perf : avoided repeated bounds checks in `take` kernel by Rich-T-kid · Pull Request #10441 · apache/arrow-rs

Rich-T-kid · 2026-07-27T02:17:45Z

Which issue does this PR close?

Closes Reduce bound checks in take kernels #8879.
this is currently blocked by benchmark/perf: add wide column take benchmarks #10442. After that is merged into main we can compare

Rationale for this change

What changes are included in this PR?

instead of checking length bound for each array we only do this for the first array and then remaining arrays avoid validation. The rational behind this is that record batches are all of the same size.

Are these changes tested?

logically nothing has changed

Are there any user-facing changes?

no

Rich-T-kid · 2026-07-27T03:04:54Z

I may have mis-understood #8879 then.

# Which issue does this PR close?  - works towards #8879. # Rationale for this change see #8879 (comment) the point of these benchmarks are to measure the change of #10441  # What changes are included in this PR? adds benchmarks for `take_record_batch()` .  # Are these changes tested? n/a  # Are there any user-facing changes? no

alamb · 2026-07-27T16:00:39Z

Update to get #10442

Rich-T-kid · 2026-07-27T17:14:01Z

take_record_batch 5 primitive cols no nulls 1024 1.00 3.4±0.02µs ? ?/sec 1.00 3.4±0.01µs ? ?/sec
take_record_batch 6 string/binary cols no nulls 1024 1.04 28.0±0.03µs ? ?/sec 1.00 26.9±0.04µs ? ?/sec
take_record_batch 7 mixed cols null values 1024 1.01 17.7±0.11µs ? ?/sec 1.00 17.6±0.09µs ? ?/sec
take_record_batch 7 mixed cols null values null indices 1024 1.00 14.8±0.05µs ? ?/sec 1.00 14.8±0.04µs ? ?/sec

yup like @Jefffrey mentioned this actually adds an extra validation step so there isn't much performance gains.

maybe what #8879 was referring to was actually the bounds checks in each of the take_impl branches

Rich-T-kid · 2026-07-27T19:10:26Z

local benchmarks for https://github.com/apache/arrow-rs/commit/1a626c58c524e30c945937be9baeac73932abad5 compared to `main`

the latest commit adds 5 unsafe blocks but they are all within loops which is causing a nice perf boost.

this avoids the length check array[index] does since the bounds are validated before take_impl() is called

Jefffrey · 2026-07-28T00:39:24Z

                // SAFETY: idx is a valid index in indices.nulls() --> idx<indices.len()
-                if values.value(unsafe { indices.value_unchecked(idx).as_usize() }) {
+                // SAFETY: caller guarantees all valid indices are in-bounds
+                if unsafe { values.value_unchecked(indices.value_unchecked(idx).as_usize()) } {


im not sure we can do this, since callers can call take without an unsafe block and with check bounds off and pass in an invalid index; this would cause UB via safe interfaces i believe?

I agree -- though it might be faster to validate the indices once at the start of take_impl and then use the unchecked accessors

We could also potentially add a take_unchecked variant or something to skip the acces check

Rich-T-kid · 2026-07-28T03:45:37Z

im not sure we can do this, since callers can call take without an unsafe block and with check bounds off and pass in an invalid index; this would cause UB via safe interfaces i believe?

aah, I think ive come full circle to understand what @Dandandan meant in

We can run consider having both bound-checked and non-bound-checked versions of the private kernels, and use them for take_record_batch for doing the take on >1 arrays.

I can update the PR to introduce a new unsafe method that uses unsafe array access methods. The main issue with this would be the code bloat.

Rich-T-kid · 2026-07-28T03:50:14Z

out of curiosity could you run the benchmarks @Jefffrey. if the perf difference is < +10% This may not even be worth it

