This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Gcc 3.1 performance regressions with respect to 2.95.3


 In message <200205290253.WAA30328@makai.watson.ibm.com>, David Edelsohn 
writes:
 > 	If it works to not clear SCHED_GROUP_P at the beginning of the
 > group, that's great.  I was concerned whether those objects were allocated
 > with the field zero so that it might randomly be set prior to the first
 > sched_analyze pass.
The only other way for the bit in question to get set for an INSN, CALL_INSN
or JUMP_INSN would be during SSA-DCE where we use it to mark dead code
(which is later deleted).

If we were really worried about this, then we'd just have a pass over all
the insns at the start of scheduling to clear SCHED_GROUP_P rather than 
clearing
it in move_insn, sched_analyze, etc.

jeff


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]