diff --git a/flang/include/flang/Optimizer/Builder/HLFIRTools.h b/flang/include/flang/Optimizer/Builder/HLFIRTools.h
--- a/flang/include/flang/Optimizer/Builder/HLFIRTools.h
+++ b/flang/include/flang/Optimizer/Builder/HLFIRTools.h
@@ -385,11 +385,14 @@
 };
 
 /// Generate a fir.do_loop nest looping from 1 to extents[i].
+/// \p isUnordered specifies whether the loops in the loop nest
+/// are unordered.
 LoopNest genLoopNest(mlir::Location loc, fir::FirOpBuilder &builder,
-                     mlir::ValueRange extents);
+                     mlir::ValueRange extents, bool isUnordered = false);
 inline LoopNest genLoopNest(mlir::Location loc, fir::FirOpBuilder &builder,
-                            mlir::Value shape) {
-  return genLoopNest(loc, builder, getIndexExtents(loc, builder, shape));
+                            mlir::Value shape, bool isUnordered = false) {
+  return genLoopNest(loc, builder, getIndexExtents(loc, builder, shape),
+                     isUnordered);
 }
 
 /// Inline the body of an hlfir.elemental at the current insertion point
diff --git a/flang/lib/Lower/ConvertCall.cpp b/flang/lib/Lower/ConvertCall.cpp
--- a/flang/lib/Lower/ConvertCall.cpp
+++ b/flang/lib/Lower/ConvertCall.cpp
@@ -1593,13 +1593,12 @@
     }
     assert(shape &&
            "elemental array calls must have at least one array arguments");
-    if (mustBeOrdered)
-      TODO(loc, "ordered elemental calls in HLFIR");
     // Push a new local scope so that any temps made inside the elemental
     // iterations are cleaned up inside the iterations.
     if (!callContext.resultType) {
       // Subroutine case. Generate call inside loop nest.
-      hlfir::LoopNest loopNest = hlfir::genLoopNest(loc, builder, shape);
+      hlfir::LoopNest loopNest =
+          hlfir::genLoopNest(loc, builder, shape, !mustBeOrdered);
       mlir::ValueRange oneBasedIndices = loopNest.oneBasedIndices;
       auto insPt = builder.saveInsertionPoint();
       builder.setInsertionPointToStart(loopNest.innerLoop.getBody());
@@ -1613,6 +1612,8 @@
       return std::nullopt;
     }
     // Function case: generate call inside hlfir.elemental
+    if (mustBeOrdered)
+      TODO(loc, "ordered elemental calls in HLFIR");
     mlir::Type elementType =
         hlfir::getFortranElementType(*callContext.resultType);
     // Get result length parameters.
diff --git a/flang/lib/Optimizer/Builder/HLFIRTools.cpp b/flang/lib/Optimizer/Builder/HLFIRTools.cpp
--- a/flang/lib/Optimizer/Builder/HLFIRTools.cpp
+++ b/flang/lib/Optimizer/Builder/HLFIRTools.cpp
@@ -807,7 +807,7 @@
 
 hlfir::LoopNest hlfir::genLoopNest(mlir::Location loc,
                                    fir::FirOpBuilder &builder,
-                                   mlir::ValueRange extents) {
+                                   mlir::ValueRange extents, bool isUnordered) {
   hlfir::LoopNest loopNest;
   assert(!extents.empty() && "must have at least one extent");
   auto insPt = builder.saveInsertionPoint();
@@ -818,7 +818,8 @@
   unsigned dim = extents.size() - 1;
   for (auto extent : llvm::reverse(extents)) {
     auto ub = builder.createConvert(loc, indexType, extent);
-    loopNest.innerLoop = builder.create<fir::DoLoopOp>(loc, one, ub, one);
+    loopNest.innerLoop =
+        builder.create<fir::DoLoopOp>(loc, one, ub, one, isUnordered);
     builder.setInsertionPointToStart(loopNest.innerLoop.getBody());
     // Reverse the indices so they are in column-major order.
