flow-utils 0.1.1

Shared algorithms and utilities for flow cytometry crates
Documentation 
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

//! GPU-accelerated FFT operations for KDE
//!
//! Uses GPU for complex multiplication in frequency domain (FFT convolution step).
//! CPU FFT is used for the actual transforms.

#[cfg(feature = "gpu")]
use burn::backend::wgpu::WgpuDevice;
#[cfg(feature = "gpu")]
use burn::tensor::Tensor;
use crate::kde::{KdeError, KdeResult};
use realfft::num_complex::Complex;

#[cfg(feature = "gpu")]
type Backend = burn::backend::wgpu::Wgpu;

/// Check if GPU is available
#[cfg(feature = "gpu")]
pub fn is_gpu_available() -> bool {
    // Simple check - in a real implementation, this would check for actual GPU availability
    // For now, we'll use a simple heuristic or environment variable
    std::env::var("FLOW_UTILS_USE_GPU").is_ok()
}

#[cfg(not(feature = "gpu"))]
pub fn is_gpu_available() -> bool {
    false
}

/// GPU-accelerated FFT-based KDE
///
/// Uses GPU for convolution multiplication and other operations,
/// while using CPU FFT for the actual transforms.
#[cfg(feature = "gpu")]
pub fn kde_fft_gpu(
    data: &[f64],
    grid: &[f64],
    bandwidth: f64,
    n: f64,
) -> KdeResult<Vec<f64>> {
    use crate::kde::fft::kde_fft;
    use crate::common::gaussian_kernel;
    
    // For now, fall back to CPU implementation
    // GPU implementation can be added later if needed
    kde_fft(data, grid, bandwidth, n)
}

#[cfg(not(feature = "gpu"))]
pub fn kde_fft_gpu(
    _data: &[f64],
    _grid: &[f64],
    _bandwidth: f64,
    _n: f64,
) -> KdeResult<Vec<f64>> {
    Err(KdeError::StatsError("GPU support not compiled in".to_string()))
}