Using ComputeTaskPool for a par_for_each query only uses half of available logical cores

[MJohnson459]

Bevy version

0.5

Operating system & version

Linux

What you did

I have a CPU heavy task and running it as part of a query par_for_each loop in a ComputeTaskPool and only half of my cores are being utilised. I originally used an AsyncTaskPool but this has even fewer threads assigned.

use bevy::tasks::ComputeTaskPool;
pub fn how_many_threads(pool: Res<ComputeTaskPool>, query: Query<(&DummyComponent)>) {
    info!("TaskPool threads: {}", pool.thread_num());
    info!("Available cores: {}", num_cpus::get());

    query.par_for_each_mut(&pool, 1, |_dummy| {
        // Will only run on
    });
}

TaskPool threads: 6
Available cores: 12

I know why this happens but it was unexpected:

bevy/crates/bevy_core/src/task_pool_options.rs

Line 54 in 75ae20d

impl Default for DefaultTaskPoolOptions {

What you expected to happen

When not using the IoTaskPool or AsyncComputeTaskPool I would expect the ComputeTaskPool to use all logical cores. Likewise I would expect the other pools to expand to optimally use the CPU available.

Alternatively, documentation should be updated to make clear what the current behaviour is and steps to create a new task pool.

What actually happened

Only half of the available core were used.

[james7132]

This seems like it would be resolved by #4740. Probably helps that #4791 was resolved.