mirror of
https://github.com/hanzoai/datastore.git
synced 2026-09-26 13:55:07 +00:00
Raúl Marín
064130bb36
Most performance tests already disable JIT via `compile_expressions=0`, `compile_aggregate_expressions=0` and `compile_sort_description=0` in `perf-comparison-tweaks-users.xml`. Tests that explicitly opt back into JIT were still bimodal: the JIT cache is global and shared across queries, so whether the first measured runs hit it or not depends on what ran before. Set `min_count_to_compile_expression`, `min_count_to_compile_aggregate_expression` and `min_count_to_compile_sort_description` to 0 in the same profile so that when a test does enable JIT, compilation happens on the very first call and the existing prewarm run populates the cache before measurements start. Drop now-redundant per-test overrides of the same settings.
33 lines
1.1 KiB
XML
33 lines
1.1 KiB
XML
<test>
|
|
<query>
|
|
WITH
|
|
bitXor(number, 0x4CF2D2BAAE6DA887) AS x0,
|
|
bitXor(x0, bitShiftRight(x0, 33)) AS x1,
|
|
x1 * 0xff51afd7ed558ccd AS x2,
|
|
bitXor(x2, bitShiftRight(x2, 33)) AS x3,
|
|
x3 * 0xc4ceb9fe1a85ec53 AS x4,
|
|
bitXor(x4, bitShiftRight(x4, 33)) AS x5
|
|
SELECT count() FROM numbers(100000000) WHERE NOT ignore(x5)
|
|
SETTINGS
|
|
compile_expressions = 0
|
|
</query>
|
|
|
|
<query>
|
|
WITH
|
|
bitXor(number, 0x4CF2D2BAAE6DA887) AS x0,
|
|
bitXor(x0, bitShiftRight(x0, 33)) AS x1,
|
|
x1 * 0xff51afd7ed558ccd AS x2,
|
|
bitXor(x2, bitShiftRight(x2, 33)) AS x3,
|
|
x3 * 0xc4ceb9fe1a85ec53 AS x4,
|
|
bitXor(x4, bitShiftRight(x4, 33)) AS x5
|
|
SELECT count() FROM numbers(100000000) WHERE NOT ignore(x5)
|
|
SETTINGS
|
|
compile_expressions = 1
|
|
</query>
|
|
|
|
<!-- The same expression written as ClickHouse builtin function. -->
|
|
<query>
|
|
SELECT count() FROM numbers(100000000) WHERE NOT ignore(intHash64(number))
|
|
</query>
|
|
</test>
|