[flang][cuda] Support memory cleanup at a return statement (#116304)

We generate `cuf.free` and `func.return` twice if a return statement
exists at the end of program.

```f90
program test
  integer, device :: a(10)
  return
end
```

```
% flang -x cuda test.cuf -mmlir --mlir-print-ir-after-all
error: loc("/path/to/test.cuf":3:3): 'func.return' op must be the last operation in the parent block
// -----// IR Dump After Fortran::lower::VerifierPass Failed () //----- //
```

Dumped IR:
```mlir
  "func.func"() <{function_type = () -> (), sym_name = "_QQmain"}> ({
...
    "cuf.free"(%5#1) <{data_attr = #cuf.cuda<device>}> : (!fir.ref<!fir.array<10xi32>>) -> ()
    "func.return"() : () -> ()
    "cuf.free"(%5#1) <{data_attr = #cuf.cuda<device>}> : (!fir.ref<!fir.array<10xi32>>) -> ()
    "func.return"() : () -> ()
}
...
```

The routine `genExitRoutine` in `Bridge.cpp` is guarded by
`blockIsUnterminated()` to make sure that `func.return` is generated
only at the end of a block. However, we redundantly run
`bridge.fctCtx().finalizeAndKeep()` before `genExitRoutine` in this
case, resulting in two pairs of `cuf.free` and `func.return`. This PR
fixes `Bridge.cpp` by using `blockIsUnterminated()` to guard
`finalizeAndKeep` as well.

Author: khaki3

flang/include/flang/Lower/StatementContext.h

@@ -92,10 +92,13 @@ class StatementContext {
     cufs.back().reset();
   }
 
+  /// Pop the stack top list.
+  void pop() { cufs.pop_back(); }
+
   /// Make cleanup calls. Pop the stack top list.
   void finalizeAndPop() {
     finalizeAndKeep();
-    cufs.pop_back();
+    pop();
   }
 
   bool hasCode() const {

flang/lib/Lower/Bridge.cpp

@@ -1621,13 +1621,19 @@ class FirConverter : public Fortran::lower::AbstractConverter {
   // Termination of symbolically referenced execution units
   //===--------------------------------------------------------------------===//
 
-  /// END of program
+  /// Exit of a routine
   ///
-  /// Generate the cleanup block before the program exits
-  void genExitRoutine() {
-
-    if (blockIsUnterminated())
-      builder->create<mlir::func::ReturnOp>(toLocation());
+  /// Generate the cleanup block before the routine exits
+  void genExitRoutine(bool earlyReturn, mlir::ValueRange retval = {}) {
+    if (blockIsUnterminated()) {
+      bridge.openAccCtx().finalizeAndKeep();
+      bridge.fctCtx().finalizeAndKeep();
+      builder->create<mlir::func::ReturnOp>(toLocation(), retval);
+    }
+    if (!earlyReturn) {
+      bridge.openAccCtx().pop();
+      bridge.fctCtx().pop();
+    }
   }
 
   /// END of procedure-like constructs
@@ -1684,9 +1690,7 @@ class FirConverter : public Fortran::lower::AbstractConverter {
             resultRef = builder->createConvert(loc, resultRefType, resultRef);
           return builder->create<fir::LoadOp>(loc, resultRef);
         });
-    bridge.openAccCtx().finalizeAndPop();
-    bridge.fctCtx().finalizeAndPop();
-    builder->create<mlir::func::ReturnOp>(loc, resultVal);
+    genExitRoutine(false, resultVal);
   }
 
   /// Get the return value of a call to \p symbol, which is a subroutine entry
@@ -1712,13 +1716,9 @@ class FirConverter : public Fortran::lower::AbstractConverter {
     } else if (Fortran::semantics::HasAlternateReturns(symbol)) {
       mlir::Value retval = builder->create<fir::LoadOp>(
           toLocation(), getAltReturnResult(symbol));
-      bridge.openAccCtx().finalizeAndPop();
-      bridge.fctCtx().finalizeAndPop();
-      builder->create<mlir::func::ReturnOp>(toLocation(), retval);
+      genExitRoutine(false, retval);
     } else {
-      bridge.openAccCtx().finalizeAndPop();
-      bridge.fctCtx().finalizeAndPop();
-      genExitRoutine();
+      genExitRoutine(false);
     }
   }
 
