diff --git a/libcxx/CMakeLists.txt b/libcxx/CMakeLists.txt --- a/libcxx/CMakeLists.txt +++ b/libcxx/CMakeLists.txt @@ -806,9 +806,11 @@ config_define(1 _LIBCPP_PSTL_CPU_BACKEND_SERIAL) elseif(LIBCXX_PSTL_CPU_BACKEND STREQUAL "std_thread") config_define(1 _LIBCPP_PSTL_CPU_BACKEND_THREAD) +elseif(LIBCXX_PSTL_CPU_BACKEND STREQUAL "libdispatch") + config_define(1 _LIBCPP_PSTL_CPU_BACKEND_LIBDISPATCH) else() message(FATAL_ERROR "LIBCXX_PSTL_CPU_BACKEND is set to ${LIBCXX_PSTL_CPU_BACKEND}, which is not a valid backend. - Valid backends are: serial, std_thread") + Valid backends are: serial, std_thread and libdispatch") endif() if (LIBCXX_ABI_DEFINES) diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt --- a/libcxx/include/CMakeLists.txt +++ b/libcxx/include/CMakeLists.txt @@ -78,6 +78,7 @@ __algorithm/pstl_backends/cpu_backends/fill.h __algorithm/pstl_backends/cpu_backends/find_if.h __algorithm/pstl_backends/cpu_backends/for_each.h + __algorithm/pstl_backends/cpu_backends/libdispatch.h __algorithm/pstl_backends/cpu_backends/merge.h __algorithm/pstl_backends/cpu_backends/serial.h __algorithm/pstl_backends/cpu_backends/stable_sort.h diff --git a/libcxx/include/__algorithm/pstl_backend.h b/libcxx/include/__algorithm/pstl_backend.h --- a/libcxx/include/__algorithm/pstl_backend.h +++ b/libcxx/include/__algorithm/pstl_backend.h @@ -169,7 +169,8 @@ }; # endif -# if defined(_LIBCPP_PSTL_CPU_BACKEND_SERIAL) || defined(_LIBCPP_PSTL_CPU_BACKEND_THREAD) +# if defined(_LIBCPP_PSTL_CPU_BACKEND_SERIAL) || defined(_LIBCPP_PSTL_CPU_BACKEND_THREAD) || \ + defined(_LIBCPP_PSTL_CPU_BACKEND_LIBDISPATCH) template <> struct __select_backend { using type = __cpu_backend_tag; diff --git a/libcxx/include/__algorithm/pstl_backends/cpu_backends/backend.h b/libcxx/include/__algorithm/pstl_backends/cpu_backends/backend.h --- a/libcxx/include/__algorithm/pstl_backends/cpu_backends/backend.h +++ b/libcxx/include/__algorithm/pstl_backends/cpu_backends/backend.h @@ -16,6 +16,8 @@ # include <__algorithm/pstl_backends/cpu_backends/serial.h> #elif defined(_LIBCPP_PSTL_CPU_BACKEND_THREAD) # include <__algorithm/pstl_backends/cpu_backends/thread.h> +#elif defined(_LIBCPP_PSTL_CPU_BACKEND_LIBDISPATCH) +# include <__algorithm/pstl_backends/cpu_backends/libdispatch.h> #else # error "Invalid CPU backend choice" #endif diff --git a/libcxx/include/__algorithm/pstl_backends/cpu_backends/libdispatch.h b/libcxx/include/__algorithm/pstl_backends/cpu_backends/libdispatch.h new file mode 100644 --- /dev/null +++ b/libcxx/include/__algorithm/pstl_backends/cpu_backends/libdispatch.h @@ -0,0 +1,225 @@ +//===----------------------------------------------------------------------===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// + +#ifndef _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_LIBDISPATCH_H +#define _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_LIBDISPATCH_H + +#include <__algorithm/lower_bound.h> +#include <__algorithm/upper_bound.h> +#include <__atomic/atomic.h> +#include <__config> +#include <__exception/terminate.h> +#include <__iterator/iterator_traits.h> +#include <__memory/construct_at.h> +#include <__memory/uninitialized_buffer.h> +#include <__memory/unique_ptr.h> +#include <__memory_resource/memory_resource.h> +#include <__numeric/transform_reduce.h> +#include <__utility/exception_guard.h> +#include <__utility/move.h> +#include <__utility/terminate_on_exception.h> +#include +#include +#include + +_LIBCPP_BEGIN_NAMESPACE_STD + +namespace __par_backend { +inline namespace __libdispatch { + +// ::dispatch_apply is marked as __attribute__((nothrow)) because it doesn't let exceptions propagate, and neither do +// we. +[[_Clang::__callback__(__func, __context, __)]] _LIBCPP_EXPORTED_FROM_ABI void +__dispatch_apply(size_t __chunk_count, void* __context, void (*__func)(void* __context, size_t __chunk)) noexcept; + +template +_LIBCPP_HIDE_FROM_ABI void __dispatch_apply(size_t __chunk_count, _Func __func) noexcept { + __libdispatch::__dispatch_apply(__chunk_count, &__func, [](void* __context, size_t __chunk) { + (*static_cast<_Func*>(__context))(__chunk); + }); +} + +// Preliminary size of each chunk: requires further discussion +constexpr ptrdiff_t __default_chunk_size = 2048; + +struct __chunk_partitions { + ptrdiff_t __chunk_count_; // includes the first chunk + ptrdiff_t __chunk_size_; + ptrdiff_t __first_chunk_size_; +}; + +[[__gnu__::__const__]] _LIBCPP_EXPORTED_FROM_ABI pmr::memory_resource* __get_memory_resource(); +[[__gnu__::__const__]] _LIBCPP_EXPORTED_FROM_ABI __chunk_partitions __partition_chunks(ptrdiff_t __size); + +template +_LIBCPP_HIDE_FROM_ABI void __parallel_for(_RandomAccessIterator __first, _RandomAccessIterator __last, _Functor __f) { + auto __partitions = __libdispatch::__partition_chunks(__last - __first); + + // Perform the chunked execution. + __libdispatch::__dispatch_apply(__partitions.__chunk_count_, [&](size_t __chunk) { + auto __this_chunk_size = __chunk == 0 ? __partitions.__first_chunk_size_ : __partitions.__chunk_size_; + auto __index = + __chunk == 0 + ? 0 + : (__chunk * __partitions.__chunk_size_) + (__partitions.__first_chunk_size_ - __partitions.__chunk_size_); + __f(__first + __index, __first + __index + __this_chunk_size); + }); +} + +template +_LIBCPP_HIDE_FROM_ABI void __libdispatch_invoke_parallel(_Func1&& __f1, _Func2&& __f2) { + __libdispatch::__dispatch_apply(2, [&](size_t __index) { + if (__index == 0) + __f1(); + else + __f2(); + }); +} + +template +struct __merge_range { + __merge_range(_RandomAccessIterator1 __mid1, _RandomAccessIterator2 __mid2, _RandomAccessIteratorOut __result) + : __mid1_(__mid1), __mid2_(__mid2), __result_(__result) {} + + _RandomAccessIterator1 __mid1_; + _RandomAccessIterator2 __mid2_; + _RandomAccessIteratorOut __result_; +}; + +template +_LIBCPP_HIDE_FROM_ABI void __parallel_merge( + _RandomAccessIterator1 __first1, + _RandomAccessIterator1 __last1, + _RandomAccessIterator2 __first2, + _RandomAccessIterator2 __last2, + _RandomAccessIterator3 __result, + _Compare __comp, + _LeafMerge __leaf_merge) { + __chunk_partitions __partitions = + __libdispatch::__partition_chunks(std::max(__last1 - __first1, __last2 - __first2)); + + if (__partitions.__chunk_count_ == 0) + return; + + if (__partitions.__chunk_count_ == 1) { + __leaf_merge(__first1, __last1, __first2, __last2, __result, __comp); + return; + } + + using __merge_range_t = __merge_range<_RandomAccessIterator1, _RandomAccessIterator2, _RandomAccessIterator3>; + + vector<__merge_range_t> __ranges; + __ranges.reserve(__partitions.