pub fn fill_slots_from_block_table(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
slots: DevicePtr,
block_table: DevicePtr,
start_pos: u32,
count: u32,
block_size: u32,
stream: u64,
) -> Result<()>Expand description
Fill paged-KV slot mappings on-device from a persistent block table.
Kernel: fill_slots_from_block_table(slots, block_table, start_pos, count, block_size)
Grid: (ceil(count/256), 1, 1) Block: (256, 1, 1)