From 3d3dd3c45782cebe0151e24eafd49b21dd9da065 Mon Sep 17 00:00:00 2001
From: Bradley Dice <bdice@bradleydice.com>
Date: Thu, 7 Jul 2022 14:09:51 -0500
Subject: [PATCH] Clean up Thrust includes. (#4675)

## Description

This PR cleans up some `#include`s for Thrust. This is meant to help ease the transition to Thrust 1.17 when that is updated in rapids-cmake.

## Context

I opened a PR https://github.com/rapidsai/cudf/pull/10489 that updates cuDF to Thrust 1.16. Notably, Thrust reduced the number of internal header inclusions:
> [#1572](https://github.com/NVIDIA/thrust/pull/1572) Removed several unnecessary header includes. Downstream projects may need to update their includes if they were relying on this behavior.

I spoke with @robertmaynard and he recommended making similar changes to clean up includes ("include what we use," in essence) to make sure we have compatibility with future versions of Thrust across all RAPIDS libraries.

This changeset also removes dependence on `thrust/detail` headers.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - William Hicks (https://github.com/wphicks)

URL: https://github.com/rapidsai/cuml/pull/4675
---
 cpp/src/arima/batched_arima.cu                |  1 +
 cpp/src/arima/batched_kalman.cu               |  1 +
 cpp/src/dbscan/corepoints/compute.cuh         |  5 ++++-
 cpp/src/dbscan/runner.cuh                     |  5 +++++
 .../batched-levelalgo/kernels/quantiles.cu    |  3 +++
 cpp/src/explainer/tree_shap.cu                |  5 +++++
 cpp/src/fil/fil.cu                            |  2 +-
 cpp/src/genetic/fitness.cuh                   |  2 ++
 cpp/src/glm/qn/glm_base.cuh                   |  5 +++++
 cpp/src/hdbscan/condensed_hierarchy.cu        |  8 ++++++++
 cpp/src/hdbscan/detail/condense.cuh           |  4 ++++
 cpp/src/hdbscan/detail/extract.cuh            |  2 ++
 cpp/src/hdbscan/detail/membership.cuh         |  2 ++
 cpp/src/hdbscan/detail/reachability.cuh       |  3 +++
 cpp/src/hdbscan/detail/select.cuh             |  7 +++++++
 cpp/src/hdbscan/detail/stabilities.cuh        |  4 ++++
 cpp/src/hdbscan/detail/utils.h                |  7 ++++++-
 cpp/src/hdbscan/runner.h                      |  4 ++++
 cpp/src/hierarchy/pw_dist_graph.cuh           |  5 +++++
 cpp/src/kmeans/kmeans_mg_impl.cuh             |  6 ++++++
 cpp/src/kmeans/sg_impl.cuh                    |  3 +++
 cpp/src/pca/sign_flip_mg.cu                   |  2 ++
 cpp/src/randomforest/randomforest.cuh         |  8 +++++---
 cpp/src/solver/lars_impl.cuh                  |  3 +++
 cpp/src/svm/linear.cu                         |  2 ++
 cpp/src/svm/smosolver.cuh                     |  3 +++
 cpp/src/svm/svc_impl.cuh                      |  1 +
 cpp/src/svm/workingset.cuh                    |  2 ++
 cpp/src/tsne/barnes_hut_tsne.cuh              |  4 ++++
 cpp/src/tsne/distances.cuh                    |  3 +++
 cpp/src/tsne/exact_kernels.cuh                |  4 ++++
 cpp/src/tsne/exact_tsne.cuh                   |  4 ++++
 cpp/src/tsne/fft_tsne.cuh                     |  6 ++++++
 cpp/src/tsne/tsne_runner.cuh                  |  2 ++
 cpp/src/tsvd/tsvd.cuh                         |  2 ++
 cpp/src/umap/init_embed/spectral_algo.cuh     |  9 ++++++---
 cpp/src/umap/runner.cuh                       |  1 +
 cpp/src/umap/simpl_set_embed/algo.cuh         | 15 ++++++++------
 cpp/src/umap/supervised.cuh                   |  1 +
 cpp/src_prims/linalg/init.h                   |  1 +
 cpp/src_prims/metrics/adjusted_rand_index.cuh |  4 ++++
 .../metrics/batched/silhouette_score.cuh      |  2 ++
 cpp/src_prims/metrics/contingencyMatrix.cuh   |  2 ++
 cpp/src_prims/metrics/scores.cuh              |  6 +++++-
 cpp/src_prims/timeSeries/stationarity.cuh     | 20 +++++++++++--------
 cpp/test/prims/knn_regression.cu              | 11 ++++++++--
 cpp/test/sg/fil_test.cu                       |  1 +
 cpp/test/sg/kmeans_test.cu                    |  1 +
 cpp/test/sg/multi_sum_test.cu                 |  1 +
 cpp/test/sg/rf_test.cu                        |  7 +++++++
 cpp/test/sg/shap_kernel.cu                    |  3 +++
 cpp/test/sg/svc_test.cu                       |  3 +++
 cpp/test/sg/tsne_test.cu                      |  1 +
 53 files changed, 193 insertions(+), 26 deletions(-)

diff --git a/cpp/src/arima/batched_arima.cu b/cpp/src/arima/batched_arima.cu
index 9bf5cf3225..3bf5384342 100644
--- a/cpp/src/arima/batched_arima.cu
+++ b/cpp/src/arima/batched_arima.cu
@@ -20,6 +20,7 @@
 #include <vector>
 