     loopNest.oneBasedIndices[dim--] = loopNest.innerLoop.getInductionVar();
diff --git a/flang/test/HLFIR/extents-of-shape-of.f90 b/flang/test/HLFIR/extents-of-shape-of.f90
--- a/flang/test/HLFIR/extents-of-shape-of.f90
+++ b/flang/test/HLFIR/extents-of-shape-of.f90
@@ -20,8 +20,8 @@
 ! CHECK-HLFIR-NEXT:    %[[EXT0:.*]] = arith.constant 2 : index
 ! CHECK-HLFIR-NEXT:    %[[EXT1:.*]] = hlfir.get_extent %[[SHAPE]] {dim = 1 : index} : (!fir.shape<2>) -> index
 ! CHECK-HLFIR-NEXT:    %[[C1:.*]] = arith.constant 1 : index
-! CHECK-HLFIR-NEXT:    fir.do_loop %[[ARG2:.*]] = %[[C1]] to %[[EXT1]] step %[[C1]] {
-! CHECK-HLFIR-NEXT:      fir.do_loop %[[ARG3:.*]] = %[[C1]] to %[[EXT0]] step %[[C1]] {
+! CHECK-HLFIR-NEXT:    fir.do_loop %[[ARG2:.*]] = %[[C1]] to %[[EXT1]] step %[[C1]] unordered {
+! CHECK-HLFIR-NEXT:      fir.do_loop %[[ARG3:.*]] = %[[C1]] to %[[EXT0]] step %[[C1]] unordered {
 ! CHECK-HLFIR-NEXT:        %[[ELE:.*]] = hlfir.apply %[[MUL]], %[[ARG3]], %[[ARG2]] : (!hlfir.expr<2x?xf32>, index, index) -> f32
 ! CHECK-HLFIR-NEXT:        %[[ASSOC:.*]]:3 = hlfir.associate %[[ELE]] {uniq_name = "adapt.valuebyref"} : (f32) -> (!fir.ref<f32>, !fir.ref<f32>, i1)
 ! CHECK-HLFIR-NEXT:        fir.call
@@ -41,8 +41,8 @@
 ! CHECK-FIR:           %[[SHAPE:.*]] = fir.shape %[[DIMS0]]#1, %[[DIMS1]]#1
 ! CHECK-FIR-NEXT:      %[[C2:.*]] = arith.constant 2 : index
 ! CHECK-FIR-NEXT:      %[[C1_1:.*]] = arith.constant 1 : index
-! CHECK-FIR-NEXT:      fir.do_loop %[[ARG2:.*]] = %[[C1_1]] to %[[DIMS1]]#1 step %[[C1_1]] {
-! CHECK-FIR-NEXT:        fir.do_loop %[[ARG3:.*]] = %[[C1_1]] to %[[C2]] step %[[C1_1]] {
+! CHECK-FIR-NEXT:      fir.do_loop %[[ARG2:.*]] = %[[C1_1]] to %[[DIMS1]]#1 step %[[C1_1]] unordered {
+! CHECK-FIR-NEXT:        fir.do_loop %[[ARG3:.*]] = %[[C1_1]] to %[[C2]] step %[[C1_1]] unordered {
 ! ...
