rm device_helpers.hip.h from cu

src/common/hist_util.cu

@@ -19,13 +19,7 @@
 #include <vector>
 
 #include "categorical.h"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "device_helpers.hip.h"
-#endif
-
 #include "hist_util.cuh"
 #include "hist_util.h"
 #include "math.h"  // NOLINT

src/common/host_device_vector.cu

@@ -11,12 +11,7 @@
 #include "xgboost/data.h"
 #include "xgboost/host_device_vector.h"
 #include "xgboost/tree_model.h"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "device_helpers.hip.h"
-#endif
 
 namespace xgboost {
 

src/common/numeric.cu

@@ -3,12 +3,7 @@
  */
 #include <thrust/execution_policy.h>
 
-#if defined(XGBOOST_USE_CUDA)
 #include "device_helpers.cuh"            // dh::Reduce, dh::XGBCachingDeviceAllocator
-#elif defined(XGBOOST_USE_HIP)
-#include "device_helpers.hip.h"          // dh::Reduce, dh::XGBCachingDeviceAllocator
-#endif
-
 #include "numeric.h"
 #include "xgboost/context.h"             // Context
 #include "xgboost/host_device_vector.h"  // HostDeviceVector

src/common/quantile.cu

@@ -16,13 +16,7 @@
 #include "../collective/device_communicator.cuh"
 #include "categorical.h"
 #include "common.h"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "device_helpers.hip.h"
-#endif
-
 #include "hist_util.h"
 #include "quantile.cuh"
 #include "quantile.h"

src/common/stats.cu

@@ -7,13 +7,7 @@
 #include <cstddef>                              // size_t
 
 #include "cuda_context.cuh"                     // CUDAContext
-
-#if defined(XGBOOST_USE_CUDA)
 #include "device_helpers.cuh"                   // dh::MakeTransformIterator, tcbegin, tcend
-#elif defined(XGBOOST_USE_HIP)
-#include "device_helpers.hip.h"                 // dh::MakeTransformIterator, tcbegin, tcend
-#endif
-
 #include "optional_weight.h"                    // common::OptionalWeights
 #include "stats.cuh"          // common::SegmentedQuantile, common::SegmentedWeightedQuantile
 #include "xgboost/base.h"     // XGBOOST_DEVICE

src/data/data.cu

@@ -5,13 +5,7 @@
  * \brief Handles setting metainfo from array interface.
  */
 #include "../common/cuda_context.cuh"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"
-#endif
-
 #include "../common/linalg_op.cuh"
 #include "array_interface.h"
 #include "device_adapter.cuh"

src/gbm/gbtree.cu

@@ -2,12 +2,7 @@
  * Copyright 2021 by Contributors
  */
 
-#if defined(XGBOOST_USE_CUDA)
 #include "../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"
-#endif
-
 #include "xgboost/context.h"
 #include "xgboost/linalg.h"
 #include "xgboost/span.h"

src/linear/updater_gpu_coordinate.cu

@@ -11,13 +11,7 @@
 
 #include "coordinate_common.h"
 #include "../common/common.h"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"
-#endif
-
 #include "../common/timer.h"
 #include "./param.h"
 

src/metric/elementwise_metric.cu

@@ -20,23 +20,14 @@
 #include "metric_common.h"
 #include "xgboost/metric.h"
 
-#if defined(XGBOOST_USE_CUDA)
+#if defined(XGBOOST_USE_CUDA) || defined(XGBOOST_USE_HIP)
 #include <thrust/execution_policy.h>  // thrust::cuda::par
 #include <thrust/functional.h>        // thrust::plus<>
 #include <thrust/iterator/counting_iterator.h>
 #include <thrust/transform_reduce.h>
 
 #include "../common/device_helpers.cuh"
-#endif  // XGBOOST_USE_CUDA
+#endif  // defined(XGBOOST_USE_CUDA) || defined(XGBOOST_USE_HIP)
-
-#if defined(XGBOOST_USE_HIP)
-#include <thrust/execution_policy.h>  // thrust::hip::par
-#include <thrust/functional.h>        // thrust::plus<>
-#include <thrust/iterator/counting_iterator.h>
-#include <thrust/transform_reduce.h>
-
-#include "../common/device_helpers.hip.h"
-#endif  // XGBOOST_USE_HIP
 
 namespace xgboost {
 namespace metric {

src/metric/multiclass_metric.cu

@@ -14,23 +14,14 @@
 #include "../common/threading_utils.h"
 #include "metric_common.h"  // MetricNoCache
 