__chunk_count_); + + // TODO: Improve the case where the smaller range is merged into just a few (or even one) chunks of the larger case + std::__terminate_on_exception([&] { + auto __orig_first1 = __first1; + auto __orig_first2 = __first2; + + bool __iterate_first_range = __last1 - __first1 > __last2 - __first2; + + auto __compute_chunk = [&__last1, &__last2, __iterate_first_range, &__comp]( + auto& __iter1, auto& __iter2, auto& __out, size_t __increment_count) -> __merge_range_t { + auto [__mid1, __mid2] = [&] { + if (__iterate_first_range) { + auto __m1 = __iter1 + __increment_count; + auto __m2 = std::lower_bound(__iter2, __last2, __m1[-1], __comp); + return std::make_pair(__m1, __m2); + } else { + auto __m2 = __iter2 + __increment_count; + auto __m1 = std::lower_bound(__iter1, __last1, __m2[-1], __comp); + return std::make_pair(__m1, __m2); + } + }(); + __merge_range_t __ret{__mid1, __mid2, __out}; + __out += (__mid1 - __iter1) + (__mid2 - __iter2); + __iter1 = __mid1; + __iter2 = __mid2; + return __ret; + }; + + // handle first chunk + __ranges.emplace_back(__compute_chunk(__first1, __first2, __result, __partitions.__first_chunk_size_)); + + // handle 2 -> N - 1 chunks + for (ptrdiff_t __i = 1; __i != __partitions.__chunk_count_ - 1; ++__i) + __ranges.emplace_back(__compute_chunk(__first1, __first2, __result, __partitions.__chunk_size_)); + + // handle last chunk + __ranges.emplace_back(__last1, __last2, __result); + + __libdispatch::__dispatch_apply(__ranges.size(), [&](size_t __index) { + auto __last_iters = __ranges[__index]; + if (__index == 0) { + __leaf_merge( + __orig_first1, __last_iters.__mid1_, __orig_first2, __last_iters.__mid2_, __last_iters.__result_, __comp); + } else { + auto __first_iters = __ranges[__index - 1]; + __leaf_merge( + __first_iters.__mid1_, + __last_iters.__mid1_, + __first_iters.__mid2_, + __last_iters.__mid2_, + __last_iters.__result_, + __comp); + } + }); + }); +} + +template +_LIBCPP_HIDE_FROM_ABI _Value __parallel_transform_reduce( + _RandomAccessIterator __first, + _RandomAccessIterator __last, + _Transform __transform, + _Value __init, + _Combiner __combiner, + _Reduction __reduction) { + auto __partitions = __libdispatch::__partition_chunks(__first, __last); + auto __values = std::__make_uninitialized_buffer<_Value[]>( + nothrow, __partitions.__chunk_count_, [](_Value* __ptr, size_t __count) { std::destroy_n(__ptr, __count); }); + + if (__values == nullptr) + std::__throw_pstl_bad_alloc(); + + // __dispatch_apply is noexcept + __libdispatch::__dispatch_apply(__partitions.__chunk_count_, [&](size_t __chunk) { + auto __this_chunk_size = __chunk == 0 ? __partitions.__first_chunk_size_ : __partitions.__chunk_size_; + auto __index = + __chunk == 0 + ? 0 + : (__chunk * __partitions.__chunk_size_) + (__partitions.__first_chunk_size_ - __partitions.__chunk_size_); + std::__construct_at( + __values + __index, + __reduction(__first + __index + 2, + __first + __index + __this_chunk_size, + __combiner(__transform((__first + __index)[0], (__first + __index)[1])))); + }); + + return std::transform_reduce(__values, __values + __partitions.