http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52934
Bug #: 52934 Summary: enhancement: cshift0 should be inlined Classification: Unclassified Product: gcc Version: 4.8.0 Status: UNCONFIRMED Severity: minor Priority: P3 Component: fortran AssignedTo: unassig...@gcc.gnu.org ReportedBy: m...@gcc.gnu.org polyhedrons protein would improve quite a bit when the cshift0 intrinsic would be inlined. With the patch below on my machine the runtime for protein drops from 0m21.367s to 0m19.227s (i.e. by 10%). Inlining a generic cshift probably wouldn't bring the same improvement, but still. (a bit related is my patch to improve cshift0 at http://gcc.gnu.org/ml/gcc-patches/2012-04/msg00533.html ) --- protein.f90.mm 2011-12-02 16:22:47.000000000 +0100 +++ protein.f90 2011-12-02 16:16:38.000000000 +0100 @@ -250,7 +250,20 @@ out3: do n1 = 1, polypeptide_len ! total_surface_energy = 0 true_total_surface_energy = 0.0_LONGreal - shifted_surface_sequence = cshift(surface_sequence(:,l),m-1) + if (m == 1) then + shifted_surface_sequence = surface_sequence(:,l) + else if (m == 2) then + shifted_surface_sequence(1:3) = surface_sequence(2:4,l) + shifted_surface_sequence(4) = surface_sequence(1,l) + else if (m == 3) then + shifted_surface_sequence(1:2) = surface_sequence(3:4,l) + shifted_surface_sequence(3:4) = surface_sequence(1:2,l) + else if (m == 4) then + shifted_surface_sequence(1) = surface_sequence(4,l) + shifted_surface_sequence(2:4) = surface_sequence(1:3,l) + else + shifted_surface_sequence = cshift(surface_sequence(:,l),m-1) + end if do k = 1, number_of_polypeptides surface_energy = 0 true_surface_energy = 0.0_LONGreal