make genattrtab generate smaller code

[Geoff Keating]

law@redhat.com writes:

> In message <Pine.LNX.4.44.0309052057230.13736-100000@wotan.suse.de>, Michael Ma
> tz writes:
>  >Hi,
>  >
>  >this patch makes the code in insn-attrtab.c cache the attributes.
>  >It currently generates code like:
> [ ... ]
> Cool.
> 
> Any chance you could take on the gt-*.h files which have some commoning
> opportunities:
> 
> For example gt_ggc_mx_lang_tree_node:
> 
>             case TS_COMMON:
>               gt_ggc_m_9tree_node ((*x).generic.common.chain);
>               gt_ggc_m_9tree_node ((*x).generic.common.type);
>               gt_ggc_m_10tree_ann_d ((*x).generic.common.ann);
>               gt_ggc_m_10location_s ((*x).generic.common.locus);
>               break;
>             case TS_INT_CST:
>               gt_ggc_m_9tree_node ((*x).generic.int_cst.common.chain);
>               gt_ggc_m_9tree_node ((*x).generic.int_cst.common.type);
>               gt_ggc_m_10tree_ann_d ((*x).generic.int_cst.common.ann);
>               gt_ggc_m_10location_s ((*x).generic.int_cst.common.locus);
>  
> 
> Even if we don't want to commonize across the case statements, there's
> clearly stuff within each case which could be commonized.
> 
> This kind of code is generated pretty regularly by the gt* generator.

The problem is that while you know and I know that generic.common.type
is the same as generic.int_cst.common.type, gengtype doesn't know.
The best way to do this would be to write an optimisation pass in the
compiler that could detect identical code and commonize it---that
would be a good general space optimisation, and it's not really any
harder than doing it in gengtype.

-- 
- Geoffrey Keating <geoffk@geoffk.org>