perf(arrow-arith): vectorize checked decimal `add`, `sub` and `mul` by mbutrovich · Pull Request #12062 · apache/arrow-rs

mbutrovich · 2026-11-03T21:35:09Z

Which issue does this PR close?

N/A.

Rationale for this change

decimal_op runs decimal add, sub and mul, including add_wrapping, sub_wrapping and mul_wrapping, through try_op! with the checked operations, which return at the first overflow. DataFusion calls add_wrapping for + (binary.rs), so its default decimal arithmetic runs one row at a time. #12058 does the same change for integers.

What changes are included in this PR?

Decimal add, sub and mul compute the wrapping result and an overflow flag for every slot in one pass, with equal scales and when one side is rescaled.
If a slot overflowed and the result has nulls, they recompute the overflow of every row into a bitmap and mask it with the validity bits. They rerun the existing checked kernel only if a valid row overflowed, so errors don't change.
Decimal32 and Decimal64 add and sub vectorize. Decimal128 stays scalar and loses the per-row branch and the operand spills of the error path. i128 mul multiplies in i64 when both operands fit. Decimal64 mul uses overflowing_mul, as in perf(arrow-arith): vectorize checked integer add, sub and mul #12058.
Decimal256 keeps the checked kernel. With -C target-cpu=x86-64-v3, LLVM vectorizes the one-pass i256 add loop two rows at a time, packing the carries into vector registers with about 20 stack spills per iteration, and it measured 0.75x (perf(arrow-arith): vectorize checked decimal add, sub and mul #12062 (comment)).
With unequal scales, only the operand with the smaller scale is multiplied, because the other multiplier is one. A scalar with the smaller scale is multiplied once, before the loop.
OverflowingOp, overflowing_op and collect_overflowing are the same as in perf(arrow-arith): vectorize checked integer add, sub and mul #12058, so whichever PR lands second drops its copy.
collect_overflowing writes into the Vec's spare capacity and calls set_len with the number of slots it wrote. With collect, LLVM sometimes left the loop inside collect out of line, where it loaded and stored the overflow flag on every row. Which instances it left out of line changed with the closure and the target: on generic aarch64 and neoverse-n1 it was rescaled Decimal128 add, which the bot measured at 0.83x on Neoverse-V2. Writing the values in our own loop without unsafe cost 30% to 40% with a zero-filled Vec, lost vectorization with push, and was slower in a two-pass, unzip or chunked form.
The output is reserved with Vec::try_reserve_exact, so an allocation failure returns ArrowError::MemoryError.
Null slots of the result hold the wrapping result instead of zero, as they already do for integer add_wrapping.
362eb22 adds the benchmark cases, so the benchmark bot can compare against it.

Are these changes tested?

New tests in arrow-arith/src/numeric.rs compare the kernels with the scalar checked operations at the edge values of every decimal type, with equal scales and with either side rescaled, as arrays, as one-row slices, and with a scalar on either side. They also cover Decimal128 mul on both sides of the i64 bounds. Another test moves one overflowing row through a 100-row batch of every decimal type, including rows where only the rescale overflows. A third checks a rescaled scalar, including one whose rescale overflows, against valid, all-null, and empty arrays. Miri passes these tests, except the position test, which checks vectorized code that Miri doesn't run.

The overflow cases are expected to be slower, because they pay for the new pass, the null-slot check, and then the checked kernel. With fail_on_overflow or ANSI mode, an overflow fails the query, so that cost is paid at most once per partition. A batch that overflows only in null slots pays for the pass and the null-slot check: decimal128_add_null_slot_overflow measured 0.88x on Neoverse-V2 and 0.86x on Xeon 8581C, because Decimal128 doesn't vectorize, and 1.45x with x86-64-v3.

The extra pass also costs code size: 166 KiB for decimal_op and the new helpers across the four decimal types, against 30 KiB before, in the thin-LTO bench binary on aarch64. The checked kernels that the fallback calls are unchanged.

decimal128_multiply_wide(0.1), where half the values don't fit in i64, measured 0.89x on Neoverse-V2. On that input the i64 branch in i128 mul goes either way at random, and the one pass also multiplies the null slots that the checked kernel skips.

Bot runs: Neoverse-V2 default and aligned, Xeon 8581C default and aligned, and Xeon 8581C with -C target-cpu=x86-64-v3 default and aligned. Those runs measured 41647fd. f933cfa changes only the i64 multiply, and its Decimal64 mul runs are: Neoverse-V2 default and aligned, Xeon 8581C default and aligned, and x86-64-v3 default and aligned.

Are there any significant user-facing changes?

No.

AI Disclosure

I wrote this PR with help from Claude but understand and support the core approach, implementation, and test coverage.

mbutrovich · 2026-11-03T21:35:39Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"

adriangbot · 2026-11-03T21:35:45Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

mbutrovich · 2026-11-03T21:35:46Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
  RUSTFLAGS: "-C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6"

mbutrovich · 2026-11-03T21:35:51Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
resources:
  arch: amd64

adriangbot · 2026-11-03T21:35:52Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

mbutrovich · 2026-11-03T21:35:56Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
  RUSTFLAGS: "-C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6"
resources:
  arch: amd64

adriangbot · 2026-11-03T21:35:56Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-03T21:36:00Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

mbutrovich · 2026-11-03T21:36:04Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
  RUSTFLAGS: "-C target-cpu=x86-64-v3"
resources:
  arch: amd64

adriangbot · 2026-11-03T21:36:08Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C target-cpu=x86-64-v3
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

mbutrovich · 2026-11-03T21:36:09Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
  RUSTFLAGS: "-C target-cpu=x86-64-v3 -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6"
resources:
  arch: amd64

adriangbot · 2026-11-03T21:47:21Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
shards: 1

Details

group                                base                                   changed
-----                                ----                                   -------
decimal128_add(0)                    1.01      8.5±0.01µs        ? ?/sec    1.00      8.4±0.00µs        ? ?/sec
decimal128_add(0.1)                  1.82     15.4±0.13µs        ? ?/sec    1.00      8.5±0.02µs        ? ?/sec
decimal128_add_null_slot_overflow    1.00     15.2±0.22µs        ? ?/sec    1.48     22.5±1.26µs        ? ?/sec
decimal128_add_overflow              1.00      9.2±0.04µs        ? ?/sec    1.80     16.6±0.07µs        ? ?/sec
decimal128_add_rescale(0)            1.00     16.7±0.01µs        ? ?/sec    1.20     20.0±0.02µs        ? ?/sec
decimal128_add_rescale(0.1)          1.00     20.2±0.25µs        ? ?/sec    1.00     20.2±0.03µs        ? ?/sec
decimal128_divide(0)                 1.00     51.7±0.03µs        ? ?/sec    1.00     51.7±0.09µs        ? ?/sec
decimal128_divide(0.1)               1.00     53.7±0.46µs        ? ?/sec    1.00     53.6±0.38µs        ? ?/sec
decimal128_multiply(0)               2.27     24.5±0.02µs        ? ?/sec    1.00     10.8±0.02µs        ? ?/sec
decimal128_multiply(0.1)             1.44     15.7±0.21µs        ? ?/sec    1.00     10.9±0.04µs        ? ?/sec
decimal128_subtract(0)               1.00      8.5±0.00µs        ? ?/sec    1.00      8.5±0.02µs        ? ?/sec
decimal128_subtract(0.1)             1.79     15.3±0.20µs        ? ?/sec    1.00      8.5±0.03µs        ? ?/sec
decimal256_add(0)                    1.66     20.1±1.25µs        ? ?/sec    1.00     12.1±0.02µs        ? ?/sec
decimal256_add(0.1)                  1.81     22.2±2.02µs        ? ?/sec    1.00     12.2±0.03µs        ? ?/sec
decimal256_multiply(0)               1.07     69.9±0.04µs        ? ?/sec    1.00     65.1±0.15µs        ? ?/sec
decimal256_multiply(0.1)             1.16     72.6±0.04µs        ? ?/sec    1.00     62.6±0.06µs        ? ?/sec
decimal64_add(0)                     3.42      8.5±0.02µs        ? ?/sec    1.00      2.5±0.00µs        ? ?/sec
decimal64_add(0.1)                   5.92     15.2±0.12µs        ? ?/sec    1.00      2.6±0.01µs        ? ?/sec
decimal64_multiply(0)                1.27      8.5±0.02µs        ? ?/sec    1.00      6.7±0.11µs        ? ?/sec
decimal64_multiply(0.1)              2.20     15.2±0.11µs        ? ?/sec    1.00      6.9±0.13µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-oclc6q2e-df121e2a-ln85

