[Mesa-dev] [PATCH 07/15] i965/fs: Create the COPY() set for use in copy propagation dataflow.

[Paul Berry]

On 12 August 2013 13:11, Kenneth Graunke <kenneth at whitecape.org> wrote:

> This is the "COPY" set from Muchnick's textbook, which is necessary
> to do the dataflow algorithm correctly.
>

I believe this can be done more easily.  The only ACP entries that are in
the hastable on exit from opt_copy_propagate_local() are the ones that made
it to the end of the block.  So we should be able to simply populate the
new "COPY" set in the "for" loop at the bottom of the fs_copy_prop_dataflow
constructor.  We don't need to do the extra work you're doing in
setup_initial_values().


>
> Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
> ---
>  .../drivers/dri/i965/brw_fs_copy_propagation.cpp   | 27
> ++++++++++++++++++----
>  1 file changed, 23 insertions(+), 4 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
> b/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
> index 7aff36b..2970af6 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
> @@ -64,6 +64,13 @@ struct block_data {
>     BITSET_WORD *liveout;
>
>     /**
> +    * Which entries in the fs_copy_prop_dataflow acp table are generated
> by
> +    * instructions in this block which reach the end of the block without
> +    * being killed.
> +    */
> +   BITSET_WORD *copy;
> +
> +   /**
>      * Which entries in the fs_copy_prop_dataflow acp table are killed
> over the
>      * course of this block.
>      */
> @@ -113,6 +120,7 @@ fs_copy_prop_dataflow::fs_copy_prop_dataflow(void
> *mem_ctx, cfg_t *cfg,
>     for (int b = 0; b < cfg->num_blocks; b++) {
>        bd[b].livein = rzalloc_array(bd, BITSET_WORD, bitset_words);
>        bd[b].liveout = rzalloc_array(bd, BITSET_WORD, bitset_words);
> +      bd[b].copy = rzalloc_array(bd, BITSET_WORD, bitset_words);
>        bd[b].kill = rzalloc_array(bd, BITSET_WORD, bitset_words);
>
>        for (int i = 0; i < ACP_HASH_SIZE; i++) {
> @@ -148,15 +156,26 @@ fs_copy_prop_dataflow::setup_initial_values()
>           if (inst->dst.file != GRF)
>              continue;
>
> -         /* Mark ACP entries which are killed by this instruction. */
>           for (int i = 0; i < num_acp; i++) {
> -            if (inst != acp[i]->inst &&
> -                (inst->overwrites_reg(acp[i]->dst) ||
> -                 inst->overwrites_reg(acp[i]->src))) {
> +            if (inst == acp[i]->inst) {
> +               /* Add this entry to the COPY set. */
> +               BITSET_SET(bd[b].copy, i);
> +            } else if (inst->overwrites_reg(acp[i]->dst) ||
> +                       inst->overwrites_reg(acp[i]->src)) {
> +               /* The current instruction kills this copy.  Add the entry
> to
> +                * the KILL set.
> +                */
>                 BITSET_SET(bd[b].kill, i);
>              }
>           }
>        }
> +
> +      /* Anything killed did not make it to the end of the block, so it
> +       * shouldn't be in COPY.
> +       */
> +      for (int i = 0; i < bitset_words; i++) {
> +         bd[b].copy[i] &= ~bd[b].kill[i];
> +      }
>     }
>  }
>
> --
> 1.8.3.4
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20130815/8634c458/attachment-0001.html>