poc_kokkos_rs/functor/
mod.rs

1mod for_kernel;
2mod policies;
3mod reduce_kernel;
4mod scan_kernel;
5
6pub use for_kernel::{ForFunctor, parallel_for};
7pub(crate) use policies::ExecutionPolicy;
8pub use policies::{
9    ExecutionSpace, MDRange, PerTeam, PerThread, Range, Schedule, TeamHandle, TeamPolicy,
10    TeamThreadMDRange, TeamThreadRange, TeamVectorMDRange, TeamVectorRange, ThreadVectorMDRange,
11    ThreadVectorRange,
12};
13
14// internal routines
15
16// / Builds a N-depth nested loop executing a kernel using the N resulting indices.
17// / Technically, this should be replaced by a tiling function, for both serial and parallel
18// / implementations.
19// fn recursive_loop<const N: usize>(ranges: &[usize; N], mut kernel: impl Fn([usize; N])) {
20//     // handles recursions
21//     fn inner<const N: usize>(
22//         current_depth: usize,
23//         ranges: &[usize; N],
24//         kernel: &mut impl Fn([usize; N]),
25//         indices: &mut [usize; N],
26//     ) {
27//         if current_depth == N {
28//             // all loops unraveled
29//             // call the kernel
30//             kernel(*indices)
31//         } else {
32//             // loop on next dimension; update indices
33//             // can we avoid a clone by passing a slice starting one element
34//             // after the unraveled range ?
35//             ranges[current_depth].clone().for_each(|i_current| {
36//                 indices[current_depth] = i_current;
37//                 inner(current_depth + 1, ranges, kernel, indices);
38//             });
39//         }
40//     }
41
42//     let mut indices = [0; N];
43//     inner(0, ranges, &mut kernel, &mut indices);
44// }
poc_kokkos_rs/functor/mod.rs

poc_kokkos_rs/functor/
mod.rs
