buffer allocators

[onursatici]

this PR introduces buffer allocators to the scan, which the caller can provide its own impl to scan directly into device buffer handles.

TODO:

• device buffer alignment

example cuda scan using this:

let session = VortexSession::default();
let allocator = Arc::new(HostToDeviceAllocator::from_session(&session)?);

let file = session
    .open_options()
    .with_allocator(allocator)
    .open_path("data.vortex")
    .await?;

file.scan()?.into_array_stream()?

[joseph-isaacs]

write destination

[joseph-isaacs]

remove?

[joseph-isaacs]

This is a CPU only thing?

[onursatici]

we would not have this wrapper normally but push the allocators to the leaf ReadAt impl's. This is here to keep the diff small and get the device copy working when using a device allocator

[codspeed-hq]

CodSpeed Performance Report

Merging this PR will degrade performance by 33.09%

_{Comparing os/gpu-scan (d665f60) with develop (176e340)}

⚠️ 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

⚡ 8 improved benchmarks
❌ 38 regressed benchmarks
✅ 1219 untouched benchmarks
⏩ 1219 skipped benchmarks¹

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

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	u8_FoR[10M]	6.4 µs	9.6 µs	-33.09%
❌	Simulation	bench_compare_primitive[(10000, 128)]	165.9 µs	184.8 µs	-10.24%
❌	Simulation	bench_compare_primitive[(10000, 2)]	162 µs	180.9 µs	-10.47%
❌	Simulation	bench_compare_primitive[(10000, 32)]	162.9 µs	181.8 µs	-10.42%
❌	Simulation	bench_compare_primitive[(10000, 4)]	161.3 µs	180.3 µs	-10.52%
❌	Simulation	bench_compare_primitive[(100000, 128)]	903.7 µs	1,094.1 µs	-17.4%
❌	Simulation	bench_compare_primitive[(10000, 8)]	161.7 µs	180.6 µs	-10.49%
❌	Simulation	bench_compare_primitive[(100000, 2)]	899.4 µs	1,089.9 µs	-17.47%
❌	Simulation	bench_compare_primitive[(100000, 2048)]	1 ms	1.2 ms	-15.96%
❌	Simulation	bench_compare_primitive[(100000, 4)]	899.6 µs	1,090 µs	-17.47%
❌	Simulation	bench_compare_primitive[(100000, 32)]	901.4 µs	1,091.8 µs	-17.44%
❌	Simulation	bench_compare_primitive[(100000, 512)]	961.6 µs	1,152 µs	-16.53%
❌	Simulation	bench_compare_primitive[(100000, 8)]	900 µs	1,090.4 µs	-17.46%
❌	Simulation	bench_compare_varbin[(10000, 32)]	170.7 µs	190.3 µs	-10.3%
❌	Simulation	bench_compare_varbin[(10000, 2)]	166.3 µs	185.9 µs	-10.53%
❌	Simulation	bench_compare_varbin[(10000, 4)]	166.8 µs	186.4 µs	-10.52%
❌	Simulation	bench_compare_varbin[(10000, 8)]	167.2 µs	186.8 µs	-10.49%
❌	Simulation	bench_compare_varbin[(100000, 128)]	921 µs	1,112 µs	-17.18%
❌	Simulation	bench_compare_varbin[(100000, 2)]	904.1 µs	1,095.2 µs	-17.45%
❌	Simulation	bench_compare_varbin[(100000, 2048)]	1.2 ms	1.4 ms	-13.53%
...	...	...	...	...	...

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

Footnotes

1. 1219 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. ↩