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/fill.h>
 #include <thrust/for_each.h>
 #include <thrust/iterator/counting_iterator.h>
diff --git a/cpp/src/arima/batched_kalman.cu b/cpp/src/arima/batched_kalman.cu
index f0c042f9c4..c05cce367f 100644
--- a/cpp/src/arima/batched_kalman.cu
+++ b/cpp/src/arima/batched_kalman.cu
@@ -20,6 +20,7 @@
 #include <cuml/tsa/batched_kalman.hpp>
 
 #include <cub/cub.cuh>
+#include <thrust/execution_policy.h>
 #include <thrust/for_each.h>
 #include <thrust/iterator/counting_iterator.h>
 
diff --git a/cpp/src/dbscan/corepoints/compute.cuh b/cpp/src/dbscan/corepoints/compute.cuh
index 5945f00280..547604b508 100644
--- a/cpp/src/dbscan/corepoints/compute.cuh
+++ b/cpp/src/dbscan/corepoints/compute.cuh
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, NVIDIA CORPORATION.
+ * Copyright (c) 2021-2022, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -18,6 +18,9 @@
 
 #include <raft/handle.hpp>
 
+#include <thrust/for_each.h>
+#include <thrust/iterator/counting_iterator.h>
+
 namespace ML {
 namespace Dbscan {
 namespace CorePoints {
diff --git a/cpp/src/dbscan/runner.cuh b/cpp/src/dbscan/runner.cuh
index e3e8dcd8aa..7eaa901d2e 100644
--- a/cpp/src/dbscan/runner.cuh
+++ b/cpp/src/dbscan/runner.cuh
@@ -34,6 +34,11 @@
 
 #include <label/classlabels.cuh>
 
+#include <thrust/copy.h>
+#include <thrust/device_ptr.h>
+#include <thrust/fill.h>
+#include <thrust/iterator/counting_iterator.h>
+
 #include <cstddef>
 
 namespace ML {
diff --git a/cpp/src/decisiontree/batched-levelalgo/kernels/quantiles.cu b/cpp/src/decisiontree/batched-levelalgo/kernels/quantiles.cu
index f8661dfcc5..e6281810ca 100644
--- a/cpp/src/decisiontree/batched-levelalgo/kernels/quantiles.cu
+++ b/cpp/src/decisiontree/batched-levelalgo/kernels/quantiles.cu
@@ -16,6 +16,9 @@
  */
 #include "../quantiles.cuh"
 
+#include <thrust/execution_policy.h>
+#include <thrust/unique.h>
+
 namespace ML {
 namespace DT {
 
diff --git a/cpp/src/explainer/tree_shap.cu b/cpp/src/explainer/tree_shap.cu
index 99d8bf8ad7..cd57800ba2 100644
--- a/cpp/src/explainer/tree_shap.cu
+++ b/cpp/src/explainer/tree_shap.cu
@@ -28,6 +28,11 @@
 #include <raft/span.hpp>
 #include <rmm/device_uvector.hpp>
 #include <thrust/device_ptr.h>
+#include <thrust/device_vector.h>
+#include <thrust/execution_policy.h>
+#include <thrust/for_each.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/memory.h>
 #include <treelite/tree.h>
 #include <type_traits>
 #include <variant>
diff --git a/cpp/src/fil/fil.cu b/cpp/src/fil/fil.cu
index 887a79e7dc..204b2b61f1 100644
--- a/cpp/src/fil/fil.cu
+++ b/cpp/src/fil/fil.cu
@@ -581,7 +581,7 @@ void check_params(const forest_params_t* params, bool dense)
   ASSERT(params->blocks_per_sm >= 0, "blocks_per_sm must be nonnegative");
   ASSERT(params->n_items >= 0, "n_items must be non-negative");
   ASSERT(params->threads_per_tree > 0, "threads_per_tree must be positive");
-  ASSERT(thrust::detail::is_power_of_2(params->threads_per_tree),
+  ASSERT((params->threads_per_tree & (params->threads_per_tree - 1)) == 0,
          "threads_per_tree must be a power of 2");
   ASSERT(params->threads_per_tree <= FIL_TPB,
          "threads_per_tree must not "
diff --git a/cpp/src/genetic/fitness.cuh b/cpp/src/genetic/fitness.cuh
index a15fb96b54..a149ed23b9 100644
--- a/cpp/src/genetic/fitness.cuh
+++ b/cpp/src/genetic/fitness.cuh
@@ -34,6 +34,8 @@
 #include <thrust/device_vector.h>
 #include <thrust/functional.h>
 #include <thrust/iterator/permutation_iterator.h>
+#include <thrust/memory.h>
+#include <thrust/scan.h>
 #include <thrust/sequence.h>
 #include <thrust/sort.h>
 #include <thrust/transform.h>
diff --git a/cpp/src/glm/qn/glm_base.cuh b/cpp/src/glm/qn/glm_base.cuh
index ef935b317e..73faac80ed 100644
--- a/cpp/src/glm/qn/glm_base.cuh
+++ b/cpp/src/glm/qn/glm_base.cuh
@@ -24,6 +24,11 @@
 #include <raft/linalg/map_then_reduce.hpp>
 #include <raft/linalg/matrix_vector_op.hpp>
 #include <raft/stats/mean.hpp>
+
+#include <thrust/execution_policy.h>
+#include <thrust/functional.h>
+#include <thrust/reduce.h>
+
 #include <vector>
 
