Intel® oneAPI DPC++ Library (oneDPL) Release Notes

Overview

The Intel® oneAPI DPC++ Library (oneDPL) accompanies the Intel® oneAPI DPC++/C++ Compiler and provides high-productivity APIs aimed to minimize programming efforts of C++ developers creating efficient heterogeneous applications.

New in 2022.12.0

Changes to CMake

• Removed the ONEDPL_DEVICE_TYPE and ONEDPL_DEVICE_BACKEND CMake options. Use ONEAPI_DEVICE_SELECTOR environment variable or compiler options for device selection instead.
• Removed the ONEDPL_USE_AOT_COMPILATION, ONEDPL_AOT_ARCH, and ONEDPL_FPGA_STATIC_REPORT CMake options. Use the appropriate compiler flags to control ahead-of-time compilation.

New Features

• exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment, and histogram algorithms are added to the <oneapi/dpl/numeric> header, and this header is now recommended to use for these algorithms.
• Improved performance of sort, stable_sort, sort_by_key, and stable_sort_by_key when using Radix sort [1] and device policies.
• Moved philox_engine to the oneapi::dpl namespace and fixed incorrect results for instantiations with non-predefined word size values and std::uint_fast64_t.
• Added experimental radix_sort and radix_sort_by_key algorithms in the oneapi::dpl::experimental::kt::gpu namespace. The implementation has been verified on Intel® Arc B580 Graphics and Intel® Data Center GPU Max Series.
• The experimental::ranges::zip_view range adaptor can now be used with the oneDPL parallel range algorithms and C++20 random access ranges.
• Improved performance of multiple (30+) algorithms with par and par_unseq execution policies and data sizes from 50K to 4M elements when built with the OpenMP backend and Intel® oneAPI DPC++/C++ Compiler.

Deprecation Notice

Support for FPGA devices in oneDPL algorithms, the corresponding execution policies fpga_policy and dpcpp_fpga, the make_fpga_policy function, and the enabling macros ONEDPL_FPGA_DEVICE and ONEDPL_FPGA_EMULATOR are deprecated and will be removed in a future release.

Fixed Issues

• Fixed validation of minimal requirements for range-based algorithms. They require clang 16 and newer instead of the corresponding libc++ versions.
• Fixed ranges::unique_copy to allow output ranges of any size.
• Fixed the default template argument for the new value type in ranges::replace and ranges::replace_if to not use projections.
• Removed excessive data copying when using device policies and host allocated data with several algorithms: fill, generate, transform_if, binary_search, lower_bound, upper_bound, histogram, unique_copy, uninitialized_copy, uninitialized_move, uninitialized_fill, uninitialized_value_construct, uninitialized_default_construct, and destroy.

Known Issues and Limitations

New in This Release

• ranges::unique_copy with the output size smaller than the input size may lose performance on GPU devices.
• kt::gpu::radix_sort_by_key function may produce incorrect results on RHEL 10 or earlier when run on Intel® Data Center GPU Max Series with SYCL buffer passed as input data and no optimization flags passed to the device compiler.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• ranges::copy_if with the output size smaller than the input size may lose performance on GPU devices.
• set_union, set_intersection, set_difference, set_symmetric_difference range algorithms require the output range to have sufficient size to hold all resulting elements.
• histogram algorithm requires the output value type to be an integral type no larger than four bytes when used with a device policy on hardware that does not support 64-bit atomic operations.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.

New in 2022.11.0

New Features

• Added tools for easier customization of policies and backends for experimental dynamic selection feature, see documentation for more details.
• Simplified public submission API for experimental dynamic selection feature, removing oneapi::dpl::experimental::select and adding oneapi::dpl::experimental::try_submit.
• Enabled list-initialization for algorithms. This aligns the API with C++26 but is supported for C++17 and above.

Fixed Issues

• Fixed ranges::copy_if to allow output ranges of any size.
• Fixed a compilation error that occurs with device policies when calling oneapi::dpl::reduce multiple times with differing parameter types.
• Removed the requirement that ranges passed to range-based algorithms support the subscipt operator. This did not comply with the requirement of the C++ standard.

Known Issues and Limitations

New in This Release

• ranges::copy_if with the output size smaller than the input size may lose performance on GPU devices.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• unique_copy, set_union, set_intersection, set_difference, set_symmetric_difference range algorithms require the output range to have sufficient size to hold all resulting elements.
• histogram algorithm requires the output value type to be an integral type no larger than four bytes when used with a device policy on hardware that does not support 64-bit atomic operations.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.
• Incorrect results may be produced by exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment with unseq or par_unseq policy when compiled by Intel® oneAPI DPC++/C++ Compiler 2024.1 or earlier with -fiopenmp, -fiopenmp-simd, -qopenmp, -qopenmp-simd options on Linux. To avoid the issue, pass -fopenmp or -fopenmp-simd option instead.

New in 2022.10.0

Deprecation Notices

The ONEDPL_USE_AOT_COMPILATION and ONEDPL_AOT_ARCH CMake options are deprecated and will be removed in a future release. Please use the relevant compiler flags to enable this feature.

New Features

• Added parallel range algorithms in namespace oneapi::dpl::ranges: set_intersection, set_union, set_difference, set_symmetric_difference, includes, unique, unique_copy, destroy, uninitialized_fill, uninitialized_move, uninitialized_copy, uninitialized_value_construct, uninitialized_default_construct, reverse, reverse_copy, swap_ranges. These algorithms operate with C++20 random access ranges.
• Improved performance of gpu::inclusive_scan kernel template and added support for binary operator and type combinations which do not have a SYCL known identity.
• Improved performance of inclusive_scan_by_segment, exclusive_scan_by_segment, set_union, set_difference, set_intersection, and set_symmetric_difference when using device policies.
• Improved performance of search operations (e.g., find, all_of, equal, search, etc.), is_heap and is_heap_until algorithms on Intel® Arc™ B-series GPU devices.