Jefffrey · 2026-07-28T05:16:56Z

run benchmark take_kernels

adriangbot · 2026-07-28T05:34:59Z

🤖 Arrow 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                                   rich-t-kid_avoid-take-bounds-checks
-----                                                                     ----                                   -----------------------------------
take bool 1024                                                            1.17   1039.0±0.76ns        ? ?/sec    1.00    889.1±0.40ns        ? ?/sec
take bool 512                                                             1.17    577.0±1.52ns        ? ?/sec    1.00    494.7±1.36ns        ? ?/sec
take bool null indices 1024                                               1.01   904.2±33.83ns        ? ?/sec    1.00    891.7±1.53ns        ? ?/sec
take bool null values 1024                                                1.17      2.0±0.00µs        ? ?/sec    1.00   1739.0±0.75ns        ? ?/sec
take bool null values null indices 1024                                   1.00  1602.8±12.81ns        ? ?/sec    1.00  1605.7±12.46ns        ? ?/sec
take check bounds i32 1024                                                1.02    667.2±1.72ns        ? ?/sec    1.00    653.3±0.55ns        ? ?/sec
take check bounds i32 512                                                 1.00    391.5±0.51ns        ? ?/sec    1.15    450.8±0.73ns        ? ?/sec
take fsb value len: 12, indices: 1024                                     1.00      2.5±0.01µs        ? ?/sec    1.00      2.5±0.00µs        ? ?/sec
take fsb value len: 12, null values, indices: 1024                        1.06      3.5±0.00µs        ? ?/sec    1.00      3.3±0.00µs        ? ?/sec
take fsb value optimized len: 16, indices: 1024                           1.08    638.6±2.47ns        ? ?/sec    1.00    591.9±1.02ns        ? ?/sec
take fsb value optimized len: 16, null values, indices: 1024              1.14   1668.7±3.30ns        ? ?/sec    1.00   1464.5±2.11ns        ? ?/sec
take i32 1024                                                             1.03    527.2±0.79ns        ? ?/sec    1.00    512.6±0.74ns        ? ?/sec
take i32 512                                                              1.00    351.4±0.84ns        ? ?/sec    1.00    350.4±0.92ns        ? ?/sec
take i32 null indices 1024                                                1.00    866.8±3.43ns        ? ?/sec    1.02    886.5±1.60ns        ? ?/sec
take i32 null values 1024                                                 1.12   1548.3±1.24ns        ? ?/sec    1.00   1388.0±1.33ns        ? ?/sec
take i32 null values null indices 1024                                    1.00   1729.2±2.11ns        ? ?/sec    1.00   1734.5±2.24ns        ? ?/sec
take list i32 1024                                                        1.01      9.0±0.03µs        ? ?/sec    1.00      8.8±0.03µs        ? ?/sec
take list i32 512                                                         1.01      4.8±0.02µs        ? ?/sec    1.00      4.7±0.01µs        ? ?/sec
take list i32 null indices 1024                                           1.03     10.5±0.02µs        ? ?/sec    1.00     10.2±0.03µs        ? ?/sec
take list i32 null values 1024                                            1.02      6.2±0.01µs        ? ?/sec    1.00      6.0±0.02µs        ? ?/sec
take list i32 null values null indices 1024                               1.02      7.2±0.07µs        ? ?/sec    1.00      7.0±0.04µs        ? ?/sec
take listview i32 1024                                                    1.00    969.4±0.74ns        ? ?/sec    1.10   1067.4±0.88ns        ? ?/sec
take listview i32 512                                                     1.00    600.0±1.12ns        ? ?/sec    1.02    609.2±0.46ns        ? ?/sec
take listview i32 null indices 1024                                       1.04   1677.0±2.02ns        ? ?/sec    1.00   1607.4±2.83ns        ? ?/sec
take listview i32 null values 1024                                        1.11      2.1±0.00µs        ? ?/sec    1.00   1842.5±1.03ns        ? ?/sec
take listview i32 null values null indices 1024                           1.04      2.5±0.00µs        ? ?/sec    1.00      2.4±0.00µs        ? ?/sec
take primitive run logical len: 1024, physical len: 512, indices: 1024    1.00     15.5±0.01µs        ? ?/sec    1.00     15.6±0.01µs        ? ?/sec
take str 1024                                                             1.00      8.0±0.00µs        ? ?/sec    1.02      8.2±0.00µs        ? ?/sec
take str 512                                                              1.00      3.8±0.00µs        ? ?/sec    1.02      3.8±0.00µs        ? ?/sec
take str null indices 1024                                                1.00      4.3±0.07µs        ? ?/sec    1.11      4.8±0.15µs        ? ?/sec
take str null indices 512                                                 1.00      2.1±0.03µs        ? ?/sec    1.10      2.4±0.07µs        ? ?/sec
take str null values 1024                                                 1.00      4.9±0.06µs        ? ?/sec    1.05      5.2±0.09µs        ? ?/sec
take str null values null indices 1024                                    1.00      3.2±0.02µs        ? ?/sec    1.08      3.4±0.05µs        ? ?/sec
take stringview 1024                                                      1.13    905.5±0.96ns        ? ?/sec    1.00    798.9±1.51ns        ? ?/sec
take stringview 512                                                       1.06    565.3±0.69ns        ? ?/sec    1.00    534.0±1.67ns        ? ?/sec
take stringview null indices 1024                                         1.07    979.9±2.25ns        ? ?/sec    1.00    918.0±1.67ns        ? ?/sec
take stringview null indices 512                                          1.00    560.1±0.43ns        ? ?/sec    1.01    568.5±1.30ns        ? ?/sec
take stringview null values 1024                                          1.21   1978.9±1.74ns        ? ?/sec    1.00   1632.9±1.59ns        ? ?/sec
take stringview null values null indices 1024                             1.01   1791.8±2.21ns        ? ?/sec    1.00   1771.2±3.36ns        ? ?/sec