 namespace ML {
diff --git a/cpp/src/hdbscan/condensed_hierarchy.cu b/cpp/src/hdbscan/condensed_hierarchy.cu
index a7a047263f..15afb14366 100644
--- a/cpp/src/hdbscan/condensed_hierarchy.cu
+++ b/cpp/src/hdbscan/condensed_hierarchy.cu
@@ -27,9 +27,17 @@
 #include <raft/sparse/convert/csr.hpp>
 #include <raft/sparse/op/sort.hpp>
 
+#include <thrust/copy.h>
+#include <thrust/device_ptr.h>
 #include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/zip_iterator.h>
 #include <thrust/reduce.h>
+#include <thrust/sort.h>
+#include <thrust/transform.h>
 #include <thrust/transform_reduce.h>
+#include <thrust/tuple.h>
 
 #include <cuml/cluster/hdbscan.hpp>
 
diff --git a/cpp/src/hdbscan/detail/condense.cuh b/cpp/src/hdbscan/detail/condense.cuh
index ccedd9c1b2..50202b6130 100644
--- a/cpp/src/hdbscan/detail/condense.cuh
+++ b/cpp/src/hdbscan/detail/condense.cuh
@@ -30,7 +30,11 @@
 
 #include <cuml/cluster/hdbscan.hpp>
 
+#include <thrust/copy.h>
+#include <thrust/device_ptr.h>
 #include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
+#include <thrust/fill.h>
 #include <thrust/reduce.h>
 
 namespace ML {
diff --git a/cpp/src/hdbscan/detail/extract.cuh b/cpp/src/hdbscan/detail/extract.cuh
index 717a922739..e778124216 100644
--- a/cpp/src/hdbscan/detail/extract.cuh
+++ b/cpp/src/hdbscan/detail/extract.cuh
@@ -32,7 +32,9 @@
 #include <rmm/device_uvector.hpp>
 #include <rmm/exec_policy.hpp>
 
+#include <thrust/device_ptr.h>
 #include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
 #include <thrust/for_each.h>
 #include <thrust/reduce.h>
 #include <thrust/sort.h>
diff --git a/cpp/src/hdbscan/detail/membership.cuh b/cpp/src/hdbscan/detail/membership.cuh
index 8170a1dc77..87d26235ed 100644
--- a/cpp/src/hdbscan/detail/membership.cuh
+++ b/cpp/src/hdbscan/detail/membership.cuh
@@ -33,7 +33,9 @@
 #include <algorithm>
 
 #include <thrust/execution_policy.h>
+#include <thrust/fill.h>
 #include <thrust/for_each.h>
+#include <thrust/iterator/counting_iterator.h>
 #include <thrust/reduce.h>
 #include <thrust/sort.h>
 #include <thrust/transform.h>
diff --git a/cpp/src/hdbscan/detail/reachability.cuh b/cpp/src/hdbscan/detail/reachability.cuh
index 73756c3bfa..c0a0b8e90e 100644
--- a/cpp/src/hdbscan/detail/reachability.cuh
+++ b/cpp/src/hdbscan/detail/reachability.cuh
@@ -34,7 +34,10 @@
 #include <cuml/neighbors/knn.hpp>
 #include <raft/distance/distance.hpp>
 
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/zip_iterator.h>
 #include <thrust/transform.h>
+#include <thrust/tuple.h>
 
 namespace ML {
 namespace HDBSCAN {
diff --git a/cpp/src/hdbscan/detail/select.cuh b/cpp/src/hdbscan/detail/select.cuh
index d5b0495ff3..8236e22a7c 100644
--- a/cpp/src/hdbscan/detail/select.cuh
+++ b/cpp/src/hdbscan/detail/select.cuh
@@ -32,11 +32,18 @@
 
 #include <algorithm>
 
+#include <thrust/copy.h>
 #include <thrust/execution_policy.h>
+#include <thrust/fill.h>
 #include <thrust/for_each.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/zip_iterator.h>
 #include <thrust/reduce.h>
 #include <thrust/sort.h>
 #include <thrust/transform.h>
+#include <thrust/transform_reduce.h>
+#include <thrust/tuple.h>
 
