scirs2-fft 0.4.0

use scirs2_core::ndarray::Array2;
use scirs2_core::Complex64;
use scirs2_fft::fft::{fft, fft2};
use scirs2_fft::memory_efficient::{fft2_efficient, fft_inplace, FftMode};
use scirs2_fft::PlanCache;
use std::f64::consts::PI;
use std::time::{Duration, Instant};

#[allow(dead_code)]
fn main() {
    println!("Memory Usage Benchmarking for FFT Operations");
    println!("=============================================\n");

    // Test sizes for 1D FFT
    let sizes_1d = [1024, 4096, 16384, 65536];

    // Test sizes for 2D FFT
    let sizes_2d = [(32, 32), (64, 64), (128, 128), (256, 256)];

    // Number of iterations for each test
    let iterations = 10;

    benchmark_1d_ffts(&sizes_1d, iterations);
    benchmark_2d_ffts(&sizes_2d, iterations);
}

/// Benchmark 1D FFT operations
#[allow(dead_code)]
fn benchmark_1d_ffts(sizes: &[usize], iterations: usize) {
    println!("\n1D FFT Memory Usage Benchmarking");
    println!("--------------------------------");
    println!(
        "{:>10} | {:>15} | {:>15} | {:>15} | {:>15}",
        "Size", "Standard (ms)", "Efficient (ms)", "Planned (ms)", "Best Speedup"
    );
    println!("{:-<75}", "");

    // Create plan cache for reuse
    let mut _plan_cache = PlanCache::new();

    for &size in sizes {
        // Create test signal
        let signal = create_test_signal(size);

        // Warmup - convert to complex for FFT functions
        let complex_signal: Vec<Complex64> =
            signal.iter().map(|&x| Complex64::new(x, 0.0)).collect();
        let _ = fft(&complex_signal, None).expect("Operation failed");
        let _ = optimized_fft(&complex_signal).expect("Operation failed");

        // Third warmup - using standard FFT since we can't use plan_cache API directly anymore
        let _ = fft(&complex_signal, None).expect("Operation failed");

        // Benchmark standard FFT
        let mut total_standard = Duration::from_nanos(0);
        for _ in 0..iterations {
            let start = Instant::now();
            let _ = fft(&complex_signal, None).expect("Operation failed");
            total_standard += start.elapsed();
        }

        // Benchmark memory-efficient FFT
        let mut total_efficient = Duration::from_nanos(0);
        for _ in 0..iterations {
            let start = Instant::now();
            let _ = optimized_fft(&complex_signal).expect("Operation failed");
            total_efficient += start.elapsed();
        }

        // Reset plan cache for fresh measurement - just for consistency with original
        _plan_cache = PlanCache::new();

        // First run with plan (using standard FFT as proxy)
        let start_first = Instant::now();
        let _ = fft(&complex_signal, None).expect("Operation failed");
        let first_plan_time = start_first.elapsed();

        // Benchmark plan-cached FFT (subsequent runs)
        let mut total_planned = Duration::from_nanos(0);
        for _ in 0..iterations {
            let start = Instant::now();
            let _ = fft(&complex_signal, None).expect("Operation failed");
            total_planned += start.elapsed();
        }

        // Calculate averages
        let avg_standard = total_standard.as_secs_f64() * 1000.0 / iterations as f64;
        let avg_efficient = total_efficient.as_secs_f64() * 1000.0 / iterations as f64;
        let avg_planned = total_planned.as_secs_f64() * 1000.0 / iterations as f64;

        // Find best approach
        let min_time = avg_standard.min(avg_efficient).min(avg_planned);
        let best_speedup = if min_time > 0.0 {
            avg_standard / min_time
        } else {
            0.0
        };

        println!(
            "{:>10} | {:>15.2} | {:>15.2} | {:>15.2} | {:>15.2}x",
            size, avg_standard, avg_efficient, avg_planned, best_speedup
        );

        // Show first plan creation time vs cached time
        println!(
            "           Plan creation: {:.2} ms, Cached: {:.2} ms, Cache speedup: {:.2}x",
            first_plan_time.as_secs_f64() * 1000.0,
            avg_planned,
            if avg_planned > 0.0 {
                first_plan_time.as_secs_f64() * 1000.0 / avg_planned
            } else {
                0.0
            }
        );
    }

    // Note that we can't show cache statistics since the API has changed
    println!("\nPlan cache used internal caching mechanism");
}

/// Benchmark 2D FFT operations
#[allow(dead_code)]
fn benchmark_2d_ffts(sizes: &[(usize, usize)], iterations: usize) {
    println!("\n2D FFT Memory Usage Benchmarking");
    println!("--------------------------------");
    println!(
        "{:>15} | {:>15} | {:>15} | {:>15} | {:>15}",
        "Size", "Standard (ms)", "Efficient (ms)", "Planned (ms)", "Best Speedup"
    );
    println!("{:-<80}", "");

