Merge branch 'main' into dev/skopienko/tag_dispatching

oneapi-src · Nov 16, 2023 · 19a0f87 · 19a0f87
2 parents efafbb2 + 53a4a2c
commit 19a0f87
Show file tree

Hide file tree

Showing 170 changed files with 15,780 additions and 108 deletions.
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -15,6 +15,10 @@ env:
   TEST_TIMEOUT: 360
   MACOS_ONEAPI_DOWNLOAD_LINK: https://registrationcenter-download.intel.com/akdlm/irc_nas/18358/m_cpp-compiler-classic_p_2022.0.0.62_offline.dmg
 
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
 jobs:
   clang-format:
     if: github.event_name == 'pull_request'

diff --git a/documentation/release_notes.rst b/documentation/release_notes.rst
@@ -36,6 +36,10 @@ New in This Release
   ``libstdc++`` version 8 or ``libc++``, ``oneapi::dpl::execution::par_unseq`` offloads
   standard parallel algorithms to the SYCL device similarly to ``std::execution::par_unseq``
   in accordance with the ``-fsycl-pstl-offload`` option value.
+- When using the dpl modulefile to initialize the user's environment and compiling with ``-fsycl-pstl-offload``
+  option of Intel® oneAPI DPC++/C++ compiler, a linking issue or program crash may be encountered due to the directory
+  containing libpstloffload.so not being included in the search path. Use the env/vars.sh to configure the working
+  environment to avoid the issue.
 - Compilation issues may be encountered when passing zip iterators to ``exclusive_scan_by_segment`` on Windows.
 - Incorrect results may be produced by ``set_intersection`` with a DPC++ execution policy,
   where elements are copied from the second input range rather than the first input range. 

diff --git a/include/oneapi/dpl/pstl/algorithm_fwd.h b/include/oneapi/dpl/pstl/algorithm_fwd.h
@@ -249,6 +249,24 @@ __pattern_walk3(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _Ran
                 _RandomAccessIterator2 __first2, _RandomAccessIterator3 __first3, _Function __f, _IsVector __is_vector,
                 /*parallel=*/::std::true_type);
 
+//------------------------------------------------------------------------
+// transform_if
+//------------------------------------------------------------------------
+
+template <typename _ExecutionPolicy, typename _ForwardIterator1, typename _ForwardIterator2, typename _Function,
+          class _IsVector, class _IsParallel>
+oneapi::dpl::__internal::__enable_if_host_execution_policy<_ExecutionPolicy, _ForwardIterator2>
+__pattern_walk2_transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                             _ForwardIterator2 __first2, _Function __func, _IsVector __is_vector,
+                             _IsParallel __is_parallel) noexcept;
+
+template <typename _ExecutionPolicy, typename _ForwardIterator1, typename _ForwardIterator2, typename _ForwardIterator3,
+          typename _Function, class _IsVector, class _IsParallel>
+oneapi::dpl::__internal::__enable_if_host_execution_policy<_ExecutionPolicy, _ForwardIterator3>
+__pattern_walk3_transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                             _ForwardIterator2 __first2, _ForwardIterator3 __first3, _Function __func,
+                             _IsVector __is_vector, _IsParallel __is_parallel) noexcept;
+
 //------------------------------------------------------------------------
 // equal
 //------------------------------------------------------------------------

diff --git a/include/oneapi/dpl/pstl/algorithm_impl.h b/include/oneapi/dpl/pstl/algorithm_impl.h
@@ -594,6 +594,32 @@ __pattern_walk3(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardI
     });
 }
 
