add hip.h

src/collective/device_communicator.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2022 XGBoost contributors
+ */
+#pragma once
+
+#include "device_communicator.cuh"

src/collective/device_communicator_adapter.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2022 XGBoost contributors
+ */
+#pragma once
+
+#include "device_communicator_adapter.cuh"

src/collective/nccl_device_communicator.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2022 XGBoost contributors
+ */
+#pragma once
+
+#include "nccl_device_communicator.cuh"

src/common/algorithm.hip.h

@@ -0,0 +1,6 @@
+/**
+ * Copyright 2022-2023 by XGBoost Contributors
+ */
+#pragma once
+
+#include "algorithm.cuh"      // Span,byte

src/common/cuda_context.hip.h

@@ -0,0 +1,6 @@
+/**
+ * Copyright 2022 by XGBoost Contributors
+ */
+#pragma once
+
+#include "cuda_context.cuh"

src/common/deterministic.hip.h

@@ -0,0 +1,6 @@
+/**
+ * Copyright 2020-2023 by XGBoost Contributors
+ */
+#pragma once
+
+#include "deterministic.cuh"  // XGBOOST_DEVICE

src/common/hist_util.hip.h

@@ -0,0 +1,9 @@
+/**
+ * Copyright 2020-2023 by XGBoost contributors
+ *
+ * \brief Front end and utilities for GPU based sketching.  Works on sliding window
+ *        instead of stream.
+ */
+#pragma once
+
+#include "hist_util.cuh"

src/common/linalg_op.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2021-2022 by XGBoost Contributors
+ */
+#pragma once
+
+#include "linalg_op.cuh"

src/common/quantile.hip.h

@@ -0,0 +1,3 @@
+#pragma once
+
+#include "quantile.cuh"

src/common/ranking_utils.hip.h

@@ -0,0 +1,6 @@
+/**
+ * Copyright 2023 by XGBoost Contributors
+ */
+#pragma once
+
+#include "ranking_utils.cuh"     // for Span

src/common/stats.hip.h

@@ -0,0 +1,6 @@
+/**
+ * Copyright 2022-2023 by XGBoost Contributors
+ */
+#pragma once
+
+#include "stats.cuh"                          // Span

src/common/threading_utils.hip.h

@@ -0,0 +1,6 @@
+/**
+ * Copyright 2021-2023 by XGBoost Contributors
+ */
+#pragma once
+
+#include "threading_utils.cuh"      // Span

src/data/device_adapter.hip.h

@@ -0,0 +1,7 @@
+/**
+ *  Copyright 2019-2023 by XGBoost Contributors
+ * \file device_adapter.cuh
+ */
+#pragma once
+
+#include "device_adapter.cuh"

src/data/ellpack_page.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2019 by XGBoost Contributors
+ */
+#pragma once
+
+#include "ellpack_page.cuh"

src/data/proxy_dmatrix.hip.h

@@ -0,0 +1,6 @@
+/**
+ * Copyright 2021-2023 XGBoost contributors
+ */
+#pragma once
+
+#include "proxy_dmatrix.cuh"

src/data/simple_dmatrix.hip.h

@@ -0,0 +1,7 @@
+/**
+ * Copyright 2019-2023 by XGBoost Contributors
+ * \file simple_dmatrix.cuh
+ */
+#pragma once
+
+#include "simple_dmatrix.cuh"     // for HasInfInData

src/objective/lambdarank_obj.hip.h

@@ -0,0 +1,6 @@
+/**
+ * Copyright 2023 XGBoost contributors
+ */
+#pragma once
+
+#include "lambdarank_obj.cuh"    // for Span

src/tree/constraints.hip.h

@@ -0,0 +1,8 @@
+/*!
+ * Copyright 2019 XGBoost contributors
+ *
+ * \file Various constraints used in GPU_Hist.
+ */
+#pragma once
+
+#include "constraints.cuh"

src/tree/gpu_hist/evaluate_splits.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2020 by XGBoost Contributors
+ */
+#pragma once
+
+#include "evaluate_splits.cuh"

src/tree/gpu_hist/expand_entry.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2020 by XGBoost Contributors
+ */
+#pragma once
+
+#include "expand_entry.cuh"

src/tree/gpu_hist/feature_groups.hip.h

@@ -0,0 +1,7 @@
+/*!
+ * Copyright 2020 by XGBoost Contributors
+ */
+
+#pragma once
+
+#include "feature_groups.cuh"

src/tree/gpu_hist/gradient_based_sampler.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2019 by XGBoost Contributors
+ */
+#pragma once
+
+#include "gradient_based_sampler.cuh"

src/tree/gpu_hist/histogram.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2020-2021 by XGBoost Contributors
+ */
+#pragma once
+
+#include "histogram.cuh"

src/tree/gpu_hist/row_partitioner.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2017-2022 XGBoost contributors
+ */
+#pragma once
+
+#include "row_partitioner.cuh"

