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: RFA: patch modifying ColdFire, power4, an power5 pipeline descriptions


David Edelsohn wrote:
Vlad,

If I understand correctly from you and Pat, your patch fixes a consistency
check that you are adding to the scheduler, but does not affect the current,
inaccurate behavior of the POWER4/5 machine description.  As you found
and we found before, if the pipeline is described accurately, the automaton
is unreasonably large.

Assuming that you patches to power4.md and power5.md do not affect
the size and build times (by retaining the inaccurate behavior) and produce
the same instruction scheduling, the patch is okay.

The compiler size is increased by 4% with these changes and building insn-automata.c is increased by 2 min (from 1min to 3 min) on 1GHz power4.

There is no difference in code for power5 on SPEC2000. For power4, the code is different but there is no visible difference on SPEC2000 scores. The SPEC2000 code size is a bit smaller with the new power4 description.

I don't know did I answer your question. Are these changes ok?


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