[PATCH] PR23948 et al., take 2
Diego Novillo
dnovillo@redhat.com
Tue Jan 10 13:25:00 GMT 2006
On Monday 09 January 2006 00:33, Paolo Bonzini wrote:
> Â Â Â Â Â Â Â Â PR tree-optimization/23109
> Â Â Â Â Â Â Â Â PR tree-optimization/23948
> Â Â Â Â Â Â Â Â PR tree-optimization/24123
>
> Â Â Â Â Â Â Â Â * Makefile.in (tree-ssa-math-opts.o): Adjust dependencies.
> Â Â Â Â * tree-cfg.c (single_noncomplex_succ): New.
> Â Â Â Â * tree-flow.h (single_noncomplex_succ): Declare it.
> Â Â Â Â * tree-ssa-math-opts.c (enum place_reciprocal): Remove.
> Â Â Â Â * tree-ssa-math-opts.c (enum place_reciprocal): Remove.
> Â Â Â Â (struct occurrence, occ_head, occ_pool, is_divide_by,
> compute_merit, insert_bb, register_division_in, insert_reciprocals,
> Â Â Â Â Â Â Â Â replace_reciprocal, free_bb): New.
> Â Â Â Â (execute_cse_reciprocals_1): Rewritten.
> Â Â Â Â (execute_cse_reciprocals): Adjust calls to
> execute_cse_reciprocals_1. Do not commit any edge insertion. Â Always
> compute dominators and create the allocation pool.
> Â Â Â Â * params.def (PARAM_MIN_DIVISIONS_FOR_RECIP_MUL): New.
> Â Â Â Â * params.h (MIN_DIVISIONS_FOR_RECIP_MUL): New.
> Â Â Â Â * passes.c (init_optimization_passes): Run recip after tree loop
> Â Â Â Â optimizations.
> Â Â Â Â * doc/invoke.texi (--param): Add the min-divisions-for-recip-mul
> Â Â Â Â parameter.
>
OK, but I agree with Ian: --param min-divisions-for-recip-mul ought to be a
target hook, not a --param. No need to resubmit the patch. Once you make
it a target hook, the patch can go in.
Another minor point. Should the gate function trigger only for -O2 and up?
I don't feel too strongly about it because this optimization shouldn't
obfuscate the code too much.
More information about the Gcc-patches
mailing list