[Mlir-commits] [mlir] [mlir][SCF] Skip dynamic peeling for non-index loops (PR #217909)

[purnima shrivastava]

https://github.com/purnima-nlp created https://github.com/llvm/llvm-project/pull/217909

`scf-for-loop-peeling` constructs an `affine.apply` on its dynamic path to compute the split bound. However, `affine.apply` requires `index` operands, while `scf.for` also permits signless integer induction variables and bounds.

Skip the dynamic peeling path when the loop induction variable is not `index`. The existing constant-bounds path remains unchanged, so constant integer loops can still be peeled.

Add a regression test for a loop with dynamic `i32` bounds, verifying that the pass leaves it unchanged rather than producing invalid IR.

to fix : #216631


>From 0d2c1ab5eef5585123e72a3a0243465b396c435b Mon Sep 17 00:00:00 2001
From: Purnima Shrivastava <purnimashrivastava05 at .com>
Date: Fri, 21 Aug 2026 18:22:30 +0530
Subject: [PATCH] [mlir][SCF] Skip dynamic peeling for non-index loops

---
 .../SCF/Transforms/LoopSpecialization.cpp     |  7 ++++++
 mlir/test/Dialect/SCF/for-loop-peeling.mlir   | 22 +++++++++++++++++++
 2 files changed, 29 insertions(+)

diff --git a/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp b/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp
index a39e5520a144b..ba5375bb42d94 100644
--- a/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp
+++ b/mlir/lib/Dialect/SCF/Transforms/LoopSpecialization.cpp
@@ -136,6 +136,13 @@ static LogicalResult peelForLoop(RewriterBase &b, ForOp forOp,
   // Fast path: lb, ub and step are constants.
   if (lbInt && ubInt && stepInt && (*ubInt - *lbInt) % *stepInt == 0)
     return failure();
+
+  // Only the dynamic path computes the peeling bound with affine.apply, which
+  // accepts only index operands.
+  if ((!lbInt || !ubInt || !stepInt) &&
+      !forOp.getInductionVar().getType().isIndex())
+    return failure();
+
   // Slow path: Examine the ops that define lb, ub and step.
   AffineExpr sym0, sym1, sym2;
   bindSymbols(b.getContext(), sym0, sym1, sym2);
diff --git a/mlir/test/Dialect/SCF/for-loop-peeling.mlir b/mlir/test/Dialect/SCF/for-loop-peeling.mlir
index 4337f3b8d4d6f..d1026f5090cb0 100644
--- a/mlir/test/Dialect/SCF/for-loop-peeling.mlir
+++ b/mlir/test/Dialect/SCF/for-loop-peeling.mlir
@@ -377,3 +377,25 @@ func.func @zero_step(%arg0: memref<i64>) {
   }
   return
 }
+
+// -----
+
+// CHECK-LABEL:   func.func @non_index_loop_bounds(
+// CHECK-SAME:                                         %[[INIT:.*]]: i32,
+// CHECK-SAME:                                         %[[N:.*]]: i32) -> i32 {
+// CHECK:           %[[C0:.*]] = arith.constant 0 : i32
+// CHECK:           %[[C2:.*]] = arith.constant 2 : i32
+// CHECK:           %[[RESULT:.*]] = scf.for %[[IV:.*]] = %[[C0]] to %[[N]] step %[[C2]] iter_args(%[[ACC:.*]] = %[[INIT]]) -> (i32) : i32 {
+// CHECK:             %[[ADD:.*]] = arith.addi %[[ACC]], %[[IV]] : i32
+// CHECK:             scf.yield %[[ADD]] : i32
+// CHECK:           }
+// CHECK:           return %[[RESULT]] : i32
+func.func @non_index_loop_bounds(%init: i32, %n: i32) -> i32 {
+  %c0 = arith.constant 0 : i32
+  %c2 = arith.constant 2 : i32
+  %r = scf.for %i = %c0 to %n step %c2 iter_args(%a = %init) -> (i32) : i32 {
+    %t = arith.addi %a, %i : i32
+    scf.yield %t : i32
+  }
+  return %r : i32
+}