http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59469
--- Comment #11 from Markus Trippelsdorf <octoploid at yandex dot com> --- (In reply to Jan Hubicka from comment #8) > OK, the unmangled name of symbol is: > 000000000004c8c0 T > _ZN4llvm21SymbolTableListTraitsINS_10BasicBlockENS_8FunctionEE21transferNodes > FromListERNS_12ilist_traitsIS1_EENS_14ilist_iteratorIS1_EES8_ > > and obviously for me it is T for -O2 and for -O3 it is optimized out. Both > seems legitimate. Can you please re-check with current mainline? > > I get: > df live regs : 7.74 (12%) usr 0.00 ( 0%) sys 7.72 (12%) > wall 0 kB ( 0%) ggc > df live&initialized regs: 11.02 (17%) usr 0.00 ( 0%) sys 10.91 (17%) > wall 0 kB ( 0%) ggc > CPROP : 20.03 (32%) usr 0.44 (65%) sys 20.51 (32%) > wall 7968 kB ( 5%) ggc > > (with checking on, but still seems something for steve, I do not think there > is particularly expensive checking in CPROP) > > I am re-checking with 4.8 4.7 and 4.8 are also slow. perf top shows: 12.80% cc1plus cc1plus [.] compute_cprop_data() 8.64% cc1plus cc1plus [.] df_worklist_dataflow(dataflow*, bitmap_head*, int*, int) 5.80% cc1plus cc1plus [.] bitmap_set_bit(bitmap_head*, int) 4.61% cc1plus cc1plus [.] bitmap_ior_into(bitmap_head*, bitmap_head const*) ...