diff --git a/tensorflow/core/kernels/cwise_op_abs.cc b/tensorflow/core/kernels/cwise_op_abs.cc
index fb22293d4a2..ff9db97fe9d 100644
--- a/tensorflow/core/kernels/cwise_op_abs.cc
+++ b/tensorflow/core/kernels/cwise_op_abs.cc
@@ -16,14 +16,14 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER5(UnaryOp, CPU, "Abs", functor::abs, Eigen::half, float, double, int32,
+REGISTER5(UnaryOp, CPU, "Abs", functor::abs, float, Eigen::half, double, int32,
           int64);
 #if !defined(__ANDROID__)
 REGISTER_KERNEL_BUILDER(Name("ComplexAbs").Device(DEVICE_CPU),
                         UnaryOp<CPUDevice, functor::abs<complex64>>);
 #endif
 #if GOOGLE_CUDA
-REGISTER4(UnaryOp, GPU, "Abs", functor::abs, Eigen::half, float, double, int64);
+REGISTER4(UnaryOp, GPU, "Abs", functor::abs, float, Eigen::half, double, int64);
 
 // A special GPU kernel for int32.
 // TODO(b/25387198): Also enable int32 in device memory. This kernel
diff --git a/tensorflow/core/kernels/cwise_op_add.cc b/tensorflow/core/kernels/cwise_op_add.cc
index 638f0d9a75d..1457b74e6fd 100644
--- a/tensorflow/core/kernels/cwise_op_add.cc
+++ b/tensorflow/core/kernels/cwise_op_add.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER9(BinaryOp, CPU, "Add", functor::add, Eigen::half, float, double, int32,
+REGISTER9(BinaryOp, CPU, "Add", functor::add, float, Eigen::half, double, int32,
           int64, int8, int16, complex64, string);
 #if GOOGLE_CUDA
-REGISTER4(BinaryOp, GPU, "Add", functor::add, Eigen::half, float, double,
+REGISTER4(BinaryOp, GPU, "Add", functor::add, float, Eigen::half, double,
           int64);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_ceil.cc b/tensorflow/core/kernels/cwise_op_ceil.cc
index 71841aed50b..c68395437d4 100644
--- a/tensorflow/core/kernels/cwise_op_ceil.cc
+++ b/tensorflow/core/kernels/cwise_op_ceil.cc
@@ -16,8 +16,8 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "Ceil", functor::ceil, Eigen::half, float, double);
+REGISTER3(UnaryOp, CPU, "Ceil", functor::ceil, float, Eigen::half, double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Ceil", functor::ceil, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Ceil", functor::ceil, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_cos.cc b/tensorflow/core/kernels/cwise_op_cos.cc
index c5835803c8a..6958fa22b89 100644
--- a/tensorflow/core/kernels/cwise_op_cos.cc
+++ b/tensorflow/core/kernels/cwise_op_cos.cc
@@ -16,9 +16,9 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Cos", functor::cos, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Cos", functor::cos, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Cos", functor::cos, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Cos", functor::cos, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_digamma.cc b/tensorflow/core/kernels/cwise_op_digamma.cc
index 07ed7f0ad78..415dbc952f3 100644
--- a/tensorflow/core/kernels/cwise_op_digamma.cc
+++ b/tensorflow/core/kernels/cwise_op_digamma.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "Digamma", functor::digamma, Eigen::half, float,
+REGISTER3(UnaryOp, CPU, "Digamma", functor::digamma, float, Eigen::half,
           double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Digamma", functor::digamma, Eigen::half, float,
+REGISTER3(UnaryOp, GPU, "Digamma", functor::digamma, float, Eigen::half,
           double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_div.cc b/tensorflow/core/kernels/cwise_op_div.cc
index f5e992d9466..0af0fc21e58 100644
--- a/tensorflow/core/kernels/cwise_op_div.cc
+++ b/tensorflow/core/kernels/cwise_op_div.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER8(BinaryOp, CPU, "Div", functor::div, Eigen::half, float, double, uint8,
+REGISTER8(BinaryOp, CPU, "Div", functor::div, float, Eigen::half, double, uint8,
           int16, int32, int64, complex64);
 #if GOOGLE_CUDA
-REGISTER6(BinaryOp, GPU, "Div", functor::div, Eigen::half, float, double, uint8,
+REGISTER6(BinaryOp, GPU, "Div", functor::div, float, Eigen::half, double, uint8,
           int16, int64);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_equal_to.cc b/tensorflow/core/kernels/cwise_op_equal_to.cc
index 79dc9edf07e..03053733c14 100644
--- a/tensorflow/core/kernels/cwise_op_equal_to.cc
+++ b/tensorflow/core/kernels/cwise_op_equal_to.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER10(BinaryOp, CPU, "Equal", functor::equal_to, Eigen::half, float,