Resource Usage

base (merge-base)

Metric	Value
Wall time	375.1s
Peak memory	14.8 MiB
Avg memory	12.2 MiB
CPU user	371.2s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	375.1s
Peak memory	15.6 MiB
Avg memory	13.0 MiB
CPU user	369.4s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

alamb

Thanks for pushing on this @Rich-T-kid

alamb · 2026-07-28T06:51:04Z

10% performance on take which gets called a lot I think is worth pursing if possible

Rich-T-kid · 2026-07-28T14:09:22Z

10% performance on take which gets called a lot I think is worth pursing if possible

i'll continue to look into this 🚀

Rich-T-kid · 2026-07-28T14:10:29Z

I noticed this benchmarks run #10441 (comment) doesn't include the new benchmarks from #10442 for some reason 🤔

Jefffrey · 2026-07-28T14:23:28Z

run benchmark take_kernels

adriangbot · 2026-07-28T14:40:46Z

🤖 Arrow 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                                   rich-t-kid_avoid-take-bounds-checks
-----                                                                     ----                                   -----------------------------------
take bool 1024                                                            1.16   1036.0±0.61ns        ? ?/sec    1.00    893.5±1.55ns        ? ?/sec
take bool 512                                                             1.16    579.1±0.41ns        ? ?/sec    1.00    497.6±1.53ns        ? ?/sec
take bool null indices 1024                                               1.05   852.1±30.78ns        ? ?/sec    1.00    813.4±2.50ns        ? ?/sec
take bool null values 1024                                                1.16      2.0±0.00µs        ? ?/sec    1.00   1747.7±2.13ns        ? ?/sec
take bool null values null indices 1024                                   1.12  1623.7±13.14ns        ? ?/sec    1.00  1450.4±13.89ns        ? ?/sec
take check bounds i32 1024                                                1.02    669.2±0.98ns        ? ?/sec    1.00    657.9±2.17ns        ? ?/sec
take check bounds i32 512                                                 1.00    395.9±1.06ns        ? ?/sec    1.15    455.1±2.31ns        ? ?/sec
take fsb value len: 12, indices: 1024                                     1.00      2.5±0.00µs        ? ?/sec    1.00      2.5±0.00µs        ? ?/sec
take fsb value len: 12, null values, indices: 1024                        1.05      3.5±0.01µs        ? ?/sec    1.00      3.4±0.00µs        ? ?/sec
take fsb value optimized len: 16, indices: 1024                           1.00    590.9±0.68ns        ? ?/sec    1.01    597.7±1.97ns        ? ?/sec
take fsb value optimized len: 16, null values, indices: 1024              1.10   1617.2±0.77ns        ? ?/sec    1.00   1475.2±3.74ns        ? ?/sec
take i32 1024                                                             1.03    528.6±1.95ns        ? ?/sec    1.00    514.7±1.92ns        ? ?/sec
take i32 512                                                              1.00    352.8±1.07ns        ? ?/sec    1.01    356.1±4.24ns        ? ?/sec
take i32 null indices 1024                                                1.00    857.8±1.17ns        ? ?/sec    1.02    875.5±3.92ns        ? ?/sec
take i32 null values 1024                                                 1.11   1546.9±2.45ns        ? ?/sec    1.00   1395.7±1.77ns        ? ?/sec
take i32 null values null indices 1024                                    1.04   1730.9±2.91ns        ? ?/sec    1.00   1668.8±4.13ns        ? ?/sec
take list i32 1024                                                        1.00      8.8±0.03µs        ? ?/sec    1.00      8.8±0.03µs        ? ?/sec
take list i32 512                                                         1.00      4.7±0.02µs        ? ?/sec    1.02      4.8±0.01µs        ? ?/sec
take list i32 null indices 1024                                           1.02     10.5±0.02µs        ? ?/sec    1.00     10.3±0.02µs        ? ?/sec
take list i32 null values 1024                                            1.00      6.1±0.01µs        ? ?/sec    1.01      6.1±0.01µs        ? ?/sec
take list i32 null values null indices 1024                               1.03      7.1±0.04µs        ? ?/sec    1.00      6.9±0.04µs        ? ?/sec
