RamakrishnaChilaka opened a new pull request, #15358:
URL: https://github.com/apache/lucene/pull/15358
### Description
This change optimizes `ForUtil` bulk integer encoding by introducing a
batched `writeInts()` method in the `DataOutput` interface, allowing integers
to be written in a single bulk operation.
It also precomputes masks to reduce branching overhead and improve encoding
efficiency.
### Benchmarks
```
baseline
Benchmark (bitsPerValue) Mode Cnt Score
Error Units
ForUtilEncodeBulkIntsBenchmark.encode 2 thrpt 5 15.331 ±
0.304 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 4 thrpt 5 12.394 ±
0.034 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 8 thrpt 5 8.274 ±
0.053 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 12 thrpt 5 4.321 ±
0.009 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 16 thrpt 5 4.409 ±
0.177 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 20 thrpt 5 2.329 ±
0.009 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 24 thrpt 5 2.187 ±
0.032 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 28 thrpt 5 1.959 ±
0.008 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 32 thrpt 5 2.356 ±
0.005 ops/us
candidate
Benchmark (bitsPerValue) Mode Cnt Score
Error Units
ForUtilEncodeBulkIntsBenchmark.encode 2 thrpt 5 19.555 ±
0.073 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 4 thrpt 5 18.772 ±
0.046 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 8 thrpt 5 15.424 ±
0.010 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 12 thrpt 5 7.651 ±
0.014 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 16 thrpt 5 9.577 ±
0.045 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 20 thrpt 5 3.980 ±
0.005 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 24 thrpt 5 3.821 ±
0.010 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 28 thrpt 5 4.001 ±
0.002 ops/us
ForUtilEncodeBulkIntsBenchmark.encode 32 thrpt 5 5.917 ±
0.009 ops/us
```
### Summary of benchmark results
Benchmark results show significant encode throughput improvements:
- +25–50% for low bitsPerValue (2–4 bits)
- ~2× speedup for 8–28 bits
- Up to 2.5× faster at 32 bits
Overall, encoding throughput improved by ~1.8× on average across all tested
bit widths.
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
