[flang] Upstream fix for performance issue

[schweitz]

Summary

Lowering of assignments in forall constructs could result in too many temporaries and copies. These changes eliminate this duplication.

[PeteSteinfeld]

All builds and tests correctly and looks good.

[jeanPerier]

clang-format should be run on lib/Lower/ConvertExpr.cpp, otherwise LGTM.

[schweitz]

Fixes clang-format issues on the entire file.

flang/lib/Lower/ConvertExpr.cpp

if (fir::isDerivedWithLenParameters(exv))
     TODO(loc, "copy derived type with length parameters");
   mlir::Type type = fir::unwrapPassByRefType(fir::getBase(exv).getType());
   fir::ExtendedValue temp = builder.createTemporary(loc, type);
   fir::factory::genScalarAssignment(builder, loc, temp, exv);
   return temp;
 }
 
 // An expression with non-zero rank is an array expression.
-template <typename A>
-static bool isArray(const A &x) {
-  return x.Rank() != 0;
-}
+template <typename A> static bool isArray(const A &x) { return x.Rank() != 0; }
 
 /// Is this a variable wrapped in parentheses?
-template <typename A>
-static bool isParenthesizedVariable(const A &) {
+template <typename A> static bool isParenthesizedVariable(const A &) {
   return false;
 }
 template <typename T>
 static bool isParenthesizedVariable(const Fortran::evaluate::Expr<T> &expr) {
   using ExprVariant = decltype(Fortran::evaluate::Expr<T>::u);
   using Parentheses = Fortran::evaluate::Parentheses<T>;
   if constexpr (Fortran::common::HasMember<Parentheses, ExprVariant>) {
     if (const auto *parentheses = std::get_if<Parentheses>(&expr.u))
ExtValue exv = std::visit(
         [&](const auto &x) { return genMutableBoxValueImpl(x); }, expr.u);
     const fir::MutableBoxValue *mutableBox =
         exv.getBoxOf<fir::MutableBoxValue>();
     if (!mutableBox)
       fir::emitFatalError(getLoc(), "expr was not lowered to MutableBoxValue");
     return *mutableBox;
   }
 
-  template <typename T>
-  ExtValue genMutableBoxValueImpl(const T &) {
+  template <typename T> ExtValue genMutableBoxValueImpl(const T &) {
     // NULL() case should not be handled here.
     fir::emitFatalError(getLoc(), "NULL() must be lowered in its context");
   }
 
   /// A `NULL()` in a position where a mutable box is expected has the same
   /// semantics as an absent optional box value.
   ExtValue genMutableBoxValueImpl(const Fortran::evaluate::NullPointer &) {
     mlir::Location loc = getLoc();
public:
   template <typename T>
   ExtValue genMutableBoxValueImpl(const Fortran::evaluate::Expr<T> &expr) {
     return std::visit([&](const auto &x) { return genMutableBoxValueImpl(x); },
                       expr.u);
   }
 
   mlir::Location getLoc() { return location; }
 
-  template <typename A>
-  mlir::Value genunbox(const A &expr) {
+  template <typename A> mlir::Value genunbox(const A &expr) {
     ExtValue e = genval(expr);
     if (const fir::UnboxedValue *r = e.getUnboxed())
       return *r;
     fir::emitFatalError(getLoc(), "unboxed expression expected");
   }
 
   /// Generate an integral constant of `value`
   template <int KIND>
public:
   mlir::Value createBinaryOp(const ExtValue &left, const ExtValue &right) {
     assert(fir::isUnboxedValue(left) && fir::isUnboxedValue(right));
     mlir::Value lhs = fir::getBase(left);
     mlir::Value rhs = fir::getBase(right);
     assert(lhs.getType() == rhs.getType() && "types must be the same");
     return builder.create<OpTy>(getLoc(), lhs, rhs);
   }
 
-  template <typename OpTy, typename A>
-  mlir::Value createBinaryOp(const A &ex) {
+  template <typename OpTy, typename A> mlir::Value createBinaryOp(const A &ex) {
     ExtValue left = genval(ex.left());
     return createBinaryOp<OpTy>(left, genval(ex.right()));
   }
 
 #undef GENBIN
 #define GENBIN(GenBinEvOp, GenBinTyCat, GenBinFirOp)                           \
   template <int KIND>                                                          \
   ExtValue genval(const Fortran::evaluate::GenBinEvOp<Fortran::evaluate::Type< \
#define GENBIN(GenBinEvOp, GenBinTyCat, GenBinFirOp)                           \
   ExtValue genval(const Fortran::evaluate::Parentheses<A> &op) {
     ExtValue input = genval(op.left());
     mlir::Value base = fir::getBase(input);
     mlir::Value newBase =
         builder.create<fir::NoReassocOp>(getLoc(), base.getType(), base);
     return fir::substBase(input, newBase);
   }
 
-  template <int KIND>
-  ExtValue genval(const Fortran::evaluate::Not<KIND> &op) {
+  template <int KIND> ExtValue genval(const Fortran::evaluate::Not<KIND> &op) {
     mlir::Value logical = genunbox(op.left());
     mlir::Value one = genBoolConstant(true);
     mlir::Value val =
         builder.createConvert(getLoc(), builder.getI1Type(), logical);
     return builder.create<mlir::arith::XOrIOp>(getLoc(), val, one);
   }
 
   template <int KIND>
for (const auto &[arg, dummy] :
       }
       llvm_unreachable("bad switch");
     }
     // Let the intrinsic library lower the intrinsic procedure call
     return Fortran::lower::genIntrinsicCall(builder, getLoc(), name, resultType,
                                             operands, stmtCtx);
   }
 
-  template <typename A>
-  bool isCharacterType(const A &exp) {
+  template <typename A> bool isCharacterType(const A &exp) {
     if (auto type = exp.GetType())
       return type->category() == Fortran::common::TypeCategory::Character;
     return false;
   }
 
   /// helper to detect statement functions
   static bool
   isStatementFunctionCall(const Fortran::evaluate::ProcedureRef &procRef) {

 
   ExtValue genval(const Fortran::evaluate::ProcedureRef &procRef) {
     llvm::Optional<mlir::Type> resTy;
     if (procRef.hasAlternateReturns())
       resTy = builder.getIndexType();
     return genProcedureRef(procRef, resTy);
   }
 
-  template <typename A>
-  bool isScalar(const A &x) {
-    return x.Rank() == 0;
-  }
+  template <typename A> bool isScalar(const A &x) { return x.Rank() == 0; }
 
   /// Helper to detect Transformational function reference.
-  template <typename T>
-  bool isTransformationalRef(const T &) {
-    return false;
-  }
+  template <typename T> bool isTransformationalRef(const T &) { return false; }
   template <typename T>
   bool isTransformationalRef(const Fortran::evaluate::FunctionRef<T> &funcRef) {
     return !funcRef.IsElemental() && funcRef.Rank();
   }
   template <typename T>
   bool isTransformationalRef(Fortran::evaluate::Expr<T> expr) {
     return std::visit([&](const auto &e) { return isTransformationalRef(e); },
                       expr.u);
   }
 
-  template <typename A>
-  ExtValue asArray(const A &x) {
+  template <typename A> ExtValue asArray(const A &x) {
     return Fortran::lower::createSomeArrayTempValue(converter, toEvExpr(x),
                                                     symMap, stmtCtx);
   }
 
   /// Lower an array value as an argument. This argument can be passed as a box
   /// value, so it may be possible to avoid making a temporary.
   template <typename A>
   ExtValue asArrayArg(const Fortran::evaluate::Expr<A> &x) {
     return std::visit([&](const auto &e) { return asArrayArg(e, x); }, x.u);
   }
   template <typename A, typename B>
   ExtValue asArrayArg(const Fortran::evaluate::Expr<A> &x, const B &y) {
     return std::visit([&](const auto &e) { return asArrayArg(e, y); }, x.u);
   }
   template <typename A, typename B>
   ExtValue asArrayArg(const Fortran::evaluate::Designator<A> &, const B &x) {
     // Designator is being passed as an argument to a procedure. Lower the
     // expression to a boxed value.
     auto someExpr = toEvExpr(x);
     return Fortran::lower::createBoxValue(getLoc(), converter, someExpr, symMap,
                                           stmtCtx);
   }
-  template <typename A, typename B>
-  ExtValue asArrayArg(const A &, const B &x) {
+  template <typename A, typename B> ExtValue asArrayArg(const A &, const B &x) {
     // If the expression to pass as an argument is not a designator, then create
     // an array temp.
     return asArray(x);
   }
 
-  template <typename A>
-  ExtValue gen(const Fortran::evaluate::Expr<A> &x) {
+  template <typename A> ExtValue gen(const Fortran::evaluate::Expr<A> &x) {
     // Whole array symbols or components, and results of transformational
     // functions already have a storage and the scalar expression lowering path
     // is used to not create a new temporary storage.
     if (isScalar(x) ||
         Fortran::evaluate::UnwrapWholeSymbolOrComponentDataRef(x) ||
         isTransformationalRef(x))
       return std::visit([&](const auto &e) { return genref(e); }, x.u);
     if (useBoxArg)
       return asArrayArg(x);
     return asArray(x);
   }
-  template <typename A>
-  ExtValue genval(const Fortran::evaluate::Expr<A> &x) {
+  template <typename A> ExtValue genval(const Fortran::evaluate::Expr<A> &x) {
     if (isScalar(x) || Fortran::evaluate::UnwrapWholeSymbolDataRef(x) ||
         inInitializer)
       return std::visit([&](const auto &e) { return genval(e); }, x.u);
     return asArray(x);
   }
 
   template <int KIND>
   ExtValue genval(const Fortran::evaluate::Expr<Fortran::evaluate::Type<
                       Fortran::common::TypeCategory::Logical, KIND>> &exp) {
     return std::visit([&](const auto &e) { return genval(e); }, exp.u);
   }
 
   using RefSet =
       std::tuple<Fortran::evaluate::ComplexPart, Fortran::evaluate::Substring,
                  Fortran::evaluate::DataRef, Fortran::evaluate::Component,
                  Fortran::evaluate::ArrayRef, Fortran::evaluate::CoarrayRef,
                  Fortran::semantics::SymbolRef>;
   template <typename A>
   static constexpr bool inRefSet = Fortran::common::HasMember<A, RefSet>;
 
   template <typename A, typename = std::enable_if_t<inRefSet<A>>>
   ExtValue genref(const A &a) {
     return gen(a);
   }
-  template <typename A>
-  ExtValue genref(const A &a) {
+  template <typename A> ExtValue genref(const A &a) {
     if (inInitializer) {
       // Initialization expressions can never allocate memory.
       return genval(a);
     }
     mlir::Type storageType = converter.genType(toEvExpr(a));
     return placeScalarValueInMemory(builder, getLoc(), genval(a), storageType);
   }
 
private:
   /// numeric types are treated as values.
   static mlir::Type adjustedArraySubtype(mlir::Type ty,
                                          mlir::ValueRange indices) {
     mlir::Type pathTy = fir::applyPathToType(ty, indices);
     assert(pathTy && "indices failed to apply to type");
     return adjustedArrayElementType(pathTy);
   }
 
+  /// Lower rhs of an array expression.
   ExtValue lowerArrayExpression(const Fortran::lower::SomeExpr &exp) {
     mlir::Type resTy = converter.genType(exp);
     return std::visit(
         [&](const auto &e) { return lowerArrayExpression(genarr(e), resTy); },
         exp.u);
   }
   ExtValue lowerArrayExpression(const ExtValue &exv) {
     assert(!explicitSpace);
fir::ShapeOp genShapeOp(llvm::ArrayRef<mlir::Value> shape) {
     return genShapeOp(getLoc(), builder, shape);
   }
 
   //===--------------------------------------------------------------------===//
   // Expression traversal and lowering.
   //===--------------------------------------------------------------------===//
 