take listview i32 1024                                                    1.17   1130.3±0.84ns        ? ?/sec    1.00    962.5±2.46ns        ? ?/sec
take listview i32 512                                                     1.04    629.2±2.25ns        ? ?/sec    1.00    607.0±2.39ns        ? ?/sec
take listview i32 null indices 1024                                       1.00   1687.9±2.03ns        ? ?/sec    1.00   1689.8±2.56ns        ? ?/sec
take listview i32 null values 1024                                        1.18      2.1±0.00µs        ? ?/sec    1.00   1822.3±2.52ns        ? ?/sec
take listview i32 null values null indices 1024                           1.02      2.4±0.01µs        ? ?/sec    1.00      2.4±0.00µs        ? ?/sec
take primitive run logical len: 1024, physical len: 512, indices: 1024    1.01     15.6±0.01µs        ? ?/sec    1.00     15.4±0.01µs        ? ?/sec
take str 1024                                                             1.00      8.0±0.00µs        ? ?/sec    1.01      8.1±0.00µs        ? ?/sec
take str 512                                                              1.01      3.7±0.00µs        ? ?/sec    1.00      3.7±0.01µs        ? ?/sec
take str null indices 1024                                                1.00      4.3±0.07µs        ? ?/sec    1.12      4.8±0.13µs        ? ?/sec
take str null indices 512                                                 1.00      2.0±0.03µs        ? ?/sec    1.18      2.4±0.06µs        ? ?/sec
take str null values 1024                                                 1.00      4.9±0.05µs        ? ?/sec    1.03      5.0±0.11µs        ? ?/sec
take str null values null indices 1024                                    1.00      3.2±0.02µs        ? ?/sec    1.01      3.2±0.04µs        ? ?/sec
take stringview 1024                                                      1.09    804.0±1.34ns        ? ?/sec    1.00    738.3±3.35ns        ? ?/sec
take stringview 512                                                       1.00    471.3±1.02ns        ? ?/sec    1.01    474.9±3.00ns        ? ?/sec
take stringview null indices 1024                                         1.07    981.7±2.61ns        ? ?/sec    1.00    920.0±3.18ns        ? ?/sec
take stringview null indices 512                                          1.00    566.6±1.75ns        ? ?/sec    1.00    565.0±3.05ns        ? ?/sec
take stringview null values 1024                                          1.08   1721.5±1.43ns        ? ?/sec    1.00   1593.7±5.30ns        ? ?/sec
take stringview null values null indices 1024                             1.04   1757.0±2.72ns        ? ?/sec    1.00   1689.3±5.88ns        ? ?/sec
take_record_batch 5 primitive cols no nulls 1024                          1.04      3.4±0.00µs        ? ?/sec    1.00      3.3±0.01µs        ? ?/sec
take_record_batch 6 string/binary cols no nulls 1024                      1.00     27.2±0.05µs        ? ?/sec    1.00     27.1±0.05µs        ? ?/sec
take_record_batch 7 mixed cols null values 1024                           1.03     17.7±0.08µs        ? ?/sec    1.00     17.2±0.29µs        ? ?/sec
take_record_batch 7 mixed cols null values null indices 1024              1.03     15.0±0.04µs        ? ?/sec    1.00     14.6±0.11µs        ? ?/sec

Resource Usage

base (merge-base)

Metric	Value
Wall time	410.1s
Peak memory	16.0 MiB
Avg memory	12.7 MiB
CPU user	406.2s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	410.1s
Peak memory	14.6 MiB
Avg memory	11.9 MiB
CPU user	403.5s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

Rich-T-kid · 2026-07-30T17:45:55Z

2947869 compared to main

we should also run adriabot again.

This PR (2947869) threads a generic parameter through a couple of functions. It's just a simple flag that determines whether value_unchecked() is used instead of array[index]. This was the most minimal change I could think of without duplicating code. The approach has zero overhead for the current safe code path due to Rust's monomorphization, each const gets replaced with its true/false branch, and dead code analysis removes the alternative branch, meaning there won't be any actual repeated comparisons within the functions.

any thoughts @alamb @Jefffrey ?

