Struct ThreadPoolManager

pub struct ThreadPoolManager { /* private fields */ }

Thread pool manager for parallel neural network operations

Manages a pool of worker threads for parallel execution of neural network operations, providing load balancing and efficient resource utilization.

Implementations

impl ThreadPoolManager

pub fn new(num_threads: Option<usize>) -> Result<Self>

Create a new thread pool manager

Arguments

• num_threads - Number of threads in the pool (None for automatic detection)

Examples

use scirs2_neural::performance::threading::ThreadPoolManager;

// Auto-detect thread count
let pool = ThreadPoolManager::new(None).unwrap();

// Specify thread count
let pool = ThreadPoolManager::new(Some(8)).unwrap();

Examples found in repository

examples/new_features_showcase.rs (line 134)

fn demonstrate_performance_optimizations() -> Result<(), Box<dyn std::error::Error>> {
    println!("⚡ Performance Optimizations Demonstration");
    println!("========================================\n");

    // 1. Thread Pool Manager
    println!("1. Thread Pool Manager:");
    let thread_pool = ThreadPoolManager::new(Some(4))?;
    println!(
        "   Created thread pool with {} threads",
        thread_pool.num_threads()
    );

    // Demonstrate parallel matrix multiplication
    let matrix_a = Array::from_elem((100, 50), 2.0f32).into_dyn();
    let matrix_b = Array::from_elem((50, 75), 3.0f32).into_dyn();

    let start_time = std::time::Instant::now();
    let result = thread_pool.parallel_matmul(&matrix_a, &matrix_b)?;
    let elapsed = start_time.elapsed();

    println!(
        "   Parallel matrix multiplication: {}x{} * {}x{} = {}x{}",
        matrix_a.shape()[0],
        matrix_a.shape()[1],
        matrix_b.shape()[0],
        matrix_b.shape()[1],
        result.shape()[0],
        result.shape()[1]
    );
    println!("   Time elapsed: {:.3}ms", elapsed.as_secs_f64() * 1000.0);
    println!("   Result sample: {:.1}", result[[0, 0]]);

    // 2. Performance Profiler
    println!("\n2. Performance Profiler:");
    let mut optimizer = PerformanceOptimizer::new(Some(1024), Some(512), Some(4), true)?;

    // Simulate some operations with profiling
    {
        let timer = optimizer.profiler_mut().start_timer("matrix_setup");
        let test_matrix_a = Array::from_elem((200, 100), 1.5f32).into_dyn();
        let test_matrix_b = Array::from_elem((100, 150), 2.5f32).into_dyn();
        optimizer
            .profiler_mut()
            .end_timer("matrix_setup".to_string(), timer);

        let _result = optimizer.optimized_matmul(&test_matrix_a, &test_matrix_b)?;
    }

    println!("   Performance profile summary:");
    optimizer.profiler().print_summary();

    // 3. Optimization Capabilities
    println!("\n3. Optimization Capabilities:");
    let capabilities = optimizer.get_capabilities();
    println!("{}", capabilities);

    println!("✅ Performance optimizations demonstration completed!\n");
    Ok(())
}

pub fn execute<F, R>(&self, f: F) -> R

where F: FnOnce() -> R + Send, R: Send,

Execute a function in the thread pool (no-op without parallel)

pub fn parallel_matmul( &self, a: &ArrayD<f32>, b: &ArrayD<f32>, ) -> Result<ArrayD<f32>>

Parallel matrix multiplication using thread pool

Performs matrix multiplication with automatic parallelization across available threads for improved performance on large matrices.

Examples found in repository

examples/new_features_showcase.rs (line 145)

fn demonstrate_performance_optimizations() -> Result<(), Box<dyn std::error::Error>> {
    println!("⚡ Performance Optimizations Demonstration");
    println!("========================================\n");

    // 1. Thread Pool Manager
    println!("1. Thread Pool Manager:");
    let thread_pool = ThreadPoolManager::new(Some(4))?;
    println!(
        "   Created thread pool with {} threads",
        thread_pool.num_threads()
    );

