2011年02月_wuhui_gdnt

10月 08月 07月 06月 05月 04月 03月 02月 01月

原创 Studying note of GCC-3.4.6 source (169)

5.13.5.2.2.1.1.2. Select output section After finding out or creating the constant descriptor, next in output_constant_definvokes below function to try to output this constant in memory. 2509static void 2510maybe_outp

2011-02-26 15:52:00 593

原创 GCC-3.4.6源代码学习笔记（169）

5.13.5.2.2.1.1.2. 确定输出节 找到或准备好了常量描述符之后，接着在output_constant_def中调用下面的函数，尝试在内存中输出这个常量。 2509static void 2510maybe_output_constant_def_contents (struct constant_descriptor_tree*desc, in varasm.c 2511

2011-02-26 15:50:00 1226

原创 Studying note of GCC-3.4.6 source (168)

5.13.5.2.2.1. The initializer Here on our target Linux/x86, the output file would be of formatELF. ELF is the format supports relocatable and PIC (position independentcode). Chapter 3 of [10] gives a good description about ELF. As summeray

2011-02-26 15:48:00 1396 2

原创 GCC-3.4.6源代码学习笔记（168）

5.13.5.2.2.1. 初始值 这里在我们的目标Linux/x86，输出文件将具有格式ELF。ELF是支持重定位及PIC（位置无关代码）的格式。【10】的第三章给出了关于ELF的一个很好的解释。从【10】的总结中，我们得到： ELF文件有3种稍有不同的风格的形式：可重定位（relocatable），可执行（executable）及共享对象。可重定位文件由编译器及汇编器创建，但在运行前需要链接器的处理。可执行文件已完成所有的重定位及确定了所有的符号，除了在运行时确定的共

2011-02-26 15:45:00 1726

原创 Studying note of GCC-3.4.6 source (167)

5.13.5.2. Output assemble for pending variables Return finish_file, we have seen that flag_unit_at_a_timeis set at level above ‘-O2’.This flag also can be set by option -funit-at-a-time which described in [6] asbelow:

2011-02-26 15:36:00 815

原创 GCC-3.4.6源代码学习笔记（167）

5.13.5.2. 为待定变量发布汇编 回到finish_file，我们已经看到flag_unit_at_a_time在‘-O2’以上优化级别中设置为1。这个标记也可以由选项-funit-at-a-time来设定，【6】对其描述如下： 在开始产生代码之前，解析整个编译单元。这允许进行某些额外的优化，不过会消耗更多内存（总体而言）。使用unit-at-a-time模式有一些兼容性问题： 使能unit-at-a-

2011-02-26 15:34:00 1437

原创 Studying note of GCC-3.4.6 source (166)

5.13.5.1.3. Preparealias set However, the entry point for creating alias set is get_alias_setbelow. Both node of type and node of declaration can claim for alias analysis,but declaration’s alias set is determined by the alias set o

2011-02-19 11:58:00 750

原创 GCC-3.4.6源代码学习笔记（166)

5.13.5.1.3. 准备别名集 构建别名集的入口是下面的get_alias_set。类型节点及声明节点可以要求别名分析，不过声明的别名集由其底层类型的别名集确定。在467行的条件选出声明的节点。在C++前端，475行的钩子get_alias_set指向cxx_get_alias_set。这个函数主要处理类型定义节点，我们暂时不管这个函数。 449 HOST_WIDE_INT 450 get_alias_set (tree t)

2011-02-19 11:56:00 1427

原创 Studying note of GCC-3.4.6 source (165)

5.13.5. Code analysis and optimization5.13.5.1. Preliminaries- alias set analysis5.13.5.1.1. Conceptof alias set In wikipedia, there is agood explaination of alias set given below. Alias analysis is a techni