+REGISTER10(BinaryOp, CPU, "Equal", functor::equal_to, float, Eigen::half,
            double, uint8, int8, int16, int32, int64, complex64, string);
 #if GOOGLE_CUDA
-REGISTER7(BinaryOp, GPU, "Equal", functor::equal_to, Eigen::half, float, double,
+REGISTER7(BinaryOp, GPU, "Equal", functor::equal_to, float, Eigen::half, double,
           uint8, int8, int16, int64);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_erf.cc b/tensorflow/core/kernels/cwise_op_erf.cc
index e784ae0f75c..ad6f8d63aaa 100644
--- a/tensorflow/core/kernels/cwise_op_erf.cc
+++ b/tensorflow/core/kernels/cwise_op_erf.cc
@@ -16,8 +16,8 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "Erf", functor::erf, Eigen::half, float, double);
+REGISTER3(UnaryOp, CPU, "Erf", functor::erf, float, Eigen::half, double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Erf", functor::erf, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Erf", functor::erf, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_erfc.cc b/tensorflow/core/kernels/cwise_op_erfc.cc
index 2d0664c37d2..9acf712b1d9 100644
--- a/tensorflow/core/kernels/cwise_op_erfc.cc
+++ b/tensorflow/core/kernels/cwise_op_erfc.cc
@@ -16,8 +16,8 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "Erfc", functor::erfc, Eigen::half, float, double);
+REGISTER3(UnaryOp, CPU, "Erfc", functor::erfc, float, Eigen::half, double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Erfc", functor::erfc, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Erfc", functor::erfc, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_exp.cc b/tensorflow/core/kernels/cwise_op_exp.cc
index cb913930806..2d7df89149d 100644
--- a/tensorflow/core/kernels/cwise_op_exp.cc
+++ b/tensorflow/core/kernels/cwise_op_exp.cc
@@ -16,9 +16,9 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Exp", functor::exp, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Exp", functor::exp, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Exp", functor::exp, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Exp", functor::exp, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_floor.cc b/tensorflow/core/kernels/cwise_op_floor.cc
index 7205399888c..8d4ae1b22a7 100644
--- a/tensorflow/core/kernels/cwise_op_floor.cc
+++ b/tensorflow/core/kernels/cwise_op_floor.cc
@@ -16,8 +16,8 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "Floor", functor::floor, Eigen::half, float, double);
+REGISTER3(UnaryOp, CPU, "Floor", functor::floor, float, Eigen::half, double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Floor", functor::floor, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Floor", functor::floor, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_greater.cc b/tensorflow/core/kernels/cwise_op_greater.cc
index 86e5e25da3a..d50bfcf267c 100644
--- a/tensorflow/core/kernels/cwise_op_greater.cc
+++ b/tensorflow/core/kernels/cwise_op_greater.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER8(BinaryOp, CPU, "Greater", functor::greater, Eigen::half, float,
+REGISTER8(BinaryOp, CPU, "Greater", functor::greater, float, Eigen::half,
           double, int32, int64, uint8, int8, int16);
 #if GOOGLE_CUDA
-REGISTER7(BinaryOp, GPU, "Greater", functor::greater, Eigen::half, float,
+REGISTER7(BinaryOp, GPU, "Greater", functor::greater, float, Eigen::half,
           double, int64, uint8, int8, int16);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_greater_equal.cc b/tensorflow/core/kernels/cwise_op_greater_equal.cc
index e6f4a057fbb..368165b9d1d 100644
--- a/tensorflow/core/kernels/cwise_op_greater_equal.cc
+++ b/tensorflow/core/kernels/cwise_op_greater_equal.cc
@@ -16,11 +16,11 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER8(BinaryOp, CPU, "GreaterEqual", functor::greater_equal, Eigen::half,
-          float, double, int32, int64, uint8, int8, int16);
+REGISTER8(BinaryOp, CPU, "GreaterEqual", functor::greater_equal, float,
+          Eigen::half, double, int32, int64, uint8, int8, int16);
 #if GOOGLE_CUDA
-REGISTER7(BinaryOp, GPU, "GreaterEqual", functor::greater_equal, Eigen::half,
-          float, double, int64, uint8, int8, int16);
+REGISTER7(BinaryOp, GPU, "GreaterEqual", functor::greater_equal, float,
+          Eigen::half, double, int64, uint8, int8, int16);
 
 // A special GPU kernel for int32.
