fix clang formatting issues

cpp/include/raft/distance/detail/fused_distance_nn.cuh

@@ -16,8 +16,6 @@
 
 #pragma once
 
-#include <cstddef>                                       // size_t
-#include <limits>                                        // std::numeric_limits
 #include <raft/core/kvp.hpp>                             // raft::KeyValuePair
 #include <raft/core/operators.hpp>                       // raft::identity_op
 #include <raft/distance/detail/distance_ops/l2_exp.cuh>  // ops::l2_exp_distance_op
@@ -32,6 +30,9 @@
 #include <raft/util/arch.cuh>            // raft::util::arch::SM_*
 #include <raft/util/cuda_utils.cuh>      // raft::ceildiv, raft::shfl
 
+#include <cstddef>  // size_t
+#include <limits>   // std::numeric_limits
+
 namespace raft {
 namespace distance {
 

cpp/include/raft/distance/detail/fused_distance_nn/cutlass_base.cuh

@@ -26,25 +26,15 @@
 #define cutlass raft_cutlass
 #endif
 
-#include <cutlass/cutlass.h>
-#include <cutlass/gemm/device/gemm.h>
-#include <cutlass/gemm/device/gemm_grouped.h>
-#include <cutlass/gemm/device/gemm_universal_adapter.h>
-#include <rmm/device_uvector.hpp>
-
-#include <cutlass/layout/matrix.h>
-#include <cutlass/layout/tensor.h>
-#include <cutlass/matrix_coord.h>
-#include <cutlass/tensor_view.h>
-
-#include <cuda/semaphore>
 #include <raft/distance/detail/fused_distance_nn/epilogue_elementwise.cuh>  // FusedDistanceNNEpilogueElementwise
 #include <raft/distance/detail/fused_distance_nn/gemm.h>                    // FusedDistanceNNGemm
 #include <raft/util/cudart_utils.hpp>   // getMultiProcessorCount
 #include <raft/util/cutlass_utils.cuh>  // RAFT_CUTLASS_TRY
 
 #include <rmm/device_uvector.hpp>
 
+#include <cuda/semaphore>
+
 #include <cutlass/cutlass.h>
 #include <cutlass/gemm/device/gemm.h>
 #include <cutlass/gemm/device/gemm_grouped.h>

cpp/include/raft/distance/detail/fused_distance_nn/epilogue_elementwise.cuh

@@ -56,16 +56,15 @@
 
 #pragma once
 
+#include <cuda/semaphore>
+
 #include <cutlass/array.h>
 #include <cutlass/cutlass.h>
 #include <cutlass/epilogue/thread/activation.h>
 #include <cutlass/functional.h>
 #include <cutlass/numeric_conversion.h>
 #include <cutlass/numeric_types.h>
 
-#include <cuda/semaphore>
-#include <cutlass/epilogue/thread/activation.h>
-
 /////////////////////////////////////////////////////////////////////////////////////////////////
 
 namespace cutlass {

cpp/include/raft/distance/detail/fused_distance_nn/fused_cosine_nn.cuh

@@ -16,8 +16,6 @@
 
 #pragma once
 
-#include <cstddef>                                       // size_t
-#include <limits>                                        // std::numeric_limits
 #include <raft/core/kvp.hpp>                             // raft::KeyValuePair
 #include <raft/core/operators.hpp>                       // raft::identity_op
 #include <raft/distance/detail/distance_ops/cosine.cuh>  // ops::l2_exp_distance_op
@@ -29,6 +27,9 @@
 #include <raft/util/arch.cuh>                               // raft::util::arch::SM_*
 #include <raft/util/cuda_utils.cuh>                         // raft::ceildiv, raft::shfl
 
+#include <cstddef>  // size_t
+#include <limits>   // std::numeric_limits
+
 namespace raft {
 namespace distance {
 

cpp/include/raft/distance/detail/fused_distance_nn/fused_l2_nn.cuh

@@ -16,8 +16,6 @@
 
 #pragma once
 
-#include <cstddef>                                       // size_t
-#include <limits>                                        // std::numeric_limits
 #include <raft/core/kvp.hpp>                             // raft::KeyValuePair
 #include <raft/core/operators.hpp>                       // raft::identity_op
 #include <raft/distance/detail/distance_ops/l2_exp.cuh>  // ops::l2_exp_distance_op
@@ -29,6 +27,9 @@
 #include <raft/util/arch.cuh>                               // raft::util::arch::SM_*
 #include <raft/util/cuda_utils.cuh>                         // raft::ceildiv, raft::shfl
 
+#include <cstddef>  // size_t
+#include <limits>   // std::numeric_limits
+
 namespace raft {
 namespace distance {
 

cpp/include/raft/distance/detail/fused_distance_nn/helper_structs.cuh

@@ -16,8 +16,6 @@
 
 #pragma once
 
-#include <cstddef>                                       // size_t
-#include <limits>                                        // std::numeric_limits
 #include <raft/core/kvp.hpp>                             // raft::KeyValuePair
 #include <raft/core/operators.hpp>                       // raft::identity_op
 #include <raft/distance/detail/distance_ops/l2_exp.cuh>  // ops::l2_exp_distance_op
@@ -29,6 +27,9 @@
 #include <raft/util/cuda_utils.cuh>                         // raft::ceildiv, raft::shfl
 #include <raft/util/device_atomics.cuh>
 
