This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [RFC] Extend ipa-bitwise-cp with pointer alignment propagation
- From: Jan Hubicka <hubicka at ucw dot cz>
- To: Prathamesh Kulkarni <prathamesh dot kulkarni at linaro dot org>
- Cc: gcc Patches <gcc-patches at gcc dot gnu dot org>, Jan Hubicka <hubicka at ucw dot cz>, Martin Jambor <mjambor at suse dot cz>, Richard Biener <rguenther at suse dot de>
- Date: Thu, 22 Sep 2016 13:56:35 +0200
- Subject: Re: [RFC] Extend ipa-bitwise-cp with pointer alignment propagation
- Authentication-results: sourceware.org; auth=none
- References: <CAAgBjMkR_Am6Zn26zx+Ah7AHZjhKe8J+156VWxZG74C+QZFADw@mail.gmail.com>
> Hi,
> The attached patch tries to extend ipa bits propagation to handle
> pointer alignment propagation.
> The patch just disables ipa-cp-alignment pass, I suppose we want to
> eventually remove it ?
Yes, can you please verify that alignments it computes are monotonously
worse than those your new code computes and include the removal in the
next iteration of the patch?
>
> Bootstrap+tested on x86_64-unknown-linux-gnu.
> Cross-tested on arm*-*-*, aarch64*-*-*.
> Does the patch look OK ?
>
> Thanks,
> Prathamesh
> @@ -2258,8 +2271,8 @@ propagate_constants_accross_call (struct cgraph_edge *cs)
> &dest_plats->itself);
> ret |= propagate_context_accross_jump_function (cs, jump_func, i,
> &dest_plats->ctxlat);
> - ret |= propagate_alignment_accross_jump_function (cs, jump_func,
> - &dest_plats->alignment);
> +// ret |= propagate_alignment_accross_jump_function (cs, jump_func,
> +// &dest_plats->alignment);
obviously we do not want commented out ocde..
> ret |= propagate_bits_accross_jump_function (cs, i, jump_func,
> &dest_plats->bits_lattice);
> ret |= propagate_aggs_accross_jump_function (cs, jump_func,
> diff --git a/gcc/ipa-prop.c b/gcc/ipa-prop.c
> index 1629781..5cee27b 100644
> --- a/gcc/ipa-prop.c
> +++ b/gcc/ipa-prop.c
> @@ -1701,6 +1701,16 @@ ipa_compute_jump_functions_for_edge (struct ipa_func_body_info *fbi,
> jfunc->bits.mask = 0;
> }
> }
> + else if (POINTER_TYPE_P (TREE_TYPE (arg)))
> + {
> + unsigned HOST_WIDE_INT bitpos;
> + unsigned align;
> +
> + jfunc->bits.known = true;
> + get_pointer_alignment_1 (arg, &align, &bitpos);
> + jfunc->bits.mask = wi::mask<widest_int>(TYPE_PRECISION (TREE_TYPE (arg)), false).and_not (align / BITS_PER_UNIT - 1);
... and long lines :)
> + jfunc->bits.value = bitpos / BITS_PER_UNIT;
> + }
> else
> gcc_assert (!jfunc->bits.known);
>
> @@ -5534,7 +5544,7 @@ ipcp_update_bits (struct cgraph_node *node)
> next_parm = DECL_CHAIN (parm);
>
> if (!bits[i].known
> - || !INTEGRAL_TYPE_P (TREE_TYPE (parm))
> + || !(INTEGRAL_TYPE_P (TREE_TYPE (parm)) || POINTER_TYPE_P (TREE_TYPE (parm)))
I suppose eventually we may want to enable other types, too.
It does even make sense to propagate this on aggregates, but definitly on
vectors and complex numbers.
Otherwise the patch seems fine to me (modulo Richard's comments)
Honza