Fixed Issues

• Removed requirement of GPU double precision support to use set_union, set_difference, set_intersection, and set_symmetric_difference on Windows operating systems.
• Removed default-constructible requirements from the value type for reduce and transform_reduce algorithms, as well as copy-constructible requirements when these algorithms are used with a native ("host") policy.
• Fixed an issue with ranges::merge when projections of the two input ranges were not the same.
• Fixed equal returning a false for empty input sequences; now it returns true.
• Fixed a compilation error SYCL kernel cannot use exceptions occurring with libstdc++ version 10 when calling adjacent_find, is_sorted and is_sorted_until range algorithms with device policies.
• Fixed an issue with PSTL_USE_NONTEMPORAL_STORES macro having no effect.
• Fixed a bug where unique called with a device policy returned an incorrect result iterator.
• Fixed a bug in exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, and inclusive_scan_by_segment algorithms when using device policies with different input and output value types.
• Fixed a bug in return value types of minmax_element and mismatch range algorithms.
• Fixed compile errors in set_union and set_symmetric_difference when using device policies with different second-input and output value types.

Known Issues and Limitations

New in This Release

• copy_if, unique_copy, set_union, set_intersection, set_difference, set_symmetric_difference range algorithms require the output range to have sufficient size to hold all resulting elements.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• histogram algorithm requires the output value type to be an integral type no larger than four bytes when used with a device policy on hardware that does not support 64-bit atomic operations.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.
• Incorrect results may be produced by exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment with unseq or par_unseq policy when compiled by Intel® oneAPI DPC++/C++ Compiler 2024.1 or earlier with -fiopenmp, -fiopenmp-simd, -qopenmp, -qopenmp-simd options on Linux. To avoid the issue, pass -fopenmp or -fopenmp-simd option instead.

New in 2022.9.0

New Features

• Added parallel range algorithms in namespace oneapi::dpl::ranges: fill, move, replace, replace_if, remove, remove_if, mismatch, minmax_element, min, max, find_first_of, find_end, is_sorted_until. These algorithms operate with C++20 random access ranges.
• Improved performance of set operation algorithms when using device policies: set_union, set_difference, set_intersection, set_symmetric_difference.
• Improved performance of copy, fill, for_each, replace, reverse, rotate, transform and 30+ other algorithms with device policies on GPUs when using std::reverse_iterator.
• Added ADL-based customization point is_onedpl_indirectly_device_accessible, which can be used to mark iterator types as indirectly device accessible. Added public trait oneapi::dpl::is_directly_device_accessible[_v] to query if types are indirectly device accessible.

Fixed Issues

• Eliminated runtime exceptions encountered when compiling code that called inclusive_scan, copy_if, partition, unique, reduce_by_segment, and related algorithms with device policies using the open source oneAPI DPC++ Compiler without specifying an optimization flag.
• Fixed a compilation error in reduce_by_segment regarding return type deduction when called with a device policy.
• Eliminated multiple compile time warnings throughout the library.

Known Issues and Limitations

New in This Release

• The set_intersection, set_difference, set_symmetric_difference, and set_union algorithms with a device policy require GPUs with double-precision support on Windows, regardless of the value type of the input sequences.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• histogram algorithm requires the output value type to be an integral type no larger than four bytes when used with a device policy on hardware that does not support 64-bit atomic operations.
• histogram may provide incorrect results with device policies in a program built with -O0 option and the driver version is 2448.13 or older.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.
• Incorrect results may be produced by exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment with unseq or par_unseq policy when compiled by Intel® oneAPI DPC++/C++ Compiler 2024.1 or earlier with -fiopenmp, -fiopenmp-simd, -qopenmp, -qopenmp-simd options on Linux. To avoid the issue, pass -fopenmp or -fopenmp-simd option instead.
• With libstdc++ version 10, the compilation error SYCL kernel cannot use exceptions occurs when calling the range-based adjacent_find, is_sorted or is_sorted_until algorithms with device policies.

New in 2022.8.0

New Features

• Added support of host policies for histogram algorithms.
• Added support for an undersized output range in the range-based merge algorithm.
• Improved performance of the merge and sorting algorithms (sort, stable_sort, sort_by_key, stable_sort_by_key) that rely on Merge sort [1], with device policies for large data sizes.
• Improved performance of copy, fill, for_each, replace, reverse, rotate, transform and 30+ other algorithms with device policies on GPUs.
• Improved oneDPL use with SYCL implementations other than Intel® oneAPI DPC++/C++ Compiler.

Fixed Issues

• Fixed an issue with drop_view in the experimental range-based API.

• Fixed compilation errors in find_if and find_if_not with device policies where the user provided predicate is device copyable but not trivially copyable.

• Fixed incorrect results or synchronous SYCL exceptions for several algorithms when compiled with -O0 and executed on a GPU device.

• Fixed an issue preventing inclusion of the <numeric> header after <execution> and <algorithm> headers.

• Fixed several issues in the sort, stable_sort, sort_by_key and stable_sort_by_key algorithms that:

  • Allows the use of non-trivially-copyable comparators.
  • Eliminates duplicate kernel names.
  • Resolves incorrect results on devices with sub-group sizes smaller than four.
  • Resolved synchronization errors that were seen on Intel® Arc™ B-series GPU devices.

