[mlir][Linalg] Do not return failure when all tile sizes are zero.

Returning failure when tile sizes are all zero prevents the change in
the marker. This makes pattern rewriter run the pattern multiple times
only to exit when it hits a limit. Instead just clone the operation
(since tiling is essentially cloning in this case). Then the
transformation filter kicks in to avoid the pattern rewriter to be
invoked many times.

Differential Revision: https://reviews.llvm.org/D113949

mlir/lib/Dialect/Linalg/Transforms/Tiling.cpp

@@ -159,8 +159,13 @@ tileLinalgOpImpl(OpBuilder &b, LinalgOp op, ValueRange tileSizes,
   // Initial tile sizes may be too big, only take the first nLoops.
   tileSizes = tileSizes.take_front(nLoops);
 
-  if (llvm::all_of(tileSizes, isZero))
-    return failure();
+  if (llvm::all_of(tileSizes, isZero)) {
+    TiledLinalgOp tiledOp;
+    tiledOp.op = cast<LinalgOp>(b.clone(*op.getOperation()));
+    tiledOp.tensorResults.assign(tiledOp.op->result_begin(),
+                                 tiledOp.op->result_end());
+    return tiledOp;
+  }
 
   // 1. Build the tiled loop ranges.
   auto allShapeSizes = op.createFlatListOfOperandDims(b, op.getLoc());

mlir/test/Dialect/Linalg/tile-zero.mlir

@@ -0,0 +1,12 @@
+// RUN: mlir-opt -test-linalg-transform-patterns=test-tile-pattern %s | FileCheck %s
+
+func @matmul_zero_tile(
+  %arg0: tensor<?x?xf32>, %arg1 : tensor<?x?xf32>, %arg2 : tensor<?x?xf32>) -> tensor<?x?xf32> {
+  %0 = linalg.matmul {__internal_linalg_transform__ = "tile"}
+      ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>)
+      outs(%arg2 : tensor<?x?xf32>) -> tensor<?x?xf32>
+  return %0 : tensor<?x?xf32>
+}
+// CHECK-LABEL: matmul_zero_tile
+//       CHECK:   linalg.matmul
+//   CHECK-NOT:   __internal_linalg_transform__