diff --git a/tensorflow/core/kernels/quantize_and_dequantize_op.cc b/tensorflow/core/kernels/quantize_and_dequantize_op.cc
index b5d416bb396d3db6b0711a2f36b4bb91c85d3e6b..4dd5e24e5a162d332a6b50000dbe8e982ac68237 100644
--- a/tensorflow/core/kernels/quantize_and_dequantize_op.cc
+++ b/tensorflow/core/kernels/quantize_and_dequantize_op.cc
@@ -164,6 +164,8 @@ TF_CALL_double(REGISTER_CPU_KERNEL);
 #define REGISTER_GPU_KERNEL(T)                                                 \
   REGISTER_KERNEL_BUILDER(Name("QuantizeAndDequantizeV2")                      \
                               .Device(DEVICE_GPU)                              \
+                              .HostMemory("input_max")                         \
+                              .HostMemory("input_min")                         \
                               .TypeConstraint<T>("T"),                         \
                           QuantizeAndDequantizeV2Op<GPUDevice, T>);            \
   REGISTER_KERNEL_BUILDER(                                                     \