[Mlir-commits] [mlir] [MLIR] Add continuous tiling to TileUsingForOp (PR #82792)

[llvmlistbot at llvm.org]

================
@@ -2581,6 +2643,157 @@ DiagnosedSilenceableFailure transform::TileReductionUsingForallOp::applyToOne(
   return DiagnosedSilenceableFailure::success();
 }
 
+//===----------------------------------------------------------------------===//
+// ContinuousTileSizesOp
+//===----------------------------------------------------------------------===//
+
+DiagnosedSilenceableFailure
+transform::ContinuousTileSizesOp::apply(transform::TransformRewriter &rewriter,
+                                        TransformResults &transformResults,
+                                        TransformState &state) {
+
+  SmallVector<Operation *> targetOps =
+      llvm::to_vector(state.getPayloadOps(getTarget()));
+
+  if (!llvm::hasSingleElement(targetOps)) {
+    return emitDefiniteFailure() << "requires exactly one target (got "
+                                 << llvm::range_size(targetOps) << ")";
+  }
+
+  auto target = dyn_cast<LinalgOp>(*targetOps.begin());
+
+  OpBuilder builder(target.getContext());
----------------
muneebkhan85 wrote:

The `OpBuilder` is needed for operations such as `makeComposedAffineApply` and `getValueOrCreateConstantIndexOp`. The provided rewriter would be usable only partly.

https://github.com/llvm/llvm-project/pull/82792