t4a-cubecl-runtime 0.11.0

The runtime API of CubeCL: clients, devices, handles and everything a kernel launch needs.
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

//! The device type of each runtime, named without naming the runtime crate.
//!
//! Each one is a value and nothing more — which GPU, which index — so they
//! live here rather than in the runtime crates that act on them. That is what
//! lets a caller hold one, and lets `cubecl` put a single enum over them all,
//! without depending on any runtime.

mod cpu;
mod cuda;
mod hip;
mod metal;
mod wgpu;

pub use cpu::CpuDevice;
pub use cuda::CudaDevice;
pub use hip::AmdDevice;
pub use metal::MetalDevice;
pub use wgpu::{WgpuBackend, WgpuDevice, WgpuDeviceKind};

pub use cubecl_common::device::DeviceId;

/// A device index as a [`DeviceId`] carries it.
///
/// Kept at the largest the id holds rather than wrapped: wrapped, an index
/// past the end lands on a device that exists; kept, it names one no machine
/// has, and naming it is refused.
fn index_id(index: usize) -> u16 {
    index.min(u16::MAX as usize) as u16
}