How to use double in kernel function?

I read section 9.3 of OpenCL specification,how can I use double in a kernel function? I tried to add “#pragma OPENCL EXTENSION cl_khr_fp64:enable"on the top of kernel function,but my program failed to compile,what can I do ?
By the way, I used “clGetDeviceInfo” to query"CL_DEVICE_PREFERRED_VECTOR_WIDTH_DOUBLE”, it returned 0,does it means my GPU support double precision or not ?

You should use the CL_DEVICE_EXTENSIONS flag to query the extensions of your device. Then check if cl_khr_fp64 is supported or not. But almost sounds like it’s not.