[llvm] [clang] [NVPTX] Add 'activemask' builtin and intrinsic support (PR #79768)

[Artem Belevich via cfe-commits]

================
@@ -65,7 +65,7 @@ def : Proc<"sm_61", [SM61, PTX50]>;
 def : Proc<"sm_62", [SM62, PTX50]>;
 def : Proc<"sm_70", [SM70, PTX60]>;
 def : Proc<"sm_72", [SM72, PTX61]>;
-def : Proc<"sm_75", [SM75, PTX63]>;
+def : Proc<"sm_75", [SM75, PTX62, PTX63]>;
----------------
Artem-B wrote:

I'm confused a bit here. Constraints on PTX version for GPU and for instrunctions are independent. You need both satisfied in order to use a given instruction on a given GPU.

So, to use activemask on  sm_75, you do need PTX63.
To use it on sm_52, you only need PTX62.

You do not need to change anything here. You already have correct predicates applied to the instruction itself and to the target builtin.

https://github.com/llvm/llvm-project/pull/79768