   /// Lower the expression, \p x, in a scalar context.
-  template <typename A>
-  ExtValue asScalar(const A &x) {
+  template <typename A> ExtValue asScalar(const A &x) {
     return ScalarExprLowering{getLoc(), converter, symMap, stmtCtx}.genval(x);
   }
 
   /// Lower the expression, \p x, in a scalar context. If this is an explicit
   /// space, the expression may be scalar and refer to an array. We want to
   /// raise the array access to array operations in FIR to analyze potential
   /// conflicts even when the result is a scalar element.
-  template <typename A>
-  ExtValue asScalarArray(const A &x) {
+  template <typename A> ExtValue asScalarArray(const A &x) {
     return explicitSpaceIsActive() && !isPointerAssignment()
                ? genarr(x)(IterationSpace{})
                : asScalar(x);
   }
 
   /// Lower the expression in a scalar context to a memory reference.
-  template <typename A>
-  ExtValue asScalarRef(const A &x) {
+  template <typename A> ExtValue asScalarRef(const A &x) {
     return ScalarExprLowering{getLoc(), converter, symMap, stmtCtx}.gen(x);
   }
 
   /// Lower an expression without dereferencing any indirection that may be
   /// a nullptr (because this is an absent optional or unallocated/disassociated
   /// descriptor). The returned expression cannot be addressed directly, it is
   /// meant to inquire about its status before addressing the related entity.
-  template <typename A>
-  ExtValue asInquired(const A &x) {
+  template <typename A> ExtValue asInquired(const A &x) {
     return ScalarExprLowering{getLoc(), converter, symMap, stmtCtx}
         .lowerIntrinsicArgumentAsInquired(x);
   }
 
   /// Some temporaries are allocated on an element-by-element basis during the
   /// array expression evaluation. Collect the cleanups here so the resources
   /// can be freed before the next loop iteration, avoiding memory leaks. etc.
   Fortran::lower::StatementContext &getElementCtx() {
if (elementCtx) {
       stmtCtx.finalize(/*popScope=*/true);
       elementCtx = false;
     }
   }
 
   /// Lower an elemental function array argument. This ensures array
   /// sub-expressions that are not variables and must be passed by address
   /// are lowered by value and placed in memory.
-  template <typename A>
-  CC genElementalArgument(const A &x) {
+  template <typename A> CC genElementalArgument(const A &x) {
     // Ensure the returned element is in memory if this is what was requested.
     if ((semant == ConstituentSemantics::RefOpaque ||
          semant == ConstituentSemantics::DataAddr ||
          semant == ConstituentSemantics::ByValueArg)) {
       if (!Fortran::evaluate::IsVariable(x)) {
         PushSemantics(ConstituentSemantics::DataValue);
         CC cc = genarr(x);
         mlir::Location loc = getLoc();
CC genarr(const Fortran::evaluate::ProcedureDesignator &) {
     TODO(getLoc(), "procedure designator");
   }
   CC genarr(const Fortran::evaluate::ProcedureRef &x) {
     if (x.hasAlternateReturns())
       fir::emitFatalError(getLoc(),
                           "array procedure reference with alt-return");
     return genProcRef(x, llvm::None);
   }
-  template <typename A>
-  CC genScalarAndForwardValue(const A &x) {
+  template <typename A> CC genScalarAndForwardValue(const A &x) {
     ExtValue result = asScalar(x);
     return [=](IterSpace) { return result; };
   }
   template <typename A, typename = std::enable_if_t<Fortran::common::HasMember<
                             A, Fortran::evaluate::TypelessExpression>>>
   CC genarr(const A &x) {
     return genScalarAndForwardValue(x);
   }
 
-  template <typename A>
-  CC genarr(const Fortran::evaluate::Expr<A> &x) {
+  template <typename A> CC genarr(const Fortran::evaluate::Expr<A> &x) {
     LLVM_DEBUG(Fortran::lower::DumpEvaluateExpr::dump(llvm::dbgs(), x));
-    if (isArray(x) || explicitSpaceIsActive() ||
+    if (isArray(x) || (explicitSpaceIsActive() && isLeftHandSide()) ||
         isElementalProcWithArrayArgs(x))
       return std::visit([&](const auto &e) { return genarr(e); }, x.u);
+    if (explicitSpaceIsActive()) {
+      assert(!isArray(x) && !isLeftHandSide());
+      auto cc = std::visit([&](const auto &e) { return genarr(e); }, x.u);
+      auto result = cc(IterationSpace{});
+      return [=](IterSpace) { return result; };
+    }
     return genScalarAndForwardValue(x);
   }
 
   // Converting a value of memory bound type requires creating a temp and
   // copying the value.
   static ExtValue convertAdjustedType(fir::FirOpBuilder &builder,
                                       mlir::Location loc, mlir::Type toType,
                                       const ExtValue &exv) {
CC genarr(const Fortran::evaluate::ComplexComponent<KIND> &x) {
     bool isImagPart = x.isImaginaryPart;
     return [=](IterSpace iters) -> ExtValue {
       mlir::Value lhs = fir::getBase(lambda(iters));
       return fir::factory::Complex{builder, loc}.extractComplexPart(lhs,
                                                                     isImagPart);
     };
   }
 
-  template <typename T>
-  CC genarr(const Fortran::evaluate::Parentheses<T> &x) {
+  template <typename T> CC genarr(const Fortran::evaluate::Parentheses<T> &x) {
     mlir::Location loc = getLoc();
     if (isReferentiallyOpaque()) {
       // Context is a call argument in, for example, an elemental procedure
       // call. TODO: all array arguments should use array_load, array_access,
       // array_amend, and INTENT(OUT), INTENT(INOUT) arguments should have
       // array_merge_store ops.
       TODO(loc, "parentheses on argument in elemental call");
     }
return [=](IterSpace iters) -> ExtValue {
       return builder.create<fir::NegcOp>(loc, fir::getBase(f(iters)));
     };
   }
 
   //===--------------------------------------------------------------------===//
   // Binary elemental ops
   //===--------------------------------------------------------------------===//
 
-  template <typename OP, typename A>
-  CC createBinaryOp(const A &evEx) {
+  template <typename OP, typename A> CC createBinaryOp(const A &evEx) {
     mlir::Location loc = getLoc();
     auto lambda = genarr(evEx.left());
     auto rf = genarr(evEx.right());
     return [=](IterSpace iters) -> ExtValue {
       mlir::Value left = fir::getBase(lambda(iters));
       mlir::Value right = fir::getBase(rf(iters));
       return builder.create<OP>(loc, left, right);
     };
CC genarr(const Fortran::evaluate::ComplexConstructor<KIND> &x) {
     return [=](IterSpace iters) -> ExtValue {
       mlir::Value lhs = fir::getBase(lf(iters));
       mlir::Value rhs = fir::getBase(rf(iters));
       return fir::factory::Complex{builder, loc}.createComplex(KIND, lhs, rhs);
     };
   }
 
   /// Fortran's concatenation operator `//`.
-  template <int KIND>
-  CC genarr(const Fortran::evaluate::Concat<KIND> &x) {
+  template <int KIND> CC genarr(const Fortran::evaluate::Concat<KIND> &x) {
     mlir::Location loc = getLoc();
     auto lf = genarr(x.left());
     auto rf = genarr(x.right());
     return [=](IterSpace iters) -> ExtValue {
       auto lhs = lf(iters);
       auto rhs = rf(iters);
       const fir::CharBoxValue *lchr = lhs.getCharBox();
       const fir::CharBoxValue *rchr = rhs.getCharBox();
       if (lchr && rchr) {
         return fir::factory::CharacterExprHelper{builder, loc}
             .createConcatenate(*lchr, *rchr);
       }
       TODO(loc, "concat on unexpected extended values");
       return mlir::Value{};
     };
   }
 
-  template <int KIND>
-  CC genarr(const Fortran::evaluate::SetLength<KIND> &x) {
+  template <int KIND> CC genarr(const Fortran::evaluate::SetLength<KIND> &x) {
     auto lf = genarr(x.left());
     mlir::Value rhs = fir::getBase(asScalar(x.right()));
     return [=](IterSpace iters) -> ExtValue {
       mlir::Value lhs = fir::getBase(lf(iters));
       return fir::CharBoxValue{lhs, rhs};
     };
   }
 
-  template <typename A>
-  CC genarr(const Fortran::evaluate::Constant<A> &x) {
-    if (/*explicitSpaceIsActive() &&*/ x.Rank() == 0)
+  template <typename A> CC genarr(const Fortran::evaluate::Constant<A> &x) {
+    if (x.Rank() == 0)
       return genScalarAndForwardValue(x);
     mlir::Location loc = getLoc();
     mlir::IndexType idxTy = builder.getIndexType();
     mlir::Type arrTy = converter.genType(toEvExpr(x));
     std::string globalName = Fortran::lower::mangle::mangleArrayLiteral(x);
     fir::GlobalOp global = builder.getNamedGlobal(globalName);
     if (!global) {
       mlir::Type symTy = arrTy;
mlir::Value getUBound(const ExtValue &x, unsigned dim, mlir::Value one) {
     return builder.create<mlir::arith::SubIOp>(loc, add, one);
   }
 
   /// Return the extent of the boxed array `x` in dimesion `dim`.
   mlir::Value getExtent(const ExtValue &x, unsigned dim) {
     return fir::factory::readExtent(builder, getLoc(), x, dim);
   }
 
-  template <typename A>
-  ExtValue genArrayBase(const A &base) {
+  template <typename A> ExtValue genArrayBase(const A &base) {
     ScalarExprLowering sel{getLoc(), converter, symMap, stmtCtx};
     return base.IsSymbol() ? sel.gen(getFirstSym(base))
                            : sel.gen(base.GetComponent());
   }
 
-  template <typename A>
-  bool hasEvArrayRef(const A &x) {
+  template <typename A> bool hasEvArrayRef(const A &x) {
     struct HasEvArrayRefHelper
         : public Fortran::evaluate::AnyTraverse<HasEvArrayRefHelper> {
       HasEvArrayRefHelper()
           : Fortran::evaluate::AnyTraverse<HasEvArrayRefHelper>(*this) {}
       using Fortran::evaluate::AnyTraverse<HasEvArrayRefHelper>::operator();
       bool operator()(const Fortran::evaluate::ArrayRef &) const {
         return true;
       }
#define GENBIN(GenBinEvOp, GenBinTyCat, GenBinFirOp)                           \
 
   /// Example: <code>array%RE</code>
   CC genarr(const Fortran::evaluate::ComplexPart &x,
             ComponentPath &components) {
     components.reversePath.push_back(&x);
     return genarr(x.complex(), components);
   }
 
-  template <typename A>
-  CC genSlicePath(const A &x, ComponentPath &components) {
+  template <typename A> CC genSlicePath(const A &x, ComponentPath &components) {
     return genarr(x, components);
   }
 
   CC genarr(const Fortran::evaluate::StaticDataObject::Pointer &,
             ComponentPath &components) {
     fir::emitFatalError(getLoc(), "substring of static array object");
   }
 
CC genarr(const Fortran::evaluate::StructureConstructor &x) {
     TODO(getLoc(), "structure constructor");
     return [](IterSpace iters) -> ExtValue { return mlir::Value{}; };
   }
 
   //===--------------------------------------------------------------------===//
   // LOCICAL operators (.NOT., .AND., .EQV., etc.)
   //===--------------------------------------------------------------------===//
 
