pub fn k4_ssd_state_passing<B: Backend>(
intra_chunk_state_bnhpr: Tensor<B, 5>,
da_chunk_end_bhn: Tensor<B, 3>,
initial_state_bhpr: Tensor<B, 4>,
) -> (Tensor<B, 5>, Tensor<B, 4>)Expand description
Based on the Kernel 4 Triton reference _state_passing_fwd_kernel (ssd_state_passing.py).
Returns:
- chunk_input_state_bnhpr used in K5.
- final_state_bhpr [final output].