Continue optimizing branch miss of if function when result type is float*/decimal*/int*

[taiyang-li]

Changelog category (leave one):

• Performance Improvement

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Continue optimizing branch miss of if function when result type is float*/decimal*/int* , follow up of #57885

The POCs helps me prove my ideas: https://github.com/bigo-sg/gluten/commits/improve_if/ (the last two commits)

Q1: with rand32() % 2 as x select if(x, materialize(1.234), materialize(2.456)) from numbers(1000000000) format Null;
Old: 0 rows in set. Elapsed: 8.865 sec. Processed 1.00 billion rows, 8.00 GB (112.81 million rows/s., 902.47 MB/s.)
New: 0 rows in set. Elapsed: 3.324 sec. Processed 1.00 billion rows, 8.00 GB (300.82 million rows/s., 2.41 GB/s.)

Q2: with rand32() % 2 as x, 1.234::Decimal64(3) as a, 2.456::Decimal64(3) as b select if(x, materialize(a), materialize(b)) from numbers(1000000000) format Null;
Old: 0 rows in set. Elapsed: 5.892 sec. Processed 1.00 billion rows, 8.00 GB (169.74 million rows/s., 1.36 GB/s.)
New: 0 rows in set. Elapsed: 5.237 sec. Processed 1.00 billion rows, 8.00 GB (190.93 million rows/s., 1.53 GB/s.)

with rand32() % 2 as x, 1::Int8 as a, -1::Int8 as b select if(x, a, b) from numbers(1000000000) format Null; 
0 rows in set. Elapsed: 8.973 sec. Processed 1.00 billion rows, 8.00 GB (111.44 million rows/s., 891.55 MB/s.)
0 rows in set. Elapsed: 2.419 sec. Processed 1.00 billion rows, 8.00 GB (413.38 million rows/s., 3.31 GB/s.)

with rand32() % 2 as x, 1::Int64 as a, -1::Int64 as b select if(x, a, b) from numbers(1000000000) format Null; 
0 rows in set. Elapsed: 2.268 sec. Processed 1.00 billion rows, 8.00 GB (440.90 million rows/s., 3.53 GB/s.)
0 rows in set. Elapsed: 1.785 sec. Processed 1.00 billion rows, 8.00 GB (560.32 million rows/s., 4.48 GB/s.) 

with rand32() % 2 as x, 1::Decimal64(3) as a, -1::Decimal64(3) as b select if(x, a, b) from numbers(1000000000) format Null; 
0 rows in set. Elapsed: 2.286 sec. Processed 1.00 billion rows, 8.00 GB (437.45 million rows/s., 3.50 GB/s.)
0 rows in set. Elapsed: 1.831 sec. Processed 1.00 billion rows, 8.00 GB (546.11 million rows/s., 4.37 GB/s.)

with rand32() % 2 as x, 1::Decimal128(3) as a, -1::Decimal128(3) as b select if(x, a, b) from numbers(1000000000) format Null; 
0 rows in set. Elapsed: 3.315 sec. Processed 1.00 billion rows, 8.00 GB (301.69 million rows/s., 2.41 GB/s.)
0 rows in set. Elapsed: 2.774 sec. Processed 1.00 billion rows, 8.00 GB (360.47 million rows/s., 2.88 GB/s.)

with rand32() % 2 as x, 1::Decimal256(3) as a, -1::Decimal256(3) as b select if(x, a, b) from numbers(100000000) format Null; 
0 rows in set. Elapsed: 1.004 sec. Processed 100.00 million rows, 800.00 MB (99.60 million rows/s., 796.83 MB/s.)
0 rows in set. Elapsed: 0.353 sec. Processed 100.00 million rows, 800.00 MB (283.16 million rows/s., 2.27 GB/s.)

with rand32() % 2 as x, 1::Int128 as a, -1::Int128 as b select if(x, a, b) from numbers(1000000000) format Null; 
0 rows in set. Elapsed: 3.330 sec. Processed 1.00 billion rows, 8.00 GB (300.33 million rows/s., 2.40 GB/s.)
0 rows in set. Elapsed: 2.755 sec. Processed 1.00 billion rows, 8.00 GB (363.00 million rows/s., 2.90 GB/s.)

with rand32() % 2 as x, 1::Int256 as a, -1::Int256 as b select if(x, a, b) from numbers(1000000000) format Null; 
0 rows in set. Elapsed: 9.825 sec. Processed 1.00 billion rows, 8.00 GB (101.78 million rows/s., 814.27 MB/s.)
0 rows in set. Elapsed: 3.406 sec. Processed 1.00 billion rows, 8.00 GB (293.59 million rows/s., 2.35 GB/s.)