+#include <cstddef>  // size_t
+#include <limits>   // std::numeric_limits
+
 namespace raft {
 namespace distance {
 

cpp/include/raft/distance/detail/fused_distance_nn/simt_kernel.cuh

@@ -16,13 +16,14 @@
 
 #pragma once
 
-#include <cstddef>                                          // size_t
-#include <limits>                                           // std::numeric_limits
 #include <raft/core/kvp.hpp>                                // raft::KeyValuePair
 #include <raft/distance/detail/distance_ops/l2_exp.cuh>     // ops::l2_exp_distance_op
 #include <raft/distance/detail/pairwise_distance_base.cuh>  // PairwiseDistances
 #include <raft/linalg/contractions.cuh>                     // Policy
 
+#include <cstddef>  // size_t
+#include <limits>   // std::numeric_limits
+
 namespace raft {
 namespace distance {
 namespace detail {

cpp/include/raft/distance/fused_distance_nn-ext.cuh

@@ -16,12 +16,13 @@
 
 #pragma once
 
-#include <cstdint>                                      // int64_t
 #include <raft/core/kvp.hpp>                            // raft::KeyValuePair
 #include <raft/core/resources.hpp>                      // raft::resources
 #include <raft/distance/fused_distance_nn_helpers.cuh>  // include initialize and reduce operations
 #include <raft/util/raft_explicit.hpp>                  // RAFT_EXPLICIT
 
+#include <cstdint>  // int64_t
+
 #ifdef RAFT_EXPLICIT_INSTANTIATE_ONLY
 
 namespace raft {

cpp/include/raft/distance/fused_distance_nn-inl.cuh

@@ -19,14 +19,17 @@
 
 #pragma once
 
-#include <cub/cub.cuh>
-#include <limits>
 #include <raft/core/resources.hpp>
 #include <raft/distance/detail/fused_distance_nn.cuh>
 #include <raft/distance/fused_distance_nn_helpers.cuh>
 #include <raft/linalg/contractions.cuh>
 #include <raft/util/cuda_utils.cuh>
+
+#include <cub/cub.cuh>
+
 #include <stdint.h>
+
+#include <limits>
 #include <type_traits>
 
 namespace raft {

cpp/include/raft/distance/fused_l2_nn-ext.cuh

@@ -16,10 +16,10 @@
 
 #pragma once
 
-#include <raft/core/kvp.hpp>                      // raft::KeyValuePair
-#include <raft/core/resources.hpp>                // raft::resources
+#include <raft/core/kvp.hpp>                            // raft::KeyValuePair
+#include <raft/core/resources.hpp>                      // raft::resources
 #include <raft/distance/fused_distance_nn_helpers.cuh>  // include initialize and reduce operations
-#include <raft/util/raft_explicit.hpp>            // RAFT_EXPLICIT
+#include <raft/util/raft_explicit.hpp>                  // RAFT_EXPLICIT
 
 #include <cstdint>  // int64_t
 

cpp/src/distance/fused_distance_nn.cu

@@ -14,10 +14,11 @@
  * limitations under the License.
  */
 
-#include <cstdint>            // int64_t
 #include <raft/core/kvp.hpp>  // raft::KeyValuePair
 #include <raft/distance/fused_distance_nn-inl.cuh>
 
+#include <cstdint>  // int64_t
+
 #define instantiate_raft_distance_fusedDistanceNNMinReduce(DataT, OutT, IdxT) \
   template void raft::distance::fusedDistanceNNMinReduce<DataT, OutT, IdxT>(  \
     OutT * min,                                                               \

cpp/src/raft_runtime/distance/fused_distance_min_arg.cu

@@ -15,13 +15,15 @@
  */
 
 #include "fused_distance_min_arg.hpp"
+
 #include <raft/core/device_mdarray.hpp>
 #include <raft/core/kvp.hpp>
 #include <raft/core/resource/cuda_stream.hpp>
 #include <raft/core/resource/thrust_policy.hpp>
 #include <raft/core/resources.hpp>
 #include <raft/distance/distance_types.hpp>
 #include <raft/linalg/norm.cuh>
+
 #include <thrust/for_each.h>
 #include <thrust/tuple.h>
 

cpp/src/raft_runtime/distance/fused_distance_min_arg.hpp

@@ -25,6 +25,7 @@
 #include <raft/distance/fused_distance_nn.cuh>
 #include <raft/distance/fused_l2_nn.cuh>
 #include <raft/linalg/norm.cuh>
+
 #include <thrust/for_each.h>
 #include <thrust/tuple.h>
 

cpp/src/raft_runtime/distance/fused_l2_min_arg.cu

@@ -15,6 +15,7 @@
  */
 
 #include "fused_distance_min_arg.hpp"
+
 #include <raft/core/device_mdarray.hpp>
 #include <raft/core/kvp.hpp>
 #include <raft/core/resource/cuda_stream.hpp>

cpp/test/distance/fused_cosine_nn.cu

@@ -17,7 +17,7 @@
 #undef RAFT_EXPLICIT_INSTANTIATE_ONLY  // Search with filter instantiation
 
 #include "../test_utils.cuh"
-#include <gtest/gtest.h>
+
 #include <raft/core/kvp.hpp>
 #include <raft/core/resource/cuda_stream.hpp>
 #include <raft/distance/detail/fused_distance_nn.cuh>
@@ -27,6 +27,8 @@
 #include <raft/util/cuda_utils.cuh>
 #include <raft/util/cudart_utils.hpp>
 
+#include <gtest/gtest.h>
+
 namespace raft {
 namespace distance {