1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111

/*
 * opencl stream executor
 * Copyright (C) 2021 trivernis
 * See LICENSE for more information
 */

use crate::executor::context::ExecutorContext;
use crate::executor::stream::{OCLStream, OCLStreamSender};
use crate::utils::result::OCLStreamResult;
use ocl::ProQue;
use scheduled_thread_pool::ScheduledThreadPool;
use std::sync::Arc;

pub mod context;
pub mod stream;

/// Stream executor for OpenCL Programs
#[derive(Clone)]
pub struct OCLStreamExecutor {
    pro_que: ProQue,
    pool: Arc<ScheduledThreadPool>,
    concurrency: usize,
}

impl OCLStreamExecutor {
    /// Creates a new OpenCL Stream executor
    /// ```rust
    /// use ocl::ProQue;
    /// use ocl_stream::OCLStreamExecutor;
    /// let pro_que = ProQue::builder().src("__kernel void bench_int() {}").build().unwrap();
    /// let executor = OCLStreamExecutor::new(pro_que);
    /// ```
    pub fn new(pro_que: ProQue) -> Self {
        Self {
            pro_que,
            pool: Arc::new(ScheduledThreadPool::new(num_cpus::get())),
            concurrency: 1,
        }
    }

    /// Sets how many threads should be used to schedule kernels on
    /// the gpu. Using multiple threads reduces the idle time of the gpu.
    /// While one kernel is running, the next one can be prepared in a
    /// different thread. A value of 0 means that the number of cpu cores should be used.
    pub fn set_concurrency(&mut self, mut num_tasks: usize) {
        if num_tasks == 0 {
            num_tasks = num_cpus::get();
        }
        self.concurrency = num_tasks;
    }

    /// Replaces the used pool with a new one
    pub fn set_pool(&mut self, pool: ScheduledThreadPool) {
        self.pool = Arc::new(pool);
    }

    /// Executes a closure in the ocl context with a bounded channel
    pub fn execute_bounded<F, T>(&self, size: usize, func: F) -> OCLStream<T>
    where
        F: Fn(ExecutorContext<T>) -> OCLStreamResult<()> + Send + Sync + 'static,
        T: Send + Sync + 'static,
    {
        let (stream, sender) = stream::bounded(size);
        self.execute(func, sender);

        stream
    }

    /// Executes a closure in the ocl context with an unbounded channel
    /// for streaming
    pub fn execute_unbounded<F, T>(&self, func: F) -> OCLStream<T>
    where
        F: Fn(ExecutorContext<T>) -> OCLStreamResult<()> + Send + Sync + 'static,
        T: Send + Sync + 'static,
    {
        let (stream, sender) = stream::unbounded();
        self.execute(func, sender);

        stream
    }

    /// Executes a closure in the ocl context
    fn execute<F, T>(&self, func: F, sender: OCLStreamSender<T>)
    where
        F: Fn(ExecutorContext<T>) -> OCLStreamResult<()> + Send + Sync + 'static,
        T: Send + Sync + 'static,
    {
        let func = Arc::new(func);

        for task_id in 0..(self.concurrency) {
            let func = Arc::clone(&func);
            let context = self.build_context(task_id, sender.clone());

            self.pool.execute(move || {
                let sender2 = context.sender().clone();

                if let Err(e) = func(context) {
                    sender2.err(e).unwrap();
                }
            });
        }
    }

    /// Builds the executor context for the executor
    fn build_context<T>(&self, task_id: usize, sender: OCLStreamSender<T>) -> ExecutorContext<T>
    where
        T: Send + Sync,
    {
        ExecutorContext::new(self.pro_que.clone(), task_id, sender)
    }
}