[llvm-branch-commits] [llvm] [X86] keep upper-16-bit f32 extractions in XMM for f16/v8i16 (PR #230698)

[Emilio Cota via llvm-branch-commits]

================
@@ -47241,6 +47241,22 @@ static SDValue combineBitcast(SDNode *N, SelectionDAG &DAG,
                          DAG.getBitcast(MVT::v8f16, Src.getOperand(0)),
                          DAG.getVectorIdxConstant(0, DL));
     }
+    using namespace SDPatternMatch;
+    SDValue FpOp;
+    if (TLI.isTypeLegal(MVT::v8f16) && N0.hasOneUse() &&
----------------
cota wrote:

Done, thanks

https://github.com/llvm/llvm-project/pull/230698