-  template <int KIND>
-  CC genarr(const Fortran::evaluate::Not<KIND> &x) {
+  template <int KIND> CC genarr(const Fortran::evaluate::Not<KIND> &x) {
     mlir::Location loc = getLoc();
     mlir::IntegerType i1Ty = builder.getI1Type();
     auto lambda = genarr(x.left());
     mlir::Value truth = builder.createBool(loc, true);
     return [=](IterSpace iters) -> ExtValue {
       mlir::Value logical = fir::getBase(lambda(iters));
       mlir::Value val = builder.createConvert(loc, i1Ty, logical);
       return builder.create<mlir::arith::XOrIOp>(loc, val, truth);
     };
   }
-  template <typename OP, typename A>
-  CC createBinaryBoolOp(const A &x) {
+  template <typename OP, typename A> CC createBinaryBoolOp(const A &x) {
     mlir::Location loc = getLoc();
     mlir::IntegerType i1Ty = builder.getI1Type();
     auto lf = genarr(x.left());
     auto rf = genarr(x.right());
     return [=](IterSpace iters) -> ExtValue {
       mlir::Value left = fir::getBase(lf(iters));
       mlir::Value right = fir::getBase(rf(iters));
       mlir::Value lhs = builder.createConvert(loc, i1Ty, left);
CC genarr(const Fortran::evaluate::Relational<Fortran::evaluate::Type<
                 Fortran::common::TypeCategory::Complex, KIND>> &x) {
     return createCompareOp<fir::CmpcOp>(translateFloatRelational(x.opr), x);
   }
   CC genarr(
       const Fortran::evaluate::Relational<Fortran::evaluate::SomeType> &r) {
     return std::visit([&](const auto &x) { return genarr(x); }, r.u);
   }
 
-  template <typename A>
-  CC genarr(const Fortran::evaluate::Designator<A> &des) {
+  template <typename A> CC genarr(const Fortran::evaluate::Designator<A> &des) {
     ComponentPath components(des.Rank() > 0);
     return std::visit([&](const auto &x) { return genarr(x, components); },
                       des.u);
   }
 
   /// Is the path component rank > 0?
   static bool ranked(const PathComponent &x) {
     return std::visit(Fortran::common::visitors{

   }
   CC genImplicitArrayAccess(const Fortran::evaluate::NamedEntity &x,
                             ComponentPath &components) {
     if (x.IsSymbol())
       return genImplicitArrayAccess(getFirstSym(x), components);
     return genImplicitArrayAccess(x.GetComponent(), components);
   }
 
-  template <typename A>
-  CC genAsScalar(const A &x) {
+  template <typename A> CC genAsScalar(const A &x) {
     mlir::Location loc = getLoc();
     if (isProjectedCopyInCopyOut()) {
       return [=, &x, builder = &converter.getFirOpBuilder()](
                  IterSpace iters) -> ExtValue {
         ExtValue exv = asScalarRef(x);
         mlir::Value val = fir::getBase(exv);
         mlir::Type eleTy = fir::unwrapRefType(val.getType());
         if (isAdjustedArrayElementType(eleTy)) {
inline bool isProjectedCopyInCopyOut() {
     return semant == ConstituentSemantics::ProjectedCopyInCopyOut;
   }
 
   // ???: Do we still need this?
   inline bool isCustomCopyInCopyOut() {
     return semant == ConstituentSemantics::CustomCopyInCopyOut;
   }
 
+  /// Are we lowering in a left-hand side context?
+  inline bool isLeftHandSide() {
+    return isCopyInCopyOut() || isProjectedCopyInCopyOut() ||
+           isCustomCopyInCopyOut();
+  }
+
   /// Array appears in a context where it must be boxed.
   inline bool isBoxValue() { return semant == ConstituentSemantics::BoxValue; }
 
   /// Array appears in a context where differences in the memory reference can
   /// be observable in the computational results. For example, an array
   /// element is passed to an impure procedure.
   inline bool isReferentiallyOpaque() {
     return semant == ConstituentSemantics::RefOpaque;

