oneapi-src/oneTBB
OneTBB
oneTBB 是一个 C++ 并行库和框架，旨在为应用程序添加多核并行性。它提供了一种基于任务的并行模型，将逻辑计算任务映射到可用的硬件核心，从而消除了手动线程管理的需要。
该库作为多核扩展工具，利用通用模板跨处理器扩展数据并行操作，以实现可移植的性能。它采用基于任务的框架，确保计算工作负载分布在硬件资源上。
该项目涵盖共享内存并行、多核任务调度和数据并行扩展。它利用工作窃取（work-stealing）任务调度器、递归范围拆分和动态负载均衡，在运行时管理核心间的工作分配。
Features
- C++ Parallel Programming - Provides a comprehensive framework for implementing multi-core execution in C++ applications without manual thread management.
- Data Parallelism Scaling - Enables running operations across large datasets using templates to ensure portable performance across different multi-core processors.
- Dynamic Load Balancing - Implements runtime adjustments to workload distribution across cores to prevent individual processor bottlenecks.
- Work-Stealing Schedulers - Ships a work-stealing scheduler that dynamically balances computational tasks across hardware threads.
- Logical Parallelism Mapping - Automatically assigns computational tasks to available hardware cores to eliminate the need for manual thread management.
- Multi-Core Scheduling - Distributes computational workloads across available processor cores to optimize hardware utilization and execution speed.
- Multi-Core Scaling Libraries - Offers a set of generic templates and tools for scaling data-parallel operations across processors for portable performance.
- C++ Parallelism Libraries - Provides a library for adding multi-core parallelism to applications through task scheduling and data-parallel templates.
- Parallelism Templates - Utilizes generic C++ templates to decouple algorithmic logic from underlying hardware resource management for portable performance.
- Parallel Computing Implementation - Provides strategies for scaling computational throughput across multiple CPU cores in C++ applications.
- Data-Parallel Scaling - Executes data-parallel operations across multi-core processors using generic templates for portable and scalable performance.
- Parallel Task Execution - Implements a model for executing multiple asynchronous computations simultaneously and aggregating their results.
- Task-Based Concurrency Frameworks - Implements a framework that maps logical computational tasks to available hardware cores to avoid manual thread management.
- Recursive Range Splitting - Divides large datasets into smaller sub-ranges recursively until they reach a grain size suitable for parallel execution.
- Shared Memory Parallelism - Facilitates the development of applications that execute multiple tasks simultaneously within a single shared memory space.
- Thread Pool Abstractions - Provides an abstraction layer that maps logical parallel tasks to a pool of worker threads, hiding physical thread management.
- Development Environments - Template library for implementing parallel programming on multi-core processors.
