RFR: 8315066: Add unsigned bounds and known bits to TypeInt/Long [v47]

[Quan Anh Mai]

On Tue, 22 Apr 2025 16:01:59 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

>> Quan Anh Mai has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains 61 commits:
>> 
>>  - Merge branch 'master' into unsignedbounds
>>  - Merge branch 'master' into unsignedbounds
>>  - reviews
>>  - Merge branch 'master' into unsignedbounds
>>  - refine comments
>>  - Merge branch 'master' into unsignedbounds
>>  - Merge branch 'master' into unsignedbounds
>>  - harden SimpleCanonicalResult
>>  - number lemmas
>>  - include
>>  - ... and 51 more: https://git.openjdk.org/jdk/compare/0995b940...cdab1911
>
> src/hotspot/share/opto/rangeinference.cpp line 293:
> 
>> 291:     // different bit between the result and lo must be the 3rd bit. As a result,
>> 292:     // the result must not be smaller than:
>> 293:     //           1 0 1 0 0 0 0 0
> 
> Oh, I'm starting to get an intuition here. I think we can make it a bit more "intuitive".
> 
> We start at the "first violation". We would like to flip the bit from 1 to 0, but since we are only allowed to increase the number, we also need to flip the 5th bit. But that is already 1 too, so we need to go to the 4th. That one we cannot flip from 0 to 1, because it is forced to be a 0 by zeros. So the first bit we can flip is the 3rd.
> 
> Maybe this formulation with "which is the highest bit we can flip" could be helpful for the intuition?

That's a better guidance for sure, I have modified this part.

> src/hotspot/share/opto/rangeinference.cpp line 307:
> 
>> 305:     // violation, which is the last set bit of tmp
>> 306:     //           0 1 1 0 0 0 0 0
>> 307:     U tmp = ~either & find_mask;
> 
> Did I understand that right: `tmp` is the bits that we cannot flip? Or is it the ones we can flip? A better name would be appreciated :)
> Same for `either`. worst case you call it `lo_or_zeros`... but that's not great either.
> 
> I'm not fully seeing through the logic here yet, so I struggle to make good suggestions.

We can say that `tmp` is all the bits we can flip, although I think that is too ambiguous, what does "can" mean here. It is better to think of it as all the bits that are not lower than `first_violation` and are 0 in both `lo` and `zeros`.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/17508#discussion_r2054469385
PR Review Comment: https://git.openjdk.org/jdk/pull/17508#discussion_r2054468470