+//------------------------------------------------------------------------
+// transform_if
+//------------------------------------------------------------------------
+
+template <typename _ExecutionPolicy, typename _ForwardIterator1, typename _ForwardIterator2, typename _Function,
+          class _IsVector, class _IsParallel>
+oneapi::dpl::__internal::__enable_if_host_execution_policy<_ExecutionPolicy, _ForwardIterator2>
+__pattern_walk2_transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                             _ForwardIterator2 __first2, _Function __func, _IsVector __is_vector,
+                             _IsParallel __is_parallel) noexcept
+{
+    return __pattern_walk2(::std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __func, __is_vector,
+                           __is_parallel);
+}
+
+template <typename _ExecutionPolicy, typename _ForwardIterator1, typename _ForwardIterator2, typename _ForwardIterator3,
+          typename _Function, class _IsVector, class _IsParallel>
+oneapi::dpl::__internal::__enable_if_host_execution_policy<_ExecutionPolicy, _ForwardIterator3>
+__pattern_walk3_transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                             _ForwardIterator2 __first2, _ForwardIterator3 __first3, _Function __func,
+                             _IsVector __is_vector, _IsParallel __is_parallel) noexcept
+{
+    return __pattern_walk3(::std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __first3, __func,
+                           __is_vector, __is_parallel);
+}
+
 //------------------------------------------------------------------------
 // equal
 //------------------------------------------------------------------------

diff --git a/include/oneapi/dpl/pstl/glue_algorithm_defs.h b/include/oneapi/dpl/pstl/glue_algorithm_defs.h
@@ -173,6 +173,20 @@ oneapi::dpl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _Forward
 transform(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result,
           _UnaryOperation __op);
 
+// [alg.transform_if]
+
+template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _UnaryOperation,
+          class _UnaryPredicate>
+oneapi::dpl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator2>
+transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result,
+             _UnaryOperation __op, _UnaryPredicate __pred);
+
+template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator3,
+          class _BinaryOperation, class _BinaryPredicate>
+oneapi::dpl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator3>
+transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+             _ForwardIterator2 __first2, _ForwardIterator3 __result, _BinaryOperation __op, _BinaryPredicate __pred);
+
 // [alg.replace]
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _Tp>

diff --git a/include/oneapi/dpl/pstl/glue_algorithm_impl.h b/include/oneapi/dpl/pstl/glue_algorithm_impl.h
@@ -327,6 +327,7 @@ swap_ranges(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardItera
 }
 
 // [alg.transform]
+
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _UnaryOperation>
 oneapi::dpl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator2>
 transform(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result,
@@ -356,6 +357,40 @@ transform(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterato
         __exec.__allow_parallel());
 }
 
+// [alg.transform_if]
+
+template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _UnaryOperation,
+          class _UnaryPredicate>
+oneapi::dpl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator2>
+transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result,
+             _UnaryOperation __op, _UnaryPredicate __pred)
+{
+    return oneapi::dpl::__internal::__pattern_walk2_transform_if(
+        ::std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
+        oneapi::dpl::__internal::__transform_if_unary_functor<_UnaryOperation, _UnaryPredicate>(::std::move(__op),
+                                                                                                ::std::move(__pred)),
+        oneapi::dpl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
+            __exec),
+        oneapi::dpl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
+            __exec));
+}
+
+template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator3,
+          class _BinaryOperation, class _BinaryPredicate>
+oneapi::dpl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator3>
+transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+             _ForwardIterator2 __first2, _ForwardIterator3 __result, _BinaryOperation __op, _BinaryPredicate __pred)
+{
+    return oneapi::dpl::__internal::__pattern_walk3_transform_if(
+        ::std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __result,
+        oneapi::dpl::__internal::__transform_if_binary_functor<_BinaryOperation, _BinaryPredicate>(::std::move(__op),
+                                                                                                   ::std::move(__pred)),
+        oneapi::dpl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
+                                                              _ForwardIterator3>(__exec),
+        oneapi::dpl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
+                                                                _ForwardIterator3>(__exec));
+}
+
 // [alg.replace]
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _Tp>

diff --git a/include/oneapi/dpl/pstl/hetero/algorithm_impl_hetero.h b/include/oneapi/dpl/pstl/hetero/algorithm_impl_hetero.h
@@ -155,7 +155,10 @@ __pattern_swap(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIt
 // walk3
 //------------------------------------------------------------------------
 
