Re: [patch] Armv7

[Richard Earnshaw <rearnsha at arm dot com>]

On Thu, 2006-02-23 at 15:46, Paul Brook wrote:
> On Thursday 23 February 2006 15:32, Daniel Jacobowitz wrote:
> > On Thu, Feb 23, 2006 at 03:25:40PM +0000, Paul Brook wrote:
> > > The attached patch implements Armv7 binutils support. There are three
> > > different "profiles" in armv7: v7A, v7R and v7M. The A and R profiles are
> > > basically armv6T2 plus a handful of extra instructions.
> > > The v7M profile removes the traditional Arm state (leaving only Thumb-2),
> > > and also removes some of the v5te and v6 instructions.
> > >
> > > I'm not entirely sure what "-march=armv7" should mean. I can think of 2
> > > options:
> >
> > Do we need to have -march=armv7?  Why, if they're so different?
> 
> My main motivation is so that I can build a single gcc multilib that will work 
> (and generate reasonable code) on all Armv7 CPUs. The common subset of Armv7 
> still generates significantly better Thumb code than Armv5. 

I think it should be option a, as you've already suggested.

There are other potentially interesting combinations of the variants. 
For example, v7M & v7R and v7R & v7A (where & is the intersection). 

R.