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

[mravishankar]

Summary

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.

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

 template <typename LoopTy>
 static FailureOr<TiledLinalgOp>
 tileLinalgOpImpl(OpBuilder &b, LinalgOp op, ValueRange tileSizes,
                  const LinalgTilingOptions &options) {
   auto nLoops = op.getNumLoops();
   // 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());
   AffineMap shapeSizesToLoopsMap = op.getShapesToLoopsMap();
   if (!shapeSizesToLoopsMap)
     return failure();
 
   SmallVector<Range, 4> loopRanges;

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

+// 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__