Skip to main content

gpu_slice_write

ferrotorch_gpu::kernels

Function gpu_slice_write 

Source 
pub fn gpu_slice_write(
    src: &CudaBuffer<f32>,
    dst: &mut CudaBuffer<f32>,
    n_batch: usize,
    d: usize,
    max_len: usize,
    pos: usize,
    device: &GpuDevice,
) -> GpuResult<()>
Expand description

Write src of shape [N, D] into row pos of dst of shape [N, max_len, D]. This is an in-place GPU operation — dst is modified.