MKL elementwise add backward: Initial implementation with vector copy

paddle/fluid/operators/elementwise_add_op.cc

@@ -25,6 +25,6 @@ REGISTER_OP_CPU_KERNEL(
 REGISTER_OP_CPU_KERNEL(
     elementwise_add_grad,
     ops::ElementwiseAddGradKernel<paddle::platform::CPUDeviceContext, float>,
-    ops::ElementwiseAddGradKernel<paddle::platform::CPUDeviceContext, double>,
-    ops::ElementwiseAddGradKernel<paddle::platform::CPUDeviceContext, int>,
-    ops::ElementwiseAddGradKernel<paddle::platform::CPUDeviceContext, int64_t>);
+    ops::ElementwiseAddGradKernel<paddle::platform::CPUDeviceContext, double>);
+//    ops::ElementwiseAddGradKernel<paddle::platform::CPUDeviceContext, int>,
+//    ops::ElementwiseAddGradKernel<paddle::platform::CPUDeviceContext, int64_t>);

paddle/fluid/operators/elementwise_add_op.h

@@ -98,9 +98,22 @@ class ElementwiseAddGradKernel : public framework::OpKernel<T> {
     auto* dx = ctx.Output<Tensor>(framework::GradVarName("X"));
     auto* dy = ctx.Output<Tensor>(framework::GradVarName("Y"));
     int axis = ctx.Attr<int>("axis");
-    ElemwiseGradCompute<DeviceContext, T, IdentityGrad<T>, IdentityGrad<T>>(
-        ctx, *x, *y, *out, *dout, axis, dx, dy, IdentityGrad<T>(),
-        IdentityGrad<T>());
+
+    if (platform::is_cpu_place(ctx.GetPlace()) && (x->dims() == y->dims())) {
+      auto blas = math::GetBlas<DeviceContext, T>(ctx);
+
+      if (dx) 
+        dx->mutable_data<T>(ctx.GetPlace());
+      if (dy)
+        dy->mutable_data<T>(ctx.GetPlace());
+
+      blas.VCOPY(dout->numel(), dout->data<T>(), dx->data<T>());
+      blas.VCOPY(dout->numel(), dout->data<T>(), dy->data<T>());
+    } else {
+      ElemwiseGradCompute<DeviceContext, T, IdentityGrad<T>, IdentityGrad<T>>(
+          ctx, *x, *y, *out, *dout, axis, dx, dy, IdentityGrad<T>(),
+          IdentityGrad<T>());
+    }
   }
 };