[Mlir-commits] [mlir] [mlir][ArmSME] More precisely model dataflow in ArmSME to SCF lowerings (PR #73922)

[Cullen Rhodes]

================
@@ -121,14 +123,17 @@ struct TileLoadOpConversion : public OpRewritePattern<arm_sme::TileLoadOp> {
     getMemrefIndices(tileLoadOp.getIndices(),
                      tileLoadOp.getMemRefType().getRank(), tileSliceIndex,
                      numTileSlices, memrefIndices, loc, rewriter);
-    tileLoadOp.createOpAndForwardTileId<arm_sme::LoadTileSliceOp>(
-        rewriter, loc, tileType, tileLoadOp.getBase(), allTruePredicate, tile,
-        memrefIndices, tileSliceIndex, tileLoadOp.getLayout());
+    auto currentTile = forOp.getRegionIterArg(0);
+    auto loadSlice =
+        tileLoadOp.createOpAndForwardTileId<arm_sme::LoadTileSliceOp>(
+            rewriter, loc, tileType, tileLoadOp.getBase(), allTruePredicate,
+            currentTile, memrefIndices, tileSliceIndex, tileLoadOp.getLayout());
+    rewriter.create<scf::YieldOp>(loc, ValueRange{loadSlice});
----------------
c-rhodes wrote:

use `getResult()`? (and below)
```suggestion
    rewriter.create<scf::YieldOp>(loc, loadSlice.getResult());
```

https://github.com/llvm/llvm-project/pull/73922