vsop-479 commented on issue #12788:
URL: https://github.com/apache/lucene/issues/12788#issuecomment-1809865124
> To benchmark then use the benchmark-jmh Gradle module
I measured max with scalar, unroll, vector implementation by benchmark-jmh:
Benchmark (size) Mode Cnt Score Error Units
VectorUtilBenchmark.maxScalar 256 thrpt 15 51.056 ± 0.310 ops/us
VectorUtilBenchmark.maxUnroll 256 thrpt 15 12.799 ± 0.036 ops/us
VectorUtilBenchmark.maxVector 256 thrpt 15 53.381 ± 3.112 ops/us
Assembly code indicates that:
Scalar get auto-vectorized, and auto-unrolled. @jpountz
Unroll doesn't get auto-vectorized.
Vector get auto-unrolled.
# Assembly code sample
````
// for scalar and vector.
smax v16.4s, v16.4s, v17.4s
add w10, w2, #0x10 // 4 times smax, 1 add.
// for unroll.
cmp w12, w11
csel w11, w12, w11, gt // cmp, csel.
````
--
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]
