update CHANGES

CHANGES

@@ -4,6 +4,7 @@
 Notable User Facing Changes
 ---------------------------
 
+- Support for Clang/LLVM 16.0
 - Dropped support for Clang/LLVM versions 6.0 to 9.0 (inclusive)
 - Added support for program-scope variables in the CPU drivers
 - Added support for generic address spaces in the CPU drivers
@@ -35,12 +36,18 @@ Notable User Facing Changes
   local memory size for the CPU devices.
 - CUDA: Added Direct Peer-to-Peer buffer migrations for much better performance
   scaling in multi-GPU scenarios
+- Partial support for cl_khr_fp16 when compiled with LLVM 16+. The
+  implementation relies on Clang, and may result in emulation if
+  the CPU does not support the required instruction set.
 
 Notable Fixes
 -------------
 
 - Fix handling of clBuildProgram() -I directories given with quotes or
   with leading/trailing white space.
+- CPU devices now flush printf() output after each call instead of waiting
+  for the end of the kernel command. This makes it (again) useful for debugging
+  kernel segfaults.
 
 3.1 December 2022
 =================