src/tree/updater_gpu_common.hip.h

@@ -0,0 +1,6 @@
+/*!
+ * Copyright 2017-2019 XGBoost contributors
+ */
+#pragma once
+
+#include "updater_gpu_common.cuh"

tests/cpp/common/test_algorithm.cu

@@ -9,8 +9,13 @@
 #include <algorithm>          // is_sorted
 #include <cstddef>            // size_t
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/algorithm.cuh"
 #include "../../../src/common/device_helpers.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/common/algorithm.hip.h"
+#include "../../../src/common/device_helpers.hip.h"
+#endif
 #include "../helpers.h"  // CreateEmptyGenericParam
 
 namespace xgboost {

tests/cpp/common/test_hist_util.cu

@@ -10,11 +10,19 @@
 #include <cmath>
 
 #include "../../../include/xgboost/logging.h"
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/device_helpers.cuh"
 #include "../../../src/common/hist_util.cuh"
 #include "../../../src/common/hist_util.h"
 #include "../../../src/common/math.h"
 #include "../../../src/data/device_adapter.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/common/device_helpers.hip.h"
+#include "../../../src/common/hist_util.hip.h"
+#include "../../../src/common/hist_util.h"
+#include "../../../src/common/math.h"
+#include "../../../src/data/device_adapter.hip.h"
+#endif
 #include "../../../src/data/simple_dmatrix.h"
 #include "../data/test_array_interface.h"
 #include "../filesystem.h"  // dmlc::TemporaryDirectory

tests/cpp/common/test_linalg.cu

@@ -3,7 +3,11 @@
  */
 #include <gtest/gtest.h>
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/linalg_op.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/common/linalg_op.hip.h"
+#endif
 #include "xgboost/context.h"
 #include "xgboost/linalg.h"
 

tests/cpp/common/test_quantile.cu

@@ -1,9 +1,15 @@
 #include <gtest/gtest.h>
 #include "test_quantile.h"
 #include "../helpers.h"
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/collective/device_communicator.cuh"
 #include "../../../src/common/hist_util.cuh"
 #include "../../../src/common/quantile.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/collective/device_communicator.hip.h"
+#include "../../../src/common/hist_util.hip.h"
+#include "../../../src/common/quantile.hip.h"
+#endif
 
 namespace xgboost {
 namespace {

tests/cpp/common/test_ranking_utils.cu

@@ -11,10 +11,17 @@
 #include <numeric>                                 // for iota
 #include <vector>                                  // for vector
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/algorithm.cuh"       // for SegmentedSequence
 #include "../../../src/common/cuda_context.cuh"    // for CUDAContext
 #include "../../../src/common/device_helpers.cuh"  // for device_vector, ToSpan
 #include "../../../src/common/ranking_utils.cuh"   // for CalcQueriesInvIDCG
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/common/algorithm.hip.h"       // for SegmentedSequence
+#include "../../../src/common/cuda_context.hip.h"    // for CUDAContext
+#include "../../../src/common/device_helpers.hip.h"  // for device_vector, ToSpan
+#include "../../../src/common/ranking_utils.hip.h"   // for CalcQueriesInvIDCG
+#endif
 #include "../../../src/common/ranking_utils.h"     // for LambdaRankParam, RankingCache
 #include "../helpers.h"                            // for EmptyDMatrix
 #include "test_ranking_utils.h"                    // for TestNDCGCache

tests/cpp/common/test_stats.cu

@@ -7,8 +7,13 @@
 #include <utility>                            // std::pair
 #include <vector>                             // std::vector
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/linalg_op.cuh"  // ElementWiseTransformDevice
 #include "../../../src/common/stats.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/common/linalg_op.hip.h"  // ElementWiseTransformDevice
+#include "../../../src/common/stats.hip.h"
+#endif
 #include "xgboost/base.h"                     // XGBOOST_DEVICE
 #include "xgboost/context.h"                  // Context
 #include "xgboost/host_device_vector.h"       // HostDeviceVector

tests/cpp/common/test_threading_utils.cu

@@ -4,8 +4,13 @@
 #include <gtest/gtest.h>
 #include <thrust/copy.h>  // thrust::copy
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/device_helpers.cuh"
 #include "../../../src/common/threading_utils.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/common/device_helpers.hip.h"
+#include "../../../src/common/threading_utils.hip.h"
+#endif
 
 namespace xgboost {
 namespace common {

tests/cpp/data/test_device_adapter.cu

@@ -7,7 +7,11 @@
 #include "../helpers.h"
 #include <thrust/device_vector.h>
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/data/device_adapter.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/data/device_adapter.hip.h"
+#endif
 #include "test_array_interface.h"
 
 using namespace xgboost;  // NOLINT

tests/cpp/data/test_iterative_dmatrix.cu

@@ -3,8 +3,13 @@
  */
 #include <gtest/gtest.h>
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/data/device_adapter.cuh"
 #include "../../../src/data/ellpack_page.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/data/device_adapter.hip.h"
+#include "../../../src/data/ellpack_page.hip.h"
+#endif
 #include "../../../src/data/iterative_dmatrix.h"
 #include "../../../src/tree/param.h"  // TrainParam
 #include "../helpers.h"