 // TODO(b/25387198): Also enable int32 in device memory. This kernel
diff --git a/tensorflow/core/kernels/cwise_op_inverse.cc b/tensorflow/core/kernels/cwise_op_inverse.cc
index 796f1d50b74..05834996c0c 100644
--- a/tensorflow/core/kernels/cwise_op_inverse.cc
+++ b/tensorflow/core/kernels/cwise_op_inverse.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Inv", functor::inverse, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Inv", functor::inverse, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER4(UnaryOp, GPU, "Inv", functor::inverse, Eigen::half, float, double,
+REGISTER4(UnaryOp, GPU, "Inv", functor::inverse, float, Eigen::half, double,
           int64);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_isfinite.cc b/tensorflow/core/kernels/cwise_op_isfinite.cc
index 33e2e20d38f..ebdfd8e13ff 100644
--- a/tensorflow/core/kernels/cwise_op_isfinite.cc
+++ b/tensorflow/core/kernels/cwise_op_isfinite.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "IsFinite", functor::isfinite, Eigen::half, float,
+REGISTER3(UnaryOp, CPU, "IsFinite", functor::isfinite, float, Eigen::half,
           double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "IsFinite", functor::isfinite, Eigen::half, float,
+REGISTER3(UnaryOp, GPU, "IsFinite", functor::isfinite, float, Eigen::half,
           double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_isinf.cc b/tensorflow/core/kernels/cwise_op_isinf.cc
index d4d821cea6c..50a4c7e76ae 100644
--- a/tensorflow/core/kernels/cwise_op_isinf.cc
+++ b/tensorflow/core/kernels/cwise_op_isinf.cc
@@ -16,8 +16,8 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "IsInf", functor::isinf, Eigen::half, float, double);
+REGISTER3(UnaryOp, CPU, "IsInf", functor::isinf, float, Eigen::half, double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "IsInf", functor::isinf, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "IsInf", functor::isinf, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_isnan.cc b/tensorflow/core/kernels/cwise_op_isnan.cc
index c7e3edfe056..0443f592318 100644
--- a/tensorflow/core/kernels/cwise_op_isnan.cc
+++ b/tensorflow/core/kernels/cwise_op_isnan.cc
@@ -16,8 +16,8 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "IsNan", functor::isnan, Eigen::half, float, double);
+REGISTER3(UnaryOp, CPU, "IsNan", functor::isnan, float, Eigen::half, double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "IsNan", functor::isnan, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "IsNan", functor::isnan, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_less.cc b/tensorflow/core/kernels/cwise_op_less.cc
index 18922f1ee72..aa9990cbfd9 100644
--- a/tensorflow/core/kernels/cwise_op_less.cc
+++ b/tensorflow/core/kernels/cwise_op_less.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER8(BinaryOp, CPU, "Less", functor::less, Eigen::half, float, double,
+REGISTER8(BinaryOp, CPU, "Less", functor::less, float, Eigen::half, double,
           int32, int64, uint8, int8, int16);
 #if GOOGLE_CUDA
-REGISTER7(BinaryOp, GPU, "Less", functor::less, Eigen::half, float, double,
+REGISTER7(BinaryOp, GPU, "Less", functor::less, float, Eigen::half, double,
           int64, uint8, int8, int16);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_less_equal.cc b/tensorflow/core/kernels/cwise_op_less_equal.cc
index 4ba840ad390..87681279e36 100644
--- a/tensorflow/core/kernels/cwise_op_less_equal.cc
+++ b/tensorflow/core/kernels/cwise_op_less_equal.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER8(BinaryOp, CPU, "LessEqual", functor::less_equal, Eigen::half, float,
