This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: RFA: Patch to constrain maximal number of tries for incomplete dfa descriptions.
Richard Henderson wrote:
> On Fri, Jun 13, 2003 at 05:04:51PM -0400, Vladimir N. Makarov wrote:
> > I don't think so (I don't know how to name the constant). I've chosen it
> > arbitrary (actually it is minimal constant to have only one case in the test
> > when max_look_ahead_tries constraints the search). The parameter has no
> > sense for descriptions when all insn are described. All dfa descriptions
> > but sparc ones describe all insns.
>
> Hmm, ok.
>
> I wonder, then, if we should consider this a bug in the Sparc
> description? If so, is there a nice way for us to detect this
> condition at compiler build-time?
>
Yes, the right way to fix this is to describe all insns. But I don't think it is
possible to make it for gcc3.3.1. I should have done the check in another place
(in insn-attrtab.c). It will be right place to check the description
completeness. Any undescribed insn would trigger abort. Also I have some idea
to use a dynamic programming algorithm to speed up the 1st cycle multipass insn
scheduling.
Taking all this into account, I think the patch is an adequate solution.
>
> > PR bootstrap/10835
> > * haifa-sched.c (max_lookahead_tries,
> > cached_first_cycle_multipass_dfa_lookahead,
> > cached_issue_rate): New variables.
> > (max_issue): Check the number of tries.
> > (choose_ready): Calculate max_lookahead_tries.
> > (sched_init): Check cached_issue_rate.
> In the mean-time, this is ok.
Thanks.