https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123144

--- Comment #7 from Avinash Jayakar <avinashd at gcc dot gnu.org> ---
Created attachment 63343
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=63343&action=edit
Proposed patch for haifa-sched.cc

I see that the register pressure values are dropping below zero in certain
scenarios with register pressure sensitive scheduling in haifa-sched.cc. Is
this expected?

Because of this there is a mismatch in max_pressure and model's max pressure. I
clipped the value to 0, if it ever drops below it in mark_regno_birth_or_death,
and the ICE does not happen. 
Regtested this change in trunk and I dont see any issues, but want to confirm
if pressure value dropping below 0 is expected.

Reply via email to