Known Issues and Limitations

New in This Release

• Incorrect results may be observed when calling sort with a device policy on Intel® Arc™ graphics 140V with data sizes of 4-8 million elements on Windows. This issue is resolved in Intel® oneAPI DPC++/C++ Compiler 2025.1 or later and Intel® Graphics Driver 32.0.101.6647 or later.
• sort, stable_sort, sort_by_key and stable_sort_by_key algorithms fail to compile when using Clang 17 and earlier versions, as well as compilers based on these versions, such as Intel® oneAPI DPC++/C++ Compiler 2023.2.0.
• When compiling code that uses device policies with the open source oneAPI DPC++ Compiler (clang++ driver), synchronous SYCL runtime exceptions regarding unfound kernels may be encountered unless an optimization flag is specified (for example -O1) as opposed to relying on the compiler's default optimization level.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• histogram algorithm requires the output value type to be an integral type no larger than four bytes when used with an FPGA policy.
• histogram may provide incorrect results with device policies in a program built with -O0 option.
• Compilation issues may be encountered when passing zip iterators to exclusive_scan_by_segment on Windows.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.
• Incorrect results may be produced by exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment with unseq or par_unseq policy when compiled by Intel® oneAPI DPC++/C++ Compiler with -fiopenmp, -fiopenmp-simd, -qopenmp, -qopenmp-simd options on Linux. To avoid the issue, pass -fopenmp or -fopenmp-simd option instead.

New in 2022.7.0

New Features

• Improved performance of the adjacent_find, all_of, any_of, copy_if, exclusive_scan, equal, find, find_if, find_end, find_first_of, find_if_not, inclusive_scan, includes, is_heap, is_heap_until, is_partitioned, is_sorted, is_sorted_until, lexicographical_compare, max_element, min_element, minmax_element, mismatch, none_of, partition, partition_copy, reduce, remove, remove_copy, remove_copy_if, remove_if, search, search_n, stable_partition, transform_exclusive_scan, transform_inclusive_scan, unique, and unique_copy algorithms with device policies.
• Improved performance of sort, stable_sort and sort_by_key algorithms with device policies when using Merge sort [1].
• Added stable_sort_by_key algorithm in namespace oneapi::dpl.
• Added parallel range algorithms in namespace oneapi::dpl::ranges: all_of, any_of, none_of, for_each, find, find_if, find_if_not, adjacent_find, search, search_n, transform, sort, stable_sort, is_sorted, merge, count, count_if, equal, copy, copy_if, min_element, max_element. These algorithms operate with C++20 random access ranges and views while also taking an execution policy similarly to other oneDPL algorithms.
• Added support for operators ==, !=, << and >> for RNG engines and distributions.
• Added experimental support for the Philox RNG engine in namespace oneapi::dpl::experimental.
• Added the <oneapi/dpl/version> header containing oneDPL version macros and new feature testing macros.

Fixed Issues

• Fixed unused variable and unused type warnings.
• Fixed memory leaks when using sort and stable_sort algorithms with the oneTBB backend.
• Fixed a build error for oneapi::dpl::begin and oneapi::dpl::end functions used with the Microsoft* Visual C++ standard library and with C++20.
• Reordered template parameters of the histogram algorithm to match its function parameter order. For affected histogram calls we recommend to remove explicit specification of template parameters and instead add explicit type conversions of the function arguments as necessary.
• gpu::esimd::radix_sort and gpu::esimd::radix_sort_by_key kernel templates now throw std::bad_alloc if they fail to allocate global memory.
• Fixed a potential hang occurring with gpu::esimd::radix_sort and gpu::esimd::radix_sort_by_key kernel templates.
• Fixed documentation for sort_by_key algorithm, which used to be mistakenly described as stable, despite being possibly unstable for some execution policies. If stability is required, use stable_sort_by_key instead.
• Fixed an error when calling sort with device execution policies on CUDA devices.
• Allow passing C++20 random access iterators to oneDPL algorithms.
• Fixed issues caused by initialization of SYCL queues in the predefined device execution policies. These policies have been updated to be immutable (const) objects.

Known Issues and Limitations

New in This Release

• histogram may provide incorrect results with device policies in a program built with -O0 option.
• Inclusion of <oneapi/dpl/dynamic_selection> prior to <oneapi/dpl/random> may result in compilation errors. Include <oneapi/dpl/random> first as a workaround.
• Incorrect results may occur when using oneapi::dpl::experimental::philox_engine with no predefined template parameters and with word_size values other than 64 and 32.
• Incorrect results or a synchronous SYCL exception may be observed with the following algorithms built with -O0 option and executed on a GPU device: exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, copy_if, remove, remove_copy, remove_copy_if, remove_if, partition, partition_copy, stable_partition, unique, unique_copy, and sort.
• The value type of the input sequence should be convertible to the type of the initial element for the following algorithms with device execution policies: transform_inclusive_scan, transform_exclusive_scan, inclusive_scan, and exclusive_scan.
• The following algorithms with device execution policies may exceed the C++ standard requirements on the number of applications of user-provided predicates or equality operators: copy_if, remove, remove_copy, remove_copy_if, remove_if, partition_copy, unique, and unique_copy. In all cases, the predicate or equality operator is applied O(n) times.
• The adjacent_find, all_of, any_of, equal, find, find_if, find_end, find_first_of, find_if_not, includes, is_heap, is_heap_until, is_sorted, is_sorted_until, mismatch, none_of, search, and search_n algorithms may cause a segmentation fault when used with a device execution policy on a CPU device, and built on Linux with Intel® oneAPI DPC++/C++ Compiler 2025.0.0 and -O0 -g compiler options.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• histogram algorithm requires the output value type to be an integral type no larger than 4 bytes when used with an FPGA policy.
• Compilation issues may be encountered when passing zip iterators to exclusive_scan_by_segment on Windows.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.
• sort, stable_sort, sort_by_key, stable_sort_by_key, partial_sort_copy algorithms may work incorrectly or cause a segmentation fault when used a device execution policy on a CPU device, and built on Linux with Intel® oneAPI DPC++/C++ Compiler and -O0 -g compiler options. To avoid the issue, pass -fsycl-device-code-split=per_kernel option to the compiler.
• Incorrect results may be produced by exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment with unseq or par_unseq policy when compiled by Intel® oneAPI DPC++/C++ Compiler with -fiopenmp, -fiopenmp-simd, -qopenmp, -qopenmp-simd options on Linux. To avoid the issue, pass -fopenmp or -fopenmp-simd option instead.
• Incorrect results may be produced by reduce, reduce_by_segment, and transform_reduce with 64-bit data types when compiled by Intel® oneAPI DPC++/C++ Compiler versions 2021.3 and newer and executed on a GPU device. For a workaround, define the ONEDPL_WORKAROUND_FOR_IGPU_64BIT_REDUCTION macro to 1 before including oneDPL header files.
• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.
• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2022.6.0