@@ -5018,8 +5018,7 @@ class FirConverter : public Fortran::lower::AbstractConverter {
       it->stmtCtx.finalizeAndKeep();
     }
     if (funit->isMainProgram()) {
-      bridge.fctCtx().finalizeAndKeep();
-      genExitRoutine();
+      genExitRoutine(true);
       return;
     }
     mlir::Location loc = toLocation();
@@ -5478,9 +5477,7 @@ class FirConverter : public Fortran::lower::AbstractConverter {
   void endNewFunction(Fortran::lower::pft::FunctionLikeUnit &funit) {
     setCurrentPosition(Fortran::lower::pft::stmtSourceLoc(funit.endStmt));
     if (funit.isMainProgram()) {
-      bridge.openAccCtx().finalizeAndPop();
-      bridge.fctCtx().finalizeAndPop();
-      genExitRoutine();
+      genExitRoutine(false);
     } else {
       genFIRProcedureExit(funit, funit.getSubprogramSymbol());
     }

flang/test/Lower/CUDA/cuda-return01.cuf

@@ -0,0 +1,14 @@
+! RUN: bbc -emit-hlfir -fcuda %s -o - | FileCheck %s
+
+! Check if finalization works with a return statement
+
+program main
+  integer, device :: a(10)
+  return
+end
+
+! CHECK: func.func @_QQmain() attributes {fir.bindc_name = "main"} {
+! CHECK: %[[DECL:.*]]:2 = hlfir.declare
+! CHECK-NEXT: cuf.free %[[DECL]]#1 : !fir.ref<!fir.array<10xi32>>
+! CHECK-NEXT: return
+! CHECK-NEXT: }

flang/test/Lower/CUDA/cuda-return02.cuf

@@ -0,0 +1,48 @@
+! RUN: bbc -emit-hlfir -fcuda %s -o - | FileCheck %s
+
+! Check if finalization works with multiple return statements
+
+program test
+  integer, device :: a(10)
+  logical :: l
+
+  if (l) then
+    return
+  end if
+
+  return
+end
+
+! CHECK: func.func @_QQmain() attributes {fir.bindc_name = "test"} {
+! CHECK: %[[DECL:.*]]:2 = hlfir.declare
+! CHECK: cf.cond_br %{{.*}}, ^bb1, ^bb2
+! CHECK-NEXT: ^bb1:
+! CHECK-NEXT: cuf.free %[[DECL]]#1 : !fir.ref<!fir.array<10xi32>>
+! CHECK-NEXT: return
+! CHECK-NEXT: ^bb2:
+! CHECK-NEXT: cuf.free %[[DECL]]#1 : !fir.ref<!fir.array<10xi32>>
+! CHECK-NEXT: return
+! CHECK-NEXT: }
+
+subroutine sub(l)
+  integer, device :: a(10)
+  logical :: l
+
+  if (l) then
+    l = .false.
+    return
+  end if
+
+  return
+end
+
+! CHECK: func.func @_QPsub(%arg0: !fir.ref<!fir.logical<4>> {fir.bindc_name = "l"}) {
+! CHECK: %[[DECL:.*]]:2 = hlfir.declare
+! CHECK: cf.cond_br %6, ^bb1, ^bb2
+! CHECK: ^bb1:
+! CHECK: cf.br ^bb3
+! CHECK: ^bb2:
+! CHECK: cf.br ^bb3
+! CHECK: ^bb3:
+! CHECK: cuf.free %[[DECL]]#1 : !fir.ref<!fir.array<10xi32>>
+! CHECK: }