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

uname:

Linux bench-c6616002276-4385-qdv7t 6.12.105+ #1 SMP Sat Sep 19 07:55:25 UTC 2026 aarch64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

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

Resource Usage

base (merge-base)

Metric	Value
Wall time	195.0s
Peak memory	14.4 MiB
Avg memory	13.5 MiB
CPU user	190.0s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	195.0s
Peak memory	13.8 MiB
Avg memory	12.9 MiB
CPU user	191.2s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-03T21:47:25Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C target-cpu=x86-64-v3 -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-03T21:48:19Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
shards: 1

Details

group                                base                                   changed
-----                                ----                                   -------
decimal128_add(0)                    1.01      8.5±0.01µs        ? ?/sec    1.00      8.4±0.01µs        ? ?/sec
decimal128_add(0.1)                  1.81     15.4±0.16µs        ? ?/sec    1.00      8.5±0.01µs        ? ?/sec
decimal128_add_null_slot_overflow    1.00     15.1±0.19µs        ? ?/sec    1.53     23.1±0.79µs        ? ?/sec
decimal128_add_overflow              1.00      9.2±0.01µs        ? ?/sec    1.80     16.6±0.01µs        ? ?/sec
decimal128_add_rescale(0)            1.00     16.7±0.02µs        ? ?/sec    1.20     20.0±0.04µs        ? ?/sec
decimal128_add_rescale(0.1)          1.00     20.1±0.34µs        ? ?/sec    1.00     20.1±0.02µs        ? ?/sec
decimal128_divide(0)                 1.00     51.7±0.03µs        ? ?/sec    1.00     51.7±0.03µs        ? ?/sec
decimal128_divide(0.1)               1.00     53.8±0.49µs        ? ?/sec    1.00     53.7±0.46µs        ? ?/sec
decimal128_multiply(0)               2.18     24.6±0.02µs        ? ?/sec    1.00     11.3±0.14µs        ? ?/sec
decimal128_multiply(0.1)             1.44     15.7±0.19µs        ? ?/sec    1.00     10.9±0.01µs        ? ?/sec
decimal128_subtract(0)               1.00      8.5±0.01µs        ? ?/sec    1.00      8.5±0.03µs        ? ?/sec
decimal128_subtract(0.1)             1.79     15.3±0.23µs        ? ?/sec    1.00      8.5±0.01µs        ? ?/sec
decimal256_add(0)                    1.66     20.2±1.33µs        ? ?/sec    1.00     12.2±0.06µs        ? ?/sec
decimal256_add(0.1)                  1.82     22.2±2.10µs        ? ?/sec    1.00     12.2±0.01µs        ? ?/sec
decimal256_multiply(0)               1.08     69.9±0.06µs        ? ?/sec    1.00     65.0±0.10µs        ? ?/sec
decimal256_multiply(0.1)             1.16     72.5±0.04µs        ? ?/sec    1.00     62.6±0.10µs        ? ?/sec
decimal64_add(0)                     3.40      8.5±0.02µs        ? ?/sec    1.00      2.5±0.01µs        ? ?/sec
decimal64_add(0.1)                   5.92     15.2±0.12µs        ? ?/sec    1.00      2.6±0.00µs        ? ?/sec
decimal64_multiply(0)                1.28      8.5±0.01µs        ? ?/sec    1.00      6.6±0.01µs        ? ?/sec
decimal64_multiply(0.1)              2.26     15.2±0.12µs        ? ?/sec    1.00      6.7±0.02µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-oclc6q2e-64c98ba8-w28v

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

uname:

Linux bench-c6616004420-4386-5sjw9 6.12.105+ #1 SMP Sat Sep 19 07:55:25 UTC 2026 aarch64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

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

Resource Usage

base (merge-base)

Metric	Value
Wall time	195.0s
Peak memory	15.7 MiB
Avg memory	14.8 MiB
CPU user	189.9s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	200.0s
Peak memory	15.0 MiB
Avg memory	14.0 MiB
CPU user	194.2s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-03T21:48:23Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C target-cpu=x86-64-v3
resources:
  arch: amd64
shards: 1

Details

group                                base                                   changed
-----                                ----                                   -------
decimal128_add(0)                    1.94     12.7±0.79µs        ? ?/sec    1.00      6.6±0.22µs        ? ?/sec
decimal128_add(0.1)                  2.53     17.1±0.76µs        ? ?/sec    1.00      6.7±0.21µs        ? ?/sec
decimal128_add_null_slot_overflow    1.00     18.0±0.54µs        ? ?/sec    1.42     25.5±1.07µs        ? ?/sec
decimal128_add_overflow              1.00     14.3±0.50µs        ? ?/sec    1.36     19.4±0.76µs        ? ?/sec
decimal128_add_rescale(0)            1.62     39.7±1.70µs        ? ?/sec    1.00     24.5±0.69µs        ? ?/sec
decimal128_add_rescale(0.1)          1.95     48.2±1.98µs        ? ?/sec    1.00     24.7±0.82µs        ? ?/sec
decimal128_divide(0)                 1.00    104.0±4.77µs        ? ?/sec    1.02    106.2±4.53µs        ? ?/sec
decimal128_divide(0.1)               1.01    101.1±5.61µs        ? ?/sec    1.00    100.5±4.70µs        ? ?/sec
decimal128_multiply(0)               2.07     21.0±1.08µs        ? ?/sec    1.00     10.2±0.47µs        ? ?/sec
decimal128_multiply(0.1)             2.62     27.3±1.22µs        ? ?/sec    1.00     10.4±0.37µs        ? ?/sec
decimal128_subtract(0)               1.98     13.0±0.59µs        ? ?/sec    1.00      6.6±0.24µs        ? ?/sec
decimal128_subtract(0.1)             2.54     17.3±0.77µs        ? ?/sec    1.00      6.8±0.14µs        ? ?/sec
decimal256_add(0)                    1.00     22.7±1.09µs        ? ?/sec    1.04     23.5±2.10µs        ? ?/sec
decimal256_add(0.1)                  1.28     30.6±1.03µs        ? ?/sec    1.00     23.9±1.93µs        ? ?/sec
decimal256_multiply(0)               1.00    106.5±6.41µs        ? ?/sec    1.02    108.9±2.53µs        ? ?/sec
decimal256_multiply(0.1)             1.00    101.3±4.68µs        ? ?/sec    1.05    106.2±2.71µs        ? ?/sec
decimal64_add(0)                     3.74      9.7±0.43µs        ? ?/sec    1.00      2.6±0.13µs        ? ?/sec
decimal64_add(0.1)                   3.88     10.7±1.18µs        ? ?/sec    1.00      2.8±0.13µs        ? ?/sec
decimal64_multiply(0)                1.10      9.4±0.50µs        ? ?/sec    1.00      8.5±0.36µs        ? ?/sec
decimal64_multiply(0.1)              1.23     11.0±1.11µs        ? ?/sec    1.00      9.0±0.16µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-11j2ashw-1b098063-44mm

