This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
Re: Unbreak md_estimate_size_before_relax for multi-pass relaxation,committed (tc-cris.c)
- To: alan at linuxcare dot com dot au
- Subject: Re: Unbreak md_estimate_size_before_relax for multi-pass relaxation,committed (tc-cris.c)
- From: Hans-Peter Nilsson <hans-peter dot nilsson at axis dot com>
- Date: Wed, 4 Apr 2001 06:42:51 +0200
- CC: hans-peter dot nilsson at axis dot com, binutils at sources dot redhat dot com
> Date: Wed, 4 Apr 2001 14:19:12 +1000 (EST)
> From: Alan Modra <alan@linuxcare.com.au>
> On Wed, 4 Apr 2001, Hans-Peter Nilsson wrote:
>
> > It is now necessary to set fr_var according to fr_subtype also
>
> I believe this is only necessary if a port makes use of fr_var in
> machine-dependent code. The generic code doesn't care about fr_var in
> rs_machine_dependent frags.
>
> I suspect that if fr_var was set
> correctly first time around, you wouldn't need to do anything on
> subsequent relax passes.
The port needs to return the size of the variant part for states
where it didn't matter before. Some ports, particularly among
those I listed, return the same unmodified fr_var, or at least
do not return a value based on fr_subtype as necessary.
As you know, a correct return-value on subsequent rounds is
necessary for correct fr_address calculation.
brgds, H-P