__chunk_count_, __init, __transform, __reduction); +} + +// TODO: parallelize this +template +_LIBCPP_HIDE_FROM_ABI void __parallel_stable_sort( + _RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp, _LeafSort __leaf_sort) { + __leaf_sort(__first, __last, __comp); +} + +_LIBCPP_HIDE_FROM_ABI inline void __cancel_execution() {} + +} // namespace __libdispatch +} // namespace __par_backend + +_LIBCPP_END_NAMESPACE_STD + +#endif // _LIBCPP___ALGORITHM_PSTL_BACKENDS_CPU_BACKENDS_LIBDISPATCH_H diff --git a/libcxx/include/__algorithm/pstl_backends/cpu_backends/transform_reduce.h b/libcxx/include/__algorithm/pstl_backends/cpu_backends/transform_reduce.h --- a/libcxx/include/__algorithm/pstl_backends/cpu_backends/transform_reduce.h +++ b/libcxx/include/__algorithm/pstl_backends/cpu_backends/transform_reduce.h @@ -163,8 +163,8 @@ std::move(__last), [__transform](_ForwardIterator __iter) { return __transform(*__iter); }, std::move(__init), - std::move(__reduce), - [=](_ForwardIterator __brick_first, _ForwardIterator __brick_last, _Tp __brick_init) { + __reduce, + [__transform, __reduce](auto __brick_first, auto __brick_last, _Tp __brick_init) { return std::__pstl_transform_reduce<__remove_parallel_policy_t<_ExecutionPolicy>>( __cpu_backend_tag{}, std::move(__brick_first), diff --git a/libcxx/include/__config_site.in b/libcxx/include/__config_site.in --- a/libcxx/include/__config_site.in +++ b/libcxx/include/__config_site.in @@ -34,6 +34,7 @@ // PSTL backends #cmakedefine _LIBCPP_PSTL_CPU_BACKEND_SERIAL #cmakedefine _LIBCPP_PSTL_CPU_BACKEND_THREAD +#cmakedefine _LIBCPP_PSTL_CPU_BACKEND_LIBDISPATCH // __USE_MINGW_ANSI_STDIO gets redefined on MinGW #ifdef __clang__ diff --git a/libcxx/include/__utility/terminate_on_exception.h b/libcxx/include/__utility/terminate_on_exception.h --- a/libcxx/include/__utility/terminate_on_exception.h +++ b/libcxx/include/__utility/terminate_on_exception.h @@ -11,6 +11,7 @@ #include <__config> #include <__exception/terminate.h> +#include #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER) # pragma GCC system_header diff --git a/libcxx/src/CMakeLists.txt b/libcxx/src/CMakeLists.txt --- a/libcxx/src/CMakeLists.txt +++ b/libcxx/src/CMakeLists.txt @@ -313,6 +313,11 @@ experimental/memory_resource.cpp ) +if (LIBCXX_PSTL_CPU_BACKEND STREQUAL "libdispatch") + set(LIBCXX_EXPERIMENTAL_SOURCES ${LIBCXX_EXPERIMENTAL_SOURCES} + pstl/libdispatch.cpp) +endif() + add_library(cxx_experimental STATIC ${LIBCXX_EXPERIMENTAL_SOURCES}) target_link_libraries(cxx_experimental PUBLIC cxx-headers) if (LIBCXX_ENABLE_SHARED) diff --git a/libcxx/src/pstl/libdispatch.cpp b/libcxx/src/pstl/libdispatch.cpp new file mode 100644 --- /dev/null +++ b/libcxx/src/pstl/libdispatch.cpp @@ -0,0 +1,70 @@ +//===----------------------------------------------------------------------===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// + +#include <__algorithm/min.h> +#include <__algorithm/pstl_backends/cpu_backends/libdispatch.h> +#include <__config> +#include +#include +#include + +_LIBCPP_BEGIN_NAMESPACE_STD + +namespace __par_backend::inline __libdispatch { + +pmr::memory_resource* __get_memory_resource() { + static std::pmr::synchronized_pool_resource pool{pmr::new_delete_resource()}; + return &pool; +} + +void __dispatch_apply(size_t chunk_count, void* context, void (*func)(void* context, size_t chunk)) noexcept { + ::dispatch_apply_f(chunk_count, DISPATCH_APPLY_AUTO, context, func); +} + +__chunk_partitions __partition_chunks(ptrdiff_t element_count) { + __chunk_partitions partitions; + partitions.