    // Demonstrate parallel matrix multiplication
    let matrix_a = Array::from_elem((100, 50), 2.0f32).into_dyn();
    let matrix_b = Array::from_elem((50, 75), 3.0f32).into_dyn();

    let start_time = std::time::Instant::now();
    let result = thread_pool.parallel_matmul(&matrix_a, &matrix_b)?;
    let elapsed = start_time.elapsed();

    println!(
        "   Parallel matrix multiplication: {}x{} * {}x{} = {}x{}",
        matrix_a.shape()[0],
        matrix_a.shape()[1],
        matrix_b.shape()[0],
        matrix_b.shape()[1],
        result.shape()[0],
        result.shape()[1]
    );
    println!("   Time elapsed: {:.3}ms", elapsed.as_secs_f64() * 1000.0);
    println!("   Result sample: {:.1}", result[[0, 0]]);

    // 2. Performance Profiler
    println!("\n2. Performance Profiler:");
    let mut optimizer = PerformanceOptimizer::new(Some(1024), Some(512), Some(4), true)?;

    // Simulate some operations with profiling
    {
        let timer = optimizer.profiler_mut().start_timer("matrix_setup");
        let test_matrix_a = Array::from_elem((200, 100), 1.5f32).into_dyn();
        let test_matrix_b = Array::from_elem((100, 150), 2.5f32).into_dyn();
        optimizer
            .profiler_mut()
            .end_timer("matrix_setup".to_string(), timer);

        let _result = optimizer.optimized_matmul(&test_matrix_a, &test_matrix_b)?;
    }

    println!("   Performance profile summary:");
    optimizer.profiler().print_summary();

    // 3. Optimization Capabilities
    println!("\n3. Optimization Capabilities:");
    let capabilities = optimizer.get_capabilities();
    println!("{}", capabilities);

    println!("✅ Performance optimizations demonstration completed!\n");
    Ok(())
}

pub fn parallel_conv2d( &self, input: &ArrayD<f32>, kernel: &ArrayD<f32>, bias: Option<&[f32]>, stride: (usize, usize), padding: (usize, usize), ) -> Result<ArrayD<f32>>

Parallel convolution operation

pub fn num_threads(&self) -> usize

Get the number of threads in the pool

Examples found in repository

examples/new_features_showcase.rs (line 137)

fn demonstrate_performance_optimizations() -> Result<(), Box<dyn std::error::Error>> {
    println!("⚡ Performance Optimizations Demonstration");
    println!("========================================\n");

    // 1. Thread Pool Manager
    println!("1. Thread Pool Manager:");
    let thread_pool = ThreadPoolManager::new(Some(4))?;
    println!(
        "   Created thread pool with {} threads",
        thread_pool.num_threads()
    );

    // Demonstrate parallel matrix multiplication
    let matrix_a = Array::from_elem((100, 50), 2.0f32).into_dyn();
    let matrix_b = Array::from_elem((50, 75), 3.0f32).into_dyn();

    let start_time = std::time::Instant::now();
    let result = thread_pool.parallel_matmul(&matrix_a, &matrix_b)?;
    let elapsed = start_time.elapsed();

    println!(
        "   Parallel matrix multiplication: {}x{} * {}x{} = {}x{}",
        matrix_a.shape()[0],
        matrix_a.shape()[1],
        matrix_b.shape()[0],
        matrix_b.shape()[1],
        result.shape()[0],
        result.shape()[1]
    );
    println!("   Time elapsed: {:.3}ms", elapsed.as_secs_f64() * 1000.0);
    println!("   Result sample: {:.1}", result[[0, 0]]);

    // 2. Performance Profiler
    println!("\n2. Performance Profiler:");
    let mut optimizer = PerformanceOptimizer::new(Some(1024), Some(512), Some(4), true)?;

    // Simulate some operations with profiling
    {
        let timer = optimizer.profiler_mut().start_timer("matrix_setup");
        let test_matrix_a = Array::from_elem((200, 100), 1.5f32).into_dyn();
        let test_matrix_b = Array::from_elem((100, 150), 2.5f32).into_dyn();
        optimizer
            .profiler_mut()
            .end_timer("matrix_setup".to_string(), timer);

        let _result = optimizer.optimized_matmul(&test_matrix_a, &test_matrix_b)?;
    }

    println!("   Performance profile summary:");
    optimizer.profiler().print_summary();

    // 3. Optimization Capabilities
    println!("\n3. Optimization Capabilities:");
    let capabilities = optimizer.get_capabilities();
    println!("{}", capabilities);

    println!("✅ Performance optimizations demonstration completed!\n");
    Ok(())
}

pub fn get_stats(&self) -> ThreadPoolStats

Get thread pool statistics