perf: optimize `array_replace` for scalar needle by lyne7-sc · Pull Request #22387 · apache/datafusion

lyne7-sc · 2026-05-20T07:18:31Z

Which issue does this PR close?

Closes #.

Rationale for this change

Currently, array_replace / array_replace_n / array_replace_all perform element-wise comparison by invoking compare_element_to_list against each row's sub-array individually. When the needle is a scalar, this can be optimized by performing a single vectorized not_distinct comparison over the entire flattened values buffer.

What changes are included in this PR?

Add a specialized replacement kernel that uses arrow_ord::cmp::not_distinct with Scalar wrapper for a single bulk comparison pass over the flat values buffer.
Extend SLT tests with multi-row scalar-argument coverage, empty-array edge cases, NULL needle replacement, and boundary n values for LargeList/FixedSizeList types.

Benchmarks

group                                                                      baseline                                optimized
-----                                                                      --------                                ---------
array_replace_all_int64/replace/list size: 10, num_rows: 4000              5.04  1124.5±146.98µs        ? ?/sec    1.00    223.1±2.79µs        ? ?/sec
array_replace_all_int64/replace/list size: 100, num_rows: 10000            1.64      7.2±0.59ms        ? ?/sec     1.00      4.4±0.12ms        ? ?/sec
array_replace_all_int64/replace/list size: 500, num_rows: 10000            1.16     25.3±4.09ms        ? ?/sec     1.00     21.8±0.69ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 10, num_rows: 4000       1.00      7.5±0.30ms        ? ?/sec     1.01      7.5±0.24ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 100, num_rows: 3000      1.00     38.5±0.52ms        ? ?/sec     1.02     39.2±1.02ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 300, num_rows: 1500      1.00     55.4±1.73ms        ? ?/sec     1.02     56.5±2.13ms        ? ?/sec
array_replace_boolean/replace/list size: 10, num_rows: 4000                4.57  1072.4±82.05µs        ? ?/sec     1.00    234.6±7.55µs        ? ?/sec
array_replace_boolean/replace/list size: 100, num_rows: 10000              2.38      3.7±0.43ms        ? ?/sec     1.00  1536.5±47.67µs        ? ?/sec
array_replace_boolean/replace/list size: 500, num_rows: 10000              1.51      6.5±0.51ms        ? ?/sec     1.00      4.3±0.12ms        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 10, num_rows: 4000      3.61  1174.3±90.82µs        ? ?/sec     1.00   325.2±26.75µs        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 100, num_rows: 10000    1.45      7.2±0.88ms        ? ?/sec     1.00      4.9±0.11ms        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 500, num_rows: 10000    1.05     25.9±2.34ms        ? ?/sec     1.00     24.6±0.71ms        ? ?/sec
array_replace_int64/replace/list size: 10, num_rows: 4000                  5.49  1025.4±24.08µs        ? ?/sec     1.00   186.7±18.10µs        ? ?/sec
array_replace_int64/replace/list size: 100, num_rows: 10000                2.46      3.6±0.13ms        ? ?/sec     1.00  1455.7±138.70µs        ? ?/sec
array_replace_int64/replace/list size: 500, num_rows: 10000                1.26      7.0±0.75ms        ? ?/sec     1.00      5.6±0.77ms        ? ?/sec
array_replace_int64_nested/replace/list size: 10, num_rows: 4000           1.03      7.3±0.14ms        ? ?/sec     1.00      7.2±0.21ms        ? ?/sec
array_replace_int64_nested/replace/list size: 100, num_rows: 3000          1.03     37.8±1.62ms        ? ?/sec     1.00     36.7±0.43ms        ? ?/sec
array_replace_int64_nested/replace/list size: 300, num_rows: 1500          1.03     53.2±1.16ms        ? ?/sec     1.00     51.7±1.87ms        ? ?/sec
array_replace_n_int64/replace/list size: 10, num_rows: 4000                5.02  1074.4±30.92µs        ? ?/sec     1.00    214.1±2.22µs        ? ?/sec
array_replace_n_int64/replace/list size: 100, num_rows: 10000              1.83      5.0±0.15ms        ? ?/sec     1.00      2.7±0.06ms        ? ?/sec
array_replace_n_int64/replace/list size: 500, num_rows: 10000              1.17     15.5±1.11ms        ? ?/sec     1.00     13.3±0.24ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 10, num_rows: 4000         1.05      7.5±0.45ms        ? ?/sec     1.00      7.1±0.07ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 100, num_rows: 3000        1.02     37.4±0.51ms        ? ?/sec     1.00     36.5±0.62ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 300, num_rows: 1500        1.02     54.9±4.97ms        ? ?/sec     1.00     53.8±3.15ms        ? ?/sec
array_replace_strings/replace/list size: 10, num_rows: 4000                2.78  1408.8±44.99µs        ? ?/sec     1.00   506.6±16.32µs        ? ?/sec
array_replace_strings/replace/list size: 100, num_rows: 10000              1.32     11.0±1.25ms        ? ?/sec     1.00      8.3±0.37ms        ? ?/sec
array_replace_strings/replace/list size: 500, num_rows: 10000              1.14     42.4±6.39ms        ? ?/sec     1.00     37.2±0.74ms        ? ?/sec

