This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: RFC: New approach to --with-cpu
- From: Daniel Jacobowitz <drow at mvista dot com>
- To: John David Anglin <dave at hiauly1 dot hia dot nrc dot ca>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Tue, 30 Jul 2002 23:47:32 -0400
- Subject: Re: RFC: New approach to --with-cpu
- References: <200207310131.g6V1VfNE021308@hiauly1.hia.nrc.ca>
On Tue, Jul 30, 2002 at 09:31:41PM -0400, John David Anglin wrote:
> >From a PA standpoint, the main objection that I have to this approach
> is that there seems to be a slight loss of flexibility. You use "with_cpu"
> to set the default scheduling model. However, I was also planning to set
> the default arch based on the target machine when a linker bug in handling
> PA2.0 code is fixed. I view these options as independent although the
> configure defaults would both be based on the target triplet. Thus,
> it would be necessary to introduce a new configure option to handle
> the arch default. In the future, I can see other default code generation
> options being set-table.
Well, that's no problem. A first cut of this patch offered defaults
for -march= and -mtune= separately on MIPS. I could do the same for
PA easily.
> I must admit that I don't really like the multilib concept and find
> it a rather idea to build libraries for all scheduling models,
> archs, etc. Thus, I don't find this justification attractive.
I don't like multilibs either. Eric mentioned some concerns that this
would interfere with them when I discussed it with him, which is why I
brought it up.
The real advantage of my approach is getting anything specs-based
correct. For instance, the preprocessor flags related to -msoft-float
are set correctly if you use --with-float=soft.
--
Daniel Jacobowitz Carnegie Mellon University
MontaVista Software Debian GNU/Linux Developer