 #include <rmm/device_uvector.hpp>
 #include <rmm/exec_policy.hpp>
diff --git a/cpp/src/hdbscan/detail/stabilities.cuh b/cpp/src/hdbscan/detail/stabilities.cuh
index 417b36381e..3a145b200b 100644
--- a/cpp/src/hdbscan/detail/stabilities.cuh
+++ b/cpp/src/hdbscan/detail/stabilities.cuh
@@ -33,10 +33,14 @@
 #include <algorithm>
 
 #include <thrust/execution_policy.h>
+#include <thrust/fill.h>
 #include <thrust/for_each.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/zip_iterator.h>
 #include <thrust/reduce.h>
 #include <thrust/sort.h>
 #include <thrust/transform.h>
+#include <thrust/tuple.h>
 
 #include <rmm/device_uvector.hpp>
 #include <rmm/exec_policy.hpp>
diff --git a/cpp/src/hdbscan/detail/utils.h b/cpp/src/hdbscan/detail/utils.h
index 3bfe06c061..3e1595e7ec 100644
--- a/cpp/src/hdbscan/detail/utils.h
+++ b/cpp/src/hdbscan/detail/utils.h
@@ -31,11 +31,16 @@
 
 #include "../condensed_hierarchy.cu"
 
+#include <thrust/copy.h>
 #include <thrust/execution_policy.h>
 #include <thrust/for_each.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/zip_iterator.h>
 #include <thrust/reduce.h>
 #include <thrust/sort.h>
 #include <thrust/transform.h>
+#include <thrust/transform_reduce.h>
+#include <thrust/tuple.h>
 
 #include <rmm/device_uvector.hpp>
 #include <rmm/exec_policy.hpp>
@@ -186,4 +191,4 @@ void parent_csr(const raft::handle_t& handle,
 };  // namespace Utils
 };  // namespace detail
 };  // namespace HDBSCAN
-};  // namespace ML
\ No newline at end of file
+};  // namespace ML
diff --git a/cpp/src/hdbscan/runner.h b/cpp/src/hdbscan/runner.h
index acb0762baa..e5baf5a445 100644
--- a/cpp/src/hdbscan/runner.h
+++ b/cpp/src/hdbscan/runner.h
@@ -32,6 +32,10 @@
 #include "detail/reachability.cuh"
 #include <cuml/cluster/hdbscan.hpp>
 
+#include <thrust/device_ptr.h>
+#include <thrust/extrema.h>
+#include <thrust/transform.h>
+
 namespace ML {
 namespace HDBSCAN {
 
diff --git a/cpp/src/hierarchy/pw_dist_graph.cuh b/cpp/src/hierarchy/pw_dist_graph.cuh
index 35dd840308..716a6a3e00 100644
--- a/cpp/src/hierarchy/pw_dist_graph.cuh
+++ b/cpp/src/hierarchy/pw_dist_graph.cuh
@@ -33,8 +33,13 @@
 
 #include <thrust/device_ptr.h>
 #include <thrust/execution_policy.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/zip_iterator.h>
 #include <thrust/scan.h>
+#include <thrust/sequence.h>
 #include <thrust/sort.h>
+#include <thrust/transform.h>
+#include <thrust/tuple.h>
 
 #include <limits>
 
diff --git a/cpp/src/kmeans/kmeans_mg_impl.cuh b/cpp/src/kmeans/kmeans_mg_impl.cuh
index 871ca53cd4..eb85141738 100644
--- a/cpp/src/kmeans/kmeans_mg_impl.cuh
+++ b/cpp/src/kmeans/kmeans_mg_impl.cuh
@@ -20,6 +20,12 @@
 #include <rmm/device_scalar.hpp>
 #include <rmm/device_uvector.hpp>
 
+#include <thrust/execution_policy.h>
+#include <thrust/fill.h>
+#include <thrust/reduce.h>
+#include <thrust/scan.h>
+#include <thrust/transform.h>
+
 #include "common.cuh"
 #include "sg_impl.cuh"
 
diff --git a/cpp/src/kmeans/sg_impl.cuh b/cpp/src/kmeans/sg_impl.cuh
index f8db188ab9..65b7b66ec7 100644
--- a/cpp/src/kmeans/sg_impl.cuh
+++ b/cpp/src/kmeans/sg_impl.cuh
@@ -22,6 +22,9 @@
 #include <rmm/device_scalar.hpp>
 #include <rmm/device_uvector.hpp>
 
+#include <thrust/fill.h>
+#include <thrust/transform.h>
+
 namespace ML {
 
 namespace kmeans {
diff --git a/cpp/src/pca/sign_flip_mg.cu b/cpp/src/pca/sign_flip_mg.cu
index f86e9e3db2..4ef824c901 100644
--- a/cpp/src/pca/sign_flip_mg.cu
+++ b/cpp/src/pca/sign_flip_mg.cu
@@ -27,6 +27,8 @@
 
 #include <thrust/device_vector.h>
 #include <thrust/execution_policy.h>
+#include <thrust/for_each.h>
+#include <thrust/iterator/counting_iterator.h>
 
 #include <cstddef>
 
diff --git a/cpp/src/randomforest/randomforest.cuh b/cpp/src/randomforest/randomforest.cuh
index f4cfb24d68..d631a3c366 100644
--- a/cpp/src/randomforest/randomforest.cuh
+++ b/cpp/src/randomforest/randomforest.cuh
@@ -16,17 +16,19 @@
 
