feature: enabling oneDPL and sort primitive refactoring

.ci/pipeline/ci.yml

@@ -29,7 +29,8 @@ variables:
   VM_IMAGE : 'ubuntu-22.04'
   SYSROOT_OS: 'jammy'
   WINDOWS_BASEKIT_URL: 'https://registrationcenter-download.intel.com/akdlm/IRC_NAS/b380d914-366b-4b77-a74a-05e3c38b3514/intel-oneapi-base-toolkit-2025.0.0.882_offline.exe'
-  WINDOWS_DPCPP_COMPONENTS: 'intel.oneapi.win.mkl.devel:intel.oneapi.win.tbb.devel'
+  WINDOWS_DPCPP_COMPONENTS: 'intel.oneapi.win.mkl.devel:intel.oneapi.win.tbb.devel:intel.oneapi.win.dpl'
+  LINUX_DPL_URL: 'https://registrationcenter-download.intel.com/akdlm/IRC_NAS/de3c613f-829c-4bdc-aa2b-6129eece3bd9/intel-onedpl-2022.7.1.15_offline.sh'
 
 resources:
   repositories:
@@ -71,6 +72,9 @@ jobs:
   - script: |
       .ci/env/apt.sh mkl
     displayName: 'mkl installation'
+  - script: 
+      chmod +x .ci/scripts/install_dpl.sh && .ci/scripts/install_dpl.sh $(LINUX_DPL_URL)
+    displayName: 'oneDPL installation'
   - script: |
       source /opt/intel/oneapi/setvars.sh
       .ci/scripts/describe_system.sh
@@ -393,6 +397,9 @@ jobs:
   - script: |
       .ci/env/apt.sh mkl
     displayName: 'mkl installation'
+  - script: 
+      chmod +x .ci/scripts/install_dpl.sh && .ci/scripts/install_dpl.sh $(LINUX_DPL_URL)
+    displayName: 'oneDPL installation'
   - script: |
       source /opt/intel/oneapi/setvars.sh
       .ci/scripts/describe_system.sh

.ci/scripts/install_dpl.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+#===============================================================================
+# Copyright contributors to the oneDAL project
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#===============================================================================
+
+URL=$1
+
+# Download the installation script
+curl --output installer.sh --url "$URL" --retry 5 --retry-delay 5
+chmod +x installer.sh
+
+# Execute the installation script
+sudo sh installer.sh -a --silent --eula accept
+installer_exit_code=$?
+
+# Clean up
+rm -f installer.sh
+
+exit $installer_exit_code

INSTALL.md

@@ -23,6 +23,7 @@ Required Software:
 * BLAS and LAPACK libraries - both provided by oneMKL
 * Python version 3.9 or higher
 * TBB library (repository contains script to download it)
+* oneDPL library
 * Microsoft Visual Studio\* (Windows\* only)
 * [MSYS2](http://msys2.github.io) (Windows\* only)
 * `make` and `dos2unix` tools; install these packages using MSYS2 on Windows\* as follows:
@@ -113,9 +114,25 @@ is available as an alternative to the manual setup.
 
             ./dev/download_tbb.sh
 
-6. Download and install Python (version 3.9 or higher).
+6. Set up Intel(R) Threading Building Blocks (Intel(R) TBB):
 
-7. Build oneDAL via command-line interface. Choose the appropriate commands based on the interface, platform, and the compiler you use. Interface and platform are required arguments of makefile while others are optional. Below you can find the set of examples for building oneDAL. You may use a combination of them to get the desired build configuration:
+    _Note: if you used the general oneAPI setvars script from a Base Toolkit installation, this step will not be necessary as oneDPL will already have been set up._
+
+    Download and install [Intel(R) oneDPL](https://www.intel.com/content/www/us/en/developer/tools/oneapi/dpc-library.html).
+    Set the environment variables for for Intel(R) oneDPL. For example:
+
+    - oneDPL (Windows\*):
+
+            call "C:\Program Files (x86)\Intel\oneAPI\dpl\latest\env\vars.bat" intel64
+
+    - oneDPL (Linux\*):
+
+            source /opt/intel/oneapi/dpl/latest/env/vars.sh intel64
+
+
+7. Download and install Python (version 3.9 or higher).
+
+8. Build oneDAL via command-line interface. Choose the appropriate commands based on the interface, platform, and the compiler you use. Interface and platform are required arguments of makefile while others are optional. Below you can find the set of examples for building oneDAL. You may use a combination of them to get the desired build configuration:
 
     - DAAL interfaces on **Linux\*** using **Intel(R) C++ Compiler**:
 

