[Mlir-commits] [mlir] [mlir][x86vector] Shuffle BF16 vector.contract output for Flat layout. (PR #174590)

[Arun Thangamani]

================
@@ -297,6 +372,104 @@ struct VectorContractBF16ToFMA
     VectorType dstType =
         VectorType::get(nonUnitDimAcc.front(), rewriter.getF32Type());
 
+    if (!isVnni) {
+
+      // Validate and shuffle the accumulator
+      Operation *accReadOp0 =
+          traceToVectorReadLikeParentOperation(contractOp.getAcc());
+      Operation *accReadOp1 =
+          traceToVectorReadLikeParentOperation(pairContractOp.getAcc());
+
+      // Iterate dowm to find the users of contact operations until it is store
+      // or transfer_write.
+      Operation *resultWriteOp0 =
+          traceToVectorWriteLikeUserOperation(contractOp.getResult());
+      Operation *resultWriteOp1 =
+          traceToVectorWriteLikeUserOperation(pairContractOp.getResult());
+
+      if (!accReadOp0 || !accReadOp1)
+        return rewriter.notifyMatchFailure(
+            contractOp,
+            "Operands doesn't have load or transfer_read as it's parent op");
----------------
arun-thmn wrote:

okay. Fixed.

https://github.com/llvm/llvm-project/pull/174590