-template <typename _ExecutionPolicy, typename _ForwardIterator1, typename _ForwardIterator2, typename _ForwardIterator3,
+template <__par_backend_hetero::access_mode __acc_mode1 = __par_backend_hetero::access_mode::read,
+          __par_backend_hetero::access_mode __acc_mode2 = __par_backend_hetero::access_mode::read,
+          __par_backend_hetero::access_mode __acc_mode3 = __par_backend_hetero::access_mode::write,
+          typename _ExecutionPolicy, typename _ForwardIterator1, typename _ForwardIterator2, typename _ForwardIterator3,
           typename _Function>
 oneapi::dpl::__internal::__enable_if_hetero_execution_policy<_ExecutionPolicy, _ForwardIterator3>
 __pattern_walk3(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
@@ -166,14 +169,11 @@ __pattern_walk3(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardI
     if (__n <= 0)
         return __first3;
 
-    auto __keep1 =
-        oneapi::dpl::__ranges::__get_sycl_range<__par_backend_hetero::access_mode::read, _ForwardIterator1>();
+    auto __keep1 = oneapi::dpl::__ranges::__get_sycl_range<__acc_mode1, _ForwardIterator1>();
     auto __buf1 = __keep1(__first1, __last1);
-    auto __keep2 =
-        oneapi::dpl::__ranges::__get_sycl_range<__par_backend_hetero::access_mode::read, _ForwardIterator2>();
+    auto __keep2 = oneapi::dpl::__ranges::__get_sycl_range<__acc_mode2, _ForwardIterator2>();
     auto __buf2 = __keep2(__first2, __first2 + __n);
-    auto __keep3 =
-        oneapi::dpl::__ranges::__get_sycl_range<__par_backend_hetero::access_mode::write, _ForwardIterator3>();
+    auto __keep3 = oneapi::dpl::__ranges::__get_sycl_range<__acc_mode3, _ForwardIterator3>();
     auto __buf3 = __keep3(__first3, __first3 + __n);
 
     oneapi::dpl::__par_backend_hetero::__parallel_for(::std::forward<_ExecutionPolicy>(__exec),
@@ -262,6 +262,55 @@ __pattern_walk2_brick_n(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _
         /*vector=*/::std::true_type{}, /*parallel*/ ::std::true_type{});
 }
 
+//------------------------------------------------------------------------
+// transform_if
+//------------------------------------------------------------------------
+
+template <typename _Name>
+struct __walk2_transform_if_wrapper
+{
+};
+
+template <typename _ExecutionPolicy, typename _ForwardIterator1, typename _ForwardIterator2, typename _Function>
+oneapi::dpl::__internal::__enable_if_hetero_execution_policy<_ExecutionPolicy, _ForwardIterator2>
+__pattern_walk2_transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                             _ForwardIterator2 __first2, _Function __func,
+                             /*vector=*/::std::true_type,
+                             /*parallel=*/::std::true_type)
+{
+    // Require `read_write` access mode for output sequence to force a copy in for host iterators to capture incoming
+    // values of the output sequence for elements where the predicate is false.
+    return __pattern_walk2</*_IsSync=*/::std::true_type, __par_backend_hetero::access_mode::read,
+                           __par_backend_hetero::access_mode::read_write>(
+        __par_backend_hetero::make_wrapped_policy<__walk2_transform_if_wrapper>(
+            ::std::forward<_ExecutionPolicy>(__exec)),
+        __first1, __last1, __first2, __func,
+        /*vector=*/::std::true_type{}, /*parallel*/ ::std::true_type{});
+}
+
+template <typename _Name>
+struct __walk3_transform_if_wrapper
+{
+};
+
+template <typename _ExecutionPolicy, typename _ForwardIterator1, typename _ForwardIterator2, typename _ForwardIterator3,
+          typename _Function>
+oneapi::dpl::__internal::__enable_if_hetero_execution_policy<_ExecutionPolicy, _ForwardIterator3>
+__pattern_walk3_transform_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                             _ForwardIterator2 __first2, _ForwardIterator3 __first3, _Function __func,
+                             /*vector=*/::std::true_type,
+                             /*parallel=*/::std::true_type)
+{
+    // Require `read_write` access mode for output sequence to force a copy in for host iterators to capture incoming
+    // values of the output sequence for elements where the predicate is false.
+    return __pattern_walk3<__par_backend_hetero::access_mode::read, __par_backend_hetero::access_mode::read,
+                           __par_backend_hetero::access_mode::read_write>(
+        __par_backend_hetero::make_wrapped_policy<__walk3_transform_if_wrapper>(
+            ::std::forward<_ExecutionPolicy>(__exec)),
+        __first1, __last1, __first2, __first3, __func,
+        /*vector=*/::std::true_type{}, /*parallel*/ ::std::true_type{});
+}
+
 //------------------------------------------------------------------------
 // fill
 //------------------------------------------------------------------------