__chunk_count_ = [&] { + ptrdiff_t cores = std::max(1u, thread::hardware_concurrency()); + + auto medium = [&](ptrdiff_t n) { return cores + ((n - cores) / cores); }; + + // This is an approximation of `log(1.01, sqrt(n))` which seemes to be reasonable for `n` larger than 500 and tops + // at 800 tasks for n ~ 8 million + auto large = [](ptrdiff_t n) { return static_cast(100.499 * std::log(std::sqrt(n))); }; + + if (element_count < cores) + return element_count; + else if (element_count < 500) + return medium(element_count); + else + return std::min(medium(element_count), large(element_count)); // provide a "smooth" transition + }(); + partitions.__chunk_size_ = element_count / partitions.__chunk_count_; + partitions.__first_chunk_size_ = partitions.__chunk_size_; + + const ptrdiff_t leftover_item_count = element_count - (partitions.__chunk_count_ * partitions.__chunk_size_); + + if (leftover_item_count == 0) + return partitions; + + if (leftover_item_count == partitions.__chunk_size_) { + partitions.__chunk_count_ += 1; + return partitions; + } + + const ptrdiff_t n_extra_items_per_chunk = leftover_item_count / partitions.__chunk_count_; + const ptrdiff_t n_final_leftover_items = leftover_item_count - (n_extra_items_per_chunk * partitions.__chunk_count_); + + partitions.__chunk_size_ += n_extra_items_per_chunk; + partitions.__first_chunk_size_ = partitions.__chunk_size_ + n_final_leftover_items; + return partitions; +} + +} // namespace __par_backend::inline __libdispatch + +_LIBCPP_END_NAMESPACE_STD diff --git a/libcxx/test/libcxx/algorithms/pstl.libdispatch.chunk_partitions.pass.cpp b/libcxx/test/libcxx/algorithms/pstl.libdispatch.chunk_partitions.pass.cpp new file mode 100644 --- /dev/null +++ b/libcxx/test/libcxx/algorithms/pstl.libdispatch.chunk_partitions.pass.cpp @@ -0,0 +1,26 @@ +//===----------------------------------------------------------------------===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// + +// + +// ADDITIONAL_COMPILE_FLAGS: -Wno-private-header + +// __chunk_partitions __partition_chunks(ptrdiff_t); + +#include <__algorithm/pstl_backends/cpu_backends/libdispatch.h> +#include +#include + +int main(int, char**) { + for (std::ptrdiff_t i = 0; i != 2ll << 20; ++i) { + auto chunks = std::__par_backend::__libdispatch::__partition_chunks(i); + assert(chunks.__chunk_count_ <= i); + assert((chunks.__chunk_count_ - 1) * chunks.__chunk_size_ + chunks.__first_chunk_size_ == i); + } + return 0; +} diff --git a/libcxx/test/std/algorithms/alg.modifying.operations/alg.copy/fuzz.pstl.copy.sh.cpp b/libcxx/test/std/algorithms/alg.modifying.operations/alg.copy/fuzz.pstl.copy.sh.cpp new file mode 100644 --- /dev/null +++ b/libcxx/test/std/algorithms/alg.modifying.operations/alg.copy/fuzz.pstl.copy.sh.cpp @@ -0,0 +1,49 @@ +//===----------------------------------------------------------------------===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// + +// UNSUPPORTED: c++03, c++11, c++14 + +// UNSUPPORTED: libcpp-has-no-incomplete-pstl + +// REQUIRES: clang || apple-clang +// RUN: %{build} -fsanitize=fuzzer -O3 + +// To make sure we run all possible code paths, we use inputs which are up to 1GB large. +// RUN: %{run} -max_total_time=10 -max_len=1073741824 + +// + +// template +// ForwardIterator2 copy(ExecutionPolicy&& policy, +// ForwardIterator1 first, ForwardIterator1 last, +// ForwardIterator2 result); + +#include +#include +#include +#include + +struct NonTrivial { + std::uint8_t val; + + NonTrivial() = default; + NonTrivial(const