2009/8/18 Pekka Enberg <penb...@cs.helsinki.fi>:
> Lets use radix_tree_lookup() and get rid of the nasty loop in
> bb_last_insn(). For some reason, this seems to fix the infinite loop
> triggered by empty basic blocks.
>
> Cc: Arthur HUILLET <arthur.huil...@free.fr>
> Cc: Tomek Grabiec <tgrab...@gmail.com>
> Signed-off-by: Pekka Enberg <penb...@cs.helsinki.fi>
> ---
> jit/basic-block.c | 16 +---------------
> jit/spill-reload.c | 2 +-
> 2 files changed, 2 insertions(+), 16 deletions(-)
>
> diff --git a/jit/basic-block.c b/jit/basic-block.c
> index 6de5c3f..dbf3675 100644
> --- a/jit/basic-block.c
> +++ b/jit/basic-block.c
> @@ -176,21 +176,7 @@ struct insn *bb_first_insn(struct basic_block *bb)
>
> struct insn *bb_last_insn(struct basic_block *bb)
> {
> - struct insn *this = list_entry(bb->insn_list.prev, struct insn,
> insn_list_node);
> -
> - /*
> - * We want to return the last "real" instruction of the basic block.
> Taking the
> - * last of the insn_list will not work in case a live interval has
> been spilled
> - * right after the final jump of the basic block.
> - * This is a side effect of the linear scan algorithm.
> - *
> - * As a result, we browse instructions starting from the last, in
> order to find the one
> - * that has a LIR position matching the position for the end of the
> block.
> - */
> - while (this->lir_pos != bb->end_insn - 1) {
> - this = list_entry(this->insn_list_node.prev, struct insn,
> insn_list_node);
> - }
> - return this;
> + return list_entry(bb->insn_list.prev, struct insn, insn_list_node);
> }
>
> static int __bb_add_neighbor(void *new, void **array, unsigned long *nb)
> diff --git a/jit/spill-reload.c b/jit/spill-reload.c
> index 5ced80e..515db8a 100644
> --- a/jit/spill-reload.c
> +++ b/jit/spill-reload.c
> @@ -183,7 +183,7 @@ static void insert_mov_insns(struct compilation_unit *cu,
> struct insn *spill_at_insn;
> int i;
>
> - spill_at_insn = bb_last_insn(from_bb);
> + spill_at_insn = radix_tree_lookup(cu->lir_insn_map,
> from_bb->end_insn - 1);
>
> /* Spill all intervals that have to be resolved */
> for (i = 0; i < nr_mapped; i++) {
> --
> 1.5.6.3
>
>
Like I said on IRC, this will not work for empty basic blocks, because
spill_at_insn will belong to the preceding
basic block. This causes that instructions will be added to different
(preceding) basic block and might not be executed on some
execution paths.
--
Tomek Grabiec
------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Jatovm-devel mailing list
Jatovm-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/jatovm-devel