Jefffrey · 2026-07-30T22:12:19Z

run benchmark take_kernels

adriangbot · 2026-07-30T22:29:34Z

🤖 Arrow 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                                   rich-t-kid_avoid-take-bounds-checks
-----                                                                     ----                                   -----------------------------------
take bool 1024                                                            1.00   1032.6±0.69ns        ? ?/sec    1.00   1036.0±1.06ns        ? ?/sec
take bool 512                                                             1.00    573.5±0.71ns        ? ?/sec    1.00    575.0±0.79ns        ? ?/sec
take bool null indices 1024                                               1.00   767.5±28.10ns        ? ?/sec    1.16    893.8±3.89ns        ? ?/sec
take bool null values 1024                                                1.00      2.0±0.00µs        ? ?/sec    1.00      2.0±0.00µs        ? ?/sec
take bool null values null indices 1024                                   1.00  1437.5±11.61ns        ? ?/sec    1.10   1582.9±8.48ns        ? ?/sec
take check bounds i32 1024                                                1.02    668.0±2.74ns        ? ?/sec    1.00    655.1±1.39ns        ? ?/sec
take check bounds i32 512                                                 1.00    391.3±1.90ns        ? ?/sec    1.16    453.2±2.18ns        ? ?/sec
take fsb value len: 12, indices: 1024                                     1.01      2.5±0.00µs        ? ?/sec    1.00      2.4±0.00µs        ? ?/sec
take fsb value len: 12, null values, indices: 1024                        1.00      3.5±0.00µs        ? ?/sec    1.01      3.5±0.00µs        ? ?/sec
take fsb value optimized len: 16, indices: 1024                           1.00    592.5±1.90ns        ? ?/sec    1.08    638.2±1.79ns        ? ?/sec
take fsb value optimized len: 16, null values, indices: 1024              1.00   1613.1±2.45ns        ? ?/sec    1.03   1664.1±2.46ns        ? ?/sec
take i32 1024                                                             1.02    525.7±1.98ns        ? ?/sec    1.00    513.1±1.14ns        ? ?/sec
take i32 512                                                              1.00    351.2±1.88ns        ? ?/sec    1.00    352.8±0.87ns        ? ?/sec
take i32 null indices 1024                                                1.00    858.2±1.52ns        ? ?/sec    1.02    873.8±1.44ns        ? ?/sec
take i32 null values 1024                                                 1.00   1542.5±3.86ns        ? ?/sec    1.00   1541.5±1.26ns        ? ?/sec
take i32 null values null indices 1024                                    1.00   1652.6±4.66ns        ? ?/sec    1.05   1732.1±3.05ns        ? ?/sec
take list i32 1024                                                        1.00      8.8±0.01µs        ? ?/sec    1.00      8.8±0.02µs        ? ?/sec
take list i32 512                                                         1.00      4.6±0.01µs        ? ?/sec    1.01      4.7±0.01µs        ? ?/sec
take list i32 null indices 1024                                           1.00     10.2±0.03µs        ? ?/sec    1.01     10.2±0.02µs        ? ?/sec
take list i32 null values 1024                                            1.00      6.1±0.02µs        ? ?/sec    1.01      6.2±0.02µs        ? ?/sec
take list i32 null values null indices 1024                               1.00      7.1±0.09µs        ? ?/sec    1.01      7.1±0.08µs        ? ?/sec
take listview i32 1024                                                    1.10   1080.5±1.51ns        ? ?/sec    1.00    979.6±1.79ns        ? ?/sec
take listview i32 512                                                     1.00    610.5±1.01ns        ? ?/sec    1.05    638.5±0.91ns        ? ?/sec
take listview i32 null indices 1024                                       1.00   1607.5±0.91ns        ? ?/sec    1.01   1622.2±1.87ns        ? ?/sec
take listview i32 null values 1024                                        1.00   1968.0±2.13ns        ? ?/sec    1.01   1991.5±2.05ns        ? ?/sec
take listview i32 null values null indices 1024                           1.00      2.4±0.00µs        ? ?/sec    1.07      2.5±0.00µs        ? ?/sec
take primitive run logical len: 1024, physical len: 512, indices: 1024    1.00     15.3±0.01µs        ? ?/sec    1.01     15.4±0.02µs        ? ?/sec
take str 1024                                                             1.00      8.0±0.00µs        ? ?/sec    1.04      8.4±0.00µs        ? ?/sec
take str 512                                                              1.01      3.7±0.00µs        ? ?/sec    1.00      3.7±0.00µs        ? ?/sec
take str null indices 1024                                                1.00      4.2±0.05µs        ? ?/sec    1.03      4.4±0.04µs        ? ?/sec
take str null indices 512                                                 1.00      2.1±0.03µs        ? ?/sec    1.06      2.2±0.02µs        ? ?/sec
take str null values 1024                                                 1.00      4.9±0.04µs        ? ?/sec    1.00      4.9±0.05µs        ? ?/sec
take str null values null indices 1024                                    1.00      3.2±0.02µs        ? ?/sec    1.04      3.3±0.01µs        ? ?/sec
take stringview 1024                                                      1.00    809.2±3.66ns        ? ?/sec    1.13    910.8±1.62ns        ? ?/sec
take stringview 512                                                       1.00    471.8±4.29ns        ? ?/sec    1.18    557.2±1.51ns        ? ?/sec
take stringview null indices 1024                                         1.00    916.2±1.61ns        ? ?/sec    1.08    986.0±2.93ns        ? ?/sec
take stringview null indices 512                                          1.00    564.5±1.61ns        ? ?/sec    1.00    565.1±2.00ns        ? ?/sec
take stringview null values 1024                                          1.00   1717.0±1.92ns        ? ?/sec    1.15   1980.6±2.08ns        ? ?/sec
take stringview null values null indices 1024                             1.00   1691.0±1.83ns        ? ?/sec    1.04   1752.5±1.69ns        ? ?/sec
take_record_batch 5 primitive cols no nulls 1024                          1.05      3.4±0.01µs        ? ?/sec    1.00      3.2±0.00µs        ? ?/sec
take_record_batch 6 string/binary cols no nulls 1024                      1.00     26.9±0.08µs        ? ?/sec    1.00     26.9±0.05µs        ? ?/sec
take_record_batch 7 mixed cols null values 1024                           1.03     17.7±0.08µs        ? ?/sec    1.00     17.1±0.36µs        ? ?/sec
take_record_batch 7 mixed cols null values null indices 1024              1.00     14.2±0.05µs        ? ?/sec    1.01     14.3±0.12µs        ? ?/sec

