claude/bitpacking-transpose-optimization-tM1U4 - Branch - vortex-data/vortex

feat[fastlanes]: add optimized 1024-bit transpose implementations

#6135

Comparing

claude/bitpacking-transpose-optimization-tM1U4

(

2cbd439

) with

develop

(

13f120f

)

-30%

Improvement: 3

Regression: 7

Untouched: 1252

New: 16

Skipped: 1290

Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data. For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

Benchmarks

2568 total

canonical_into_non_nullable[(10000, 100, 0.0)]

encodings/fastlanes/benches/canonicalize_bench.rs

-30%

1.9 ms2.7 ms

into_canonical_non_nullable[(10000, 100, 0.0)]

encodings/fastlanes/benches/canonicalize_bench.rs

-29%

1.9 ms2.7 ms

canonical_into_non_nullable[(10000, 100, 0.01)]

encodings/fastlanes/benches/canonicalize_bench.rs

-28%

2.1 ms3 ms

into_canonical_non_nullable[(10000, 100, 0.01)]

encodings/fastlanes/benches/canonicalize_bench.rs

-27%

2.2 ms3 ms

canonical_into_non_nullable[(10000, 100, 0.1)]

encodings/fastlanes/benches/canonicalize_bench.rs

-18%

3.7 ms4.5 ms

into_canonical_non_nullable[(10000, 100, 0.1)]

encodings/fastlanes/benches/canonicalize_bench.rs

-18%

3.8 ms4.6 ms

into_canonical_nullable[(10000, 100, 0.0)]

encodings/fastlanes/benches/canonicalize_bench.rs

-16%

4.4 ms5.2 ms

u8_FoR[10M]

vortex-cuda/benches/for_cuda.rs::benches::cuda_benchmarks::benchmark_for_cuda::FoR_cuda_u8

×13

71.7 µs5.6 µs

canonical_into_nullable[(10000, 100, 0.0)]

encodings/fastlanes/benches/canonicalize_bench.rs

+20%

4.9 ms4.1 ms

canonical_into_nullable[(10000, 10, 0.0)]

encodings/fastlanes/benches/canonicalize_bench.rs

+19%

528.5 µs445.6 µs

transpose_baseline_throughput

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A2.5 ms

transpose_baseline

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A10.9 µs

transpose_best_throughput

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A92.8 µs

transpose_best

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A2 µs

transpose_scalar

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A3.4 µs

untranspose_best

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A2.8 µs

transpose_scalar_throughput

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A661 µs

transpose_scalar_fast

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A1.7 µs

untranspose_baseline

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A10.9 µs

transpose_scalar_fast_throughput

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A64.2 µs

transpose_avx2

encodings/fastlanes/benches/transpose_bench.rs::x86_benches

N/A

N/A2.8 µs

untranspose_scalar

encodings/fastlanes/benches/transpose_bench.rs

N/A

N/A3.2 µs

transpose_bmi2

encodings/fastlanes/benches/transpose_bench.rs::x86_benches

N/A

N/A1.8 µs

transpose_avx2_throughput

encodings/fastlanes/benches/transpose_bench.rs::x86_throughput_benches

N/A

N/A314.3 µs

transpose_bmi2_throughput

encodings/fastlanes/benches/transpose_bench.rs::x86_throughput_benches

N/A

N/A90.1 µs

Commits

Click on a commit to change the comparison range

Base

develop

13f120f

-29.9%

feat[fastlanes]: add BMI2 PEXT/PDEP transpose and fix GFNI implementations

338b2bc

7 hours ago

by claude

feat[fastlanes]: add scalar_fast and ARM64 NEON transpose implementations

f13f085

7 hours ago

by claude

perf[fastlanes]: fully unroll BMI2 transpose for 12% performance gain

c471853

7 hours ago

by claude

-15.84%

test[fastlanes]: add verification against fastlanes crate transpose

7282427

7 hours ago

by claude

+15.84%

feat[fastlanes]: add AVX-512 VBMI transpose with 7.5x speedup

ccf3a19

6 hours ago

by claude

-15.69%

feat[fastlanes]: add dual-block VBMI transpose for 10% more throughput

99299a5

6 hours ago

by claude

+15.69%

feat[fastlanes]: add 4-block VBMI transpose for 7% additional speedup

2cbd439

5 hours ago

by claude

Home Terms Privacy Docs