atomr-accel-cuda 0.10.0

GPU acceleration via the actor model. Wraps NVIDIA CUDA libraries (cuBLAS, cuDNN, cuFFT, cuRAND, cuSOLVER, cuSPARSE, cuTENSOR, cuBLASLt, NVRTC, NCCL) as supervised atomr actors with generation-validated buffers and a uniform async surface.
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

//! `SingleStreamAllocator` (§5.7) — every `KernelActor` shares one
//! stream. Useful for resource-constrained edge devices or
//! deterministic-replay testing.

use std::sync::Arc;

use cudarc::driver::CudaStream;

use super::{ActorHints, StreamAllocator};

pub struct SingleStreamAllocator {
    stream: Arc<CudaStream>,
}

impl SingleStreamAllocator {
    pub fn new(stream: Arc<CudaStream>) -> Self {
        Self { stream }
    }
}

impl StreamAllocator for SingleStreamAllocator {
    fn acquire(&self, _hints: ActorHints) -> Arc<CudaStream> {
        self.stream.clone()
    }
}