This is the mail archive of the 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: [committed] R3000 scheduling description

On Thu, 2004-06-24 at 02:34, Richard Sandiford wrote:
> Eric Christopher <> writes:
> > Adds a description for the r3000 scheduling description. No noticable
> > change in scheduling.
> Can you be more specific?  Like: how did you test for differences?

I had a set of .i files that I ran through. It strangely didn't have any

> More importantly, the original description had a separate memory unit,
> but here you're using the ALU to do loads.  It really does make a
> difference, especially when you're reserving units for more than one
> cycle, as you're doing here.

> If the original description is to be believed, there's a separate
> unit for floating-point multiplications ("mult").

> ...and a separate unit again for division.

I'll just clip these together and fix them. Thanks.

> > -       (eq_attr "cpu" "!r3000,r3900,r4600,r4650,r4100,r4120,r4300,r5000"))
> > +       (eq_attr "cpu" "r3900,r4600,r4650,r4100,r4120,r4300,r5000"))
> Erm, surely you don't want to remove "!"?  Lots of other instances
> in the patch as well.

Actually, I think so. I believe the !r3000 is so that we don't match
r3000 on that. Now that we're not going to be using that type of
description for r3000 it shouldn't matter. Of course, I could be wrong.


Eric Christopher <>

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