[Mlir-commits] [mlir] [mlir][scf] Fix unrolling when the yielded value is defined above the… (PR #122177)

llvmlistbot at llvm.org llvmlistbot at llvm.org
Wed Jan 8 14:06:30 PST 2025


llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-mlir-scf

Author: Alexander Belyaev (pifon2a)

<details>
<summary>Changes</summary>

… loop.

---
Full diff: https://github.com/llvm/llvm-project/pull/122177.diff


2 Files Affected:

- (modified) mlir/lib/Dialect/SCF/Utils/Utils.cpp (+1-1) 
- (modified) mlir/test/Dialect/SCF/loop-unroll.mlir (+28) 


``````````diff
diff --git a/mlir/lib/Dialect/SCF/Utils/Utils.cpp b/mlir/lib/Dialect/SCF/Utils/Utils.cpp
index 6cda7100fe073d..fa82bcb816a2a2 100644
--- a/mlir/lib/Dialect/SCF/Utils/Utils.cpp
+++ b/mlir/lib/Dialect/SCF/Utils/Utils.cpp
@@ -361,7 +361,7 @@ static void generateUnrolledLoop(
 
     // Update yielded values.
     for (unsigned i = 0, e = lastYielded.size(); i < e; i++)
-      lastYielded[i] = operandMap.lookup(yieldedValues[i]);
+      lastYielded[i] = operandMap.lookupOrDefault(yieldedValues[i]);
   }
 
   // Make sure we annotate the Ops in the original body. We do this last so that
diff --git a/mlir/test/Dialect/SCF/loop-unroll.mlir b/mlir/test/Dialect/SCF/loop-unroll.mlir
index 68a11fb6a72c64..329ca46efdc4a5 100644
--- a/mlir/test/Dialect/SCF/loop-unroll.mlir
+++ b/mlir/test/Dialect/SCF/loop-unroll.mlir
@@ -489,3 +489,31 @@ func.func @static_loop_unroll_with_integer_iv() -> (f32, f32) {
 //  UNROLL-BY-3-NEXT:     scf.yield %[[EADD]], %[[EMUL]] : f32, f32
 //  UNROLL-BY-3-NEXT:   }
 //  UNROLL-BY-3-NEXT:   return %[[EFOR]]#0, %[[EFOR]]#1 : f32, f32
+
+// -----
+
+// Test loop unrolling when the yielded value is defined above the loop.
+func.func @loop_unroll_static_yield_value_defined_above(%init: i32) {
+  %c42 = arith.constant 42 : i32
+  %c0 = arith.constant 0 : index
+  %c1 = arith.constant 1 : index
+  %c4 = arith.constant 4 : index
+  %103:2 = scf.for %i = %c0 to %c4 step %c1
+      iter_args(%iter1 = %c42, %iter2 = %init) -> (i32, i32) {
+    %0 = arith.andi %iter2, %iter1 : i32
+    scf.yield %0, %init : i32, i32
+  }
+  return
+}
+// UNROLL-OUTER-BY-2-LABEL: @loop_unroll_static_yield_value_defined_above(
+// UNROLL-OUTER-BY-2-SAME:    %[[INIT:.*]]: i32) {
+// UNROLL-OUTER-BY-2-DAG:   %[[C42:.*]] = arith.constant 42 : i32
+// UNROLL-OUTER-BY-2-DAG:   %[[C0:.*]] = arith.constant 0 : index
+// UNROLL-OUTER-BY-2-DAG:   %[[C4:.*]] = arith.constant 4 : index
+// UNROLL-OUTER-BY-2-DAG:   %[[C2:.*]] = arith.constant 2 : index
+// UNROLL-OUTER-BY-2:       scf.for %{{.*}} = %[[C0]] to %[[C4]] step %[[C2]]
+// UNROLL-OUTER-BY-2-SAME:      iter_args(%[[ITER1:.*]] = %[[C42]],
+// UNROLL-OUTER-BY-2-SAME:                %[[ITER2:.*]] = %[[INIT]])
+// UNROLL-OUTER-BY-2:         %[[SUM:.*]] = arith.andi %[[ITER2]], %[[ITER1]]
+// UNROLL-OUTER-BY-2:         %[[SUM1:.*]] = arith.andi %[[INIT]], %[[SUM]]
+// UNROLL-OUTER-BY-2:         scf.yield %[[SUM1]], %[[INIT]] : i32, i32
\ No newline at end of file

``````````

</details>


https://github.com/llvm/llvm-project/pull/122177


More information about the Mlir-commits mailing list