Do not add nulls to the hash table when null_equality::NOT_EQUAL is passed to left_semi_join and left_anti_join #8277

nvdbaranec · 2021-05-18T22:25:46Z

This is fundamentally the same issue and fix as https://github.com/rapidsai/cudf/pull/6943/files from @hyperbolic2346

When nulls are considered not equal (null_equality::NOT_EQUAL) there is no point in adding them to the hash table used for the join as they will never compare as true against anything. Adding large numbers of nulls was causing huge performance issues.

Includes a fix to doxygen comments for left_anti_join

Performance gain is tremendous.

Before:

Benchmark                                                                             Time             CPU   Iterations
-----------------------------------------------------------------------------------------------------------------------
Join<int32_t, int32_t>/left_anti_join_32bit_nulls/100000/100000/manual_time        1072 ms         1072 ms            1
Join<int32_t, int32_t>/left_anti_join_32bit_nulls/200000/400000/manual_time        4253 ms         4253 ms            1
Join<int32_t, int32_t>/left_anti_join_32bit_nulls/300000/1000000/manual_time      14016 ms        14016 ms            1
Join<int32_t, int32_t>/left_semi_join_32bit_nulls/100000/100000/manual_time         932 ms          932 ms            1
Join<int32_t, int32_t>/left_semi_join_32bit_nulls/200000/400000/manual_time        4481 ms         4481 ms            1
Join<int32_t, int32_t>/left_semi_join_32bit_nulls/300000/1000000/manual_time      14172 ms        14172 ms            1

After:

-----------------------------------------------------------------------------------------------------------------------
Benchmark                                                                             Time             CPU   Iterations
-----------------------------------------------------------------------------------------------------------------------
Join<int32_t, int32_t>/left_anti_join_32bit_nulls/100000/100000/manual_time       0.143 ms        0.162 ms         4996
Join<int32_t, int32_t>/left_anti_join_32bit_nulls/200000/400000/manual_time       0.255 ms        0.275 ms         2780
Join<int32_t, int32_t>/left_anti_join_32bit_nulls/300000/1000000/manual_time      0.514 ms        0.532 ms         1368
Join<int32_t, int32_t>/left_semi_join_32bit_nulls/100000/100000/manual_time       0.135 ms        0.155 ms         5203
Join<int32_t, int32_t>/left_semi_join_32bit_nulls/200000/400000/manual_time       0.206 ms        0.224 ms         3325
Join<int32_t, int32_t>/left_semi_join_32bit_nulls/300000/1000000/manual_time      0.368 ms        0.385 ms         1903

…assed to left_anti_join and left_semi_join. Major performance gains. Added benchmarks.

nvdbaranec · 2021-05-18T22:27:16Z

I noticed as I was doing this that we have no tests for left_anti_join. Will add.

…owered.

cpp/src/join/semi_join.cu

harrism

Nice speedup!

cpp/src/join/semi_join.cu

cpp/benchmarks/join/join_benchmark.cu

vuule · 2021-05-18T23:23:41Z

cpp/src/join/semi_join.cu

-                     [hash_table] __device__(size_type idx) mutable {
-                       hash_table.insert(thrust::make_pair(idx, true));
-                     });
+  // if compare_nulls == NOT_EQUAL, we can simply ignore any rows that are


Could we also adjust the hash table size based on the number of valid rows? Not sure how important this is for join operations, just a thought.

Something like this seems to work. I have to do a count_unset_bits() on the aggregate bitmask though. This also feels mildly spooky to change as I'm trying to treat lightly in this unfamiliar module :)

auto const row_bitmask = (compare_nulls == null_equality::EQUAL) ? rmm::device_buffer{} : cudf::detail::bitmask_and(right_flattened_keys, stream); size_type right_row_ignore_count = compare_nulls == null_equality::EQUAL ? 0 : count_unset_bits(static_cast<const bitmask_type*>(row_bitmask.data()), 0, right_num_rows); ... size_t const hash_table_size = compute_hash_table_size(right_num_rows - right_row_ignore_count);

cpp/src/join/semi_join.cu

…ith both nulls EQUAL and UNEQUAL.

