sort stable_sort partial_sort nth_element is_sorted reduce transform_reduce inclusive_scan exclusive_scan Some of the algorithmsCppCon 2023 33 David Olsen - Back to Basics: Concurrency C++ PARALLEL sort stable_sort partial_sort nth_element is_sorted reduce transform_reduce inclusive_scan exclusive_scan Some of the algorithmsCppCon 2023 34 David Olsen - Back to Basics: Concurrency C++ PARALLEL sort stable_sort partial_sort nth_element is_sorted reduce transform_reduce inclusive_scan exclusive_scan Some of the algorithmsCppCon 2023 35 David Olsen - Back to Basics: Concurrency C++ PARALLEL

{ layout_vertices::mapping> scan{grid_.extents(), endpoints, p, theta}; return {data_.data(), scan}; } 69github.com/griswaldbrooks/spanny2 layout_vertices std::mdspan> scan{grid_.extents(), endpoints, p, theta}; return {data_.data(), scan}; } creates std::mdspan with one dimensional extent 70github. :dextents> scan{grid_.extents(), endpoints, p, theta}; return {data_.data(), scan}; } used by the mapping template parameter to create the laser scan from endpoints 71github.com

p1689 20 https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2022/p1689r5.htmlPer target scanning scan: done by compiler (i.e. cl -scanDependencies) collate: cmake -E cmake_ninja_dyndep 21Where we are Clang/ninja 32 $hoffman@caprica b % ninja -v [1/10] "/Users/hoffman/Work/llvm/llvm-inst/bin/clang-scan-deps" -format=p1689 -- /Users/hoffman/Work/llvm/llvm-inst/bin/clang++ -std=gnu++20 -arch arm64 -isysroot o.ddi.tmp CMakeFiles/hello.dir/main.cxx.o.ddi [2/10] "/Users/hoffman/Work/llvm/llvm-inst/bin/clang-scan-deps" -format=p1689 -- /Users/hoffman/Work/llvm/llvm-inst/bin/clang++ -std=gnu++20 -arch arm64 -isysroot

OpenMP 与 Intel TBB 7.被忽视的访存优化：内存带宽与 cpu 缓存机制 8.GPU 专题： wrap 调度，共享内存， barrier 9.并行算法实战： reduce ， scan ，矩阵乘法等 10.存储大规模三维数据的关键：稀疏数据结构 11.物理仿真实战：邻居搜索表实现 pbf 流体求解 12.C++ 在 ZENO 中的工程实践：从 primitive 说起 13 。 封装好了： parallel_reduce 保证每次运行结果一致： parallel_deterministic_reduce 并行缩并的额外好处：能避免浮点误差，例如求平均值 扫描（ scan ） 如图所示，扫描和缩并差不多，只不过他会把求和的中间结果存到数组里去 1 个线程，依次处理 8 个元素的扫描，花了 7 秒 用电量： 1*7=7 度电 总用时： 1*7=7 秒 结论：串行扫描的时间复杂度为 O(nlogn) 。 可见，并行后虽然降低了时间复杂度，但是以提升工作复杂度为代价！ 更多细节，敬请期待 GPU 专题，我们会以 CUDA 为例详细探讨两全方案。 封装好了： parallel_scan 第 3 章：性能测试 案例： map 与 reduce 的组合 测试所花费时间： tbb::tick_count::now() 并行和串行的速度比较 如何评价 • 公式：加速比 = 串行用时

JULICH Forschungszentrum 7/ Improving std::execution example Stdexec: :SendeT auto async_inclusiVe_scan (stdexec: :Scheduler auto scheduler， std: :Span input_range， std: :Span< double> auto output_chunk = ， input_chunks [chunk_index] ; = output_chunks [chunk_index] ; std: :inclusive_scan(input_chunk.begin() ， input_chunk .end () ，output_chunk.begin()); partials[chunk_index + 1] = output_chunk output_chunk.back(); ]) | stdexec: :then([] (std: :vector&gw partials) { std: :inclusive_scan(partials.begin() ， partials. end ()，Ppartials.begin()); return std: :move(partials) ; J) | stdexec:

including open-source •Incorporate fuzzing in some CI runs •Scan code for references to vulnerable packages •Produce scripts to scan company code for matches against known open- source codeProblem Establish incremental milestones with success at each step Step Action Benefit Cost 1 Develop tool to scan company repos/code for open-source Identify legal & security risks X months 2 Choose a project

1COMPLETENESS VIOLATIONS (3) COMPLETENESS VIOLATIONS (3) transform_{in,ex}clusive_scan transform_{in,ex}clusive_scan can't zip two ranges? can't zip two ranges? transform transform can, can, transform_reduce class T, class BinaryOperation, class UnaryOperation> OutputIterator transform_exclusive_scan( InputIterator first, InputIterator last, OutputIterator result, T init, BinaryOperation

nextplatform.com/2021/02/03/can-sycl-slice-into-broader-supercomputing/ https://www.phoronix.com/scan.php?page=news_item&px=hipSYCL-New-Lite-Runtime https://software.intel.com/content/www/us/en/devel shift_group_right() ○ permute_group_by_xor() ○ select_from_group() ○ reduce() / inclusive_scan() / exclusive_scan() template < typename Group, typename Ptr, typename BinaryOperation> std::iterator

drop_while filter flatten map mask pairwise pairwise_map prescan read_only reverse scan scan_first set_difference set_intersection set_symmetric_difference set_union slide stride split

write( DataTag tag, const void* data); error_code read(DataTag tag, void* data) const; error_code scan(size_t offset); //... }; Settings.cpp FileSystem.hppStub out what you need #include "FileSystem write( DataTag tag, const void* data); error_code read(DataTag tag, void* data) const; error_code scan(size_t offset); //... }; Settings.cpp FileSystem.hpp #include