+REGISTER8(BinaryOp, CPU, "LessEqual", functor::less_equal, float, Eigen::half,
           double, int32, int64, uint8, int8, int16);
 #if GOOGLE_CUDA
-REGISTER7(BinaryOp, GPU, "LessEqual", functor::less_equal, Eigen::half, float,
+REGISTER7(BinaryOp, GPU, "LessEqual", functor::less_equal, float, Eigen::half,
           double, int64, uint8, int8, int16);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_lgamma.cc b/tensorflow/core/kernels/cwise_op_lgamma.cc
index 2a63ea64f63..d5f84eb620c 100644
--- a/tensorflow/core/kernels/cwise_op_lgamma.cc
+++ b/tensorflow/core/kernels/cwise_op_lgamma.cc
@@ -16,8 +16,8 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER3(UnaryOp, CPU, "Lgamma", functor::lgamma, Eigen::half, float, double);
+REGISTER3(UnaryOp, CPU, "Lgamma", functor::lgamma, float, Eigen::half, double);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Lgamma", functor::lgamma, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Lgamma", functor::lgamma, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_log.cc b/tensorflow/core/kernels/cwise_op_log.cc
index ca7e5b360f2..ab6a1f97783 100644
--- a/tensorflow/core/kernels/cwise_op_log.cc
+++ b/tensorflow/core/kernels/cwise_op_log.cc
@@ -16,9 +16,9 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Log", functor::log, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Log", functor::log, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Log", functor::log, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Log", functor::log, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_maximum.cc b/tensorflow/core/kernels/cwise_op_maximum.cc
index 1e3ffb74476..c3792e1a53c 100644
--- a/tensorflow/core/kernels/cwise_op_maximum.cc
+++ b/tensorflow/core/kernels/cwise_op_maximum.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER5(BinaryOp, CPU, "Maximum", functor::maximum, Eigen::half, float,
+REGISTER5(BinaryOp, CPU, "Maximum", functor::maximum, float, Eigen::half,
           double, int32, int64);
 #if GOOGLE_CUDA
-REGISTER4(BinaryOp, GPU, "Maximum", functor::maximum, Eigen::half, float,
+REGISTER4(BinaryOp, GPU, "Maximum", functor::maximum, float, Eigen::half,
           double, int64);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_minimum.cc b/tensorflow/core/kernels/cwise_op_minimum.cc
index 7a000f36552..3586f4cf48e 100644
--- a/tensorflow/core/kernels/cwise_op_minimum.cc
+++ b/tensorflow/core/kernels/cwise_op_minimum.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER5(BinaryOp, CPU, "Minimum", functor::minimum, Eigen::half, float,
+REGISTER5(BinaryOp, CPU, "Minimum", functor::minimum, float, Eigen::half,
           double, int32, int64);
 #if GOOGLE_CUDA
-REGISTER4(BinaryOp, GPU, "Minimum", functor::minimum, Eigen::half, float,
+REGISTER4(BinaryOp, GPU, "Minimum", functor::minimum, float, Eigen::half,
           double, int64);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_mul.cc b/tensorflow/core/kernels/cwise_op_mul.cc
index 7939f0d74d1..395cea5d7fd 100644
--- a/tensorflow/core/kernels/cwise_op_mul.cc
+++ b/tensorflow/core/kernels/cwise_op_mul.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER9(BinaryOp, CPU, "Mul", functor::mul, Eigen::half, float, double, uint8,
+REGISTER9(BinaryOp, CPU, "Mul", functor::mul, float, Eigen::half, double, uint8,
           int8, int16, int32, int64, complex64);
 #if GOOGLE_CUDA
-REGISTER7(BinaryOp, GPU, "Mul", functor::mul, Eigen::half, float, double, uint8,
+REGISTER7(BinaryOp, GPU, "Mul", functor::mul, float, Eigen::half, double, uint8,
           int8, int16, int64);
 // A special GPU kernel for int32.
