[flang][cuda] Allow DO CONCURRENT in cuf kernel (#124190)

flang/lib/Semantics/check-cuda.cpp

@@ -520,7 +520,8 @@ void CUDAChecker::Enter(const parser::SeparateModuleSubprogram &x) {
 
 static int DoConstructTightNesting(
     const parser::DoConstruct *doConstruct, const parser::Block *&innerBlock) {
-  if (!doConstruct || !doConstruct->IsDoNormal()) {
+  if (!doConstruct ||
+      (!doConstruct->IsDoNormal() && !doConstruct->IsDoConcurrent())) {
     return 0;
   }
   innerBlock = &std::get<parser::Block>(doConstruct->t);

flang/test/Semantics/cuf09.cuf

@@ -130,7 +130,6 @@ program main
   do
     exit
   end do
-  !ERROR: !$CUF KERNEL DO (1) must be followed by a DO construct with tightly nested outer levels of counted DO loops
   !$cuf kernel do <<< 1, 2 >>>
   do concurrent (j=1:10)
   end do