On Thu, 19 Nov 2020, Jan Hubicka wrote:
> Hi,
> after fixing few issues I gotto stage where 1.4M icf mismatches are due to
> comparing two gimple clobber. The problem is that operand_equal_p match
> clobber
>
> case CONSTRUCTOR:
> /* In GIMPLE empty constructors are allowed in initializers of
> aggregates. */
> return !CONSTRUCTOR_NELTS (arg0) && !CONSTRUCTOR_NELTS (arg1);
>
> But this happens too late after comparing semantics of its type (that
> are not very relevant for memory store and fails way too often).
>
> In the context of ipa-icf we do not really need to match RHS of gimple
> clobbers:
> it is enough to know that the LHS stores can be considered equivalent.
>
> I this added logic to hash them all the same way and compare using
> TREE_CLOBBER_P flag. I see other option in extending operand_equal_p
> in fold-const to handle them more generously or making stmt hash and compare
> to skip comparing/hashing RHS of gimple_clobber_p.
>
> I am now down to 1453 opernad_equal_p mismatches so it seems we are getting
> into shape.
>
> Bootstrapped/regtested x86_64, looks reasonable?
> Honza
>
> * ipa-icf-gimple.c (func_checker::hash_operand): Hash gimple clobber.
> (func_checker::operand_equal_p): Special case gimple clobber.
> diff --git a/gcc/ipa-icf-gimple.c b/gcc/ipa-icf-gimple.c
> index ffb1baddbdb..69bc9ab5b34 100644
> --- a/gcc/ipa-icf-gimple.c
> +++ b/gcc/ipa-icf-gimple.c
> @@ -245,6 +245,14 @@ func_checker::hash_operand (const_tree arg,
> inchash::hash &hstate,
> break;
> }
>
> + /* In gimple all clobbers can be considered equal: while comparaing two
> + gimple clobbers we match the left hand memory accesses. */
> + if (TREE_CLOBBER_P (arg))
> + {
> + hstate.add_int (0xc10bbe5);
> + return;
> + }
> +
> return operand_compare::hash_operand (arg, hstate, flags);
> }
>
> @@ -306,6 +314,10 @@ func_checker::operand_equal_p (const_tree t1, const_tree
> t2,
> default:
> break;
> }
> + /* In gimple all clobbers can be considered equal. We match the right hand
left hand
otherwise yes, I guess this will work for ICF.
> + memory accesses. */
> + if (TREE_CLOBBER_P (t1) || TREE_CLOBBER_P (t2))
> + return TREE_CLOBBER_P (t1) == TREE_CLOBBER_P (t2);
>
> return operand_compare::operand_equal_p (t1, t2, flags);
> }
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imend