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6616009972-4389-t6k59 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	195.0s
Peak memory	15.0 MiB
Avg memory	13.9 MiB
CPU user	188.9s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	195.0s
Peak memory	14.8 MiB
Avg memory	13.6 MiB
CPU user	191.2s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-03T21:48:37Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
resources:
  arch: amd64
shards: 1

Details

group                                base                                   changed
-----                                ----                                   -------
decimal128_add(0)                    2.02     11.4±0.40µs        ? ?/sec    1.00      5.7±0.09µs        ? ?/sec
decimal128_add(0.1)                  2.57     14.9±0.29µs        ? ?/sec    1.00      5.8±0.09µs        ? ?/sec
decimal128_add_null_slot_overflow    1.00     15.1±0.35µs        ? ?/sec    1.39     21.0±0.54µs        ? ?/sec
decimal128_add_overflow              1.00     12.2±0.21µs        ? ?/sec    1.38     16.9±0.26µs        ? ?/sec
decimal128_add_rescale(0)            1.89     40.6±1.25µs        ? ?/sec    1.00     21.5±0.54µs        ? ?/sec
decimal128_add_rescale(0.1)          2.11     45.5±1.67µs        ? ?/sec    1.00     21.5±0.27µs        ? ?/sec
decimal128_divide(0)                 1.00     97.6±3.43µs        ? ?/sec    1.00     97.2±2.72µs        ? ?/sec
decimal128_divide(0.1)               1.00     91.8±1.05µs        ? ?/sec    1.00     92.1±1.98µs        ? ?/sec
decimal128_multiply(0)               2.41     22.3±0.65µs        ? ?/sec    1.00      9.2±0.15µs        ? ?/sec
decimal128_multiply(0.1)             2.81     25.9±0.60µs        ? ?/sec    1.00      9.2±0.31µs        ? ?/sec
decimal128_subtract(0)               1.99     11.4±0.33µs        ? ?/sec    1.00      5.7±0.21µs        ? ?/sec
decimal128_subtract(0.1)             2.58     14.9±0.33µs        ? ?/sec    1.00      5.8±0.09µs        ? ?/sec
decimal256_add(0)                    2.08     20.9±1.27µs        ? ?/sec    1.00     10.0±0.17µs        ? ?/sec
decimal256_add(0.1)                  2.56     26.2±1.66µs        ? ?/sec    1.00     10.2±0.21µs        ? ?/sec
decimal256_multiply(0)               1.02     93.9±2.54µs        ? ?/sec    1.00     92.1±3.07µs        ? ?/sec
decimal256_multiply(0.1)             1.00     89.3±2.34µs        ? ?/sec    1.00     89.1±2.36µs        ? ?/sec
decimal64_add(0)                     3.15      8.5±0.14µs        ? ?/sec    1.00      2.7±0.21µs        ? ?/sec
decimal64_add(0.1)                   3.36      9.2±0.91µs        ? ?/sec    1.00      2.8±0.09µs        ? ?/sec
decimal64_multiply(0)                1.32      8.5±0.19µs        ? ?/sec    1.00      6.5±0.33µs        ? ?/sec
decimal64_multiply(0.1)              1.42      9.3±0.91µs        ? ?/sec    1.00      6.5±0.09µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-11j2ashw-5b2a8cd0-98fd

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6616005830-4387-xn8p5 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	195.0s
Peak memory	15.2 MiB
Avg memory	14.2 MiB
CPU user	190.1s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	200.1s
Peak memory	14.9 MiB
Avg memory	13.7 MiB
CPU user	197.2s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-03T21:48:38Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Details

group                                base                                   changed
-----                                ----                                   -------
decimal128_add(0)                    1.42      7.8±0.14µs        ? ?/sec    1.00      5.5±0.07µs        ? ?/sec
decimal128_add(0.1)                  1.97     10.9±0.11µs        ? ?/sec    1.00      5.5±0.05µs        ? ?/sec
decimal128_add_null_slot_overflow    1.00     10.9±0.25µs        ? ?/sec    1.52     16.6±0.27µs        ? ?/sec
decimal128_add_overflow              1.00      9.0±0.03µs        ? ?/sec    1.51     13.6±0.19µs        ? ?/sec
decimal128_add_rescale(0)            2.23     44.7±1.07µs        ? ?/sec    1.00     20.1±0.25µs        ? ?/sec
decimal128_add_rescale(0.1)          2.06     41.8±0.50µs        ? ?/sec    1.00     20.3±0.19µs        ? ?/sec
decimal128_divide(0)                 1.01     89.1±0.77µs        ? ?/sec    1.00     88.2±0.78µs        ? ?/sec
decimal128_divide(0.1)               1.00     84.3±0.72µs        ? ?/sec    1.00     84.2±0.78µs        ? ?/sec
decimal128_multiply(0)               2.60     23.1±1.38µs        ? ?/sec    1.00      8.9±0.09µs        ? ?/sec
decimal128_multiply(0.1)             2.67     24.1±0.36µs        ? ?/sec    1.00      9.0±0.22µs        ? ?/sec
decimal128_subtract(0)               1.38      7.7±0.07µs        ? ?/sec    1.00      5.6±0.10µs        ? ?/sec
decimal128_subtract(0.1)             1.98     11.0±0.16µs        ? ?/sec    1.00      5.6±0.06µs        ? ?/sec
decimal256_add(0)                    1.36     14.3±0.12µs        ? ?/sec    1.00     10.5±0.11µs        ? ?/sec
decimal256_add(0.1)                  1.80     19.7±0.34µs        ? ?/sec    1.00     10.9±0.11µs        ? ?/sec
decimal256_multiply(0)               1.00     86.0±1.20µs        ? ?/sec    1.03     88.2±1.26µs        ? ?/sec
decimal256_multiply(0.1)             1.00     84.6±1.06µs        ? ?/sec    1.01     85.5±0.82µs        ? ?/sec
decimal64_add(0)                     2.72      6.8±0.10µs        ? ?/sec    1.00      2.5±0.03µs        ? ?/sec
decimal64_add(0.1)                   3.08      8.3±0.86µs        ? ?/sec    1.00      2.7±0.02µs        ? ?/sec
decimal64_multiply(0)                1.08      6.7±0.05µs        ? ?/sec    1.00      6.2±0.05µs        ? ?/sec
decimal64_multiply(0.1)              1.32      8.3±0.90µs        ? ?/sec    1.00      6.3±0.14µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-11j2ashw-69d831de-tpgp

