Remote Execution Caching Compiler (RECC) CppCon 2024 September 19, 2024 Shivam Bairoliya Software Engineer© 2024 Bloomberg Finance L.P. All rights reserved. What is RECC? ● Remote Execution Caching source build tool that wraps compiler commands and optionally forwards them to a remote build execution service ○ Encompasses the capabilities of both ccache and distcc ○ Supports remote linking and CC) ○ Supports multiple operating systems (Linux, macOS, Solaris) ● Compatible with any remote execution API server supported by Bazel ○ Single Host Server/Proxy: BuildBox-CASD ○ Distributed Server:

product lines, features (D***n, Ba**ta, IELTS, etc.) • Increasingly sophisticated systems(E.g., adaptive learning system) • Ever tighter integration • Algorithm, backend, data, all part of one system google/protobuf grpc/grpc-go ************************************************************ cd $(bazel info execution_root) && bazel-out/host/bin/external/com_google_protobuf/protoc \ --plugin=protoc-gen-go=baze --experimental_platforms=@io_bazel_rules_go//go/ toolchain:linux_amd64 //:demo # list all targets bazel query @com_github_grpc_grpc//... --output label_kind # ignore dir bazel build //... -//ignored_dir/

索深度学习、强化学习、人工智能、大数据、知 识图谱、NLP 和计算机视觉前沿的技术发展，探索本地生活服务电商的价值。主要工作方向 包括： ● 触发策略：用户意图识别、广告商家数据理解，Query 改写，深度匹配，相关性建模。 ● 质量预估：广告质量度建模。点击率、转化率、客单价、交易额预估。 ● 机制设计：广告排序机制、竞价机制、出价建议、流量预估、预算分配。 ● 创意 问题，SIM4 提出把行为中 的重要相关信息搜索出来。具体来说，在拿到需要被预估的商品信息后，可以像信息 检索一样，对用户行为商品构建一个快速查询的索引。待预估商品的信息可以当做是 一个 Query，从用户的所有行为中，查询与其相关的行为子序列。 因此，受启发于 MIMN 的超长序列和 SIM 的检索思路，我们设计出情境化序列检索 方法，依据 Cube 内的情境，从用户超长的历史行为序列中检索出的与当次访问情境 一系列探索。 最初，我们直接将近三年的点击行为直接引入到模型中来，发现离线效果提升显著， 但是带来的训练和推理的压力不可承受。在此基础上，借鉴了 SIM4，将候选商家的 品类 ID 当作 Query，先从用户的行为序列中检索出相同品类的商家，再进行兴趣建 模，离线取得了不错的收益。 具体的，尝试过使用二级品类和叶子品类来分别做检索，在检索后根据分位点进行最 大长度截断的情况下，二级品类检索出来的序列平均长度大约为

allowing to specify multiple product names per bean definition * added database product name "Adaptive Server Enterprise" to Sybase definition in default sql-error-codes.xml file * changed CustomSQ in case of a pre-bound Session * HibernateTemplate proxies exposed Sessions by default, applying query cache settings and transaction timeouts * deprecated HibernateTemplate's "createQuery", "getNamedQuery" TimerFactoryBean accepts SchedulerTimerTasks with "period" <= 0 now, scheduling them for one-time execution * added "schedulerListeners", "(global)JobListeners", "(global)TriggerListeners" bean properties

default impls now, throwing exceptions * TransactionSynchronization objects can influence their execution order through implementing the Ordered interface * MethodMap/NameMatchTransactionAttributeSource allowing to specify multiple product names per bean definition * added database product name "Adaptive Server Enterprise" to Sybase definition in default sql-error-codes.xml file * changed CustomSQ in case of a pre-bound Session * HibernateTemplate proxies exposed Sessions by default, applying query cache settings and transaction timeouts * deprecated HibernateTemplate's "createQuery", "getNamedQuery"

all? ● Typical embedded systems are small and follow static execution patterns ● Even a multi-core automotive chip may have parallel execution but no concurrency if components scheduled on different cores a predetermined time slice in the schedule for execution ● The complete schedule is configured upfront as part of the system design ● Temporal execution of tasks is completely deterministic ⇒ Tasks not OS at the bottom. ⇒ Lots of concurrency, asynchronous APIs as the default Asynchronous APIs in Adaptive Autosar ara::com::FindServiceHandle find_service_handle = RadarServiceProxy::StartFindService(

function calls • PEP 659: Specializing Adaptive Interpreter 动态语言的虚拟机可以根据执行中出现过的类型和值对代码进行特化，以提高运行效率。这种特化通常与 “JIT ”编译 器联系在一起。但即使没有编译成机器代码，这种优化也是有益的。 Execution Quicken Adaptive Instruction Specialization Deopt Deopt def get_url(path): return "https://example.com" + path print(dis.dis(get_url, adaptive=True)) 3 0 RESUME 0 4 2 LOAD_CONST 1 ('https://example.com') 4 LOAD_FAST 4 LOAD_FAST 0 (path) 6 BINARY_OP_ADAPTIVE 0 (+) 10 RETURN_VALUE PEP 659: Specializing Adaptive Interpreter 3 0 RESUME_QUICK 0 4 2 LOAD_CONST__LOAD_FAST

in "getRequestURI" vs "getContextPath" (can happen on Resin) * CommonsMultipartResolver exposes query parameters in multipart requests too, not just multipart POST parameters * AbstractFormControlle default impls now, throwing exceptions * TransactionSynchronization objects can influence their execution order through implementing the Ordered interface * MethodMap/NameMatchTransactionAttributeSource allowing to specify multiple product names per bean definition * added database product name "Adaptive Server Enterprise" to Sybase definition in default sql-error-codes.xml file * changed CustomSQ

in "getRequestURI" vs "getContextPath" (can happen on Resin) * CommonsMultipartResolver exposes query parameters in multipart requests too, not just multipart POST parameters * AbstractFormControlle default impls now, throwing exceptions * TransactionSynchronization objects can influence their execution order through implementing the Ordered interface * MethodMap/NameMatchTransactionAttributeSource allowing to specify multiple product names per bean definition * added database product name "Adaptive Server Enterprise" to Sybase definition in default sql-error-codes.xml file * changed CustomSQ

in "getRequestURI" vs "getContextPath" (can happen on Resin) * CommonsMultipartResolver exposes query parameters in multipart requests too, not just multipart POST parameters * AbstractFormControlle default impls now, throwing exceptions * TransactionSynchronization objects can influence their execution order through implementing the Ordered interface * MethodMap/NameMatchTransactionAttributeSource allowing to specify multiple product names per bean definition * added database product name "Adaptive Server Enterprise" to Sybase definition in default sql-error-codes.xml file * changed CustomSQ