 // TODO(b/25387198): Also enable int32 in device memory. This kernel
diff --git a/tensorflow/core/kernels/cwise_op_neg.cc b/tensorflow/core/kernels/cwise_op_neg.cc
index 97bbbb20ca4..2b672285b3e 100644
--- a/tensorflow/core/kernels/cwise_op_neg.cc
+++ b/tensorflow/core/kernels/cwise_op_neg.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER6(UnaryOp, CPU, "Neg", functor::neg, Eigen::half, float, double, int32,
+REGISTER6(UnaryOp, CPU, "Neg", functor::neg, float, Eigen::half, double, int32,
           complex64, int64);
 #if GOOGLE_CUDA
-REGISTER4(UnaryOp, GPU, "Neg", functor::neg, Eigen::half, float, double, int64);
+REGISTER4(UnaryOp, GPU, "Neg", functor::neg, float, Eigen::half, double, int64);
 
 // A special GPU kernel for int32.
 // TODO(b/25387198): Also enable int32 in device memory. This kernel
diff --git a/tensorflow/core/kernels/cwise_op_not_equal_to.cc b/tensorflow/core/kernels/cwise_op_not_equal_to.cc
index 17688cce628..3ca522dd297 100644
--- a/tensorflow/core/kernels/cwise_op_not_equal_to.cc
+++ b/tensorflow/core/kernels/cwise_op_not_equal_to.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER10(BinaryOp, CPU, "NotEqual", functor::not_equal_to, Eigen::half, float,
+REGISTER10(BinaryOp, CPU, "NotEqual", functor::not_equal_to, float, Eigen::half,
            double, uint8, int8, int16, int32, int64, complex64, string);
 #if GOOGLE_CUDA
-REGISTER7(BinaryOp, GPU, "NotEqual", functor::not_equal_to, Eigen::half, float,
+REGISTER7(BinaryOp, GPU, "NotEqual", functor::not_equal_to, float, Eigen::half,
           double, uint8, int8, int16, int64);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_pow.cc b/tensorflow/core/kernels/cwise_op_pow.cc
index 8f8c1000fe3..8bb71c03d8e 100644
--- a/tensorflow/core/kernels/cwise_op_pow.cc
+++ b/tensorflow/core/kernels/cwise_op_pow.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER6(BinaryOp, CPU, "Pow", functor::pow, Eigen::half, float, double, int32,
+REGISTER6(BinaryOp, CPU, "Pow", functor::pow, float, Eigen::half, double, int32,
           int64, complex64);
 #if GOOGLE_CUDA
-REGISTER4(BinaryOp, GPU, "Pow", functor::pow, Eigen::half, float, double,
+REGISTER4(BinaryOp, GPU, "Pow", functor::pow, float, Eigen::half, double,
           int64);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_rsqrt.cc b/tensorflow/core/kernels/cwise_op_rsqrt.cc
index bd2b21da2b6..ff3e8d778f7 100644
--- a/tensorflow/core/kernels/cwise_op_rsqrt.cc
+++ b/tensorflow/core/kernels/cwise_op_rsqrt.cc
@@ -16,9 +16,9 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Rsqrt", functor::rsqrt, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Rsqrt", functor::rsqrt, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Rsqrt", functor::rsqrt, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Rsqrt", functor::rsqrt, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_sigmoid.cc b/tensorflow/core/kernels/cwise_op_sigmoid.cc
index 7748d4934f8..574866b3f09 100644
--- a/tensorflow/core/kernels/cwise_op_sigmoid.cc
+++ b/tensorflow/core/kernels/cwise_op_sigmoid.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Sigmoid", functor::sigmoid, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Sigmoid", functor::sigmoid, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Sigmoid", functor::sigmoid, Eigen::half, float,
+REGISTER3(UnaryOp, GPU, "Sigmoid", functor::sigmoid, float, Eigen::half,
           double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_sign.cc b/tensorflow/core/kernels/cwise_op_sign.cc
index 78e03dd32d1..7107970332d 100644
--- a/tensorflow/core/kernels/cwise_op_sign.cc
+++ b/tensorflow/core/kernels/cwise_op_sign.cc
