[flang] Avoid copyin/copyout if the actual argument is contiguous at runtime

[clementval]

Summary

This patch adds contiguity check with the runtime to avoid copyin/copyout
in case the actual argument is actually contiguous.

[clementval]

flang/lib/Lower/ConvertExpr.cpp

2852

Currently a single call to IsContiguous is made.

call runtime
if present
  if contiguous
    // no copy
  else
    copy
end


if present && !contiguous
  copyout

Would it make sense to nest it in the first if so it is not performed when an arg is not present or so on? It would require an extra call when the first condition is true but no call when it is false.

if present
  call runtime
  if contiguous
    no copy
  else
    copy
  end
end

if present
  call runtime
  if !contiguous
    copyout
  end
end

[jeanPerier]

Logic looks good to me. One suggestion to maybe simplify the copy-out lowering code a bit.

flang/lib/Lower/ConvertExpr.cpp

2819

I do not think a new value here, you can still use "restrictCopyAndFreeAtRuntime" and leave the copy-out code unmodified if you compute restrictCopyAndFreeAtRuntime = isContiguous || restrictCopyAndFreeAtRuntime before building the CopyOut pairs.

[jeanPerier]

flang/lib/Lower/ConvertExpr.cpp

2852

The runtime should not spend too much time on arguments that are not present since a nullptr descriptor should be passed. I am not sure doing the runtime call twice in the other cases wouldn't be a worse drawback.
So I think it is best to keep it the way you did it.

[clementval]

Get rid of extra value in the CopyOutPair.

[jeanPerier]

LGTM

flang/lib/Lower/ConvertExpr.cpp

 #include "flang/Optimizer/Builder/Runtime/Character.h"
 #include "flang/Optimizer/Builder/Runtime/RTBuilder.h"
 #include "flang/Optimizer/Builder/Runtime/Ragged.h"
 #include "flang/Optimizer/Builder/Todo.h"
 #include "flang/Optimizer/Dialect/FIRAttr.h"
 #include "flang/Optimizer/Dialect/FIRDialect.h"
 #include "flang/Optimizer/Dialect/FIROpsSupport.h"
 #include "flang/Optimizer/Support/FatalError.h"
+#include "flang/Runtime/support.h"
 #include "flang/Semantics/expression.h"
 #include "flang/Semantics/symbol.h"
 #include "flang/Semantics/tools.h"
 #include "flang/Semantics/type.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "llvm/ADT/TypeSwitch.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 #include <algorithm>
 
 #define DEBUG_TYPE "flang-lower-expr"
 
+using namespace Fortran::runtime;
+
 //===----------------------------------------------------------------------===//
 // The composition and structure of Fortran::evaluate::Expr is defined in
 // the various header files in include/flang/Evaluate. You are referred
 // there for more information on these data structures. Generally speaking,
 // these data structures are a strongly typed family of abstract data types
 // that, composed as trees, describe the syntax of Fortran expressions.
 //
 // This part of the bridge can traverse these tree structures and lower them
struct CopyOutPair {
     // callee, in which case it must be copied-out to the variable.
     bool argMayBeModifiedByCall;
     // Optional boolean value that, if present and false, prevents
     // the copy-out and temp deallocation.
     llvm::Optional<mlir::Value> restrictCopyAndFreeAtRuntime;
   };
   using CopyOutPairs = llvm::SmallVector<CopyOutPair, 4>;
 
   /// Helper to read any fir::BoxValue into other fir::ExtendedValue categories
   /// not based on fir.box.
   /// This will lose any non contiguous stride information and dynamic type and
   /// should only be called if \p exv is known to be contiguous or if its base
   /// address will be replaced by a contiguous one. If \p exv is not a
   /// fir::BoxValue, this is a no-op.
   ExtValue readIfBoxValue(const ExtValue &exv) {
     if (const auto *box = exv.getBoxOf<fir::BoxValue>())
       return fir::factory::readBoxValue(builder, getLoc(), *box);
     return exv;
   }
 
   /// Generate a contiguous temp to pass \p actualArg as argument \p arg. The
   /// creation of the temp and copy-in can be made conditional at runtime by
   /// providing a runtime boolean flag \p restrictCopyAtRuntime (in which case
   /// the temp and copy will only be made if the value is true at runtime).
   ExtValue genCopyIn(const ExtValue &actualArg,
                      const Fortran::lower::CallerInterface::PassedEntity &arg,
                      CopyOutPairs &copyOutPairs,
                      llvm::Optional<mlir::Value> restrictCopyAtRuntime,
                      bool byValue) {
     const bool doCopyOut = !byValue && arg.mayBeModifiedByCall();
     llvm::StringRef tempName = byValue ? ".copy" : ".copyinout";
-    if (!restrictCopyAtRuntime) {
+    mlir::Location loc = getLoc();
+    bool isActualArgBox = fir::isa_box_type(fir::getBase(actualArg).getType());
+    mlir::Value isContiguousResult;
+    mlir::Type addrType = fir::HeapType::get(
+        fir::unwrapPassByRefType(fir::getBase(actualArg).getType()));
+
+    if (isActualArgBox) {
+      // Check at runtime if the argument is contiguous so no copy is needed.
+      mlir::func::FuncOp isContiguousFct =
+          fir::runtime::getRuntimeFunc<mkRTKey(IsContiguous)>(loc, builder);
+      fir::CallOp isContiguous = builder.create<fir::CallOp>(
+          loc, isContiguousFct,
+          mlir::ValueRange{builder.createConvert(
+              loc, isContiguousFct.getFunctionType().getInput(0),
+              fir::getBase(actualArg))});
+      isContiguousResult = isContiguous.getResult(0);
+    }
+
+    auto doCopyIn = [&]() -> ExtValue {
       ExtValue temp = genArrayTempFromMold(actualArg, tempName);
       if (arg.mayBeReadByCall())
         genArrayCopy(temp, actualArg);
+      return temp;
+    };
+
+    auto noCopy = [&]() {
+      mlir::Value box = fir::getBase(actualArg);
+      mlir::Value boxAddr = builder.create<fir::BoxAddrOp>(loc, addrType, box);
+      builder.create<fir::ResultOp>(loc, boxAddr);
+    };
+
+    auto combinedCondition = [&]() {
+      if (isActualArgBox) {
+        mlir::Value zero =
+            builder.createIntegerConstant(loc, builder.getI1Type(), 0);
+        mlir::Value notContiguous = builder.create<mlir::arith::CmpIOp>(
+            loc, mlir::arith::CmpIPredicate::eq, isContiguousResult, zero);
+        if (!restrictCopyAtRuntime) {
+          restrictCopyAtRuntime = notContiguous;
+        } else {
+          mlir::Value cond = builder.create<mlir::arith::AndIOp>(
+              loc, *restrictCopyAtRuntime, notContiguous);
+          restrictCopyAtRuntime = cond;
+        }
+      }
+    };
+
+    if (!restrictCopyAtRuntime) {
+      if (isActualArgBox) {
+        // isContiguousResult = genIsContiguousCall();
+        mlir::Value addr =
+            builder
+                .genIfOp(loc, {addrType}, isContiguousResult,
+                         /*withElseRegion=*/true)
+                .genThen([&]() { noCopy(); })
+                .genElse([&] {
+                  ExtValue temp = doCopyIn();
+                  builder.create<fir::ResultOp>(loc, fir::getBase(temp));
+                })
+                .getResults()[0];
+        fir::ExtendedValue temp =
+            fir::substBase(readIfBoxValue(actualArg), addr);
+        combinedCondition();
         copyOutPairs.emplace_back(
             CopyOutPair{actualArg, temp, doCopyOut, restrictCopyAtRuntime});
         return temp;
       }
+
+      ExtValue temp = doCopyIn();
+      copyOutPairs.emplace_back(CopyOutPair{actualArg, temp, doCopyOut});
+      return temp;
+    }
+
     // Otherwise, need to be careful to only copy-in if allowed at runtime.
-    mlir::Location loc = getLoc();
-    auto addrType = fir::HeapType::get(
-        fir::unwrapPassByRefType(fir::getBase(actualArg).getType()));
     mlir::Value addr =
         builder
             .genIfOp(loc, {addrType}, *restrictCopyAtRuntime,
                      /*withElseRegion=*/true)
             .genThen([&]() {
-              auto temp = genArrayTempFromMold(actualArg, tempName);
-              if (arg.mayBeReadByCall())
-                genArrayCopy(temp, actualArg);
+              if (isActualArgBox) {
+                // isContiguousResult = genIsContiguousCall();
+                // Avoid copyin if the argument is contiguous at runtime.
+                mlir::Value addr1 =
+                    builder
+                        .genIfOp(loc, {addrType}, isContiguousResult,
+                                 /*withElseRegion=*/true)
+                        .genThen([&]() { noCopy(); })
+                        .genElse([&]() {
+                          ExtValue temp = doCopyIn();
+                          builder.create<fir::ResultOp>(loc,
+                                                        fir::getBase(temp));
+                        })
+                        .getResults()[0];
+                builder.create<fir::ResultOp>(loc, addr1);
+              } else {
+                ExtValue temp = doCopyIn();
                 builder.create<fir::ResultOp>(loc, fir::getBase(temp));
+              }
             })
             .genElse([&]() {
-              auto nullPtr = builder.createNullConstant(loc, addrType);
+              mlir::Value nullPtr = builder.createNullConstant(loc, addrType);
               builder.create<fir::ResultOp>(loc, nullPtr);
             })
             .getResults()[0];
-    // Associate the temp address with actualArg lengths and extents.
+    // Associate the temp address with actualArg lengths and extents if a
+    // temporary is generated. Otherwise the same address is associated.
     fir::ExtendedValue temp = fir::substBase(readIfBoxValue(actualArg), addr);
+    combinedCondition();
     copyOutPairs.emplace_back(
         CopyOutPair{actualArg, temp, doCopyOut, restrictCopyAtRuntime});
     return temp;
   }
 
