Commits
Sanjiban Sengupta authored and Antoine Pitrou committed c6d844755ac
ARROW-16014: [C++] Create more benchmarks for measuring expression evaluation overhead This PR adds an Expression Execution benchmark in `arrow-compute` which uses the `rows_per_batch` as a parameter and measures the `rows_per_second` for executing the scalar expression. ``` Running ./release/arrow-compute-expression-benchmark Run on (10 X 24.1211 MHz CPU s) CPU Caches: L1 Data 64 KiB (x10) L1 Instruction 128 KiB (x10) L2 Unified 4096 KiB (x5) ----------------------------------------------------------------------------------------------------------------------------------------------------------- Benchmark Time CPU Iterations UserCounters... ----------------------------------------------------------------------------------------------------------------------------------------------------------- ExecuteScalarExpressionOverhead/complex_expression/rows_per_batch:1000/real_time/threads:1 2996512 ns 2995935 ns 232 batches_per_second=333.721k/s rows_per_second=333.721M/s ExecuteScalarExpressionOverhead/complex_expression/rows_per_batch:1000/real_time/threads:10 2808554 ns 26825320 ns 250 batches_per_second=356.055k/s rows_per_second=356.055M/s ExecuteScalarExpressionOverhead/complex_expression/rows_per_batch:10000/real_time/threads:1 745300 ns 745197 ns 914 batches_per_second=134.174k/s rows_per_second=1.34174G/s ExecuteScalarExpressionOverhead/complex_expression/rows_per_batch:10000/real_time/threads:10 274030 ns 2621425 ns 2570 batches_per_second=364.923k/s rows_per_second=3.64923G/s ExecuteScalarExpressionOverhead/complex_expression/rows_per_batch:100000/real_time/threads:1 509337 ns 509270 ns 1371 batches_per_second=19.6334k/s rows_per_second=1.96334G/s ExecuteScalarExpressionOverhead/complex_expression/rows_per_batch:100000/real_time/threads:10 67678 ns 645726 ns 10530 batches_per_second=147.758k/s rows_per_second=14.7758G/s ExecuteScalarExpressionOverhead/complex_expression/rows_per_batch:1000000/real_time/threads:1 498521 ns 498480 ns 1397 batches_per_second=2.00593k/s rows_per_second=2.00593G/s ExecuteScalarExpressionOverhead/complex_expression/rows_per_batch:1000000/real_time/threads:10 131368 ns 1231142 ns 5300 batches_per_second=7.61221k/s rows_per_second=7.61221G/s ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:1000/real_time/threads:1 1099162 ns 1098507 ns 635 batches_per_second=909.784k/s rows_per_second=909.784M/s ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:1000/real_time/threads:10 992847 ns 9585644 ns 700 batches_per_second=1007.2k/s rows_per_second=1007.2M/s ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:10000/real_time/threads:1 465254 ns 465121 ns 1447 batches_per_second=214.936k/s rows_per_second=2.14936G/s ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:10000/real_time/threads:10 100351 ns 891244 ns 7060 batches_per_second=996.503k/s rows_per_second=9.96503G/s ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:100000/real_time/threads:1 393506 ns 393395 ns 1784 batches_per_second=25.4126k/s rows_per_second=2.54126G/s ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:100000/real_time/threads:10 57732 ns 545456 ns 12070 batches_per_second=173.215k/s rows_per_second=17.3215G/s ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:1000000/real_time/threads:1 392542 ns 392370 ns 1797 batches_per_second=2.5475k/s rows_per_second=2.5475G/s ExecuteScalarExpressionOverhead/simple_expression/rows_per_batch:1000000/real_time/threads:10 103905 ns 945614 ns 6740 batches_per_second=9.6242k/s rows_per_second=9.6242G/s ExecuteScalarExpressionOverhead/zero_copy_expression/rows_per_batch:1000/real_time/threads:1 556220 ns 556173 ns 1265 batches_per_second=1.79785M/s rows_per_second=1.79785G/s ExecuteScalarExpressionOverhead/zero_copy_expression/rows_per_batch:1000/real_time/threads:10 600231 ns 5753634 ns 1210 batches_per_second=1.66603M/s rows_per_second=1.66603G/s ExecuteScalarExpressionOverhead/zero_copy_expression/rows_per_batch:10000/real_time/threads:1 55670 ns 55645 ns 12314 batches_per_second=1.79629M/s rows_per_second=17.9629G/s ExecuteScalarExpressionOverhead/zero_copy_expression/rows_per_batch:10000/real_time/threads:10 59506 ns 563823 ns 12210 batches_per_second=1.68052M/s rows_per_second=16.8052G/s ExecuteScalarExpressionOverhead/zero_copy_expression/rows_per_batch:100000/real_time/threads:1 5644 ns 5643 ns 123957 batches_per_second=1.77176M/s rows_per_second=177.176G/s ExecuteScalarExpressionOverhead/zero_copy_expression/rows_per_batch:100000/real_time/threads:10 5932 ns 57219 ns 118590 batches_per_second=1.68579M/s rows_per_second=168.579G/s ExecuteScalarExpressionOverhead/zero_copy_expression/rows_per_batch:1000000/real_time/threads:1 558 ns 557 ns 1259506 batches_per_second=1.79198M/s rows_per_second=1.79198T/s ExecuteScalarExpressionOverhead/zero_copy_expression/rows_per_batch:1000000/real_time/threads:10 594 ns 5712 ns 1223120 batches_per_second=1.68243M/s rows_per_second=1.68243T/s ExecuteScalarExpressionOverhead/ref_only_expression/rows_per_batch:1000/real_time/threads:1 20897 ns 20877 ns 33585 batches_per_second=47.8538M/s rows_per_second=47.8538G/s ExecuteScalarExpressionOverhead/ref_only_expression/rows_per_batch:1000/real_time/threads:10 146133 ns 1390005 ns 5150 batches_per_second=6.84306M/s rows_per_second=6.84306G/s ExecuteScalarExpressionOverhead/ref_only_expression/rows_per_batch:10000/real_time/threads:1 2084 ns 2084 ns 335070 batches_per_second=47.9808M/s rows_per_second=479.808G/s ExecuteScalarExpressionOverhead/ref_only_expression/rows_per_batch:10000/real_time/threads:10 14656 ns 139612 ns 48420 batches_per_second=6.82311M/s rows_per_second=68.2311G/s ExecuteScalarExpressionOverhead/ref_only_expression/rows_per_batch:100000/real_time/threads:1 209 ns 209 ns 3342442 batches_per_second=47.755M/s rows_per_second=4.7755T/s ExecuteScalarExpressionOverhead/ref_only_expression/rows_per_batch:100000/real_time/threads:10 1404 ns 13089 ns 481520 batches_per_second=7.12084M/s rows_per_second=712.084G/s ExecuteScalarExpressionOverhead/ref_only_expression/rows_per_batch:1000000/real_time/threads:1 20.9 ns 20.9 ns 33575027 batches_per_second=47.8822M/s rows_per_second=47.8822T/s ExecuteScalarExpressionOverhead/ref_only_expression/rows_per_batch:1000000/real_time/threads:10 145 ns 1371 ns 5094890 batches_per_second=6.89397M/s rows_per_second=6.89397T/s ``` Closes #12755 from sanjibansg/scheduler/ScalarExecution Authored-by: Sanjiban Sengupta <sanjiban.sg@gmail.com> Signed-off-by: Antoine Pitrou <antoine@python.org>