[Mlir-commits] [mlir] [mlir][TilingInterface] Use `LoopLikeOpInterface` in tiling using SCF to unify tiling with `scf.for` and `scf.forall`. (PR #77874)

[Matthias Springer]

================
@@ -622,6 +626,47 @@ LogicalResult scf::ForallOp::promoteIfSingleIteration(RewriterBase &rewriter) {
   return success();
 }
 
+Block::BlockArgListType ForallOp::getRegionIterArgs() {
+  return getBody()->getArguments().drop_front(getRank());
+}
+
+MutableArrayRef<OpOperand> ForallOp::getInitsMutable() {
+  return getOutputsMutable();
+}
+
+FailureOr<LoopLikeOpInterface>
+ForallOp::replaceWithAdditionalYields(RewriterBase &rewriter,
+                                      ValueRange newInitOperands,
+                                      bool replaceInitOperandUsesInLoop,
+                                      const NewYieldValuesFn &newYieldValueFn) {
+  OpBuilder::InsertionGuard g(rewriter);
+  rewriter.setInsertionPoint(getOperation());
+  auto inits = llvm::to_vector(getOutputs());
+  inits.append(newInitOperands.begin(), newInitOperands.end());
+  auto newLoop = rewriter.create<scf::ForallOp>(
+      getLoc(), getMixedLowerBound(), getMixedUpperBound(), getMixedStep(),
+      inits, getMapping(), [](OpBuilder &, Location, ValueRange) {});
+
+  // Move the region of the current block to the newly created op.
+  Block *newLoopBody = newLoop.getBody();
+  rewriter.mergeBlocks(
+      getBody(), newLoopBody,
+      newLoopBody->getArguments().take_front(getBody()->getNumArguments()));
+
+  // Update the terminator.
+  {
+    OpBuilder::InsertionGuard g(rewriter);
+    auto terminator = cast<scf::InParallelOp>(newLoopBody->getTerminator());
+    rewriter.setInsertionPointToEnd(terminator.getBody());
+    newYieldValueFn(
----------------
matthias-springer wrote:

I couldn't find the place in the code, but I think this lambda creates the `parallel_insert_slice` ops. It probably returns an empty `SmallVector`. This does not fit with the comment in `LoopLikeInterface.h`:

```c++
/// A function that returns the additional yielded values during
/// `replaceWithAdditionalYields`. `newBbArgs` are the newly added region
/// iter_args. This function should return as many values as there are block
/// arguments in `newBbArgs`.
using NewYieldValuesFn = std::function<SmallVector<Value>(
    OpBuilder &b, Location loc, ArrayRef<BlockArgument> newBbArgs)>;
```

It is a bit confusing that for `scf.for` we expect the function to return a list of `Value` and for `scf.forall` we expect the function to build new IR and not return anything.

How about we generalize `NewYieldValuesFn` to `RewriteTerminatorFn = std::function<void(RewriterBase &rewriter, Operation *terminator, ArrayRef<BlockArgument> newBbArgs)>`?


https://github.com/llvm/llvm-project/pull/77874