   /// Generate copy-out if needed and free the temporary for an argument that
   /// has been copied-in into a contiguous temp.
   void genCopyOut(const CopyOutPair &copyOutPair) {
     mlir::Location loc = getLoc();
     if (!copyOutPair.restrictCopyAndFreeAtRuntime) {
       if (copyOutPair.argMayBeModifiedByCall)
         genArrayCopy(copyOutPair.var, copyOutPair.temp);
       builder.create<fir::FreeMemOp>(loc, fir::getBase(copyOutPair.temp));
       return;
     }
+
     builder.genIfThen(loc, *copyOutPair.restrictCopyAndFreeAtRuntime)
         .genThen([&]() {
           if (copyOutPair.argMayBeModifiedByCall)
             genArrayCopy(copyOutPair.var, copyOutPair.temp);
           builder.create<fir::FreeMemOp>(loc, fir::getBase(copyOutPair.temp));
         })
         .end();
   }

flang/test/Lower/call-by-value-attr.f90

!CHECK-LABEL: func @_QQmain()
   !CHECK: %[[CONV_5:.*]] = fir.convert %[[CONST_5]] : (i64) -> index
   !CHECK: %[[CONST_1:.*]] = arith.constant 1 : i64
   !CHECK: %[[CONV_1:.*]] = fir.convert %[[CONST_1]] : (i64) -> index
   !CHECK: %[[CONST_15_3:.*]] = arith.constant 15 : i64
   !CHECK: %[[CONV_15:.*]] = fir.convert %[[CONST_15_3]] : (i64) -> index
   !CHECK: %[[SHAPE_7:.*]]  = fir.shape %[[CONST_15_1]] : (index) -> !fir.shape<1>
   !CHECK: %[[SLICE:.*]] = fir.slice %[[CONV_5]], %[[CONV_15]], %[[CONV_1]] : (index, index, index) -> !fir.slice<1>
   !CHECK: %[[BOX:.*]] = fir.embox %[[ARRAY_B]](%[[SHAPE_7]]) [%[[SLICE]]] : (!fir.ref<!fir.array<15xi32>>, !fir.shape<1>, !fir.slice<1>) -> !fir.box<!fir.array<11xi32>>
+  !CHECK: %[[BOX_NONE:.*]] = fir.convert %[[BOX]] : (!fir.box<!fir.array<11xi32>>) -> !fir.box<none>
+  !CHECK: %[[IS_CONTIGUOUS:.*]] = fir.call @_FortranAIsContiguous(%[[BOX_NONE]]) : (!fir.box<none>) -> i1
+  !CHECK: %[[ADDR:.*]] = fir.if %[[IS_CONTIGUOUS]] -> (!fir.heap<!fir.array<11xi32>>) {
+  !CHECK: %[[BOX_ADDR:.*]] = fir.box_addr %[[BOX]] : (!fir.box<!fir.array<11xi32>>) -> !fir.heap<!fir.array<11xi32>>
+  !CHECKL fir.result %[[BOXADDR]] : !fir.heap<!fir.array<11xi32>>
   !CHECK: %[[CONST_0:.*]] = arith.constant 0 : index
   !CHECK: %[[DIMS:.*]]:3 = fir.box_dims %[[BOX]], %[[CONST_0]] : (!fir.box<!fir.array<11xi32>>, index) -> (index, index, index)
   !CHECK: %[[ARRAY_COPY_2:.*]] = fir.allocmem !fir.array<11xi32>, %[[DIMS]]#1 {uniq_name = ".copy"}
   !CHECK: %[[SHAPE_8:.*]] = fir.shape %[[DIMS]]#1 : (index) -> !fir.shape<1>
   !CHECK: %[[ARRAY_LOAD_7:.*]] = fir.array_load %[[ARRAY_COPY_2]](%[[SHAPE_8]]) : (!fir.heap<!fir.array<11xi32>>, !fir.shape<1>) -> !fir.array<11xi32>
   !CHECK: %[[ARRAY_LOAD_8:.*]] = fir.array_load %[[BOX]] : (!fir.box<!fir.array<11xi32>>) -> !fir.array<11xi32>
   !CHECK: %[[DO_4:.*]] = fir.do_loop {{.*}} {
   !CHECK: }
-  !CHECK fir.array_merge_store %[[ARRAY_LOAD_7]], %[[DO_4]] to %[[ARRAY_COPY_2]] : !fir.array<11xi32>, !fir.array<11xi32>, !fir.heap<!fir.array<11xi32>>
-  !CHECK: %[[CONVERT_B:.*]] = fir.convert %[[ARRAY_COPY_2]] : (!fir.heap<!fir.array<11xi32>>) -> !fir.ref<!fir.array<10xi32>>
+  !CHECK: fir.array_merge_store %[[ARRAY_LOAD_7]], %[[DO_4]] to %[[ARRAY_COPY_2]] : !fir.array<11xi32>, !fir.array<11xi32>, !fir.heap<!fir.array<11xi32>>
+  !CHECK: fir.result %[[ARRAY_COPY_2]] : !fir.heap<!fir.array<11xi32>>
+  !CHECK: %[[CONVERT_B:.*]] = fir.convert %[[ADDR]] : (!fir.heap<!fir.array<11xi32>>) -> !fir.ref<!fir.array<10xi32>>
   !CHECK: fir.call @_QPsubra(%[[CONVERT_B]])
   call subra(b(5:15))
 end program call_by_value_attr
 
 
 ! CHECK-LABEL: func @_QPtest_litteral_copies_1
 subroutine test_litteral_copies_1
   ! VALUE arguments  can be modified by the callee, so the static storage of

flang/test/Lower/call-copy-in-out.f90

 ! Test copy-in / copy-out of non-contiguous variable passed as F77 array arguments.
 ! RUN: bbc -emit-fir %s -o - | FileCheck %s
 
 ! Nominal test
 ! CHECK-LABEL: func @_QPtest_assumed_shape_to_array(
 ! CHECK-SAME: %[[x:.*]]: !fir.box<!fir.array<?xf32>>{{.*}}) {
 subroutine test_assumed_shape_to_array(x)
   real :: x(:)
+
+! CHECK: %[[box_none:.*]] = fir.convert %[[x]] : (!fir.box<!fir.array<?xf32>>) -> !fir.box<none>
+! CHECK: %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
+! CHECK: %[[addr:.*]] = fir.if %[[is_contiguous]] -> (!fir.heap<!fir.array<?xf32>>) {
+! CHECK:   %[[box_addr:.*]] = fir.box_addr %[[x]] : (!fir.box<!fir.array<?xf32>>) -> !fir.heap<!fir.array<?xf32>>
+! CHECK:   fir.result %[[box_addr]] : !fir.heap<!fir.array<?xf32>>
+! CHECK: } else {
 ! Creating temp
 ! CHECK:  %[[dim:.*]]:3 = fir.box_dims %[[x:.*]], %c0{{.*}} : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
 ! CHECK:  %[[temp:.*]] = fir.allocmem !fir.array<?xf32>, %[[dim]]#1 {uniq_name = ".copyinout"}
 
 ! Copy-in
 ! CHECK-DAG:  %[[shape:.*]] = fir.shape %[[dim]]#1 : (index) -> !fir.shape<1>
 ! CHECK-DAG:  %[[temp_load:.*]] = fir.array_load %[[temp]](%[[shape]]) : (!fir.heap<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
 ! CHECK-DAG:  %[[x_load:.*]] = fir.array_load %[[x]] : (!fir.box<!fir.array<?xf32>>) -> !fir.array<?xf32>
 ! CHECK:  %[[copyin:.*]] = fir.do_loop %[[i:.*]] = %{{.*}} to %{{.*}} step %{{.*}} iter_args(%[[res:.*]] = %[[temp_load]]) -> (!fir.array<?xf32>) {
 ! CHECK:    %[[fetch:.*]] = fir.array_fetch %[[x_load]], %[[i]] : (!fir.array<?xf32>, index) -> f32
 ! CHECK:    %[[update:.*]] = fir.array_update %[[res]], %[[fetch]], %[[i]] : (!fir.array<?xf32>, f32, index) -> !fir.array<?xf32>
 ! CHECK:    fir.result %[[update]] : !fir.array<?xf32>
 ! CHECK:  }
 ! CHECK:  fir.array_merge_store %[[temp_load]], %[[copyin:.*]] to %[[temp]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.heap<!fir.array<?xf32>>
+! CHECK:  fir.result %[[temp]] : !fir.heap<!fir.array<?xf32>>
 
-! CHECK:  %[[cast:.*]] = fir.convert %[[temp]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<?xf32>>
+! CHECK:  %[[dim:.*]]:3 = fir.box_dims %[[x]], %c0{{.*}} : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
+! CHECK:  %[[cast:.*]] = fir.convert %[[addr]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<?xf32>>
 ! CHECK:  fir.call @_QPbar(%[[cast]]) : (!fir.ref<!fir.array<?xf32>>) -> ()
 
