Re: [PATCH] path solver: only compute ranges at the current path position [PR126876]

On 8/18/26 08:47, Richard Biener wrote:
> On Tue, Aug 18, 2026 at 2:40 PM Aldy Hernandez <[email protected]> wrote:
>> On Tue, Aug 18, 2026 at 12:53:20PM +0200, Richard Biener wrote:
>>> On Tue, Aug 18, 2026 at 12:30 PM Aldy Hernandez <[email protected]> wrote:
>>> Are there not in-BB side-effects to ranges that might or might not be reflected?
>>>
>>> I'm not sure it's sensible to use path ranger as current query.
>> Side effects are block granular, in both rangers AFAIK, so a mid fold
>> query through the path ranger either gets the global range or a cached
>> value from everything seen so far in the path.  So I think we should
>> be OK.
>>
>> However, if we're using the path ranger to ask things from the path,
>> but expecting a root ranger response, that's bound to fail.  To be
>> honest, we haven't stressed the path ranger outside of the backwards
>> threader much (maybe loop-ch and loop unswitching??).  So the
>> conservative option may be not to ask the path ranger itself, or bail
>> if the cache hasn't been set up?
> I think it's weird to end up with path ranger as query and we should
> have fallen back to a "below" active ranger or global ranges.  I'd
> have expected path ranger queries to explicitly go to it, not via
> the get_range_query () active one.  Or is the gimple_ranger
> instantiated by the back threader "tainted" somehow, even
> when not using a path_range_query constructed from it?
>
> Richard.

THe path ranger using the a main ranger for all queries which extend 
outside the path.. this allows all known info leading into the path to 
be handled properly.  It seems reasonable that any request to the path 
ranger for values outside the path should go the root ranger.

  in tree-ssa-dom.cc I see:

   /* Recursively walk the dominator tree optimizing statements. */
   gimple_ranger *ranger = enable_ranger (fun);
   path_range_query path_query (*ranger);

This would be what was happening before my change..   the path ranger 
doing its thing, and then called into fold and the simplifier was then 
invoking  get_range_query() which would then be the root ranger.

We should se the same results if the path ranger queries simply invoke 
the root ranger if the request is from outside the block.

If its from within the path, i would think you would want the values 
from within the path..   but thats above my paygrade :-P

   Andrew