flang/test/Lower/forall/array-constructor.f90

+! RUN: bbc -emit-fir %s -o - | FileCheck %s
+
+subroutine ac1(arr,n)
+  integer :: arr(:), n
+  forall (i=1:n:2)
+     arr(i:i+2) = func((/i/))
+  end forall
+contains
+   pure integer function func(a)
+    integer, intent(in) :: a(:)
+    func = a(1)
+  end function func
+end subroutine ac1
+
+! CHECK-LABEL: func @_QPac1(
+! CHECK-SAME:               %[[VAL_0:.*]]: !fir.box<!fir.array<?xi32>> {fir.bindc_name = "arr"},
+! CHECK-SAME:               %[[VAL_1:.*]]: !fir.ref<i32> {fir.bindc_name = "n"}) {
+! CHECK:         %[[VAL_2:.*]] = fir.alloca index {bindc_name = ".buff.pos"}
+! CHECK:         %[[VAL_3:.*]] = fir.alloca index {bindc_name = ".buff.size"}
+! CHECK:         %[[VAL_4:.*]] = fir.alloca i32 {adapt.valuebyref, bindc_name = "i"}
+! CHECK:         %[[VAL_5:.*]] = arith.constant 1 : i32
+! CHECK:         %[[VAL_6:.*]] = fir.convert %[[VAL_5]] : (i32) -> index
+! CHECK:         %[[VAL_7:.*]] = fir.load %[[VAL_1]] : !fir.ref<i32>
+! CHECK:         %[[VAL_8:.*]] = fir.convert %[[VAL_7]] : (i32) -> index
+! CHECK:         %[[VAL_9:.*]] = arith.constant 2 : i32
+! CHECK:         %[[VAL_10:.*]] = fir.convert %[[VAL_9]] : (i32) -> index
+! CHECK:         %[[VAL_11:.*]] = fir.array_load %[[VAL_0]] : (!fir.box<!fir.array<?xi32>>) -> !fir.array<?xi32>
+! CHECK:         %[[VAL_12:.*]] = fir.do_loop %[[VAL_13:.*]] = %[[VAL_6]] to %[[VAL_8]] step %[[VAL_10]] unordered iter_args(%[[VAL_14:.*]] = %[[VAL_11]]) -> (!fir.array<?xi32>) {
+! CHECK:           %[[VAL_15:.*]] = fir.convert %[[VAL_13]] : (index) -> i32
+! CHECK:           fir.store %[[VAL_15]] to %[[VAL_4]] : !fir.ref<i32>
+! CHECK:           %[[VAL_16:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_17:.*]] = fir.load %[[VAL_4]] : !fir.ref<i32>
+! CHECK:           %[[VAL_18:.*]] = fir.convert %[[VAL_17]] : (i32) -> i64
+! CHECK:           %[[VAL_19:.*]] = fir.convert %[[VAL_18]] : (i64) -> index
+! CHECK:           %[[VAL_20:.*]] = arith.constant 1 : i64
+! CHECK:           %[[VAL_21:.*]] = fir.convert %[[VAL_20]] : (i64) -> index
+! CHECK:           %[[VAL_22:.*]] = fir.load %[[VAL_4]] : !fir.ref<i32>
+! CHECK:           %[[VAL_23:.*]] = arith.constant 2 : i32
+! CHECK:           %[[VAL_24:.*]] = arith.addi %[[VAL_22]], %[[VAL_23]] : i32
+! CHECK:           %[[VAL_25:.*]] = fir.convert %[[VAL_24]] : (i32) -> i64
+! CHECK:           %[[VAL_26:.*]] = fir.convert %[[VAL_25]] : (i64) -> index
+! CHECK:           %[[VAL_27:.*]] = arith.constant 0 : index
+! CHECK:           %[[VAL_28:.*]] = arith.subi %[[VAL_26]], %[[VAL_19]] : index
+! CHECK:           %[[VAL_29:.*]] = arith.addi %[[VAL_28]], %[[VAL_21]] : index
+! CHECK:           %[[VAL_30:.*]] = arith.divsi %[[VAL_29]], %[[VAL_21]] : index
+! CHECK:           %[[VAL_31:.*]] = arith.cmpi sgt, %[[VAL_30]], %[[VAL_27]] : index
+! CHECK:           %[[VAL_32:.*]] = arith.select %[[VAL_31]], %[[VAL_30]], %[[VAL_27]] : index
+! CHECK:           %[[VAL_33:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_34:.*]] = arith.constant 0 : index
+! CHECK:           fir.store %[[VAL_34]] to %[[VAL_2]] : !fir.ref<index>
+! CHECK:           %[[VAL_35:.*]] = fir.allocmem !fir.array<1xi32>
+! CHECK:           %[[VAL_36:.*]] = arith.constant 1 : index
+! CHECK:           fir.store %[[VAL_36]] to %[[VAL_3]] : !fir.ref<index>
+! CHECK:           %[[VAL_37:.*]] = fir.load %[[VAL_4]] : !fir.ref<i32>
+! CHECK:           %[[VAL_38:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_39:.*]] = fir.zero_bits !fir.ref<!fir.array<1xi32>>
+! CHECK:           %[[VAL_40:.*]] = fir.coordinate_of %[[VAL_39]], %[[VAL_38]] : (!fir.ref<!fir.array<1xi32>>, index) -> !fir.ref<i32>
+! CHECK:           %[[VAL_41:.*]] = fir.convert %[[VAL_40]] : (!fir.ref<i32>) -> index
+! CHECK:           %[[VAL_42:.*]] = fir.load %[[VAL_2]] : !fir.ref<index>
+! CHECK:           %[[VAL_43:.*]] = fir.load %[[VAL_3]] : !fir.ref<index>
+! CHECK:           %[[VAL_44:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_45:.*]] = arith.addi %[[VAL_42]], %[[VAL_44]] : index
+! CHECK:           %[[VAL_46:.*]] = arith.cmpi sle, %[[VAL_43]], %[[VAL_45]] : index
+! CHECK:           %[[VAL_47:.*]] = fir.if %[[VAL_46]] -> (!fir.heap<!fir.array<1xi32>>) {
+! CHECK:             %[[VAL_48:.*]] = arith.constant 2 : index
+! CHECK:             %[[VAL_49:.*]] = arith.muli %[[VAL_45]], %[[VAL_48]] : index
+! CHECK:             fir.store %[[VAL_49]] to %[[VAL_3]] : !fir.ref<index>
+! CHECK:             %[[VAL_50:.*]] = arith.muli %[[VAL_49]], %[[VAL_41]] : index
+! CHECK:             %[[VAL_51:.*]] = fir.convert %[[VAL_35]] : (!fir.heap<!fir.array<1xi32>>) -> !fir.ref<i8>
+! CHECK:             %[[VAL_52:.*]] = fir.convert %[[VAL_50]] : (index) -> i64
+! CHECK:             %[[VAL_53:.*]] = fir.call @realloc(%[[VAL_51]], %[[VAL_52]]) : (!fir.ref<i8>, i64) -> !fir.ref<i8>
+! CHECK:             %[[VAL_54:.*]] = fir.convert %[[VAL_53]] : (!fir.ref<i8>) -> !fir.heap<!fir.array<1xi32>>
+! CHECK:             fir.result %[[VAL_54]] : !fir.heap<!fir.array<1xi32>>
+! CHECK:           } else {
+! CHECK:             fir.result %[[VAL_35]] : !fir.heap<!fir.array<1xi32>>
+! CHECK:           }
+! CHECK:           %[[VAL_55:.*]] = fir.coordinate_of %[[VAL_56:.*]], %[[VAL_42]] : (!fir.heap<!fir.array<1xi32>>, index) -> !fir.ref<i32>
+! CHECK:           fir.store %[[VAL_37]] to %[[VAL_55]] : !fir.ref<i32>
+! CHECK:           fir.store %[[VAL_45]] to %[[VAL_2]] : !fir.ref<index>
+! CHECK:           %[[VAL_57:.*]] = fir.load %[[VAL_2]] : !fir.ref<index>
+! CHECK:           %[[VAL_58:.*]] = fir.shape %[[VAL_57]] : (index) -> !fir.shape<1>
+! CHECK:           %[[VAL_59:.*]] = fir.array_load %[[VAL_56]](%[[VAL_58]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.array<1xi32>
+! CHECK:           %[[VAL_60:.*]] = fir.allocmem !fir.array<1xi32>
+! CHECK:           %[[VAL_61:.*]] = fir.shape %[[VAL_33]] : (index) -> !fir.shape<1>
+! CHECK:           %[[VAL_62:.*]] = fir.array_load %[[VAL_60]](%[[VAL_61]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.array<1xi32>
+! CHECK:           %[[VAL_63:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_64:.*]] = arith.constant 0 : index
+! CHECK:           %[[VAL_65:.*]] = arith.subi %[[VAL_33]], %[[VAL_63]] : index
+! CHECK:           %[[VAL_66:.*]] = fir.do_loop %[[VAL_67:.*]] = %[[VAL_64]] to %[[VAL_65]] step %[[VAL_63]] unordered iter_args(%[[VAL_68:.*]] = %[[VAL_62]]) -> (!fir.array<1xi32>) {
+! CHECK:             %[[VAL_69:.*]] = fir.array_fetch %[[VAL_59]], %[[VAL_67]] : (!fir.array<1xi32>, index) -> i32
+! CHECK:             %[[VAL_70:.*]] = fir.array_update %[[VAL_68]], %[[VAL_69]], %[[VAL_67]] : (!fir.array<1xi32>, i32, index) -> !fir.array<1xi32>
+! CHECK:             fir.result %[[VAL_70]] : !fir.array<1xi32>
+! CHECK:           }
+! CHECK:           fir.array_merge_store %[[VAL_62]], %[[VAL_71:.*]] to %[[VAL_60]] : !fir.array<1xi32>, !fir.array<1xi32>, !fir.heap<!fir.array<1xi32>>
+! CHECK:           %[[VAL_72:.*]] = fir.shape %[[VAL_33]] : (index) -> !fir.shape<1>
+! CHECK:           %[[VAL_73:.*]] = fir.embox %[[VAL_60]](%[[VAL_72]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<1xi32>>
+! CHECK:           %[[VAL_74:.*]] = fir.convert %[[VAL_73]] : (!fir.box<!fir.array<1xi32>>) -> !fir.box<!fir.array<?xi32>>
+! CHECK:           %[[VAL_75:.*]] = fir.call @_QFac1Pfunc(%[[VAL_74]]) : (!fir.box<!fir.array<?xi32>>) -> i32
+! CHECK:           %[[VAL_76:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_77:.*]] = arith.constant 0 : index
+! CHECK:           %[[VAL_78:.*]] = arith.subi %[[VAL_32]], %[[VAL_76]] : index
+! CHECK:           %[[VAL_79:.*]] = fir.do_loop %[[VAL_80:.*]] = %[[VAL_77]] to %[[VAL_78]] step %[[VAL_76]] unordered iter_args(%[[VAL_81:.*]] = %[[VAL_14]]) -> (!fir.array<?xi32>) {
+! CHECK:             %[[VAL_82:.*]] = arith.subi %[[VAL_19]], %[[VAL_16]] : index
+! CHECK:             %[[VAL_83:.*]] = arith.muli %[[VAL_80]], %[[VAL_21]] : index
+! CHECK:             %[[VAL_84:.*]] = arith.addi %[[VAL_82]], %[[VAL_83]] : index
+! CHECK:             %[[VAL_85:.*]] = fir.array_update %[[VAL_81]], %[[VAL_75]], %[[VAL_84]] : (!fir.array<?xi32>, i32, index) -> !fir.array<?xi32>
+! CHECK:             fir.result %[[VAL_85]] : !fir.array<?xi32>
+! CHECK:           }
+! CHECK:           fir.freemem %[[VAL_60]] : !fir.heap<!fir.array<1xi32>>
+! CHECK:           fir.freemem %[[VAL_56]] : !fir.heap<!fir.array<1xi32>>
+! CHECK:           fir.result %[[VAL_86:.*]] : !fir.array<?xi32>
+! CHECK:         }
+! CHECK:         fir.array_merge_store %[[VAL_11]], %[[VAL_87:.*]] to %[[VAL_0]] : !fir.array<?xi32>, !fir.array<?xi32>, !fir.box<!fir.array<?xi32>>
+! CHECK:         return
+! CHECK:       }
+
+! CHECK-LABEL: func @_QFac1Pfunc(
+! CHECK-SAME:                    %[[VAL_0:.*]]: !fir.box<!fir.array<?xi32>> {fir.bindc_name = "a"}) -> i32 {
+! CHECK:         %[[VAL_1:.*]] = fir.alloca i32 {bindc_name = "func", uniq_name = "_QFfuncEfunc"}
+! CHECK:         %[[VAL_2:.*]] = arith.constant 1 : i64
+! CHECK:         %[[VAL_3:.*]] = arith.constant 1 : i64
+! CHECK:         %[[VAL_4:.*]] = arith.subi %[[VAL_2]], %[[VAL_3]] : i64
+! CHECK:         %[[VAL_5:.*]] = fir.coordinate_of %[[VAL_0]], %[[VAL_4]] : (!fir.box<!fir.array<?xi32>>, i64) -> !fir.ref<i32>
+! CHECK:         %[[VAL_6:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
+! CHECK:         fir.store %[[VAL_6]] to %[[VAL_1]] : !fir.ref<i32>
+! CHECK:         %[[VAL_7:.*]] = fir.load %[[VAL_1]] : !fir.ref<i32>
+! CHECK:         return %[[VAL_7]] : i32
+! CHECK:       }
+
+subroutine ac2(arr,n)
+  integer :: arr(:), n
+  forall (i=1:n:2)
+     arr(i:i+2) = func((/i/))
+  end forall
+contains
+  pure function func(a)
+    integer :: func(3)
+    integer, intent(in) :: a(:)
+    func = a(1:3)
+  end function func
+end subroutine ac2
+
+! CHECK-LABEL: func @_QPac2(
+! CHECK-SAME:               %[[VAL_0:.*]]: !fir.box<!fir.array<?xi32>> {fir.bindc_name = "arr"},
+! CHECK-SAME:               %[[VAL_1:.*]]: !fir.ref<i32> {fir.bindc_name = "n"}) {
+! CHECK:         %[[VAL_2:.*]] = fir.alloca !fir.array<3xi32> {bindc_name = ".result"}
+! CHECK:         %[[VAL_3:.*]] = fir.alloca index {bindc_name = ".buff.pos"}
+! CHECK:         %[[VAL_4:.*]] = fir.alloca index {bindc_name = ".buff.size"}
+! CHECK:         %[[VAL_5:.*]] = fir.alloca i32 {adapt.valuebyref, bindc_name = "i"}
+! CHECK:         %[[VAL_6:.*]] = arith.constant 1 : i32
+! CHECK:         %[[VAL_7:.*]] = fir.convert %[[VAL_6]] : (i32) -> index
+! CHECK:         %[[VAL_8:.*]] = fir.load %[[VAL_1]] : !fir.ref<i32>
+! CHECK:         %[[VAL_9:.*]] = fir.convert %[[VAL_8]] : (i32) -> index
+! CHECK:         %[[VAL_10:.*]] = arith.constant 2 : i32
+! CHECK:         %[[VAL_11:.*]] = fir.convert %[[VAL_10]] : (i32) -> index
+! CHECK:         %[[VAL_12:.*]] = fir.array_load %[[VAL_0]] : (!fir.box<!fir.array<?xi32>>) -> !fir.array<?xi32>