2011-02-19 11:54:00 789

原创 GCC-3.4.6源代码学习笔记（165）

5.13.5. 代码分析及优化5.13.5.1. 预备知识- 别名集分析5.13.5.1.1. 别名集概念 在维基百科（wikipedia）中，给出了如下的别名集的一个很好的解释。 别名分析是在编译器理论中的一个技术，用于确定一个存储位置是否可能以多个方式访问。两个指针通常被认为别名，如果它们指向同一个位置。 别名分析技术通常分为流-敏感（flow-sensitivity）及上下文-敏感（context-se

2011-02-19 11:53:00 1521

原创 Studying note of GCC-3.4.6 source (164)

5.13.4.7.1.3. Buildnodes for the function call Return from build_function_call, coerced_params at line 2494 holdsthe converted arguments. If the function invoked is builtin funciton, thebuiltin functions can be expanded here and no n

2011-02-19 11:51:00 804

原创 GCC-3.4.6源代码学习笔记（164）

5.13.4.7.1.3. 为函数调用构建节点从build_function_call返回，在2494行的coerced_params保存了转换后的实参。如果被调用的函数是内建函数，这个内建函数可以在这里被展开，因为没有函数可以调用，不需要创建CALL_EXPR节点。3716 tree3717 expand_tree_builtin (tree function, tree params, tree coerced_params) in c-common.c3718 {3719

2011-02-19 11:49:00 911

原创 Studying note of GCC-3.4.6 source (163)

5.13.4.7.1.2. Doappropriate conversion for arguments In this section, we go a little deeper to see in detail how to matchthe argument with the parameter, and do necessary conversion appropriately. In C++ front-end, function parameters

2011-02-19 11:47:00 783

原创 GCC-3.4.6源代码学习笔记（163）

5.13.4.7.1.2. 为实参执行适合的转换 在这一节中，我们要深入一些来看一下如何把实参匹配形参，并执行必须的转换。 在C++前端中，函数形参被记录在FUNCTION_TYPE的TYPE_ARG_TYPES域中，它是一个tree_list类型的串，其节点中的TREE_VALUE是对应形参的类型；TREE_PURPOSE是缺省实参值的一个表达式，如果有的话。如果在该链表中最后的节点是void_list_node（一个TREE_LIST节点，其中的TREE_VALUE是

2011-02-19 11:46:00 1123

原创 Studying note of GCC-3.4.6 source (162 - continue)

At line 5350 current_class_ptr is the PARM_DECL for the `this'pointer, when we're processing a member function. See that in below only the typeof instancecan be known for sure it returns non-null. Compared with condition at line5281,

2011-02-19 11:44:00 708

原创 GCC-3.4.6源代码学习笔记（162-续）

在5350行，当我们正在处理一个成员函数时，current_class_ptr是用于‘this’指针的PARM_DECL。看到在下面，仅当instance的类型是可以明确知道的情形，才返回非空值。与5281行的条件相比，5281行的条件表示一个指针，而在C++里，它可以是该类层次中不同的类型，以提供多态。因此作为结果，它返回NULL_TREE。 5275 static tree 5276 fixed_type_or_null (tree in

2011-02-19 11:36:00 1028

原创 Studying note of GCC-3.4.6 source (162)

5.13.4.7.1. Generatingcode for function invocation5.13.4.7.1.1. Buildexpression to take function’s address Now, we begin a new long offshoot away from our main route. When wewrite code like: f (a, b); to call function f, the co

2011-02-19 11:32:00 788

原创 GCC-3.4.6源代码学习笔记（162）

5.13.4.7.1. 为函数调用产生代码5.13.4.7.1.1. 构建获取函数地址的表达式 当我们写出类似：f (a, b);的代码来调用f，编译器需要努力使得这个调用以期望的方式可行。下面的函数泄露了编译器的所作所为。 2420 tree 2421 build_function_call (tree function, treeparams) i

2011-02-19 11:28:00 1220

原创 Studying note of GCC-3.4.6 source (161)

5.13.4.7. Generate code for invoking initializer functionaccording to priority If iteration taken by code between line 2573 to 2800 turns outstable – that is nothing new is generated or in other words all objectsexplicitly or

2011-02-13 17:27:00 795

原创 GCC-3.4.6源代码学习笔记（161）

5.13.4.7. 发布代码根据优先级调用初始化函数 如果由代码2573至2800行执行的迭代稳定了下来——即不再产生新的东西，或者换句话说，所有显式或隐式涉及的对象都已经处理了，就可以继续完成机器代码分别这个目标。 finish_file (continue) 2801 /* All used inlinefunctions must have a definition at this point.

