diff --git a/cpp/src/wholegraph_ops/block_topk_with_raft.cuh b/cpp/src/wholegraph_ops/block_topk_with_raft.cuh
deleted file mode 100644
index db5a436b5..000000000
--- a/cpp/src/wholegraph_ops/block_topk_with_raft.cuh
+++ /dev/null
@@ -1,109 +0,0 @@
-#pragma once
-#include <raft/matrix/detail/select_k-inl.cuh>
-namespace wholegraph_ops {
-template <typename T, typename IdxT>
-constexpr auto calc_smem_size_for_block_wide(int num_of_warp, int k) -> int
-{
-  return raft::Pow2<256>::roundUp(raft::ceildiv(num_of_warp, 2) * sizeof(T) * k) +
-         raft::ceildiv(num_of_warp, 2) * sizeof(IdxT) * k;
-}
-
-template <typename KeyT,
-          int BLOCK_SIZE,
-          int ITEMS_PER_THREAD,
-          int MAXK,
-          bool ASCENDING  = false,
-          typename ValueT = cub::NullType,
-          template <int, bool, typename, typename> class WarpSortClassT =
-            raft::matrix::detail::select::warpsort::warp_sort_distributed_ext>
-class BlockTopkRaftWarpSort {
-  static_assert(MAXK <= raft::matrix::detail::select::warpsort::kMaxCapacity,
-                "MAXK should be smaller than warpsort::kMaxCapacity");
-  static_assert(MAXK >= 1 && raft::is_a_power_of_two(MAXK),
-                "MAXK should >=1 and is a power of two ");
-
-  using bq_t = raft::matrix::detail::select::warpsort::
-    block_sort<WarpSortClassT, MAXK, ASCENDING, KeyT, ValueT>;
-
-  static constexpr int WARP_SIZE = 32;
-  static constexpr int CAL_SMEM_SIZE =
-    calc_smem_size_for_block_wide<KeyT, ValueT>(BLOCK_SIZE / WARP_SIZE, MAXK);
-  static constexpr int SMEM_REQUIRED = bq_t::queue_t::mem_required(BLOCK_SIZE);
-  struct _TempStorage {
-    union {
-      __align__(256) uint8_t smem_buf_bytes0[CAL_SMEM_SIZE];
-      __align__(256) uint8_t smem_buf_bytes1[SMEM_REQUIRED];
-      struct {
-        KeyT store_keys[MAXK];
-        ValueT store_values[MAXK];
-      };
-    };
-  };
-
- public:
-  struct TempStorage : cub::Uninitialized<_TempStorage> {};
-
-  __device__ __forceinline__ BlockTopkRaftWarpSort(TempStorage& temp_storage)
-    : temp_storage_{temp_storage.Alias()}, tid_(threadIdx.x){};
-
-  __device__ __forceinline__ void TopKToStriped(KeyT (&keys)[ITEMS_PER_THREAD],
-                                                ValueT (&values)[ITEMS_PER_THREAD],
-                                                const int k,
-                                                const int valid_count)
-  {
-    bq_t queue(k, temp_storage_.smem_buf_bytes1);
-
-#pragma unroll
-    for (int i = 0; i < ITEMS_PER_THREAD; i++) {
-      KeyT key = (i * BLOCK_SIZE + tid_) < valid_count
-                   ? keys[i]
-                   : WarpSortClassT<MAXK, ASCENDING, KeyT, ValueT>::kDummy;
-      queue.add(key, values[i]);
-    }
-    queue.done(temp_storage_.smem_buf_bytes0);
-    __syncthreads();
-    queue.store(temp_storage_.store_keys, temp_storage_.store_values);
-    __syncthreads();
-#pragma unroll
-    for (int i = 0; i < ITEMS_PER_THREAD; i++) {
-      int id = i * BLOCK_SIZE + tid_;
-      if (id < k) {
-        keys[i]   = temp_storage_.store_keys[id];
-        values[i] = temp_storage_.store_values[id];
-      }
-    }
-  }
-
-  __device__ __forceinline__ void TopKToStriped(KeyT (&keys)[ITEMS_PER_THREAD],
-
-                                                const int k,
-                                                const int valid_count)
-  {
-    bq_t queue(k, temp_storage_.smem_buf_bytes1);
-
-#pragma unroll
-    for (int i = 0; i < ITEMS_PER_THREAD; i++) {
-      KeyT key = (i * BLOCK_SIZE + tid_) < valid_count
-                   ? keys[i]
-                   : WarpSortClassT<MAXK, ASCENDING, KeyT, ValueT>::kDummy;
-      queue.add(key, i);
-    }
-    queue.done(temp_storage_.smem_buf_bytes0);
-    __syncthreads();
-
-    queue.store(temp_storage_.store_keys, temp_storage_.store_values);
-    __syncthreads();
-
-#pragma unroll
-    for (int i = 0; i < ITEMS_PER_THREAD; i++) {
-      int id = i * BLOCK_SIZE + tid_;
-      if (id < k) { keys[i] = temp_storage_.store_keys[id]; }
-    }
-  }
-
- private:
-  _TempStorage& temp_storage_;
-  int tid_;
-};
-
-};  // namespace wholegraph_ops