[flang] Add a REDUCE clause to each nested loop (llvm#95555)

For DO CONCURRENT REDUCE, every nested loop should have a REDUCE clause
so that we can lower reduction without analysis.

Author: khaki3

flang/lib/Lower/Bridge.cpp

@@ -1801,14 +1801,16 @@ class FirConverter : public Fortran::lower::AbstractConverter {
               std::get_if<Fortran::parser::LocalitySpec::LocalInit>(&x.u))
         for (const Fortran::parser::Name &x : localInitList->v)
           info.localInitSymList.push_back(x.symbol);
-      if (const auto *reduceList =
-              std::get_if<Fortran::parser::LocalitySpec::Reduce>(&x.u)) {
-        fir::ReduceOperationEnum reduce_operation = getReduceOperationEnum(
-            std::get<Fortran::parser::ReductionOperator>(reduceList->t));
-        for (const Fortran::parser::Name &x :
-             std::get<std::list<Fortran::parser::Name>>(reduceList->t)) {
-          info.reduceSymList.push_back(
-              std::make_pair(reduce_operation, x.symbol));
+      for (IncrementLoopInfo &info : incrementLoopNestInfo) {
+        if (const auto *reduceList =
+                std::get_if<Fortran::parser::LocalitySpec::Reduce>(&x.u)) {
+          fir::ReduceOperationEnum reduce_operation = getReduceOperationEnum(
+              std::get<Fortran::parser::ReductionOperator>(reduceList->t));
+          for (const Fortran::parser::Name &x :
+               std::get<std::list<Fortran::parser::Name>>(reduceList->t)) {
+            info.reduceSymList.push_back(
+                std::make_pair(reduce_operation, x.symbol));
+          }
         }
       }
       if (const auto *sharedList =

flang/test/Lower/loops3.f90

@@ -12,8 +12,8 @@ subroutine loop_test
 
   ! CHECK: %[[VAL_0:.*]] = fir.alloca f32 {bindc_name = "m", uniq_name = "_QFloop_testEm"}
   ! CHECK: %[[VAL_1:.*]] = fir.address_of(@_QFloop_testEsum) : !fir.ref<i32>
-  ! CHECK: fir.do_loop %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} unordered {
-  ! CHECK: fir.do_loop %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} unordered {
+  ! CHECK: fir.do_loop %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} unordered reduce(#fir.reduce_attr<add> -> %[[VAL_1:.*]] : !fir.ref<i32>, #fir.reduce_attr<max> -> %[[VAL_0:.*]] : !fir.ref<f32>) {
+  ! CHECK: fir.do_loop %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} unordered reduce(#fir.reduce_attr<add> -> %[[VAL_1:.*]] : !fir.ref<i32>, #fir.reduce_attr<max> -> %[[VAL_0:.*]] : !fir.ref<f32>) {
   ! CHECK: fir.do_loop %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} unordered reduce(#fir.reduce_attr<add> -> %[[VAL_1:.*]] : !fir.ref<i32>, #fir.reduce_attr<max> -> %[[VAL_0:.*]] : !fir.ref<f32>) {
   do concurrent (i=1:5, j=1:5, k=1:5) local(tmp) reduce(+:sum) reduce(max:m)
     tmp = i + j + k