//! Benachmarking funcitonality for [Criterion.rs](https://github.com/bheisler/criterion.rs)
//! This benchmark will compare the performance of various thread pools launched with different amounts of
//! maximum threads.
//! Each thread will calculate a partial dot product of two different vectors composed of 1,000,000 64-bit
//! double precision floating point values.

use std::sync::Arc;

use criterion::{black_box, criterion_group, criterion_main, BenchmarkId, Criterion, Throughput};
use imsearch::multithreading::ThreadPool;

/// Amount of elements per vector used to calculate the dot product
const VEC_ELEM_COUNT: usize = 1_000_000;
/// Number of threads to test
const THREAD_COUNTS: [usize; 17] = [
    1, 2, 4, 6, 8, 10, 12, 16, 18, 20, 22, 26, 28, 32, 40, 56, 64,
];
/// seeds used to scramble up the values produced by the hash function for each vector
/// these are just some pseudo random numbers
const VEC_SEEDS: [u64; 2] = [0xa3f8347abce16, 0xa273048ca9dea];

/// Compute the dot product of two vectors
/// # Panics
/// this function assumes both vectors to be of exactly the same length.
/// If this is not the case the function will panic.
fn dot(a: &[f64], b: &[f64]) -> f64 {
    let mut sum = 0.0;

    for i in 0..a.len() {
        sum += a[i] * b[i];
    }

    sum
}

/// Computes the dot product using a thread pool with varying number of threads. The vectors will be both splitted into equally
/// sized slices which then get passed ot their own thread to compute the partial dot product. After all threads have
/// finished the partial dot products will be summed to create the final result.
fn dot_parallel(a: Arc<Vec<f64>>, b: Arc<Vec<f64>>, threads: usize) {

    let mut pool = ThreadPool::with_limit(threads);

    // number of elements in each vector for each thread
    let steps = a.len() / threads;

    for i in 0..threads {
        // offset of the first element for the thread local vec
        let chunk = i * steps;
        // create a new strong reference to the vector
        let aa = a.clone();
        let bb = b.clone();
        // launch a new thread
        pool.enqueue(move || {
            let a = &aa[chunk..(chunk + steps)];
            let b = &bb[chunk..(chunk + steps)];
            dot(a, b)
        });
    }
  
    pool.join_all();

    black_box(pool.get_results().iter().sum::<f64>());
}

/// Compute a simple hash value for the given index value.
/// This function will return a value between [0, 1].
#[inline]
fn hash(x: f64) -> f64 {
    ((x * 234.8743 + 3.8274).sin() * 87624.58376).fract()
}

/// Create a vector filled with `size` elements of 64-bit floating point numbers
/// each initialized with the function `hash` and the given seed. The vector will
/// be filled with values between [0, 1].
fn create_vec(size: usize, seed: u64) -> Arc<Vec<f64>> {
    let mut vec = Vec::with_capacity(size);

    for i in 0..size {
        vec.push(hash(i as f64 + seed as f64));
    }

    Arc::new(vec)
}

/// Function for executing the thread pool benchmarks using criterion.rs.
/// It will create two different vectors and benchmark the single thread performance
/// as well as the multi threadded performance for varying amounts of threads.
pub fn bench_threadpool(c: &mut Criterion) {
    let vec_a = create_vec(VEC_ELEM_COUNT, VEC_SEEDS[0]);
    let vec_b = create_vec(VEC_ELEM_COUNT, VEC_SEEDS[1]);

    let mut group = c.benchmark_group("threadpool with various number of threads");

    for threads in THREAD_COUNTS.iter() {
        group.throughput(Throughput::Bytes(*threads as u64));
        group.bench_with_input(BenchmarkId::from_parameter(threads), threads, |b, _| {
            b.iter(|| {
                dot_parallel(vec_a.clone(), vec_b.clone(), *threads);
            });
        });
    }
    group.finish();
}

/// Benchmark the effects of over and underusing a thread pools thread capacity.
/// The thread pool will automatically choose the number of threads to use.
/// We will then run a custom number of jobs with that pool that may be smaller or larger
/// than the amount of threads the pool can offer.
fn pool_overusage(a: Arc<Vec<f64>>, b: Arc<Vec<f64>>, threads: usize) {
    // automatically choose the number of threads
    let mut pool = ThreadPool::new();

    // number of elements in each vector for each thread
    let steps = a.len() / threads;

    for i in 0..threads {
        // offset of the first element for the thread local vec
        let chunk = i * steps;
        // create a new strong reference to the vector
        let aa = a.clone();
        let bb = b.clone();
        // launch a new thread
        pool.enqueue(move || {
            let a = &aa[chunk..(chunk + steps)];
            let b = &bb[chunk..(chunk + steps)];
            dot(a, b)
        });
    }

    pool.join_all();

    black_box(pool.get_results().iter().sum::<f64>());
}

/// Benchmark the effects of over and underusing a thread pools thread capacity.
/// The thread pool will automatically choose the number of threads to use.
/// We will then run a custom number of jobs with that pool that may be smaller or larger
/// than the amount of threads the pool can offer.
pub fn bench_overusage(c: &mut Criterion) {
    let vec_a = create_vec(VEC_ELEM_COUNT, VEC_SEEDS[0]);
    let vec_b = create_vec(VEC_ELEM_COUNT, VEC_SEEDS[1]);

    let mut group = c.benchmark_group("threadpool overusage");

    for threads in THREAD_COUNTS.iter() {
        group.throughput(Throughput::Bytes(*threads as u64));
        group.bench_with_input(BenchmarkId::from_parameter(threads), threads, |b, _| {
            b.iter(|| {
                pool_overusage(vec_a.clone(), vec_b.clone(), *threads);
            });
        });
    }
    group.finish();
}

/// Benchmark the performance of a single thread used to calculate the dot product.
pub fn bench_single_threaded(c: &mut Criterion) {
    let vec_a = create_vec(VEC_ELEM_COUNT, VEC_SEEDS[0]);
    let vec_b = create_vec(VEC_ELEM_COUNT, VEC_SEEDS[1]);

    c.bench_function("single threaded", |s| {
        s.iter(|| {
            black_box(dot(&vec_a, &vec_b));
        });
    });
}

criterion_group!(
    benches,
    bench_single_threaded,
    bench_threadpool,
    bench_overusage
);
criterion_main!(benches);