WORKSPACE

@@ -79,6 +79,21 @@ ccl_repo(
     ]
 )
 
+load("@onedal//dev/bazel/deps:dpl.bzl", "dpl_repo")
+dpl_repo(
+    name = "dpl",
+    root_env_var = "DPL_ROOT",
+    urls = [
+        "https://files.pythonhosted.org/packages/95/f6/18f78cb933e01ecd9e99d37a10da4971a795fcfdd1d24640799b4050fdbb/onedpl_devel-2022.7.1-py2.py3-none-manylinux_2_28_x86_64.whl",
+    ],
+    sha256s = [
+        "3b270999d2464c5151aa0e7995dda9e896d072c75069ccee1efae9dc56bdc417",
+    ],
+    strip_prefixes = [
+        "onedpl_devel-2022.7.1.data/data",
+    ],
+)
+
 load("@onedal//dev/bazel/deps:mkl.bzl", "mkl_repo")
 mkl_repo(
     name = "mkl",

cpp/oneapi/dal/BUILD

@@ -31,6 +31,7 @@ dal_module(
     ],
     dpc_deps = [
         "@mkl//:mkl_dpc",
+        "@dpl//:headers",
     ],
 )
 

cpp/oneapi/dal/algo/decision_forest/backend/gpu/train_feature_type_dpc.cpp

@@ -35,7 +35,7 @@ inline sycl::event sort_inplace(sycl::queue& queue_,
                                 const bk::event_vector& deps = {}) {
     ONEDAL_ASSERT(src.get_count() > 0);
     auto src_ind = pr::ndarray<Index, 1>::empty(queue_, { src.get_count() });
-    return pr::radix_sort_indices_inplace<Float, Index>{ queue_ }(src, src_ind, deps);
+    return pr::radix_sort_indices_inplace<Float, Index>(queue_, src, src_ind, deps);
 }
 
 template <typename Float, typename Bin, typename Index>
@@ -429,13 +429,14 @@ sycl::event indexed_features<Float, Bin, Index>::operator()(const table& tbl,
             pr::ndarray<Bin, 1>::empty(queue_, { row_count_ }, sycl::usm::alloc::device);
     }
 
-    pr::radix_sort_indices_inplace<Float, Index> sort{ queue_ };
-
     sycl::event last_event;
 
     for (Index i = 0; i < column_count_; i++) {
         last_event = extract_column(data_nd_, values_nd, indices_nd, i, { last_event });
-        last_event = sort(values_nd, indices_nd, { last_event });
+        last_event = pr::radix_sort_indices_inplace<Float, Index>(queue_,
+                                                                  values_nd,
+                                                                  indices_nd,
+                                                                  { last_event });
         last_event =
             compute_bins(values_nd, indices_nd, column_bin_vec_[i], entries_[i], i, { last_event });
     }

cpp/oneapi/dal/algo/decision_forest/backend/gpu/train_kernel_hist_impl.hpp

@@ -21,7 +21,7 @@
 #include "oneapi/dal/backend/primitives/utils.hpp"
 #include "oneapi/dal/algo/decision_forest/train_types.hpp"
 
-#include "oneapi/dal/backend/primitives/rng/host_engine_collection.hpp"
+#include "oneapi/dal/backend/primitives/rng/device_engine.hpp"
 
 #include "oneapi/dal/algo/decision_forest/backend/gpu/train_misc_structs.hpp"
 #include "oneapi/dal/algo/decision_forest/backend/gpu/train_impurity_data.hpp"
@@ -50,8 +50,7 @@ class train_kernel_hist_impl {
     using model_manager_t = train_model_manager<Float, Index, Task>;
     using train_context_t = train_context<Float, Index, Task>;
     using imp_data_t = impurity_data<Float, Index, Task>;
-    using rng_engine_t = pr::host_engine;
-    using rng_engine_list_t = std::vector<rng_engine_t>;
+    using rng_engine_list_t = ::oneapi::dal::backend::primitives::device_engine;
     using msg = dal::detail::error_messages;
     using comm_t = bk::communicator<spmd::device_memory_access::usm>;
     using node_t = node<Index>;
@@ -575,7 +574,7 @@ class train_kernel_hist_impl {
                                 pr::ndarray<hist_type_t, 1>& oob_per_obs_list,
                                 pr::ndarray<Float, 1>& var_imp,
                                 pr::ndarray<Float, 1>& var_imp_variance,
-                                const rng_engine_list_t& rng_engine_arr,
+                                rng_engine_list_t& rng_engine_arr,
                                 Index tree_idx,
                                 Index tree_in_block,
                                 Index built_tree_count,