Re: [PR] feat(perf): BMI and per word for bit filtering [arrow-rs]

["devanbenz (via GitHub)" <gi...@apache.org>]

devanbenz commented on PR #10136:
URL: https://github.com/apache/arrow-rs/pull/10136#issuecomment-5798305000

   @sdf-jkl I borrowed some ideas from #11055 
   
   I'm seeing a single instance where #11055 is outperforming which is above the noise margin 
   
   Specifically `filter context short string view (kept 1/2)` BMI2 path. 
   
   If you have a chance could you check your benches locally again using the updated commits?  
   
   | Case | Build | main | #11055 | #10136 | #11055 vs main | #10136 vs main | #10136 vs #11055 |
   |---|---|---:|---:|---:|---:|---:|---:|
   | filter context i32 w NULLs (kept 1/2) | scalar | 47.54 µs | 36.99 µs | 28.52 µs | -22.2% | -40.0% | -22.9% |
   | | bmi2 | 53.28 µs | 10.75 µs | 10.63 µs | -79.8% | -80.1% | -1.2% |
   | filter context u8 w NULLs (kept 1/2) | scalar | 39.32 µs | 35.82 µs | 26.41 µs | -8.9% | -32.9% | -26.3% |
   | | bmi2 | 37.90 µs | 9.12 µs | 9.10 µs | -75.9% | -76.0% | -0.3% |
   | filter context string dictionary w NULLs (kept 1/2) | scalar | 36.39 µs | 36.15 µs | 28.40 µs | -0.7% | -21.9% | -21.4% |
   | | bmi2 | 44.88 µs | 10.85 µs | 10.69 µs | -75.8% | -76.2% | -1.5% |
   | filter f32 (kept 1/2) | scalar | 61.87 µs | 49.56 µs | 39.97 µs | -19.9% | -35.4% | -19.4% |
   | | bmi2 | 61.96 µs | 22.16 µs | 22.74 µs | -64.2% | -63.3% | +2.6% |
   | filter context f32 (kept 1/2) | scalar | 44.43 µs | 38.29 µs | 28.40 µs | -13.8% | -36.1% | -25.8% |
   | | bmi2 | 36.93 µs | 10.74 µs | 10.80 µs | -70.9% | -70.8% | +0.6% |
   | filter context short string view (kept 1/2) | scalar | 51.30 µs | 53.05 µs | 44.48 µs | +3.4% | -13.3% | -16.1% |
   | | bmi2 | 50.56 µs | 25.74 µs | 29.11 µs | -49.1% | -42.4% | +13.1% |
   | filter context mixed string view (kept 1/2) | scalar | 54.59 µs | 53.86 µs | 43.77 µs | -1.3% | -19.8% | -18.7% |
   | | bmi2 | 56.13 µs | 29.01 µs | 26.35 µs | -48.3% | -53.1% | -9.2% |
   | boolean 65 536 bits, kept 1/2 (filter context fsb …, 9 rows) | scalar | 28.74–39.63 µs | 27.08–28.72 µs | 19.15–19.39 µs | -29…-2% | -51…-33% | -33…-29% |
   | | bmi2 | 32.59–41.92 µs | 1.87–1.94 µs | 1.71–1.77 µs | -95…-94% | -96…-95% | -11…-6% |


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org