diff --git a/include/oneapi/dpl/pstl/utils.h b/include/oneapi/dpl/pstl/utils.h
@@ -289,6 +289,48 @@ class __transform_functor
     }
 };
 
+template <typename _UnaryOper, typename _UnaryPred>
+class __transform_if_unary_functor
+{
+    mutable _UnaryOper _M_oper;
+    mutable _UnaryPred _M_pred;
+
+  public:
+    explicit __transform_if_unary_functor(_UnaryOper&& __op, _UnaryPred&& __pred)
+        : _M_oper(::std::forward<_UnaryOper>(__op)), _M_pred(::std::forward<_UnaryPred>(__pred))
+    {
+    }
+
+    template <typename _Input1Type, typename _OutputType>
+    void
+    operator()(const _Input1Type& x, _OutputType& y) const
+    {
+        if (_M_pred(x))
+            y = _M_oper(x);
+    }
+};
+
+template <typename _BinaryOper, typename _BinaryPred>
+class __transform_if_binary_functor
+{
+    mutable _BinaryOper _M_oper;
+    mutable _BinaryPred _M_pred;
+
+  public:
+    explicit __transform_if_binary_functor(_BinaryOper&& __op, _BinaryPred&& __pred)
+        : _M_oper(::std::forward<_BinaryOper>(__op)), _M_pred(::std::forward<_BinaryPred>(__pred))
+    {
+    }
+
+    template <typename _Input1Type, typename _Input2Type, typename _OutputType>
+    void
+    operator()(const _Input1Type& x, const _Input2Type& y, _OutputType& z) const
+    {
+        if (_M_pred(x, y))
+            z = _M_oper(x, y);
+    }
+};
+
 template <typename _Tp, typename _Pred>
 class __replace_functor
 {

diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
@@ -15,6 +15,19 @@
 set (ranlux_24_48_test.pass_timeout_debug "900") # 15min
 set (ranlux_24_48_test.pass_timeout_release "720") # 12min
 
+# If TBB headers are available, libstdc++ enables TBB backend; some TBB symbols become undefined unless it is linked.
+# The undefined symbol error arises only with GCC compiler.
+# There is no known way to limit the workaround to libstdc++ only.
+if (CMAKE_CXX_COMPILER_ID STREQUAL GNU AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 11)
+    if (ONEDPL_BACKEND MATCHES "^(serial|omp|dpcpp_only)$")
+        find_package(TBB 2021 QUIET COMPONENTS tbb OPTIONAL_COMPONENTS tbbmalloc)
+        if (TBB_FOUND)
+            message(STATUS "Tests are linked against TBB to avoid undefined symbol errors due to TBB usage in libstdc++")
+            set(_gcc_link_tbb TRUE)
+        endif()
+    endif()
+endif()
+
 # disable fast math for ALL TESTS
 # icpx and dpcpp accept -fno-fast-math, but icx-cl and dpcpp-cl only accept /clang:-fno-fast-math
 foreach(_fno_fast_math_flag -fno-fast-math /clang:-fno-fast-math)
@@ -91,6 +104,9 @@ macro(onedpl_add_test test_source_file switch_off_checked_iterators)
 
     target_include_directories(${_test_name} PRIVATE "${CMAKE_CURRENT_LIST_DIR}")
     target_link_libraries(${_test_name} PRIVATE oneDPL)
+    if (_gcc_link_tbb)
+        target_link_libraries(${_test_name} PRIVATE TBB::tbb)
+    endif()
     set_target_properties(${_test_name} PROPERTIES CXX_EXTENSIONS NO)
 
     add_test(NAME ${_test_name} COMMAND ${CMAKE_CURRENT_BINARY_DIR}/${_test_name})