> Am 01.12.2024 um 03:00 schrieb Andrew Pinski <quic_apin...@quicinc.com>:
>
> While looking into PR 117859, I noticed that LIM
> sometimes would produce `bool_var2 = bool_var1 != 0` instead
> of just using bool_var2. This patch allows LIM to reuse bool_var1
> in the place where bool_var2 was going to be used.
>
> Bootstrapped and tested on x86_64-linux-gnu.
Ok
Richard
> gcc/ChangeLog:
>
> * tree-ssa-loop-im.cc (move_computations_worker): While moving
> phi, reuse the lhs of the conditional if it is a boolean type.
>
> Signed-off-by: Andrew Pinski <quic_apin...@quicinc.com>
> ---
> gcc/tree-ssa-loop-im.cc | 21 ++++++++++++++++-----
> 1 file changed, 16 insertions(+), 5 deletions(-)
>
> diff --git a/gcc/tree-ssa-loop-im.cc b/gcc/tree-ssa-loop-im.cc
> index 538a0588d33..0130a809e30 100644
> --- a/gcc/tree-ssa-loop-im.cc
> +++ b/gcc/tree-ssa-loop-im.cc
> @@ -1304,11 +1304,22 @@ move_computations_worker (basic_block bb)
> edges of COND. */
> extract_true_false_args_from_phi (dom, stmt, &arg0, &arg1);
> gcc_assert (arg0 && arg1);
> - t = make_ssa_name (boolean_type_node);
> - new_stmt = gimple_build_assign (t, gimple_cond_code (cond),
> - gimple_cond_lhs (cond),
> - gimple_cond_rhs (cond));
> - gsi_insert_on_edge (loop_preheader_edge (level), new_stmt);
> + /* For `bool_val != 0`, reuse bool_val. */
> + if (gimple_cond_code (cond) == NE_EXPR
> + && integer_zerop (gimple_cond_rhs (cond))
> + && types_compatible_p (TREE_TYPE (gimple_cond_lhs (cond)),
> + boolean_type_node))
> + {
> + t = gimple_cond_lhs (cond);
> + }
> + else
> + {
> + t = make_ssa_name (boolean_type_node);
> + new_stmt = gimple_build_assign (t, gimple_cond_code (cond),
> + gimple_cond_lhs (cond),
> + gimple_cond_rhs (cond));
> + gsi_insert_on_edge (loop_preheader_edge (level), new_stmt);
> + }
> new_stmt = gimple_build_assign (gimple_phi_result (stmt),
> COND_EXPR, t, arg0, arg1);
> todo |= TODO_cleanup_cfg;
> --
> 2.43.0
>
