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: rtlopt merge part 2 - bb-reorder rewrite (updated)


On Sat, Feb 08, 2003 at 01:32:06PM +0100, Jan Hubicka wrote:
> It does return minimal size.  get_attr_length is having code to set
> 0 addresses.  PA port does have C function that access addresses
> unconditionally, but I sent patch for that some time ago.
> Minimal size is probably good estimate.

Ok.

> > (1) At present this is being run after sched2, which means that the block
> >     duplication we're doing isn't as useful as it could be.  If we duplicate
> >     block C and place it after block B, we'd like to be able to interleave
> >     the insns from C and B.
> 
> I sent patch for that already in the merge patch adding switch for
> superblock and trace scheduling.  it is disabled for normal scheduling
> but we can run bb-reorder before sched2 uncodnitonally and only rerun it
> when reg-stack or crossjumping happent.

Ok.  I'll get to that soon enough then.

> >     The current placement of bb-reorder stems from wanting to do the reorder
> >     after reg-stack does it's nastiness.  Have you looked into what the
> >     effects of running bb-reorder twice?  If this is too painful, we could
> 
> Yes, it works well in general.

Good.


r~


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