[llvm] [DAGCombine] Add DAG optimisation for BF16_TO_FP (PR #69426)

[Simon Pilgrim via llvm-commits]

================
@@ -26219,6 +26223,21 @@ SDValue DAGCombiner::visitFP_TO_BF16(SDNode *N) {
   return SDValue();
 }
 
+SDValue DAGCombiner::visitBF16_TO_FP(SDNode *N) {
----------------
RKSimon wrote:

SimplifyDemandedBits should handle shl(lshr(x,c),c) mask patterns already - are there bitcasts in the way?

https://github.com/llvm/llvm-project/pull/69426