[RNG] add RNG armpl backend

CMakeLists.txt

@@ -106,7 +106,8 @@ endif()
 if(ENABLE_MKLCPU_BACKEND
         OR ENABLE_MKLGPU_BACKEND
         OR ENABLE_CURAND_BACKEND
-        OR ENABLE_ROCRAND_BACKEND)
+        OR ENABLE_ROCRAND_BACKEND
+        OR ENABLE_ARMPL_BACKEND)
   list(APPEND DOMAINS_LIST "rng")
 endif()
 if(ENABLE_MKLGPU_BACKEND

README.md

@@ -276,12 +276,18 @@ Supported compilers include:
             <td align="center">Dynamic, Static</td>
         </tr>
         <tr>
-            <td rowspan=4 align="center">RNG</td>
+            <td rowspan=5 align="center">RNG</td>
             <td align="center">x86 CPU</td>
             <td align="center">Intel(R) oneMKL</td>
             <td align="center">Intel DPC++</br>AdaptiveCpp</td>
             <td align="center">Dynamic, Static</td>
         </tr>
+        <tr>
+            <td align="center">aarch64 CPU</td>
+            <td align="center">Arm Performance Libraries</td>
+            <td align="center">Open DPC++</br>AdaptiveCpp</td>
+            <td align="center">Dynamic, Static</td>
+        </tr>
         <tr>
             <td align="center">Intel GPU</td>
             <td align="center">Intel(R) oneMKL</td>

include/oneapi/math/detail/backends_table.hpp

@@ -175,6 +175,12 @@ static std::map<domain, std::map<device, std::vector<const char*>>> libraries =
           {
 #ifdef ONEMATH_ENABLE_MKLCPU_BACKEND
               LIB_NAME("rng_mklcpu")
+#endif
+          } },
+        { device::aarch64cpu,
+          {
+#ifdef ONEMATH_ENABLE_ARMPL_BACKEND
+              LIB_NAME("rng_armpl")
 #endif
           } },
         { device::intelgpu,

include/oneapi/math/rng/detail/armpl/onemath_rng_armpl.hpp

@@ -0,0 +1,56 @@
+/*******************************************************************************
+* Copyright 2025 SiPearl
+* Copyright 2020-2021 Intel Corporation
+*
+* 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.
+*
+*
+* SPDX-License-Identifier: Apache-2.0
+*******************************************************************************/
+
+#ifndef _ONEMATH_RNG_ARMPL_HPP_
+#define _ONEMATH_RNG_ARMPL_HPP_
+
+#include <cstdint>
+#if __has_include(<sycl/sycl.hpp>)
+#include <sycl/sycl.hpp>
+#else
+#include <CL/sycl.hpp>
+#endif
+
+#include "oneapi/math/detail/export.hpp"
+#include "oneapi/math/rng/detail/engine_impl.hpp"
+
+namespace oneapi {
+namespace math {
+namespace rng {
+namespace armpl {
+
+ONEMATH_EXPORT oneapi::math::rng::detail::engine_impl* create_philox4x32x10(sycl::queue queue,
+                                                                            std::uint64_t seed);
+
+ONEMATH_EXPORT oneapi::math::rng::detail::engine_impl* create_philox4x32x10(
+    sycl::queue queue, std::initializer_list<std::uint64_t> seed);
+
+ONEMATH_EXPORT oneapi::math::rng::detail::engine_impl* create_mrg32k3a(sycl::queue queue,
+                                                                       std::uint32_t seed);
+
+ONEMATH_EXPORT oneapi::math::rng::detail::engine_impl* create_mrg32k3a(
+    sycl::queue queue, std::initializer_list<std::uint32_t> seed);
+
+} // namespace armpl
+} // namespace rng
+} // namespace math
+} // namespace oneapi
+
+#endif //_ONEMATH_RNG_ARMPL_HPP_

include/oneapi/math/rng/engines.hpp

@@ -47,6 +47,9 @@
 #ifdef ONEMATH_ENABLE_ROCRAND_BACKEND
 #include "oneapi/math/rng/detail/rocrand/onemath_rng_rocrand.hpp"
 #endif
+#ifdef ONEMATH_ENABLE_ARMPL_BACKEND
+#include "oneapi/math/rng/detail/armpl/onemath_rng_armpl.hpp"
+#endif
 
 namespace oneapi {
 namespace math {
@@ -103,6 +106,15 @@ class philox4x32x10 {
             : pimpl_(rocrand::create_philox4x32x10(selector.get_queue(), seed)) {}
 #endif
 
