Compact StringView buffer during sparse (<50%) take to avoid holding the original buffers alive

[lyang24]

Which issue does this PR close?

On clickbench query profiles Noticed the buffers allocated on producer thread, dropped on consumer thread. Mimalloc's deferred-free mechanism caused up to 9.26% CPU in _mi_heap_delayed_free_partial.

Rationale for this change

The previous implementation treated take as a purely generic gather, leaving buffer compaction to the downstream coalescing step. While correct, this meant that in sparse cases we would:

1. Read and copy string data twice (once during take, again during coalescing), and
2. Keep large backing buffers alive across thread boundaries via Arc, even when only a small subset of the data was referenced.

This PR optimizes the sparse case by recognizing that compaction is effectively inevitable downstream. When take selects a small fraction of rows, we fuse gather and compaction into a single pass. This:

1. Avoids the second read/copy of string data
2. Prevents retaining large unused buffers
3. Reduces cross-thread memory retention

The dense case remains unchanged, so the existing fast path is preserved. The heuristic is intentionally conservative and only triggers when sparsity suggests compaction will be beneficial.

What changes are included in this PR?

The optimization fuses the take and compaction into a single pass via take_byte_view_compact. It kicks in automatically when the take is sparse (indices.len() < array.len() / 2):

Are these changes tested?

arrow select test passed

Are there any user-facing changes?

no

[Dandandan]

run benchmark take_kernels

[alamb-ghbot]

🤖 ./gh_compare_arrow.sh gh_compare_arrow.sh Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing take_byte_view_compact (c4d8432) to 7dbe58a diff
BENCH_NAME=take_kernels
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental,object_store --bench take_kernels
BENCH_FILTER=
BENCH_BRANCH_NAME=take_byte_view_compact
Results will be posted here when complete

[alamb-ghbot]

🤖: Benchmark completed

Details

group                                                                     main                                   take_byte_view_compact
-----                                                                     ----                                   ----------------------
take bool 1024                                                            1.00  1338.6±10.73ns        ? ?/sec    1.00   1332.0±2.99ns        ? ?/sec
take bool 512                                                             1.01    734.4±2.15ns        ? ?/sec    1.00    729.3±1.27ns        ? ?/sec
take bool null indices 1024                                               1.20  1256.1±45.38ns        ? ?/sec    1.00   1043.1±5.96ns        ? ?/sec
take bool null values 1024                                                1.00      2.6±0.01µs        ? ?/sec    1.00      2.6±0.00µs        ? ?/sec
take bool null values null indices 1024                                   1.44      3.0±0.03µs        ? ?/sec    1.00      2.1±0.07µs        ? ?/sec
take check bounds i32 1024                                                1.00   849.9±19.98ns        ? ?/sec    1.00    848.9±3.09ns        ? ?/sec
take check bounds i32 512                                                 1.01    592.5±3.26ns        ? ?/sec    1.00   589.1±12.71ns        ? ?/sec
take i32 1024                                                             1.00    718.7±1.15ns        ? ?/sec    1.00   715.7±10.72ns        ? ?/sec
take i32 512                                                              1.01    444.9±2.09ns        ? ?/sec    1.00    442.5±1.05ns        ? ?/sec
take i32 null indices 1024                                                1.00    997.1±2.51ns        ? ?/sec    1.01  1009.0±86.37ns        ? ?/sec
take i32 null values 1024                                                 1.00      2.0±0.01µs        ? ?/sec    1.01      2.1±0.02µs        ? ?/sec
take i32 null values null indices 1024                                    1.02      2.2±0.08µs        ? ?/sec    1.00      2.2±0.02µs        ? ?/sec
take primitive fsb value len: 12, indices: 1024                           1.00      3.5±0.04µs        ? ?/sec    1.00      3.4±0.01µs        ? ?/sec
take primitive fsb value len: 12, null values, indices: 1024              1.00      4.8±0.06µs        ? ?/sec    1.00      4.8±0.01µs        ? ?/sec
take primitive run logical len: 1024, physical len: 512, indices: 1024    1.02     20.7±0.07µs        ? ?/sec    1.00     20.3±0.09µs        ? ?/sec
take str 1024                                                             1.00     11.1±0.04µs        ? ?/sec    1.00     11.2±0.05µs        ? ?/sec
take str 512                                                              1.01      5.4±0.03µs        ? ?/sec    1.00      5.4±0.03µs        ? ?/sec
take str null indices 1024                                                1.00      7.8±0.03µs        ? ?/sec    1.01      7.8±0.03µs        ? ?/sec
take str null indices 512                                                 1.00      3.7±0.02µs        ? ?/sec    1.01      3.8±0.02µs        ? ?/sec
take str null values 1024                                                 1.00      8.8±0.08µs        ? ?/sec    1.00      8.8±0.05µs        ? ?/sec
take str null values null indices 1024                                    1.02      7.0±0.11µs        ? ?/sec    1.00      6.8±0.16µs        ? ?/sec
take stringview 1024                                                      1.10   898.5±10.53ns        ? ?/sec    1.00    819.5±7.92ns        ? ?/sec
take stringview 512                                                       1.18    599.4±8.94ns        ? ?/sec    1.00    509.6±4.78ns        ? ?/sec
take stringview null indices 1024                                         1.00  1431.0±26.67ns        ? ?/sec    1.00  1431.8±15.38ns        ? ?/sec
take stringview null indices 512                                          1.00    738.8±2.44ns        ? ?/sec    1.00    738.6±2.89ns        ? ?/sec
take stringview null values 1024                                          1.00      2.1±0.02µs        ? ?/sec    1.04      2.2±0.01µs        ? ?/sec
take stringview null values null indices 1024                             1.07      2.5±0.03µs        ? ?/sec    1.00      2.3±0.02µs        ? ?/sec