Are these changes tested?

Yes, existing and new slt edge-case tests in array_replace.slt.

Are there any user-facing changes?

No.

Jefffrey

One observation I have is this is a fast path for if from is a scalar, but it would be likely that to (and max too) might also be scalars in that case 🤔

Should we have the paths just be:

-- scalar fast path
select array_replace(array, scalar, scalar[, scalar]);
-- general fallback
select array_replace(array, scalar, array[, scalar]);
-- or any other combination

Thoughts?

lyne7-sc · 2026-05-21T11:22:10Z

@Jefffrey Thanks for the review! I agree that to and max are also commonly scalar in practice and have already specialized them. but I think the performance gains here are relatively minor.

The other suggestions have already been fixed.

Jefffrey · 2026-05-22T09:28:04Z

I was more thinking along the lines of explicitly expecting from/to/max to be scalars in the scalar path, instead of allowing them to be both scalar and array and having those code paths within the fast path. I guess what I'm trying to get at, is what behaviour is this optimization targeting? Like is there a common query pattern that this accelerates (where only from is scalar but to and max may be non-scalar)? Knowing this would help us know which code paths to support (otherwise we might introduce a lot of complexity to handle a code path that rarely occurs in practice 🤔)

lyne7-sc · 2026-05-23T03:16:54Z

I see your concern, all scalars is the most common pattern in practice, and scoping the optimization to just that case makes sense. Weighing in the code complexity of supporting those edge cases, I agree it's better to keep things simple — I'll narrow the fast path to all-scalars and let the general path handle the rest.

Jefffrey · 2026-05-24T00:29:24Z

run benchmark array_replace

adriangbot · 2026-05-24T00:32:38Z

🤖 Criterion benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c4526922152-305-wcmsk 6.12.68+ #1 SMP Wed Apr 1 02:23:28 UTC 2026 aarch64 GNU/Linux

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Comparing perf/replace (7e348c2) to 4f45193 (merge-base) diff
BENCH_NAME=array_replace
BENCH_COMMAND=cargo bench --features=parquet --bench array_replace
BENCH_FILTER=
Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-05-24T00:47:58Z

🤖 Criterion benchmark completed (GKE) | trigger

Instance: c4a-highmem-16 (12 vCPU / 65 GiB)