 
 ! CHECK-ALL:           return
diff --git a/flang/test/Lower/HLFIR/calls-optional.f90 b/flang/test/Lower/HLFIR/calls-optional.f90
--- a/flang/test/Lower/HLFIR/calls-optional.f90
+++ b/flang/test/Lower/HLFIR/calls-optional.f90
@@ -77,7 +77,7 @@
 ! CHECK:  %[[VAL_10:.*]]:3 = fir.box_dims %[[VAL_2]]#0, %[[VAL_9]] : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
 ! CHECK:  %[[VAL_11:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>
 ! CHECK:  %[[VAL_12:.*]] = arith.constant 1 : index
-! CHECK:  fir.do_loop %[[VAL_13:.*]] = %[[VAL_12]] to %[[VAL_10]]#1 step %[[VAL_12]] {
+! CHECK:  fir.do_loop %[[VAL_13:.*]] = %[[VAL_12]] to %[[VAL_10]]#1 step %[[VAL_12]] unordered {
 ! CHECK:    %[[VAL_14:.*]] = hlfir.designate %[[VAL_2]]#0 (%[[VAL_13]])  : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
 ! CHECK:    %[[VAL_15:.*]] = fir.if %[[VAL_8]] -> (!fir.ref<f32>) {
 ! CHECK:      %[[VAL_16:.*]] = arith.constant 0 : index
@@ -109,7 +109,7 @@
 ! CHECK:  %[[VAL_2:.*]] = arith.constant 0 : index
 ! CHECK:  %[[VAL_3:.*]]:3 = fir.box_dims %[[VAL_1]]#0, %[[VAL_2]] : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
 ! CHECK:  %[[VAL_4:.*]] = arith.constant 1 : index
-! CHECK:  fir.do_loop %[[VAL_5:.*]] = %[[VAL_4]] to %[[VAL_3]]#1 step %[[VAL_4]] {
+! CHECK:  fir.do_loop %[[VAL_5:.*]] = %[[VAL_4]] to %[[VAL_3]]#1 step %[[VAL_4]] unordered {
 ! CHECK:    %[[VAL_6:.*]] = hlfir.designate %[[VAL_1]]#0 (%[[VAL_5]])  : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
 ! CHECK:    fir.call @_QPelem_takes_one_optional(%[[VAL_6]]) {{.*}} : (!fir.ref<f32>) -> ()
 ! CHECK:  }
@@ -131,7 +131,7 @@
 ! CHECK:  %[[VAL_5:.*]] = arith.constant 0 : index
 ! CHECK:  %[[VAL_6:.*]]:3 = fir.box_dims %[[VAL_2]]#0, %[[VAL_5]] : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
 ! CHECK:  %[[VAL_7:.*]] = arith.constant 1 : index
-! CHECK:  fir.do_loop %[[VAL_8:.*]] = %[[VAL_7]] to %[[VAL_6]]#1 step %[[VAL_7]] {
+! CHECK:  fir.do_loop %[[VAL_8:.*]] = %[[VAL_7]] to %[[VAL_6]]#1 step %[[VAL_7]] unordered {
 ! CHECK:    %[[VAL_9:.*]] = hlfir.designate %[[VAL_2]]#0 (%[[VAL_8]])  : (!fir.box<!fir.array<?xf32>>, index) -> !fir.ref<f32>
 ! CHECK:    %[[VAL_10:.*]] = fir.embox %[[VAL_9]] : (!fir.ref<f32>) -> !fir.box<f32>
 ! CHECK:    %[[VAL_11:.*]] = fir.rebox %[[VAL_10]] : (!fir.box<f32>) -> !fir.class<none>
diff --git a/flang/test/Lower/HLFIR/elemental-user-procedure-ref.f90 b/flang/test/Lower/HLFIR/elemental-user-procedure-ref.f90
--- a/flang/test/Lower/HLFIR/elemental-user-procedure-ref.f90
+++ b/flang/test/Lower/HLFIR/elemental-user-procedure-ref.f90
@@ -90,9 +90,59 @@
 ! CHECK:  %[[VAL_4:.*]] = arith.constant 20 : index
 ! CHECK:  %[[VAL_6:.*]]:2 = hlfir.declare %[[VAL_1:.*]](%[[VAL_5:[^)]*]]) {{.*}}y
 ! CHECK:  %[[VAL_7:.*]] = arith.constant 1 : index
-! CHECK:  fir.do_loop %[[VAL_8:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_7]] {
-! CHECK:    fir.do_loop %[[VAL_9:.*]] = %[[VAL_7]] to %[[VAL_3]] step %[[VAL_7]] {
+! CHECK:  fir.do_loop %[[VAL_8:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_7]] unordered {
+! CHECK:    fir.do_loop %[[VAL_9:.*]] = %[[VAL_7]] to %[[VAL_3]] step %[[VAL_7]] unordered {
