[arm][kernel]fix: remove unused kernel to reduce build size for CI. test=develop

lite/backends/arm/math/elementwise.cc

@@ -59,15 +59,6 @@ void elementwise_add<int32_t>(const int32_t* dinx,
   naive_elementwise_op<int32_t>(dinx, diny, dout, num, naive_add<int32_t>);
 }
 
-// todo: use arm intrinsics
-template <>
-void elementwise_add<int64_t>(const int64_t* dinx,
-                              const int64_t* diny,
-                              int64_t* dout,
-                              int num) {
-  naive_elementwise_op<int64_t>(dinx, diny, dout, num, naive_add<int64_t>);
-}
-
 template <>
 void elementwise_add<float>(const float* dinx,
                             const float* diny,
@@ -261,18 +252,6 @@ void elementwise_add_broadcast<int32_t>(const int32_t* dinx,
       dinx, diny, dout, batch, channels, num, naive_add<int32_t>);
 }
 
-// todo: use arm intrinsics
-template <>
-void elementwise_add_broadcast<int64_t>(const int64_t* dinx,
-                                        const int64_t* diny,
-                                        int64_t* dout,
-                                        int batch,
-                                        int channels,
-                                        int num) {
-  naive_elementwise_op_broadcast<int64_t>(
-      dinx, diny, dout, batch, channels, num, naive_add<int64_t>);
-}
-
 template <>
 void elementwise_add_broadcast<float>(const float* dinx,
                                       const float* diny,
@@ -494,15 +473,6 @@ void elementwise_sub<int32_t>(const int32_t* dinx,
   naive_elementwise_op<int32_t>(dinx, diny, dout, num, naive_sub<int32_t>);
 }
 
-// todo: use arm intrinsics
-template <>
-void elementwise_sub<int64_t>(const int64_t* dinx,
-                              const int64_t* diny,
-                              int64_t* dout,
-                              int num) {
-  naive_elementwise_op<int64_t>(dinx, diny, dout, num, naive_sub<int64_t>);
-}
-
 template <>
 void elementwise_sub<float>(const float* dinx,
                             const float* diny,
@@ -615,18 +585,6 @@ void elementwise_sub_broadcast<int32_t>(const int32_t* dinx,
       dinx, diny, dout, batch, channels, num, naive_sub<int32_t>);
 }
 
-// todo: use arm intrinsics
-template <>
-void elementwise_sub_broadcast<int64_t>(const int64_t* dinx,
-                                        const int64_t* diny,
-                                        int64_t* dout,
-                                        int batch,
-                                        int channels,
-                                        int num) {
-  naive_elementwise_op_broadcast<int64_t>(
-      dinx, diny, dout, batch, channels, num, naive_sub<int64_t>);
-}
-
 template <>
 void elementwise_sub_broadcast<float>(const float* dinx,
                                       const float* diny,

lite/kernels/arm/elementwise_compute.cc

@@ -398,16 +398,6 @@ REGISTER_LITE_KERNEL(
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
     .Finalize();
 
-using elementwise_add_int64_t =
-    paddle::lite::kernels::arm::ElementwiseAddCompute<int64_t,
-                                                      PRECISION(kInt64)>;
-REGISTER_LITE_KERNEL(
-    elementwise_add, kARM, kInt64, kNCHW, elementwise_add_int64_t, def)
-    .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .BindInput("Y", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .Finalize();
-
 REGISTER_LITE_KERNEL(
     fusion_elementwise_add_activation,
     kARM,
@@ -439,16 +429,6 @@ REGISTER_LITE_KERNEL(
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
     .Finalize();
 
-using elementwise_sub_int64_t =
-    paddle::lite::kernels::arm::ElementwiseSubCompute<int64_t,
-                                                      PRECISION(kInt64)>;
-REGISTER_LITE_KERNEL(
-    elementwise_sub, kARM, kInt64, kNCHW, elementwise_sub_int64_t, def)
-    .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .BindInput("Y", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .Finalize();
-
 REGISTER_LITE_KERNEL(
     fusion_elementwise_sub_activation,
     kARM,
@@ -479,16 +459,6 @@ REGISTER_LITE_KERNEL(
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
     .Finalize();
 
-using elementwise_mul_int64_t =
-    paddle::lite::kernels::arm::ElementwiseMulCompute<int64_t,
-                                                      PRECISION(kInt64)>;
-REGISTER_LITE_KERNEL(
-    elementwise_mul, kARM, kInt64, kNCHW, elementwise_mul_int64_t, def)
-    .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .BindInput("Y", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .Finalize();
-
 using fusion_elementwise_mul_activation_float_t = paddle::lite::kernels::arm::
     ElementwiseMulActivationCompute<float, PRECISION(kFloat)>;
 REGISTER_LITE_KERNEL(fusion_elementwise_mul_activation,
@@ -558,17 +528,6 @@ REGISTER_LITE_KERNEL(
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt32))})
     .Finalize();
 
-using elementwise_div_int64_t =
-    paddle::lite::kernels::arm::ElementwiseDivCompute<int64_t,
-                                                      PRECISION(kInt64)>;
-
-REGISTER_LITE_KERNEL(
-    elementwise_div, kARM, kInt64, kNCHW, elementwise_div_int64_t, def)
-    .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .BindInput("Y", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kInt64))})
-    .Finalize();
-
 REGISTER_LITE_KERNEL(
     fusion_elementwise_div_activation,
     kARM,