Internal testing shows improvements in some SPEC HPC benchmarks with this change.
omp.loop
omp.target