News

• oneAPI DPC++ Library Manual Migration Guide to simplify the migration of Thrust* and CUB* APIs from CUDA*.
• radix_sort and radix_sort_by_key kernel templates were moved into oneapi::dpl::experimental::kt::gpu::esimd namespace. The former oneapi::dpl::experimental::kt::esimd namespace is deprecated and will be removed in a future release.
• The for_loop, for_loop_strided, for_loop_n, for_loop_n_strided algorithms in namespace oneapi::dpl::experimental are enforced to fail with device execution policies.

New Features

• Added experimental inclusive_scan kernel template algorithm residing in the oneapi::dpl::experimental::kt::gpu namespace.
• radix_sort and radix_sort_by_key kernel templates are extended with overloads for out-of-place sorting. These overloads preserve the input sequence and sort data into the user provided output sequence.
• Improved performance of the reduce, min_element, max_element, minmax_element, is_partitioned, lexicographical_compare, binary_search, lower_bound, and upper_bound algorithms with device policies.
• sort, stable_sort, sort_by_key algorithms now use Radix sort [1] for sorting sycl::half elements compared with std::less or std::greater.

Fixed Issues

• Fixed compilation errors when using reduce, min_element, max_element, minmax_element, is_partitioned, and lexicographical_compare with Intel oneAPI DPC++/C++ compiler 2023.0 and earlier.
• Fixed possible data races in the following algorithms used with device execution policies: remove_if, unique, inplace_merge, stable_partition, partial_sort_copy, rotate.
• Fixed excessive copying of data in std::vector allocated with a USM allocator for standard library implementations which have allocator information in the std::vector::iterator type.
• Fixed an issue where checking std::is_default_constructible for transform_iterator with a functor that is not default-constructible could cause a build error or an incorrect result.
• Fixed handling of sycl device copyable for internal and public oneDPL types.
• Fixed handling of std::reverse_iterator as input to oneDPL algorithms using a device policy.
• Fixed set_intersection to always copy from the first input sequence to the output, where previously some calls would copy from the second input sequence.
• Fixed compilation errors when using oneapi::dpl::zip_iterator with the oneTBB backend and C++20.

Known Issues and Limitations

New in This Release

• histogram algorithm requires the output value type to be an integral type no larger than 4 bytes when used with an FPGA policy.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• When compiled with -fsycl-pstl-offload option of Intel oneAPI DPC++/C++ compiler and with 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.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.
• sort, stable_sort, sort_by_key, partial_sort_copy algorithms may work incorrectly or cause a segmentation fault when used a DPC++ execution policy for CPU device, and built on Linux with Intel® oneAPI DPC++/C++ Compiler and -O0 -g compiler options. To avoid the issue, pass -fsycl-device-code-split=per_kernel option to the compiler.
• Incorrect results may be produced by exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment with unseq or par_unseq policy when compiled by Intel® oneAPI DPC++/C++ Compiler with -fiopenmp, -fiopenmp-simd, -qopenmp, -qopenmp-simd options on Linux. To avoid the issue, pass -fopenmp or -fopenmp-simd option instead.
• Incorrect results may be produced by reduce, reduce_by_segment, and transform_reduce with 64-bit data types when compiled by Intel® oneAPI DPC++/C++ Compiler versions 2021.3 and newer and executed on GPU devices. For a workaround, define the ONEDPL_WORKAROUND_FOR_IGPU_64BIT_REDUCTION macro to 1 before including oneDPL header files.
• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.
• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2022.5.0

New Features

• Added new histogram algorithms for generating a histogram from an input sequence into an output sequence representing either equally spaced or user-defined bins. These algorithms are currently only available for device execution policies.
• Supported zip_iterator for transform algorithm.

Fixed Issues

• Fixed handling of permutation_iterator as input to oneDPL algorithms for a variety of source iterator and permutation types which caused issues.
• Fixed zip_iterator to be sycl device copyable for trivially copyable source iterator types.
• Added a workaround for reduction algorithm failures with 64-bit data types. Define the ONEDPL_WORKAROUND_FOR_IGPU_64BIT_REDUCTION macro to 1 before including oneDPL header files.

Known Issues and Limitations

New in This Release

• Crashes or incorrect results may occur when using oneapi::dpl::reverse_iterator or std::reverse_iterator as input to oneDPL algorithms with device execution policies.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• When compiled with -fsycl-pstl-offload option of Intel oneAPI DPC++/C++ compiler and with 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.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.
• sort, stable_sort, sort_by_key, partial_sort_copy algorithms may work incorrectly or cause a segmentation fault when used a DPC++ execution policy for CPU device, and built on Linux with Intel® oneAPI DPC++/C++ Compiler and -O0 -g compiler options. To avoid the issue, pass -fsycl-device-code-split=per_kernel option to the compiler.
• Incorrect results may be produced by exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment with unseq or par_unseq policy when compiled by Intel® oneAPI DPC++/C++ Compiler with -fiopenmp, -fiopenmp-simd, -qopenmp, -qopenmp-simd options on Linux. To avoid the issue, pass -fopenmp or -fopenmp-simd option instead.
• Incorrect results may be produced by reduce, reduce_by_segment, and transform_reduce with 64-bit data types when compiled by Intel® oneAPI DPC++/C++ Compiler versions 2021.3 and newer and executed on GPU devices. For a workaround, define the ONEDPL_WORKAROUND_FOR_IGPU_64BIT_REDUCTION macro to 1 before including oneDPL header files.
• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.
• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2022.4.0