@@ -19,7 +19,7 @@ namespace tensorflow {
 REGISTER6(UnaryOp, CPU, "Sign", functor::sign, float, double, int32, int64,
           complex64, Eigen::half);
 #if GOOGLE_CUDA
-REGISTER4(UnaryOp, GPU, "Sign", functor::sign, Eigen::half, float, double,
+REGISTER4(UnaryOp, GPU, "Sign", functor::sign, float, Eigen::half, double,
           int64);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_sin.cc b/tensorflow/core/kernels/cwise_op_sin.cc
index 48c4e01aac1..123a251c620 100644
--- a/tensorflow/core/kernels/cwise_op_sin.cc
+++ b/tensorflow/core/kernels/cwise_op_sin.cc
@@ -16,9 +16,9 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Sin", functor::sin, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Sin", functor::sin, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Sin", functor::sin, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Sin", functor::sin, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_sqrt.cc b/tensorflow/core/kernels/cwise_op_sqrt.cc
index 7af3c42c655..daaa4c55c6b 100644
--- a/tensorflow/core/kernels/cwise_op_sqrt.cc
+++ b/tensorflow/core/kernels/cwise_op_sqrt.cc
@@ -16,9 +16,9 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Sqrt", functor::sqrt, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Sqrt", functor::sqrt, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Sqrt", functor::sqrt, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Sqrt", functor::sqrt, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow
diff --git a/tensorflow/core/kernels/cwise_op_square.cc b/tensorflow/core/kernels/cwise_op_square.cc
index 71154a8553b..80c8423cf9b 100644
--- a/tensorflow/core/kernels/cwise_op_square.cc
+++ b/tensorflow/core/kernels/cwise_op_square.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER6(UnaryOp, CPU, "Square", functor::square, Eigen::half, float, double,
+REGISTER6(UnaryOp, CPU, "Square", functor::square, float, Eigen::half, double,
           int32, complex64, int64);
 #if GOOGLE_CUDA
-REGISTER4(UnaryOp, GPU, "Square", functor::square, Eigen::half, float, double,
+REGISTER4(UnaryOp, GPU, "Square", functor::square, float, Eigen::half, double,
           int64);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_sub.cc b/tensorflow/core/kernels/cwise_op_sub.cc
index cd2f1ae753e..245f13ad68b 100644
--- a/tensorflow/core/kernels/cwise_op_sub.cc
+++ b/tensorflow/core/kernels/cwise_op_sub.cc
@@ -16,10 +16,10 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER6(BinaryOp, CPU, "Sub", functor::sub, Eigen::half, float, double, int32,
+REGISTER6(BinaryOp, CPU, "Sub", functor::sub, float, Eigen::half, double, int32,
           int64, complex64);
 #if GOOGLE_CUDA
-REGISTER4(BinaryOp, GPU, "Sub", functor::sub, Eigen::half, float, double,
+REGISTER4(BinaryOp, GPU, "Sub", functor::sub, float, Eigen::half, double,
           int64);
 
 // A special GPU kernel for int32.
diff --git a/tensorflow/core/kernels/cwise_op_tanh.cc b/tensorflow/core/kernels/cwise_op_tanh.cc
index a4cbb150a0f..a42b6fe6e41 100644
--- a/tensorflow/core/kernels/cwise_op_tanh.cc
+++ b/tensorflow/core/kernels/cwise_op_tanh.cc
@@ -16,9 +16,9 @@ limitations under the License.
 #include "tensorflow/core/kernels/cwise_ops_common.h"
 
 namespace tensorflow {
-REGISTER4(UnaryOp, CPU, "Tanh", functor::tanh, Eigen::half, float, double,
+REGISTER4(UnaryOp, CPU, "Tanh", functor::tanh, float, Eigen::half, double,
           complex64);
 #if GOOGLE_CUDA
-REGISTER3(UnaryOp, GPU, "Tanh", functor::tanh, Eigen::half, float, double);
+REGISTER3(UnaryOp, GPU, "Tanh", functor::tanh, float, Eigen::half, double);
 #endif
 }  // namespace tensorflow