7 @@ struct DivGradXFunctor<ComplexType<T>> {
 template <typename T>
 struct DivGradYFunctor {
   inline HOSTDEVICE T operator()(const T a, const T b, const T c) const {
-    return -a * b / c;
+    return -a * ((b / c) / c);
   }
 };
 
@@ -167,7 +167,7 @@ struct DivGradYFunctor<ComplexType<T>> {
   inline HOSTDEVICE ComplexType<T> operator()(const ComplexType<T> a,
                                               const ComplexType<T> b,
                                               const ComplexType<T> c) const {
-    ComplexType<T> out_div_c_conj((b / c).real, -(b / c).imag);
+    ComplexType<T> out_div_c_conj(((b / c) / c).real, -((b / c) / c).imag);
     return -a * out_div_c_conj;
   }
 };
diff --git a/paddle/phi/kernels/gpu/elementwise_divide_grad_kernel.cu b/paddle/phi/kernels/gpu/elementwise_divide_grad_kernel.cu
index 57bf6da4060d34acdbf44ac80d4577e24247417d..d63841f17e62ea0ea25e5888b8db9c164f4dab49 100644
--- a/paddle/phi/kernels/gpu/elementwise_divide_grad_kernel.cu
+++ b/paddle/phi/kernels/gpu/elementwise_divide_grad_kernel.cu
@@ -36,7 +36,7 @@ void DivideGradKernel(const Context& dev_ctx,
                       DenseTensor* dy) {
   const auto place = dev_ctx.GetPlace();
   if (dx != nullptr && dy != nullptr) {
-    std::vector<const DenseTensor*> ins = {&dout, &out, &y};
+    std::vector<const DenseTensor*> ins = {&dout, &x, &y};
     GetGradXAndYOut<ElementwiseType::kTernary, T>(
         dev_ctx,
         place,
@@ -51,7 +51,7 @@ void DivideGradKernel(const Context& dev_ctx,
     GetGradXOrYOut<ElementwiseType::kBinary, T>(
         dev_ctx, place, axis, ins, dout, dx, funcs::DivGradXFunctor<T>());
   } else if (dy != nullptr && dx == nullptr) {
-    std::vector<const DenseTensor*> ins = {&dout, &out, &y};
+    std::vector<const DenseTensor*> ins = {&dout, &x, &y};
     GetGradXOrYOut<ElementwiseType::kTernary, T>(
         dev_ctx, place, axis, ins, dout, dy, funcs::DivGradYFunctor<T>());
   }