 ! Copy-out
-
 ! CHECK-DAG:  %[[x_load:.*]] = fir.array_load %[[x]] : (!fir.box<!fir.array<?xf32>>) -> !fir.array<?xf32>
+! CHECK-DAG:  %[[c0:.*]] = arith.constant 0 : index
+
 ! CHECK-DAG:  %[[shape:.*]] = fir.shape %[[dim]]#1 : (index) -> !fir.shape<1>
-! CHECK-DAG:  %[[temp_load:.*]] = fir.array_load %[[temp]](%[[shape]]) : (!fir.heap<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
+! CHECK-DAG:  %[[temp_load:.*]] = fir.array_load %[[addr]](%[[shape]]) : (!fir.heap<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
 ! CHECK:  %[[copyout:.*]] = fir.do_loop %[[i:.*]] = %{{.*}} to %{{.*}} step %{{.*}} iter_args(%[[res:.*]] = %[[x_load]]) -> (!fir.array<?xf32>) {
 ! CHECK:    %[[fetch:.*]] = fir.array_fetch %[[temp_load]], %[[i]] : (!fir.array<?xf32>, index) -> f32
 ! CHECK:    %[[update:.*]] = fir.array_update %[[res]], %[[fetch]], %[[i]] : (!fir.array<?xf32>, f32, index) -> !fir.array<?xf32>
 ! CHECK:    fir.result %[[update]] : !fir.array<?xf32>
 ! CHECK:  }
 ! CHECK:  fir.array_merge_store %[[x_load]], %[[copyout:.*]] to %[[x]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.box<!fir.array<?xf32>>
 
-! CHECK: fir.freemem %[[temp]] : !fir.heap<!fir.array<?xf32>>
+! CHECK: fir.freemem %[[addr]] : !fir.heap<!fir.array<?xf32>>
 
   call bar(x)
 end subroutine
 
 ! Test that copy-in/copy-out does not trigger the re-evaluation of
 ! the designator expression.
 ! CHECK-LABEL: func @_QPeval_expr_only_once(
 ! CHECK-SAME: %[[x:.*]]: !fir.ref<!fir.array<200xf32>>{{.*}}) {
 subroutine eval_expr_only_once(x)
   integer :: only_once
   real :: x(200)
 ! CHECK: fir.call @_QPonly_once()
 ! CHECK: %[[x_section:.*]] = fir.embox %[[x]](%{{.*}}) [%{{.*}}] : (!fir.ref<!fir.array<200xf32>>, !fir.shape<1>, !fir.slice<1>) -> !fir.box<!fir.array<?xf32>>
+! CHECK: %[[box_none:.*]] = fir.convert %[[x_section]] : (!fir.box<!fir.array<?xf32>>) -> !fir.box<none>
+! CHECK: %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
+! CHECK: %[[addr:.*]] = fir.if %[[is_contiguous]] -> (!fir.heap<!fir.array<?xf32>>) {
+
 ! CHECK: %[[temp:.*]] = fir.allocmem !fir.array<?xf32>
 ! CHECK-NOT: fir.call @_QPonly_once()
 ! CHECK:  fir.array_merge_store %{{.*}}, %{{.*}} to %[[temp]]
 ! CHECK-NOT: fir.call @_QPonly_once()
 
-! CHECK:  %[[cast:.*]] = fir.convert %[[temp]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<?xf32>>
+! CHECK:  %[[cast:.*]] = fir.convert %[[addr]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<?xf32>>
 ! CHECK:  fir.call @_QPbar(%[[cast]]) : (!fir.ref<!fir.array<?xf32>>) -> ()
   call bar(x(1:200:only_once()))
 
 ! CHECK-NOT: fir.call @_QPonly_once()
 ! CHECK:  fir.array_merge_store %{{.*}}, %{{.*}} to %[[x_section]]
 ! CHECK-NOT: fir.call @_QPonly_once()
-! CHECK: fir.freemem %[[temp]] : !fir.heap<!fir.array<?xf32>>
+
+! CHECK: fir.freemem %[[addr]] : !fir.heap<!fir.array<?xf32>>
 end subroutine
 
 ! Test no copy-in/copy-out is generated for contiguous assumed shapes.
 ! CHECK-LABEL: func @_QPtest_contiguous(
 ! CHECK-SAME: %[[x:.*]]: !fir.box<!fir.array<?xf32>>
 subroutine test_contiguous(x)
   real, contiguous :: x(:)
 ! CHECK: %[[addr:.*]] = fir.box_addr %[[x]] : (!fir.box<!fir.array<?xf32>>) -> !fir.ref<!fir.array<?xf32>>

 ! CHECK: %[[x:.*]]: !fir.box<!fir.array<?xf32>>{{.*}}) {
 subroutine test_intent_out(x)
   real :: x(:)
   interface
   subroutine bar_intent_out(x)
     real, intent(out) :: x(100)
   end subroutine
   end interface
+! CHECK: %[[box_none:.*]] = fir.convert %[[x]] : (!fir.box<!fir.array<?xf32>>) -> !fir.box<none>
+! CHECK: %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
+! CHECK: %[[addr:.*]] = fir.if %[[is_contiguous]]
+! CHECK: } else {
 ! CHECK: %[[dim:.*]]:3 = fir.box_dims %[[x]], %c0{{.*}} : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
 ! CHECK: %[[temp:.*]] = fir.allocmem !fir.array<?xf32>, %[[dim]]#1
 ! CHECK-NOT:  fir.array_merge_store
-! CHECK:  %[[cast:.*]] = fir.convert %[[temp]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
+! CHECK: %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false{{.*}} : i1
+! CHECK:  %[[cast:.*]] = fir.convert %[[addr]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
 ! CHECK:  fir.call @_QPbar_intent_out(%[[cast]]) : (!fir.ref<!fir.array<100xf32>>) -> ()
   call bar_intent_out(x)
-! CHECK:  fir.array_merge_store %{{.*}}, %{{.*}} to %[[x]]
-! CHECK: fir.freemem %[[temp]] : !fir.heap<!fir.array<?xf32>>
+  
+! CHECK: fir.if %[[not_contiguous]]
+! CHECK: fir.array_merge_store %{{.*}}, %{{.*}} to %[[x]]
+! CHECK: fir.freemem %[[addr]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK: return
 end subroutine
 
 ! Test copy-out is skipped for intent(out) arguments.
-! CHECK: func @_QPtest_intent_in(
+! CHECK-LABEL: func.func @_QPtest_intent_in(
 ! CHECK: %[[x:.*]]: !fir.box<!fir.array<?xf32>>{{.*}}) {
 subroutine test_intent_in(x)
   real :: x(:)
   interface
   subroutine bar_intent_in(x)
     real, intent(in) :: x(100)
   end subroutine
   end interface
+! CHECK: %[[box_none:.*]] = fir.convert %[[x]] : (!fir.box<!fir.array<?xf32>>) -> !fir.box<none>
+! CHECK: %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
+! CHECK: %[[addr:.*]] = fir.if %[[is_contiguous]]
+! CHECK: } else {
 ! CHECK: %[[dim:.*]]:3 = fir.box_dims %[[x]], %c0{{.*}} : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
 ! CHECK: %[[temp:.*]] = fir.allocmem !fir.array<?xf32>, %[[dim]]#1
 ! CHECK:  fir.array_merge_store %{{.*}}, %{{.*}} to %[[temp]]
-! CHECK:  %[[cast:.*]] = fir.convert %[[temp]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
+! CHECK: %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false{{.*}} : i1
+! CHECK:  %[[cast:.*]] = fir.convert %[[addr]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
 ! CHECK:  fir.call @_QPbar_intent_in(%[[cast]]) : (!fir.ref<!fir.array<100xf32>>) -> ()
   call bar_intent_in(x)
+! CHECK: fir.if %[[not_contiguous]]
 ! CHECK-NOT:  fir.array_merge_store
-! CHECK: fir.freemem %[[temp]] : !fir.heap<!fir.array<?xf32>>
+! CHECK: fir.freemem %[[addr]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK: return
 end subroutine
 
 ! Test copy-in/copy-out is done for intent(inout)
 ! CHECK: func @_QPtest_intent_inout(
 ! CHECK: %[[x:.*]]: !fir.box<!fir.array<?xf32>>{{.*}}) {
 subroutine test_intent_inout(x)
   real :: x(:)
   interface
   subroutine bar_intent_inout(x)
     real, intent(inout) :: x(100)
   end subroutine
   end interface
+! CHECK: %[[box_none:.*]] = fir.convert %[[x]] : (!fir.box<!fir.array<?xf32>>) -> !fir.box<none>
+! CHECK: %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
+! CHECK: %[[addr:.*]] = fir.if %[[is_contiguous]]
+! CHECK: } else {
 ! CHECK: %[[dim:.*]]:3 = fir.box_dims %[[x]], %c0{{.*}} : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
 ! CHECK: %[[temp:.*]] = fir.allocmem !fir.array<?xf32>, %[[dim]]#1
 ! CHECK:  fir.array_merge_store %{{.*}}, %{{.*}} to %[[temp]]
-! CHECK:  %[[cast:.*]] = fir.convert %[[temp]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
+! CHECK: %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false{{.*}} : i1
+! CHECK:  %[[cast:.*]] = fir.convert %[[addr]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
 ! CHECK:  fir.call @_QPbar_intent_inout(%[[cast]]) : (!fir.ref<!fir.array<100xf32>>) -> ()
   call bar_intent_inout(x)
+! CHECK: fir.if %[[not_contiguous]]
 ! CHECK:  fir.array_merge_store %{{.*}}, %{{.*}} to %[[x]]
-! CHECK: fir.freemem %[[temp]] : !fir.heap<!fir.array<?xf32>>
+! CHECK: fir.freemem %[[addr]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK: return
 end subroutine
 
