[mlir][transform] Emit error message with emitSilenceableFailure (#86146)

The previous implementation used a `notifyMatchFailure` to emit failure
message inappropriately and then used the
`emitDefaultSilenceableFailure`. This patch changes this to use the more
appropriate `emitSilenceableFailure` with error message. Additionally a
failure test has been added.

mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp

@@ -3269,22 +3269,24 @@ DiagnosedSilenceableFailure transform::FlattenElementwiseLinalgOp::applyToOne(
     transform::ApplyToEachResultList &results,
     transform::TransformState &state) {
   rewriter.setInsertionPoint(target);
-  if (!isElementwise(target)) {
+  if (!isElementwise(target))
-    failed(rewriter.notifyMatchFailure(
+    return mlir::emitSilenceableFailure(target->getLoc())
-        target, "only elementwise flattening is supported"));
+           << "only elementwise flattening is supported";
-    return emitDefaultSilenceableFailure(target);
+
-  }
   // If rank <= 1, do nothing
   if (target.getNumLoops() <= 1) {
     results.push_back(target);
     return DiagnosedSilenceableFailure::success();
   }
+
+  // Attempt to flatten all dims to one.
   ReassociationIndices reassociation(target.getNumLoops());
   std::iota(reassociation.begin(), reassociation.end(), 0);
   auto maybeFlattened =
       collapseOpIterationDims(target, reassociation, rewriter);
   if (failed(maybeFlattened))
-    return emitDefaultSilenceableFailure(target);
+    return mlir::emitSilenceableFailure(target->getLoc())
+           << "attempted to flatten, but failed";
   results.push_back(maybeFlattened->collapsedOp);
   rewriter.replaceOp(target, maybeFlattened->results);
   return DiagnosedSilenceableFailure::success();

mlir/test/Dialect/Linalg/flatten-unsupported.mlir

@@ -0,0 +1,33 @@
+// RUN: mlir-opt %s -transform-interpreter -split-input-file -verify-diagnostics
+
+func.func @non_elementwise(%arg0: memref<2x3xf32>, %arg1: memref<3x4xf32>, %arg2: memref<2x4xf32>) {
+  // expected-error @below {{only elementwise flattening is supported}}
+  linalg.matmul ins(%arg0, %arg1 : memref<2x3xf32>, memref<3x4xf32>) outs(%arg2: memref<2x4xf32>)
+  return
+}
+
+module attributes {transform.with_named_sequence} {
+  transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+    %0 = transform.structured.match interface{LinalgOp} in %arg1 : (!transform.any_op) -> !transform.any_op
+    %flattened = transform.structured.flatten_elementwise %0
+      : (!transform.any_op) -> !transform.any_op
+    transform.yield
+  }
+}
+
+// -----
+
+func.func @unsupported_memref(%arg0: memref<32x7xf32, strided<[7, 2]>>, %arg1: memref<32x7xf32, strided<[7, 2]>>, %arg2: memref<32x7xf32, strided<[7, 2]>>) {
+  // expected-error @below {{attempted to flatten, but failed}}
+    linalg.map {arith.addf} ins(%arg0, %arg1: memref<32x7xf32, strided<[7, 2]>>, memref<32x7xf32, strided<[7, 2]>>) outs(%arg2: memref<32x7xf32, strided<[7, 2]>>)
+    return
+}
+
+module attributes {transform.with_named_sequence} {
+  transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+    %0 = transform.structured.match interface{LinalgOp} in %arg1 : (!transform.any_op) -> !transform.any_op
+    %flattened = transform.structured.flatten_elementwise %0
+      : (!transform.any_op) -> !transform.any_op
+    transform.yield
+  }
+}