 #pragma once
 
-#include <raft/common/nvtx.hpp>
-
 #include <decisiontree/batched-levelalgo/quantiles.cuh>
 #include <decisiontree/decisiontree.cuh>
 #include <decisiontree/treelite_util.h>
 
 #include <metrics/scores.cuh>
+
+#include <raft/common/nvtx.hpp>
+#include <raft/cudart_utils.h>
 #include <raft/random/permute.hpp>
 #include <raft/random/rng.cuh>
 
-#include <raft/cudart_utils.h>
+#include <thrust/execution_policy.h>
+#include <thrust/sequence.h>
 
 #ifdef _OPENMP
 #include <omp.h>
diff --git a/cpp/src/solver/lars_impl.cuh b/cpp/src/solver/lars_impl.cuh
index 49dfd04152..7810eccc21 100644
--- a/cpp/src/solver/lars_impl.cuh
+++ b/cpp/src/solver/lars_impl.cuh
@@ -37,6 +37,9 @@
 #include <rmm/device_uvector.hpp>
 #include <thrust/copy.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
+#include <thrust/fill.h>
 #include <thrust/iterator/counting_iterator.h>
 #include <thrust/sequence.h>
 #include <thrust/sort.h>
diff --git a/cpp/src/svm/linear.cu b/cpp/src/svm/linear.cu
index 924fe86689..65a1fc6758 100644
--- a/cpp/src/svm/linear.cu
+++ b/cpp/src/svm/linear.cu
@@ -36,8 +36,10 @@
 #include <rmm/device_uvector.hpp>
 #include <thrust/copy.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/fill.h>
 #include <thrust/iterator/counting_iterator.h>
+#include <thrust/tuple.h>
 
 #include <glm/ols.cuh>
 #include <glm/qn/qn.cuh>
diff --git a/cpp/src/svm/smosolver.cuh b/cpp/src/svm/smosolver.cuh
index a6bc8a6968..0ee698a948 100644
--- a/cpp/src/svm/smosolver.cuh
+++ b/cpp/src/svm/smosolver.cuh
@@ -32,8 +32,11 @@
 #include <raft/cuda_utils.cuh>
 #include <raft/cudart_utils.h>
 #include <string>
+#include <thrust/copy.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/fill.h>
+#include <thrust/iterator/constant_iterator.h>
 #include <type_traits>
 
 #include "kernelcache.cuh"
diff --git a/cpp/src/svm/svc_impl.cuh b/cpp/src/svm/svc_impl.cuh
index ca5eb0f11a..3aeeb60882 100644
--- a/cpp/src/svm/svc_impl.cuh
+++ b/cpp/src/svm/svc_impl.cuh
@@ -38,6 +38,7 @@
 #include <rmm/mr/device/per_device_resource.hpp>
 #include <thrust/copy.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/iterator/counting_iterator.h>
 
 namespace ML {
diff --git a/cpp/src/svm/workingset.cuh b/cpp/src/svm/workingset.cuh
index 6b112ffd86..6240ea7594 100644
--- a/cpp/src/svm/workingset.cuh
+++ b/cpp/src/svm/workingset.cuh
@@ -36,7 +36,9 @@
 #include <thrust/device_ptr.h>
 #include <thrust/iterator/permutation_iterator.h>
 
+#include <thrust/copy.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/iterator/permutation_iterator.h>
 
 #include <cub/cub.cuh>
diff --git a/cpp/src/tsne/barnes_hut_tsne.cuh b/cpp/src/tsne/barnes_hut_tsne.cuh
index 6f3435eceb..1ca21d35ba 100644
--- a/cpp/src/tsne/barnes_hut_tsne.cuh
+++ b/cpp/src/tsne/barnes_hut_tsne.cuh
@@ -22,6 +22,10 @@
 #include <raft/cudart_utils.h>
 #include <raft/linalg/eltwise.hpp>
 
+#include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/fill.h>
+
 namespace ML {
 namespace TSNE {
 
diff --git a/cpp/src/tsne/distances.cuh b/cpp/src/tsne/distances.cuh
index 828e93f3f7..faac40cbfe 100644
--- a/cpp/src/tsne/distances.cuh
+++ b/cpp/src/tsne/distances.cuh
@@ -34,6 +34,9 @@
 #include <rmm/device_uvector.hpp>
 #include <rmm/exec_policy.hpp>
 
+#include <thrust/functional.h>
+#include <thrust/transform_reduce.h>
+
 namespace ML {
 namespace TSNE {
 
diff --git a/cpp/src/tsne/exact_kernels.cuh b/cpp/src/tsne/exact_kernels.cuh
index 6ad3da2120..81270a44d0 100644
--- a/cpp/src/tsne/exact_kernels.cuh
+++ b/cpp/src/tsne/exact_kernels.cuh
@@ -21,6 +21,10 @@
 #include <raft/cudart_utils.h>
 #include <raft/linalg/eltwise.hpp>
 
