HoustonPutman commented on PR #12311: URL: https://github.com/apache/lucene/pull/12311#issuecomment-1561571022
Just ran the benchmarks on an m2 max machine, to see if it improved from the m1 performance (it didn't): java -jar target/vectorbench.jar -p size=1024 ``` Benchmark (size) Mode Cnt Score Error Units BinaryCosineBenchmark.cosineDistanceNew 1024 thrpt 5 2.442 ± 0.017 ops/us BinaryCosineBenchmark.cosineDistanceOld 1024 thrpt 5 1.016 ± 0.359 ops/us BinaryDotProductBenchmark.dotProductNew 1024 thrpt 5 6.495 ± 0.269 ops/us BinaryDotProductBenchmark.dotProductOld 1024 thrpt 5 3.150 ± 0.120 ops/us BinarySquareBenchmark.squareDistanceNew 1024 thrpt 5 6.544 ± 0.504 ops/us BinarySquareBenchmark.squareDistanceOld 1024 thrpt 5 3.102 ± 0.088 ops/us FloatCosineBenchmark.cosineNew 1024 thrpt 5 8.113 ± 0.168 ops/us FloatCosineBenchmark.cosineOld 1024 thrpt 5 1.147 ± 0.030 ops/us FloatDotProductBenchmark.dotProductNew 1024 thrpt 5 16.719 ± 0.287 ops/us FloatDotProductBenchmark.dotProductOld 1024 thrpt 5 3.814 ± 0.198 ops/us FloatSquareBenchmark.squareNew 1024 thrpt 5 14.941 ± 0.529 ops/us FloatSquareBenchmark.squareOld 1024 thrpt 5 3.321 ± 0.068 ops/us ``` java -jar target/vectorbench.jar -p size=1024 -t max ``` Benchmark (size) Mode Cnt Score Error Units BinaryCosineBenchmark.cosineDistanceNew 1024 thrpt 5 20.388 ± 3.046 ops/us BinaryCosineBenchmark.cosineDistanceOld 1024 thrpt 5 10.483 ± 1.104 ops/us BinaryDotProductBenchmark.dotProductNew 1024 thrpt 5 56.619 ± 4.635 ops/us BinaryDotProductBenchmark.dotProductOld 1024 thrpt 5 30.796 ± 0.797 ops/us BinarySquareBenchmark.squareDistanceNew 1024 thrpt 5 55.987 ± 3.258 ops/us BinarySquareBenchmark.squareDistanceOld 1024 thrpt 5 28.751 ± 2.661 ops/us FloatCosineBenchmark.cosineNew 1024 thrpt 5 65.090 ± 6.555 ops/us FloatCosineBenchmark.cosineOld 1024 thrpt 5 11.113 ± 1.230 ops/us FloatDotProductBenchmark.dotProductNew 1024 thrpt 5 135.249 ± 8.125 ops/us FloatDotProductBenchmark.dotProductOld 1024 thrpt 5 33.419 ± 1.708 ops/us FloatSquareBenchmark.squareNew 1024 thrpt 5 110.014 ± 7.691 ops/us FloatSquareBenchmark.squareOld 1024 thrpt 5 26.892 ± 15.677 ops/us ``` It'd be interesting to see how this/java plays with the Gravitron3 which is ARMv8 with 2x256 bit SVE. Thanks for the amazing work here y'all! -- 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: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
