This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH]: Structure aliasing, part 1
On Sat, 2005-03-12 at 10:33 -0500, Diego Novillo wrote:
> On 03/11/05 14:32, Daniel Berlin wrote:
>
> > Note that ada will bootstrap about 10-15% slower right now, with some
> > stragglers being 30% slower.
> >
> Two options. We either disable SFTs for Ada or tweak --param
> global-var-threshold to be much lower for Ada.
I've simply disabled SFT's for ada for now, through making it a flag and
making the ada frontend set the flag to 0.
This is because it looks like the threshold would have to be a *lot*
lower than it is now (it's currently half a million, and i'd probably
have to set it to 10000 or less), and that seems more significant a
change to the patch than simply not enabling it for Ada for now.
> In the meantime we need
> to figure out better heuristics for both .global_var and SFT generation.
>
> There's a FIXME note in maybe_create_global_var about creating several
> .GVs, one per type or alias set. Perhaps that's another idea to try.
> The thing is that once we created .GV it's game over for the aliaser.
> Suddenly everything is related to everything.
>
> The patch is fine with those Ada tweaks to the GV threshold.
>
>
> Diego.