[robot-ch-test-poll4]

This is an automated comment for commit 818fb98 with description of existing statuses. It's updated for the latest CI running

❌ Click here to open a full report in a separate page

Successful checks

Check name	Description	Status
AST fuzzer	Runs randomly generated queries to catch program errors. The build type is optionally given in parenthesis. If it fails, ask a maintainer for help	✅ success
ClickBench	Runs [ClickBench](https://github.com/ClickHouse/ClickBench/) with instant-attach table	✅ success
ClickHouse build check	Builds ClickHouse in various configurations for use in further steps. You have to fix the builds that fail. Build logs often has enough information to fix the error, but you might have to reproduce the failure locally. The cmake options can be found in the build log, grepping for cmake. Use these options and follow the general build process	✅ success
Compatibility check	Checks that clickhouse binary runs on distributions with old libc versions. If it fails, ask a maintainer for help	✅ success
Docker server and keeper images	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ success
Docs check	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ success
Fast tests	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ success
Flaky tests	Checks if new added or modified tests are flaky by running them repeatedly, in parallel, with more randomization. Functional tests are run 100 times with address sanitizer, and additional randomization of thread scheduling. Integrational tests are run up to 10 times. If at least once a new test has failed, or was too long, this check will be red. We don't allow flaky tests, read the doc	✅ success
Install packages	Checks that the built packages are installable in a clear environment	✅ success
Integration tests	The integration tests report. In parenthesis the package type is given, and in square brackets are the optional part/total tests	✅ success
Mergeable Check	Checks if all other necessary checks are successful	✅ success
Performance Comparison	Measure changes in query performance. The performance test report is described in detail here. In square brackets are the optional part/total tests	✅ success
SQLancer	Fuzzing tests that detect logical bugs with SQLancer tool	✅ success
Sqllogic	Run clickhouse on the sqllogic test set against sqlite and checks that all statements are passed	✅ success
Stateful tests	Runs stateful functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	✅ success
Stateless tests	Runs stateless functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	✅ success
Style check	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ success
Unit tests	Runs the unit tests for different release types	✅ success
Upgrade check	Runs stress tests on server version from last release and then tries to upgrade it to the version from the PR. It checks if the new server can successfully startup without any errors, crashes or sanitizer asserts	✅ success

Check name	Description	Status
CI running	A meta-check that indicates the running CI. Normally, it's in success or pending state. The failed status indicates some problems with the PR	⏳ pending
Stress test	Runs stateless functional tests concurrently from several clients to detect concurrency-related errors	❌ failure

[taiyang-li]

Performance report on x86-64

Performance report on aarch64

Some bad cases

• if: SELECT count() FROM zeros(1000000000) WHERE NOT ignore(if(rand32() < 42949673, 1, 2)). Slow down.
• if_to_multiif: WITH number AS x SELECT sum(x < 1 ? 1 : (x < 5 ? 2 : 3)) FROM numbers(100000000). Not permanent.
• if_to_multiif: WITH number AS x SELECT sum(x < 1 ? 1 : (x < 5 ? 2 : (x < 10 ? 3 : (x % 2 ? 4 : 5)))) FROM numbers(100000000). Not permanent.

[taiyang-li]

Bad cases are solved

[Algunenano]

Fast test failures are unrelated (unpinned dependency update). Please rebase with master

[Algunenano]

Interestingly, we have a similar approach in AggregateFunctionSum:

ClickHouse/src/AggregateFunctions/AggregateFunctionSum.h

Lines 171 to 187 in 76abd32

	static_assert(sizeof(Value) == 4 || sizeof(Value) == 8);
	using equivalent_integer = typename std::conditional_t<sizeof(Value) == 4, UInt32, UInt64>;
	constexpr size_t unroll_count = 128 / sizeof(T);
	T partial_sums[unroll_count]{};
	const auto * unrolled_end = ptr + (count / unroll_count * unroll_count);
	while (ptr < unrolled_end)
	{
	for (size_t i = 0; i < unroll_count; ++i)
	{
	equivalent_integer value;
	std::memcpy(&value, &ptr[i], sizeof(Value));
	value &= (!condition_map[i] != add_if_zero) - 1;
	Value d;
	std::memcpy(&d, &value, sizeof(Value));

[Avogar]

@Algunenano I know you like such oprimizations, so, please be a reviewer instead of me :)

[taiyang-li]

https://s3.amazonaws.com/clickhouse-test-reports/59148/49fc8a7099041cb0e9ce7c76fe32c11997868ca8/upgrade_check__debug_.html seems not related to this pr

[Algunenano]

Looks really nice to me. Thanks a lot, specially for adding the extra parenthesis in for+if situations (helps readability IMO). I've left a few comments. Can you have a look, please?

[taiyang-li]

BTW, this is another opt of if function when input type is map: #59413 @Algunenano