New Features

• Added experimental radix_sort and radix_sort_by_key algorithms residing in the oneapi::dpl::experimental::kt::esimd namespace. These algorithms are first in the family of kernel templates that allow configuring a variety of parameters including the number of elements to process by a work item, and the size of a workgroup. The algorithms only work with Intel® Data Center GPU Max Series.
• Added new transform_if algorithm for applying a transform function conditionally based on a predicate, with overloads provided for one and two input sequences that use correspondingly unary and binary operations and predicates.
• Optimizations used with Intel® oneAPI DPC++/C++ Compiler are expanded to the open source oneAPI DPC++ compiler.

Known Issues and Limitations

New in This Release

• esimd::radix_sort and esimd::radix_sort_by_key kernel templates fail to compile when a program is built with -g, -O0, -O1 compiler options.
• esimd::radix_sort_by_key kernel template produces wrong results with the following combinations of kernel_param and types of keys and values: - sizeof(key_type) + sizeof(val_type) == 12, kernel_param::workgroup_size == 64, and kernel_param::data_per_workitem == 96 - sizeof(key_type) + sizeof(val_type) == 16, kernel_param::workgroup_size == 64, and kernel_param::data_per_workitem == 64

New in 2022.3.0

New Features

• Added an experimental feature to dynamically select an execution context, e.g., a SYCL queue. The feature provides selection functions such as select, submit and submit_and_wait, and several selection policies: fixed_resource_policy, round_robin_policy, dynamic_load_policy, and auto_tune_policy.
• unseq and par_unseq policies now enable vectorization also for Intel oneAPI DPC++/C++ Compiler.
• Added support for passing zip iterators as segment value data in reduce_by_segment, exclusive_scan_by_segment, and inclusive_scan_by_segment.
• Improved performance of the merge, sort, stable_sort, sort_by_key, reduce, min_element, max_element, minmax_element, is_partitioned, and lexicographical_compare algorithms with DPC++ execution policies.

Fixed Issues

• Fixed the reduce_async function to not ignore the provided binary operation.

Known Issues and Limitations

New in This Release

• When compiled with -fsycl-pstl-offload option of Intel oneAPI DPC++/C++ compiler and with 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.
• For transform_exclusive_scan and exclusive_scan to run in-place (that is, with the same data used for both input and destination) and with an execution policy of unseq or par_unseq, it is required that the provided input and destination iterators are equality comparable. Furthermore, the equality comparison of the input and destination iterator must evaluate to true. If these conditions are not met, the result of these algorithm calls is undefined.
• sort, stable_sort, sort_by_key, partial_sort_copy algorithms may work incorrectly or cause a segmentation fault when used a DPC++ execution policy for CPU device, and built on Linux with Intel® oneAPI DPC++/C++ Compiler and -O0 -g compiler options. To avoid the issue, pass -fsycl-device-code-split=per_kernel option to the compiler.
• Incorrect results may be produced by exclusive_scan, inclusive_scan, transform_exclusive_scan, transform_inclusive_scan, exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment with unseq or par_unseq policy when compiled by Intel® oneAPI DPC++/C++ Compiler with -fiopenmp, -fiopenmp-simd, -qopenmp, -qopenmp-simd options on Linux. To avoid the issue, pass -fopenmp or -fopenmp-simd option instead.
• Incorrect results may be produced by reduce, reduce_by_segment, and transform_reduce with 64-bit data types when compiled by Intel® oneAPI DPC++/C++ Compiler versions 2021.3 and newer and executed on GPU devices.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.
• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2022.2.0

New Features

• Added sort_by_key algorithm for key-value sorting.
• Improved performance of the reduce, min_element, max_element, minmax_element, is_partitioned, and lexicographical_compare algorithms with DPC++ execution policies.
• Improved performance of the reduce_by_segment, inclusive_scan_by_segment, and exclusive_scan_by_segment algorithms for binary operators with known identities when using DPC++ execution policies.
• Added value_type to all views in oneapi::dpl::experimental::ranges.
• Extended oneapi::dpl::experimental::ranges::sort to support projections applied to the range elements prior to comparison.

Fixed Issues

• The minimally required CMake version is raised to 3.11 on Linux and 3.20 on Windows.
• Added new CMake package oneDPLIntelLLVMConfig.cmake to resolve issues using CMake 3.20+ on Windows for icx and icx-cl.
• Fixed an error in the sort and stable_sort algorithms when performing a descending sort on signed numeric types with negative values.
• Fixed an error in reduce_by_segment algorithm when a non-commutative predicate is used.
• Fixed an error in sort and stable_sort algorithms for integral types wider than 4 bytes.
• Fixed an error for some compilers where OpenMP or SYCL backend was selected by CMake scripts without full compiler support.

Known Issues and Limitations

New in This Release

• Incorrect results may be produced with in-place scans using unseq and par_unseq policies on CPUs with the Intel® C++ Compiler 2021.8.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.
• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2022.1.1

New Features

• Improved sort algorithm performance for the arithmetic data types with std::less or std::greater comparison operator and DPC++ policy.

Fixes Issues

• Fixed an error that caused segmentation faults in transform_reduce, minmax_element, and related algorithms when ran on CPU devices.
• Fixed a compilation error in transform_reduce, minmax_element, and related algorithms on FPGAs.
• Fixed permutation_iterator to support C-style array as a permutation map.
• Fixed a radix-sort issue with 64-bit signed integer types.

