add int and int64 dtype for gather_op (#14175)

test=develop

paddle/fluid/operators/gather_op.cc

@@ -102,7 +102,9 @@ REGISTER_OPERATOR(gather, ops::GatherOp, ops::GatherOpMaker,
                   paddle::framework::DefaultGradOpDescMaker<true>);
 REGISTER_OPERATOR(gather_grad, ops::GatherGradOp);
 REGISTER_OP_CPU_KERNEL(gather, ops::GatherOpKernel<float>,
-                       ops::GatherOpKernel<int>, ops::GatherOpKernel<double>);
+                       ops::GatherOpKernel<double>, ops::GatherOpKernel<int>,
+                       ops::GatherOpKernel<int64_t>);
 REGISTER_OP_CPU_KERNEL(gather_grad, ops::GatherGradientOpKernel<float>,
+                       ops::GatherGradientOpKernel<double>,
                        ops::GatherGradientOpKernel<int>,
-                       ops::GatherGradientOpKernel<double>);
+                       ops::GatherGradientOpKernel<int64_t>);

paddle/fluid/operators/gather_op.cu

@@ -61,5 +61,11 @@ class GatherGradOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OP_CUDA_KERNEL(gather, ops::GatherOpCUDAKernel<float>);
-REGISTER_OP_CUDA_KERNEL(gather_grad, ops::GatherGradOpCUDAKernel<float>);
+REGISTER_OP_CUDA_KERNEL(gather, ops::GatherOpCUDAKernel<float>,
+                        ops::GatherOpCUDAKernel<double>,
+                        ops::GatherOpCUDAKernel<int64_t>,
+                        ops::GatherOpCUDAKernel<int>);
+REGISTER_OP_CUDA_KERNEL(gather_grad, ops::GatherGradOpCUDAKernel<float>,
+                        ops::GatherGradOpCUDAKernel<double>,
+                        ops::GatherGradOpCUDAKernel<int64_t>,
+                        ops::GatherGradOpCUDAKernel<int>);