 ! Test characters are handled correctly
 ! CHECK-LABEL: func @_QPtest_char(
 ! CHECK-SAME:    %[[VAL_0:.*]]: !fir.box<!fir.array<?x!fir.char<1,10>>>{{.*}}) {
 subroutine test_char(x)
   ! CHECK: %[[VAL_1:.*]] = arith.constant 10 : index
+  ! CHECK: %[[box_none:.*]] = fir.convert %[[VAL_0]] : (!fir.box<!fir.array<?x!fir.char<1,10>>>) -> !fir.box<none>
+  ! CHECK: %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
+  ! CHECK: %[[addr:.*]] = fir.if %[[is_contiguous]]
+  ! CHECK: } else {
   ! CHECK: %[[VAL_2:.*]] = arith.constant 0 : index
   ! CHECK: %[[VAL_3:.*]]:3 = fir.box_dims %[[VAL_0]], %[[VAL_2]] : (!fir.box<!fir.array<?x!fir.char<1,10>>>, index) -> (index, index, index)
   ! CHECK: %[[VAL_4:.*]] = fir.allocmem !fir.array<?x!fir.char<1,10>>, %[[VAL_3]]#1 {uniq_name = ".copyinout"}
   ! CHECK: %[[VAL_5:.*]] = fir.shape %[[VAL_3]]#1 : (index) -> !fir.shape<1>
   ! CHECK: %[[VAL_6:.*]] = fir.array_load %[[VAL_4]](%[[VAL_5]]) : (!fir.heap<!fir.array<?x!fir.char<1,10>>>, !fir.shape<1>) -> !fir.array<?x!fir.char<1,10>>
   ! CHECK: %[[VAL_7:.*]] = fir.array_load %[[VAL_0]] : (!fir.box<!fir.array<?x!fir.char<1,10>>>) -> !fir.array<?x!fir.char<1,10>>
   ! CHECK: %[[VAL_8:.*]] = arith.constant 1 : index
   ! CHECK: %[[VAL_9:.*]] = arith.constant 0 : index
   ! CHECK: %[[VAL_10:.*]] = arith.subi %[[VAL_3]]#1, %[[VAL_8]] : index
   ! CHECK: %[[VAL_11:.*]] = fir.do_loop %[[VAL_12:.*]] = %[[VAL_9]] to %[[VAL_10]] step %[[VAL_8]] unordered iter_args(%[[VAL_13:.*]] = %[[VAL_6]]) -> (!fir.array<?x!fir.char<1,10>>) {
   ! CHECK: %[[VAL_14:.*]] = fir.array_access %[[VAL_7]], %[[VAL_12]] : (!fir.array<?x!fir.char<1,10>>, index) -> !fir.ref<!fir.char<1,10>>
   ! CHECK: %[[VAL_15:.*]] = fir.array_access %[[VAL_13]], %[[VAL_12]] : (!fir.array<?x!fir.char<1,10>>, index) -> !fir.ref<!fir.char<1,10>>
   ! CHECK: %[[VAL_16:.*]] = arith.constant 10 : index
   ! CHECK: %[[VAL_17:.*]] = arith.constant 1 : i64
   ! CHECK: %[[VAL_18:.*]] = fir.convert %[[VAL_16]] : (index) -> i64
   ! CHECK: %[[VAL_19:.*]] = arith.muli %[[VAL_17]], %[[VAL_18]] : i64
   ! CHECK: %[[VAL_20:.*]] = arith.constant false
   ! CHECK: %[[VAL_21:.*]] = fir.convert %[[VAL_15]] : (!fir.ref<!fir.char<1,10>>) -> !fir.ref<i8>
   ! CHECK: %[[VAL_22:.*]] = fir.convert %[[VAL_14]] : (!fir.ref<!fir.char<1,10>>) -> !fir.ref<i8>
   ! CHECK: fir.call @llvm.memmove.p0.p0.i64(%[[VAL_21]], %[[VAL_22]], %[[VAL_19]], %[[VAL_20]]) : (!fir.ref<i8>, !fir.ref<i8>, i64, i1) -> ()
   ! CHECK: %[[VAL_23:.*]] = fir.array_amend %[[VAL_13]], %[[VAL_15]] : (!fir.array<?x!fir.char<1,10>>, !fir.ref<!fir.char<1,10>>) -> !fir.array<?x!fir.char<1,10>>
   ! CHECK: fir.result %[[VAL_23]] : !fir.array<?x!fir.char<1,10>>
   ! CHECK: }
   ! CHECK: fir.array_merge_store %[[VAL_6]], %[[VAL_24:.*]] to %[[VAL_4]] : !fir.array<?x!fir.char<1,10>>, !fir.array<?x!fir.char<1,10>>, !fir.heap<!fir.array<?x!fir.char<1,10>>>
-  ! CHECK: %[[VAL_25:.*]] = fir.convert %[[VAL_4]] : (!fir.heap<!fir.array<?x!fir.char<1,10>>>) -> !fir.ref<!fir.char<1,?>>
+  ! CHECK: %[[dim:.*]]:3 = fir.box_dims %[[VAL_0]], %c0{{.*}} : (!fir.box<!fir.array<?x!fir.char<1,10>>>, index) -> (index, index, index)
+  ! CHECK: %[[VAL_25:.*]] = fir.convert %[[addr]] : (!fir.heap<!fir.array<?x!fir.char<1,10>>>) -> !fir.ref<!fir.char<1,?>>
   ! CHECK: %[[VAL_26:.*]] = fir.emboxchar %[[VAL_25]], %[[VAL_1]] : (!fir.ref<!fir.char<1,?>>, index) -> !fir.boxchar<1>
   ! CHECK: fir.call @_QPbar_char(%[[VAL_26]]) : (!fir.boxchar<1>) -> ()
   ! CHECK: %[[VAL_27:.*]] = fir.array_load %[[VAL_0]] : (!fir.box<!fir.array<?x!fir.char<1,10>>>) -> !fir.array<?x!fir.char<1,10>>
   ! CHECK: %[[VAL_28:.*]] = arith.constant 0 : index
   ! CHECK: %[[VAL_29:.*]]:3 = fir.box_dims %[[VAL_0]], %[[VAL_28]] : (!fir.box<!fir.array<?x!fir.char<1,10>>>, index) -> (index, index, index)
-  ! CHECK: %[[VAL_30:.*]] = fir.shape %[[VAL_3]]#1 : (index) -> !fir.shape<1>
-  ! CHECK: %[[VAL_31:.*]] = fir.array_load %[[VAL_4]](%[[VAL_30]]) : (!fir.heap<!fir.array<?x!fir.char<1,10>>>, !fir.shape<1>) -> !fir.array<?x!fir.char<1,10>>
+  ! CHECK: %[[VAL_30:.*]] = fir.shape %[[dim]]#1 : (index) -> !fir.shape<1>
+  ! CHECK: %[[VAL_31:.*]] = fir.array_load %[[addr]](%[[VAL_30]]) : (!fir.heap<!fir.array<?x!fir.char<1,10>>>, !fir.shape<1>) -> !fir.array<?x!fir.char<1,10>>
   ! CHECK: %[[VAL_32:.*]] = arith.constant 1 : index
   ! CHECK: %[[VAL_33:.*]] = arith.constant 0 : index
   ! CHECK: %[[VAL_34:.*]] = arith.subi %[[VAL_29]]#1, %[[VAL_32]] : index
   ! CHECK: %[[VAL_35:.*]] = fir.do_loop %[[VAL_36:.*]] = %[[VAL_33]] to %[[VAL_34]] step %[[VAL_32]] unordered iter_args(%[[VAL_37:.*]] = %[[VAL_27]]) -> (!fir.array<?x!fir.char<1,10>>) {
   ! CHECK: %[[VAL_38:.*]] = fir.array_access %[[VAL_31]], %[[VAL_36]] : (!fir.array<?x!fir.char<1,10>>, index) -> !fir.ref<!fir.char<1,10>>
   ! CHECK: %[[VAL_39:.*]] = fir.array_access %[[VAL_37]], %[[VAL_36]] : (!fir.array<?x!fir.char<1,10>>, index) -> !fir.ref<!fir.char<1,10>>
   ! CHECK: %[[VAL_40:.*]] = arith.constant 10 : index
   ! CHECK: %[[VAL_41:.*]] = arith.constant 1 : i64
   ! CHECK: %[[VAL_42:.*]] = fir.convert %[[VAL_40]] : (index) -> i64
   ! CHECK: %[[VAL_43:.*]] = arith.muli %[[VAL_41]], %[[VAL_42]] : i64
   ! CHECK: %[[VAL_44:.*]] = arith.constant false
   ! CHECK: %[[VAL_45:.*]] = fir.convert %[[VAL_39]] : (!fir.ref<!fir.char<1,10>>) -> !fir.ref<i8>
   ! CHECK: %[[VAL_46:.*]] = fir.convert %[[VAL_38]] : (!fir.ref<!fir.char<1,10>>) -> !fir.ref<i8>
   ! CHECK: fir.call @llvm.memmove.p0.p0.i64(%[[VAL_45]], %[[VAL_46]], %[[VAL_43]], %[[VAL_44]]) : (!fir.ref<i8>, !fir.ref<i8>, i64, i1) -> ()
   ! CHECK: %[[VAL_47:.*]] = fir.array_amend %[[VAL_37]], %[[VAL_39]] : (!fir.array<?x!fir.char<1,10>>, !fir.ref<!fir.char<1,10>>) -> !fir.array<?x!fir.char<1,10>>
   ! CHECK: fir.result %[[VAL_47]] : !fir.array<?x!fir.char<1,10>>
   ! CHECK: }
   ! CHECK: fir.array_merge_store %[[VAL_27]], %[[VAL_48:.*]] to %[[VAL_0]] : !fir.array<?x!fir.char<1,10>>, !fir.array<?x!fir.char<1,10>>, !fir.box<!fir.array<?x!fir.char<1,10>>>
-  ! CHECK: fir.freemem %[[VAL_4]] : !fir.heap<!fir.array<?x!fir.char<1,10>>>
+  ! CHECK: fir.freemem %[[addr]] : !fir.heap<!fir.array<?x!fir.char<1,10>>>
 