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6616007503-4388-cfx5g 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	205.1s
Peak memory	16.4 MiB
Avg memory	15.6 MiB
CPU user	203.1s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	210.0s
Peak memory	16.0 MiB
Avg memory	14.6 MiB
CPU user	206.2s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-03T21:59:10Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (c5774ac) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C target-cpu=x86-64-v3 -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Details

group                                base                                   changed
-----                                ----                                   -------
decimal128_add(0)                    2.09     13.6±0.50µs        ? ?/sec    1.00      6.5±0.31µs        ? ?/sec
decimal128_add(0.1)                  2.63     18.0±0.84µs        ? ?/sec    1.00      6.8±0.14µs        ? ?/sec
decimal128_add_null_slot_overflow    1.00     17.9±0.65µs        ? ?/sec    1.38     24.8±1.19µs        ? ?/sec
decimal128_add_overflow              1.00     14.6±0.34µs        ? ?/sec    1.37     20.0±0.55µs        ? ?/sec
decimal128_add_rescale(0)            1.77     41.9±1.58µs        ? ?/sec    1.00     23.6±1.17µs        ? ?/sec
decimal128_add_rescale(0.1)          2.12     50.6±1.62µs        ? ?/sec    1.00     23.9±1.16µs        ? ?/sec
decimal128_divide(0)                 1.07    111.9±1.65µs        ? ?/sec    1.00    104.3±6.30µs        ? ?/sec
decimal128_divide(0.1)               1.06    106.2±3.83µs        ? ?/sec    1.00     99.8±4.60µs        ? ?/sec
decimal128_multiply(0)               2.10     21.6±0.37µs        ? ?/sec    1.00     10.3±0.41µs        ? ?/sec
decimal128_multiply(0.1)             2.63     27.9±1.36µs        ? ?/sec    1.00     10.6±0.31µs        ? ?/sec
decimal128_subtract(0)               2.16     13.6±0.37µs        ? ?/sec    1.00      6.3±0.33µs        ? ?/sec
decimal128_subtract(0.1)             2.63     18.0±0.29µs        ? ?/sec    1.00      6.8±0.11µs        ? ?/sec
decimal256_add(0)                    1.01     23.8±0.45µs        ? ?/sec    1.00     23.5±1.99µs        ? ?/sec
decimal256_add(0.1)                  1.38     31.2±0.99µs        ? ?/sec    1.00     22.5±2.04µs        ? ?/sec
decimal256_multiply(0)               1.03    112.2±3.98µs        ? ?/sec    1.00    108.4±3.71µs        ? ?/sec
decimal256_multiply(0.1)             1.01    105.6±1.57µs        ? ?/sec    1.00    104.1±4.22µs        ? ?/sec
decimal64_add(0)                     3.71     10.1±0.20µs        ? ?/sec    1.00      2.7±0.07µs        ? ?/sec
decimal64_add(0.1)                   3.65     10.0±1.01µs        ? ?/sec    1.00      2.7±0.13µs        ? ?/sec
decimal64_multiply(0)                1.14     10.1±0.20µs        ? ?/sec    1.00      8.9±0.14µs        ? ?/sec
decimal64_multiply(0.1)              1.25     10.4±0.98µs        ? ?/sec    1.00      8.3±0.45µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-11j2ashw-4803721b-kr8v

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6616011753-4390-d5h9z 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	190.0s
Peak memory	16.3 MiB
Avg memory	15.3 MiB
CPU user	185.9s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	195.0s
Peak memory	15.7 MiB
Avg memory	14.6 MiB
CPU user	193.2s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

mbutrovich · 2026-11-04T00:01:59Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"

mbutrovich · 2026-11-04T00:02:03Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
  RUSTFLAGS: "-C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6"

adriangbot · 2026-11-04T00:02:03Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (cffdb05) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

mbutrovich · 2026-11-04T00:02:08Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
resources:
  arch: amd64

adriangbot · 2026-11-04T00:02:08Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (cffdb05) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

mbutrovich · 2026-11-04T00:02:13Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
  RUSTFLAGS: "-C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6"
resources:
  arch: amd64

adriangbot · 2026-11-04T00:02:13Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (cffdb05) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-04T00:02:18Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (cffdb05) to 5cf1987 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

mbutrovich · 2026-11-04T00:02:18Z

run benchmark arithmetic_kernels

baseline:
  ref: 5cf1987a91
env:
  BENCH_FILTER: "^decimal"
  RUSTFLAGS: "-C target-cpu=x86-64-v3"
resources:
  arch: amd64

adriangbot · 2026-11-13T19:50:19Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (41647fd) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C target-cpu=x86-64-v3 -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-13T19:50:42Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (41647fd) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal
shards: 1

Details

group                                 base                                   changed
-----                                 ----                                   -------
decimal128_add(0)                     1.00      8.5±0.03µs        ? ?/sec    1.00      8.5±0.00µs        ? ?/sec
decimal128_add(0.1)                   1.55     13.4±0.22µs        ? ?/sec    1.00      8.7±0.01µs        ? ?/sec
decimal128_add_null_slot_overflow     1.00     15.0±0.19µs        ? ?/sec    1.13     17.0±0.01µs        ? ?/sec
decimal128_add_overflow               1.00      9.2±0.01µs        ? ?/sec    1.82     16.7±0.01µs        ? ?/sec
decimal128_add_rescale(0)             1.11     16.8±0.04µs        ? ?/sec    1.00     15.1±0.05µs        ? ?/sec
decimal128_add_rescale(0.1)           1.20     17.9±0.23µs        ? ?/sec    1.00     14.8±0.01µs        ? ?/sec
decimal128_add_rescale_scalar(0)      4.52     34.2±3.81µs        ? ?/sec    1.00      7.6±0.01µs        ? ?/sec
decimal128_add_rescale_scalar(0.1)    4.09     30.9±0.48µs        ? ?/sec    1.00      7.6±0.12µs        ? ?/sec
decimal128_divide(0)                  1.00     51.5±0.04µs        ? ?/sec    1.00     51.5±0.02µs        ? ?/sec
decimal128_divide(0.1)                1.00     48.0±0.35µs        ? ?/sec    1.00     48.1±0.40µs        ? ?/sec
decimal128_multiply(0)                2.23     24.6±0.02µs        ? ?/sec    1.00     11.0±0.01µs        ? ?/sec
decimal128_multiply(0.1)              1.27     14.2±0.13µs        ? ?/sec    1.00     11.1±0.01µs        ? ?/sec
decimal128_multiply_wide(0)           1.18     24.5±0.10µs        ? ?/sec    1.00     20.8±0.01µs        ? ?/sec
decimal128_multiply_wide(0.1)         1.00     17.7±0.01µs        ? ?/sec    1.12     19.8±0.01µs        ? ?/sec
decimal128_subtract(0)                1.00      8.5±0.01µs        ? ?/sec    1.00      8.5±0.01µs        ? ?/sec
decimal128_subtract(0.1)              1.55     13.5±0.23µs        ? ?/sec    1.00      8.7±0.04µs        ? ?/sec
decimal256_add(0)                     1.06     22.8±1.27µs        ? ?/sec    1.00     21.5±1.03µs        ? ?/sec
decimal256_add(0.1)                   1.00     20.7±1.52µs        ? ?/sec    1.00     20.8±1.79µs        ? ?/sec
decimal256_multiply(0)                1.00     70.8±0.36µs        ? ?/sec    1.00     71.1±0.31µs        ? ?/sec
decimal256_multiply(0.1)              1.00     67.2±0.06µs        ? ?/sec    1.00     67.2±0.09µs        ? ?/sec
decimal64_add(0)                      3.34      8.5±0.01µs        ? ?/sec    1.00      2.5±0.00µs        ? ?/sec
decimal64_add(0.1)                    5.03     13.7±0.09µs        ? ?/sec    1.00      2.7±0.00µs        ? ?/sec
decimal64_multiply(0)                 1.25      8.5±0.01µs        ? ?/sec    1.00      6.8±0.05µs        ? ?/sec
decimal64_multiply(0.1)               2.00     13.6±0.09µs        ? ?/sec    1.00      6.8±0.00µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-ez8gwx9u-a9459978-96wb