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Details

group                                                                      main                                   perf_replace
-----                                                                      ----                                   ------------
array_replace_all_int64/replace/list size: 10, num_rows: 4000              7.05  1697.9±11.78µs        ? ?/sec    1.00    240.7±1.63µs        ? ?/sec
array_replace_all_int64/replace/list size: 100, num_rows: 10000            2.65      8.4±0.04ms        ? ?/sec    1.00      3.2±0.02ms        ? ?/sec
array_replace_all_int64/replace/list size: 500, num_rows: 10000            1.69     26.6±0.16ms        ? ?/sec    1.00     15.7±0.13ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 10, num_rows: 4000       1.00     12.3±0.10ms        ? ?/sec    1.02     12.5±0.15ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 100, num_rows: 3000      1.06     72.7±1.16ms        ? ?/sec    1.00     68.9±0.84ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 300, num_rows: 1500      1.00     99.2±1.21ms        ? ?/sec    1.00     99.3±1.61ms        ? ?/sec
array_replace_boolean/replace/list size: 10, num_rows: 4000                4.81   1525.8±4.68µs        ? ?/sec    1.00    317.2±1.59µs        ? ?/sec
array_replace_boolean/replace/list size: 100, num_rows: 10000              2.90      4.8±0.01ms        ? ?/sec    1.00   1672.6±4.64µs        ? ?/sec
array_replace_boolean/replace/list size: 500, num_rows: 10000              1.58      8.4±0.02ms        ? ?/sec    1.00      5.3±0.01ms        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 10, num_rows: 4000      5.56   1665.8±8.51µs        ? ?/sec    1.00    299.8±1.91µs        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 100, num_rows: 10000    2.41      8.7±0.15ms        ? ?/sec    1.00      3.6±0.14ms        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 500, num_rows: 10000    1.12     37.6±0.22ms        ? ?/sec    1.00     33.6±0.23ms        ? ?/sec
array_replace_int64/replace/list size: 10, num_rows: 4000                  7.25   1593.3±6.46µs        ? ?/sec    1.00    219.8±2.17µs        ? ?/sec
array_replace_int64/replace/list size: 100, num_rows: 10000                3.02      5.5±0.11ms        ? ?/sec    1.00  1822.2±88.37µs        ? ?/sec
array_replace_int64/replace/list size: 500, num_rows: 10000                1.00      9.8±0.19ms        ? ?/sec    1.44     14.2±0.12ms        ? ?/sec
array_replace_int64_nested/replace/list size: 10, num_rows: 4000           1.00     12.3±0.12ms        ? ?/sec    1.00     12.2±0.10ms        ? ?/sec
array_replace_int64_nested/replace/list size: 100, num_rows: 3000          1.00     67.8±0.83ms        ? ?/sec    1.00     67.9±0.78ms        ? ?/sec
array_replace_int64_nested/replace/list size: 300, num_rows: 1500          1.00     96.9±1.77ms        ? ?/sec    1.00     96.6±1.75ms        ? ?/sec
array_replace_n_int64/replace/list size: 10, num_rows: 4000                7.22   1705.5±7.10µs        ? ?/sec    1.00    236.2±1.74µs        ? ?/sec
array_replace_n_int64/replace/list size: 100, num_rows: 10000              3.19      6.8±0.10ms        ? ?/sec    1.00      2.1±0.03ms        ? ?/sec
array_replace_n_int64/replace/list size: 500, num_rows: 10000              1.72     19.1±0.11ms        ? ?/sec    1.00     11.1±0.20ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 10, num_rows: 4000         1.00     12.4±0.12ms        ? ?/sec    1.00     12.3±0.12ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 100, num_rows: 3000        1.00     68.1±0.82ms        ? ?/sec    1.00     68.2±1.04ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 300, num_rows: 1500        1.03    100.9±1.24ms        ? ?/sec    1.00     98.3±1.72ms        ? ?/sec
array_replace_strings/replace/list size: 10, num_rows: 4000                4.77   1996.2±6.94µs        ? ?/sec    1.00    418.1±1.47µs        ? ?/sec
array_replace_strings/replace/list size: 100, num_rows: 10000              1.66     11.3±0.08ms        ? ?/sec    1.00      6.8±0.06ms        ? ?/sec
array_replace_strings/replace/list size: 500, num_rows: 10000              1.16     43.0±1.64ms        ? ?/sec    1.00     37.1±0.37ms        ? ?/sec

Resource Usage

base (merge-base)

Metric	Value
Wall time	310.1s
Peak memory	4.7 GiB
Avg memory	4.6 GiB
CPU user	358.1s
CPU sys	10.2s
Peak spill	0 B

branch

Metric	Value
Wall time	295.1s
Peak memory	4.8 GiB
Avg memory	4.6 GiB
CPU user	337.0s
CPU sys	18.2s
Peak spill	0 B

File an issue against this benchmark runner

Jefffrey · 2026-05-24T00:50:13Z

run benchmark array_replace

adriangbot · 2026-05-24T00:52:51Z

🤖 Criterion benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c4526965097-306-z8468 6.12.68+ #1 SMP Wed Apr 1 02:23:28 UTC 2026 aarch64 GNU/Linux

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Comparing perf/replace (7e348c2) to 4f45193 (merge-base) diff
BENCH_NAME=array_replace
BENCH_COMMAND=cargo bench --features=parquet --bench array_replace
BENCH_FILTER=
Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-05-24T01:04:33Z

🤖 Criterion benchmark completed (GKE) | trigger

Instance: c4a-highmem-16 (12 vCPU / 65 GiB)