 ! CHECK:      %[[VAL_10:.*]] = hlfir.designate %[[VAL_6]]#0 (%[[VAL_9]], %[[VAL_8]])  : (!fir.ref<!fir.array<10x20xf32>>, index, index) -> !fir.ref<f32>
 ! CHECK:      fir.call @_QPelem_sub(%[[VAL_2]]#1, %[[VAL_10]]) fastmath<contract> : (!fir.ref<i32>, !fir.ref<f32>) -> ()
 ! CHECK:    }
 ! CHECK:  }
+
+subroutine impure_elemental(x)
+  real :: x(10, 20)
+  interface
+    impure elemental subroutine impure_elem(a)
+      real, intent(in) :: a
+    end subroutine
+  end interface
+  call impure_elem(x)
+end subroutine
+! CHECK-LABEL:   func.func @_QPimpure_elemental(
+! CHECK-SAME:      %[[VAL_0:.*]]: !fir.ref<!fir.array<10x20xf32>> {fir.bindc_name = "x"}) {
+! CHECK:           %[[VAL_1:.*]] = arith.constant 10 : index
+! CHECK:           %[[VAL_2:.*]] = arith.constant 20 : index
+! CHECK:           %[[VAL_3:.*]] = fir.shape %[[VAL_1]], %[[VAL_2]] : (index, index) -> !fir.shape<2>
+! CHECK:           %[[VAL_4:.*]]:2 = hlfir.declare %[[VAL_0]](%[[VAL_3]]) {uniq_name = "_QFimpure_elementalEx"} : (!fir.ref<!fir.array<10x20xf32>>, !fir.shape<2>) -> (!fir.ref<!fir.array<10x20xf32>>, !fir.ref<!fir.array<10x20xf32>>)
+! CHECK:           %[[VAL_5:.*]] = arith.constant 1 : index
+! CHECK:           fir.do_loop %[[VAL_6:.*]] = %[[VAL_5]] to %[[VAL_2]] step %[[VAL_5]] {
+! CHECK:             fir.do_loop %[[VAL_7:.*]] = %[[VAL_5]] to %[[VAL_1]] step %[[VAL_5]] {
+! CHECK:               %[[VAL_8:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_7]], %[[VAL_6]])  : (!fir.ref<!fir.array<10x20xf32>>, index, index) -> !fir.ref<f32>
+! CHECK:               fir.call @_QPimpure_elem(%[[VAL_8]]) fastmath<contract> : (!fir.ref<f32>) -> ()
+! CHECK:             }
+! CHECK:           }
+! CHECK:           return
+! CHECK:         }
+
+subroutine ordered_elemental(x)
+  real :: x(10, 20)
+  interface
+    elemental subroutine ordered_elem(a)
+      real, intent(inout) :: a
+    end subroutine
+  end interface
+  call ordered_elem(x)
+end subroutine
+! CHECK-LABEL:   func.func @_QPordered_elemental(
+! CHECK-SAME:      %[[VAL_0:.*]]: !fir.ref<!fir.array<10x20xf32>> {fir.bindc_name = "x"}) {
+! CHECK:           %[[VAL_1:.*]] = arith.constant 10 : index
+! CHECK:           %[[VAL_2:.*]] = arith.constant 20 : index
+! CHECK:           %[[VAL_3:.*]] = fir.shape %[[VAL_1]], %[[VAL_2]] : (index, index) -> !fir.shape<2>
+! CHECK:           %[[VAL_4:.*]]:2 = hlfir.declare %[[VAL_0]](%[[VAL_3]]) {uniq_name = "_QFordered_elementalEx"} : (!fir.ref<!fir.array<10x20xf32>>, !fir.shape<2>) -> (!fir.ref<!fir.array<10x20xf32>>, !fir.ref<!fir.array<10x20xf32>>)
+! CHECK:           %[[VAL_5:.*]] = arith.constant 1 : index
+! CHECK:           fir.do_loop %[[VAL_6:.*]] = %[[VAL_5]] to %[[VAL_2]] step %[[VAL_5]] {
+! CHECK:             fir.do_loop %[[VAL_7:.*]] = %[[VAL_5]] to %[[VAL_1]] step %[[VAL_5]] {
+! CHECK:               %[[VAL_8:.*]] = hlfir.designate %[[VAL_4]]#0 (%[[VAL_7]], %[[VAL_6]])  : (!fir.ref<!fir.array<10x20xf32>>, index, index) -> !fir.ref<f32>
+! CHECK:               fir.call @_QPordered_elem(%[[VAL_8]]) fastmath<contract> : (!fir.ref<f32>) -> ()
+! CHECK:             }
+! CHECK:           }
+! CHECK:           return
+! CHECK:         }