+! CHECK:         %[[VAL_13:.*]] = fir.do_loop %[[VAL_14:.*]] = %[[VAL_7]] to %[[VAL_9]] step %[[VAL_11]] unordered iter_args(%[[VAL_15:.*]] = %[[VAL_12]]) -> (!fir.array<?xi32>) {
+! CHECK:           %[[VAL_16:.*]] = fir.convert %[[VAL_14]] : (index) -> i32
+! CHECK:           fir.store %[[VAL_16]] to %[[VAL_5]] : !fir.ref<i32>
+! CHECK:           %[[VAL_17:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_18:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
+! CHECK:           %[[VAL_19:.*]] = fir.convert %[[VAL_18]] : (i32) -> i64
+! CHECK:           %[[VAL_20:.*]] = fir.convert %[[VAL_19]] : (i64) -> index
+! CHECK:           %[[VAL_21:.*]] = arith.constant 1 : i64
+! CHECK:           %[[VAL_22:.*]] = fir.convert %[[VAL_21]] : (i64) -> index
+! CHECK:           %[[VAL_23:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
+! CHECK:           %[[VAL_24:.*]] = arith.constant 2 : i32
+! CHECK:           %[[VAL_25:.*]] = arith.addi %[[VAL_23]], %[[VAL_24]] : i32
+! CHECK:           %[[VAL_26:.*]] = fir.convert %[[VAL_25]] : (i32) -> i64
+! CHECK:           %[[VAL_27:.*]] = fir.convert %[[VAL_26]] : (i64) -> index
+! CHECK:           %[[VAL_28:.*]] = arith.constant 0 : index
+! CHECK:           %[[VAL_29:.*]] = arith.subi %[[VAL_27]], %[[VAL_20]] : index
+! CHECK:           %[[VAL_30:.*]] = arith.addi %[[VAL_29]], %[[VAL_22]] : index
+! CHECK:           %[[VAL_31:.*]] = arith.divsi %[[VAL_30]], %[[VAL_22]] : index
+! CHECK:           %[[VAL_32:.*]] = arith.cmpi sgt, %[[VAL_31]], %[[VAL_28]] : index
+! CHECK:           %[[VAL_33:.*]] = arith.select %[[VAL_32]], %[[VAL_31]], %[[VAL_28]] : index
+! CHECK:           %[[VAL_34:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_35:.*]] = arith.constant 0 : index
+! CHECK:           fir.store %[[VAL_35]] to %[[VAL_3]] : !fir.ref<index>
+! CHECK:           %[[VAL_36:.*]] = fir.allocmem !fir.array<1xi32>
+! CHECK:           %[[VAL_37:.*]] = arith.constant 1 : index
+! CHECK:           fir.store %[[VAL_37]] to %[[VAL_4]] : !fir.ref<index>
+! CHECK:           %[[VAL_38:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
+! CHECK:           %[[VAL_39:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_40:.*]] = fir.zero_bits !fir.ref<!fir.array<1xi32>>
+! CHECK:           %[[VAL_41:.*]] = fir.coordinate_of %[[VAL_40]], %[[VAL_39]] : (!fir.ref<!fir.array<1xi32>>, index) -> !fir.ref<i32>
+! CHECK:           %[[VAL_42:.*]] = fir.convert %[[VAL_41]] : (!fir.ref<i32>) -> index
+! CHECK:           %[[VAL_43:.*]] = fir.load %[[VAL_3]] : !fir.ref<index>
+! CHECK:           %[[VAL_44:.*]] = fir.load %[[VAL_4]] : !fir.ref<index>
+! CHECK:           %[[VAL_45:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_46:.*]] = arith.addi %[[VAL_43]], %[[VAL_45]] : index
+! CHECK:           %[[VAL_47:.*]] = arith.cmpi sle, %[[VAL_44]], %[[VAL_46]] : index
+! CHECK:           %[[VAL_48:.*]] = fir.if %[[VAL_47]] -> (!fir.heap<!fir.array<1xi32>>) {
+! CHECK:             %[[VAL_49:.*]] = arith.constant 2 : index
+! CHECK:             %[[VAL_50:.*]] = arith.muli %[[VAL_46]], %[[VAL_49]] : index
+! CHECK:             fir.store %[[VAL_50]] to %[[VAL_4]] : !fir.ref<index>
+! CHECK:             %[[VAL_51:.*]] = arith.muli %[[VAL_50]], %[[VAL_42]] : index
+! CHECK:             %[[VAL_52:.*]] = fir.convert %[[VAL_36]] : (!fir.heap<!fir.array<1xi32>>) -> !fir.ref<i8>
+! CHECK:             %[[VAL_53:.*]] = fir.convert %[[VAL_51]] : (index) -> i64
+! CHECK:             %[[VAL_54:.*]] = fir.call @realloc(%[[VAL_52]], %[[VAL_53]]) : (!fir.ref<i8>, i64) -> !fir.ref<i8>
+! CHECK:             %[[VAL_55:.*]] = fir.convert %[[VAL_54]] : (!fir.ref<i8>) -> !fir.heap<!fir.array<1xi32>>
+! CHECK:             fir.result %[[VAL_55]] : !fir.heap<!fir.array<1xi32>>
+! CHECK:           } else {
+! CHECK:             fir.result %[[VAL_36]] : !fir.heap<!fir.array<1xi32>>
+! CHECK:           }
+! CHECK:           %[[VAL_56:.*]] = fir.coordinate_of %[[VAL_57:.*]], %[[VAL_43]] : (!fir.heap<!fir.array<1xi32>>, index) -> !fir.ref<i32>
+! CHECK:           fir.store %[[VAL_38]] to %[[VAL_56]] : !fir.ref<i32>
+! CHECK:           fir.store %[[VAL_46]] to %[[VAL_3]] : !fir.ref<index>
+! CHECK:           %[[VAL_58:.*]] = fir.load %[[VAL_3]] : !fir.ref<index>
+! CHECK:           %[[VAL_59:.*]] = fir.shape %[[VAL_58]] : (index) -> !fir.shape<1>
+! CHECK:           %[[VAL_60:.*]] = fir.array_load %[[VAL_57]](%[[VAL_59]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.array<1xi32>
+! CHECK:           %[[VAL_61:.*]] = fir.allocmem !fir.array<1xi32>
+! CHECK:           %[[VAL_62:.*]] = fir.shape %[[VAL_34]] : (index) -> !fir.shape<1>
+! CHECK:           %[[VAL_63:.*]] = fir.array_load %[[VAL_61]](%[[VAL_62]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.array<1xi32>
+! CHECK:           %[[VAL_64:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_65:.*]] = arith.constant 0 : index
+! CHECK:           %[[VAL_66:.*]] = arith.subi %[[VAL_34]], %[[VAL_64]] : index
+! CHECK:           %[[VAL_67:.*]] = fir.do_loop %[[VAL_68:.*]] = %[[VAL_65]] to %[[VAL_66]] step %[[VAL_64]] unordered iter_args(%[[VAL_69:.*]] = %[[VAL_63]]) -> (!fir.array<1xi32>) {
+! CHECK:             %[[VAL_70:.*]] = fir.array_fetch %[[VAL_60]], %[[VAL_68]] : (!fir.array<1xi32>, index) -> i32
+! CHECK:             %[[VAL_71:.*]] = fir.array_update %[[VAL_69]], %[[VAL_70]], %[[VAL_68]] : (!fir.array<1xi32>, i32, index) -> !fir.array<1xi32>
+! CHECK:             fir.result %[[VAL_71]] : !fir.array<1xi32>
+! CHECK:           }
+! CHECK:           fir.array_merge_store %[[VAL_63]], %[[VAL_72:.*]] to %[[VAL_61]] : !fir.array<1xi32>, !fir.array<1xi32>, !fir.heap<!fir.array<1xi32>>
+! CHECK:           %[[VAL_73:.*]] = fir.shape %[[VAL_34]] : (index) -> !fir.shape<1>
+! CHECK:           %[[VAL_74:.*]] = fir.embox %[[VAL_61]](%[[VAL_73]]) : (!fir.heap<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<1xi32>>
+! CHECK:           %[[VAL_75:.*]] = arith.constant 3 : i64
+! CHECK:           %[[VAL_76:.*]] = arith.constant 1 : i64
+! CHECK:           %[[VAL_77:.*]] = arith.subi %[[VAL_75]], %[[VAL_76]] : i64
+! CHECK:           %[[VAL_78:.*]] = arith.constant 1 : i64
+! CHECK:           %[[VAL_79:.*]] = arith.addi %[[VAL_77]], %[[VAL_78]] : i64
+! CHECK:           %[[VAL_80:.*]] = fir.convert %[[VAL_79]] : (i64) -> index
+! CHECK:           %[[VAL_81:.*]] = fir.call @llvm.stacksave() : () -> !fir.ref<i8>
+! CHECK:           %[[VAL_82:.*]] = fir.shape %[[VAL_80]] : (index) -> !fir.shape<1>
+! CHECK:           %[[VAL_83:.*]] = fir.convert %[[VAL_74]] : (!fir.box<!fir.array<1xi32>>) -> !fir.box<!fir.array<?xi32>>
+! CHECK:           %[[VAL_84:.*]] = fir.call @_QFac2Pfunc(%[[VAL_83]]) : (!fir.box<!fir.array<?xi32>>) -> !fir.array<3xi32>
+! CHECK:           fir.save_result %[[VAL_84]] to %[[VAL_2]](%[[VAL_82]]) : !fir.array<3xi32>, !fir.ref<!fir.array<3xi32>>, !fir.shape<1>
+! CHECK:           %[[VAL_85:.*]] = fir.shape %[[VAL_80]] : (index) -> !fir.shape<1>
+! CHECK:           %[[VAL_86:.*]] = fir.array_load %[[VAL_2]](%[[VAL_85]]) : (!fir.ref<!fir.array<3xi32>>, !fir.shape<1>) -> !fir.array<3xi32>
+! CHECK:           %[[VAL_87:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_88:.*]] = arith.constant 0 : index
+! CHECK:           %[[VAL_89:.*]] = arith.subi %[[VAL_33]], %[[VAL_87]] : index
+! CHECK:           %[[VAL_90:.*]] = fir.do_loop %[[VAL_91:.*]] = %[[VAL_88]] to %[[VAL_89]] step %[[VAL_87]] unordered iter_args(%[[VAL_92:.*]] = %[[VAL_15]]) -> (!fir.array<?xi32>) {
+! CHECK:             %[[VAL_93:.*]] = fir.array_fetch %[[VAL_86]], %[[VAL_91]] : (!fir.array<3xi32>, index) -> i32
+! CHECK:             %[[VAL_94:.*]] = arith.subi %[[VAL_20]], %[[VAL_17]] : index
+! CHECK:             %[[VAL_95:.*]] = arith.muli %[[VAL_91]], %[[VAL_22]] : index
+! CHECK:             %[[VAL_96:.*]] = arith.addi %[[VAL_94]], %[[VAL_95]] : index
+! CHECK:             %[[VAL_97:.*]] = fir.array_update %[[VAL_92]], %[[VAL_93]], %[[VAL_96]] : (!fir.array<?xi32>, i32, index) -> !fir.array<?xi32>
+! CHECK:             fir.result %[[VAL_97]] : !fir.array<?xi32>
+! CHECK:           }
+! CHECK:           fir.call @llvm.stackrestore(%[[VAL_81]]) : (!fir.ref<i8>) -> ()
+! CHECK:           fir.freemem %[[VAL_61]] : !fir.heap<!fir.array<1xi32>>
+! CHECK:           fir.freemem %[[VAL_57]] : !fir.heap<!fir.array<1xi32>>
+! CHECK:           fir.result %[[VAL_98:.*]] : !fir.array<?xi32>
+! CHECK:         }
+! CHECK:         fir.array_merge_store %[[VAL_12]], %[[VAL_99:.*]] to %[[VAL_0]] : !fir.array<?xi32>, !fir.array<?xi32>, !fir.box<!fir.array<?xi32>>
+! CHECK:         return
+! CHECK:       }
+
+! CHECK-LABEL: func @_QFac2Pfunc(
+! CHECK-SAME:                    %[[VAL_0:.*]]: !fir.box<!fir.array<?xi32>> {fir.bindc_name = "a"}) -> !fir.array<3xi32> {
+! CHECK:         %[[VAL_1:.*]] = arith.constant 3 : index
+! CHECK:         %[[VAL_2:.*]] = fir.alloca !fir.array<3xi32> {bindc_name = "func", uniq_name = "_QFfuncEfunc"}
+! CHECK:         %[[VAL_3:.*]] = fir.shape %[[VAL_1]] : (index) -> !fir.shape<1>
+! CHECK:         %[[VAL_4:.*]] = fir.array_load %[[VAL_2]](%[[VAL_3]]) : (!fir.ref<!fir.array<3xi32>>, !fir.shape<1>) -> !fir.array<3xi32>
+! CHECK:         %[[VAL_5:.*]] = arith.constant 1 : i64
+! CHECK:         %[[VAL_6:.*]] = fir.convert %[[VAL_5]] : (i64) -> index
+! CHECK:         %[[VAL_7:.*]] = arith.constant 1 : i64
+! CHECK:         %[[VAL_8:.*]] = fir.convert %[[VAL_7]] : (i64) -> index
+! CHECK:         %[[VAL_9:.*]] = arith.constant 3 : i64
+! CHECK:         %[[VAL_10:.*]] = fir.convert %[[VAL_9]] : (i64) -> index
+! CHECK:         %[[VAL_11:.*]] = fir.slice %[[VAL_6]], %[[VAL_10]], %[[VAL_8]] : (index, index, index) -> !fir.slice<1>
+! CHECK:         %[[VAL_12:.*]] = fir.array_load %[[VAL_0]] {{\[}}%[[VAL_11]]] : (!fir.box<!fir.array<?xi32>>, !fir.slice<1>) -> !fir.array<?xi32>
+! CHECK:         %[[VAL_13:.*]] = arith.constant 1 : index
+! CHECK:         %[[VAL_14:.*]] = arith.constant 0 : index
+! CHECK:         %[[VAL_15:.*]] = arith.subi %[[VAL_1]], %[[VAL_13]] : index
+! CHECK:         %[[VAL_16:.*]] = fir.do_loop %[[VAL_17:.*]] = %[[VAL_14]] to %[[VAL_15]] step %[[VAL_13]] unordered iter_args(%[[VAL_18:.*]] = %[[VAL_4]]) -> (!fir.array<3xi32>) {
+! CHECK:           %[[VAL_19:.*]] = fir.array_fetch %[[VAL_12]], %[[VAL_17]] : (!fir.array<?xi32>, index) -> i32
+! CHECK:           %[[VAL_20:.*]] = fir.array_update %[[VAL_18]], %[[VAL_19]], %[[VAL_17]] : (!fir.array<3xi32>, i32, index) -> !fir.array<3xi32>
+! CHECK:           fir.result %[[VAL_20]] : !fir.array<3xi32>
+! CHECK:         }
+! CHECK:         fir.array_merge_store %[[VAL_4]], %[[VAL_21:.*]] to %[[VAL_2]] : !fir.array<3xi32>, !fir.array<3xi32>, !fir.ref<!fir.array<3xi32>>
+! CHECK:         %[[VAL_22:.*]] = fir.load %[[VAL_2]] : !fir.ref<!fir.array<3xi32>>
+! CHECK:         return %[[VAL_22]] : !fir.array<3xi32>
+! CHECK:       }

