wesm commented on PR #13364: URL: https://github.com/apache/arrow/pull/13364#issuecomment-1156654534
Here is the raw benchmark results (contender is this patch merged to master versus the commit before it) on my Intel i9 (has avx512) with cpu frequency scaling disabled, seems to have yielded some speedups but also some regressions: https://gist.githubusercontent.com/wesm/e103586caa0b4144350b2a754a7c96a4/raw/374383d9e1af5186e072a327277d40d427bd5430/benchmark_diff_execspan.txt Some of the regressions looks spurious, but some others we ought to profile to better understand, for example `Unique` seems to have been affected but I don't know why ``` UniqueUInt8/1 1.732 GiB/sec 1.255 GiB/sec -27.544 ArrayArrayKernel<KleeneAnd>/32768/0 11.853 GiB/sec 8.581 GiB/sec -27.600 BitmapEqualsWithOffset/8192 6.547 GiB/sec 4.735 GiB/sec -27.671 ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:100000/real_time/threads:36 227450.971042 164478.721836 -27.686 UniqueInt64/5 11.135 GiB/sec 7.692 GiB/sec -30.919 TakeInt64RandomIndicesWithNulls/1048576/1 2.787G items/sec 1.917G items/sec -31.231 ArrayScalarKernel<SubtractChecked, Int32Type>/1048576/0 8.185 GiB/sec 5.553 GiB/sec -32.153 ArrayScalarKernel<SubtractChecked, UInt32Type>/1048576/0 8.199 GiB/sec 5.550 GiB/sec -32.312 ArrayArrayKernel<KleeneAnd>/1048576/0 6.401 GiB/sec 4.306 GiB/sec -32.731 UniqueUInt8/0 2.121 GiB/sec 1.421 GiB/sec -33.013 ArrayScalarKernel<AddChecked, UInt16Type>/1048576/0 4.203 GiB/sec 2.813 GiB/sec -33.064 ArrayScalarKernel<SubtractChecked, UInt16Type>/1048576/0 4.204 GiB/sec 2.811 GiB/sec -33.131 ArrayRangeEqualsFloat32/32768/100 1.449G items/sec 958.666M items/sec -33.837 BitmapWriter/8192 164.977 MiB/sec 100.866 MiB/sec -38.861 ArrayRangeEqualsFloat32/32768/10000 1.905G items/sec 1.088G items/sec -42.904 BM_PlainDecodingSpacedFloat/32768/10000 82.335 GiB/sec 46.504 GiB/sec -43.518 BenchmarkTemporalBinary<MonthDayNanoBetween, time32_type>/1048576/100 118.598 MiB/sec 44.037 MiB/sec -62.868 UniqueUInt8/6 24.516 GiB/sec 1.994 GiB/sec -91.865 UniqueInt64/13 195.807 GiB/sec 15.848 GiB/sec -91.906 UniqueInt64/6 196.137 GiB/sec 15.835 GiB/sec -91.927 ``` -- 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]
