bench_mask_from_bits.rs

#![feature(portable_simd)]

use core_simd::*;
use criterion::{black_box, criterion_group, criterion_main, Criterion};
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
use std::ops::Add;

#[inline(never)]
fn mask8x8_from_bits(bits: u64) -> mask8x8 {
    mask8x8::from_bitmask(bits as u8)
}

#[inline(never)]
fn mask8x16_from_bits(bits: u64) -> mask8x16 {
    mask8x16::from_bitmask(bits as u16)
}

#[inline(never)]
fn mask16x8_from_bits(bits: u64) -> mask16x8 {
    mask16x8::from_bitmask(bits as u8)
}

#[inline(never)]
fn mask16x16_from_bits(bits: u64) -> mask16x16 {
    mask16x16::from_bitmask(bits as u16)
}

#[inline(never)]
fn mask8x8_add(bits: u64, a: i8x8, b: i8x8) -> i8x8 {
    a.add(mask8x8::from_bitmask(bits as u8).select(b, i8x8::splat(0)))
}

#[inline(never)]
fn mask8x16_add(bits: u64, a: i8x16, b: i8x16) -> i8x16 {
    a.add(mask8x16::from_bitmask(bits as u16).select(b, i8x16::splat(0)))
}

#[inline(never)]
fn mask16x8_add(bits: u64, a: i16x8, b: i16x8) -> i16x8 {
    a.add(mask16x8::from_bitmask(bits as u8).select(b, i16x8::splat(0)))
}

#[inline(never)]
fn mask16x16_add(bits: u64, a: i16x16, b: i16x16) -> i16x16 {
    a.add(mask16x16::from_bitmask(bits as u16).select(b, i16x16::splat(0)))
}

pub fn bench_mask_from_bits(c: &mut Criterion) {
    let mut rng = StdRng::seed_from_u64(42);

    let bits = black_box(rng.gen_range(0..0x01_0000_u64));

    c.bench_function("mask8x8_from_bits", |bencher| {
        bencher.iter(|| mask8x8_from_bits(bits))
    });
    c.bench_function("mask8x16_from_bits", |bencher| {
        bencher.iter(|| mask8x16_from_bits(bits))
    });
    c.bench_function("mask16x8_from_bits", |bencher| {
        bencher.iter(|| mask16x8_from_bits(bits))
    });
    c.bench_function("mask16x16_from_bits", |bencher| {
        bencher.iter(|| mask16x16_from_bits(bits))
    });

    c.bench_function("mask8x8_add", |bencher| {
        let a = black_box(i8x8::splat(rng.gen_range(0..=127_i8)));
        let b = black_box(i8x8::splat(rng.gen_range(0..=127_i8)));
        bencher.iter(|| mask8x8_add(bits, a, b))
    });
    c.bench_function("mask8x16_add", |bencher| {
        let a = black_box(i8x16::splat(rng.gen_range(0..=127_i8)));
        let b = black_box(i8x16::splat(rng.gen_range(0..=127_i8)));
        bencher.iter(|| mask8x16_add(bits, a, b))
    });
    c.bench_function("mask16x8_add", |bencher| {
        let a = black_box(i16x8::splat(rng.gen_range(0..=32767_i16)));
        let b = black_box(i16x8::splat(rng.gen_range(0..=32767_i16)));
        bencher.iter(|| mask16x8_add(bits, a, b))
    });
    c.bench_function("mask16x16_add", |bencher| {
        let a = black_box(i16x16::splat(rng.gen_range(0..=32767_i16)));
        let b = black_box(i16x16::splat(rng.gen_range(0..=32767_i16)));
        bencher.iter(|| mask16x16_add(bits, a, b))
    });
}

criterion_group!(benches, bench_mask_from_bits);
criterion_main!(benches);