GH-43926: [C++] Compute: RowEncoder eliminates offsets when all columns are fixed-sized

[mapleFU]

Rationale for this change

Not use offsets_ vector when all columns in RowEncoder is fixed-width.

This:

1. Reducing the memory usage
2. Reducing calling to AddLength

This might enlarge 8-byte memory in RowEncoder.

What changes are included in this PR?

Not use offsets_ vector when all columns in RowEncoder is fixed-width.

Are these changes tested?

Covered by existing

Are there any user-facing changes?

no

• GitHub Issue: [C++] Compute: RowEncoder eliminates offsets when all columns are fixed-sized #43926

[github-actions]

⚠️ GitHub issue #43926 has been automatically assigned in GitHub to PR creator.

[pitrou]

What does performance look like?

[mapleFU]

CI need fix

D:\a\arrow\arrow\build\cpp\src\arrow\CMakeFiles\arrow_compute_shared.dir\Unity\unity_4_cxx.cxx
D:/a/arrow/arrow/cpp/src/arrow/compute/row/row_encoder_internal.cc(359): error C2220: the following warning is treated as an error
D:/a/arrow/arrow/cpp/src/arrow/compute/row/row_encoder_internal.cc(359): warning C4244: '+=': conversion from 'const int64_t' to 'int32_t', possible loss of data

[zanmato1984]

Some preliminary comments, mostly nits.

[zanmato1984]

What's this line for?

[zanmato1984]

I would suggest using a more descriptive name - XxInfo generally doesn't give much information about Xx. Maybe something like EncodeColumnMeta?

[mapleFU]

I would go out for tour from 9.14 to 9.21. I'll focus on List Join these days, mark this as wip first

[mapleFU]

With:

struct BenchmarkSettings {
  int num_threads = 1;
  JoinType join_type = JoinType::INNER;
  // Change to 'true' to benchmark alternative, non-default and less optimized version of
  // a hash join node implementation.
  bool use_basic_implementation = true;

After:

BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:1        166477 ns       166453 ns         4154 rows/sec=6.15188M/s
BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:8       1439597 ns      1439500 ns          487 rows/sec=5.69086M/s
BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:64     13310182 ns     13310009 ns           54 rows/sec=4.92381M/s
BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:512   354529054 ns    354520400 ns            2 rows/sec=1.47887M/s
BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:4096 6766162518 ns   6763357800 ns            1 rows/sec=620.151k/s
BM_HashJoinBasic_KeyTypes/"{utf8}"/Hashtable krows:1         191892 ns       191763 ns         3584 rows/sec=5.33993M/s
BM_HashJoinBasic_KeyTypes/"{utf8}"/Hashtable krows:4         791714 ns       791462 ns          875 rows/sec=5.17523M/s
BM_HashJoinBasic_KeyTypes/"{utf8}"/Hashtable krows:16       3627607 ns      3627186 ns          196 rows/sec=4.517M/s
BM_HashJoinBasic_KeyTypes/"{utf8}"/Hashtable krows:64      22743811 ns     22741736 ns           33 rows/sec=2.88175M/s

Before:

BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:1        162843 ns       162781 ns         4207 rows/sec=6.29067M/s
BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:8       1421683 ns      1421186 ns          493 rows/sec=5.7642M/s
BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:64     13439032 ns     13438304 ns           52 rows/sec=4.87681M/s
BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:512   408234871 ns    408206400 ns            2 rows/sec=1.28437M/s
BM_HashJoinBasic_KeyTypes/"{int32}"/Hashtable krows:4096 6854617796 ns   6852387200 ns            1 rows/sec=612.094k/s
BM_HashJoinBasic_KeyTypes/"{utf8}"/Hashtable krows:1         188940 ns       188805 ns         3716 rows/sec=5.42359M/s
BM_HashJoinBasic_KeyTypes/"{utf8}"/Hashtable krows:4         759144 ns       758925 ns          918 rows/sec=5.39711M/s
BM_HashJoinBasic_KeyTypes/"{utf8}"/Hashtable krows:16       3477411 ns      3476909 ns          202 rows/sec=4.71223M/s
BM_HashJoinBasic_KeyTypes/"{utf8}"/Hashtable krows:64      20586176 ns     20585156 ns           34 rows/sec=3.18365M/s