[llvm] [NVPTX] support packed f32 instructions for sm_100+ (PR #126337)

[Alex MacLean via llvm-commits]

================
@@ -2536,6 +2559,10 @@ foreach vt = [v2f16, v2bf16, v2i16, v4i8] in {
   def: Pat<(vt (ProxyReg  vt:$src)), (ProxyRegI32 $src)>;
 }
 
+def: Pat<(v2f32 (ProxyReg v2f32:$src)), (ProxyRegI64 $src)>;
+
+def: Pat<(v2f32 (bitconvert i64:$src)), (ProxyRegI64 $src)>;
----------------
AlexMaclean wrote:

Can we just use `(v2f32 Int64Regs:$src)` here?

https://github.com/llvm/llvm-project/pull/126337