This is the mail archive of the
mailing list for the GCC project.
Re: Ping: [PATCH] Enable bbro for -Os
- From: Steven Bosscher <stevenb dot gcc at gmail dot com>
- To: Zhenqiang Chen <zhenqiang dot chen at arm dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Fri, 24 Aug 2012 14:17:01 +0200
- Subject: Re: Ping: [PATCH] Enable bbro for -Os
- References: <email@example.com>
On Wed, Aug 22, 2012 at 8:49 AM, Zhenqiang Chen <firstname.lastname@example.org> wrote:
>> The patch is to enable bbro for -Os. When optimizing for size, it
>> * avoid duplicating block.
>> * keep its original order if there is no chance to fall through.
>> * ignore edge frequency and probability.
>> * handle predecessor first if its index is smaller to break long trace.
You do this by inserting the index as a key. I don't fully understand
this change. You're assuming that a block with a lower index has a
lower pre-order number in the CFG's DFS spanning tree, IIUC (i.e. the
blocks are numbered sequentially)? I'm not sure that's always true. I
think you should add an explanation for this heuristic.
>> * only connect Trace n with Trace n + 1 to reduce long jump.
>> * bb-reorder.c (connect_better_edge_p): New added.
>> (find_traces_1_round): When optimizing for size, ignore edge
>> and probability, and handle all in one round.
>> (bb_to_key): Use bb->index as key for size.
>> (better_edge_p): The smaller bb index is better for size.
>> (connect_traces): Connect block n with block n + 1;
>> connect trace m with trace m + 1 if falling through.
>> (copy_bb_p): Avoid duplicating blocks.
>> (gate_handle_reorder_blocks): Enable bbro when optimizing for -Os.
This probably fixes PR54364.
> @@ -1169,6 +1272,10 @@ copy_bb_p (const_basic_block bb, int code_may_grow)
> int max_size = uncond_jump_length;
> rtx insn;
> + /* Avoid duplicating blocks for size. */
> + if (optimize_function_for_size_p (cfun))
> + return false;
> if (!bb->frequency)
> return false;
This shouldn't be necessary, due to the CODE_MAY_GROW argument, and
this change should result in a code size increase because jumps to
conditional jumps aren't removed anymore. What did you make this
change for, do you have a test case where code size increases if you
allow copy_bb_p to return true?