On Fri, Aug 13, 2021 at 06:22:48PM -0700, apinski--- via Gcc-patches wrote:
> From: Andrew Pinski <apin...@marvell.com>
>
> Even though this does not change the generated code,
> it does improve the initial RTL generation.
>
> gcc/ChangeLog:
>
> * tree-ssa-math-opts.c (match_arith_overflow):
> Add range and nonzero bits information to
> the new overflow ssa name. Also fold
> the use statement.
> ---
> gcc/tree-ssa-math-opts.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/gcc/tree-ssa-math-opts.c b/gcc/tree-ssa-math-opts.c
> index c4a6492b50d..bb7edeaa6f7 100644
> --- a/gcc/tree-ssa-math-opts.c
> +++ b/gcc/tree-ssa-math-opts.c
> @@ -4221,6 +4221,8 @@ match_arith_overflow (gimple_stmt_iterator *gsi, gimple
> *stmt,
> }
> }
> tree ovf = make_ssa_name (type);
> + set_range_info (ovf, VR_RANGE, wi::zero (TYPE_PRECISION (type)), wi::one
> (TYPE_PRECISION (type)));
Too long line.
> + set_nonzero_bits (ovf, wi::one (TYPE_PRECISION (type)));
Are you sure this is really needed?
set_range_info should have computed that one from the range already:
/* If it is a range, try to improve nonzero_bits from the min/max. */
if (range_type == VR_RANGE)
{
wide_int xorv = ri->get_min () ^ ri->get_max ();
if (xorv != 0)
xorv = wi::mask (precision - wi::clz (xorv), false, precision);
ri->set_nonzero_bits (ri->get_nonzero_bits () & (ri->get_min () | xorv));
}
> @@ -4279,6 +4281,8 @@ match_arith_overflow (gimple_stmt_iterator *gsi, gimple
> *stmt,
> gimple_assign_set_rhs1 (use_stmt, cond);
> }
> }
> + gimple_stmt_iterator gsi1 = gsi_for_stmt (use_stmt);
> + fold_stmt (&gsi1);
> update_stmt (use_stmt);
I don't think it is safe to do that here.
First of all, the update_stmt is right after it, then some further code
that still uses use_stmt right below and this is in a loop that iterates imm
uses of that use_stmt too.
If you want to fold use_stmt, can't it be done after the whole loop?
> if (code == MULT_EXPR && use_stmt != orig_use_stmt)
> {
> --
> 2.27.0
Jakub
