Remove deprecated kernels (#3316)

* remove

* remove

* fix

* remove

* remove

src/kernel/cuda/utils.cu

-/*!
- *  Copyright (c) 2019 by Contributors
- * \file kernel/cuda/utils.cu
- * \brief Utility function implementations on CUDA
- */
-#include "../../runtime/cuda/cuda_common.h"
-#include "../utils.h"
-
-namespace dgl {
-namespace kernel {
-namespace utils {
-
-template <typename DType>
-__global__ void _FillKernel(DType* ptr, size_t length, DType val) {
-  int tx = blockIdx.x * blockDim.x + threadIdx.x;
-  int stride_x = gridDim.x * blockDim.x;
-  while (tx < length) {
-    ptr[tx] = val;
-    tx += stride_x;
-  }
-}
-
-template <int XPU, typename DType>
-void Fill(const DLContext& ctx, DType* ptr, size_t length, DType val) {
-  auto* thr_entry = runtime::CUDAThreadEntry::ThreadLocal();
-  int nt = utils::FindNumThreads(length, 1024);
-  int nb = (length + nt - 1) / nt;
-  CUDA_KERNEL_CALL(_FillKernel, nb, nt, 0, thr_entry->stream, ptr, length, val);
-}
-
-template void Fill<kDLGPU, float>(const DLContext& ctx, float* ptr, size_t length, float val);
-template void Fill<kDLGPU, double>(const DLContext& ctx, double* ptr, size_t length, double val);
-
-}  // namespace utils
-}  // namespace kernel
-}  // namespace dgl

src/runtime/cuda/cuda_hashtable.cu

@@ -7,10 +7,10 @@
 #include <cassert>
 
 #include "cuda_hashtable.cuh"
-#include "../../kernel/cuda/atomic.cuh"
+#include "../../array/cuda/atomic.cuh"
 #include "../../array/cuda/dgl_cub.cuh"
 
-using namespace dgl::kernel::cuda;
+using namespace dgl::aten::cuda;
 
 namespace dgl {
 namespace runtime {

src/runtime/cuda/nccl_api.cu

@@ -29,7 +29,6 @@
 #include "cuda_common.h"
 #include "../../runtime/workspace.h"
 #include "../../partition/ndarray_partition.h"
-#include "../../kernel/cuda/atomic.cuh"
 #include "../../array/cuda/dgl_cub.cuh"
 #include "../../array/cuda/array_index_select.cuh"
 
@@ -44,7 +43,6 @@
 
 namespace dgl {
 
-using namespace kernel::cuda;
 using namespace partition;
 
 namespace runtime {