[HIPClang] Switch ordering on math functions

Header math_functions.h should be included after including __clang_cuda_math_forward_declares.h to avoid warning: attribute declaration must precede definition.

hipamd/include/hip/hcc_detail/hip_runtime.h

@@ -108,10 +108,10 @@ extern int HIP_TRACE_API;
 #endif
 #include <hip/hcc_detail/hip_atomic.h>
 #include <hip/hcc_detail/host_defines.h>
-#include <hip/hcc_detail/math_functions.h>
 #include <hip/hcc_detail/device_functions.h>
 #include <hip/hcc_detail/surface_functions.h>
 #if __HCC__
+#include <hip/hcc_detail/math_functions.h>
 #include <hip/hcc_detail/texture_functions.h>
 #endif // __HCC__
 
@@ -438,8 +438,6 @@ extern const __device__ __attribute__((weak)) __hip_builtin_gridDim_t gridDim;
 #define hipGridDim_y gridDim.y
 #define hipGridDim_z gridDim.z
 
-#include <hip/hcc_detail/math_functions.h>
-
 // Support std::complex.
 #pragma push_macro("__CUDA__")
 #define __CUDA__
@@ -450,6 +448,8 @@ extern const __device__ __attribute__((weak)) __hip_builtin_gridDim_t gridDim;
 #undef __CUDA__
 #pragma pop_macro("__CUDA__")
 
+#include <hip/hcc_detail/math_functions.h>
+
 #endif
 
 #endif  // HIP_HCC_DETAIL_RUNTIME_H