[Core] direct indexing on self.block_table_np in compute_slot_mapping (#22940)

Signed-off-by: linzebing <[email protected]>

Author: linzebing

vllm/v1/worker/block_table.py

@@ -91,8 +91,7 @@ def compute_slot_mapping(self, req_indices: np.ndarray,
         # block_size.
         block_table_indices = (req_indices * self.max_num_blocks_per_req +
                                positions // self.block_size)
-        block_table_cpu = self.get_cpu_tensor()
-        block_numbers = block_table_cpu.flatten()[block_table_indices].numpy()
+        block_numbers = self.block_table_np.ravel()[block_table_indices]
         block_offsets = positions % self.block_size
         np.add(block_numbers * self.block_size,
                block_offsets,