> >> Different targets use the machine reorg pass for all sorts of different > >> things. Most of the code in reorg.c is actually not the machine reorg > >> pass, it is the delay slots pass (pass_delay_slots). The machine reorg > >> pass (pass_machine_reorg) simply calls targetm.machine_dependent_reorg, > >> which is what a backend (in config/*) calls > >> TARGET_MACHINE_DEPENDENT_REORG. > > > > Which means if config/arch does NOT define TARGET_MACHINE_DEPENDENT_REORG > > , then pass_machine_reorg does NOTHING for that arch. Am I right? > > Correct.
I am looking into config/arch which defines TARGET_MACHINE_DEPENDENT_REORG and trying to figure out what kind of operations might change the CFG. Take the function ix86_reorg in config/i386/i386.c as an example. 1. Functions like ix86_pad_short_function, ix86_pad_returns and ix86_avoid_jump_mispredicts add padding, i.e., nop instruction. 2. Function move_or_delete_vzeroupper move or delete vzeroupper in different cases. I think above operations only modify basic blocks in the CFG but NOT the edges between basic blocks. And the CFG should be the same in this case, right? Although I am not 100% sure if they do exactly what the comments say. ;-) Regards, chenwj -- Wei-Ren Chen (陳韋任) Computer Systems Lab, Institute of Information Science, Academia Sinica, Taiwan (R.O.C.) Tel:886-2-2788-3799 #1667