New in 2022.1.0

New Features

• Added generate, generate_n, transform algorithms to Tested Standard C++ API.
• Improved performance of the inclusive_scan, exclusive_scan, reduce and max_element algorithms with DPC++ execution policies.

Fixed Issues

• Added a workaround for the TBB headers not found issue occurring with libstdc++ version 9 when oneTBB headers are not present in the environment. The workaround requires inclusion of the oneDPL headers before the libstdc++ headers.
• When possible, oneDPL CMake scripts now enforce C++17 as the minimally required language version.
• Fixed an error in the exclusive_scan algorithm when the output iterator is equal to the input iterator (in-place scan).

Known Issues and Limitations

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.
• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2022.0.0

New Features

• Added the functionality from <complex> and more APIs from <cmath> and <limits> standard headers to Tested Standard C++ API.
• Improved performance of sort and stable_sort algorithms on GPU devices when using Radix sort [1].

Fixed Issues

• Fixed permutation_iterator to work with C++ lambda functions for index permutation.
• Fixed an error in oneapi::dpl::experimental::ranges::guard_view and oneapi::dpl::experimental::ranges::zip_view when using operator[] with an index exceeding the limits of a 32 bit integer type.
• Fixed errors when data size is 0 in upper_bound, lower_bound and binary_search algorithms.

Changes affecting backward compatibility

• Removed support of C++11 and C++14.

• Changed the size and the layout of the discard_block_engine class template.

  For further details, please refer to 2022.0 Changes.

Known Issues and Limitations

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.
• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2021.7.1

New Features

• Added possibility to construct a zip_iterator out of a std::tuple of iterators.
• Added 9 more serial-based versions of algorithms: is_heap, is_heap_until, make_heap, push_heap, pop_heap, is_sorted, is_sorted_until, partial_sort, partial_sort_copy. Please refer to Tested Standard C++ API.

Fixed Issues

• Added namespace alias dpl = oneapi::dpl into all public headers.
• Fixed error in reduce_by_segment algorithm.
• Fixed wrong results error in algorithms call with permutation iterator.

Known Issues and Limitations

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.
• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2021.7.0

Deprecation Notice

• Deprecated support of C++11 for Parallel API with host execution policies (seq, unseq, par, par_unseq). C++17 is the minimal required version going forward.

Fixed Issues

• Fixed a kernel name definition error in range-based algorithms and reduce_by_segment used with a device_policy object that has no explicit kernel name.

Known Issues and Limitations

New in This Release

• STL algorithm functions (such as std::for_each) used in DPC++ kernels do not compile with the debug version of the Microsoft* Visual C++ standard library.

New in 2021.6.1

Fixed Issues

• Fixed compilation errors with C++20.
• Fixed CL_OUT_OF_RESOURCES issue for Radix sort algorithm executed on CPU devices.
• Fixed crashes in exclusive_scan_by_segment, inclusive_scan_by_segment, reduce_by_segment algorithms applied to device-allocated USM.

Known Issues and Limitations

• No new issues in this release.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.

New in 2021.6

New Features

• Added a new implementation for par and par_unseq execution policies based on OpenMP* 4.5 pragmas. It can be enabled with the ONEDPL_USE_OPENMP_BACKEND macro. For more details, see Macros page in oneDPL Guide.
• Added the range-based version of the reduce_by_segment algorithm and improved performance of the iterator-based reduce_by_segment APIs. Please note that the use of the reduce_by_segment algorithm requires C++17.
• Added the following algorithms (serial versions) to Tested Standard C++ API: for_each_n, copy, copy_backward, copy_if, copy_n, is_permutation, fill, fill_n, move, move_backward.

Changes affecting backward compatibility

• Fixed param_type API of random number distributions to satisfy C++ standard requirements. The new definitions of param_type are not compatible with incorrect definitions in previous library versions. Recompilation is recommended for all codes that might use param_type.

Fixed Issues

• Fixed hangs and errors when oneDPL is used together with oneAPI Math Kernel Library (oneMKL) in Data Parallel C++ (DPC++) programs.
• Fixed possible data races in the following algorithms used with DPC++ execution policies: sort, stable_sort, partial_sort, nth_element.

Known Issues and Limitations

• No new issues in this release.

Existing Issues

See oneDPL Guide for other restrictions and known limitations.

• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.

New in 2021.5

New Features

• Added new random number distributions: exponential_distribution, bernoulli_distribution, geometric_distribution, lognormal_distribution, weibull_distribution, cachy_distribution, extreme_value_distribution.
• Added the following algorithms (serial versions) to Tested Standard C++ API: all_of, any_of, none_of, count, count_if, for_each, find, find_if, find_if_not.
• Improved performance of search and find_end algorithms on GPU devices.

Fixed Issues

• Fixed SYCL* 2020 features deprecation warnings.
• Fixed some corner cases of normal_distribution functionality.
• Fixed a floating point exception occurring on CPU devices when a program uses a lot of oneDPL algorithms and DPC++ kernels.
• Fixed possible hanging and data races of the following algorithms used with DPC++ execution policies: count, count_if, is_partitioned, lexicographical_compare, max_element, min_element, minmax_element, reduce, transform_reduce.

Known Issues and Limitations

New in This Release

• The definition of lambda functions used with parallel algorithms should not depend on preprocessor macros that makes it different for the host and the device. Otherwise, the behavior is undefined.

