[clang] [OpenCL] Add cl_khr_fp64 guard for atomic_double/double in cl_ext_float_atomics functions (PR #169252)

[Wenju He via cfe-commits]

================
@@ -13781,6 +13781,7 @@ float __ovld atomic_fetch_max_explicit(volatile atomic_float *,
     defined(__opencl_c_ext_fp32_local_atomic_min_max)
 
 #if defined(cl_khr_int64_base_atomics) && defined(cl_khr_int64_extended_atomics)
+#ifdef cl_khr_fp64
 #if defined(__opencl_c_ext_fp64_global_atomic_min_max)
----------------
wenju-he wrote:

> Interesting... maybe `__opencl_c_ext_fp64_global_atomic_min_max` should not be defined if `cl_khr_fp64` is not supported.

thanks for the suggestion. I've reverted the first commit and disabled `__opencl_c_ext_fp64_global_atomic_min_max` when `cl_khr_fp64` is not supported in commit https://github.com/llvm/llvm-project/pull/169252/commits/9f397aa93a29ede5fce75f75fb74b5e2dda88dcc

https://github.com/llvm/llvm-project/pull/169252