[flang][hlfir] Support fir.declare in AbstractResult pass

The AbstractResult pass replaces allocation of function result on the callee
side per an extra argument so that the allocation of the result can be
done on the caller stack.
It looks for the result allocation from the fir.return op, so it needs
to handle (in a transparent way) a fir.declare in the chain between the
allocation and the fir.return.

Reviewed By: vzakhari, clementval

Differential Revision: https://reviews.llvm.org/D149057

flang/lib/Optimizer/Transforms/AbstractResult.cpp

@@ -217,6 +217,10 @@ class ReturnOpConversion : public mlir::OpRewritePattern<mlir::func::ReturnOp> {
     if (auto *op = returnedValue.getDefiningOp())
       if (auto load = mlir::dyn_cast<fir::LoadOp>(op)) {
         auto resultStorage = load.getMemref();
+        // The result alloca may be behind a fir.declare, if any.
+        if (auto declare = mlir::dyn_cast_or_null<fir::DeclareOp>(
+                resultStorage.getDefiningOp()))
+          resultStorage = declare.getMemref();
         // TODO: This should be generalized for derived types, and it is
         // architecture and OS dependent.
         if (fir::isa_builtin_cptr_type(returnedValue.getType())) {
@@ -232,7 +236,7 @@ class ReturnOpConversion : public mlir::OpRewritePattern<mlir::func::ReturnOp> {
               ret, mlir::ValueRange{retValue});
           return mlir::success();
         }
-        load.getMemref().replaceAllUsesWith(newArg);
+        resultStorage.replaceAllUsesWith(newArg);
         replacedStorage = true;
         if (auto *alloc = resultStorage.getDefiningOp())
           if (alloc->use_empty())

flang/test/Fir/abstract-results.fir

@@ -106,6 +106,33 @@ func.func @retcptr() -> !fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__addres
   // FUNC-BOX: return %[[VAL]] : i64
 }
 
+// FUNC-REF-LABEL:  func private @arrayfunc_callee_declare(
+// FUNC-REF-SAME: %[[buffer:.*]]: !fir.ref<!fir.array<?xf32>>, %[[n:.*]]: index) {
+// FUNC-BOX-LABEL:  func private @arrayfunc_callee_declare(
+// FUNC-BOX-SAME: %[[box:.*]]: !fir.box<!fir.array<?xf32>>, %[[n:.*]]: index) {
+func.func private @arrayfunc_callee_declare(%n : index) -> !fir.array<?xf32> {
+  %buffer_alloc = fir.alloca !fir.array<?xf32>, %n
+  %shape = fir.shape %n : (index) -> !fir.shape<1>
+  %buffer = fir.declare %buffer_alloc(%shape) {uniq_name = "x"}: (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.ref<!fir.array<?xf32>>
+  // Do something with result (res(4) = 42.)
+  %c4 = arith.constant 4 : i64
+  %coor = fir.coordinate_of %buffer, %c4 : (!fir.ref<!fir.array<?xf32>>, i64) -> !fir.ref<f32>
+  %cst = arith.constant 4.200000e+01 : f32
+  fir.store %cst to %coor : !fir.ref<f32>
+  %res = fir.load %buffer : !fir.ref<!fir.array<?xf32>>
+  return %res : !fir.array<?xf32>
+
+  // FUNC-REF-DAG: %[[buffer_declare:.*]] = fir.declare %[[buffer]](%{{.*}}) {uniq_name = "x"} : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.ref<!fir.array<?xf32>>
+  // FUNC-REF-DAG: %[[coor:.*]] = fir.coordinate_of %[[buffer_declare]], %{{.*}} : (!fir.ref<!fir.array<?xf32>>, i64) -> !fir.ref<f32>
+  // FUNC-REF-DAG: fir.store %{{.*}} to %[[coor]] : !fir.ref<f32>
+  // FUNC-REF: return
+
+  // FUNC-BOX: %[[buffer:.*]] = fir.box_addr %[[box]] : (!fir.box<!fir.array<?xf32>>) -> !fir.ref<!fir.array<?xf32>>
+  // FUNC-BOX-DAG: %[[buffer_declare:.*]] = fir.declare %[[buffer]](%{{.*}}) {uniq_name = "x"} : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.ref<!fir.array<?xf32>>
+  // FUNC-BOX-DAG: %[[coor:.*]] = fir.coordinate_of %[[buffer_declare]], %{{.*}} : (!fir.ref<!fir.array<?xf32>>, i64) -> !fir.ref<f32>
+  // FUNC-BOX-DAG: fir.store %{{.*}} to %[[coor]] : !fir.ref<f32>
+  // FUNC-BOX: return
+}
 
 // ------------------------ Test caller rewrite --------------------------------