[llvm] [NVPTX] Add intrinsics for narrow-fp to bf16 conversions (PR #191376)

[Rajat Bajpai via llvm-commits]

================
@@ -666,6 +670,16 @@ let hasSideEffects = false in {
 
   defm CVT_e4m3x2 : CVT_TO_F8X2<"e4m3">;
   defm CVT_e5m2x2 : CVT_TO_F8X2<"e5m2">;
+  
----------------
rajatbajpai wrote:

This could be merged with below `fp8x2 -> f16x2 (unscaled)` under a single multiclass.

https://github.com/llvm/llvm-project/pull/191376