|
| 1 | +// SPDX-License-Identifier: Apache-2.0 |
| 2 | +// SPDX-FileCopyrightText: Copyright the Vortex contributors |
| 3 | + |
| 4 | +//! Benchmarks `bitpack_encode_blocked`, which packs every 1024-value block at its own bit width. |
| 5 | +
|
| 6 | +#![expect(clippy::unwrap_used)] |
| 7 | + |
| 8 | +use std::sync::LazyLock; |
| 9 | + |
| 10 | +use divan::Bencher; |
| 11 | +use divan::counter::ItemsCount; |
| 12 | +use mimalloc::MiMalloc; |
| 13 | +use vortex_array::VortexSessionExecute; |
| 14 | +use vortex_array::arrays::PrimitiveArray; |
| 15 | +use vortex_fastlanes::bitpack_compress::bitpack_encode_blocked; |
| 16 | +use vortex_session::VortexSession; |
| 17 | + |
| 18 | +#[global_allocator] |
| 19 | +static GLOBAL: MiMalloc = MiMalloc; |
| 20 | + |
| 21 | +fn main() { |
| 22 | + divan::main(); |
| 23 | +} |
| 24 | + |
| 25 | +static SESSION: LazyLock<VortexSession> = LazyLock::new(|| { |
| 26 | + let session = vortex_array::array_session(); |
| 27 | + vortex_fastlanes::initialize(&session); |
| 28 | + session |
| 29 | +}); |
| 30 | + |
| 31 | +/// 64 blocks of `u32`, 256 KiB in all: shorter iterations are noisy on the walltime legs. |
| 32 | +const NUM_BLOCKS: usize = 64; |
| 33 | + |
| 34 | +#[vortex_bench_support::cpu_features] |
| 35 | +#[divan::bench] |
| 36 | +fn bitpack_blocked_compress(bencher: Bencher) { |
| 37 | + // Block widths cycle from 1 to 16 bits, so the blocks take different packing kernels. |
| 38 | + let bit_widths: Vec<u8> = (1..=16).cycle().take(NUM_BLOCKS).collect(); |
| 39 | + let array = PrimitiveArray::from_iter(bit_widths.iter().flat_map(|&bit_width| { |
| 40 | + (0..1024u32).map(move |i| i.wrapping_mul(7919) & ((1 << bit_width) - 1)) |
| 41 | + })); |
| 42 | + |
| 43 | + bencher.counter(ItemsCount::new(array.len())).bench(|| { |
| 44 | + bitpack_encode_blocked( |
| 45 | + &array, |
| 46 | + &bit_widths, |
| 47 | + Some(0), |
| 48 | + &mut SESSION.create_execution_ctx(), |
| 49 | + ) |
| 50 | + .unwrap() |
| 51 | + }); |
| 52 | +} |
0 commit comments