这两次也是经常见的单词，从字面理解也好理解：从上往下，从下往上。
举一个例子《A Top-Down Method for Performance Analysis and Counters Architecture》Ahmad Yasin。
这个文章里有这么一句：
The primary aim of performance monitoring units(PMUs) is to enable software developers to effectively tune their workload for maximum performance on a given system. …A main contributor to this, is the fact that these performance events were historically defined in an ad-hoc bottom-up fashion.
这里的ad-hoc bottom-up，让人想起有时候调试软件问题的时候，经常性的会加一些打印输出，看看这个变量对不对，看看那个变量对不对，从具体的函数里的某些变量开始，慢慢，这些打印就变得非常多。和这个例子非常类似。 bottom-up的含义就是没有从全局出发，来分析问题，解决问题，看待问题，所以多多少少有遗漏的地方。而反衬到top-down的概念上，就是从全局，整体来看问题，需要在哪里加打印输出，再打印输出，不是关键的地方不加，一面造成开日志的负担。从全局出发的缺点是，是比较全，有时感觉无从下手，前期工作比较多，而且有时候，可能考虑不全，导致有所遗漏。
上面这句话的出处是：《A Performance Counter Architecture for Computing Accurate CPI Components》
The reason is that performance counters have historically been constructed in a bottom up fashion by focusing on the events that aect performance, for example the various cache miss rates, without regard for how the performance counts should be combined to form an
overall picture of CPI.