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Details

group                                                                      main                                   perf_replace
-----                                                                      ----                                   ------------
array_replace_all_int64/replace/list size: 10, num_rows: 4000              7.09   1711.3±8.51µs        ? ?/sec    1.00    241.3±1.89µs        ? ?/sec
array_replace_all_int64/replace/list size: 100, num_rows: 10000            2.79      8.6±0.32ms        ? ?/sec    1.00      3.1±0.02ms        ? ?/sec
array_replace_all_int64/replace/list size: 500, num_rows: 10000            2.33     35.3±0.21ms        ? ?/sec    1.00     15.2±0.14ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 10, num_rows: 4000       1.00     12.4±0.12ms        ? ?/sec    1.00     12.4±0.13ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 100, num_rows: 3000      1.01     69.2±0.89ms        ? ?/sec    1.00     68.9±0.92ms        ? ?/sec
array_replace_all_int64_nested/replace/list size: 300, num_rows: 1500      1.02    100.4±2.51ms        ? ?/sec    1.00     98.9±1.08ms        ? ?/sec
array_replace_boolean/replace/list size: 10, num_rows: 4000                4.90  1552.6±14.01µs        ? ?/sec    1.00    316.9±1.61µs        ? ?/sec
array_replace_boolean/replace/list size: 100, num_rows: 10000              2.92      4.9±0.01ms        ? ?/sec    1.00   1672.1±5.03µs        ? ?/sec
array_replace_boolean/replace/list size: 500, num_rows: 10000              1.58      8.4±0.02ms        ? ?/sec    1.00      5.3±0.01ms        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 10, num_rows: 4000      5.73  1684.1±11.23µs        ? ?/sec    1.00    293.7±2.16µs        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 100, num_rows: 10000    2.23      8.0±0.19ms        ? ?/sec    1.00      3.6±0.06ms        ? ?/sec
array_replace_fixed_size_binary/replace/list size: 500, num_rows: 10000    1.08     35.6±0.16ms        ? ?/sec    1.00     32.9±0.53ms        ? ?/sec
array_replace_int64/replace/list size: 10, num_rows: 4000                  7.34  1609.7±14.74µs        ? ?/sec    1.00    219.3±2.34µs        ? ?/sec
array_replace_int64/replace/list size: 100, num_rows: 10000                4.52      5.2±0.04ms        ? ?/sec    1.00  1153.7±15.17µs        ? ?/sec
array_replace_int64/replace/list size: 500, num_rows: 10000                1.32     17.1±0.23ms        ? ?/sec    1.00     13.0±0.17ms        ? ?/sec
array_replace_int64_nested/replace/list size: 10, num_rows: 4000           1.01     12.3±0.12ms        ? ?/sec    1.00     12.2±0.10ms        ? ?/sec
array_replace_int64_nested/replace/list size: 100, num_rows: 3000          1.00     67.7±0.92ms        ? ?/sec    1.03     69.6±0.92ms        ? ?/sec
array_replace_int64_nested/replace/list size: 300, num_rows: 1500          1.01     98.1±1.73ms        ? ?/sec    1.00     97.0±1.71ms        ? ?/sec
array_replace_n_int64/replace/list size: 10, num_rows: 4000                7.27  1710.2±10.04µs        ? ?/sec    1.00    235.2±1.77µs        ? ?/sec
array_replace_n_int64/replace/list size: 100, num_rows: 10000              3.01      6.7±0.05ms        ? ?/sec    1.00      2.2±0.03ms        ? ?/sec
array_replace_n_int64/replace/list size: 500, num_rows: 10000              1.00     18.4±0.12ms        ? ?/sec    1.03     19.0±0.15ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 10, num_rows: 4000         1.00     12.4±0.13ms        ? ?/sec    1.00     12.4±0.14ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 100, num_rows: 3000        1.00     68.2±0.81ms        ? ?/sec    1.04     71.3±1.80ms        ? ?/sec
array_replace_n_int64_nested/replace/list size: 300, num_rows: 1500        1.01     98.2±1.52ms        ? ?/sec    1.00     97.5±1.19ms        ? ?/sec
array_replace_strings/replace/list size: 10, num_rows: 4000                4.87      2.0±0.01ms        ? ?/sec    1.00    412.8±1.37µs        ? ?/sec
array_replace_strings/replace/list size: 100, num_rows: 10000              1.62     10.9±0.04ms        ? ?/sec    1.00      6.8±0.04ms        ? ?/sec
array_replace_strings/replace/list size: 500, num_rows: 10000              1.25     47.2±2.12ms        ? ?/sec    1.00     37.7±0.99ms        ? ?/sec