+#include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/reduce.h>
+
 #define restrict __restrict__
 
 namespace ML {
diff --git a/cpp/src/tsne/exact_tsne.cuh b/cpp/src/tsne/exact_tsne.cuh
index c3c64090c6..8f034d27d4 100644
--- a/cpp/src/tsne/exact_tsne.cuh
+++ b/cpp/src/tsne/exact_tsne.cuh
@@ -20,6 +20,10 @@
 #include <cuml/common/logger.hpp>
 #include <raft/cudart_utils.h>
 
+#include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/fill.h>
+
 namespace ML {
 namespace TSNE {
 
diff --git a/cpp/src/tsne/fft_tsne.cuh b/cpp/src/tsne/fft_tsne.cuh
index cdd0223232..0eacec6cf0 100644
--- a/cpp/src/tsne/fft_tsne.cuh
+++ b/cpp/src/tsne/fft_tsne.cuh
@@ -34,6 +34,12 @@
 #include <rmm/device_scalar.hpp>
 #include <rmm/device_uvector.hpp>
 
+#include <thrust/device_ptr.h>
+#include <thrust/fill.h>
+#include <thrust/functional.h>
+#include <thrust/reduce.h>
+#include <thrust/transform.h>
+
 namespace ML {
 namespace TSNE {
 
diff --git a/cpp/src/tsne/tsne_runner.cuh b/cpp/src/tsne/tsne_runner.cuh
index d7a1a2453a..776a705d06 100644
--- a/cpp/src/tsne/tsne_runner.cuh
+++ b/cpp/src/tsne/tsne_runner.cuh
@@ -23,6 +23,8 @@
 #include <raft/cudart_utils.h>
 #include <rmm/device_uvector.hpp>
 
+#include <thrust/transform.h>
+
 #include "barnes_hut_tsne.cuh"
 #include "exact_tsne.cuh"
 #include "fft_tsne.cuh"
diff --git a/cpp/src/tsvd/tsvd.cuh b/cpp/src/tsvd/tsvd.cuh
index 0090b22131..ea77c74391 100644
--- a/cpp/src/tsvd/tsvd.cuh
+++ b/cpp/src/tsvd/tsvd.cuh
@@ -35,6 +35,8 @@
 #include <rmm/exec_policy.hpp>
 #include <thrust/device_vector.h>
 #include <thrust/execution_policy.h>
+#include <thrust/for_each.h>
+#include <thrust/iterator/counting_iterator.h>
 
 namespace ML {
 
diff --git a/cpp/src/umap/init_embed/spectral_algo.cuh b/cpp/src/umap/init_embed/spectral_algo.cuh
index ef09165173..2b191aa9a2 100644
--- a/cpp/src/umap/init_embed/spectral_algo.cuh
+++ b/cpp/src/umap/init_embed/spectral_algo.cuh
@@ -16,15 +16,18 @@
 
 #pragma once
 
+#include <cuml/cluster/spectral.hpp>
 #include <cuml/manifold/umapparams.h>
 
 #include <raft/linalg/add.hpp>
-#include <raft/sparse/coo.hpp>
-
 #include <raft/linalg/transpose.hpp>
 #include <raft/random/rng.hpp>
+#include <raft/sparse/coo.hpp>
+
+#include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
 
-#include <cuml/cluster/spectral.hpp>
 #include <iostream>
 
 namespace UMAPAlgo {
diff --git a/cpp/src/umap/runner.cuh b/cpp/src/umap/runner.cuh
index 02f970d348..af304c3fce 100644
--- a/cpp/src/umap/runner.cuh
+++ b/cpp/src/umap/runner.cuh
@@ -31,6 +31,7 @@
 
 #include <thrust/count.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/extrema.h>
 #include <thrust/reduce.h>
 #include <thrust/scan.h>
diff --git a/cpp/src/umap/simpl_set_embed/algo.cuh b/cpp/src/umap/simpl_set_embed/algo.cuh
index 1f358f177c..2241ce26f5 100644
--- a/cpp/src/umap/simpl_set_embed/algo.cuh
+++ b/cpp/src/umap/simpl_set_embed/algo.cuh
@@ -24,11 +24,6 @@
 
 #include <curand.h>
 #include <math.h>
-#include <thrust/device_ptr.h>
-#include <thrust/extrema.h>
-#include <thrust/for_each.h>
-#include <thrust/iterator/counting_iterator.h>
-#include <thrust/system/cuda/execution_policy.h>
 
 #include <common/fast_int_div.cuh>
 #include <cstdlib>
@@ -42,8 +37,16 @@
 #include <string>
 
 #include <raft/sparse/op/filter.hpp>
+
+#include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
+#include <thrust/for_each.h>
+#include <thrust/iterator/constant_iterator.h>
+#include <thrust/iterator/counting_iterator.h>
 #include <thrust/iterator/discard_iterator.h>
-#pragma once
+#include <thrust/reduce.h>
+#include <thrust/system/cuda/execution_policy.h>
 
 namespace UMAPAlgo {
 namespace SimplSetEmbed {
diff --git a/cpp/src/umap/supervised.cuh b/cpp/src/umap/supervised.cuh
index abc69d3345..fb1fc32e9c 100644
--- a/cpp/src/umap/supervised.cuh
+++ b/cpp/src/umap/supervised.cuh
@@ -29,6 +29,7 @@
 