codecov · 2021-05-20T19:58:25Z

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.06@8406522). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-21.06    #8277   +/-   ##
===============================================
  Coverage                ?   82.86%           
===============================================
  Files                   ?      105           
  Lines                   ?    17861           
  Branches                ?        0           
===============================================
  Hits                    ?    14801           
  Misses                  ?     3060           
  Partials                ?        0

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8406522...0c73774. Read the comment docs.

harrism · 2021-05-24T20:25:48Z

@gpucibot merge

Do not add nulls to the hash table when null_equality::NOT_EQUAL is p…

9fae839

…assed to left_anti_join and left_semi_join. Major performance gains. Added benchmarks.

nvdbaranec added libcudf Affects libcudf (C++/CUDA) code. Spark Functionality that helps Spark RAPIDS improvement Improvement / enhancement to an existing function non-breaking Non-breaking change labels May 18, 2021

nvdbaranec requested a review from a team as a code owner May 18, 2021 22:25

nvdbaranec requested review from robertmaynard and vuule May 18, 2021 22:25

nvdbaranec added 3 commits May 18, 2021 17:32

Restore a couple of benchmark parameters that had been artificially l…

a4e3f30

…owered.

More benchmark number tweaks.

97e1a47

Newline in benchmarks

99827d1

hyperbolic2346 reviewed May 18, 2021

View reviewed changes

cpp/src/join/semi_join.cu Outdated Show resolved Hide resolved

harrism requested changes May 18, 2021

View reviewed changes

cpp/src/join/semi_join.cu Outdated Show resolved Hide resolved

cpp/src/join/semi_join.cu Show resolved Hide resolved

cpp/benchmarks/join/join_benchmark.cu Outdated Show resolved Hide resolved

vuule reviewed May 18, 2021

View reviewed changes

robertmaynard reviewed May 19, 2021

View reviewed changes

cpp/src/join/semi_join.cu Show resolved Hide resolved

nvdbaranec added 2 commits May 20, 2021 10:56

Add missing join tests: Semi-join with nulls UNEQUAL, and anti-join w…

8226b0c

…ith both nulls EQUAL and UNEQUAL.

Small PR review tweaks.

5528977

nvdbaranec requested review from hyperbolic2346 and harrism May 20, 2021 16:29

github-actions bot added the CMake CMake build issue label May 20, 2021

robertmaynard self-requested a review May 20, 2021 16:50

Comment grammar fix.

0c73774

jrhemstad approved these changes May 24, 2021

View reviewed changes

hyperbolic2346 approved these changes May 24, 2021

View reviewed changes

robertmaynard approved these changes May 24, 2021

View reviewed changes

harrism approved these changes May 24, 2021

View reviewed changes

rapids-bot bot merged commit 7e725b5 into rapidsai:branch-21.06 May 24, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Do not add nulls to the hash table when null_equality::NOT_EQUAL is passed to left_semi_join and left_anti_join #8277

Do not add nulls to the hash table when null_equality::NOT_EQUAL is passed to left_semi_join and left_anti_join #8277

nvdbaranec commented May 18, 2021 •

edited

Loading

nvdbaranec commented May 18, 2021

harrism left a comment

vuule May 18, 2021

nvdbaranec May 20, 2021

codecov bot commented May 20, 2021 •

edited

Loading

harrism commented May 24, 2021

Do not add nulls to the hash table when null_equality::NOT_EQUAL is passed to left_semi_join and left_anti_join #8277

Do not add nulls to the hash table when null_equality::NOT_EQUAL is passed to left_semi_join and left_anti_join #8277

Conversation

nvdbaranec commented May 18, 2021 • edited Loading

nvdbaranec commented May 18, 2021

harrism left a comment

Choose a reason for hiding this comment

vuule May 18, 2021

Choose a reason for hiding this comment

nvdbaranec May 20, 2021

Choose a reason for hiding this comment

codecov bot commented May 20, 2021 • edited Loading

Codecov Report

harrism commented May 24, 2021

nvdbaranec commented May 18, 2021 •

edited

Loading

codecov bot commented May 20, 2021 •

edited

Loading