On Mon, Jan 20, 2020 at 2:39 AM <apin...@marvell.com> wrote:
>
> From: Andrew Pinski <apin...@marvell.com>
>
> Reported as PR 93321, prepare_block_for_update with some huge
> recusive inlining can go past the stack limit. The loop
> at the end, could be transformed such that the last iteration goes
> back to the begining of the function instead of the call.
> This reduces the stack usage and speeds up slightly
> the function.
>
> OK? Bootstrapped and tested on x86_64-linux-gnu with no regressions.
Can you instead change it to a worklist, see for example
tree-ssa-pre.c:compute_avail
for a recipe.
OK with that change.
Richard.
> ChangeLog:
> * tree-into-ssa.c (prepare_block_for_update): Manaually sibcall
> optimize to self.
> ---
> gcc/tree-into-ssa.c | 16 ++++++++++++----
> 1 file changed, 12 insertions(+), 4 deletions(-)
>
> diff --git a/gcc/tree-into-ssa.c b/gcc/tree-into-ssa.c
> index c27bf2ce121..6e139c3b056 100644
> --- a/gcc/tree-into-ssa.c
> +++ b/gcc/tree-into-ssa.c
> @@ -2616,6 +2616,7 @@ prepare_block_for_update (basic_block bb, bool
> insert_phi_p)
> edge e;
> edge_iterator ei;
>
> +again:
> mark_block_for_update (bb);
>
> /* Process PHI nodes marking interesting those that define or use
> @@ -2695,10 +2696,17 @@ prepare_block_for_update (basic_block bb, bool
> insert_phi_p)
> }
>
> /* Now visit all the blocks dominated by BB. */
> - for (son = first_dom_son (CDI_DOMINATORS, bb);
> - son;
> - son = next_dom_son (CDI_DOMINATORS, son))
> - prepare_block_for_update (son, insert_phi_p);
> + for (son = first_dom_son (CDI_DOMINATORS, bb); son; )
> + {
> + basic_block next = next_dom_son (CDI_DOMINATORS, son);
> + if (!next)
> + {
> + bb = son;
> + goto again;
> + }
> + prepare_block_for_update (son, insert_phi_p);
> + son = next;
> + }
> }
>
>
> --
> 2.17.1
>
