iceberg-kitten opened a new pull request, #1638:
URL: https://github.com/apache/arrow-go/pull/1638

   ### Why
   Casts between binary arrays and views size the output, then repeat capacity 
and validity checks on each builder append.
   
   ### What
   Write the cast result directly into Arrow buffers.
   
   ### Implementation
   Size the headers, offsets, and payload up front and copy the logical 
validity bitmap. Keep the existing UTF-8 checks and the single-buffer limit for 
view payloads.
   
   ### Benchmark
   
   Cast 65,536 values per operation. Values are 4 to 256 bytes, with either no 
nulls or a null every tenth row.
   
   | Input | Time/op before → after | Change | B/op | Allocs/op |
   | --- | ---: | ---: | ---: | ---: |
   | BinaryView → Binary, 4 B, 0% null | 2.08 → 1.11 ms | -46.5% | 827,592 → 
544,843 | 78 → 54 |
   | BinaryView → Binary, 4 B, 10% null | 2.19 → 1.44 ms | -34.3% | 791,448 → 
519,858 | 52 → 43 |
   | BinaryView → Binary, 32 B, 0% null | 2.51 → 1.55 ms | -38.1% | 2,659,243 → 
2,378,089 | 52 → 40 |
   | BinaryView → Binary, 32 B, 10% null | 2.51 → 1.73 ms | -31.1% | 2,446,251 
→ 2,174,664 | 52 → 43 |
   | BinaryView → Binary, 256 B, 0% null | 5.3 → 4.09 ms | -22.7% | 17,339,311 
→ 17,058,171 | 52 → 40 |
   | BinaryView → Binary, 256 B, 10% null | 4.71 → 3.68 ms | -21.9% | 
15,659,950 → 15,388,378 | 52 → 44 |
   | String → StringView, 8 B, 0% null | 2.29 → 1.5 ms | -34.3% | 2,126,486 → 
1,059,062 | 46 → 38 |
   | String → StringView, 8 B, 10% null | 2.31 → 1.72 ms | -25.5% | 2,126,485 → 
1,068,630 | 46 → 41 |
   | String → StringView, 32 B, 0% null | 3.75 → 2.22 ms | -40.7% | 4,232,004 → 
3,164,564 | 52 → 42 |
   | String → StringView, 32 B, 10% null | 3.65 → 2.3 ms | -37.1% | 4,019,012 → 
2,961,132 | 52 → 45 |
   | String → StringView, 256 B, 0% null | 5.07 → 4.33 ms | -14.6% | 18,912,069 
→ 17,844,645 | 52 → 43 |
   | String → StringView, 256 B, 10% null | 5.24 → 4.29 ms | -18.2% | 
17,232,711 → 16,174,830 | 52 → 45 |
   
   Medians of 8 runs per revision on Linux amd64, Intel Core i9-12900HK, Go 
1.26.11, pinned to one efficiency core. Baseline is `main` at `aba35416`. Use 
the same benchmark on both revisions. Copy the PR's benchmark file into the 
baseline checkout before running.
   
   ```sh
   go test ./arrow/compute -run '^$' -bench 
'^BenchmarkBinary(ToViewDirectBuffers|ViewToBinaryMaterialization)$/^((value-len|bytes)=(4|8|32|256))$/^null-every=(0|10)$'
 -benchmem -benchtime=120ms -count=8 -cpu=1
   ```
   


-- 
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: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to