solvr 0.2.0

Advanced computing library for real-world problem solving - optimization, differential equations, interpolation, statistics, and more
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

//! CUDA implementation of local extrema detection algorithms.
//!
//! This algorithm is CPU-only due to its sequential comparison patterns.
//! GPU implementations are not efficient for variable-order neighborhood comparisons.

use crate::signal::traits::extrema::{ExtremaAlgorithms, ExtremaResult, ExtremumMode};
use numr::error::{Error, Result};
use numr::runtime::cuda::{CudaClient, CudaRuntime};
use numr::tensor::Tensor;

impl ExtremaAlgorithms<CudaRuntime> for CudaClient {
    fn argrelmin(
        &self,
        _x: &Tensor<CudaRuntime>,
        _order: usize,
        _mode: ExtremumMode,
    ) -> Result<ExtremaResult<CudaRuntime>> {
        Err(Error::backend_limitation(
            "cuda",
            "argrelmin",
            "Local extrema detection is CPU-only due to sequential comparison patterns. Transfer data to CPU first.",
        ))
    }

    fn argrelmax(
        &self,
        _x: &Tensor<CudaRuntime>,
        _order: usize,
        _mode: ExtremumMode,
    ) -> Result<ExtremaResult<CudaRuntime>> {
        Err(Error::backend_limitation(
            "cuda",
            "argrelmax",
            "Local extrema detection is CPU-only due to sequential comparison patterns. Transfer data to CPU first.",
        ))
    }
}