    // Create plan cache for reuse
    let mut _plan_cache = PlanCache::new();

    for &(rows, cols) in sizes {
        // Create test array
        let signal = create_test_array(rows, cols);

        // Warmup
        let _ = fft2(&signal, None, None, None).expect("Operation failed");
        let _ = optimized_fft2(&signal, None).expect("Operation failed");
        // Third warmup using standard FFT2
        let _ = fft2(&signal, None, None, None).expect("Operation failed");

        // Benchmark standard FFT2
        let mut total_standard = Duration::from_nanos(0);
        for _ in 0..iterations {
            let start = Instant::now();
            let _ = fft2(&signal, None, None, None).expect("Operation failed");
            total_standard += start.elapsed();
        }

        // Benchmark memory-efficient FFT2
        let mut total_efficient = Duration::from_nanos(0);
        for _ in 0..iterations {
            let start = Instant::now();
            let _ = optimized_fft2(&signal, None).expect("Operation failed");
            total_efficient += start.elapsed();
        }

        // Reset plan cache for fresh measurement
        _plan_cache = PlanCache::new();

        // First run with plan creation (using standard FFT2 as proxy)
        let start_first = Instant::now();
        let _ = fft2(&signal, None, None, None).expect("Operation failed");
        let first_plan_time = start_first.elapsed();

        // Benchmark plan-cached FFT2 (subsequent runs)
        let mut total_planned = Duration::from_nanos(0);
        for _ in 0..iterations {
            let start = Instant::now();
            let _ = fft2(&signal, None, None, None).expect("Operation failed");
            total_planned += start.elapsed();
        }

        // Calculate averages
        let avg_standard = total_standard.as_secs_f64() * 1000.0 / iterations as f64;
        let avg_efficient = total_efficient.as_secs_f64() * 1000.0 / iterations as f64;
        let avg_planned = total_planned.as_secs_f64() * 1000.0 / iterations as f64;

        // Find best approach
        let min_time = avg_standard.min(avg_efficient).min(avg_planned);
        let best_speedup = if min_time > 0.0 {
            avg_standard / min_time
        } else {
            0.0
        };

        println!(
            "{:>6}x{:<8} | {:>15.2} | {:>15.2} | {:>15.2} | {:>15.2}x",
            rows, cols, avg_standard, avg_efficient, avg_planned, best_speedup
        );

        // Show first plan creation time vs cached time
        println!(
            "              Plan creation: {:.2} ms, Cached: {:.2} ms, Cache speedup: {:.2}x",
            first_plan_time.as_secs_f64() * 1000.0,
            avg_planned,
            if avg_planned > 0.0 {
                first_plan_time.as_secs_f64() * 1000.0 / avg_planned
            } else {
                0.0
            }
        );
    }

    // Note that we can't show cache statistics since the API has changed
    println!("\nPlan cache used internal caching mechanism");

    // Report memory usage comparison based on our estimation
    println!("\nEstimated memory usage comparison:");
    println!(
        "{:>15} | {:>15} | {:>15} | {:>15}",
        "Size", "Standard (MB)", "Efficient (MB)", "Reduction (%)"
    );
    println!("{:-<70}", "");

    for &(rows, cols) in sizes {
        let size = rows * cols;
        let std_mem =
            size as f64 * std::mem::size_of::<Complex64>() as f64 * 3.5 / (1024.0 * 1024.0);
        let eff_mem =
            size as f64 * std::mem::size_of::<Complex64>() as f64 * 2.0 / (1024.0 * 1024.0);
        let reduction = 100.0 * (std_mem - eff_mem) / std_mem;

        println!(
            "{:>6}x{:<8} | {:>15.2} | {:>15.2} | {:>15.2}%",
            rows, cols, std_mem, eff_mem, reduction
        );
    }
}

/// Create a test signal with sine waves for 1D FFT
#[allow(dead_code)]
fn create_test_signal(size: usize) -> Vec<f64> {
    let mut signal = Vec::with_capacity(size);
    for i in 0..size {
        let x = i as f64 / size as f64;
        let value = (2.0 * PI * 4.0 * x).sin() + 0.5 * (2.0 * PI * 8.0 * x).sin();
        signal.push(value);
    }
    signal
}

/// Create a test array with 2D patterns for 2D FFT
#[allow(dead_code)]
fn create_test_array(rows: usize, cols: usize) -> Array2<Complex64> {
    let mut array = Array2::zeros((rows, cols));
    for i in 0..rows {
        for j in 0..cols {
            let x = i as f64 / rows as f64;
            let y = j as f64 / cols as f64;
            let value = (2.0 * PI * 4.0 * x).sin() * (2.0 * PI * 4.0 * y).cos();
            array[[i, j]] = Complex64::new(value, 0.0);
        }
    }
    array
}