NonTrivial& v) : val(v.val) {} + NonTrivial& operator=(const NonTrivial& v) { + val = v.val; + return *this; + } + + bool operator==(NonTrivial v) const { return v.val == val; } +}; + +extern "C" int LLVMFuzzerTestOneInput(const std::uint8_t* data, std::size_t size) { + std::vector vec(size); + auto input = reinterpret_cast(data); + std::copy(std::execution::par, input, input + size, vec.begin()); + return !std::equal(input, input + size, vec.data()); +} diff --git a/libcxx/test/std/algorithms/alg.sorting/alg.merge/fuzz.pstl.merge.sh.cpp b/libcxx/test/std/algorithms/alg.sorting/alg.merge/fuzz.pstl.merge.sh.cpp new file mode 100644 --- /dev/null +++ b/libcxx/test/std/algorithms/alg.sorting/alg.merge/fuzz.pstl.merge.sh.cpp @@ -0,0 +1,55 @@ +//===----------------------------------------------------------------------===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// + +// UNSUPPORTED: c++03, c++11, c++14 + +// UNSUPPORTED: libcpp-has-no-incomplete-pstl + +// REQUIRES: clang || apple-clang +// RUN: %{build} -fsanitize=fuzzer -O3 + +// To make sure we run all possible code paths, we use inputs which are up to 1GB large. +// RUN: %{run} -max_total_time=10 -max_len=1073741824 + +// + +// template +// ForwardIterator2 copy(ExecutionPolicy&& policy, +// ForwardIterator1 first, ForwardIterator1 last, +// ForwardIterator2 result); + +#include +#include +#include +#include + +extern "C" int LLVMFuzzerTestOneInput(const std::uint8_t* data, std::size_t size) { + try { + if (size < sizeof(std::size_t)) + return -1; + + std::size_t split_at; + std::memcpy(&split_at, data, sizeof(std::size_t)); + data += sizeof(std::size_t); + size -= sizeof(std::size_t); + + if (split_at > size) + return -1; + + std::vector in(data, data + size); + std::vector out(size); + + std::sort(in.begin(), in.begin() + split_at); + std::sort(in.begin() + split_at, in.end()); + std::merge( + std::execution::par, in.data(), in.data() + split_at, in.data() + split_at, in.data() + size, out.data()); + return !std::is_sorted(out.begin(), out.end()); + } catch (...) { // allocation in std::merge failed, so the input data was too large + return -1; + } +} diff --git a/libcxx/test/std/algorithms/alg.sorting/alg.merge/pstl.merge.pass.cpp b/libcxx/test/std/algorithms/alg.sorting/alg.merge/pstl.merge.pass.cpp --- a/libcxx/test/std/algorithms/alg.sorting/alg.merge/pstl.merge.pass.cpp +++ b/libcxx/test/std/algorithms/alg.sorting/alg.merge/pstl.merge.pass.cpp @@ -49,6 +49,13 @@ assert((out == std::array{1, 2, 3, 4, 5, 6, 7, 8, 9, 10})); } + { // check that it works with both ranges being empty + std::array a; + std::array b; + std::array out; + std::merge( + policy, Iter1(std::begin(a)), Iter1(std::end(a)), Iter2(std::begin(b)), Iter2(std::end(b)), std::begin(out)); + } { // check that it works with the first range being empty std::array a; int b[] = {2, 4, 6, 8, 10}; @@ -96,8 +103,12 @@ } std::vector out(std::size(a) + std::size(b)); - std::merge( - Iter1(a.data()), Iter1(a.data() + a.size()), Iter2(b.data()), Iter2(b.data() + b.size()), std::begin(out)); + std::merge(policy, + Iter1(a.data()), + Iter1(a.data() + a.size()), + Iter2(b.data()), + Iter2(b.data() + b.size()), + std::begin(out)); std::vector expected(200); std::iota(expected.begin(), expected.end(), 0); assert(std::equal(out.begin(), out.end(), expected.begin()));