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]

Re: The DFA based scheduler.


On Tue, Sep 04, 2001 at 11:58:17AM -0400, Vladimir Makarov wrote:
> Naveen Sharma wrote:
> > 
> > Hi,
> > 
> > I have been interested by the discussion on the new DFA based scheduler
> > developed by you for gcc.I plan to study the implementation of the
> > scheduler and would like to contribute further for the scheduler.
> > My friend,Nitin Gupta,would  be joining me in the effort.
> > At present we are interested in using the DFA based scheduler
> > for SH4 target which might benefit from the new scheduler.
> > 
> > It would be helpful if you can answer some initial queries.
> > 
> > 1. What targets are tested till now &
> >    What improvements in generated code were noticed?
> 
>   There are a lot of descriptions tested in RedHat (mips, powerpc, a
> Fujitsu VLIW processor, even IA64).  They were tested for the
> correctness of the DFA-based scheduler work mainly.
> 
>   I've seen some improvements for the Fujistu VLIW processor.  The dfa
> based scheduler is used to solve VLIW insn packing too with the aid of
> the code for multipass insn scheduling (see hook
> TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD).
> 
>   I don't know about improvements for mips.  Probably there are no
> improvement.  David Edelsohn reported that there is no improvement for
> powerpc too.  As for IA64, I've seen improvement about 0.4-0.5% for all
> SPECInt95  (although some SPECInt95 tests had degradation).  The DFA
> based scheduler for IA64 was complicated and buggy (I had few time for
> this).  May be the result can be improved (if the degradation of some
> SPECint95 tests is eliminated).

On the EEMBC benchmarks, I saw a 5-15 percent speedup by going to the DFA
scheduler for a particular MIPS target.  It is possible that the original
scheduler implementation was not optimized for the particular MIPS model, so
this may be a best case scenario.

-- 
Michael Meissner, Red Hat, Inc.  (GCC group)
PMB 198, 174 Littleton Road #3, Westford, Massachusetts 01886, USA
Work:	  meissner@redhat.com		phone: +1 978-486-9304
Non-work: meissner@spectacle-pond.org	fax:   +1 978-692-4482


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