feat[gpu]: nvidia cub filter kernel

[0ax1]

No description provided.

[0ax1]

Filter_cuda_i64/1M_10pct/100000
                        time:   [35.605 µs 35.867 µs 36.041 µs]
                        thrpt:  [206.72 GiB/s 207.73 GiB/s 209.26 GiB/s]
Filter_cuda_i64/1M_50pct/500000
                        time:   [38.558 µs 38.703 µs 38.785 µs]
                        thrpt:  [192.10 GiB/s 192.51 GiB/s 193.23 GiB/s]
Filter_cuda_i64/1M_90pct/900000
                        time:   [45.271 µs 45.448 µs 45.596 µs]
                        thrpt:  [163.40 GiB/s 163.93 GiB/s 164.58 GiB/s]
Filter_cuda_i64/10M_10pct/1000000
                        time:   [216.37 µs 217.62 µs 218.51 µs]
                        thrpt:  [340.98 GiB/s 342.37 GiB/s 344.35 GiB/s]
Filter_cuda_i64/10M_50pct/5000000
                        time:   [287.96 µs 288.69 µs 289.13 µs]
                        thrpt:  [257.69 GiB/s 258.08 GiB/s 258.74 GiB/s]
Filter_cuda_i64/10M_90pct/9000000
                        time:   [346.53 µs 347.00 µs 347.57 µs]
                        thrpt:  [214.36 GiB/s 214.72 GiB/s 215.01 GiB/s]
Found 2 outliers among 10 measurements (20.00%)

[codspeed-hq]

CodSpeed Performance Report

Merging this PR will degrade performance by 29.91%

_{Comparing ad/cub-device-select (cd0dd0c) with develop (68130ce)¹}

⚠️ 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.

Summary

⚡ 7 improved benchmarks
❌ 10 regressed benchmarks
✅ 1144 untouched benchmarks
🆕 18 new benchmarks
⏩ 1323 skipped benchmarks²

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	canonical_into_non_nullable[(10000, 1, 0.0)]	25.7 µs	36.2 µs	-29.12%
❌	Simulation	canonical_into_non_nullable[(10000, 10, 0.0)]	195.5 µs	279 µs	-29.91%
❌	Simulation	canonical_into_non_nullable[(10000, 10, 0.1)]	382.1 µs	471.6 µs	-18.98%
❌	Simulation	canonical_into_non_nullable[(10000, 1, 0.01)]	32.2 µs	41.1 µs	-21.61%
❌	Simulation	canonical_into_non_nullable[(10000, 10, 0.01)]	222.6 µs	306.1 µs	-27.28%
❌	Simulation	canonical_into_non_nullable[(10000, 1, 0.1)]	48 µs	57 µs	-15.76%
⚡	Simulation	canonical_into_nullable[(10000, 100, 0.0)]	5 ms	4.4 ms	+14.03%
⚡	Simulation	into_canonical_non_nullable[(10000, 1, 0.01)]	46.3 µs	39.1 µs	+18.27%
⚡	Simulation	into_canonical_non_nullable[(10000, 1, 0.0)]	40.4 µs	33.1 µs	+22.14%
❌	Simulation	into_canonical_non_nullable[(10000, 10, 0.0)]	201.6 µs	282.4 µs	-28.59%
❌	Simulation	into_canonical_non_nullable[(10000, 10, 0.01)]	229.2 µs	309.2 µs	-25.87%
❌	Simulation	into_canonical_nullable[(10000, 10, 0.1)]	632 µs	718.9 µs	-12.09%
❌	Simulation	into_canonical_non_nullable[(10000, 10, 0.1)]	385.1 µs	471.5 µs	-18.34%
⚡	Simulation	into_canonical_nullable[(10000, 100, 0.1)]	6.9 ms	6.1 ms	+13.74%
⚡	Simulation	into_canonical_nullable[(10000, 100, 0.0)]	5.1 ms	4.3 ms	+16.84%
⚡	Simulation	into_canonical_non_nullable[(10000, 1, 0.1)]	62.8 µs	55.2 µs	+13.74%
⚡	Simulation	into_canonical_nullable[(10000, 10, 0.0)]	540.8 µs	458.2 µs	+18.02%
🆕	WallTime	1M_10pct[100000]	N/A	45.8 µs	N/A
🆕	WallTime	10M_10pct[1000000]	N/A	132.6 µs	N/A
🆕	WallTime	1M_50pct[500000]	N/A	21.9 µs	N/A
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Footnotes

1. No successful run was found on develop (6ab6b5f) during the generation of this report, so 68130ce was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

2. 1323 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[a10y]

sweet