The offload variant of PR65458 - parloops transforms omp-thread functions. AFAIU, running the parloops pass on the host version of offloaded functions, may as well speed up as degrade performance. It's probably best to disable parloops for offloaded functions, at least until we do a targeted exercise to speed up host execution of offloaded functions.
Created attachment 35052 [details] tentative patch
Confirmed.
Author: vries Date: Sat Mar 21 22:22:52 2015 New Revision: 221562 URL: https://gcc.gnu.org/viewcvs?rev=221562&root=gcc&view=rev Log: Mark offloaded functions as parallelized 2015-03-21 Tom de Vries <tom@codesourcery.com> PR tree-optimization/65460 * omp-low.c (expand_omp_target): Set parallelized_function on cgraph_node for child_fn. Modified: branches/gomp-4_0-branch/gcc/ChangeLog branches/gomp-4_0-branch/gcc/omp-low.c
stage1 ping: https://gcc.gnu.org/ml/gcc-patches/2015-04/msg00861.html
patch: https://gcc.gnu.org/ml/gcc-patches/2015-03/msg01063.html
stage1 ping^2: https://gcc.gnu.org/ml/gcc-patches/2015-06/msg00557.html
Author: vries Date: Tue Jun 9 21:14:46 2015 New Revision: 224303 URL: https://gcc.gnu.org/viewcvs?rev=224303&root=gcc&view=rev Log: Mark offloaded functions as parallelized 2015-06-09 Tom de Vries <tom@codesourcery.com> PR tree-optimization/65460 * omp-low.c (expand_omp_target): Set parallelized_function on cgraph_node for child_fn. Modified: trunk/gcc/ChangeLog trunk/gcc/omp-low.c
patch committed to trunk, marking resolved fixed.