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: Unreviewed fix for bootstrap failure

> Date: Sun, 29 Jun 2003 00:34:05 -0400 (EDT)
> From: "John David Anglin" <>

> > 2. Are you sure this is a good idea?  What happens if a label is deleted
> >    that's referred to from a REG_EQUIV note, and the note is later
> >    substituted for the register (say, by reload)?
> As I understand the current situation, the only notes that can contain
> label references are REG_EQUAL notes.  These cannot be used for
> substitution, only cse.  Thus, this isn't a problem.

I don't see why that would be the case.  Surely it's valid to have a
REG_EQUIV note that says "this is (label_ref ...)".

> I agree that if there was a more general usage of notes with symbol
> references, then they would have to be included in the label usage
> counts.
> The patch doesn't change the current semantics.  For example, label
> usage counts are not updated when a register note is deleted.  Situations
> when the labels in an instruction are replaced are relatively rare.

Yes, this is probably what's been saving us until now.

> Otherwise, the patch which added these notes would have created a lot
> more problems.  Thus, it seemed the most expediate fix was to the
> replacement.  The other alternative was to delete REG_EQUAL notes
> containing symbol references from instructions that needed label
> replacement.

Why can't we simply keep track of each replacement?  Surely it's as
easy as incrementing one counter, and decrementing another, for each
place where the replacement is done.

- Geoffrey Keating <>

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