WHIRL SSA形式中的别名和间接内存操作有效表示

需积分: 10 164 浏览量更新于2024-07-21 收藏 79KB PDF 举报

"WHIRL SSA doc 是一份关于WHIRL SSA设计的文档，主要探讨了在静态单赋值（Static Single Assignment, SSA）形式中有效地表示别名和间接内存操作的问题。该文档来源于Open64项目，并由包括Fred C. Chow在内的五位作者共同撰写。文档中提出了一种防止因别名存在而导致SSA变量版本爆炸的方法，以及一种允许全局优化间接内存操作的技术，实现了基于全局值编号的精确且紧凑的SSA表示。" 在编译器领域，SSA形式是一种用于中间表示(IR)的优化技术，它要求每个变量在任何控制流点只能被赋值一次，从而简化了数据流分析和优化。然而，别名管理和间接内存操作在SSA形式中是极具挑战性的议题。别名是指两个或多个变量可能引用相同内存位置的情况，而间接内存操作涉及通过指针访问内存，其目标地址在运行时才能确定。该文档的核心贡献在于解决这两个问题。首先，针对别名管理，作者提出的方法旨在避免在处理别名时SSA变量版本的过度增长。通常，别名的存在会导致SSA形式中的变量产生大量版本，这会增加编译器的复杂性和内存需求。作者的解决方案旨在保持SSA表示的规模，同时保持其准确性。其次，对于间接内存操作，文档介绍了一种技术，允许这些操作在全球范围内进行公共化。这意味着编译器可以识别并合并类似的间接内存访问，提高代码效率。这种全局优化对于减少冗余和提高程序性能至关重要，特别是在存在动态指针运算的场景下。文档还提到了全局值编号，这是一种用于跟踪和比较变量值的技术。通过全局值编号，编译器可以识别出不同路径上的变量是否具有相同的值，这对于SSA形式的构建和优化非常有用。这种方法可以帮助编译器更好地理解程序的行为，从而进行更精细的优化。 "WHIRL SSA doc"提供了关于如何在SSA形式中有效处理别名和间接内存操作的深入洞察，这对编译器设计者和优化专家来说是极其宝贵的资源。通过这些方法，编译器可以生成更高效、更紧凑的代码，从而提升软件的整体性能。

same variable from being referenced later in the program. On the use side, in addition to

real uses of the variable, there are places in the program where there are potential refer-

ences to the variable that need to be taken into account in analyzing the program. We

call these potential references MayUse.

To accommodate the MayDefs, we use the idea from [CCF94] in which SSA edges

for the same variable are factored over its MayDefs. This is referred to as location-fac-

tored SSA form in [Steen95]. We model this effect by introducing the χ assignment

operator in our SSA representation. χ links up the use-def edges through MayDefs. The

operand of χ is the last version of the variable, and its result is the version after the

potential deﬁnition. Thus, if variable i may be modiﬁed, we annotate the code with i

χ(i

), where i

is the current version of the variable.

To model MayUses, we introduce the µ operator in our SSA representation. µ takes

as its operand the version of the variable that may be referenced, and produces no result.

Thus, if variable i may be referenced, we annotate the code with µ(i

), where i

is the

current version of the variable.

In our internal representation, expressions cannot have side effects. Memory loca-

tions can only be modiﬁed by statements, which include direct and indirect store state-

ments and calls. Thus, χ can only be associated with store and call statements. µ is

associated with any dereferencing operation, like the unary operator * in C, which can

happen within an expression. Thus, µ arises at both statements and expressions. We also

mark return statements with µ for non-local variables to represent their liveness at func-

tion exits. Separating MayDef and MayUse allows us to model the effects of calls pre-

cisely. For example, a call that only references a variable will only cause a µ but no χ.

For our modeling purpose, the µ takes effect just before the call, and the χ takes effect

right after the call. Figure 1 gives an example of the use of µ, χ together with φ in our

SSA representation. In the example, function func uses but does not modify variable i.

The inclusion of µ and χ in the SSA form does not impact the complexity of the

algorithm that computes SSA form. A pre-pass inserts unnamed µ’s and χ’s for the

Fig. 1. Example of µ, χ and φ

← 2

if (j

)

*p ← 3

←χ(i

)

µ(i

)

call func()

←φ(i

, i

)

return i3

Original Program:

SSA Representation:

i ← 2

if (j)

*p ← 3

call func()

return i

剩余15页未读，继续阅读

qq_21255579

粉丝: 0
资源: 2

WHIRL SSA形式中的别名和间接内存操作有效表示

HP SSA 用户指南

HP 服务器SSA下常规设置

open64 WHIRL SYMTAB

campbell diagram中的whirl direction bw fw什么意思

WHIRL

whirl：用于实现LL（1）解析器的仅标头的库

ember-cli-whirl:Ember CLI 插件包含 whirl.css

旋转：Whirl，Rust中的开源世界服务器实现

nvopencc教程

whirl:用于循环作业的淡化宝石。 预装了 Redis 和 HTTParty。 考虑到 Heroku 的使用情况而构建

最新资源

whirl:用于循环作业的淡化宝石。预装了 Redis 和 HTTParty。考虑到 Heroku 的使用情况而构建