/// Memory-optimized FFT implementation using memory-efficient algorithms
#[allow(dead_code)]
fn optimized_fft(input: &[Complex64]) -> scirs2_fft::error::FFTResult<Vec<Complex64>> {
    // Use the in-place implementation with appropriate buffer sizing
    let mut input_clone = input.to_vec();
    let mut output = vec![Complex64::new(0.0, 0.0); input.len()];
    fft_inplace(&mut input_clone, &mut output, FftMode::Forward, true)?;
    Ok(output)
}

/// Memory-optimized FFT2 implementation
#[allow(dead_code)]
fn optimized_fft2(
    input: &Array2<Complex64>,
    shape: Option<(usize, usize)>,
) -> scirs2_fft::error::FFTResult<Array2<Complex64>> {
    // Get shape
    let rows_cols = match shape {
        Some(s) => s,
        None => {
            let shape = input.shape();
            (shape[0], shape[1])
        }
    };

    // Use memory-efficient fft2 implementation
    let view = input.view();
    fft2_efficient(&view, None, FftMode::Forward, true)
}

/// Memory usage tracking utilities
mod memory_tracking {
    use std::sync::atomic::{AtomicUsize, Ordering};

    // Global counter for active allocations
    #[allow(dead_code)]
    static ACTIVE_ALLOCATIONS: AtomicUsize = AtomicUsize::new(0);
    #[allow(dead_code)]
    static TOTAL_ALLOCATED: AtomicUsize = AtomicUsize::new(0);
    #[allow(dead_code)]
    static PEAK_ALLOCATED: AtomicUsize = AtomicUsize::new(0);

    /// Reset memory tracking counters
    #[allow(dead_code)]
    pub fn reset_counters() {
        ACTIVE_ALLOCATIONS.store(0, Ordering::SeqCst);
        TOTAL_ALLOCATED.store(0, Ordering::SeqCst);
        PEAK_ALLOCATED.store(0, Ordering::SeqCst);
    }

    /// Get current memory usage statistics
    #[allow(dead_code)]
    pub fn get_stats() -> MemoryStats {
        MemoryStats {
            active_allocations: ACTIVE_ALLOCATIONS.load(Ordering::SeqCst),
            total_allocated: TOTAL_ALLOCATED.load(Ordering::SeqCst),
            peak_allocated: PEAK_ALLOCATED.load(Ordering::SeqCst),
        }
    }

    /// Record a memory allocation
    #[allow(dead_code)]
    pub fn record_allocation(size: usize) {
        ACTIVE_ALLOCATIONS.fetch_add(1, Ordering::SeqCst);
        TOTAL_ALLOCATED.fetch_add(size, Ordering::SeqCst);

        // Update peak memory usage
        loop {
            let current = TOTAL_ALLOCATED.load(Ordering::SeqCst);
            let peak = PEAK_ALLOCATED.load(Ordering::SeqCst);

            if current <= peak {
                break;
            }

            if PEAK_ALLOCATED
                .compare_exchange(peak, current, Ordering::SeqCst, Ordering::SeqCst)
                .is_ok()
            {
                break;
            }
        }
    }

    /// Record a memory deallocation
    #[allow(dead_code)]
    pub fn record_deallocation(size: usize) {
        ACTIVE_ALLOCATIONS.fetch_sub(1, Ordering::SeqCst);
        TOTAL_ALLOCATED.fetch_sub(size, Ordering::SeqCst);
    }

    /// Memory usage statistics
    #[derive(Debug, Clone, Copy)]
    #[allow(dead_code)]
    pub struct MemoryStats {
        pub active_allocations: usize,
        pub total_allocated: usize,
        pub peak_allocated: usize,
    }

    impl MemoryStats {
        /// Format memory size in human-readable form
        #[allow(dead_code)]
        pub fn formatsize(size: usize) -> String {
            if size < 1024 {
                format!("{} B", size)
            } else if size < 1024 * 1024 {
                format!("{:.2} KB", size as f64 / 1024.0)
            } else if size < 1024 * 1024 * 1024 {
                format!("{:.2} MB", size as f64 / (1024.0 * 1024.0))
            } else {
                format!("{:.2} GB", size as f64 / (1024.0 * 1024.0 * 1024.0))
            }
        }

        /// Get peak memory usage in human-readable form
        #[allow(dead_code)]
        pub fn peak_memory_str(&self) -> String {
            Self::formatsize(self.peak_allocated)
        }
    }
}