   character(10) :: x(:)
   call bar_char(x)
   ! CHECK:         return
   ! CHECK:       }
 end subroutine test_char
 
 ! CHECK-LABEL: func @_QPtest_scalar_substring_does_no_trigger_copy_inout

flang/test/Lower/dummy-argument-optional-2.f90

 subroutine pass_pointer_array(i)
   real, pointer :: i(:)
   call takes_opt_explicit_shape(i)
 ! CHECK:         %[[VAL_1:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>
 ! CHECK:         %[[VAL_2:.*]] = fir.box_addr %[[VAL_1]] : (!fir.box<!fir.ptr<!fir.array<?xf32>>>) -> !fir.ptr<!fir.array<?xf32>>
 ! CHECK:         %[[VAL_3:.*]] = fir.convert %[[VAL_2]] : (!fir.ptr<!fir.array<?xf32>>) -> i64
 ! CHECK:         %[[VAL_4:.*]] = arith.constant 0 : i64
 ! CHECK:         %[[VAL_5:.*]] = arith.cmpi ne, %[[VAL_3]], %[[VAL_4]] : i64
-! CHECK:         %[[VAL_6:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>
+! CHECK:         %[[box:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>
 ! CHECK:         %[[VAL_7:.*]] = arith.constant 0 : index
+! CHECK:         %[[box_none:.*]] = fir.convert %[[box]] : (!fir.box<!fir.ptr<!fir.array<?xf32>>>) -> !fir.box<none>
+! CHECK:         %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
 ! CHECK:         %[[VAL_9:.*]] = fir.if %[[VAL_5]] -> (!fir.heap<!fir.array<?xf32>>) {
+! CHECK:           %[[addr:.*]] = fir.if %[[is_contiguous]] -> (!fir.heap<!fir.array<?xf32>>) {
+! CHECK:             %[[box_addr:.*]] = fir.box_addr %[[box]] : (!fir.box<!fir.ptr<!fir.array<?xf32>>>) -> !fir.heap<!fir.array<?xf32>>
+! CHECK:             fir.result %[[box_addr]] : !fir.heap<!fir.array<?xf32>>
+! CHECK:           } else {
 ! CHECK:           %[[VAL_10:.*]] = arith.constant 0 : index
-! CHECK:           %[[VAL_11:.*]]:3 = fir.box_dims %[[VAL_6]], %[[VAL_10]] : (!fir.box<!fir.ptr<!fir.array<?xf32>>>, index) -> (index, index, index)
+! CHECK:           %[[VAL_11:.*]]:3 = fir.box_dims %[[box]], %[[VAL_10]] : (!fir.box<!fir.ptr<!fir.array<?xf32>>>, index) -> (index, index, index)
 ! CHECK:           %[[VAL_12:.*]] = fir.allocmem !fir.array<?xf32>, %[[VAL_11]]#1 {uniq_name = ".copyinout"}
 ! CHECK:           %[[VAL_20:.*]] = fir.do_loop {{.*}} {
 ! CHECK:           }
 ! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_20]] to %[[VAL_12]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.heap<!fir.array<?xf32>>
 ! CHECK:           fir.result %[[VAL_12]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         } else {
 ! CHECK:           %[[VAL_26:.*]] = fir.zero_bits !fir.heap<!fir.array<?xf32>>
 ! CHECK:           fir.result %[[VAL_26]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
+! CHECK:         %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false : i1
+! CHECK:         %[[and:.*]] = arith.andi %[[VAL_5]], %[[not_contiguous]] : i1
 ! CHECK:         %[[VAL_29:.*]] = fir.convert %[[VAL_9]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
 ! CHECK:         fir.call @_QPtakes_opt_explicit_shape(%[[VAL_29]]) : (!fir.ref<!fir.array<100xf32>>) -> ()
-! CHECK:         fir.if %[[VAL_5]] {
+! CHECK:         fir.if %[[and]] {
 ! CHECK:           %[[VAL_40:.*]] = fir.do_loop {{.*}} {
 ! CHECK:           }
-! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_40]] to %[[VAL_6]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.box<!fir.ptr<!fir.array<?xf32>>>
+! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_40]] to %[[box]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.box<!fir.ptr<!fir.array<?xf32>>>
 ! CHECK:           fir.freemem %[[VAL_9]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
 end subroutine
 
 ! CHECK-LABEL: func @_QMoptional_testsPpass_pointer_array_char(
 ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>>{{.*}}) {
 subroutine pass_pointer_array_char(c)
   character(:), pointer :: c(:)
   call takes_opt_explicit_shape_char(c)
 ! CHECK:         %[[VAL_1:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>>
 ! CHECK:         %[[VAL_2:.*]] = fir.box_addr %[[VAL_1]] : (!fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>) -> !fir.ptr<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         %[[VAL_3:.*]] = fir.convert %[[VAL_2]] : (!fir.ptr<!fir.array<?x!fir.char<1,?>>>) -> i64
 ! CHECK:         %[[VAL_4:.*]] = arith.constant 0 : i64
 ! CHECK:         %[[VAL_5:.*]] = arith.cmpi ne, %[[VAL_3]], %[[VAL_4]] : i64
 ! CHECK:         %[[VAL_6:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>>
+! CHECK:         %[[box_none:.*]] = fir.convert %[[VAL_6]] : (!fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>) -> !fir.box<none>
+! CHECK:         %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
 ! CHECK:         %[[VAL_9:.*]] = fir.if %[[VAL_5]] -> (!fir.heap<!fir.array<?x!fir.char<1,?>>>) {
 ! CHECK:           %[[VAL_10:.*]] = arith.constant 0 : index
 ! CHECK:           %[[VAL_11:.*]]:3 = fir.box_dims %[[VAL_6]], %[[VAL_10]] : (!fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>, index) -> (index, index, index)
 ! CHECK:           %[[VAL_12:.*]] = fir.box_elesize %[[VAL_6]] : (!fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>) -> index
 ! CHECK:           %[[VAL_13:.*]] = fir.allocmem !fir.array<?x!fir.char<1,?>>(%[[VAL_12]] : index), %[[VAL_11]]#1 {uniq_name = ".copyinout"}
 ! CHECK:           %[[VAL_21:.*]] = fir.do_loop {{.*}} {
 ! CHECK:           }
 ! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_21]] to %[[VAL_13]] typeparams %[[VAL_12]] : !fir.array<?x!fir.char<1,?>>, !fir.array<?x!fir.char<1,?>>, !fir.heap<!fir.array<?x!fir.char<1,?>>>, index
 ! CHECK:           fir.result %[[VAL_13]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         } else {
 ! CHECK:           %[[VAL_46:.*]] = fir.zero_bits !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:           fir.result %[[VAL_46]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         }
 ! CHECK:         %[[VAL_47:.*]] = fir.box_elesize %[[VAL_6]] : (!fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>) -> index
+! CHECK:         %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false : i1
+! CHECK:         %[[and:.*]] = arith.andi %[[VAL_5]], %[[not_contiguous]] : i1
 ! CHECK:         %[[VAL_50:.*]] = fir.convert %[[VAL_9]] : (!fir.heap<!fir.array<?x!fir.char<1,?>>>) -> !fir.ref<!fir.char<1,?>>
 ! CHECK:         %[[VAL_52:.*]] = fir.emboxchar %[[VAL_50]], %[[VAL_47]] : (!fir.ref<!fir.char<1,?>>, index) -> !fir.boxchar<1>
 ! CHECK:         fir.call @_QPtakes_opt_explicit_shape_char(%[[VAL_52]]) : (!fir.boxchar<1>) -> ()
-! CHECK:         fir.if %[[VAL_5]] {
+! CHECK:         fir.if %[[and]] {
 ! CHECK:           %[[VAL_62:.*]] = fir.do_loop {{.*}} {
 ! CHECK:           }
 ! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_62]] to %[[VAL_6]] : !fir.array<?x!fir.char<1,?>>, !fir.array<?x!fir.char<1,?>>, !fir.box<!fir.ptr<!fir.array<?x!fir.char<1,?>>>>
 ! CHECK:           fir.freemem %[[VAL_9]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         }
 ! CHECK:         return
 ! CHECK:       }
 end subroutine
 
 ! This case is bit special because the pointer is not a symbol but a function
 ! result. Test that the copy-in/copy-out is the same as with normal pointers.
 
