} } ## I ntel Intrinsics Guide • _mm 系列指令出自 头文件。 · 指令的文档可以看这个网站： • https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html - 里面有详细说明每个指令对应的汇编，方便理解的伪代码，延迟和花费的时钟周期等。 "std: " << is_aligned << std::endl; } } ## tbbmalloc 的主要好处：能保证 64 字节对齐 • tbb::cache_aligned_ allocator 的最大好处在于他分配的内存地址，永远会对齐到缓存行（64 字节），对 SIMD 而言可以用 _mm_load_ps 而不是 _mm_loadu_ps 5; i++) { std::vectortbb::cache_aligned_allocator> arr(n); bool is_aligned = (uintptr_t)arr.data() % 64 == 0; std::cout << "tbb: " << is_aligned <<

spall https://handmade.network/p/333/spall/ geiger https://github.com/david-grs/geiger Intel IACA https://www.intel.com/content/www/us/en/developer/articles/tool/architecture-code-analyzer.html ## The Obligatory • browsable log! • filterable log! • no thread sync needed! ☑ tbb worker 56 ☑ tbb worker 55 ☑ tbb worker 54 ☑ tbb worker 53 ☑ tbb worker 52 Filter messages Clear Time 3s 909,796,918ns 3s 913 No FP16 (HALF-PREC TBB engage reading runtime_texture_binary_header_v4015 reading runtime_texture_binary_header_v4015 reading runtime_texture_binary_header_v4015 TBB engage TBB engage TDD ## Plots

https://github.com/david-grs/geiger Xpedite https://github.com/morganstanley/Xpedite Intel IACA https://www.intel.com/content/www/us/en/developer/articles/tool/architecture-code-analyzer.html ## The Obligatory log! Filter messages Clear • filterable log! • no thread sync needed! ☑ tbb worker 55 ☑ tbb worker 54 ☑ tbb worker 53 ☑ tbb worker 52 eclair_vk_framebuffer: Creating persistent pixels, la eclair_vk_framebuffer: No FP16 (HALF-PREC TBB engage reading runtime_texture_binary_header_v4015 reading runtime_texture_binary_header_v4015 reading runtime_texture_binary_header_v4015 TBB engage TBB engage TDD ## Plots

PRESENTATION AGENDA ☑ TVM @ AliOS Overview TVM @ AliOS ARM CPU TVM @ AliOS Hexagon DSP TVM @ AliOS Intel GPU ☑ Misc ## PART ONE TVM @ AliOS Overview ## AliOS Overview • AliOS (www.alios.cn) is a newly r0 = #0; jumpr r31 } ## PART FOUR AliOS TVM @ Intel GPU ## AliOS TVM @ Intel GPU • Implement the schedule from scratch • Leverage Intel Subgroup Extension ## Subgroups  ## AliOS TVM @ Intel GPU GEMM Hardware Efficiency @ Intel Apollo Lake GPU ![Image](/uploads/documents/9/0/e/a/90eab7a9909eddc3e1f4b253cda18ef6/p24_1

C++ 进阶：模板元编程与函数式编程 4. 编译器如何自动优化：从汇编角度看 C++ 5. C++11 起的多线程编程：从 mutex 到无锁并行 6. 并行编程常用框架：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 可以提供多个库，又称组件 (components)，比如 TBB 这个包，就包含了 tbb, tbbmalloc, tbbmalloc_proxy 这三个组件。 - 因此为避免冲突，每个包都享有一个独立的名字空间，以 :: 的分割（和 C++ 还挺像的）。 • 你可以指定要用哪几个组件： • find package(TBB REQUIRED COMPONENTS tbb tbbmalloc REQUIRED) REQUIRED) • target link libraries(myexec PUBLIC TBB::tbb TBB::tbbmalloc) ## 第三方库 - 常用 package 列表 1. fmt::fmt 2. spdlog::spdlog 3. range-v3::range-v3 4. TBB::tbb 5. OpenVDB::openvdb 6. Boost::iostreams 7

8ea9d54e/p14_4.jpg) ## Benchmarks: ## Test Environment • Ubuntu 24.04 GCC 13.2.0 - 13th Gen Intel(R) Core(TM) i9-13900HX - Hyperthreading disabled - All 16 “efficient” cores isolated ## Three Three established MPMC queues were benchmarked along with Work Contracts: Boost “lock free” TBB "concurrent_queue" MoodyCamel "ConcurrentQueue" • Work Contracts ## Caution: Slideware bench(std::forward(task), numThreads, numTasks); bench<tbb_queue>(std::forward(task), numThreads, numTasks); bench(std::