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

uname:

Linux bench-c6863033033-5021-2ds4r 6.12.105+ #1 SMP Sat Sep 19 07:55:25 UTC 2026 aarch64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

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

Resource Usage

base (merge-base)

Metric	Value
Wall time	225.1s
Peak memory	14.3 MiB
Avg memory	13.7 MiB
CPU user	223.0s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	240.1s
Peak memory	14.3 MiB
Avg memory	13.5 MiB
CPU user	235.3s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T19:51:47Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (41647fd) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Details

group                                 base                                   changed
-----                                 ----                                   -------
decimal128_add(0)                     1.40      8.0±0.11µs        ? ?/sec    1.00      5.7±0.10µs        ? ?/sec
decimal128_add(0.1)                   1.84     10.7±0.22µs        ? ?/sec    1.00      5.8±0.07µs        ? ?/sec
decimal128_add_null_slot_overflow     1.00     11.0±0.18µs        ? ?/sec    1.22     13.4±0.20µs        ? ?/sec
decimal128_add_overflow               1.00      9.2±0.19µs        ? ?/sec    1.51     13.9±0.23µs        ? ?/sec
decimal128_add_rescale(0)             4.00     47.5±0.98µs        ? ?/sec    1.00     11.9±0.21µs        ? ?/sec
decimal128_add_rescale(0.1)           3.24     38.8±0.59µs        ? ?/sec    1.00     12.0±0.15µs        ? ?/sec
decimal128_add_rescale_scalar(0)      10.50    51.3±0.89µs        ? ?/sec    1.00      4.9±0.12µs        ? ?/sec
decimal128_add_rescale_scalar(0.1)    10.48    50.8±0.85µs        ? ?/sec    1.00      4.8±0.09µs        ? ?/sec
decimal128_divide(0)                  1.01     91.2±1.62µs        ? ?/sec    1.00     90.7±1.30µs        ? ?/sec
decimal128_divide(0.1)                1.00     77.9±1.00µs        ? ?/sec    1.01     78.4±1.39µs        ? ?/sec
decimal128_multiply(0)                1.80     18.6±0.21µs        ? ?/sec    1.00     10.3±0.16µs        ? ?/sec
decimal128_multiply(0.1)              2.11     22.3±0.34µs        ? ?/sec    1.00     10.6±0.15µs        ? ?/sec
decimal128_multiply_wide(0)           1.10     19.5±0.30µs        ? ?/sec    1.00     17.7±0.50µs        ? ?/sec
decimal128_multiply_wide(0.1)         1.40     22.1±0.36µs        ? ?/sec    1.00     15.8±0.31µs        ? ?/sec
decimal128_subtract(0)                1.38      8.0±0.12µs        ? ?/sec    1.00      5.8±0.10µs        ? ?/sec
decimal128_subtract(0.1)              1.84     10.7±0.18µs        ? ?/sec    1.00      5.8±0.08µs        ? ?/sec
decimal256_add(0)                     1.00     15.8±0.24µs        ? ?/sec    1.01     16.0±0.29µs        ? ?/sec
decimal256_add(0.1)                   1.01     20.5±0.48µs        ? ?/sec    1.00     20.4±0.56µs        ? ?/sec
decimal256_multiply(0)                1.00     89.8±1.83µs        ? ?/sec    1.00     90.0±1.26µs        ? ?/sec
decimal256_multiply(0.1)              1.01     78.7±1.23µs        ? ?/sec    1.00     78.2±0.97µs        ? ?/sec
decimal64_add(0)                      2.28      5.9±0.98µs        ? ?/sec    1.00      2.6±0.09µs        ? ?/sec
decimal64_add(0.1)                    2.84      7.9±0.83µs        ? ?/sec    1.00      2.8±0.04µs        ? ?/sec
decimal64_multiply(0)                 1.00      6.0±1.01µs        ? ?/sec    1.06      6.3±0.12µs        ? ?/sec
decimal64_multiply(0.1)               1.20      7.8±0.87µs        ? ?/sec    1.00      6.5±0.13µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1x7drz99-ecac925a-lwl9

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6863037385-5024-8mfkp 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	240.1s
Peak memory	16.8 MiB
Avg memory	16.2 MiB
CPU user	238.1s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	235.1s
Peak memory	16.6 MiB
Avg memory	15.4 MiB
CPU user	232.3s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T19:51:51Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (41647fd) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal
resources:
  arch: amd64
shards: 1

Details

group                                 base                                   changed
-----                                 ----                                   -------
decimal128_add(0)                     1.40      7.8±0.04µs        ? ?/sec    1.00      5.6±0.01µs        ? ?/sec
decimal128_add(0.1)                   1.87     10.5±0.08µs        ? ?/sec    1.00      5.6±0.05µs        ? ?/sec
decimal128_add_null_slot_overflow     1.00     11.0±0.09µs        ? ?/sec    1.17     12.8±0.12µs        ? ?/sec
decimal128_add_overflow               1.00      9.1±0.06µs        ? ?/sec    1.46     13.3±0.12µs        ? ?/sec
decimal128_add_rescale(0)             3.25     37.1±0.31µs        ? ?/sec    1.00     11.4±0.04µs        ? ?/sec
decimal128_add_rescale(0.1)           3.32     38.2±0.85µs        ? ?/sec    1.00     11.5±0.08µs        ? ?/sec
decimal128_add_rescale_scalar(0)      10.77    51.0±0.51µs        ? ?/sec    1.00      4.7±0.04µs        ? ?/sec
decimal128_add_rescale_scalar(0.1)    10.60    48.4±0.30µs        ? ?/sec    1.00      4.6±0.03µs        ? ?/sec
decimal128_divide(0)                  1.00     89.8±0.33µs        ? ?/sec    1.01     90.4±0.77µs        ? ?/sec
decimal128_divide(0.1)                1.02     76.7±0.75µs        ? ?/sec    1.00     75.5±1.04µs        ? ?/sec
decimal128_multiply(0)                1.78     18.3±0.02µs        ? ?/sec    1.00     10.3±0.24µs        ? ?/sec
decimal128_multiply(0.1)              2.08     21.2±0.15µs        ? ?/sec    1.00     10.2±0.22µs        ? ?/sec
decimal128_multiply_wide(0)           1.14     18.3±0.03µs        ? ?/sec    1.00     16.1±0.42µs        ? ?/sec
decimal128_multiply_wide(0.1)         1.43     21.0±0.20µs        ? ?/sec    1.00     14.6±0.14µs        ? ?/sec
decimal128_subtract(0)                1.38      7.8±0.03µs        ? ?/sec    1.00      5.6±0.08µs        ? ?/sec
decimal128_subtract(0.1)              1.88     10.6±0.16µs        ? ?/sec    1.00      5.6±0.04µs        ? ?/sec
decimal256_add(0)                     1.02     15.5±0.05µs        ? ?/sec    1.00     15.2±0.10µs        ? ?/sec
decimal256_add(0.1)                   1.01     20.1±0.64µs        ? ?/sec    1.00     19.9±0.55µs        ? ?/sec
decimal256_multiply(0)                1.02     88.3±1.44µs        ? ?/sec    1.00     86.4±0.78µs        ? ?/sec
decimal256_multiply(0.1)              1.02     76.8±0.39µs        ? ?/sec    1.00     75.2±0.45µs        ? ?/sec
decimal64_add(0)                      2.28      5.8±0.96µs        ? ?/sec    1.00      2.5±0.02µs        ? ?/sec
decimal64_add(0.1)                    2.88      7.7±0.85µs        ? ?/sec    1.00      2.7±0.04µs        ? ?/sec
decimal64_multiply(0)                 1.00      5.8±0.96µs        ? ?/sec    1.07      6.2±0.03µs        ? ?/sec
decimal64_multiply(0.1)               1.24      7.7±0.82µs        ? ?/sec    1.00      6.2±0.05µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1x7drz99-ecac925a-cdz2