+#ifdef ONEMATH_ENABLE_ARMPL_BACKEND
+    philox4x32x10(backend_selector<backend::armpl> selector, std::uint64_t seed = default_seed)
+            : pimpl_(armpl::create_philox4x32x10(selector.get_queue(), seed)) {}
+
+    philox4x32x10(backend_selector<backend::armpl> selector,
+                  std::initializer_list<std::uint64_t> seed)
+            : pimpl_(armpl::create_philox4x32x10(selector.get_queue(), seed)) {}
+#endif
+
     philox4x32x10(const philox4x32x10& other) {
         pimpl_.reset(other.pimpl_.get()->copy_state());
     }
@@ -192,6 +204,14 @@ class mrg32k3a {
             : pimpl_(rocrand::create_mrg32k3a(selector.get_queue(), seed)) {}
 #endif
 
+#ifdef ONEMATH_ENABLE_ARMPL_BACKEND
+    mrg32k3a(backend_selector<backend::armpl> selector, std::uint32_t seed = default_seed)
+            : pimpl_(armpl::create_mrg32k3a(selector.get_queue(), seed)) {}
+
+    mrg32k3a(backend_selector<backend::armpl> selector, std::initializer_list<std::uint32_t> seed)
+            : pimpl_(armpl::create_mrg32k3a(selector.get_queue(), seed)) {}
+#endif
+
     mrg32k3a(const mrg32k3a& other) {
         pimpl_.reset(other.pimpl_.get()->copy_state());
     }

src/rng/backends/CMakeLists.txt

@@ -36,3 +36,8 @@ if(ENABLE_ROCRAND_BACKEND AND UNIX)
   add_subdirectory(rocrand)
 endif()
 
+if(ENABLE_ARMPL_BACKEND AND UNIX)
+  add_subdirectory(armpl)
+endif()
+
+

src/rng/backends/armpl/CMakeLists.txt

@@ -0,0 +1,76 @@
+#===============================================================================
+# Copyright 2025 SiPearl
+# Copyright 2020-2021 Intel Corporation
+#
+# 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.
+#
+#
+# SPDX-License-Identifier: Apache-2.0
+#===============================================================================
+
+set(LIB_NAME onemath_rng_armpl)
+set(LIB_OBJ ${LIB_NAME}_obj)
+
+find_package(ARMPL REQUIRED)
+
+set(SOURCES armpl_common.hpp
+  philox4x32x10.cpp
+  mrg32k3a.cpp
+  $<$<BOOL:${BUILD_SHARED_LIBS}>: armpl_rng_cpu_wrappers.cpp>
+)
+
+add_library(${LIB_NAME})
+add_library(${LIB_OBJ} OBJECT ${SOURCES})
+add_dependencies(onemath_backend_libs_rng ${LIB_NAME})
+target_include_directories(${LIB_OBJ}
+  PRIVATE ${PROJECT_SOURCE_DIR}/include
+          ${PROJECT_SOURCE_DIR}/src
+          ${CMAKE_BINARY_DIR}/bin
+          ${ARMPL_INCLUDE}
+          ${ONEMATH_GENERATED_INCLUDE_PATH}
+)
+
+target_compile_options(${LIB_OBJ} PRIVATE ${ONEMATH_BUILD_COPT} ${ARMPL_COPT})
+if (USE_ADD_SYCL_TO_TARGET_INTEGRATION)
+  add_sycl_to_target(TARGET ${LIB_OBJ} SOURCES ${SOURCES})
+endif()
+
+target_link_libraries(${LIB_OBJ} PUBLIC ONEMATH::SYCL::SYCL ${ARMPL_LINK})
+
+set_target_properties(${LIB_OBJ} PROPERTIES
+  POSITION_INDEPENDENT_CODE ON
+)
+target_link_libraries(${LIB_NAME} PUBLIC ${LIB_OBJ})
+
+# Set oneMATH libraries as not transitive for dynamic
+if(BUILD_SHARED_LIBS)
+  set_target_properties(${LIB_NAME} PROPERTIES
+	  INTERFACE_LINK_LIBRARIES ONEMATH::SYCL::SYCL
+  )
+endif()
+
+# Add major version to the library
+set_target_properties(${LIB_NAME} PROPERTIES
+  SOVERSION ${PROJECT_VERSION_MAJOR}
+)
+
+# Add dependencies rpath to the library
+list(APPEND CMAKE_BUILD_RPATH $<TARGET_FILE_DIR:${LIB_NAME}>)
+
+# Add the library to install package
+install(TARGETS ${LIB_OBJ} EXPORT oneMathTargets)
+install(TARGETS ${LIB_NAME} EXPORT oneMathTargets
+  RUNTIME DESTINATION bin
+  ARCHIVE DESTINATION lib
+  LIBRARY DESTINATION lib
+)