2011-02-13 17:26:00 1205

原创 GCC-3.4.6源代码学习笔记（160）

5.13.4.5. 迭代 – 处理名字空间中的全局对象 前面我们已经为需要构造函数及析构函数的全局或静态对象准备、注册好了初始化函数及退出函数。这一次的迭代检查是否还有没有处理的全局对象，并处理之。函数walk_namespaces从上至下以前序访问名字空间树。 848 int 849 walk_namespaces (walk_namespaces_fn f,void* data)

2011-02-11 08:11:00 953

原创 Studying note of GCC-3.4.6 source (160)

5.13.4.5. Iterate – handling globals in namespaces Before, we have prepared and registered initialize and exittingfunction for global or static objects having constructor or destructor. In thisiteration, it checks if there is a

2011-02-11 08:11:00 668

原创 Studying note of GCC-3.4.6 source (159)

5.13.4.4. Iterate – emitting code for deferred function Function generated by the compiler (via implicitly_declare_fn, mark_decl_instantiated,or build_clone)or inline method is cached in deferred_fns. Arriving here, we

2011-02-11 08:09:00 627

原创 GCC-3.4.6源代码学习笔记（159）

5.13.4.4. 迭代– 为延迟函数发布代码 由编译器产生的函数（经由implicitly_declare_fn，mark_decl_instantiated或build_clone）或内联函数缓存在deferred_fns中。因为到这里，我们已经完成了整个源文件的代码解析，处理这些延迟函数的信息都应该齐备了。 首先，对于由编译器自己产生的“人造”函数，其设置了标记DECL_ARTIFICIAL；并且FUNCTION_DECL节点，其DE

2011-02-11 08:07:00 832

原创 Studying note of GCC-3.4.6 source (158)

5.13.4.3. Iterate – emitting code for ctor/dtor of globalaggregate Next in the DO WHILEloop, at line 2651 static_aggregates is a list that holds aggregateshaving constructor or destructor and reside in the global scope. As

2011-02-11 08:05:00 1395

原创 GCC-3.4.6源代码学习笔记（158）

5.13.4.3. 迭代 – 发布全局聚集类的构造函数/析构函数 接着在这个DO WHILE循环中，在2651行static_aggregates是一个链表，它包含了在全局绑定域中具有构造函数或析构函数的聚集类。正如我们已经看过，编译器需要确保在进入“main”函数之前，所有全局变量必须已经分配了内存及被初始化。5.13.4.3.1. 修整需要初始化的变量 因此在2651行，prune_vars_needing_no_initializ

2011-02-11 08:03:00 2209

原创 Studying note of GCC-3.4.6 source (157)

5.13.4.2. Iterate - emitting tinfo Now all tinfos need code emitted have been chained into unemitted_tinfo_decls,then in finish_fileat line 2628, emit_tinfo_decldetermines whether the tinfo of decl needs be emitted, retur

2011-02-11 08:01:00 591

原创 GCC-3.4.6源代码学习笔记（157）

5.13.4.2. 迭代– 发布tinfo 现在所有需要代码发布的tinfo已经被串入unemitted_tinfo_decls中。接着在finish_file的2628行，emit_tinfo_decl确定decl的tinfo是否需要发布，如果需要，则返回true。下面，看到之前在finish_file的2571行调用的emit_support_tinfos中，doing_runtime被设为1。 那么在调用class_initializ

2011-02-11 07:57:00 987

原创 Studying note of GCC-3.4.6 source (156)

5.13.3. Generate tinfo for fundamental types After outputting PCH file if required, back in finish_file to continue thestruggle of assemble emitting. Next, the compiler should prepare tinfo for thefundamental types. finis

