feat[gpu]: dict take values prim

[joseph-isaacs]

Support cuda dict decode for primitive types

[codspeed-hq]

Merging this PR will degrade performance by 29.75%

⚠️ 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.

⚡ 2 improved benchmarks
❌ 7 regressed benchmarks
✅ 1249 untouched benchmarks
🆕 4 new benchmarks
⏩ 1290 skipped benchmarks¹

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

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
🆕	WallTime	u32_values_u8_codes[10M]	N/A	128.1 µs	N/A
🆕	WallTime	u32_values_u16_codes[10M]	N/A	144.6 µs	N/A
🆕	WallTime	u64_values_u32_codes[10M]	N/A	274.9 µs	N/A
🆕	WallTime	u64_values_u8_codes[10M]	N/A	213.7 µs	N/A
❌	Simulation	into_canonical_non_nullable[(10000, 100, 0.1)]	3.8 ms	4.6 ms	-17.7%
❌	Simulation	into_canonical_non_nullable[(10000, 100, 0.01)]	2.2 ms	3 ms	-27.05%
❌	Simulation	into_canonical_non_nullable[(10000, 100, 0.0)]	1.9 ms	2.7 ms	-29.42%
❌	Simulation	into_canonical_nullable[(10000, 100, 0.0)]	4.4 ms	5.2 ms	-15.62%
❌	Simulation	canonical_into_non_nullable[(10000, 100, 0.0)]	1.9 ms	2.7 ms	-29.75%
❌	Simulation	canonical_into_non_nullable[(10000, 100, 0.01)]	2.1 ms	2.9 ms	-27.39%
⚡	Simulation	canonical_into_nullable[(10000, 10, 0.0)]	528.5 µs	444.1 µs	+19.03%
⚡	Simulation	canonical_into_nullable[(10000, 100, 0.0)]	4.9 ms	4.1 ms	+19.79%
❌	Simulation	canonical_into_non_nullable[(10000, 100, 0.1)]	3.7 ms	4.5 ms	-18.15%

---

_{Comparing ji/dict-array-2 (0c69415) with develop (46708ba)}

Footnotes

1. 1290 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. ↩

[0ax1]

@claude review

[claude]

a

[0ax1]

1b is a very slow run. i'd rather keep 1m and 100k. expected 1b inputs seems rather "synthetic"?

[joseph-isaacs]

How slow was it?

[0ax1]

do you see any througput diff between 10, 100, 1000M?

[joseph-isaacs]

I did when we rand it before