-#if defined(XGBOOST_USE_CUDA)
+#if defined(XGBOOST_USE_CUDA) || defined(XGBOOST_USE_HIP)
 #include <thrust/execution_policy.h>  // thrust::cuda::par
 #include <thrust/functional.h>        // thrust::plus<>
 #include <thrust/iterator/counting_iterator.h>
 #include <thrust/transform_reduce.h>
 
 #include "../common/device_helpers.cuh"
-#endif  // XGBOOST_USE_CUDA
+#endif  // XGBOOST_USE_CUDA || XGBOOST_USE_HIP
-
-#if defined(XGBOOST_USE_HIP)
-#include <thrust/execution_policy.h>  // thrust::hip::par
-#include <thrust/functional.h>        // thrust::plus<>
-#include <thrust/iterator/counting_iterator.h>
-#include <thrust/transform_reduce.h>
-
-#include "../common/device_helpers.hip.h"
-#endif  // XGBOOST_USE_HIP
 
 namespace xgboost {
 namespace metric {

src/metric/survival_metric.cu

@@ -19,15 +19,10 @@
 #include "xgboost/json.h"
 #include "xgboost/metric.h"
 
-#if defined(XGBOOST_USE_CUDA)
+#if defined(XGBOOST_USE_CUDA) || defined(XGBOOST_USE_HIP)
 #include <thrust/execution_policy.h>  // thrust::cuda::par
 #include "../common/device_helpers.cuh"
-#endif  // XGBOOST_USE_CUDA
+#endif  // XGBOOST_USE_CUDA || XGBOOST_USE_HIP
-
-#if defined(XGBOOST_USE_HIP)
-#include <thrust/execution_policy.h>  // thrust::hip::par
-#include "../common/device_helpers.hip.h"
-#endif  // XGBOOST_USE_HIP
 
 using AFTParam = xgboost::common::AFTParam;
 using ProbabilityDistributionType = xgboost::common::ProbabilityDistributionType;

src/objective/adaptive.cu

@@ -12,13 +12,7 @@
 #endif
 
 #include "../common/cuda_context.cuh"  // CUDAContext
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"
-#endif
-
 #include "../common/stats.cuh"
 #include "adaptive.h"
 #include "xgboost/context.h"

src/objective/rank_obj.cu

@@ -15,27 +15,20 @@
 #include "../common/math.h"
 #include "../common/random.h"
 
-#if defined(__CUDACC__)
+#if defined(__CUDACC__) || defined(__HIP_PLATFORM_AMD__)
 #include <thrust/sort.h>
 #include <thrust/gather.h>
 #include <thrust/iterator/discard_iterator.h>
 #include <thrust/random/uniform_int_distribution.h>
 #include <thrust/random/linear_congruential_engine.h>
 
+#if defined(__CUDACC__)
 #include <cub/util_allocator.cuh>
+#elif defined(__HIP_PLATFORM_AMD__)
+#include <hipcub/util_allocator.hpp>
+#endif
 
 #include "../common/device_helpers.cuh"
-#elif defined(__HIP_PLATFORM_AMD__)
-
-#include <thrust/sort.h>
-#include <thrust/gather.h>
-#include <thrust/iterator/discard_iterator.h>
-#include <thrust/random/uniform_int_distribution.h>
-#include <thrust/random/linear_congruential_engine.h>
-
-#include <hipcub/util_allocator.hpp>
-
-#include "../common/device_helpers.hip.h"
 #endif
 
 namespace xgboost {

src/objective/regression_obj.cu

@@ -35,12 +35,9 @@
 #include "xgboost/span.h"
 #include "xgboost/tree_model.h"  // RegTree
 
-#if defined(XGBOOST_USE_CUDA)
+#if defined(XGBOOST_USE_CUDA) || defined(XGBOOST_USE_HIP)
 #include "../common/device_helpers.cuh"
 #include "../common/linalg_op.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"
-#include "../common/linalg_op.cuh"
 #endif  // defined(XGBOOST_USE_CUDA), defined(XGBOOST_USE_HIP)
 
 namespace xgboost {

src/predictor/gpu_predictor.cu

@@ -14,13 +14,7 @@
 #include "../common/bitfield.h"
 #include "../common/categorical.h"
 #include "../common/common.h"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"
-#endif
-
 #include "../data/device_adapter.cuh"
 #include "../data/ellpack_page.cuh"
 #include "../data/proxy_dmatrix.h"

src/tree/constraints.cu

@@ -14,12 +14,7 @@
 #include "xgboost/span.h"
 #include "constraints.cuh"
 #include "param.h"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"
-#endif
 
 namespace xgboost {
 

src/tree/fit_stump.cu

@@ -12,13 +12,7 @@
 #include <cstddef>                                // std::size_t
 
