INNER CODE UNIT · Rust

nccl4rust_lsa_barrier_cta

NVIDIA/nccl · contrib/nccl4rust/crates/nccl-device-sys/src/lib.rs:97

    pub fn nccl4rust_lsa_barrier_cta(comm: *const ncclDevComm_t, index: u32, multimem: bool);

    /// LSA f32 reduce-sum-copy using all threads in the CTA.
    ///
    /// This primitive does not add entry/exit barriers. Callers must establish
    /// the ordering required by the public NCCL reduce/copy API.
    pub fn nccl4rust_lsa_reduce_sum_copy_f32_cta(
        comm: *const ncclDevComm_t,
        src_window: ncclWindow_t,
        src_offset: usize,
        dst_window: ncclWindow_t,
        dst_offset: usize,
        count: usize,
    );
}

const _: () = {
    assert!(core::mem::size_of::<ncclTeam_t>() == 12);

View source record →