Re: Rename across basic block boundaries

[Richard Sandiford <richard dot sandiford at linaro dot org>]

Bernd Schmidt <bernds@codesourcery.com> writes:
> On 09/01/11 16:16, Richard Sandiford wrote:
>> Bernd Schmidt <bernds@codesourcery.com> writes:
>>> On 08/26/11 14:57, Richard Sandiford wrote:
>>>> Wouldn't a reverse post-order (inverted_post_order_compute) allow even
>>>> more pre-opening (as well as being less code)?
>>>
>>> I can't really tell from the comments what that function is supposed to
>>> produce.
>> 
>> Sorry, I thought it was supposed to produce a reverse postorder, but...
>> 
>>> I've made a change to use it to order the bbs, but that made rr
>>> visit basic blocks without seeing any of their predecessors first,
>> 
>> ...I guess not. :-)  pre_and_rev_post_order_compute should though.
>> Could you try that instead?
>
> That seems to work for me.

Looks good to me.  Maybe here:

> +  /* The order in which we visit blocks ensures that whenever
> +     possible, we only process a block after at least one of its
> +     predecessors, which provides a "seeding" effect to make the logic
> +     in set_incoming_from_chain and init_rename_info useful.  */
> +
> +  for (i = 0; i < n_bbs; i++)
> +    {
> +      basic_block bb1 = BASIC_BLOCK (inverse_postorder[i]);
> +      struct bb_rename_info *this_info = rename_info + i;
> [...]
> +      if (bb1->aux == NULL)
> +	continue;

it would be better to use:

  this_info = (struct bb_rename_info *) bb1->aux;

  if (this_info == NULL)
    continue;

so that we don't care which order the rename_info array is.  You could
then keep the original form of the first loop:

  /* Gather some information about the blocks in this function.  */
  rename_info = XCNEWVEC (struct bb_rename_info, n_basic_blocks);
  ri_index = 0;
  FOR_EACH_BB (bb)
    {
      struct bb_rename_info *ri = rename_info + ri_index;
      ri->bb = bb;
      ri->n_preds = EDGE_COUNT (bb->preds);
      bb->aux = ri;
      ri_index++;
    }

OK with me whichever.

Richard