flang/test/Lower/forall/forall-2.f90

 ! CHECK:           %[[VAL_25:.*]] = arith.addi %[[VAL_24]], %[[VAL_19]] : index
 ! CHECK:           %[[VAL_26:.*]] = arith.divsi %[[VAL_25]], %[[VAL_19]] : index
 ! CHECK:           %[[VAL_27:.*]] = arith.cmpi sgt, %[[VAL_26]], %[[VAL_23]] : index
 ! CHECK:           %[[VAL_28:.*]] = arith.select %[[VAL_27]], %[[VAL_26]], %[[VAL_23]] : index
 ! CHECK:           %[[VAL_29:.*]] = fir.load %[[VAL_0]] : !fir.ref<i32>
 ! CHECK:           %[[VAL_30:.*]] = fir.convert %[[VAL_29]] : (i32) -> i64
 ! CHECK:           %[[VAL_31:.*]] = fir.convert %[[VAL_30]] : (i64) -> index
 ! CHECK:           %[[VAL_32:.*]] = arith.subi %[[VAL_31]], %[[VAL_15]] : index
+! CHECK:           %[[VAL_39:.*]] = fir.load %[[VAL_0]] : !fir.ref<i32>
+! CHECK:           %[[VAL_40:.*]] = arith.constant 0 : i32
+! CHECK:           %[[VAL_41:.*]] = arith.subi %[[VAL_40]], %[[VAL_39]] : i32
 ! CHECK:           %[[VAL_33:.*]] = arith.constant 1 : index
 ! CHECK:           %[[VAL_34:.*]] = arith.constant 0 : index
 ! CHECK:           %[[VAL_35:.*]] = arith.subi %[[VAL_28]], %[[VAL_33]] : index
 ! CHECK:           %[[VAL_36:.*]] = fir.do_loop %[[VAL_37:.*]] = %[[VAL_34]] to %[[VAL_35]] step %[[VAL_33]] unordered iter_args(%[[VAL_38:.*]] = %[[VAL_13]]) -> (!fir.array<10x10xi32>) {
-! CHECK:             %[[VAL_39:.*]] = fir.load %[[VAL_0]] : !fir.ref<i32>
-! CHECK:             %[[VAL_40:.*]] = arith.constant 0 : i32
-! CHECK:             %[[VAL_41:.*]] = arith.subi %[[VAL_40]], %[[VAL_39]] : i32
 ! CHECK:             %[[VAL_42:.*]] = arith.subi %[[VAL_17]], %[[VAL_15]] : index
 ! CHECK:             %[[VAL_43:.*]] = arith.muli %[[VAL_37]], %[[VAL_19]] : index
 ! CHECK:             %[[VAL_44:.*]] = arith.addi %[[VAL_42]], %[[VAL_43]] : index
 ! CHECK:             %[[VAL_45:.*]] = fir.array_update %[[VAL_38]], %[[VAL_41]], %[[VAL_44]], %[[VAL_32]] : (!fir.array<10x10xi32>, i32, index, index) -> !fir.array<10x10xi32>
 ! CHECK:             fir.result %[[VAL_45]] : !fir.array<10x10xi32>
 ! CHECK:           }
 ! CHECK:           fir.result %[[VAL_46:.*]] : !fir.array<10x10xi32>
 ! CHECK:         }

 ! CHECK:         %[[VAL_8:.*]] = arith.constant 1 : i32
 ! CHECK:         %[[VAL_9:.*]] = fir.convert %[[VAL_8]] : (i32) -> index
 ! CHECK:         %[[VAL_10:.*]] = arith.constant 1 : index
 ! CHECK:         %[[VAL_11:.*]] = fir.shape %[[VAL_3]] : (index) -> !fir.shape<1>
 ! CHECK:         %[[VAL_12:.*]] = fir.array_load %[[VAL_0]](%[[VAL_11]]) : (!fir.ref<!fir.array<1xi32>>, !fir.shape<1>) -> !fir.array<1xi32>
 ! CHECK:         %[[VAL_13:.*]] = fir.do_loop %[[VAL_14:.*]] = %[[VAL_7]] to %[[VAL_9]] step %[[VAL_10]] unordered iter_args(%[[VAL_15:.*]] = %[[VAL_12]]) -> (!fir.array<1xi32>) {
 ! CHECK:           %[[VAL_16:.*]] = fir.convert %[[VAL_14]] : (index) -> i32
 ! CHECK:           fir.store %[[VAL_16]] to %[[VAL_2]] : !fir.ref<i32>
-! CHECK:           %[[VAL_17:.*]] = arith.constant 1 : i32
-! CHECK:           %[[VAL_18:.*]] = arith.constant 1 : index
-! CHECK:           %[[VAL_19:.*]] = arith.constant 1 : i64
+! CHECK-DAG:       %[[VAL_18:.*]] = arith.constant 1 : index
+! CHECK-DAG:       %[[VAL_19:.*]] = arith.constant 1 : i64
 ! CHECK:           %[[VAL_20:.*]] = fir.convert %[[VAL_19]] : (i64) -> index
 ! CHECK:           %[[VAL_21:.*]] = arith.addi %[[VAL_18]], %[[VAL_4]] : index
 ! CHECK:           %[[VAL_22:.*]] = arith.subi %[[VAL_21]], %[[VAL_18]] : index
 ! CHECK:           %[[VAL_23:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
 ! CHECK:           %[[VAL_24:.*]] = fir.convert %[[VAL_23]] : (i32) -> i64
 ! CHECK:           %[[VAL_25:.*]] = fir.undefined index
 ! CHECK:           %[[VAL_26:.*]] = fir.shape %[[VAL_4]], %[[VAL_5]] : (index, index) -> !fir.shape<2>
 ! CHECK:           %[[VAL_27:.*]] = fir.slice %[[VAL_18]], %[[VAL_22]], %[[VAL_20]], %[[VAL_24]], %[[VAL_25]], %[[VAL_25]] : (index, index, index, i64, index, index) -> !fir.slice<2>
 ! CHECK:           %[[VAL_28:.*]] = fir.embox %[[VAL_1]](%[[VAL_26]]) {{\[}}%[[VAL_27]]] : (!fir.ref<!fir.array<2x2xi32>>, !fir.shape<2>, !fir.slice<2>) -> !fir.box<!fir.array<?xi32>>
 ! CHECK:           %[[VAL_29:.*]] = fir.call @_QPe(%[[VAL_28]]) : (!fir.box<!fir.array<?xi32>>) -> i32
+! CHECK:           %[[VAL_17:.*]] = arith.constant 1 : i32
 ! CHECK:           %[[VAL_30:.*]] = arith.addi %[[VAL_29]], %[[VAL_17]] : i32
 ! CHECK:           %[[VAL_31:.*]] = arith.constant 1 : index
 ! CHECK:           %[[VAL_32:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
 ! CHECK:           %[[VAL_33:.*]] = fir.convert %[[VAL_32]] : (i32) -> i64
 ! CHECK:           %[[VAL_34:.*]] = fir.convert %[[VAL_33]] : (i64) -> index
 ! CHECK:           %[[VAL_35:.*]] = arith.subi %[[VAL_34]], %[[VAL_31]] : index
 ! CHECK:           %[[VAL_36:.*]] = fir.array_update %[[VAL_15]], %[[VAL_30]], %[[VAL_35]] : (!fir.array<1xi32>, i32, index) -> !fir.array<1xi32>
 ! CHECK:           fir.result %[[VAL_36]] : !fir.array<1xi32>

flang/test/Lower/forall/forall-construct-2.f90

 ! CHECK:             %[[VAL_35:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
 ! CHECK:             %[[VAL_36:.*]] = fir.convert %[[VAL_35]] : (i32) -> i64
 ! CHECK:             %[[VAL_37:.*]] = fir.convert %[[VAL_36]] : (i64) -> index
 ! CHECK:             %[[VAL_38:.*]] = arith.subi %[[VAL_37]], %[[VAL_34]] : index
 ! CHECK:             %[[VAL_39:.*]] = fir.load %[[VAL_4]] : !fir.ref<i32>
 ! CHECK:             %[[VAL_40:.*]] = fir.convert %[[VAL_39]] : (i32) -> i64
 ! CHECK:             %[[VAL_41:.*]] = fir.convert %[[VAL_40]] : (i64) -> index
 ! CHECK:             %[[VAL_42:.*]] = arith.subi %[[VAL_41]], %[[VAL_34]] : index
+! CHECK:             %[[VAL_54:.*]] = fir.array_fetch %[[VAL_23]], %[[VAL_38]], %[[VAL_42]] : (!fir.array<200x200xf32>, index, index) -> f32
 ! CHECK:             %[[VAL_43:.*]] = arith.constant 1 : index
 ! CHECK-DAG:         %[[VAL_44:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
 ! CHECK-DAG:         %[[VAL_45:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_46:.*]] = arith.addi %[[VAL_44]], %[[VAL_45]] : i32
 ! CHECK:             %[[VAL_47:.*]] = fir.convert %[[VAL_46]] : (i32) -> i64
 ! CHECK:             %[[VAL_48:.*]] = fir.convert %[[VAL_47]] : (i64) -> index
 ! CHECK:             %[[VAL_49:.*]] = arith.subi %[[VAL_48]], %[[VAL_43]] : index
 ! CHECK:             %[[VAL_50:.*]] = fir.load %[[VAL_4]] : !fir.ref<i32>
 ! CHECK:             %[[VAL_51:.*]] = fir.convert %[[VAL_50]] : (i32) -> i64
 ! CHECK:             %[[VAL_52:.*]] = fir.convert %[[VAL_51]] : (i64) -> index
 ! CHECK:             %[[VAL_53:.*]] = arith.subi %[[VAL_52]], %[[VAL_43]] : index
-! CHECK:             %[[VAL_54:.*]] = fir.array_fetch %[[VAL_23]], %[[VAL_38]], %[[VAL_42]] : (!fir.array<200x200xf32>, index, index) -> f32
 ! CHECK:             %[[VAL_55:.*]] = fir.array_fetch %[[VAL_25]], %[[VAL_49]], %[[VAL_53]] : (!fir.array<200x200xf32>, index, index) -> f32
 ! CHECK:             %[[VAL_56:.*]] = arith.addf %[[VAL_54]], %[[VAL_55]] : f32
 ! CHECK:             %[[VAL_57:.*]] = arith.constant 1 : index
 ! CHECK:             %[[VAL_58:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
 ! CHECK:             %[[VAL_59:.*]] = fir.convert %[[VAL_58]] : (i32) -> i64
 ! CHECK:             %[[VAL_60:.*]] = fir.convert %[[VAL_59]] : (i64) -> index
 ! CHECK:             %[[VAL_61:.*]] = arith.subi %[[VAL_60]], %[[VAL_57]] : index
 ! CHECK:             %[[VAL_62:.*]] = fir.load %[[VAL_4]] : !fir.ref<i32>

flang/test/Lower/forall/forall-construct-3.f90

 ! CHECK:               %[[VAL_48:.*]] = fir.load %[[VAL_6]] : !fir.ref<i32>
 ! CHECK:               %[[VAL_49:.*]] = fir.convert %[[VAL_48]] : (i32) -> i64
 ! CHECK:               %[[VAL_50:.*]] = fir.convert %[[VAL_49]] : (i64) -> index
 ! CHECK:               %[[VAL_51:.*]] = arith.subi %[[VAL_50]], %[[VAL_47]] : index
 ! CHECK:               %[[VAL_52:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
 ! CHECK:               %[[VAL_53:.*]] = fir.convert %[[VAL_52]] : (i32) -> i64
 ! CHECK:               %[[VAL_54:.*]] = fir.convert %[[VAL_53]] : (i64) -> index
 ! CHECK:               %[[VAL_55:.*]] = arith.subi %[[VAL_54]], %[[VAL_47]] : index
+! CHECK:               %[[VAL_67:.*]] = fir.array_fetch %[[VAL_24]], %[[VAL_51]], %[[VAL_55]] : (!fir.array<200x200xf32>, index, index) -> f32
 ! CHECK:               %[[VAL_56:.*]] = arith.constant 1 : index
 ! CHECK-DAG:           %[[VAL_57:.*]] = fir.load %[[VAL_6]] : !fir.ref<i32>
 ! CHECK-DAG:           %[[VAL_58:.*]] = arith.constant 1 : i32
 ! CHECK:               %[[VAL_59:.*]] = arith.addi %[[VAL_57]], %[[VAL_58]] : i32
 ! CHECK:               %[[VAL_60:.*]] = fir.convert %[[VAL_59]] : (i32) -> i64
 ! CHECK:               %[[VAL_61:.*]] = fir.convert %[[VAL_60]] : (i64) -> index
 ! CHECK:               %[[VAL_62:.*]] = arith.subi %[[VAL_61]], %[[VAL_56]] : index
 ! CHECK:               %[[VAL_63:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>
 ! CHECK:               %[[VAL_64:.*]] = fir.convert %[[VAL_63]] : (i32) -> i64
 ! CHECK:               %[[VAL_65:.*]] = fir.convert %[[VAL_64]] : (i64) -> index
 ! CHECK:               %[[VAL_66:.*]] = arith.subi %[[VAL_65]], %[[VAL_56]] : index
-! CHECK:               %[[VAL_67:.*]] = fir.array_fetch %[[VAL_24]], %[[VAL_51]], %[[VAL_55]] : (!fir.array<200x200xf32>, index, index) -> f32
 ! CHECK:               %[[VAL_68:.*]] = fir.array_fetch %[[VAL_26]], %[[VAL_62]], %[[VAL_66]] : (!fir.array<200x200xf32>, index, index) -> f32
 ! CHECK:               %[[VAL_69:.*]] = arith.addf %[[VAL_67]], %[[VAL_68]] : f32
 ! CHECK:               %[[VAL_70:.*]] = arith.constant 1 : index
 ! CHECK:               %[[VAL_71:.*]] = fir.load %[[VAL_6]] : !fir.ref<i32>
 ! CHECK:               %[[VAL_72:.*]] = fir.convert %[[VAL_71]] : (i32) -> i64
 ! CHECK:               %[[VAL_73:.*]] = fir.convert %[[VAL_72]] : (i64) -> index
 ! CHECK:               %[[VAL_74:.*]] = arith.subi %[[VAL_73]], %[[VAL_70]] : index
 ! CHECK:               %[[VAL_75:.*]] = fir.load %[[VAL_5]] : !fir.ref<i32>

