[mlir][SCF] Fix invalid IR in ParallelOpSingleOrZeroIterationDimsFolder pattern

[matthias-springer]

ParallelOpSingleOrZeroIterationDimsFolder used to produce invalid IR:

within split at mlir/test/Dialect/SCF/canonicalize.mlir:1 offset :11:3: error: 'scf.parallel' op expects region #0 to have 0 or 1 blocks
  scf.parallel (%i0, %i1, %i2) = (%c0, %c3, %c7) to (%c1, %c6, %c10) step (%c1, %c2, %c3) {
  ^
within split at mlir/test/Dialect/SCF/canonicalize.mlir:1 offset :11:3: note: see current operation:
"scf.parallel"(%4, %5, %3) <{operandSegmentSizes = array<i32: 1, 1, 1, 0>}> ({
^bb0(%arg1: index):
  "memref.store"(%0, %arg0, %1, %arg1, %6) : (i32, memref<?x?x?xi32>, index, index, index) -> ()
  "scf.yield"() : () -> ()
^bb1(%8: index):  // no predecessors
  "scf.yield"() : () -> ()
}) : (index, index, index) -> ()

Together with #74551, this commit fixes mlir/test/Dialect/SCF/canonicalize.mlir when verifying the IR after each pattern application (#74270).

[llvmbot]

@llvm/pr-subscribers-mlir

Author: Matthias Springer (matthias-springer)

Changes

ParallelOpSingleOrZeroIterationDimsFolder used to produce invalid IR:

within split at mlir/test/Dialect/SCF/canonicalize.mlir:1 offset :11:3: error: 'scf.parallel' op expects region #<!-- -->0 to have 0 or 1 blocks
  scf.parallel (%i0, %i1, %i2) = (%c0, %c3, %c7) to (%c1, %c6, %c10) step (%c1, %c2, %c3) {
  ^
within split at mlir/test/Dialect/SCF/canonicalize.mlir:1 offset :11:3: note: see current operation:
"scf.parallel"(%4, %5, %3) &lt;{operandSegmentSizes = array&lt;i32: 1, 1, 1, 0&gt;}&gt; ({
^bb0(%arg1: index):
  "memref.store"(%0, %arg0, %1, %arg1, %6) : (i32, memref&lt;?x?x?xi32&gt;, index, index, index) -&gt; ()
  "scf.yield"() : () -&gt; ()
^bb1(%8: index):  // no predecessors
  "scf.yield"() : () -&gt; ()
}) : (index, index, index) -&gt; ()

Together with #74551, this commit fixes mlir/test/Dialect/SCF/canonicalize.mlir when verifying the IR after each pattern application (#74270).

---

Full diff: https://github.com/llvm/llvm-project/pull/74552.diff

1 Files Affected:

• (modified) mlir/lib/Dialect/SCF/IR/SCF.cpp (+2)

diff --git a/mlir/lib/Dialect/SCF/IR/SCF.cpp b/mlir/lib/Dialect/SCF/IR/SCF.cpp
index 3b55704c4ea07..2e7382160f8eb 100644
--- a/mlir/lib/Dialect/SCF/IR/SCF.cpp
+++ b/mlir/lib/Dialect/SCF/IR/SCF.cpp
@@ -3040,6 +3040,8 @@ struct ParallelOpSingleOrZeroIterationDimsFolder
     auto newOp =
         rewriter.create<ParallelOp>(op.getLoc(), newLowerBounds, newUpperBounds,
                                     newSteps, op.getInitVals(), nullptr);
+    // Erase the empty block that was inserted by the builder.
+    rewriter.eraseBlock(newOp.getBody());
     // Clone the loop body and remap the block arguments of the collapsed loops
     // (inlining does not support a cancellable block argument mapping).
     rewriter.cloneRegionBefore(op.getRegion(), newOp.getRegion(),

[llvmbot]

@llvm/pr-subscribers-mlir-scf

Author: Matthias Springer (matthias-springer)

Changes

ParallelOpSingleOrZeroIterationDimsFolder used to produce invalid IR:

within split at mlir/test/Dialect/SCF/canonicalize.mlir:1 offset :11:3: error: 'scf.parallel' op expects region #<!-- -->0 to have 0 or 1 blocks
  scf.parallel (%i0, %i1, %i2) = (%c0, %c3, %c7) to (%c1, %c6, %c10) step (%c1, %c2, %c3) {
  ^
within split at mlir/test/Dialect/SCF/canonicalize.mlir:1 offset :11:3: note: see current operation:
"scf.parallel"(%4, %5, %3) &lt;{operandSegmentSizes = array&lt;i32: 1, 1, 1, 0&gt;}&gt; ({
^bb0(%arg1: index):
  "memref.store"(%0, %arg0, %1, %arg1, %6) : (i32, memref&lt;?x?x?xi32&gt;, index, index, index) -&gt; ()
  "scf.yield"() : () -&gt; ()
^bb1(%8: index):  // no predecessors
  "scf.yield"() : () -&gt; ()
}) : (index, index, index) -&gt; ()

Together with #74551, this commit fixes mlir/test/Dialect/SCF/canonicalize.mlir when verifying the IR after each pattern application (#74270).

---

Full diff: https://github.com/llvm/llvm-project/pull/74552.diff

1 Files Affected:

• (modified) mlir/lib/Dialect/SCF/IR/SCF.cpp (+2)

diff --git a/mlir/lib/Dialect/SCF/IR/SCF.cpp b/mlir/lib/Dialect/SCF/IR/SCF.cpp
index 3b55704c4ea07..2e7382160f8eb 100644
--- a/mlir/lib/Dialect/SCF/IR/SCF.cpp
+++ b/mlir/lib/Dialect/SCF/IR/SCF.cpp
@@ -3040,6 +3040,8 @@ struct ParallelOpSingleOrZeroIterationDimsFolder
     auto newOp =
         rewriter.create<ParallelOp>(op.getLoc(), newLowerBounds, newUpperBounds,
                                     newSteps, op.getInitVals(), nullptr);
+    // Erase the empty block that was inserted by the builder.
+    rewriter.eraseBlock(newOp.getBody());
     // Clone the loop body and remap the block arguments of the collapsed loops
     // (inlining does not support a cancellable block argument mapping).
     rewriter.cloneRegionBefore(op.getRegion(), newOp.getRegion(),