晨读GCC

每天早晨的第一个或两个小时，用来读GCC，每天中午的第一个两个小时，用来干苦累活。。。

读代码时候的原则简单说一下：

（1）不要浮于框架，要从细处点滴积累，从一个work horse函数开始。

（2）先算法后代码，会更有效，不会一叶障目，管中窥豹。

（3）结合更多的独立思考，会更高效，异步的学习方式很自然，但是要尝试改变。

（4）提倡资料汇总，并点面结合的方法。

（5）多实验，尽管没有现成的和成熟的实验方法，但是自己写实验方案也不失好方法。

有计划，加多坚持，等于目标成功：

下面说一下计划,现在走到了优化算法这一步，但是对三层（算loop和cfg的话是五层咯）语言模式，还不是十分的了然。因此还存在一个反向巩固的过程，这会使开始阶段比较慢一些，同样，每个pass都是算法和数据结构（数据表示层）的结合，熟悉了一个算法及对应的代码以后就会对相应的实际数据结构也更加熟悉。更进一步，对别的算法比较了解以后对相应的代码读的就比较快了。

因此，计划先对某 specific pass的代码进行研读，再更有针对性的了解其他优化遍的代码。

这个启蒙代码，要力求算法清晰，资料丰富。

试试vrp pass 好了， go ！

quotes：

Instead of finding the expressions which are constant, we want to determine the weighted range of values an expression can have during the execution of a program.Branches based on this variable can then be accurately predicted simply by examining this weighted value range.

Their approach uses the static single assignment (SSA) representation, which was originally developed as a dataflow representation geared towards propagating values through a program. It therefore provides an ideal platform for this kind of analysis. Using SSA form, constant propagation can be performed in essentially linear time.

In SSA form every variable has only one assignment, and therefore each use has a single definition. In essence, the def-use chains (called SSA edges) become a one-to-many rather than a many-to-many relationship.