flang/test/Lower/forall/forall-construct.f90

 ! CHECK:               %[[VAL_54:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
 ! CHECK:               %[[VAL_55:.*]] = fir.convert %[[VAL_54]] : (i32) -> i64
 ! CHECK:               %[[VAL_56:.*]] = fir.convert %[[VAL_55]] : (i64) -> index
 ! CHECK:               %[[VAL_57:.*]] = arith.subi %[[VAL_56]], %[[VAL_53]] : index
 ! CHECK:               %[[VAL_58:.*]] = fir.load %[[VAL_3]] : !fir.ref<i32>
 ! CHECK:               %[[VAL_59:.*]] = fir.convert %[[VAL_58]] : (i32) -> i64
 ! CHECK:               %[[VAL_60:.*]] = fir.convert %[[VAL_59]] : (i64) -> index
 ! CHECK:               %[[VAL_61:.*]] = arith.subi %[[VAL_60]], %[[VAL_53]] : index
-! CHECK:               %[[VAL_62:.*]] = arith.constant 3.140000e+00 : f32
-! CHECK:               %[[VAL_63:.*]] = fir.array_fetch %[[VAL_31]], %[[VAL_57]], %[[VAL_61]] : (!fir.array<?x?xf32>, index, index) -> f32
+! CHECK-DAG:           %[[VAL_62:.*]] = arith.constant 3.140000e+00 : f32
+! CHECK-DAG:           %[[VAL_63:.*]] = fir.array_fetch %[[VAL_31]], %[[VAL_57]], %[[VAL_61]] : (!fir.array<?x?xf32>, index, index) -> f32
 ! CHECK:               %[[VAL_64:.*]] = arith.divf %[[VAL_63]], %[[VAL_62]] : f32
 ! CHECK:               %[[VAL_65:.*]] = arith.constant 1 : index
 ! CHECK:               %[[VAL_66:.*]] = fir.load %[[VAL_3]] : !fir.ref<i32>
 ! CHECK:               %[[VAL_67:.*]] = fir.convert %[[VAL_66]] : (i32) -> i64
 ! CHECK:               %[[VAL_68:.*]] = fir.convert %[[VAL_67]] : (i64) -> index
 ! CHECK:               %[[VAL_69:.*]] = arith.subi %[[VAL_68]], %[[VAL_65]] : index
 ! CHECK:               %[[VAL_70:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
 ! CHECK:               %[[VAL_71:.*]] = fir.convert %[[VAL_70]] : (i32) -> i64

flang/test/Lower/forall/forall-ranked.f90

 ! CHECK:           %[[VAL_31:.*]] = arith.select %[[VAL_30]], %[[VAL_29]], %[[VAL_26]] : index
 ! CHECK:           %[[VAL_32:.*]] = fir.field_index arr, !fir.type<_QFtest_forall_with_ranked_dimensionTt{arr:!fir.array<11xi32>}>
 ! CHECK-DAG:       %[[VAL_33:.*]] = fir.load %[[VAL_0]] : !fir.ref<i32>
 ! CHECK-DAG:       %[[VAL_34:.*]] = arith.constant 4 : i32
 ! CHECK:           %[[VAL_35:.*]] = arith.addi %[[VAL_33]], %[[VAL_34]] : i32
 ! CHECK:           %[[VAL_36:.*]] = fir.convert %[[VAL_35]] : (i32) -> i64
 ! CHECK:           %[[VAL_37:.*]] = fir.convert %[[VAL_36]] : (i64) -> index
 ! CHECK:           %[[VAL_38:.*]] = arith.subi %[[VAL_37]], %[[VAL_16]] : index
+! CHECK:           %[[VAL_45:.*]] = fir.call @_QPf(%[[VAL_0]]) : (!fir.ref<i32>) -> i32
 ! CHECK:           %[[VAL_39:.*]] = arith.constant 1 : index
 ! CHECK:           %[[VAL_40:.*]] = arith.constant 0 : index
 ! CHECK:           %[[VAL_41:.*]] = arith.subi %[[VAL_31]], %[[VAL_39]] : index
 ! CHECK:           %[[VAL_42:.*]] = fir.do_loop %[[VAL_43:.*]] = %[[VAL_40]] to %[[VAL_41]] step %[[VAL_39]] unordered iter_args(%[[VAL_44:.*]] = %[[VAL_13]]) -> (!fir.array<10x10x!fir.type<_QFtest_forall_with_ranked_dimensionTt{arr:!fir.array<11xi32>}>>) {
-! CHECK:             %[[VAL_45:.*]] = fir.call @_QPf(%[[VAL_0]]) : (!fir.ref<i32>) -> i32
 ! CHECK:             %[[VAL_46:.*]] = arith.subi %[[VAL_17]], %[[VAL_17]] : index
 ! CHECK:             %[[VAL_47:.*]] = arith.muli %[[VAL_43]], %[[VAL_23]] : index
 ! CHECK:             %[[VAL_48:.*]] = arith.addi %[[VAL_46]], %[[VAL_47]] : index
 ! CHECK:             %[[VAL_49:.*]] = fir.array_update %[[VAL_44]], %[[VAL_45]], %[[VAL_21]], %[[VAL_48]], %[[VAL_32]], %[[VAL_38]] : (!fir.array<10x10x!fir.type<_QFtest_forall_with_ranked_dimensionTt{arr:!fir.array<11xi32>}>>, i32, index, index, !fir.field, index) -> !fir.array<10x10x!fir.type<_QFtest_forall_with_ranked_dimensionTt{arr:!fir.array<11xi32>}>>
 ! CHECK:             fir.result %[[VAL_49]] : !fir.array<10x10x!fir.type<_QFtest_forall_with_ranked_dimensionTt{arr:!fir.array<11xi32>}>>
 ! CHECK:           }
 ! CHECK:           fir.result %[[VAL_50:.*]] : !fir.array<10x10x!fir.type<_QFtest_forall_with_ranked_dimensionTt{arr:!fir.array<11xi32>}>>
 ! CHECK:         }

flang/test/Lower/forall/forall-slice.f90

 ! CHECK:             %[[VAL_45:.*]] = fir.convert %[[VAL_44]] : (i32) -> i64
 ! CHECK:             %[[VAL_46:.*]] = fir.convert %[[VAL_45]] : (i64) -> index
 ! CHECK:             %[[VAL_47:.*]] = arith.constant 0 : index
 ! CHECK:             %[[VAL_48:.*]] = arith.subi %[[VAL_46]], %[[VAL_40]] : index
 ! CHECK:             %[[VAL_49:.*]] = arith.addi %[[VAL_48]], %[[VAL_43]] : index
 ! CHECK:             %[[VAL_50:.*]] = arith.divsi %[[VAL_49]], %[[VAL_43]] : index
 ! CHECK:             %[[VAL_51:.*]] = arith.cmpi sgt, %[[VAL_50]], %[[VAL_47]] : index
 ! CHECK:             %[[VAL_52:.*]] = arith.select %[[VAL_51]], %[[VAL_50]], %[[VAL_47]] : index
+! CHECK:             %[[VAL_59:.*]] = fir.call @_QPf(%[[VAL_3]]) : (!fir.ref<i32>) -> i32
 ! CHECK:             %[[VAL_53:.*]] = arith.constant 1 : index
 ! CHECK:             %[[VAL_54:.*]] = arith.constant 0 : index
 ! CHECK:             %[[VAL_55:.*]] = arith.subi %[[VAL_52]], %[[VAL_53]] : index
 ! CHECK:             %[[VAL_56:.*]] = fir.do_loop %[[VAL_57:.*]] = %[[VAL_54]] to %[[VAL_55]] step %[[VAL_53]] unordered iter_args(%[[VAL_58:.*]] = %[[VAL_25]]) -> (!fir.array<10x10x!fir.type<_QFtest_forall_with_sliceTt{arr:!fir.array<11xi32>}>>) {
-! CHECK:               %[[VAL_59:.*]] = fir.call @_QPf(%[[VAL_3]]) : (!fir.ref<i32>) -> i32
 ! CHECK:               %[[VAL_60:.*]] = arith.subi %[[VAL_40]], %[[VAL_37]] : index
 ! CHECK:               %[[VAL_61:.*]] = arith.muli %[[VAL_57]], %[[VAL_43]] : index
 ! CHECK:               %[[VAL_62:.*]] = arith.addi %[[VAL_60]], %[[VAL_61]] : index
 ! CHECK:               %[[VAL_63:.*]] = fir.array_update %[[VAL_58]], %[[VAL_59]], %[[VAL_31]], %[[VAL_35]], %[[VAL_36]], %[[VAL_62]] : (!fir.array<10x10x!fir.type<_QFtest_forall_with_sliceTt{arr:!fir.array<11xi32>}>>, i32, index, index, !fir.field, index) -> !fir.array<10x10x!fir.type<_QFtest_forall_with_sliceTt{arr:!fir.array<11xi32>}>>
 ! CHECK:               fir.result %[[VAL_63]] : !fir.array<10x10x!fir.type<_QFtest_forall_with_sliceTt{arr:!fir.array<11xi32>}>>
 ! CHECK:             }
 ! CHECK:             fir.result %[[VAL_64:.*]] : !fir.array<10x10x!fir.type<_QFtest_forall_with_sliceTt{arr:!fir.array<11xi32>}>>
 ! CHECK:           }

flang/test/Lower/forall/test9.f90

subroutine test9(a,b,n)
   real, intent(inout) :: a(n)
   real, intent(in) :: b(n)
   loopy: FORALL (i=1:n-1)
      a(i+1) = a(i) + b(i)
   END FORALL loopy
 end subroutine test9
 