 #include "../collective/device_communicator.cuh"  // DeviceCommunicator
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../common/device_helpers.cuh"           // dh::MakeTransformIterator
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"         // dh::MakeTransformIterator
-#endif
-
 #include "fit_stump.h"
 #include "xgboost/base.h"     // GradientPairPrecise, GradientPair, XGBOOST_DEVICE
 #include "xgboost/context.h"  // Context

src/tree/gpu_hist/evaluate_splits.cu

@@ -6,19 +6,14 @@
 #include <limits>
 
 #include "../../common/categorical.h"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../common/device_helpers.hip.h"
-#include <hip/hip_cooperative_groups.h>
-#endif
-
 #include "../../data/ellpack_page.cuh"
 #include "evaluate_splits.cuh"
 #include "expand_entry.cuh"
 
 #if defined(XGBOOST_USE_HIP)
+#include <hip/hip_cooperative_groups.h>
+
 #define WARP_SIZE WAVEFRONT_SIZE
 #elif defined(XGBOOST_USE_CUDA)
 #define WARP_SIZE 32

src/tree/gpu_hist/evaluator.cu

@@ -7,12 +7,7 @@
 #include <thrust/logical.h>  // thrust::any_of
 #include <thrust/sort.h>     // thrust::stable_sort
 
-#if defined(XGBOOST_USE_CUDA)
 #include "../../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../common/device_helpers.hip.h"
-#endif
-
 #include "../../common/hist_util.h"  // common::HistogramCuts
 #include "evaluate_splits.cuh"
 #include "xgboost/data.h"

src/tree/gpu_hist/feature_groups.cu

@@ -7,13 +7,7 @@
 #include <vector>
 
 #include "feature_groups.cuh"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../common/device_helpers.hip.h"
-#endif
-
 #include "../../common/hist_util.h"
 
 namespace xgboost {

src/tree/gpu_hist/histogram.cu

@@ -9,13 +9,7 @@
 #include <limits>
 
 #include "../../common/deterministic.cuh"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../common/device_helpers.hip.h"
-#endif
-
 #include "../../data/ellpack_page.cuh"
 #include "histogram.cuh"
 #include "row_partitioner.cuh"

src/tree/gpu_hist/row_partitioner.cu

@@ -7,12 +7,7 @@
 
 #include <vector>
 
-#if defined(XGBOOST_USE_CUDA)
 #include "../../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../common/device_helpers.hip.h"
-#endif
-
 #include "row_partitioner.cuh"
 
 namespace xgboost {

src/tree/updater_gpu_hist.cu

@@ -17,13 +17,7 @@
 #include "../common/categorical.h"
 
 #include "../common/cuda_context.cuh"  // CUDAContext
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../common/device_helpers.hip.h"
-#endif
-
 #include "../common/hist_util.h"
 #include "../common/io.h"
 #include "../common/timer.h"

tests/cpp/common/test_host_device_vector.cu

@@ -6,12 +6,7 @@
 #include <thrust/equal.h>
 #include <thrust/iterator/counting_iterator.h>
 
-#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../../src/common/device_helpers.hip.h"
-#endif
-
 #include <xgboost/host_device_vector.h>
 
 namespace xgboost {

tests/cpp/common/test_span.cu

@@ -7,11 +7,7 @@
 #include <thrust/device_vector.h>
 #include <thrust/execution_policy.h>
 
-#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../../src/common/device_helpers.hip.h"
-#endif
 
 #include <xgboost/span.h>
 #include "test_span.h"

tests/cpp/data/test_metainfo.cu

@@ -6,12 +6,7 @@
 #include <xgboost/data.h>
 #include <xgboost/json.h>
 
-#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../../src/common/device_helpers.hip.h"
-#endif
-
 #include "test_array_interface.h"
 #include "test_metainfo.h"
 

tests/cpp/tree/test_constraints.cu

@@ -10,12 +10,7 @@
 #include <set>
 #include "../../../src/tree/constraints.cuh"
 #include "../../../src/tree/param.h"
-
-#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/device_helpers.cuh"
-#elif defined(XGBOOST_USE_HIP)
-#include "../../../src/common/device_helpers.hip.h"
-#endif
 
 namespace xgboost {
 namespace {