dspk-core 0.0.10

Core types for dspk: recordings, buffers, time, masks, windows, and the CubeCL compute layer with run-time device choice (wgpu, CUDA, HIP, CPU)
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

//! Special float values inside kernels.
//!
//! WGSL has no literal for infinity, so `F::new(f32::INFINITY)` fails to compile on the wgpu
//! runtime. These helpers build ±∞ from their IEEE-754 bits and cast to the kernel's float type
//! (an `f32` infinity casts to infinity in every float type).

use cubecl::prelude::*;

/// IEEE-754 bits of `f32` +∞.
pub const POSITIVE_INFINITY_BITS: u32 = f32::INFINITY.to_bits();

/// IEEE-754 bits of `f32` −∞.
pub const NEGATIVE_INFINITY_BITS: u32 = f32::NEG_INFINITY.to_bits();

/// +∞ in `F`.
#[cube]
pub fn positive_infinity<F: Float>() -> F {
    F::cast_from(f32::reinterpret(POSITIVE_INFINITY_BITS))
}

/// −∞ in `F`.
#[cube]
pub fn negative_infinity<F: Float>() -> F {
    F::cast_from(f32::reinterpret(NEGATIVE_INFINITY_BITS))
}