[flang][cuda] Do not lower device variables in main program as globals (#102512)

Flang considers arrays in main program larger than 32 bytes having the
SAVE attribute and lowers them as globals. In CUDA Fortran, device
variables are not allowed to have the SAVE attribute and should be
allocated dynamically in the main program scope.
This patch updates lowering so CUDA Fortran device variables are not
considered with the SAVE attribute.

Author: clementval

flang/include/flang/Evaluate/tools.h

@@ -1243,6 +1243,18 @@ bool CheckForCoindexedObject(parser::ContextualMessages &,
     const std::optional<ActualArgument> &, const std::string &procName,
     const std::string &argName);
 
+inline bool CanCUDASymbolHasSave(const Symbol &sym) {
+  if (const auto *details =
+          sym.GetUltimate().detailsIf<semantics::ObjectEntityDetails>()) {
+    if (details->cudaDataAttr() &&
+        *details->cudaDataAttr() != common::CUDADataAttr::Pinned &&
+        *details->cudaDataAttr() != common::CUDADataAttr::Unified) {
+      return false;
+    }
+  }
+  return true;
+}
+
 inline bool IsCUDADeviceSymbol(const Symbol &sym) {
   if (const auto *details =
           sym.GetUltimate().detailsIf<semantics::ObjectEntityDetails>()) {

flang/lib/Evaluate/tools.cpp

@@ -1696,7 +1696,8 @@ bool IsSaved(const Symbol &original) {
       (features.IsEnabled(common::LanguageFeature::SaveMainProgram) ||
           (features.IsEnabled(
                common::LanguageFeature::SaveBigMainProgramVariables) &&
-              symbol.size() > 32))) {
+              symbol.size() > 32)) &&
+      Fortran::evaluate::CanCUDASymbolHasSave(symbol)) {
     // With SaveBigMainProgramVariables, keeping all unsaved main program
     // variables of 32 bytes or less on the stack allows keeping numerical and
     // logical scalars, small scalar characters or derived, small arrays, and

flang/test/Lower/CUDA/cuda-program-global.cuf

@@ -1,19 +1,22 @@
 ! RUN: bbc -emit-hlfir -fcuda %s -o - | FileCheck %s
 
-! Test lowering of program local variable that are global
+! Test lowering of program local variables. Make sure CUDA device variables are
+! not lowered as global.
 
 program test
   integer, device :: a(10)
+  integer, unified :: u(10)
   integer :: b(10)
   integer :: i
   print*,i
 end
 
 ! CHECK-LABEL: func.func @_QQmain()
-! CHECK: fir.address_of(@_QFEa) : !fir.ref<!fir.array<10xi32>>
+! CHECK: cuf.alloc !fir.array<10xi32> {bindc_name = "a", data_attr = #cuf.cuda<device>, uniq_name = "_QFEa"} -> !fir.ref<!fir.array<10xi32>>
 ! CHECK: fir.address_of(@_QFEb) : !fir.ref<!fir.array<10xi32>>
 ! CHECK: %[[ALLOCA:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFEi"}
 ! CHECK: hlfir.declare %[[ALLOCA]] {uniq_name = "_QFEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 
-! CHECK: fir.global internal @_QFEa {data_attr = #cuf.cuda<device>} : !fir.array<10xi32> {{{$}}
+! CHECK-NOT: fir.global internal @_QFEa {data_attr = #cuf.cuda<device>} : !fir.array<10xi32> {{{$}}
 ! CHECK: fir.global internal @_QFEb : !fir.array<10xi32> {{{$}}
+! CHECK: fir.global internal @_QFEu {data_attr = #cuf.cuda<unified>} : !fir.array<10xi32>