[Dandandan]

I think it makes sense to add some special cases, e.g. :

• no input buffers (i.e. source was all inlined)
• no output buffer (target is all inlined)
• a single output buffer (offset always zero)

[lyang24]

good catch added special cases especially all inlined case could be a fast path

[alamb]

run benchmark take_kernels

[alamb-ghbot]

🤖 ./gh_compare_arrow.sh gh_compare_arrow.sh Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing take_byte_view_compact (c4d8432) to 7dbe58a diff
BENCH_NAME=take_kernels
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental,object_store --bench take_kernels
BENCH_FILTER=
BENCH_BRANCH_NAME=take_byte_view_compact
Results will be posted here when complete

[alamb-ghbot]

🤖: Benchmark completed

Details

group                                                                     main                                   take_byte_view_compact
-----                                                                     ----                                   ----------------------
take bool 1024                                                            1.00   1333.8±1.21ns        ? ?/sec    1.00   1331.1±1.78ns        ? ?/sec
take bool 512                                                             1.01    734.8±1.01ns        ? ?/sec    1.00    730.8±1.32ns        ? ?/sec
take bool null indices 1024                                               1.00  1292.5±28.41ns        ? ?/sec    1.61      2.1±0.03µs        ? ?/sec
take bool null values 1024                                                1.00      2.6±0.01µs        ? ?/sec    1.00      2.6±0.01µs        ? ?/sec
take bool null values null indices 1024                                   1.00      2.9±0.11µs        ? ?/sec    1.48      4.3±0.09µs        ? ?/sec
take check bounds i32 1024                                                1.00    846.0±2.16ns        ? ?/sec    1.00    846.6±2.51ns        ? ?/sec
take check bounds i32 512                                                 1.14    588.9±7.75ns        ? ?/sec    1.00    517.2±6.61ns        ? ?/sec
take i32 1024                                                             1.01   721.6±20.07ns        ? ?/sec    1.00    716.5±1.13ns        ? ?/sec
take i32 512                                                              1.00    444.7±2.67ns        ? ?/sec    1.00    445.4±1.94ns        ? ?/sec
take i32 null indices 1024                                                1.44  1448.4±12.46ns        ? ?/sec    1.00  1007.1±79.72ns        ? ?/sec
take i32 null values 1024                                                 1.00      2.0±0.02µs        ? ?/sec    1.01      2.1±0.02µs        ? ?/sec
take i32 null values null indices 1024                                    1.00      2.2±0.02µs        ? ?/sec    1.38      3.0±0.04µs        ? ?/sec
take primitive fsb value len: 12, indices: 1024                           1.01      3.5±0.07µs        ? ?/sec    1.00      3.4±0.01µs        ? ?/sec
take primitive fsb value len: 12, null values, indices: 1024              1.00      4.8±0.12µs        ? ?/sec    1.00      4.8±0.09µs        ? ?/sec
take primitive run logical len: 1024, physical len: 512, indices: 1024    1.00     20.8±0.10µs        ? ?/sec    1.00     20.8±0.23µs        ? ?/sec
take str 1024                                                             1.00     11.1±0.04µs        ? ?/sec    1.00     11.1±0.06µs        ? ?/sec
take str 512                                                              1.00      5.4±0.03µs        ? ?/sec    1.00      5.4±0.11µs        ? ?/sec
take str null indices 1024                                                1.01      7.9±0.05µs        ? ?/sec    1.00      7.8±0.09µs        ? ?/sec
take str null indices 512                                                 1.00      3.8±0.02µs        ? ?/sec    1.00      3.8±0.02µs        ? ?/sec
take str null values 1024                                                 1.01      8.8±0.26µs        ? ?/sec    1.00      8.8±0.14µs        ? ?/sec
take str null values null indices 1024                                    1.00      7.0±0.09µs        ? ?/sec    1.09      7.6±0.05µs        ? ?/sec
take stringview 1024                                                      1.10    894.5±9.39ns        ? ?/sec    1.00    810.3±1.60ns        ? ?/sec
take stringview 512                                                       1.16    591.9±6.99ns        ? ?/sec    1.00    510.2±2.35ns        ? ?/sec
take stringview null indices 1024                                         1.00  1454.3±33.02ns        ? ?/sec    1.00  1448.6±54.65ns        ? ?/sec
take stringview null indices 512                                          1.01    739.6±2.48ns        ? ?/sec    1.00    731.9±4.32ns        ? ?/sec
take stringview null values 1024                                          1.00      2.1±0.07µs        ? ?/sec    1.01      2.1±0.11µs        ? ?/sec
take stringview null values null indices 1024                             1.00      2.5±0.05µs        ? ?/sec    1.25      3.1±0.03µs        ? ?/sec