Resource Usage

base (merge-base)

Metric	Value
Wall time	405.1s
Peak memory	14.9 MiB
Avg memory	11.1 MiB
CPU user	402.3s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	410.1s
Peak memory	15.2 MiB
Avg memory	13.0 MiB
CPU user	406.5s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

Jefffrey · 2026-07-30T22:35:50Z

run benchmark take_kernels

adriangbot · 2026-07-30T22:52:48Z

🤖 Arrow 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                                   rich-t-kid_avoid-take-bounds-checks
-----                                                                     ----                                   -----------------------------------
take bool 1024                                                            1.00   1035.2±1.22ns        ? ?/sec    1.00   1035.2±1.48ns        ? ?/sec
take bool 512                                                             1.00    571.4±0.58ns        ? ?/sec    1.01    574.8±1.43ns        ? ?/sec
take bool null indices 1024                                               1.00   780.2±31.42ns        ? ?/sec    1.21    940.9±1.47ns        ? ?/sec
take bool null values 1024                                                1.00      2.0±0.00µs        ? ?/sec    1.00      2.0±0.00µs        ? ?/sec
take bool null values null indices 1024                                   1.00  1432.1±10.38ns        ? ?/sec    1.11   1585.9±9.11ns        ? ?/sec
take check bounds i32 1024                                                1.01    664.8±1.38ns        ? ?/sec    1.00    658.9±5.17ns        ? ?/sec
take check bounds i32 512                                                 1.00    389.3±0.43ns        ? ?/sec    1.17    456.0±6.29ns        ? ?/sec
take fsb value len: 12, indices: 1024                                     1.01      2.5±0.01µs        ? ?/sec    1.00      2.5±0.00µs        ? ?/sec
take fsb value len: 12, null values, indices: 1024                        1.00      3.5±0.00µs        ? ?/sec    1.01      3.5±0.00µs        ? ?/sec
take fsb value optimized len: 16, indices: 1024                           1.00    592.6±2.13ns        ? ?/sec    1.08    639.5±2.13ns        ? ?/sec
take fsb value optimized len: 16, null values, indices: 1024              1.00   1609.5±1.69ns        ? ?/sec    1.04   1671.9±2.10ns        ? ?/sec
take i32 1024                                                             1.02    524.5±5.83ns        ? ?/sec    1.00    512.5±1.79ns        ? ?/sec
take i32 512                                                              1.00    350.4±0.80ns        ? ?/sec    1.01    352.7±1.10ns        ? ?/sec
take i32 null indices 1024                                                1.00    855.1±1.64ns        ? ?/sec    1.02    872.2±2.31ns        ? ?/sec
take i32 null values 1024                                                 1.00   1539.5±1.82ns        ? ?/sec    1.00   1544.4±1.53ns        ? ?/sec
take i32 null values null indices 1024                                    1.00   1653.2±3.63ns        ? ?/sec    1.05   1735.5±4.15ns        ? ?/sec
take list i32 1024                                                        1.01      8.8±0.01µs        ? ?/sec    1.00      8.7±0.03µs        ? ?/sec
take list i32 512                                                         1.02      4.8±0.01µs        ? ?/sec    1.00      4.7±0.02µs        ? ?/sec
take list i32 null indices 1024                                           1.00     10.2±0.02µs        ? ?/sec    1.01     10.2±0.03µs        ? ?/sec
take list i32 null values 1024                                            1.00      6.1±0.01µs        ? ?/sec    1.01      6.2±0.01µs        ? ?/sec
