Improve grouping performance via better vectorization in accumulate functions

[alamb]

Draft as I am running performance tests and will clean up the code a bit if possible.

Which issue does this PR close?

Part of #6889

This is a follow on to #6904 where some last minute changes (to track null state correctly) decreased performance slightly -- I am trying to get some back

Rationale for this change

I am trying to help the compiler to vectorize the inner loops of the accumulators. Let's do this by doing the minimal work as possible

What changes are included in this PR?

Update null state with a separate loop

Are these changes tested?

covered by existing tests

Are there any user-facing changes?

[alamb]

the idea is to use a separate loop

Note we can skip this update entirely once all the groups have been seen -- I will explore if I can figure out how to do that next.

[Dandandan]

Can't we assume seen_values is true for every group_index for "no nulls, no filter" until we observed a first null-value?

[alamb]

That is an excellent point. I will try and take that into account

[alamb]

af3f62b shows a little benefit bit maybe in the noise range

┏━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃ main_base ┃ alamb_faster_grouping ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1     │  560.29ms │              533.10ms │     no change │
│ QQuery 2     │  174.72ms │              162.55ms │ +1.07x faster │
│ QQuery 3     │  167.44ms │              161.30ms │     no change │
│ QQuery 4     │  119.56ms │              117.52ms │     no change │
│ QQuery 5     │  394.33ms │              400.22ms │     no change │
│ QQuery 6     │   45.54ms │               40.08ms │ +1.14x faster │
│ QQuery 7     │  849.06ms │              837.12ms │     no change │
│ QQuery 8     │  258.03ms │              247.16ms │     no change │
│ QQuery 9     │  595.82ms │              566.07ms │     no change │
│ QQuery 10    │  341.31ms │              337.04ms │     no change │
│ QQuery 11    │  170.28ms │              159.25ms │ +1.07x faster │
│ QQuery 12    │  175.19ms │              173.80ms │     no change │
│ QQuery 13    │  331.46ms │              315.21ms │     no change │
│ QQuery 14    │   48.67ms │               49.25ms │     no change │
│ QQuery 15    │   59.04ms │               50.88ms │ +1.16x faster │
│ QQuery 16    │  164.45ms │              164.20ms │     no change │
│ QQuery 17    │  966.84ms │              873.41ms │ +1.11x faster │
│ QQuery 18    │ 1676.43ms │             1637.90ms │     no change │
│ QQuery 19    │  166.95ms │              167.24ms │     no change │
│ QQuery 20    │  313.92ms │              313.73ms │     no change │
│ QQuery 21    │ 1062.69ms │             1060.65ms │     no change │
│ QQuery 22    │   88.17ms │               93.33ms │  1.06x slower │
└──────────────┴───────────┴───────────────────────┴───────────────┘

[alamb]

I am now benchmarking the special case of "don't track nulls they can't possibly be present

[alamb]

Special casing the nulls seems to have helped a little, though the fact that some queries now report slower is surprising to me

┏━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃ main_base ┃ alamb_faster_grouping ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1     │  574.53ms │              522.95ms │ +1.10x faster │
│ QQuery 2     │  168.33ms │              152.40ms │ +1.10x faster │
│ QQuery 3     │  173.45ms │              182.54ms │  1.05x slower │
│ QQuery 4     │  116.60ms │              116.16ms │     no change │
│ QQuery 5     │  402.46ms │              405.81ms │     no change │
│ QQuery 6     │   45.04ms │               41.94ms │ +1.07x faster │
│ QQuery 7     │  922.93ms │              916.25ms │     no change │
│ QQuery 8     │  255.73ms │              240.93ms │ +1.06x faster │
│ QQuery 9     │  577.15ms │              554.53ms │     no change │
│ QQuery 10    │  318.40ms │              330.10ms │     no change │
│ QQuery 11    │  167.56ms │              172.50ms │     no change │
│ QQuery 12    │  174.26ms │              177.85ms │     no change │
│ QQuery 13    │  324.26ms │              320.80ms │     no change │
│ QQuery 14    │   48.11ms │               48.24ms │     no change │
│ QQuery 15    │   54.18ms │               57.05ms │  1.05x slower │
│ QQuery 16    │  164.83ms │              168.32ms │     no change │
│ QQuery 17    │  897.29ms │              877.01ms │     no change │
│ QQuery 18    │ 1672.32ms │             1675.61ms │     no change │
│ QQuery 19    │  179.13ms │              166.64ms │ +1.07x faster │
│ QQuery 20    │  324.28ms │              324.27ms │     no change │
│ QQuery 21    │ 1147.35ms │             1177.47ms │     no change │
│ QQuery 22    │   93.86ms │               91.11ms │     no change │

[Dandandan]

Noise maybe? Your previous run seems to have some noisy results as well.
I only get relatively relatively stable perf results by closing all browsers / IDEs etc.

[alamb]

Noise maybe? Your previous run seems to have some noisy results as well.
I only get relatively relatively stable perf results by closing all browsers / IDEs etc.

Yeah, I am trying to control for noise by running these experiments on a GCP machine (not my laptop) 🤔

[alamb]

Tracking this basic idea in #7066 , I don't plan ti pursue this PR anymore for now