 #include <thrust/count.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/extrema.h>
 #include <thrust/reduce.h>
 #include <thrust/scan.h>
diff --git a/cpp/src_prims/linalg/init.h b/cpp/src_prims/linalg/init.h
index 3c44394548..4dfb930880 100644
--- a/cpp/src_prims/linalg/init.h
+++ b/cpp/src_prims/linalg/init.h
@@ -20,6 +20,7 @@
 
 #include <thrust/copy.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/iterator/counting_iterator.h>
 
 namespace MLCommon {
diff --git a/cpp/src_prims/metrics/adjusted_rand_index.cuh b/cpp/src_prims/metrics/adjusted_rand_index.cuh
index 3f0780e656..edc0943dde 100644
--- a/cpp/src_prims/metrics/adjusted_rand_index.cuh
+++ b/cpp/src_prims/metrics/adjusted_rand_index.cuh
@@ -33,6 +33,10 @@
 #include <rmm/device_scalar.hpp>
 #include <rmm/device_uvector.hpp>
 
+#include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
+
 namespace MLCommon {
 namespace Metrics {
 
diff --git a/cpp/src_prims/metrics/batched/silhouette_score.cuh b/cpp/src_prims/metrics/batched/silhouette_score.cuh
index 813393ce6b..79898a2f74 100644
--- a/cpp/src_prims/metrics/batched/silhouette_score.cuh
+++ b/cpp/src_prims/metrics/batched/silhouette_score.cuh
@@ -24,6 +24,8 @@
 #include <rmm/device_uvector.hpp>
 #include <rmm/exec_policy.hpp>
 #include <thrust/device_vector.h>
+#include <thrust/fill.h>
+#include <thrust/reduce.h>
 
 namespace MLCommon {
 namespace Metrics {
diff --git a/cpp/src_prims/metrics/contingencyMatrix.cuh b/cpp/src_prims/metrics/contingencyMatrix.cuh
index 9fc0526565..ca621c8ca8 100644
--- a/cpp/src_prims/metrics/contingencyMatrix.cuh
+++ b/cpp/src_prims/metrics/contingencyMatrix.cuh
@@ -20,6 +20,8 @@
 #include <raft/cudart_utils.h>
 
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/extrema.h>
 #include <thrust/reduce.h>
 
 #include <cub/cub.cuh>
diff --git a/cpp/src_prims/metrics/scores.cuh b/cpp/src_prims/metrics/scores.cuh
index 6c42f32600..782e413726 100644
--- a/cpp/src_prims/metrics/scores.cuh
+++ b/cpp/src_prims/metrics/scores.cuh
@@ -16,7 +16,6 @@
 
 #pragma once
 
-#include <memory>
 #include <raft/cudart_utils.h>
 #include <raft/distance/distance.hpp>
 #include <raft/linalg/eltwise.hpp>
@@ -24,10 +23,15 @@
 #include <raft/linalg/subtract.hpp>
 #include <raft/spatial/knn/knn.hpp>
 #include <raft/stats/mean.hpp>
+
 #include <rmm/device_scalar.hpp>
 #include <rmm/device_uvector.hpp>
+
 #include <selection/columnWiseSort.cuh>
+
+#include <thrust/count.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/reduce.h>
 
 namespace MLCommon {
diff --git a/cpp/src_prims/timeSeries/stationarity.cuh b/cpp/src_prims/timeSeries/stationarity.cuh
index 519dface1a..d06069c099 100644
--- a/cpp/src_prims/timeSeries/stationarity.cuh
+++ b/cpp/src_prims/timeSeries/stationarity.cuh
@@ -25,21 +25,25 @@
 
 #pragma once
 
-#include <math.h>
-#include <thrust/device_ptr.h>
-#include <thrust/execution_policy.h>
-#include <thrust/iterator/counting_iterator.h>
-#include <thrust/iterator/transform_iterator.h>
-#include <thrust/scan.h>
-#include <vector>
-
 #include "arima_helpers.cuh"
+
 #include <raft/cudart_utils.h>
 #include <raft/linalg/matrix_vector_op.hpp>
 #include <raft/linalg/reduce.hpp>
 #include <raft/stats/mean.hpp>
+
 #include <rmm/device_uvector.hpp>
 
+#include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
+#include <thrust/functional.h>
+#include <thrust/iterator/counting_iterator.h>
+#include <thrust/iterator/transform_iterator.h>
+#include <thrust/scan.h>
+
+#include <cmath>
+#include <vector>
+
 namespace MLCommon {
 
 namespace TimeSeries {
diff --git a/cpp/test/prims/knn_regression.cu b/cpp/test/prims/knn_regression.cu
index 01cfad6775..34090db674 100644
--- a/cpp/test/prims/knn_regression.cu
+++ b/cpp/test/prims/knn_regression.cu
@@ -15,21 +15,28 @@
  */
 