tests/cpp/data/test_proxy_dmatrix.cu

@@ -7,7 +7,11 @@
 #include <any>  // for any_cast
 #include <memory>
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/data/device_adapter.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/data/device_adapter.hip.h"
+#endif
 #include "../../../src/data/proxy_dmatrix.h"
 #include "../helpers.h"
 

tests/cpp/data/test_simple_dmatrix.cu

@@ -4,7 +4,11 @@
 
 #include <thrust/sequence.h>
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/data/device_adapter.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/data/device_adapter.hip.h"
+#endif
 #include "../helpers.h"
 #include "test_array_interface.h"
 #include "../../../src/data/array_interface.h"

tests/cpp/helpers.cu

@@ -1,7 +1,11 @@
 #include <xgboost/c_api.h>
 
 #include "helpers.h"
+#if defined(XGBOOST_USE_CUDA)
 #include "../../src/data/device_adapter.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../src/data/device_adapter.hip.h"
+#endif
 #include "../../src/data/iterative_dmatrix.h"
 
 namespace xgboost {

tests/cpp/objective/test_lambdarank_obj.cu

@@ -7,8 +7,13 @@
 #include <cstdint>                               // for uint32_t
 #include <vector>                                // for vector
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/common/cuda_context.cuh"  // for CUDAContext
 #include "../../../src/objective/lambdarank_obj.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/common/cuda_context.hip.h"  // for CUDAContext
+#include "../../../src/objective/lambdarank_obj.hip.h"
+#endif
 #include "test_lambdarank_obj.h"
 
 namespace xgboost::obj {

tests/cpp/predictor/test_gpu_predictor.cu

@@ -9,7 +9,11 @@
 
 #include <string>
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/data/device_adapter.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/data/device_adapter.hip.h"
+#endif
 #include "../../../src/data/proxy_dmatrix.h"
 #include "../../../src/gbm/gbtree_model.h"
 #include "../helpers.h"

tests/cpp/tree/gpu_hist/test_driver.cu

@@ -1,6 +1,10 @@
 #include <gtest/gtest.h>
 #include "../../../../src/tree/driver.h"
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../../src/tree/gpu_hist/expand_entry.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../../src/tree/gpu_hist/expand_entry.hip.h"
+#endif
 
 namespace xgboost {
 namespace tree {

tests/cpp/tree/gpu_hist/test_evaluate_splits.cu

@@ -3,7 +3,11 @@
  */
 #include <gtest/gtest.h>
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../../src/tree/gpu_hist/evaluate_splits.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../../src/tree/gpu_hist/evaluate_splits.hip.h"
+#endif
 #include "../../helpers.h"
 #include "../../histogram_helpers.h"
 #include "../test_evaluate_splits.h"  // TestPartitionBasedSplit

tests/cpp/tree/gpu_hist/test_histogram.cu

@@ -6,8 +6,13 @@
 #include <vector>
 
 #include "../../../../src/common/categorical.h"
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../../src/tree/gpu_hist/histogram.cuh"
 #include "../../../../src/tree/gpu_hist/row_partitioner.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../../src/tree/gpu_hist/histogram.hip.h"
+#include "../../../../src/tree/gpu_hist/row_partitioner.hip.h"
+#endif
 #include "../../../../src/tree/param.h"  // TrainParam
 #include "../../categorical_helpers.h"
 #include "../../helpers.h"

tests/cpp/tree/gpu_hist/test_row_partitioner.cu

@@ -9,7 +9,11 @@
 #include <algorithm>
 #include <vector>
 
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../../src/tree/gpu_hist/row_partitioner.cuh"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../../src/tree/gpu_hist/row_partitioner.hip.h"
+#endif
 #include "../../helpers.h"
 #include "xgboost/base.h"
 #include "xgboost/context.h"

tests/cpp/tree/test_gpu_hist.cu

@@ -12,10 +12,17 @@
 
 #include "../../../src/common/common.h"
 #include "../../../src/data/sparse_page_source.h"
+#if defined(XGBOOST_USE_CUDA)
 #include "../../../src/tree/constraints.cuh"
 #include "../../../src/tree/param.h"  // for TrainParam
 #include "../../../src/tree/updater_gpu_common.cuh"
 #include "../../../src/tree/updater_gpu_hist.cu"
+#elif defined(XGBOOST_USE_HIP)
+#include "../../../src/tree/constraints.hip.h"
+#include "../../../src/tree/param.h"  // for TrainParam
+#include "../../../src/tree/updater_gpu_common.hip.h"
+#include "../../../src/tree/updater_gpu_hist.hip"
+#endif
 #include "../filesystem.h"  // dmlc::TemporaryDirectory
 #include "../helpers.h"
 #include "../histogram_helpers.h"