take list i32 null values null indices 1024                               1.00      7.0±0.10µs        ? ?/sec    1.01      7.1±0.08µs        ? ?/sec
take listview i32 1024                                                    1.00    971.0±1.50ns        ? ?/sec    1.09   1060.2±2.38ns        ? ?/sec
take listview i32 512                                                     1.00    605.9±2.60ns        ? ?/sec    1.17    709.8±3.68ns        ? ?/sec
take listview i32 null indices 1024                                       1.04   1683.9±4.84ns        ? ?/sec    1.00   1620.7±5.24ns        ? ?/sec
take listview i32 null values 1024                                        1.05      2.2±0.00µs        ? ?/sec    1.00      2.1±0.01µs        ? ?/sec
take listview i32 null values null indices 1024                           1.01      2.4±0.00µs        ? ?/sec    1.00      2.4±0.00µs        ? ?/sec
take primitive run logical len: 1024, physical len: 512, indices: 1024    1.00     15.3±0.01µs        ? ?/sec    1.00     15.4±0.01µs        ? ?/sec
take str 1024                                                             1.00      8.1±0.01µs        ? ?/sec    1.00      8.1±0.01µs        ? ?/sec
take str 512                                                              1.00      3.7±0.00µs        ? ?/sec    1.00      3.7±0.01µs        ? ?/sec
take str null indices 1024                                                1.00      4.3±0.05µs        ? ?/sec    1.02      4.4±0.05µs        ? ?/sec
take str null indices 512                                                 1.00      2.1±0.03µs        ? ?/sec    1.05      2.2±0.02µs        ? ?/sec
take str null values 1024                                                 1.00      4.9±0.05µs        ? ?/sec    1.00      4.9±0.07µs        ? ?/sec
take str null values null indices 1024                                    1.00      3.1±0.02µs        ? ?/sec    1.03      3.2±0.01µs        ? ?/sec
take stringview 1024                                                      1.00    734.9±0.87ns        ? ?/sec    1.24    909.3±1.85ns        ? ?/sec
take stringview 512                                                       1.00    470.3±1.98ns        ? ?/sec    1.18    555.4±1.93ns        ? ?/sec
take stringview null indices 1024                                         1.00    914.6±1.57ns        ? ?/sec    1.02    931.3±2.54ns        ? ?/sec
take stringview null indices 512                                          1.00    563.6±0.97ns        ? ?/sec    1.00    561.2±2.20ns        ? ?/sec
take stringview null values 1024                                          1.00   1717.2±1.12ns        ? ?/sec    1.11   1902.8±2.78ns        ? ?/sec
take stringview null values null indices 1024                             1.00   1641.7±1.95ns        ? ?/sec    1.05   1726.1±3.21ns        ? ?/sec
take_record_batch 5 primitive cols no nulls 1024                          1.05      3.4±0.01µs        ? ?/sec    1.00      3.2±0.01µs        ? ?/sec
take_record_batch 6 string/binary cols no nulls 1024                      1.00     27.1±0.05µs        ? ?/sec    1.01     27.5±0.06µs        ? ?/sec
take_record_batch 7 mixed cols null values 1024                           1.02     17.6±0.10µs        ? ?/sec    1.00     17.2±0.37µs        ? ?/sec
take_record_batch 7 mixed cols null values null indices 1024              1.00     14.0±0.06µs        ? ?/sec    1.03     14.4±0.14µs        ? ?/sec

Resource Usage

base (merge-base)

Metric	Value
Wall time	410.1s
Peak memory	14.8 MiB
Avg memory	10.8 MiB
CPU user	406.3s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	410.1s
Peak memory	15.2 MiB
Avg memory	12.9 MiB
CPU user	407.5s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

Jefffrey · 2026-07-30T22:54:36Z

seems to be quite some real regressions?

