Uproot reads ROOT TTrees using pure Python. For numerical and (singly) jagged arrays, this is fast because a whole block of data can be interpreted as an array without modifying the data. For other cases, such as arrays of std::vector<std::vector<float>>, numerical data are interleaved with structure, and the only way to deserialize them is with a sequential algorithm. When written in Python, such algorithms are very slow. We solve this problem by writing the same logic in a language that can be executed quickly. AwkwardForth is a Domain Specific Language (DSL), based on Standard Forth with I/O extensions for making Awkward Arrays, and it can be interpreted as a fast virtual machine without requiring LLVM as a dependency. We generate code as late as possible to take advantage of optimization opportunities. All ROOT types previously implemented with Python have been converted to AwkwardForth. Double and triple-jagged arrays, for example, are 400x faster in AwkwardForth than in Python, with multithreaded scaling up to 1 second/GB because AwkwardForth releases the Python GIL. We also investigate the possibility of JIT-compiling the generated AwkwardForth code using LLVM to increase the performance gains. In this paper, we describe design aspects, performance studies, and future directions in accelerating Uproot with AwkwardForth.

翻译：Uproot使用纯Python读取ROOT TTree。对于数值数组和（单层）锯齿数组，由于整块数据可直接解释为数组而无需修改数据，因此速度较快。对于其他情况，如std::vector<std::vector<float>>类型的数组，数值数据与结构信息交错排列，反序列化的唯一方法是采用顺序算法。当用Python编写此类算法时，速度非常缓慢。我们通过使用快速执行的语言编写相同逻辑来解决此问题。AwkwardForth是一种领域特定语言（DSL），基于标准Forth并扩展了用于生成Awkward Array的I/O功能，它可作为快速虚拟机解释执行，无需依赖LLVM。我们尽可能晚地生成代码以充分利用优化机会。此前用Python实现的所有ROOT类型已转换为AwkwardForth。例如，双层和三层锯齿数组在AwkwardForth中的速度比Python快400倍，且由于AwkwardForth释放了Python GIL，多线程扩展可达1秒/GB。我们还研究了使用LLVM对生成的AwkwardForth代码进行即时编译以进一步提升性能的可能性。本文描述了使用AwkwardForth加速Uproot的设计方案、性能研究及未来方向。