This is the mail archive of the gcc@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: inerting nop's using the DFA scheduler


hello,

 The current gcc scheduler does not insert nops when
there is a stall in the pipeline.
 Since our processor does not automatically stall when
a hazard is detected we modified the function
queue_to_ready()present in haifa-sched.c, as suggested
Naveen sharma.
 Now whenever a stall is detected explicit nops are
emiited. The problem is that this technique works only
with the old scheduler and not with the DFA scheduler.
 Especially when data dependency between two
instructions is represented through define_bypass
there are no stalls and hence nop's cannot be emitted.
 Doesnt queue_to_ready() handle data hazards,
described through define_bypass or is there some other
way the scheduler handles data hazards?

regards,
danish


__________________________________________________
Do You Yahoo!?
Yahoo! Autos - Get free new car price quotes
http://autos.yahoo.com


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