src/rng/backends/armpl/armpl_common.hpp

@@ -0,0 +1,86 @@
+/*******************************************************************************
+* Copyright 2025 SiPearl
+* Copyright 2020-2021 Intel Corporation
+*
+* 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.
+*
+*
+* SPDX-License-Identifier: Apache-2.0
+*******************************************************************************/
+
+#ifndef _RNG_CPU_COMMON_HPP_
+#define _RNG_CPU_COMMON_HPP_
+
+#if __has_include(<sycl/sycl.hpp>)
+#include <sycl/sycl.hpp>
+#else
+#include <CL/sycl.hpp>
+#endif
+
+#define GET_MULTI_PTR template get_multi_ptr<sycl::access::decorated::yes>().get_raw()
+#define __fp16        _Float16
+#define INTEGER64     1
+
+#include "armpl.h"
+
+namespace oneapi {
+namespace math {
+namespace rng {
+namespace armpl {
+
+inline int check_armpl_version(armpl_int_t major_req, armpl_int_t minor_req, armpl_int_t build_req,
+                               const char* message) {
+    armpl_int_t major, minor, build;
+    char* tag;
+    armplversion(&major, &minor, &build, (const char**)&tag);
+    if (major > major_req) {
+        return 0;
+    }
+    else if (major == major_req && minor > minor_req) {
+        return 0;
+    }
+    else if (major == major_req && minor == minor_req && build >= build_req) {
+        return 0;
+    }
+    throw oneapi::math::unimplemented("rng", "version support", message);
+}
+
+template <typename K, typename H, typename F>
+static inline auto host_task_internal(H& cgh, F f, int) -> decltype(cgh.host_task(f)) {
+    return cgh.host_task(f);
+}
+
+template <typename K, typename H, typename F>
+static inline void host_task_internal(H& cgh, F f, long) {
+#ifndef __SYCL_DEVICE_ONLY__
+    cgh.template single_task<K>(f);
+#endif
+}
+
+template <typename K, typename H, typename F>
+static inline void host_task(H& cgh, F f) {
+    (void)host_task_internal<K>(cgh, f, 0);
+}
+
+template <typename Engine, typename Distr>
+class kernel_name {};
+
+template <typename Engine, typename Distr>
+class kernel_name_usm {};
+
+} // namespace armpl
+} // namespace rng
+} // namespace math
+} // namespace oneapi
+
+#endif //_RNG_CPU_COMMON_HPP_

src/rng/backends/armpl/armpl_rng_cpu_wrappers.cpp

@@ -0,0 +1,30 @@
+/*******************************************************************************
+* Copyright 2025 SiPearl
+* Copyright 2020-2021 Intel Corporation
+*
+* 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.
+*
+*
+* SPDX-License-Identifier: Apache-2.0
+*******************************************************************************/
+
+#include "rng/function_table.hpp"
+#include "oneapi/math/rng/detail/armpl/onemath_rng_armpl.hpp"
+
+#define WRAPPER_VERSION 1
+
+extern "C" ONEMATH_EXPORT rng_function_table_t onemath_rng_table = {
+    WRAPPER_VERSION, oneapi::math::rng::armpl::create_philox4x32x10,
+    oneapi::math::rng::armpl::create_philox4x32x10, oneapi::math::rng::armpl::create_mrg32k3a,
+    oneapi::math::rng::armpl::create_mrg32k3a
+};