diff --git a/mlir/lib/Dialect/Affine/IR/AffineOps.cpp b/mlir/lib/Dialect/Affine/IR/AffineOps.cpp
--- a/mlir/lib/Dialect/Affine/IR/AffineOps.cpp
+++ b/mlir/lib/Dialect/Affine/IR/AffineOps.cpp
@@ -2482,9 +2482,12 @@
                                 SmallVectorImpl<OpFoldResult> &results) {
   bool folded = succeeded(foldLoopBounds(*this));
   folded |= succeeded(canonicalizeLoopBounds(*this));
-  if (hasTrivialZeroTripCount(*this)) {
+  if (hasTrivialZeroTripCount(*this) && getNumResults() != 0) {
     // The initial values of the loop-carried variables (iter_args) are the
-    // results of the op.
+    // results of the op. But this must be avoided for an affine.for op that
+    // does not return any results. Since ops that do not return results cannot
+    // be folded away, we would enter an infinite loop of folds on the same
+    // affine.for op.
     results.assign(getIterOperands().begin(), getIterOperands().end());
     folded = true;
   }
diff --git a/mlir/test/Dialect/Affine/canonicalize.mlir b/mlir/test/Dialect/Affine/canonicalize.mlir
--- a/mlir/test/Dialect/Affine/canonicalize.mlir
+++ b/mlir/test/Dialect/Affine/canonicalize.mlir
@@ -582,14 +582,23 @@
 
 // -----
 
+// CHECK-LABEL:  func @zero_iter_loop_not_folded
+func.func @zero_iter_loop_not_folded() {
+  %A = memref.alloc() : memref<4xf32>
+  affine.for %i = 0 to 0 {
+      %load = affine.load %A[%i] : memref<4xf32>
+      affine.store %load, %A[%i] : memref<4xf32>
+  }
+  // CHECK:   affine.for {{.*}} = 0 to 0 {
+  return
+}
+
+// -----
+
 // CHECK-LABEL:  func @fold_zero_iter_loops
 // CHECK-SAME: %[[ARG:.*]]: index
 func.func @fold_zero_iter_loops(%in : index) -> index {
   %c1 = arith.constant 1 : index
-  affine.for %i = 0 to 0 {
-    affine.for %j = 0 to -1 {
-    }
-  }
   %res = affine.for %i = 0 to 0 iter_args(%loop_arg = %in) -> index {
     %yield = arith.addi %loop_arg, %c1 : index
     affine.yield %yield : index