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6863035971-5023-r7mqq 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	245.1s
Peak memory	15.3 MiB
Avg memory	14.5 MiB
CPU user	237.8s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	245.1s
Peak memory	15.6 MiB
Avg memory	14.5 MiB
CPU user	240.3s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T19:52:39Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (41647fd) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C target-cpu=x86-64-v3
resources:
  arch: amd64
shards: 1

Details

group                                 base                                   changed
-----                                 ----                                   -------
decimal128_add(0)                     1.97     13.2±0.70µs        ? ?/sec    1.00      6.7±0.23µs        ? ?/sec
decimal128_add(0.1)                   2.57     17.1±0.67µs        ? ?/sec    1.00      6.6±0.25µs        ? ?/sec
decimal128_add_null_slot_overflow     1.45     17.1±0.65µs        ? ?/sec    1.00     11.8±0.58µs        ? ?/sec
decimal128_add_overflow               1.00     14.1±0.50µs        ? ?/sec    1.33     18.8±0.84µs        ? ?/sec
decimal128_add_rescale(0)             3.39     46.0±2.06µs        ? ?/sec    1.00     13.6±0.64µs        ? ?/sec
decimal128_add_rescale(0.1)           3.30     44.2±1.73µs        ? ?/sec    1.00     13.4±0.66µs        ? ?/sec
decimal128_add_rescale_scalar(0)      10.18    56.6±1.91µs        ? ?/sec    1.00      5.6±0.23µs        ? ?/sec
decimal128_add_rescale_scalar(0.1)    9.94     53.8±2.23µs        ? ?/sec    1.00      5.4±0.24µs        ? ?/sec
decimal128_divide(0)                  1.00    108.3±6.01µs        ? ?/sec    1.00    108.4±3.31µs        ? ?/sec
decimal128_divide(0.1)                1.03     91.7±2.96µs        ? ?/sec    1.00     89.0±3.80µs        ? ?/sec
decimal128_multiply(0)                1.90     20.8±0.72µs        ? ?/sec    1.00     10.9±0.82µs        ? ?/sec
decimal128_multiply(0.1)              2.37     24.7±1.05µs        ? ?/sec    1.00     10.4±0.47µs        ? ?/sec
decimal128_multiply_wide(0)           1.11     21.3±0.95µs        ? ?/sec    1.00     19.2±1.77µs        ? ?/sec
decimal128_multiply_wide(0.1)         1.42     24.5±0.93µs        ? ?/sec    1.00     17.2±0.81µs        ? ?/sec
decimal128_subtract(0)                2.03     13.6±0.74µs        ? ?/sec    1.00      6.7±0.23µs        ? ?/sec
decimal128_subtract(0.1)              2.55     16.8±0.82µs        ? ?/sec    1.00      6.6±0.23µs        ? ?/sec
decimal256_add(0)                     1.00     26.3±1.69µs        ? ?/sec    1.01     26.6±1.35µs        ? ?/sec
decimal256_add(0.1)                   1.03     31.6±1.21µs        ? ?/sec    1.00     30.6±1.29µs        ? ?/sec
decimal256_multiply(0)                1.00    103.2±4.47µs        ? ?/sec    1.00    103.7±4.00µs        ? ?/sec
decimal256_multiply(0.1)              1.06     93.8±3.06µs        ? ?/sec    1.00     88.3±3.78µs        ? ?/sec
decimal64_add(0)                      3.39      9.1±1.38µs        ? ?/sec    1.00      2.7±0.10µs        ? ?/sec
decimal64_add(0.1)                    4.23     10.5±0.92µs        ? ?/sec    1.00      2.5±0.10µs        ? ?/sec
decimal64_multiply(0)                 1.00      9.1±1.33µs        ? ?/sec    1.01      9.2±0.30µs        ? ?/sec
decimal64_multiply(0.1)               1.14     10.5±0.89µs        ? ?/sec    1.00      9.2±0.37µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1x7drz99-4b7c01e4-6sl9

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6863038661-5025-2j8w9 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	230.1s
Peak memory	15.6 MiB
Avg memory	14.9 MiB
CPU user	226.0s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	240.1s
Peak memory	15.6 MiB
Avg memory	14.4 MiB
CPU user	234.3s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T20:01:24Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (41647fd) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal
  RUSTFLAGS: -C target-cpu=x86-64-v3 -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Details

group                                 base                                   changed
-----                                 ----                                   -------
decimal128_add(0)                     1.43      8.0±0.15µs        ? ?/sec    1.00      5.6±0.05µs        ? ?/sec
decimal128_add(0.1)                   1.87     10.6±0.12µs        ? ?/sec    1.00      5.7±0.12µs        ? ?/sec
decimal128_add_null_slot_overflow     1.01     11.0±0.19µs        ? ?/sec    1.00     10.8±0.27µs        ? ?/sec
decimal128_add_overflow               1.00      9.1±0.14µs        ? ?/sec    1.52     13.9±0.25µs        ? ?/sec
decimal128_add_rescale(0)             2.82     33.4±0.63µs        ? ?/sec    1.00     11.8±0.22µs        ? ?/sec
decimal128_add_rescale(0.1)           3.21     38.0±0.49µs        ? ?/sec    1.00     11.8±0.16µs        ? ?/sec
decimal128_add_rescale_scalar(0)      9.74     47.4±1.20µs        ? ?/sec    1.00      4.9±0.08µs        ? ?/sec
decimal128_add_rescale_scalar(0.1)    9.04     45.0±0.63µs        ? ?/sec    1.00      5.0±0.10µs        ? ?/sec
decimal128_divide(0)                  1.01     88.4±1.45µs        ? ?/sec    1.00     87.6±0.79µs        ? ?/sec
decimal128_divide(0.1)                1.00     75.9±1.10µs        ? ?/sec    1.00     75.5±0.97µs        ? ?/sec
decimal128_multiply(0)                1.76     16.8±0.26µs        ? ?/sec    1.00      9.6±0.19µs        ? ?/sec
decimal128_multiply(0.1)              2.11     20.7±0.38µs        ? ?/sec    1.00      9.8±0.20µs        ? ?/sec
decimal128_multiply_wide(0)           1.06     17.2±0.22µs        ? ?/sec    1.00     16.2±0.63µs        ? ?/sec
decimal128_multiply_wide(0.1)         1.38     20.8±0.28µs        ? ?/sec    1.00     15.1±0.30µs        ? ?/sec
decimal128_subtract(0)                1.45      8.0±0.13µs        ? ?/sec    1.00      5.5±0.06µs        ? ?/sec
decimal128_subtract(0.1)              1.89     10.6±0.12µs        ? ?/sec    1.00      5.6±0.06µs        ? ?/sec
decimal256_add(0)                     1.00     15.7±0.21µs        ? ?/sec    1.00     15.7±0.19µs        ? ?/sec
decimal256_add(0.1)                   1.00     20.8±0.64µs        ? ?/sec    1.01     20.9±0.65µs        ? ?/sec
decimal256_multiply(0)                1.01     89.8±1.68µs        ? ?/sec    1.00     89.3±1.34µs        ? ?/sec
decimal256_multiply(0.1)              1.00     78.9±1.11µs        ? ?/sec    1.02     80.1±4.47µs        ? ?/sec
decimal64_add(0)                      2.58      5.9±0.95µs        ? ?/sec    1.00      2.3±0.04µs        ? ?/sec
decimal64_add(0.1)                    3.71      7.7±0.85µs        ? ?/sec    1.00      2.1±0.03µs        ? ?/sec
decimal64_multiply(0)                 1.00      5.9±0.94µs        ? ?/sec    1.32      7.8±0.08µs        ? ?/sec
decimal64_multiply(0.1)               1.00      7.7±0.84µs        ? ?/sec    1.05      8.1±0.11µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1x7drz99-ecac925a-k54z