 ! CHECK-LABEL: func @_QMoptional_testsPforward_pointer_array() {
 subroutine forward_pointer_array()
   call takes_opt_explicit_shape(returns_pointer())
 ! CHECK:         %[[VAL_0:.*]] = fir.alloca !fir.box<!fir.ptr<!fir.array<?xf32>>> {bindc_name = ".result"}
 ! CHECK:         %[[VAL_1:.*]] = fir.call @_QPreturns_pointer() : () -> !fir.box<!fir.ptr<!fir.array<?xf32>>>
 ! CHECK:         fir.save_result %[[VAL_1]] to %[[VAL_0]] : !fir.box<!fir.ptr<!fir.array<?xf32>>>, !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>
 ! CHECK:         %[[VAL_2:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>
 ! CHECK:         %[[VAL_3:.*]] = fir.box_addr %[[VAL_2]] : (!fir.box<!fir.ptr<!fir.array<?xf32>>>) -> !fir.ptr<!fir.array<?xf32>>
 ! CHECK:         %[[VAL_4:.*]] = fir.convert %[[VAL_3]] : (!fir.ptr<!fir.array<?xf32>>) -> i64
 ! CHECK:         %[[VAL_5:.*]] = arith.constant 0 : i64
 ! CHECK:         %[[VAL_6:.*]] = arith.cmpi ne, %[[VAL_4]], %[[VAL_5]] : i64
+! CHECK:         %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%{{.*}}) : (!fir.box<none>) -> i1
 ! CHECK:         %[[VAL_7:.*]] = fir.if %[[VAL_6]] -> (!fir.heap<!fir.array<?xf32>>) {
 ! CHECK:           %[[VAL_10:.*]] = fir.allocmem !fir.array<?xf32>
 ! CHECK:           fir.do_loop {{.*}} {
 ! CHECK:           }
 ! CHECK:           fir.result %[[VAL_10]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         } else {
 ! CHECK:           %[[VAL_11:.*]] = fir.zero_bits !fir.heap<!fir.array<?xf32>>
 ! CHECK:           fir.result %[[VAL_11]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
+! CHECK:         %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false : i1
+! CHECK:         %[[and:.*]] = arith.andi %[[VAL_6]], %[[not_contiguous]] : i1
 ! CHECK:         %[[VAL_14:.*]] = fir.convert %[[VAL_7]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
 ! CHECK:         fir.call @_QPtakes_opt_explicit_shape(%[[VAL_14]]) : (!fir.ref<!fir.array<100xf32>>) -> ()
-! CHECK:         fir.if %[[VAL_6]] {
+! CHECK:         fir.if %[[and]] {
 ! CHECK:           fir.do_loop {{.*}} {
 ! CHECK:           }
 ! CHECK:           fir.freemem %[[VAL_7]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
 end subroutine
 
 ! -----------------------------------------------------------------------------
 !    Test passing assumed shape optional to explicit shape optional
 ! -----------------------------------------------------------------------------
 ! The fix.box can only be read if the assumed shape is present,
 ! and the copy-in/copy-out must also be conditional on the assumed
 ! shape presence.
 
 ! CHECK-LABEL: func @_QMoptional_testsPpass_opt_assumed_shape(
 ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.box<!fir.array<?xf32>> {fir.bindc_name = "x", fir.optional}) {
 subroutine pass_opt_assumed_shape(x)
   real, optional :: x(:)
   call takes_opt_explicit_shape(x)
 ! CHECK:         %[[VAL_1:.*]] = fir.is_present %[[VAL_0]] : (!fir.box<!fir.array<?xf32>>) -> i1
 ! CHECK:         %[[VAL_2:.*]] = fir.zero_bits !fir.ref<!fir.array<?xf32>>
 ! CHECK:         %[[VAL_3:.*]] = arith.constant 0 : index
 ! CHECK:         %[[VAL_4:.*]] = fir.shape %[[VAL_3]] : (index) -> !fir.shape<1>
 ! CHECK:         %[[VAL_5:.*]] = fir.embox %[[VAL_2]](%[[VAL_4]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.box<!fir.array<?xf32>>
 ! CHECK:         %[[VAL_6:.*]] = arith.select %[[VAL_1]], %[[VAL_0]], %[[VAL_5]] : !fir.box<!fir.array<?xf32>>
+! CHECK:         %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%{{.*}}) : (!fir.box<none>) -> i1
 ! CHECK:         %[[VAL_7:.*]] = fir.if %[[VAL_1]] -> (!fir.heap<!fir.array<?xf32>>) {
 ! CHECK:           %[[VAL_8:.*]] = arith.constant 0 : index
 ! CHECK:           %[[VAL_9:.*]]:3 = fir.box_dims %[[VAL_6]], %[[VAL_8]] : (!fir.box<!fir.array<?xf32>>, index) -> (index, index, index)
 ! CHECK:           %[[VAL_10:.*]] = fir.allocmem !fir.array<?xf32>, %[[VAL_9]]#1 {uniq_name = ".copyinout"}
 ! CHECK:           %[[VAL_17:.*]] = fir.do_loop {{.*}} {
 ! CHECK:           }
 ! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_17]] to %[[VAL_10]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.heap<!fir.array<?xf32>>
 ! CHECK:           fir.result %[[VAL_10]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         } else {
 ! CHECK:           %[[VAL_23:.*]] = fir.zero_bits !fir.heap<!fir.array<?xf32>>
 ! CHECK:           fir.result %[[VAL_23]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
+! CHECK:         %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false : i1
+! CHECK:         %[[and:.*]] = arith.andi %[[VAL_1]], %[[not_contiguous]] : i1
 ! CHECK:         %[[VAL_26:.*]] = fir.convert %[[VAL_27:.*]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
 ! CHECK:         fir.call @_QPtakes_opt_explicit_shape(%[[VAL_26]]) : (!fir.ref<!fir.array<100xf32>>) -> ()
-! CHECK:         fir.if %[[VAL_1]] {
+! CHECK:         fir.if %[[and]] {
 ! CHECK:           %[[VAL_36:.*]] = fir.do_loop {{.*}} { 
 ! CHECK:           }
 ! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_36]] to %[[VAL_6]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.box<!fir.array<?xf32>>
 ! CHECK:           fir.freemem %[[VAL_27]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
 end subroutine
 
 ! CHECK-LABEL: func @_QMoptional_testsPpass_opt_assumed_shape_char(
 ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.box<!fir.array<?x!fir.char<1,?>>> {fir.bindc_name = "c", fir.optional}) {
 subroutine pass_opt_assumed_shape_char(c)
   character(*), optional :: c(:)
   call takes_opt_explicit_shape_char(c)
 ! CHECK:         %[[VAL_1:.*]] = fir.is_present %[[VAL_0]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> i1
 ! CHECK:         %[[VAL_2:.*]] = fir.zero_bits !fir.ref<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         %[[VAL_3:.*]] = arith.constant 0 : index
 ! CHECK:         %[[VAL_4:.*]] = fir.shape %[[VAL_3]] : (index) -> !fir.shape<1>
 ! CHECK:         %[[VAL_5:.*]] = arith.constant 0 : index
 ! CHECK:         %[[VAL_6:.*]] = fir.embox %[[VAL_2]](%[[VAL_4]]) typeparams %[[VAL_5]] : (!fir.ref<!fir.array<?x!fir.char<1,?>>>, !fir.shape<1>, index) -> !fir.box<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         %[[VAL_7:.*]] = arith.select %[[VAL_1]], %[[VAL_0]], %[[VAL_6]] : !fir.box<!fir.array<?x!fir.char<1,?>>>
+! CHECK:         %[[box_none:.*]] = fir.convert %[[VAL_7]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> !fir.box<none>
+! CHECK:         %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
 ! CHECK:         %[[VAL_8:.*]] = fir.if %[[VAL_1]] -> (!fir.heap<!fir.array<?x!fir.char<1,?>>>) {
+! CHECK:         %[[addr:.*]] = fir.if %[[is_contiguous]] -> (!fir.heap<!fir.array<?x!fir.char<1,?>>>) {
+! CHECK:           %[[res:.*]] = fir.box_addr %[[VAL_7]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> !fir.heap<!fir.array<?x!fir.char<1,?>>>
+! CHECK:           fir.result %[[res]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
+! CHECK:         } else {
+! CHECK:           %[[box_elesize:.*]] = fir.box_elesize %[[VAL_7]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> index
+! CHECK:           %[[temp:.*]] = fir.allocmem !fir.array<?x!fir.char<1,?>>(%[[box_elesize]] : index), %{{.*}}#1 {uniq_name = ".copyinout"}
 ! CHECK:           %[[VAL_19:.*]] = fir.do_loop  {{.*}} {
 ! CHECK:           }
-! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_19]] to %[[VAL_12]] typeparams %[[VAL_11]] : !fir.array<?x!fir.char<1,?>>, !fir.array<?x!fir.char<1,?>>, !fir.heap<!fir.array<?x!fir.char<1,?>>>, index
+! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_19]] to %[[temp]] typeparams %[[box_elesize]] : !fir.array<?x!fir.char<1,?>>, !fir.array<?x!fir.char<1,?>>, !fir.heap<!fir.array<?x!fir.char<1,?>>>, index
 ! CHECK:           fir.result %[[VAL_12]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         } else {
 ! CHECK:           %[[VAL_44:.*]] = fir.zero_bits !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:           fir.result %[[VAL_44]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         }
 ! CHECK:         %[[VAL_45:.*]] = fir.box_elesize %[[VAL_7]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> index
+! CHECK:         %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false : i1
+! CHECK:         %[[and:.*]] = arith.andi %[[VAL_1]], %[[not_contiguous]] : i1
 ! CHECK:         %[[VAL_48:.*]] = fir.convert %[[VAL_49:.*]] : (!fir.heap<!fir.array<?x!fir.char<1,?>>>) -> !fir.ref<!fir.char<1,?>>
 ! CHECK:         %[[VAL_50:.*]] = fir.emboxchar %[[VAL_48]], %[[VAL_45]] : (!fir.ref<!fir.char<1,?>>, index) -> !fir.boxchar<1>
 ! CHECK:         fir.call @_QPtakes_opt_explicit_shape_char(%[[VAL_50]]) : (!fir.boxchar<1>) -> ()
-! CHECK:         fir.if %[[VAL_1]] {
+! CHECK:         fir.if %[[and]] {
 ! CHECK:           %[[VAL_59:.*]] = fir.do_loop {{.*}} {
 ! CHECK:           fir.array_merge_store %{{.*}}, %[[VAL_59]] to %[[VAL_7]] : !fir.array<?x!fir.char<1,?>>, !fir.array<?x!fir.char<1,?>>, !fir.box<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:           fir.freemem %[[VAL_49]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:         }
 end subroutine
 
