We see -3.17% reduction in compile time of 3mm and -17.87% when compiled with -DPOLYBENCH_USE_C99_PROTO flag.
We also do not get any regressions.
Details
Details
Diff Detail
Diff Detail
- Repository
- rL LLVM
Event Timeline
Comment Actions
We see -3.17% reduction in compile time of 3mm and -17.87% when compiled with -DPOLYBENCH_USE_C99_PROTO flag.
We also do not get any regressions.http://reviews.llvm.org/D12198
Files:
lib/Analysis/ScopInfo.cppIndex: lib/Analysis/ScopInfo.cpp
- lib/Analysis/ScopInfo.cpp
+++ lib/Analysis/ScopInfo.cpp
@@ -1302,8 +1302,8 @@Scop::MinMaxVectorTy &MinMaxAccesses) { Accesses = isl_union_map_intersect_domain(Accesses, Domains);+ isl_set_free(AssumedContext);
isl_union_set *Locations = isl_union_map_range(Accesses);
- Locations = isl_union_set_intersect_params(Locations, AssumedContext); Locations = isl_union_set_coalesce(Locations); Locations = isl_union_set_detect_equalities(Locations); bool Valid = (0 == isl_union_set_foreach_set(Locations, buildMinMaxAccess,
Hi Pratik,
please do not even pass the assumedContext to calculateMinMaxAccess.
Tobias