ReduceOps: reset result readiness flag (#3545)

In #3421, a change was made to avoid a common mistake of calling
ReduceData::value multiple times. Unfortunately, it breaks codes that
using one ReduceOps with multiple ReduceData objects. This PR fixes both
cases.

## Checklist

The proposed changes:
- [x] fix a bug or incorrect behavior in AMReX
- [ ] add new capabilities to AMReX
- [ ] changes answers in the test suite to more than roundoff level
- [ ] are likely to significantly affect the results of downstream AMReX
users
- [ ] include documentation in the code and/or rst files, if appropriate

Src/Base/AMReX_Reduce.H

@@ -258,6 +258,7 @@ public:
  * m_max_blocks * sizeof(Type)))),
  m_fn_value([&reduce_op,this] () -> Type { return this->value(reduce_op); })
  {
+ reduce_op.resetResultReadiness();
  static_assert(std::is_trivially_copyable<Type>(),
  "ReduceData::Type must be trivially copyable");
  static_assert(std::is_trivially_destructible<Type>(),
@@ -750,6 +751,9 @@ public:
 
 private:
  bool m_result_is_ready = false;
+
+public:
+ void resetResultReadiness () { m_result_is_ready = false; }
 };
 
 namespace Reduce {
@@ -993,6 +997,7 @@ public:
  : m_tuple(OpenMP::in_parallel() ? 1 : OpenMP::get_max_threads()),
  m_fn_value([&reduce_op,this] () -> Type { return this->value(reduce_op); })
  {
+ reduce_op.resetResultReadiness();
  for (auto& t : m_tuple) {
  Reduce::detail::for_each_init<0, Type, Ps...>(t);
  }
@@ -1158,6 +1163,8 @@ public:
  }
 
  bool m_result_is_ready = false;
+
+ void resetResultReadiness () { m_result_is_ready = false; }
 };
 
 namespace Reduce {