 ! -----------------------------------------------------------------------------
 !    Test passing contiguous optional assumed shape to explicit shape optional
subroutine pass_opt_assumed_shape_to_intentin(x)
   real, optional :: x(:)
   call takes_opt_explicit_shape_intentin(x)
 ! CHECK:         %[[VAL_1:.*]] = fir.is_present %[[VAL_0]] : (!fir.box<!fir.array<?xf32>>) -> i1
 ! CHECK:         %[[VAL_2:.*]] = fir.zero_bits !fir.ref<!fir.array<?xf32>>
 ! CHECK:         %[[VAL_3:.*]] = arith.constant 0 : index
 ! CHECK:         %[[VAL_4:.*]] = fir.shape %[[VAL_3]] : (index) -> !fir.shape<1>
 ! CHECK:         %[[VAL_5:.*]] = fir.embox %[[VAL_2]](%[[VAL_4]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.box<!fir.array<?xf32>>
 ! CHECK:         %[[VAL_6:.*]] = arith.select %[[VAL_1]], %[[VAL_0]], %[[VAL_5]] : !fir.box<!fir.array<?xf32>>
+! CHECK:         %[[box_none:.*]] = fir.convert %[[VAL_6]] : (!fir.box<!fir.array<?xf32>>) -> !fir.box<none>
+! CHECK:         %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
 ! CHECK:         %[[VAL_7:.*]] = fir.if %[[VAL_1]] -> (!fir.heap<!fir.array<?xf32>>) {
 ! CHECK:           %[[VAL_10:.*]] = fir.allocmem !fir.array<?xf32>
 ! CHECK:           fir.do_loop {{.*}} {
 ! CHECK:           }
 ! CHECK:           fir.result %[[VAL_10]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         } else {
 ! CHECK:           %[[VAL_23:.*]] = fir.zero_bits !fir.heap<!fir.array<?xf32>>
 ! CHECK:           fir.result %[[VAL_23]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
+! CHECK:         %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false : i1
+! CHECK:         %[[and:.*]] = arith.andi %[[VAL_1]], %[[not_contiguous]] : i1
 ! CHECK:         %[[VAL_24:.*]] = fir.convert %[[VAL_7]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
 ! CHECK:         fir.call @_QPtakes_opt_explicit_shape_intentin(%[[VAL_24]]) : (!fir.ref<!fir.array<100xf32>>) -> ()
-! CHECK:         fir.if %[[VAL_1]] {
+! CHECK:         fir.if %[[and]] {
 ! CHECK-NOT:       fir.do_loop
 ! CHECK:           fir.freemem %[[VAL_7]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
 end subroutine
 
 ! CHECK-LABEL: func @_QMoptional_testsPpass_opt_assumed_shape_to_intentout(
 ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.box<!fir.array<?xf32>> {fir.bindc_name = "x", fir.optional}) {
 subroutine pass_opt_assumed_shape_to_intentout(x)
   real, optional :: x(:)
   call takes_opt_explicit_shape_intentout(x)
 ! CHECK:         %[[VAL_1:.*]] = fir.is_present %[[VAL_0]] : (!fir.box<!fir.array<?xf32>>) -> i1
 ! CHECK:         %[[VAL_2:.*]] = fir.zero_bits !fir.ref<!fir.array<?xf32>>
 ! CHECK:         %[[VAL_3:.*]] = arith.constant 0 : index
 ! CHECK:         %[[VAL_4:.*]] = fir.shape %[[VAL_3]] : (index) -> !fir.shape<1>
 ! CHECK:         %[[VAL_5:.*]] = fir.embox %[[VAL_2]](%[[VAL_4]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.box<!fir.array<?xf32>>
 ! CHECK:         %[[VAL_6:.*]] = arith.select %[[VAL_1]], %[[VAL_0]], %[[VAL_5]] : !fir.box<!fir.array<?xf32>>
+! CHECK:         %[[box_none:.*]] = fir.convert %[[VAL_6]] : (!fir.box<!fir.array<?xf32>>) -> !fir.box<none>
+! CHECK:         %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
 ! CHECK:         %[[VAL_7:.*]] = fir.if %[[VAL_1]] -> (!fir.heap<!fir.array<?xf32>>) {
 ! CHECK:           %[[VAL_10:.*]] = fir.allocmem !fir.array<?xf32>
 ! CHECK-NOT:       fir.do_loop
 ! CHECK:           fir.result %[[VAL_10]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         } else {
 ! CHECK:           %[[VAL_11:.*]] = fir.zero_bits !fir.heap<!fir.array<?xf32>>
 ! CHECK:           fir.result %[[VAL_11]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
+! CHECK:         %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false : i1
+! CHECK:         %[[and:.*]] = arith.andi %[[VAL_1]], %[[not_contiguous]] : i1
 ! CHECK:         %[[VAL_14:.*]] = fir.convert %[[VAL_7]] : (!fir.heap<!fir.array<?xf32>>) -> !fir.ref<!fir.array<100xf32>>
 ! CHECK:         fir.call @_QPtakes_opt_explicit_shape_intentout(%[[VAL_14]]) : (!fir.ref<!fir.array<100xf32>>) -> ()
-! CHECK:         fir.if %[[VAL_1]] {
+! CHECK:         fir.if %[[and]] {
 ! CHECK:           fir.do_loop {{.*}} {
 ! CHECK:           }
 ! CHECK:           fir.freemem %[[VAL_7]] : !fir.heap<!fir.array<?xf32>>
 ! CHECK:         }
 end subroutine
 end module

flang/test/Lower/optional-value-caller.f90

subroutine test_dyn_array_from_assumed(i, n)
   integer, optional :: i(:)
   call dyn_array(i, n)
 ! CHECK:  %[[VAL_2:.*]] = fir.is_present %[[VAL_0]] : (!fir.box<!fir.array<?xi32>>) -> i1
 ! CHECK:  %[[VAL_3:.*]] = fir.zero_bits !fir.ref<!fir.array<?xi32>>
 ! CHECK:  %[[VAL_4:.*]] = arith.constant 0 : index
 ! CHECK:  %[[VAL_5:.*]] = fir.shape %[[VAL_4]] : (index) -> !fir.shape<1>
 ! CHECK:  %[[VAL_6:.*]] = fir.embox %[[VAL_3]](%[[VAL_5]]) : (!fir.ref<!fir.array<?xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<?xi32>>
 ! CHECK:  %[[VAL_7:.*]] = arith.select %[[VAL_2]], %[[VAL_0]], %[[VAL_6]] : !fir.box<!fir.array<?xi32>>
+! CHECK:  %[[box_none:.*]] = fir.convert %[[VAL_7]] : (!fir.box<!fir.array<?xi32>>) -> !fir.box<none>
+! CHECK:  %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
 ! CHECK:  %[[VAL_8:.*]] = fir.if %[[VAL_2]] -> (!fir.heap<!fir.array<?xi32>>) {
 ! CHECK:    %[[VAL_9:.*]] = arith.constant 0 : index
 ! CHECK:    %[[VAL_10:.*]]:3 = fir.box_dims %[[VAL_7]], %[[VAL_9]] : (!fir.box<!fir.array<?xi32>>, index) -> (index, index, index)
 ! CHECK:    %[[VAL_11:.*]] = fir.allocmem !fir.array<?xi32>, %[[VAL_10]]#1 {uniq_name = ".copy"}
 ! CHECK:    %[[VAL_18:.*]] = fir.do_loop
 ! CHECK:    }
 ! CHECK:    fir.array_merge_store %{{.*}}, %[[VAL_18]] to %[[VAL_11]] : !fir.array<?xi32>, !fir.array<?xi32>, !fir.heap<!fir.array<?xi32>>
 ! CHECK:    fir.result %[[VAL_11]] : !fir.heap<!fir.array<?xi32>>
 ! CHECK:  } else {
 ! CHECK:    %[[VAL_24:.*]] = fir.zero_bits !fir.heap<!fir.array<?xi32>>
 ! CHECK:    fir.result %[[VAL_24]] : !fir.heap<!fir.array<?xi32>>
 ! CHECK:  }
+! CHECK:  %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false{{.*}} : i1
+! CHECK:  %[[and:.*]] = arith.andi %[[VAL_2]], %[[not_contiguous]] : i1
 ! CHECK:  %[[VAL_25:.*]] = fir.convert %[[VAL_8]] : (!fir.heap<!fir.array<?xi32>>) -> !fir.ref<!fir.array<?xi32>>
 ! CHECK:  fir.call @_QPdyn_array(%[[VAL_25]], %[[VAL_1]]) : (!fir.ref<!fir.array<?xi32>>, !fir.ref<i64>) -> ()
-! CHECK:  fir.if %[[VAL_2]] {
+! CHECK:  fir.if %[[and]] {
 ! CHECK:    fir.freemem %[[VAL_8]] : !fir.heap<!fir.array<?xi32>>
 ! CHECK:  }
 end subroutine
 
