Use fp16 as data type in all examples

example/37_permute/CMakeLists.txt

 add_custom_target(example_permute)
 
-add_example_executable(example_permute_1xHxW_fp32 permute_1xHxW_fp32.cpp)
-add_example_executable(example_permute_NxHxW_fp32 permute_NxHxW_fp32.cpp)
+add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
+add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
 add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
 
-add_dependencies(example_permute example_permute_1xHxW_fp32)
-add_dependencies(example_permute example_permute_NxHxW_fp32)
+add_dependencies(example_permute example_permute_1xHxW_fp16)
+add_dependencies(example_permute example_permute_NxHxW_fp16)
 add_dependencies(example_permute example_permute_HxWx4_fp16)

example/37_permute/permute_1xHxW_fp32.cpp → example/37_permute/permute_1xHxW_fp16.cpp

@@ -3,8 +3,8 @@
 
 #include "common.hpp"
 
-using ADataType = F32;
-using BDataType = F32;
+using ADataType = F16;
+using BDataType = F16;
 
 // clang-format off
 using DevicePermuteInstance = ck::tensor_operation::device::DevicePermute

example/37_permute/permute_NxHxW_fp32.cpp → example/37_permute/permute_NxHxW_fp16.cpp

@@ -3,8 +3,8 @@
 
 #include "common.hpp"
 
-using ADataType = F32;
-using BDataType = F32;
+using ADataType = F16;
+using BDataType = F16;
 
 // clang-format off
 using DevicePermuteInstance = ck::tensor_operation::device::DevicePermute