Re: [PATCH v3 0/3] MR698: vkd3d-shader/ir: Introduce a better CFG structurizer, part 4.

[Conor McCarthy (＠cmccarthy)]

Conor McCarthy (@cmccarthy) commented about libs/vkd3d-shader/ir.c:
> -
>         * important that the loop break target matches
> -
>         * `successor'. So we have to move that interval's begin
> -
>         * point to the begin point of the candidate interval,
> -
>         * i.e. `begin'. But what if the interval we should extend
> -
>         * backward is not synthetic? This cannot happen,
> -
>         * fortunately, because it would mean that there is a jump
> -
>         * entering a loop via a block which is not the loop
> -
>         * header, so the CFG would not be reducible. */
> -
>        for (k = 0; k < cfg->loop_interval_count; ++k)
> -
>        {
> -
>            struct cfg_loop_interval *interval = &cfg->loop_intervals[k];
> -
> -
>            if (interval->begin < successor->order_pos && successor->order_pos < interval->end)
> -
>            {
> -
>                if (interval->synthetic)
> -
>                    interval->begin = min(begin, interval->begin);

`interval` is nested within an outer interval (`begin`, `successor->order_pos`), correct? I understand the value of moving the nested `begin` up as high as possible in case there are other branches it should contain, but are there other reasons for doing this? In particular, does the algorithm depend on it?