Re: [patch] Fix for postreload

[Chung-Lin Tang <cltang at codesourcery dot com>]

On 2010/9/14 äå 02:13, Eric Botcazou wrote:
> > I think it's because the second else-if clause exchanges base_reg (which
> > is here known to be a constant) and XEXP(src,1), the latter becoming the
> > new "base reg" below. So reg_mode[REGNO(XEXP(src,1))] is what matters here.
>
> I think that reg_mode should be checked whenever reg_offset is used as a
> replacement, so the first check is correct and the second should be changed.

From the surrounding context, I think reg_mode[] is checked in 
accordance with updating reg_base_reg[], not reg_offset[].

That said, this discussion is sort of diverging off my patch :)
My fix was only to add the required reg_symbol_ref[] tests, to ensure 
the case is 'offset' instead of 'sym + offset'. How the mode checks 
should be modified can be a later patch.