Existing Issues

• exclusive_scan and transform_exclusive_scan algorithms may provide wrong results with vector execution policies when building a program with GCC 10 and using -O0 option.
• Some algorithms may hang when a program is built with -O0 option, executed on GPU devices and large number of elements is to be processed.
• The use of oneDPL together with the GNU C++ standard library (libstdc++) version 9 or 10 may lead to compilation errors (caused by oneTBB API changes). To overcome these issues, include oneDPL header files before the standard C++ header files, or disable parallel algorithms support in the standard library. For more information, please see Intel® oneAPI Threading Building Blocks (oneTBB) Release Notes.
• The using namespace oneapi; directive in a oneDPL program code may result in compilation errors with some compilers including GCC 7 and earlier. Instead of this directive, explicitly use oneapi::dpl namespace, or create a namespace alias.
• The implementation does not yet provide namespace oneapi::std as defined in the oneDPL Specification.
• The use of the range-based API requires C++17 and the C++ standard libraries coming with GCC 8.1 (or higher) or Clang 7 (or higher).
• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• When used within DPC++ kernels or transferred to/from a device, std::array can only hold objects whose type meets DPC++ requirements for use in kernels and for data transfer, respectively.
• std::array::at member function cannot be used in kernels because it may throw an exception; use std::array::operator[] instead.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• Due to specifics of Microsoft* Visual C++, some standard floating-point math functions (including std::ldexp, std::frexp, std::sqrt(std::complex<float>)) require device support for double precision.
• The oneapi::dpl::experimental::ranges::reverse algorithm is not available with -fno-sycl-unnamed-lambda option.

New in 2021.4

New Features

• Added the range-based versions of the following algorithms: any_of, adjacent_find, copy_if, none_of, remove_copy_if, remove_copy, replace_copy, replace_copy_if, reverse, reverse_copy, rotate_copy, swap_ranges, unique, unique_copy.
• Added new asynchronous algorithms: inclusive_scan_async, exclusive_scan_async, transform_inclusive_scan_async, transform_exclusive_scan_async.
• Added structured binding support for zip_iterator::value_type.

Fixed Issues

• Fixed an issue with asynchronous algorithms returning future<ptr> with unified shared memory (USM).

Known Issues and Limitations

New in This Release

• With Intel® oneAPI DPC++/C++ Compiler, unseq and par_unseq execution policies do not use OpenMP SIMD pragmas due to compilation issues with the -fopenm-simd option, possibly resulting in suboptimal performance.
• The oneapi::dpl::experimental::ranges::reverse algorithm does not compile with -fno-sycl-unnamed-lambda option.

Existing Issues

• exclusive_scan and transform_exclusive_scan algorithms may provide wrong results with vector execution policies when building a program with GCC 10 and using -O0 option.
• Some algorithms may hang when a program is built with -O0 option, executed on GPU devices and large number of elements is to be processed.
• The use of oneDPL together with the GNU C++ standard library (libstdc++) version 9 or 10 may lead to compilation errors (caused by oneTBB API changes). To overcome these issues, include oneDPL header files before the standard C++ header files, or disable parallel algorithms support in the standard library. For more information, please see Intel® oneAPI Threading Building Blocks (oneTBB) Release Notes.
• The using namespace oneapi; directive in a oneDPL program code may result in compilation errors with some compilers including GCC 7 and earlier. Instead of this directive, explicitly use oneapi::dpl namespace, or create a namespace alias.
• The implementation does not yet provide namespace oneapi::std as defined in the oneDPL Specification.
• The use of the range-based API requires C++17 and the C++ standard libraries coming with GCC 8.1 (or higher) or Clang 7 (or higher).
• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• When used within DPC++ kernels or transferred to/from a device, std::array can only hold objects whose type meets DPC++ requirements for use in kernels and for data transfer, respectively.
• std::array::at member function cannot be used in kernels because it may throw an exception; use std::array::operator[] instead.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• Due to specifics of Microsoft* Visual C++, some standard floating-point math functions (including std::ldexp, std::frexp, std::sqrt(std::complex<float>)) require device support for double precision.

New in 2021.3

New Features

• Added the range-based versions of the following algorithms: all_of, any_of, count, count_if, equal, move, remove, remove_if, replace, replace_if.
• Added the following utility ranges (views): generate, fill, rotate.

Changes to Existing Features

• Improved performance of discard_block_engine (including ranlux24, ranlux48, ranlux24_vec, ranlux48_vec predefined engines) and normal_distribution.
• Added two constructors to transform_iterator: the default constructor and a constructor from an iterator without a transformation. transform_iterator constructed these ways uses transformation functor of type passed in template arguments.
• transform_iterator can now work on top of forward iterators.

Fixed Issues

• Fixed execution of swap_ranges algorithm with unseq, par execution policies.
• Fixed an issue causing memory corruption and double freeing in scan-based algorithms compiled with -O0 and -g options and run on CPU devices.
• Fixed incorrect behavior in the exclusive_scan algorithm that occurred when the input and output iterator ranges overlapped.
• Fixed error propagation for async runtime exceptions by consistently calling sycl::event::wait_and_throw internally.
• Fixed the warning: local variable will be copied despite being returned by name [-Wreturn-std-move].

Known Issues and Limitations

• No new issues in this release.

Existing Issues

• exclusive_scan and transform_exclusive_scan algorithms may provide wrong results with vector execution policies when building a program with GCC 10 and using -O0 option.
• Some algorithms may hang when a program is built with -O0 option, executed on GPU devices and large number of elements is to be processed.
• The use of oneDPL together with the GNU C++ standard library (libstdc++) version 9 or 10 may lead to compilation errors (caused by oneTBB API changes). To overcome these issues, include oneDPL header files before the standard C++ header files, or disable parallel algorithms support in the standard library. For more information, please see Intel® oneAPI Threading Building Blocks (oneTBB) Release Notes.
• The using namespace oneapi; directive in a oneDPL program code may result in compilation errors with some compilers including GCC 7 and earlier. Instead of this directive, explicitly use oneapi::dpl namespace, or create a namespace alias.
• The implementation does not yet provide namespace oneapi::std as defined in the oneDPL Specification.
• The use of the range-based API requires C++17 and the C++ standard libraries coming with GCC 8.1 (or higher) or Clang 7 (or higher).
• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• When used within DPC++ kernels or transferred to/from a device, std::array can only hold objects whose type meets DPC++ requirements for use in kernels and for data transfer, respectively.
• std::array::at member function cannot be used in kernels because it may throw an exception; use std::array::operator[] instead.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• Due to specifics of Microsoft* Visual C++, some standard floating-point math functions (including std::ldexp, std::frexp, std::sqrt(std::complex<float>)) require device support for double precision.

