Merge pull request #2371 from bstatcomp/bugfix_reduce_sum

Fixes bug in `reduce_sum` handling expressions with threading enabled

Author: rok-cesnovar

stan/math/prim/functor/reduce_sum.hpp

@@ -204,9 +204,9 @@ inline auto reduce_sum(Vec&& vmapped, int grainsize, std::ostream* msgs,
 
 #ifdef STAN_THREADS
   return internal::reduce_sum_impl<ReduceFunction, void, return_type, Vec,
-                                   Args...>()(std::forward<Vec>(vmapped), true,
-                                              grainsize, msgs,
-                                              std::forward<Args>(args)...);
+                                   ref_type_t<Args&&>...>()(
+      std::forward<Vec>(vmapped), true, grainsize, msgs,
+      std::forward<Args>(args)...);
 #else
   if (vmapped.empty()) {
     return return_type(0.0);

stan/math/rev/core/deep_copy_vars.hpp

@@ -20,7 +20,7 @@ namespace math {
  *  Otherwise it will be moved.
  */
 template <typename Arith, typename = require_arithmetic_t<scalar_type_t<Arith>>>
-inline decltype(auto) deep_copy_vars(Arith&& arg) {
+inline Arith deep_copy_vars(Arith&& arg) {
   return std::forward<Arith>(arg);
 }