MappedScop::reductionMapSchedule: always map all thread identifiers

A reduction is always mapped to all thread identifiers.
MappedScop::detectReductions will therefore not detect a band
as a reduction if it does not have a sufficient number
of (coincident) members.
numThreads.view.size() is therefore never greater than reductionDim + 1.
Taking the minimum (introduced in 88764d3 (split out reduction band after
thread mapping to remove nested mapping, Mon Apr 9 17:43:18 2018 +0200)),
is therefore not needed and in fact only causes confusion.

Author: Sven Verdoolaege

tc/core/polyhedral/cuda/mapped_scop.cc

@@ -299,7 +299,7 @@ isl::multi_union_pw_aff MappedScop::reductionMapSchedule(
   auto reductionSchedule = reductionBand->mupa_;
   auto nMember = reductionBand->nMember();
   auto reductionDim = reductionBand->nOuterCoincident();
-  auto nMappedThreads = std::min(numThreads.view.size(), reductionDim + 1);
+  auto nMappedThreads = numThreads.view.size();
   TC_CHECK_GE(nMember, reductionDim);
   reductionSchedule = reductionSchedule.drop_dims(
       isl::dim_type::set, reductionDim + 1, nMember - (reductionDim + 1));