pub fn ragged_prefill_bf16_hd128(
q: u64,
k: u64,
v: u64,
o: u64,
qo_indptr_h: &[i32],
kv_indptr_h: &[i32],
qo_indptr_d: u64,
kv_indptr_d: u64,
batch: u32,
total_qo_rows: u32,
total_kv_rows: u32,
num_qo_heads: u32,
num_kv_heads: u32,
head_dim: u32,
sm_scale: f32,
causal: bool,
sliding_window: Option<u32>,
stream: u64,
) -> Result<()>Expand description
2026-09-25: Ragged batched prefill attention, BF16, head_dim 128, GQA.
The contract of super::ragged_prefill_bf16_hd256, plus sliding_window
in the in-tree convention (mask when q - k >= w, as in
kernels/gb10/common/attn_prefill.cu); None or Some(0) means no window.
It is passed to FlashInfer as window_left = w - 1.