[Mlir-commits] [mlir] [mlir][Vector] Tighten up application conditions in TransferReadAfter… (PR #143869)

[Kunwar Grover]

================
@@ -4684,15 +4687,27 @@ struct TransferReadAfterWriteToBroadcast
     if (getUnusedDimsBitVector({readOp.getPermutationMap()}) !=
         getUnusedDimsBitVector({defWrite.getPermutationMap()}))
       return failure();
-    if (readOp.getIndices() != defWrite.getIndices() ||
-        readOp.getMask() != defWrite.getMask())
+    // This pattern should only catch the broadcast case, the non-broadcast case
+    // should be done separately to keep application conditions clean and
+    // separate.
+    AffineMap readMap = compressUnusedDims(readOp.getPermutationMap());
+    AffineMap writeMap = compressUnusedDims(defWrite.getPermutationMap());
+    bool bcast = !readMap.getBroadcastDims().empty() ||
+                 !writeMap.getBroadcastDims().empty();
+    if (!bcast)
+      return failure();
+    // At this point, we know we have a bcast.
+    // The masked case is too complext atm, bail.
+    if (readOp.getMask() || defWrite.getMask())
+      return failure();
----------------
Groverkss wrote:

We should document why we are bailing (i.e. we need to respect the mask for padding)

https://github.com/llvm/llvm-project/pull/143869