 #include "test_utils.h"
+
 #include <gtest/gtest.h>
-#include <iostream>
+
 #include <label/classlabels.cuh>
+
 #include <raft/cuda_utils.cuh>
 #include <raft/cudart_utils.h>
 #include <raft/linalg/reduce.hpp>
 #include <raft/random/rng.hpp>
 #include <raft/spatial/knn/knn.hpp>
+
 #include <rmm/device_uvector.hpp>
+
 #include <selection/knn.cuh>
-#include <vector>
 
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/extrema.h>
 
+#include <iostream>
+#include <vector>
+
 namespace MLCommon {
 namespace Selection {
 
diff --git a/cpp/test/sg/fil_test.cu b/cpp/test/sg/fil_test.cu
index cb3bcf8495..82955ec8ae 100644
--- a/cpp/test/sg/fil_test.cu
+++ b/cpp/test/sg/fil_test.cu
@@ -26,6 +26,7 @@
 #include <test_utils.h>
 #include <thrust/execution_policy.h>
 #include <thrust/functional.h>
+#include <thrust/iterator/counting_iterator.h>
 #include <thrust/transform.h>
 
 #include <treelite/c_api.h>
diff --git a/cpp/test/sg/kmeans_test.cu b/cpp/test/sg/kmeans_test.cu
index 679517c888..7c43f1526e 100644
--- a/cpp/test/sg/kmeans_test.cu
+++ b/cpp/test/sg/kmeans_test.cu
@@ -26,6 +26,7 @@
 #include <cuml/common/logger.hpp>
 #include <cuml/datasets/make_blobs.hpp>
 #include <cuml/metrics/metrics.hpp>
+#include <thrust/execution_policy.h>
 #include <thrust/fill.h>
 
 namespace ML {
diff --git a/cpp/test/sg/multi_sum_test.cu b/cpp/test/sg/multi_sum_test.cu
index 5adb18a1fc..dbc2c8e42e 100644
--- a/cpp/test/sg/multi_sum_test.cu
+++ b/cpp/test/sg/multi_sum_test.cu
@@ -23,6 +23,7 @@
 #include <raft/random/rng.hpp>
 
 #include <thrust/device_vector.h>
+#include <thrust/fill.h>
 #include <thrust/host_vector.h>
 
 #include <gtest/gtest.h>
diff --git a/cpp/test/sg/rf_test.cu b/cpp/test/sg/rf_test.cu
index 18923c4baa..b6cf8fcfd1 100644
--- a/cpp/test/sg/rf_test.cu
+++ b/cpp/test/sg/rf_test.cu
@@ -31,10 +31,17 @@
 #include <raft/linalg/transpose.hpp>
 
 #include <thrust/binary_search.h>
+#include <thrust/copy.h>
 #include <thrust/device_vector.h>
+#include <thrust/execution_policy.h>
+#include <thrust/for_each.h>
+#include <thrust/functional.h>
 #include <thrust/host_vector.h>
+#include <thrust/iterator/counting_iterator.h>
 #include <thrust/logical.h>
+#include <thrust/random.h>
 #include <thrust/shuffle.h>
+#include <thrust/transform.h>
 
 #include <gtest/gtest.h>
 
diff --git a/cpp/test/sg/shap_kernel.cu b/cpp/test/sg/shap_kernel.cu
index 4fa2c4edaa..327afe6cdf 100644
--- a/cpp/test/sg/shap_kernel.cu
+++ b/cpp/test/sg/shap_kernel.cu
@@ -23,8 +23,11 @@
 #include <raft/handle.hpp>
 #include <rmm/device_uvector.hpp>
 
+#include <thrust/count.h>
 #include <thrust/device_ptr.h>
 #include <thrust/device_vector.h>
+#include <thrust/execution_policy.h>
+#include <thrust/fill.h>
 
 #include <test_utils.h>
 
diff --git a/cpp/test/sg/svc_test.cu b/cpp/test/sg/svc_test.cu
index d9bbd5e7f8..3c987c452d 100644
--- a/cpp/test/sg/svc_test.cu
+++ b/cpp/test/sg/svc_test.cu
@@ -38,9 +38,12 @@
 #include <svm/workingset.cuh>
 #include <test_utils.h>
 #include <thrust/device_ptr.h>
+#include <thrust/execution_policy.h>
 #include <thrust/fill.h>
 #include <thrust/iterator/zip_iterator.h>
+#include <thrust/reduce.h>
 #include <thrust/transform.h>
+#include <thrust/tuple.h>
 #include <type_traits>
 #include <vector>
 
diff --git a/cpp/test/sg/tsne_test.cu b/cpp/test/sg/tsne_test.cu
index c7c79d8535..7224c63094 100644
--- a/cpp/test/sg/tsne_test.cu
+++ b/cpp/test/sg/tsne_test.cu
@@ -28,6 +28,7 @@
 #include <raft/cudart_utils.h>
 #include <stdio.h>
 #include <stdlib.h>
+#include <thrust/reduce.h>
 #include <tsne/distances.cuh>
 #include <tsne/tsne_runner.cuh>
 #include <tsne/utils.cuh>