I believe this is correct, and won't cause any issues with phi node
placement, etc. before dead_cf comes and gets rid any code in the
unreachable block(s).
Reviewed-by: Connor Abbott <cwabbo...@gmail.com>
BTW, I'd ask that you hold off on pushing this until I at least get a
chance to look at the phi builder stuff. I've already talked with you
about the return lowering pass, and the inlining itself shouldn't be
*that* bad, but when I glanced at the phi builder interface I couldn't
grasp quite what was going on.
On Sat, Feb 13, 2016 at 9:14 PM, Jason Ekstrand <ja...@jlekstrand.net> wrote:
> Previously, nir_dominance.c didn't properly handle unreachable blocks.
> This can happen if, for instance, you have something like this:
>
> loop {
> if (...) {
> break;
> } else {
> break;
> }
> }
>
> In this case, the block right after the if statement will be unreachable.
> This commit makes two changes to handle this. First, it removes an assert
> and allows block->imm_dom to be null if the block is unreachable. Second,
> it properly skips unreachable blocks in calc_dom_frontier_cb.
>
> Cc: Connor Abbott <cwabbo...@gmail.com>
> ---
> src/compiler/nir/nir_dominance.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/src/compiler/nir/nir_dominance.c
> b/src/compiler/nir/nir_dominance.c
> index b345b85..d95f396 100644
> --- a/src/compiler/nir/nir_dominance.c
> +++ b/src/compiler/nir/nir_dominance.c
> @@ -94,7 +94,6 @@ calc_dominance_cb(nir_block *block, void *_state)
> }
> }
>
> - assert(new_idom);
> if (block->imm_dom != new_idom) {
> block->imm_dom = new_idom;
> state->progress = true;
> @@ -112,6 +111,11 @@ calc_dom_frontier_cb(nir_block *block, void *state)
> struct set_entry *entry;
> set_foreach(block->predecessors, entry) {
> nir_block *runner = (nir_block *) entry->key;
> +
> + /* Skip unreachable predecessors */
> + if (runner->imm_dom == NULL)
> + continue;
> +
> while (runner != block->imm_dom) {
> _mesa_set_add(runner->dom_frontier, block);
> runner = runner->imm_dom;
> --
> 2.5.0.400.gff86faf
>
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
