Re: [tree-profiling] Pre-inline optimizations

[Richard Guenther <richard dot guenther at gmail dot com>]

On Thu, 3 Mar 2005 08:51:52 +0100, Jan Hubicka <jh@suse.cz> wrote:
> 
> Hi,
> this patch makes it possible to execute optimizations early (ie before IPA is
> performed).  At the moment we do that by going into SSA and leaving back, but
> in longer run it would be nice to stay in SSA form assuming that memory
> consumption issues are addressed somehow.


> +   NEXT_PASS (pass_all_early_optimizations);
> +   *p = NULL;
> +
> +   p = &pass_all_early_optimizations.sub;
> +   NEXT_PASS (pass_init_datastructures);
> +   NEXT_PASS (pass_referenced_vars);
> +   NEXT_PASS (pass_lower_memref);
> +   NEXT_PASS (pass_build_ssa);
> +   NEXT_PASS (pass_may_alias);
> +   NEXT_PASS (pass_dce);
> +   NEXT_PASS (pass_dominator);
> +   NEXT_PASS (pass_redundant_phi);
> +   NEXT_PASS (pass_tail_recursion);
> +   NEXT_PASS (pass_del_ssa);
> +   NEXT_PASS (pass_cleanup_cfg);
> +   NEXT_PASS (pass_free_datastructures);
>     *p = NULL;

I would have guessed that CCP and DCE are the most useful optimizations,
maybe forward-prop and cprop.  But dom?  That's really expensive.  But I am
again biased for templated C++ code for which it should be the most useful
thing to remove dead code from either constant template parameters or
arguments.

Richard.