Re: [REVISED][PATCH/RFT] Fix PR middle-end/PR28690, modify swap_commutative_operands_p

[Peter Bergner <bergner at vnet dot ibm dot com>]

On Tue, 2007-06-12 at 18:50 +0200, Rask Ingemann Lambertsen wrote:
> On Tue, Jun 12, 2007 at 10:46:48AM -0500, Peter Bergner wrote:
> 
> >   2) It seems like we could do better than the all or nothing alignment
> >      .p2align gives us.  Obviously, we can't change the behavior of
> >      .p2align, but it would have been nice (maybe a different/new
> >      alignment macro?) if instead of giving up completely, it would have
> >      increased the alignment as much as it could up to the requested 16
> >      bytes within the 7 byte skip maximum.  So for example, if we 
> >      currently have 2 byte alignment and we ask for 16 byte alignment,
> >      wouldn't it be nice if it gave us 8 byte alignment (which is within
> >      7 bytes) rather than giving up completely?
> 
>    That would be '.p2align 3', wouldn't it?

I'm talking about .p2align's behavior and not the args to pass to it.
Currently, you either get the alignment you ask for within the limits
you specify, or you get nothing.  So ".p2align 4,,7" will give you
16 byte alignment or it won't change it at all.  All I'm saying, is
if it can't give me 16 byte alignment because we'd have to skip more
than 7 bytes, why can't it give me 8 byte alignment instead?  ...or
4 byte alignment?  ...or whatever is more aligned than my current
alignment but still within the maximum byte skip value I specified?

Peter