I have moved the pass so that it is immediately after pass_ira - no difference, I still see register 75 (and 74). Also, df.c doesn't appear to be in my source - I assume it is now df-core.c which does have quite the documentation and I will take a look. Are you basically saying that after the hard assignments the live sets are not necessarily valid any more? That would be .. oh, the word that comes to mind is "bad"! As an aside, inserting RTL to save and restore rax regardless of whether or not it is live seems to work (need to run the full tests yet) - it just bugs me to be saving things when it may not be necessary. Bill ---------------------------- William Mahoney <wmahoney@xxxxxxxxxxxxxxxx> writes: > Anyway as I said this is called from passes > > -------------------------------- > NEXT_PASS (pass_machine_reorg); > NEXT_PASS (pass_cleanup_barriers); > NEXT_PASS (pass_delay_slots); > NEXT_PASS (pass_split_for_shorten_branches); > NEXT_PASS (pass_convert_to_eh_region_ranges); > NEXT_PASS (pass_bill_instrument); /* Bill was here */ > NEXT_PASS (pass_shorten_branches); > ----------------------------- > > So you see it IS in fact after pass_convert_to_eh_region_ranges as you > noted below. Is that where the hard registers are assigned, and if so, it > causes the live bitmap in the blocks to be wrong? Please reply to the mailing list, not just to me. Thanks. Hard registers are assigned in pass_ira, so quite a bit earlier than where you have your pass. I think you need to do your pass before pass_free_cfg. I don't know what it means to call df_get_live_in after pass_free_cfg. Also, are you calling df_live_add_problem and df_analyze at the start of your pass? See the long comment at the top of df.c. Ian