 ! CHECK-LABEL: func @_QPtest9(
-! CHECK-SAME:    %[[VAL_0:.*]]: !fir.ref<!fir.array<?xf32>>{{.*}}, %[[VAL_1:.*]]: !fir.ref<!fir.array<?xf32>>{{.*}}, %[[VAL_2:.*]]: !fir.ref<i32>{{.*}}) {
+! CHECK-SAME:                 %[[VAL_0:.*]]: !fir.ref<!fir.array<?xf32>> {fir.bindc_name = "a"},
+! CHECK-SAME:                 %[[VAL_1:.*]]: !fir.ref<!fir.array<?xf32>> {fir.bindc_name = "b"},
+! CHECK-SAME:                 %[[VAL_2:.*]]: !fir.ref<i32> {fir.bindc_name = "n"}) {
 ! CHECK:         %[[VAL_3:.*]] = fir.alloca i32 {adapt.valuebyref, bindc_name = "i"}
 ! CHECK:         %[[VAL_4:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
 ! CHECK:         %[[VAL_5:.*]] = fir.convert %[[VAL_4]] : (i32) -> i64
-! CHECK:         %[[VAL_6A:.*]] = fir.convert %[[VAL_5]] : (i64) -> index
-! CHECK:         %[[C0:.*]] = arith.constant 0 : index
-! CHECK:         %[[CMP:.*]] = arith.cmpi sgt, %[[VAL_6A]], %[[C0]] : index
-! CHECK:         %[[VAL_6:.*]] = arith.select %[[CMP]], %[[VAL_6A]], %[[C0]] : index
-! CHECK:         %[[VAL_7:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
-! CHECK:         %[[VAL_8:.*]] = fir.convert %[[VAL_7]] : (i32) -> i64
-! CHECK:         %[[VAL_9A:.*]] = fir.convert %[[VAL_8]] : (i64) -> index
-! CHECK:         %[[C0_2:.*]] = arith.constant 0 : index
-! CHECK:         %[[CMP_2:.*]] = arith.cmpi sgt, %[[VAL_9A]], %[[C0_2]] : index
-! CHECK:         %[[VAL_9:.*]] = arith.select %[[CMP_2]], %[[VAL_9A]], %[[C0_2]] : index
-! CHECK:         %[[VAL_10:.*]] = arith.constant 1 : i32
-! CHECK:         %[[VAL_11:.*]] = fir.convert %[[VAL_10]] : (i32) -> index
-! CHECK:         %[[VAL_12:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
-! CHECK:         %[[VAL_13:.*]] = arith.constant 1 : i32
-! CHECK:         %[[VAL_14:.*]] = arith.subi %[[VAL_12]], %[[VAL_13]] : i32
-! CHECK:         %[[VAL_15:.*]] = fir.convert %[[VAL_14]] : (i32) -> index
-! CHECK:         %[[VAL_16:.*]] = arith.constant 1 : index
-! CHECK:         %[[VAL_17:.*]] = fir.shape %[[VAL_6]] : (index) -> !fir.shape<1>
-! CHECK:         %[[VAL_18:.*]] = fir.array_load %[[VAL_0]](%[[VAL_17]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
-! CHECK:         %[[VAL_19:.*]] = fir.shape %[[VAL_6]] : (index) -> !fir.shape<1>
-! CHECK:         %[[VAL_20:.*]] = fir.array_load %[[VAL_0]](%[[VAL_19]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
-! CHECK:         %[[VAL_21:.*]] = fir.shape %[[VAL_9]] : (index) -> !fir.shape<1>
-! CHECK:         %[[VAL_22:.*]] = fir.array_load %[[VAL_1]](%[[VAL_21]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
-! CHECK:         %[[VAL_23:.*]] = fir.do_loop %[[VAL_24:.*]] = %[[VAL_11]] to %[[VAL_15]] step %[[VAL_16]] unordered iter_args(%[[VAL_25:.*]] = %[[VAL_18]]) -> (!fir.array<?xf32>) {
-! CHECK:           %[[VAL_26:.*]] = fir.convert %[[VAL_24]] : (index) -> i32
-! CHECK:           fir.store %[[VAL_26]] to %[[VAL_3]] : !fir.ref<i32>
-! CHECK:           %[[VAL_27:.*]] = arith.constant 1 : index
-! CHECK:           %[[VAL_28:.*]] = fir.load %[[VAL_3]] : !fir.ref<i32>
-! CHECK:           %[[VAL_29:.*]] = fir.convert %[[VAL_28]] : (i32) -> i64
-! CHECK:           %[[VAL_30:.*]] = fir.convert %[[VAL_29]] : (i64) -> index
-! CHECK:           %[[VAL_31:.*]] = arith.subi %[[VAL_30]], %[[VAL_27]] : index
-! CHECK:           %[[VAL_32:.*]] = arith.constant 1 : index
-! CHECK:           %[[VAL_33:.*]] = fir.load %[[VAL_3]] : !fir.ref<i32>
-! CHECK:           %[[VAL_34:.*]] = fir.convert %[[VAL_33]] : (i32) -> i64
-! CHECK:           %[[VAL_35:.*]] = fir.convert %[[VAL_34]] : (i64) -> index
-! CHECK:           %[[VAL_36:.*]] = arith.subi %[[VAL_35]], %[[VAL_32]] : index
-! CHECK:           %[[VAL_37:.*]] = fir.array_fetch %[[VAL_20]], %[[VAL_31]] : (!fir.array<?xf32>, index) -> f32
-! CHECK:           %[[VAL_38:.*]] = fir.array_fetch %[[VAL_22]], %[[VAL_36]] : (!fir.array<?xf32>, index) -> f32
-! CHECK:           %[[VAL_39:.*]] = arith.addf %[[VAL_37]], %[[VAL_38]] : f32
-! CHECK:           %[[VAL_40:.*]] = arith.constant 1 : index
-! CHECK-DAG:       %[[VAL_41:.*]] = fir.load %[[VAL_3]] : !fir.ref<i32>
-! CHECK-DAG:       %[[VAL_42:.*]] = arith.constant 1 : i32
-! CHECK:           %[[VAL_43:.*]] = arith.addi %[[VAL_41]], %[[VAL_42]] : i32
-! CHECK:           %[[VAL_44:.*]] = fir.convert %[[VAL_43]] : (i32) -> i64
-! CHECK:           %[[VAL_45:.*]] = fir.convert %[[VAL_44]] : (i64) -> index
-! CHECK:           %[[VAL_46:.*]] = arith.subi %[[VAL_45]], %[[VAL_40]] : index
-! CHECK:           %[[VAL_47:.*]] = fir.array_update %[[VAL_25]], %[[VAL_39]], %[[VAL_46]] : (!fir.array<?xf32>, f32, index) -> !fir.array<?xf32>
-! CHECK:           fir.result %[[VAL_47]] : !fir.array<?xf32>
+! CHECK:         %[[VAL_6:.*]] = fir.convert %[[VAL_5]] : (i64) -> index
+! CHECK:         %[[VAL_7:.*]] = arith.constant 0 : index
+! CHECK:         %[[VAL_8:.*]] = arith.cmpi sgt, %[[VAL_6]], %[[VAL_7]] : index
+! CHECK:         %[[VAL_9:.*]] = arith.select %[[VAL_8]], %[[VAL_6]], %[[VAL_7]] : index
+! CHECK:         %[[VAL_10:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
+! CHECK:         %[[VAL_11:.*]] = fir.convert %[[VAL_10]] : (i32) -> i64
+! CHECK:         %[[VAL_12:.*]] = fir.convert %[[VAL_11]] : (i64) -> index
+! CHECK:         %[[VAL_13:.*]] = arith.constant 0 : index
+! CHECK:         %[[VAL_14:.*]] = arith.cmpi sgt, %[[VAL_12]], %[[VAL_13]] : index
+! CHECK:         %[[VAL_15:.*]] = arith.select %[[VAL_14]], %[[VAL_12]], %[[VAL_13]] : index
+! CHECK:         %[[VAL_16:.*]] = arith.constant 1 : i32
+! CHECK:         %[[VAL_17:.*]] = fir.convert %[[VAL_16]] : (i32) -> index
+! CHECK:         %[[VAL_18:.*]] = fir.load %[[VAL_2]] : !fir.ref<i32>
+! CHECK:         %[[VAL_19:.*]] = arith.constant 1 : i32
+! CHECK:         %[[VAL_20:.*]] = arith.subi %[[VAL_18]], %[[VAL_19]] : i32
+! CHECK:         %[[VAL_21:.*]] = fir.convert %[[VAL_20]] : (i32) -> index
+! CHECK:         %[[VAL_22:.*]] = arith.constant 1 : index
+! CHECK:         %[[VAL_23:.*]] = fir.shape %[[VAL_9]] : (index) -> !fir.shape<1>
+! CHECK:         %[[VAL_24:.*]] = fir.array_load %[[VAL_0]](%[[VAL_23]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
+! CHECK:         %[[VAL_25:.*]] = fir.shape %[[VAL_9]] : (index) -> !fir.shape<1>
+! CHECK:         %[[VAL_26:.*]] = fir.array_load %[[VAL_0]](%[[VAL_25]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
+! CHECK:         %[[VAL_27:.*]] = fir.shape %[[VAL_15]] : (index) -> !fir.shape<1>
+! CHECK:         %[[VAL_28:.*]] = fir.array_load %[[VAL_1]](%[[VAL_27]]) : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.array<?xf32>
+! CHECK:         %[[VAL_29:.*]] = fir.do_loop %[[VAL_30:.*]] = %[[VAL_17]] to %[[VAL_21]] step %[[VAL_22]] unordered iter_args(%[[VAL_31:.*]] = %[[VAL_24]]) -> (!fir.array<?xf32>) {
+! CHECK:           %[[VAL_32:.*]] = fir.convert %[[VAL_30]] : (index) -> i32
+! CHECK:           fir.store %[[VAL_32]] to %[[VAL_3]] : !fir.ref<i32>
+! CHECK:           %[[VAL_33:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_34:.*]] = fir.load %[[VAL_3]] : !fir.ref<i32>
+! CHECK:           %[[VAL_35:.*]] = fir.convert %[[VAL_34]] : (i32) -> i64
+! CHECK:           %[[VAL_36:.*]] = fir.convert %[[VAL_35]] : (i64) -> index
+! CHECK:           %[[VAL_37:.*]] = arith.subi %[[VAL_36]], %[[VAL_33]] : index
+! CHECK:           %[[VAL_38:.*]] = fir.array_fetch %[[VAL_26]], %[[VAL_37]] : (!fir.array<?xf32>, index) -> f32
+! CHECK:           %[[VAL_39:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_40:.*]] = fir.load %[[VAL_3]] : !fir.ref<i32>
+! CHECK:           %[[VAL_41:.*]] = fir.convert %[[VAL_40]] : (i32) -> i64
+! CHECK:           %[[VAL_42:.*]] = fir.convert %[[VAL_41]] : (i64) -> index
+! CHECK:           %[[VAL_43:.*]] = arith.subi %[[VAL_42]], %[[VAL_39]] : index
+! CHECK:           %[[VAL_44:.*]] = fir.array_fetch %[[VAL_28]], %[[VAL_43]] : (!fir.array<?xf32>, index) -> f32
+! CHECK:           %[[VAL_45:.*]] = arith.addf %[[VAL_38]], %[[VAL_44]] : f32
+! CHECK:           %[[VAL_46:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_47:.*]] = fir.load %[[VAL_3]] : !fir.ref<i32>
+! CHECK:           %[[VAL_48:.*]] = arith.constant 1 : i32
+! CHECK:           %[[VAL_49:.*]] = arith.addi %[[VAL_47]], %[[VAL_48]] : i32
+! CHECK:           %[[VAL_50:.*]] = fir.convert %[[VAL_49]] : (i32) -> i64
+! CHECK:           %[[VAL_51:.*]] = fir.convert %[[VAL_50]] : (i64) -> index
+! CHECK:           %[[VAL_52:.*]] = arith.subi %[[VAL_51]], %[[VAL_46]] : index
+! CHECK:           %[[VAL_53:.*]] = fir.array_update %[[VAL_31]], %[[VAL_45]], %[[VAL_52]] : (!fir.array<?xf32>, f32, index) -> !fir.array<?xf32>
+! CHECK:           fir.result %[[VAL_53]] : !fir.array<?xf32>
 ! CHECK:         }
-! CHECK:         fir.array_merge_store %[[VAL_18]], %[[VAL_48:.*]] to %[[VAL_0]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.ref<!fir.array<?xf32>>
+! CHECK:         fir.array_merge_store %[[VAL_24]], %[[VAL_54:.*]] to %[[VAL_0]] : !fir.array<?xf32>, !fir.array<?xf32>, !fir.ref<!fir.array<?xf32>>
 ! CHECK:         return
 ! CHECK:       }