New in 2021.2

New Features

• Added support of parallel, vector and DPC++ execution policies for the following algorithms: shift_left, shift_right.
• Added the range-based versions of the following algorithms: sort, stable_sort, merge.
• Added experimental asynchronous algorithms: copy_async, fill_async, for_each_async, reduce_async, sort_async, transform_async, transform_reduce_async. These algorithms are declared in oneapi::dpl::experimental namespace and implemented only for DPC++ policies. In order to make these algorithms available the <oneapi/dpl/async> header should be included. Use of the asynchronous API requires C++11.
• Utility function wait_for_all enables waiting for completion of an arbitrary number of events.
• Added the ONEDPL_USE_PREDEFINED_POLICIES macro, which enables predefined policy objects and make_device_policy, make_fpga_policy functions without arguments. It is turned on by default.

Changes to Existing Features

• Improved performance of the following algorithms: count, count_if, is_partitioned, lexicographical_compare, max_element, min_element, minmax_element, reduce, transform_reduce, and sort, stable_sort when using Radix sort [1].
• Improved performance of the linear_congruential_engine RNG engine (including minstd_rand, minstd_rand0, minstd_rand_vec, minstd_rand0_vec predefined engines).

Fixed Issues

• Fixed runtime errors occurring with find_end, search, search_n algorithms when a program is built with -O0 option and executed on CPU devices.
• Fixed the majority of unused parameter warnings.

Known Issues and Limitations

• exclusive_scan and transform_exclusive_scan algorithms may provide wrong results with vector execution policies when building a program with GCC 10 and using -O0 option.
• Some algorithms may hang when a program is built with -O0 option, executed on GPU devices and large number of elements is to be processed.
• The use of oneDPL together with the GNU C++ standard library (libstdc++) version 9 or 10 may lead to compilation errors (caused by oneTBB API changes). To overcome these issues, include oneDPL header files before the standard C++ header files, or disable parallel algorithms support in the standard library. For more information, please see Intel® oneAPI Threading Building Blocks (oneTBB) Release Notes.
• The using namespace oneapi; directive in a oneDPL program code may result in compilation errors with some compilers including GCC 7 and earlier. Instead of this directive, explicitly use oneapi::dpl namespace, or create a namespace alias.
• The implementation does not yet provide namespace oneapi::std as defined in the oneDPL Specification.
• The use of the range-based API requires C++17 and the C++ standard libraries coming with GCC 8.1 (or higher) or Clang 7 (or higher).
• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• When used within DPC++ kernels or transferred to/from a device, std::array can only hold objects whose type meets DPC++ requirements for use in kernels and for data transfer, respectively.
• std::array::at member function cannot be used in kernels because it may throw an exception; use std::array::operator[] instead.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• Due to specifics of Microsoft* Visual C++, some standard floating-point math functions (including std::ldexp, std::frexp, std::sqrt(std::complex<float>)) require device support for double precision.

New in 2021.1 Gold

Key Features

• This version implements the oneDPL Specification v1.0, including parallel algorithms, DPC++ execution policies, special iterators, and other utilities.
• oneDPL algorithms can work with data in DPC++ buffers as well as in unified shared memory (USM).
• For several algorithms, experimental API that accepts ranges (similar to C++20) is additionally provided.
• A subset of the standard C++ libraries for Microsoft* Visual C++, GCC, and Clang is supported in DPC++ kernels, including <array>, <complex>, <functional>, <tuple>, <type_traits>, <utility> and other standard library API. For the detailed list, please refer to oneDPL Guide.
• Standard C++ random number generators and distributions for use in DPC++ kernels.

Known Issues and Limitations

• The use of oneDPL together with the GNU C++ standard library (libstdc++) version 9 or 10 may lead to compilation errors (caused by oneTBB API changes). To overcome these issues, include oneDPL header files before the standard C++ header files, or disable parallel algorithms support in the standard library. For more information, please see Intel® oneAPI Threading Building Blocks (oneTBB) Release Notes.
• The using namespace oneapi; directive in a oneDPL program code may result in compilation errors with some compilers including GCC 7 and earlier. Instead of this directive, explicitly use oneapi::dpl namespace, or create a namespace alias.
• The partial_sort_copy, sort and stable_sort algorithms are prone to CL_BUILD_PROGRAM_FAILURE when a program uses Radix sort [1], is built with -O0 option and executed on CPU devices.
• The implementation does not yet provide namespace oneapi::std as defined in the oneDPL Specification.
• The use of the range-based API requires C++17 and the C++ standard libraries coming with GCC 8.1 (or higher) or Clang 7 (or higher).
• std::tuple, std::pair cannot be used with SYCL buffers to transfer data between host and device.
• When used within DPC++ kernels or transferred to/from a device, std::array can only hold objects whose type meets DPC++ requirements for use in kernels and for data transfer, respectively.
• std::array::at member function cannot be used in kernels because it may throw an exception; use std::array::operator[] instead.
• std::array cannot be swapped in DPC++ kernels with std::swap function or swap member function in the Microsoft* Visual C++ standard library.
• Due to specifics of Microsoft* Visual C++, some standard floating-point math functions (including std::ldexp, std::frexp, std::sqrt(std::complex<float>)) require device support for double precision.

[1]	(1, 2, 3, 4, 5, 6, 7) The sorting algorithms in oneDPL use Radix sort for arithmetic data types and sycl::half (since oneDPL 2022.6) compared with std::less or std::greater, otherwise Merge sort.