Resource Usage

base (merge-base)

Metric	Value
Wall time	300.1s
Peak memory	4.7 GiB
Avg memory	4.6 GiB
CPU user	341.5s
CPU sys	19.1s
Peak spill	0 B

branch

Metric	Value
Wall time	295.1s
Peak memory	4.8 GiB
Avg memory	4.6 GiB
CPU user	333.8s
CPU sys	20.5s
Peak spill	0 B

File an issue against this benchmark runner

lyne7-sc · 2026-05-24T03:06:15Z

All comments addressed, thanks for your review! @Jefffrey

Jefffrey · 2026-05-25T02:19:34Z

Thanks @lyne7-sc

## Which issue does this PR close?  - Closes #. ## Rationale for this change  Similar to apache#22387 (array_replace scalar optimization) `array_remove` / `array_remove_n` / `array_remove_all` perform element-wise comparison by invoking `compare_element_to_list` against each row's sub-array individually. When the needle is a scalar, this can be optimized by performing a single vectorized `distinct` comparison over the entire flattened values buffer. ## What changes are included in this PR? - Add a specialized removal kernel (`general_remove_with_scalar`) that uses `arrow_ord::cmp::distinct` with `Scalar` wrapper for a single bulk comparison pass over the flat values buffer. - Extend SLT tests with multi-row scalar-argument coverage, NULL-containing arrays, empty-array edge cases, boundary `n` values, and LargeList type coverage.  ### Benchmarks ``` group main optimized ----- ---- --------- array_remove_all_int64/remove/list size: 10, num_rows: 4000 4.35 856.8±97.81µs ? ?/sec 1.00 196.9±4.48µs ? ?/sec array_remove_all_int64/remove/list size: 100, num_rows: 10000 1.90 5.5±0.09ms ? ?/sec 1.00 2.9±0.09ms ? ?/sec array_remove_all_int64/remove/list size: 500, num_rows: 10000 1.35 19.2±0.21ms ? ?/sec 1.00 14.2±0.48ms ? ?/sec array_remove_all_int64_nested/remove/list size: 10, num_rows: 4000 1.00 7.1±0.12ms ? ?/sec 1.04 7.4±0.12ms ? ?/sec array_remove_all_int64_nested/remove/list size: 100, num_rows: 3000 1.00 36.5±0.39ms ? ?/sec 1.05 38.3±2.61ms ? ?/sec array_remove_all_int64_nested/remove/list size: 300, num_rows: 1500 1.01 53.5±2.26ms ? ?/sec 1.00 53.0±0.99ms ? ?/sec array_remove_boolean/remove/list size: 10, num_rows: 4000 3.83 813.9±7.08µs ? ?/sec 1.00 212.4±2.28µs ? ?/sec array_remove_boolean/remove/list size: 100, num_rows: 10000 2.73 3.7±0.03ms ? ?/sec 1.00 1364.7±177.83µs ? ?/sec array_remove_boolean/remove/list size: 500, num_rows: 10000 2.34 9.8±0.14ms ? ?/sec 1.00 4.2±0.25ms ? ?/sec array_remove_fixed_size_binary/remove/list size: 10, num_rows: 4000 3.16 918.2±16.76µs ? ?/sec 1.00 290.6±9.79µs ? ?/sec array_remove_fixed_size_binary/remove/list size: 100, num_rows: 10000 1.56 6.9±0.13ms ? ?/sec 1.00 4.4±0.15ms ? ?/sec array_remove_fixed_size_binary/remove/list size: 500, num_rows: 10000 1.17 27.7±0.84ms ? ?/sec 1.00 23.6±2.04ms ? ?/sec array_remove_int64/remove/list size: 10, num_rows: 4000 4.55 825.7±6.30µs ? ?/sec 1.00 181.3±4.32µs ? ?/sec array_remove_int64/remove/list size: 100, num_rows: 10000 3.35 3.8±0.11ms ? ?/sec 1.00 1135.6±54.87µs ? ?/sec array_remove_int64/remove/list size: 500, num_rows: 10000 2.04 10.3±0.35ms ? ?/sec 1.00 5.1±0.39ms ? ?/sec array_remove_int64_nested/remove/list size: 10, num_rows: 4000 1.00 7.1±0.18ms ? ?/sec 1.02 7.2±0.07ms ? ?/sec array_remove_int64_nested/remove/list size: 100, num_rows: 3000 1.00 36.1±1.35ms ? ?/sec 1.07 38.5±3.67ms ? ?/sec array_remove_int64_nested/remove/list size: 300, num_rows: 1500 1.00 51.7±0.57ms ? ?/sec 1.05 54.1±2.13ms ? ?/sec array_remove_n_int64/remove/list size: 10, num_rows: 4000 4.43 845.3±5.00µs ? ?/sec 1.00 190.6±2.84µs ? ?/sec array_remove_n_int64/remove/list size: 100, num_rows: 10000 2.29 4.7±0.11ms ? ?/sec 1.00 2.0±0.12ms ? ?/sec array_remove_n_int64/remove/list size: 500, num_rows: 10000 1.63 14.8±0.42ms ? ?/sec 1.00 9.0±0.51ms ? ?/sec array_remove_n_int64_nested/remove/list size: 10, num_rows: 4000 1.00 7.0±0.09ms ? ?/sec 1.29 8.9±3.44ms ? ?/sec array_remove_n_int64_nested/remove/list size: 100, num_rows: 3000 1.00 36.6±0.42ms ? ?/sec 1.03 37.7±0.68ms ? ?/sec array_remove_n_int64_nested/remove/list size: 300, num_rows: 1500 1.00 52.7±3.68ms ? ?/sec 1.03 54.5±4.49ms ? ?/sec array_remove_strings/remove/list size: 10, num_rows: 4000 2.50 1144.6±21.95µs ? ?/sec 1.00 457.0±14.15µs ? ?/sec array_remove_strings/remove/list size: 100, num_rows: 10000 1.42 10.5±1.16ms ? ?/sec 1.00 7.4±0.34ms ? ?/sec array_remove_strings/remove/list size: 500, num_rows: 10000 1.12 39.8±0.91ms ? ?/sec 1.00 35.5±1.51ms ? ?/sec ``` ## Are these changes tested?  Yes, existing and new SLT edge-case tests in `array_remove.slt`. ## Are there any user-facing changes? No.