 ! CHECK-LABEL: func @_QMtestPtest_array_ptr(
 ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>> {fir.bindc_name = "i"}) {
 subroutine test_array_ptr(i)
   integer, pointer :: i(:)
   call array(i)
 ! CHECK:  %[[VAL_1:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>
 ! CHECK:  %[[VAL_2:.*]] = fir.box_addr %[[VAL_1]] : (!fir.box<!fir.ptr<!fir.array<?xi32>>>) -> !fir.ptr<!fir.array<?xi32>>
 ! CHECK:  %[[VAL_3:.*]] = fir.convert %[[VAL_2]] : (!fir.ptr<!fir.array<?xi32>>) -> i64
 ! CHECK:  %[[VAL_4:.*]] = arith.constant 0 : i64
 ! CHECK:  %[[VAL_5:.*]] = arith.cmpi ne, %[[VAL_3]], %[[VAL_4]] : i64
 ! CHECK:  %[[VAL_6:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>
 ! CHECK:  %[[VAL_7:.*]] = arith.constant 0 : index
 ! CHECK:  %[[VAL_8:.*]]:3 = fir.box_dims %[[VAL_6]], %[[VAL_7]] : (!fir.box<!fir.ptr<!fir.array<?xi32>>>, index) -> (index, index, index)
+! CHECK:  %[[box_none:.*]] = fir.convert %[[VAL_6]] : (!fir.box<!fir.ptr<!fir.array<?xi32>>>) -> !fir.box<none>
+! CHECK:  %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
 ! CHECK:  %[[VAL_9:.*]] = fir.if %[[VAL_5]] -> (!fir.heap<!fir.array<?xi32>>) {
 ! CHECK:    %[[VAL_10:.*]] = arith.constant 0 : index
 ! CHECK:    %[[VAL_11:.*]]:3 = fir.box_dims %[[VAL_6]], %[[VAL_10]] : (!fir.box<!fir.ptr<!fir.array<?xi32>>>, index) -> (index, index, index)
 ! CHECK:    %[[VAL_12:.*]] = fir.allocmem !fir.array<?xi32>, %[[VAL_11]]#1 {uniq_name = ".copy"}
 ! CHECK:    %[[VAL_20:.*]] = fir.do_loop
 ! CHECK:    }
 ! CHECK:    fir.array_merge_store %{{.*}}, %[[VAL_20]] to %[[VAL_12]] : !fir.array<?xi32>, !fir.array<?xi32>, !fir.heap<!fir.array<?xi32>>
 ! CHECK:    fir.result %[[VAL_12]] : !fir.heap<!fir.array<?xi32>>
 ! CHECK:  } else {
 ! CHECK:    %[[VAL_26:.*]] = fir.zero_bits !fir.heap<!fir.array<?xi32>>
 ! CHECK:    fir.result %[[VAL_26]] : !fir.heap<!fir.array<?xi32>>
 ! CHECK:  }
+! CHECK:  %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false{{.*}} : i1
+! CHECK:  %[[and:.*]] = arith.andi %[[VAL_5]], %[[not_contiguous]] : i1
 ! CHECK:  %[[VAL_27:.*]] = fir.convert %[[VAL_9]] : (!fir.heap<!fir.array<?xi32>>) -> !fir.ref<!fir.array<100xi32>>
-! CHECK:  fir.call @_QParray(%[[VAL_27]]) : (!fir.ref<!fir.array<100xi32>>) -> ()
-! CHECK:  fir.if %[[VAL_5]] {
+! CHECK:  fir.if %[[and]] {
 ! CHECK:    fir.freemem %[[VAL_9]] : !fir.heap<!fir.array<?xi32>>
 ! CHECK:  }
 end subroutine
 
 ! CHECK-LABEL: func @_QMtestPtest_char(
 ! CHECK-SAME:  %[[VAL_0:.*]]: !fir.boxchar<1> {fir.bindc_name = "c", fir.optional}) {
 subroutine test_char(c)
   character(*), optional :: c

 ! CHECK:  %[[VAL_1:.*]] = fir.alloca i64 {bindc_name = "n", uniq_name = "_QMtestFtest_char_arrayEn"}
 ! CHECK:  %[[VAL_2:.*]] = fir.is_present %[[VAL_0]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> i1
 ! CHECK:  %[[VAL_3:.*]] = fir.zero_bits !fir.ref<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:  %[[VAL_4:.*]] = arith.constant 0 : index
 ! CHECK:  %[[VAL_5:.*]] = fir.shape %[[VAL_4]] : (index) -> !fir.shape<1>
 ! CHECK:  %[[VAL_6:.*]] = arith.constant 0 : index
 ! CHECK:  %[[VAL_7:.*]] = fir.embox %[[VAL_3]](%[[VAL_5]]) typeparams %[[VAL_6]] : (!fir.ref<!fir.array<?x!fir.char<1,?>>>, !fir.shape<1>, index) -> !fir.box<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:  %[[VAL_8:.*]] = arith.select %[[VAL_2]], %[[VAL_0]], %[[VAL_7]] : !fir.box<!fir.array<?x!fir.char<1,?>>>
+! CHECK:  %[[box_none:.*]] = fir.convert %5 : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> !fir.box<none>
+! CHECK:  %[[is_contiguous:.*]] = fir.call @_FortranAIsContiguous(%[[box_none]]) : (!fir.box<none>) -> i1
 ! CHECK:  %[[VAL_9:.*]] = fir.if %[[VAL_2]] -> (!fir.heap<!fir.array<?x!fir.char<1,?>>>) {
 ! CHECK:    %[[VAL_10:.*]] = arith.constant 0 : index
 ! CHECK:    %[[VAL_11:.*]]:3 = fir.box_dims %[[VAL_8]], %[[VAL_10]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>, index) -> (index, index, index)
 ! CHECK:    %[[VAL_12:.*]] = fir.box_elesize %[[VAL_8]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> index
 ! CHECK:    %[[VAL_13:.*]] = fir.allocmem !fir.array<?x!fir.char<1,?>>(%[[VAL_12]] : index), %[[VAL_11]]#1 {uniq_name = ".copy"}
 ! CHECK:    %[[VAL_20:.*]] = fir.do_loop {{.*}}
 ! CHECK:      fir.call @llvm.memmove.p0.p0.i64
 ! CHECK:    }
 ! CHECK:    fir.array_merge_store %{{.*}}, %[[VAL_20]] to %[[VAL_13]] typeparams %[[VAL_12]] : !fir.array<?x!fir.char<1,?>>, !fir.array<?x!fir.char<1,?>>, !fir.heap<!fir.array<?x!fir.char<1,?>>>, index
 ! CHECK:    fir.result %[[VAL_13]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:  } else {
 ! CHECK:    %[[VAL_45:.*]] = fir.zero_bits !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:    fir.result %[[VAL_45]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:  }
 ! CHECK:  %[[VAL_46:.*]] = fir.box_elesize %[[VAL_8]] : (!fir.box<!fir.array<?x!fir.char<1,?>>>) -> index
+! CHECK:  %[[not_contiguous:.*]] = arith.cmpi eq, %[[is_contiguous]], %false{{.*}} : i1
+! CHECK:  %[[and:.*]] = arith.andi %[[VAL_2]], %[[not_contiguous]] : i1
 ! CHECK:  %[[VAL_47:.*]] = fir.convert %[[VAL_9]] : (!fir.heap<!fir.array<?x!fir.char<1,?>>>) -> !fir.ref<!fir.char<1,?>>
 ! CHECK:  %[[VAL_49:.*]] = fir.emboxchar %[[VAL_47]], %[[VAL_46]] : (!fir.ref<!fir.char<1,?>>, index) -> !fir.boxchar<1>
 ! CHECK:  fir.call @_QPdyn_char_array(%[[VAL_49]], %[[VAL_1]]) : (!fir.boxchar<1>, !fir.ref<i64>) -> ()
-! CHECK:  fir.if %[[VAL_2]] {
+! CHECK:  fir.if %[[and]] {
 ! CHECK:    fir.freemem %[[VAL_9]] : !fir.heap<!fir.array<?x!fir.char<1,?>>>
 ! CHECK:  }
 end subroutine
 end