Skip to main content

reflection_pad1d_backward

teeny_kernels::nn::pad::reflection_pad1d

Function reflection_pad1d_backward 

Source 
pub fn reflection_pad1d_backward<T: Triton, D: Num, const PAD_LEFT: i32, const PAD_RIGHT: i32, const BLOCK_OL: i32>(
    dy_ptr: T::Pointer<D>,
    dx_ptr: T::Pointer<D>,
    _B: i32,
    C: i32,
    L: i32,
    OL: i32,
)
where
    T::I32Tensor: Tensor<i32, 1> + Comparison<i32, BoolTensor = T::BoolTensor>,
    T::BoolTensor: BitAnd<Output = T::BoolTensor> + BitOr<Output = T::BoolTensor>,
    T::Pointer<D>: AddOffsets<i32, 1, T::I32Tensor, Output = T::Tensor<T::Pointer<D>>>,
Expand description

1-D reflection padding backward pass.

Each output gradient position maps back to one input position via the same reflection rule. Multiple output positions may map to the same input position (the boundary elements reflect), so atomic_add is used. dx must be zero-initialised before launch.