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6863040285-5026-gtrmp 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	245.1s
Peak memory	16.5 MiB
Avg memory	15.5 MiB
CPU user	239.2s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	245.1s
Peak memory	16.8 MiB
Avg memory	15.5 MiB
CPU user	241.3s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

mbutrovich · 2026-11-13T20:55:49Z

run benchmark arithmetic_kernels

baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: "^decimal(64_multiply|128_divide)"

mbutrovich · 2026-11-13T20:55:53Z

run benchmark arithmetic_kernels

baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: "^decimal(64_multiply|128_divide)"
  RUSTFLAGS: "-C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6"

mbutrovich · 2026-11-13T20:56:02Z

run benchmark arithmetic_kernels

baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: "^decimal(64_multiply|128_divide)"
resources:
  arch: amd64

mbutrovich · 2026-11-13T20:56:09Z

run benchmark arithmetic_kernels

baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: "^decimal(64_multiply|128_divide)"
  RUSTFLAGS: "-C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6"
resources:
  arch: amd64

mbutrovich · 2026-11-13T20:56:16Z

run benchmark arithmetic_kernels

baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: "^decimal(64_multiply|128_divide)"
  RUSTFLAGS: "-C target-cpu=x86-64-v3"
resources:
  arch: amd64

mbutrovich · 2026-11-13T20:56:21Z

run benchmark arithmetic_kernels

baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: "^decimal(64_multiply|128_divide)"
  RUSTFLAGS: "-C target-cpu=x86-64-v3 -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6"
resources:
  arch: amd64

adriangbot · 2026-11-13T21:02:47Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:02:49Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:02:50Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:02:51Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:05:57Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
shards: 1

Details

group                      base                                   changed
-----                      ----                                   -------
decimal128_divide(0)       1.00     51.5±0.03µs        ? ?/sec    1.00     51.5±0.02µs        ? ?/sec
decimal128_divide(0.1)     1.01     48.2±0.43µs        ? ?/sec    1.00     47.9±0.33µs        ? ?/sec
decimal64_multiply(0)      1.35      8.5±0.01µs        ? ?/sec    1.00      6.3±0.00µs        ? ?/sec
decimal64_multiply(0.1)    2.13     13.5±0.38µs        ? ?/sec    1.00      6.3±0.10µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1w1y66a2-83be8f18-trzm

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

uname:

Linux bench-c6864709006-5058-lzkj7 6.12.105+ #1 SMP Sat Sep 19 07:55:25 UTC 2026 aarch64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

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

Resource Usage

base (merge-base)

Metric	Value
Wall time	40.0s
Peak memory	14.1 MiB
Avg memory	10.5 MiB
CPU user	34.8s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	35.0s
Peak memory	14.3 MiB
Avg memory	11.6 MiB
CPU user	33.0s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:05:59Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
shards: 1

Details

group                      base                                   changed
-----                      ----                                   -------
decimal128_divide(0)       1.00     51.5±0.05µs        ? ?/sec    1.00     51.5±0.02µs        ? ?/sec
decimal128_divide(0.1)     1.00     48.2±0.42µs        ? ?/sec    1.00     48.2±0.42µs        ? ?/sec
decimal64_multiply(0)      1.35      8.5±0.01µs        ? ?/sec    1.00      6.3±0.00µs        ? ?/sec
decimal64_multiply(0.1)    2.10     13.4±0.28µs        ? ?/sec    1.00      6.4±0.01µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1w1y66a2-83be8f18-tdfb

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

uname:

Linux bench-c6864710340-5059-q28tt 6.12.105+ #1 SMP Sat Sep 19 07:55:25 UTC 2026 aarch64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

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

Resource Usage

base (merge-base)

Metric	Value
Wall time	40.0s
Peak memory	15.1 MiB
Avg memory	11.7 MiB
CPU user	35.8s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	35.0s
Peak memory	15.5 MiB
Avg memory	12.7 MiB
CPU user	33.0s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:06:04Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
  RUSTFLAGS: -C target-cpu=x86-64-v3
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:06:05Z

🤖 Benchmark starting (GKE) | trigger

Target: arithmetic_kernels

Sharding factor: 1 (1 workers).

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
  RUSTFLAGS: -C target-cpu=x86-64-v3 -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Results will be posted when all workers finish.

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:08:17Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
resources:
  arch: amd64
shards: 1

Details

group                      base                                   changed
-----                      ----                                   -------
decimal128_divide(0)       1.04    123.9±1.78µs        ? ?/sec    1.00    118.9±3.00µs        ? ?/sec
decimal128_divide(0.1)     1.02    101.2±2.71µs        ? ?/sec    1.00     99.1±2.35µs        ? ?/sec
decimal64_multiply(0)      2.40      9.0±1.41µs        ? ?/sec    1.00      3.8±0.08µs        ? ?/sec
decimal64_multiply(0.1)    2.76     11.4±1.15µs        ? ?/sec    1.00      4.1±0.11µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1jwtivje-76dcaf12-5hrm

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6864713381-5060-8pfcc 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	40.0s
Peak memory	15.9 MiB
Avg memory	12.6 MiB
CPU user	35.7s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	40.0s
Peak memory	16.2 MiB
Avg memory	13.1 MiB
CPU user	36.0s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:08:19Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
  RUSTFLAGS: -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Details

group                      base                                   changed
-----                      ----                                   -------
decimal128_divide(0)       1.00    118.4±2.53µs        ? ?/sec    1.00    118.0±1.33µs        ? ?/sec
decimal128_divide(0.1)     1.00    100.3±1.19µs        ? ?/sec    1.00    100.3±1.40µs        ? ?/sec
decimal64_multiply(0)      2.49      9.2±1.39µs        ? ?/sec    1.00      3.7±0.06µs        ? ?/sec
decimal64_multiply(0.1)    2.83     11.2±0.95µs        ? ?/sec    1.00      3.9±0.11µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1jwtivje-ad4e1533-frd7

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6864715626-5061-8c4tx 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	40.0s
Peak memory	16.8 MiB
Avg memory	13.4 MiB
CPU user	35.6s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	40.0s
Peak memory	17.3 MiB
Avg memory	13.7 MiB
CPU user	35.0s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:11:16Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
  RUSTFLAGS: -C target-cpu=x86-64-v3
resources:
  arch: amd64
shards: 1

Details

group                      base                                   changed
-----                      ----                                   -------
decimal128_divide(0)       1.00    109.6±3.65µs        ? ?/sec    1.06    115.8±3.22µs        ? ?/sec
decimal128_divide(0.1)     1.00     96.8±1.49µs        ? ?/sec    1.00     96.6±1.15µs        ? ?/sec
decimal64_multiply(0)      2.46      8.9±1.45µs        ? ?/sec    1.00      3.6±0.05µs        ? ?/sec
decimal64_multiply(0.1)    3.10     12.0±1.00µs        ? ?/sec    1.00      3.9±0.13µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1jwtivje-76dcaf12-2tws