Rich-T-kid · 2026-07-31T04:29:47Z

yea the benchmarks do not look good. even the targeted take_record_batch benchmarks do not show good results.

my understanding of rust's monomorphization may be incorrect. I'll keep looking into it

github-actions Bot added the arrow Changes to the arrow crate label Jul 27, 2026

Rich-T-kid mentioned this pull request Jul 27, 2026

benchmark/perf: add wide column take benchmarks #10442

Merged

avoided repeated bounds checks

1e0e5a4

Rich-T-kid force-pushed the rich-t-kid/avoid-take-bounds-checks branch from 6b46564 to 1e0e5a4 Compare July 27, 2026 02:28

Jefffrey reviewed Jul 27, 2026

View reviewed changes

Comment thread arrow-select/src/take.rs

alamb changed the title ~~perf : avoided repeated bounds checks~~ perf : avoided repeated bounds checks in take kernel Jul 27, 2026

This comment was marked as outdated.

Sign in to view

This comment was marked as duplicate.

Sign in to view

This comment was marked as outdated.

Sign in to view

Rich-T-kid force-pushed the rich-t-kid/avoid-take-bounds-checks branch from 5656bd2 to c27953d Compare July 27, 2026 19:05

replace get bounds checks with unsafe blocks

1a626c5

Rich-T-kid force-pushed the rich-t-kid/avoid-take-bounds-checks branch from c27953d to 1a626c5 Compare July 27, 2026 19:06

Jefffrey reviewed Jul 28, 2026

View reviewed changes

This comment was marked as duplicate.

Sign in to view

alamb reviewed Jul 28, 2026

View reviewed changes

Merge branch 'main' into rich-t-kid/avoid-take-bounds-checks

7c9c165

This comment was marked as duplicate.

Sign in to view

rich-T-kid added 5 commits July 29, 2026 19:22

introduce unsafe alternative for

5887b5b

checkmark

d9fe5cf

fix git

099b4c0

branch on const bool

e550818

wire up take_record_batch to use take_unchecked

6181df3

Rich-T-kid force-pushed the rich-t-kid/avoid-take-bounds-checks branch from 6dc9944 to 6181df3 Compare July 30, 2026 17:09

update all callsites

2947869

Merge branch 'main' into rich-t-kid/avoid-take-bounds-checks

ffc18f7

github-actions Bot added the arrow-select label Jul 30, 2026

This comment was marked as duplicate.

Sign in to view

Uh oh!

Conversation

Rich-T-kid commented Jul 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Are these changes tested?

Are there any user-facing changes?

Uh oh!

Uh oh!

Rich-T-kid commented Jul 27, 2026

Uh oh!

alamb commented Jul 27, 2026

Uh oh!

This comment was marked as outdated.

This comment was marked as duplicate.

This comment was marked as outdated.

Rich-T-kid commented Jul 27, 2026

Uh oh!

Rich-T-kid commented Jul 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Jefffrey Jul 28, 2026

Choose a reason for hiding this comment

Uh oh!

alamb Jul 28, 2026

Choose a reason for hiding this comment

Uh oh!

Rich-T-kid commented Jul 28, 2026

Uh oh!

Rich-T-kid commented Jul 28, 2026

Uh oh!

Jefffrey commented Jul 28, 2026

Uh oh!

This comment was marked as duplicate.

adriangbot commented Jul 28, 2026

Uh oh!

alamb left a comment

Choose a reason for hiding this comment

Uh oh!

alamb commented Jul 28, 2026

Uh oh!

Rich-T-kid commented Jul 28, 2026

Uh oh!

Rich-T-kid commented Jul 28, 2026

Uh oh!

Jefffrey commented Jul 28, 2026

Uh oh!

This comment was marked as duplicate.

adriangbot commented Jul 28, 2026

Uh oh!

Rich-T-kid commented Jul 30, 2026

Uh oh!

Jefffrey commented Jul 30, 2026

Uh oh!

This comment was marked as duplicate.

adriangbot commented Jul 30, 2026

Uh oh!

Jefffrey commented Jul 30, 2026

Uh oh!

This comment was marked as duplicate.

adriangbot commented Jul 30, 2026

Uh oh!

Jefffrey commented Jul 30, 2026

Uh oh!

Rich-T-kid commented Jul 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Rich-T-kid commented Jul 27, 2026 •

edited

Loading

Rich-T-kid commented Jul 27, 2026 •

edited

Loading