[llvm] [NVPTX] Add support for f16 fabs (PR #116107)

[Artem Belevich via llvm-commits]

================
@@ -0,0 +1,99 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; ## Some FP16 support but not for abs
+; RUN: llc < %s -mtriple=nvptx64-nvidia-cuda -mcpu=sm_53  \
----------------
Artem-B wrote:

Few nits on the tests -- I would explicitly set PTX version high enough on all test cases to make sure that native `abs` instruction is disabled by the GPU type or by explicit attributes, and not unintentionally so by the too-low default PTX version.

https://github.com/llvm/llvm-project/pull/116107