lyne7-sc and others added 3 commits May 20, 2026 11:16

Refactor array replace invocation

1468803

Refine array replace scalar path tests

3e79a38

Merge branch 'main' into perf/replace

47b811b

github-actions Bot added sqllogictest SQL Logic Tests (.slt) functions Changes to functions implementation labels May 20, 2026

lyne7-sc mentioned this pull request May 20, 2026

perf: optimize array_remove for scalar needle #22390

Merged

Jefffrey reviewed May 20, 2026

View reviewed changes

Comment thread datafusion/functions-nested/src/replace.rs Outdated

Comment thread datafusion/functions-nested/src/replace.rs Outdated

Comment thread datafusion/functions-nested/src/replace.rs Outdated

lyne7-sc added 3 commits May 21, 2026 15:53

better clarity

11dfa84

handle to and max arg scalar

3bae38b

Merge branch 'main' into perf/replace

e50617c

lyne7-sc and others added 3 commits May 23, 2026 11:16

Merge branch 'main' into perf/replace

2d828ce

handle all scalar path, other fallback general

6ac7c16

apply optimization suggestion from apache#22390

a4df12f

Jefffrey reviewed May 23, 2026

View reviewed changes

clean up per review feedback

7e348c2

Jefffrey approved these changes May 24, 2026

View reviewed changes

handle null value for max_arg

765f308

Jefffrey added this pull request to the merge queue May 25, 2026

Merged via the queue into apache:main with commit de41306 May 25, 2026
35 checks passed

Jefffrey mentioned this pull request May 25, 2026

array functions (remove, resize, extract, replace, repeat) not properly handling nulls for index arguments #22507

Open

lyne7-sc deleted the perf/replace branch May 26, 2026 01:48

Conversation

lyne7-sc commented May 20, 2026

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Benchmarks

Are these changes tested?

Are there any user-facing changes?

Uh oh!

Jefffrey left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

lyne7-sc commented May 21, 2026

Uh oh!

Jefffrey commented May 22, 2026

Uh oh!

lyne7-sc commented May 23, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Jefffrey commented May 24, 2026

Uh oh!

adriangbot commented May 24, 2026

Uh oh!

adriangbot commented May 24, 2026

Uh oh!

Jefffrey commented May 24, 2026

Uh oh!

adriangbot commented May 24, 2026

Uh oh!

adriangbot commented May 24, 2026

Uh oh!

lyne7-sc commented May 24, 2026

Uh oh!

Jefffrey commented May 25, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants