RFR: 8282365: Optimize divideUnsigned and remainderUnsigned for constants [v16]

[Quan Anh Mai]

On Mon, 3 Jul 2023 12:28:49 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 50 commits:
>> 
>>  - missing java_negate
>>  - Merge branch 'master' into unsignedDiv
>>  - whitespace
>>  - move asserts to use sites
>>  - windows complaints
>>  - compiler complaints
>>  - undefined internal linkage
>>  - add tests, special casing large shift
>>  - draft
>>  - Merge branch 'master' into unsignedDiv
>>  - ... and 40 more: https://git.openjdk.org/jdk/compare/5b147eb5...eb1f5dd9
>
> src/hotspot/share/opto/divnode.cpp line 906:
> 
>> 904:   }
>> 905: 
>> 906:   // TODO: Improve Value inference of both signed and unsigned division
> 
> Did you miss a `TODO` here?

Currently, we only do `Value` for constant folding, we can restrict the value range of the division in a more rigorous way, e.g `min_jint / 3 < x / y < max_jint / 3` if `y > 3`.

> src/hotspot/share/opto/divnode.cpp line 1419:
> 
>> 1417:   }
>> 1418:   juint con = ti->get_con();
>> 1419:   const Type* u = phase->type(in(1));
> 
> This is a constant foldable bailout? Why do you do it earlier here?
> 
> Generally, I'm starting to wonder if all this code duplication makes sense in all the `Ideal` methods?

Yes we bailout to avoid having to do all the calculations. Let me think how to avoid all these duplications.

> src/hotspot/share/opto/divnode.cpp line 1428:
> 
>> 1426:     return new AndINode(in(1), phase->intcon(con - 1));
>> 1427:   }
>> 1428:   // TODO: This can be calculated directly, see https://arxiv.org/abs/1902.01961
> 
> Stranded `TODO`?

A modulus can be calculated directly without going through the division transformation. Which reduces the cost a little bit.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/9947#discussion_r1261426552
PR Review Comment: https://git.openjdk.org/jdk/pull/9947#discussion_r1261428309
PR Review Comment: https://git.openjdk.org/jdk/pull/9947#discussion_r1261430069