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6864718429-5062-wsld5 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	45.0s
Peak memory	16.1 MiB
Avg memory	13.4 MiB
CPU user	41.7s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	40.0s
Peak memory	16.2 MiB
Avg memory	12.7 MiB
CPU user	35.0s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-11-13T21:11:21Z

🤖 Benchmark completed (GKE) | trigger

Target: arithmetic_kernels

Comparing vectorize-checked-decimal-arithmetic (f933cfa) to 362eb22 diff

Run configuration

run benchmark arithmetic_kernels
baseline:
  ref: 362eb2294d
env:
  BENCH_FILTER: ^decimal(64_multiply|128_divide)
  RUSTFLAGS: -C target-cpu=x86-64-v3 -C llvm-args=-align-all-functions=6 -C llvm-args=-align-all-nofallthru-blocks=6
resources:
  arch: amd64
shards: 1

Details

group                      base                                   changed
-----                      ----                                   -------
decimal128_divide(0)       1.00    111.4±2.06µs        ? ?/sec    1.01    112.2±2.15µs        ? ?/sec
decimal128_divide(0.1)     1.03     97.1±2.12µs        ? ?/sec    1.00     94.5±2.42µs        ? ?/sec
decimal64_multiply(0)      2.47      8.7±1.35µs        ? ?/sec    1.00      3.5±0.12µs        ? ?/sec
decimal64_multiply(0.1)    3.04     11.4±1.04µs        ? ?/sec    1.00      3.7±0.14µs        ? ?/sec

Per-runner information

arithmetic_kernels — shard 1/1

Node: gk3-benchmark-cluster-nap-1jwtivje-ad4e1533-dcrz

Instance: c4-highmem-16 (12 vCPU / 65 GiB)

uname:

Linux bench-c6864720043-5063-dc2pd 6.12.105+ #1 SMP Sat Sep 19 07:57:03 UTC 2026 x86_64 GNU/Linux

BENCH_COMMAND:

cargo bench --features=arrow,async,test_common,experimental,object_store --bench arithmetic_kernels

CPU Details (lscpu)

Architecture:                            x86_64
CPU op-mode(s):                          32-bit, 64-bit
Address sizes:                           52 bits physical, 57 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               GenuineIntel
Model name:                              INTEL(R) XEON(R) PLATINUM 8581C CPU @ 2.30GHz
CPU family:                              6
Model:                                   207
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                2
BogoMIPS:                                4600.00
Flags:                                   fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor:                       KVM
Virtualization type:                     full
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                16 MiB (8 instances)
L3 cache:                                260 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; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Resource Usage

base (merge-base)

Metric	Value
Wall time	40.0s
Peak memory	17.0 MiB
Avg memory	13.5 MiB
CPU user	35.7s
CPU sys	0.0s
Peak spill	0 B

branch

Metric	Value
Wall time	40.0s
Peak memory	17.2 MiB
Avg memory	13.6 MiB
CPU user	35.0s
CPU sys	0.0s
Peak spill	0 B

File an issue against this benchmark runner

mbutrovich · 2026-11-13T21:25:33Z

Decimal64 mul shares the i64 multiply that #12058 changed to overflowing_mul, and this PR now has the same change (f933cfa). Without and with 10% nulls, decimal64_multiply measured 1.35x and 2.14x on Neoverse-V2, 2.37x and 2.78x on Xeon 8581C, and 2.47x and 3.08x with x86-64-v3. Without nulls on x86-64-v3 it measured 0.76x to 0.99x in the earlier runs. The unchanged decimal128_divide measured 0.95x to 1.04x in the same runs.

andygrove

Thanks @mbutrovich, this looks good to me. I checked the sign-bit overflow logic for add/sub, the widening multiplies, the masking of overflow in null slots (including offsets), and the fallback to the checked kernel, and they all look correct. I left a few non-blocking nits.

andygrove · 2026-11-13T23:08:07Z

+        c.bench_function(&format!("decimal64_add({null_density})"), |bench| {
+            bench.iter(|| hint::black_box(add(&a64, &b64).unwrap()))
+        });
+        c.bench_function(&format!("decimal64_multiply({null_density})"), |bench| {


nit: The description says Decimal32 and Decimal64 add/sub now vectorize, but there's no Decimal32 benchmark and no decimal64_subtract. Could you add them? The Decimal32 path in particular (widening i32 to i64 for mul, sign-bit add/sub) has no coverage, so a slowdown there like the one you saw for i256 would go unnoticed.

andygrove · 2026-11-13T23:08:07Z

+/// through a pointer. The checked `add` cases of the `arithmetic_kernels`
+/// benchmark measure this.
+#[inline(never)]
+fn collect_overflowing<N>(


nit: Could you add numbers to the PR description comparing this to a safe version (e.g. Vec::with_capacity + push/extend with the same #[inline(never)] boundary)? Since the gain depends on LLVM's inlining choices for a given target, it would be good to have a record that the unsafe is what produces the gain.

andygrove · 2026-11-13T23:08:07Z

+        overflow |= overflowed;
+        written += 1;
+    }
+    debug_assert!(written <= values.capacity());


nit: written <= values.capacity() can't fail, because the zip is bounded by spare_capacity_mut(). It doesn't check the set_len precondition either. Maybe drop it, or capture iter.len() up front and debug_assert_eq!(written, len), which at least catches an iterator that reports the wrong length.

github-actions Bot added arrow-core Core crates related to representing arrow data arrow-kernels Crates that transform arrow data (e.g. compute, cast) labels Nov 3, 2026

mbutrovich mentioned this pull request Nov 3, 2026

[DISCUSSION] 2026 Q4 Roadmap apache/datafusion#26684

Open

perf(arrow-arith): use overflowing_mul for checked i64 multiply

f933cfa

Merge branch 'main' into vectorize-checked-decimal-arithmetic

9062376

mbutrovich requested review from andygrove and sunchao November 13, 2026 21:28

andygrove approved these changes Nov 13, 2026

View reviewed changes

Conversation

mbutrovich commented Nov 3, 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 significant user-facing changes?

AI Disclosure

Uh oh!

mbutrovich commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

mbutrovich commented Nov 3, 2026

Uh oh!

mbutrovich commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

mbutrovich commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

mbutrovich commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

mbutrovich commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

adriangbot commented Nov 3, 2026

Uh oh!

mbutrovich commented Nov 4, 2026

Uh oh!

mbutrovich commented Nov 4, 2026

Uh oh!

adriangbot commented Nov 4, 2026

Uh oh!

mbutrovich commented Nov 4, 2026

Uh oh!

adriangbot commented Nov 4, 2026

Uh oh!

mbutrovich commented Nov 4, 2026

Uh oh!

adriangbot commented Nov 4, 2026

Uh oh!

adriangbot commented Nov 4, 2026

Uh oh!

mbutrovich commented Nov 4, 2026

Uh oh!

adriangbot commented Nov 13, 2026

Uh oh!

adriangbot commented Nov 13, 2026

Uh oh!

adriangbot commented Nov 13, 2026

Uh oh!

adriangbot commented Nov 13, 2026

Uh oh!

adriangbot commented Nov 13, 2026

Uh oh!

adriangbot commented Nov 13, 2026

Uh oh!

mbutrovich commented Nov 13, 2026

Uh oh!

mbutrovich commented Nov 13, 2026

Uh oh!

mbutrovich commented Nov 3, 2026 •

edited

Loading