2011-02-02 09:48:00 928

原创 GCC-3.4.6源代码学习笔记（156）

5.13.3. 为基本类型产生tinfo 操持完PCH文件，回到finish_file中继续为汇编代码的生成而奋斗。接下来编译器将为基本类型准备tinfo。 finish_file (continue) 2546 /* Otherwise, GDBcan get confused, because in only knows 2547 about source for LINENO-1l

2011-02-02 09:45:00 1484

原创 Studying note of GCC-3.4.6 source (155)

5.13.2. Output PCH file Seeing something related to conversion, resumes our example inafter-parsing stage. Return from expand_or_defer_fn we are back cp_parser_function_definition_after_declaratorwhich returns the FUNCTION_DECL immedi

2011-02-02 09:36:00 908

原创 GCC-3.4.6源代码学习笔记（155）

5.13.2. 输出PCH文件 看过了转换相关的内容后，回到我们例子的解析后的阶段，从expand_or_defer_fn返回，我们回到cp_parser_function_definition_after_declarator，并立即返回FUNCTION_DECL，这个FUNCTION_DECL为cp_parser_function_definition_from_specifiers_and_declarator返回，然后cp_parser_init_declarator

2011-02-02 09:33:00 1161

Introduction to Theory of Computation

This is a free textbook for an undergraduate course on the Theory of Computation, which we have been teaching at Carleton University since 2002.Until the 2011/2012 academic year, this course was offered as a second-year course (COMP 2805) and was compulsory for all Computer Science students. Starting with the 2012/2013 academic year, the course has been downgraded to a third-year optional course (COMP 3803).

2018-09-23

Survey on Instruction Selection

Instruction selection is one of three optimization problems involved in the code generator back-end of a compiler. The instruction selector is responsible of transforming an input program from its target-independent representation into a target-specific form by making best use of the available machine instructions. Hence instruction selection is a crucial part of efficient code generation.

2018-09-23

Towards a Compilation infrastructure for network processors

Modern network processors (NPs) typically resemble a highly-multithreaded multiprocessor-ona-chip, supporting a wide variety of mechanisms for on-chip storage and inter-task communication. NP applications are themselves composed of many threads that share memory and other resources,and synchronize and communicate frequently. In contrast, studies of new NP architectures and features are often performed by benchmarking a simulation model of the new NP using independent kernel programs that neither communicate nor share memory. In this paper we present a NP simulation infrastructure that (i) uses realistic NP applications that are multithreaded, share memory, synchronize, and communicate; and (ii) automatically maps these applications to a variety of NP architectures and features. We use our infrastructure to evaluate threading and scaling, on-chip storage and communication, and to suggest future techniques for automated compilation for NPs.

2018-09-23

Crafting a Compiler

Brief Contents 1 Introduction 1 2 A Simple Compiler 31 3 Scanning—Theory and Practice 57 4 Grammars and Parsing 113 5 Top-Down Parsing 143 6 Bottom-Up Parsing 179 7 Syntax-Directed Translation 235 8 Symbol Tables and Declaration Processing 279 9 Semantic Analysis 343 10 Intermediate Representations 391 11 Code Generation for a Virtual Machine 417 12 Runtime Support 445 13 Target Code Generation 489 14 Program Optimization 547

2018-09-01

空空如也

TA创建的收藏夹 TA关注的收藏夹

TA关注的人

Introduction to Theory of Computation

Survey on Instruction Selection

Towards a Compilation infrastructure for network processors

Crafting a Compiler

How Debuggers Work

Compiler Construction-20th

Compiler Construction-19th

Compiler Construction-17th

Compiler Construction-16th

Compiler Construction-5th

Construction and Evolution of Code Generator

Compiler Construction-11th

The compiler design handbook

Data_Flow_Analysis_Theory_And_Practice(Bookos.org)

The Compiler Design Handbook Optimizations and machine code generation

C++_Coding_Standards_101_Rule

instruction scheduling for instruction level parallel processor

空空如也