diff --git a/lite/kernels/arm/cast_compute.cc b/lite/kernels/arm/cast_compute.cc
index 266ae1fc916af4303aca274c39b9b4923fdbb154..3b3ef07e105c583b7e3eb8b64b14610ca0f9e41a 100644
--- a/lite/kernels/arm/cast_compute.cc
+++ b/lite/kernels/arm/cast_compute.cc
@@ -73,7 +73,7 @@ void CastCompute::Run() {
 }  // namespace paddle
 
 REGISTER_LITE_KERNEL(
-    cast, kARM, kFloat, kNCHW, paddle::lite::kernels::arm::CastCompute, def)
+    cast, kARM, kAny, kNCHW, paddle::lite::kernels::arm::CastCompute, def)
     .BindInput("X", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kAny))})
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kARM), PRECISION(kAny))})
     .Finalize();
diff --git a/lite/kernels/arm/cast_compute.h b/lite/kernels/arm/cast_compute.h
index d342a405ad593b8457b2899fa3ee6ae843d8f792..1f8da056a8be61de20b5d6e98e455e850b9c9f8d 100644
--- a/lite/kernels/arm/cast_compute.h
+++ b/lite/kernels/arm/cast_compute.h
@@ -23,7 +23,7 @@ namespace lite {
 namespace kernels {
 namespace arm {
 
-class CastCompute : public KernelLite<TARGET(kARM), PRECISION(kFloat)> {
+class CastCompute : public KernelLite<TARGET(kARM), PRECISION(kAny)> {
  public:
   using param_t = operators::CastParam;
 
diff --git a/lite/kernels/arm/fill_constant_batch_size_like_compute.cc b/lite/kernels/arm/fill_constant_batch_size_like_compute.cc
index c7944f007e69763bc6c1902925e89b004900b122..7bb096e0223ee2c0b43edb9b6af710032008e70c 100644
--- a/lite/kernels/arm/fill_constant_batch_size_like_compute.cc
+++ b/lite/kernels/arm/fill_constant_batch_size_like_compute.cc
@@ -52,7 +52,7 @@ void FillConstantBatchSizeLikeCompute::Run() {
 REGISTER_LITE_KERNEL(
     fill_constant_batch_size_like,
     kARM,
-    kFloat,
+    kAny,
     kNCHW,
     paddle::lite::kernels::arm::FillConstantBatchSizeLikeCompute,
     def)
diff --git a/lite/kernels/arm/fill_constant_batch_size_like_compute.h b/lite/kernels/arm/fill_constant_batch_size_like_compute.h
index 2107a943f02b7fad0d2207cc6a6d9d49518a39ee..23aa64bb6417ae1ed0b551520096cf6401ec702c 100644
--- a/lite/kernels/arm/fill_constant_batch_size_like_compute.h
+++ b/lite/kernels/arm/fill_constant_batch_size_like_compute.h
@@ -22,7 +22,7 @@ namespace kernels {
 namespace arm {
 
 class FillConstantBatchSizeLikeCompute
-    : public KernelLite<TARGET(kARM), PRECISION(kFloat)> {
+    : public KernelLite<TARGET(kARM), PRECISION(kAny)> {
  public:
   using param_t = operators::FillConstantBatchSizeLikeParam;
 
diff --git a/lite/kernels/arm/fill_constant_compute.cc b/lite/kernels/arm/fill_constant_compute.cc
index a17aa33e039edbd14eb350538abdbefa12735f71..50e4f041022fa5fb946ca183fabfb079c9f21df5 100644
--- a/lite/kernels/arm/fill_constant_compute.cc
+++ b/lite/kernels/arm/fill_constant_compute.cc
@@ -52,7 +52,7 @@ void FillConstantCompute::Run() {
 // float
 REGISTER_LITE_KERNEL(fill_constant,
                      kARM,
-                     kFloat,
+                     kAny,
                      kNCHW,
                      paddle::lite::kernels::arm::FillConstantCompute,
                      def)
diff --git a/lite/kernels/arm/fill_constant_compute.h b/lite/kernels/arm/fill_constant_compute.h
index 5d5f7078a0d4469f43fe0be8aede32fe11f3c568..7717c4c2628cff5358cc2011c01cb4b02ee125dc 100644
--- a/lite/kernels/arm/fill_constant_compute.h
+++ b/lite/kernels/arm/fill_constant_compute.h
@@ -21,7 +21,7 @@ namespace lite {
 namespace kernels {
 namespace arm {
 
-class FillConstantCompute : public KernelLite<TARGET(kARM), PRECISION(kFloat)> {
+class FillConstantCompute : public KernelLite<TARGET(kARM), PRECISION(kAny)> {
  public:
   using param_t = operators::FillConstantParam;
 
diff --git a/lite/tests/kernels/cast_compute_test.cc b/lite/tests/kernels/cast_compute_test.cc
index a7316a6162ed9a1bbbaf4956d51ab19c017fd3e4..e3b27ce627843a2e66ba407cfc7f7e986d4a840e 100644
--- a/lite/tests/kernels/cast_compute_test.cc
+++ b/lite/tests/kernels/cast_compute_test.cc
@@ -163,7 +163,7 @@ TEST(Cast, precision) {
   Place place;
   float abs_error = 2e-5;
 #if defined(LITE_WITH_ARM)
-  place = TARGET(kARM);
+  place = {TARGET(kARM), PRECISION(kAny)};
 #elif defined(LITE_WITH_XPU)
   place = TARGET(kXPU);
 #else
diff --git a/lite/tests/kernels/fill_constant_batch_size_like_compute_test.cc b/lite/tests/kernels/fill_constant_batch_size_like_compute_test.cc
index 03cc99da82b1d84b5a1149be2d0a656913bf0c50..e225d81ad05db1080738979ce404350ca0507da7 100644
--- a/lite/tests/kernels/fill_constant_batch_size_like_compute_test.cc
+++ b/lite/tests/kernels/fill_constant_batch_size_like_compute_test.cc
@@ -138,7 +138,7 @@ TEST(fill_constant_batch_size_like, precision) {
   place = TARGET(kNPU);
   abs_error = 1e-2;  // use fp16 in npu
 #elif defined(LITE_WITH_ARM)
-  place = TARGET(kARM);
+  place = {TARGET(kARM), PRECISION(kAny)};
 #else
   return;
 #endif
diff --git a/lite/tests/kernels/fill_constant_compute_test.cc b/lite/tests/kernels/fill_constant_compute_test.cc
index c9ae1ebaa57aeafe59cf9aa46533aff45884e777..ef5a4fa7785d4de410fc663d98d47c0ec028feca 100644
--- a/lite/tests/kernels/fill_constant_compute_test.cc
+++ b/lite/tests/kernels/fill_constant_compute_test.cc
@@ -175,7 +175,7 @@ TEST(fill_constant, precision) {
   place = TARGET(kNPU);
   abs_error = 1e-2;  // use fp16 in npu
 #elif defined(LITE_WITH_ARM)
-  place = TARGET(kARM);
+  place = {TARGET(kARM), PRECISION(kAny)};
 #else
   return;
 #endif