CS715 :  Design and Implementation of the Gnu Compiler Generation Framework  (2008-2009)

Welcome to the course homepage. Your instructor is Uday Khedker and TA is Sameera Deshpande. Please see the proposal document of this course to get an idea of the nature of the course.

Announcements

• (14 Nov 2008). I am planning to float several R&D projects (CS 691) on GCC in the next semester. I have listed some topics below. Your suggestions about possible topics are most welcome.
  

• Extending the methodology of construction of machine descriptions to include possible optimizations through machine descriptions.
• Precise construction of call graphs using interprocedural function pointer analysis.
• Interprocedural pointer analysis.
• BDD based support for constructing lattices (intended to be used for data flow analysis).
• Extending the  generic data flow analyser gdfa to interprocedural level.

• (2 Nov 2008). Here is a report describing incremental construction of machine descriptions. See the file optabs.h in gcc subdirectory to find a list of the pattern names known to GCC.
  

• (17 Oct 2008). On second thoughts, I have decided to divide assignment 5 into two parts.

• The 5th assignment is to create spim machine description level 5 for GCC 4.0.2 by adding support for floating point types and structures to the language supported by level 4.  (Note that on the GCC workshop page, level 5 refers to the level in which instruction scheduling has been incorporated. Now we will call it level 6). The deadline for this assignment is Wed 5 Nov.
• The 6th assignment is to extend levels 0 to 5 to GCC 4.3.1 for the spim simulator. I expect the changes to be mainly confined to the header file. You can see the header file for level 0.0 for i386. You will have to replace the no-libgcc option by cc1 option. The deadline for this assignment is Sat 15 Nov.
  

• (17 Oct 2008). I will evaluate your 4th assignment on Wed 22 Oct.
  

• (17 Oct 2008). As described in the lecture on Wed 15 Oct, your 5th assignment is to extend the spim machine descriptions for level 4 in the following manner.

• Make the machine description work for GCC 4.3.1. This may require some changes to be made in the header file. You can take a look at the header file for gcc 4.3.1 for level 0.0 of i386 machine descriptions.This has been prepared by Jaishri; we do not know what changes are required for spim.
• Create level 5 machine descriptions by adding the following features to the language: Support for floating point types and structures.

• (29 Sep 2008). Your 3rd assignment was to traverse the call graph and print details of function calls and usage of global variables. Your 4th assignment is to traverse the RTL level CFG and print a more understandable RTL code by using "=" for set operations, converting prefix expressions into infix, and using familiar symbols for operations. All organizationsl details of the data structures, comments, and auxilliary information must be skipped. For example, consider the following RTL code dumped in file.c.13.cfg for an assignment statement c = a + b.


Observe that the note REG_EQUAL has been ignored. Please refer to gccint.pdf for details. The submission deadline for this assignment is 12:00 noon on Sunday 12 Oct. 2008.


• (29 Aug 2008). Here are some tips that might help you in assignment 2.

• Look for tcc_statement and tcc_expression in files tree.def and tree.h. Find out the relevant statements, expressions, and assignment statements.
• Segmentation fault results from accessing illegal memory, most often, a NULL pointer. When this occurs while building gcc, it means that a  change made by you has caused the resulting compiler in accesing illegal memory when it is used to compile the GCC source. Disabling the added code and rebuilding the compiler can help you to debug your code. This has to be done repeatedly by progressively enabling chunks of disabled code.
  
• In order the add options to your compiler, observe $SOURCE/gcc/common.opt, $BUILD/gcc/options.h, and $BUILD/gcc/options.c. These options may not be available if you compile cc1 independently as in the previous case. This is because the options are processed by the driver. Still you need not do the complete build after every small change. I have found the following approach very effective in speeding up the process: I do the complete build for the first time and then for subsequent builds, I give make cc1 command $BUILD/gcc/. Unlike first assignment, this does not require adding cc1 option in the top level make file. After this, the make install command in the top level directory copies the newly built cc1 to the libexec directory thereby making the new cc1 available to the gcc driver.


• (22 Aug 2008). Details of assingment 2. You have to add a pass that will traverse gimple cfg in GCC 4.3.1
  

• Generate the following statistics for each function processed: Number of global variables, number of local variables, number of scalar variables, number of aggregate variables (i.e. structures, arrays, pointers, unions etc.) number of assignment statements, number of function calls, number of unconditional branches, number of conditional branches, number of conditions, number and values of numeric constants, number of four arithmetic operations (+,-,*,/), number of copy statements.

• You have to submit a patch of your implementation by Wed 3 Sept.  13:00 hours.
  

• Additional work in any project will get additional credit.
  

• (22 Aug 2008). Slides on machine independent phases of GCC.
  

• (15 Aug 2008). Use the option "--enable-languages=c" option for your assignment to build cc1.
  

• (14 Aug 2008). Some of you have shown interest in knowing how I have implemented the configuration system required in your earlier assignment. I have updated the  simple compiler implementation  html pages that point to the required files. Observe the additional configuration that I have set up. 

• (14 Aug 2008). In yesterday's lecture, we compiled the programs empty.c, and add-global.c  using the command "gcc -fdump-tree-all -da -S" and examined the following dump files: AST (.c.t00.tu), Gimple (.c.t03.gimple and .c.t21.cfg), RTL (.c.00.expand, .c.13.cfg, .c.58.shorten), and Assembly (.s). Repeat the experiment for add-global-return.c, add-global-assign.c, add-local.c , add-local-return.c, and  add-local-assign.c. Subsequent steps in understanding compilation are to consider the following features: more complex expressions, arrays and structures, loops, and function calls with and without parameters.
  

• (14 Aug 2008). Assignment 1. Modify the configuration and build process of GCC 4.3.1 such that after the configure command, it is possible to give a command "make cc1" from the top level build directory to build only cc1. This assignment is due on Friday 22 Aug.
  

• (14 Aug 2008). An interesting (but old) discussion about configuration and build tools.
  

• (2 Aug 2008). For your assignment, you may want to read this seminar report on configuration and build tools.

• (1 Aug 2008). In your first assignment, you have to build a configuration system that builds ceCC by checking the following:

  • If lex, yacc, or iburg are not available, the configuration must abort with a suitable error message.
  • If spim is not available, the configuration should just give a warning but the compiler must be built.
  • If gv is not available, the configuration should check for kghostview and replace the use of gv in the code by kghostview and the compiler should be built.
  • If dot is not available, the configuration should avoid compiling the code that uses dot. The file ce-show-tree.c should also be excluded from compilation but rest of the compiler must be built.

• (1 Aug 2008). Slides that were used in today's lecture have been uploaded.

• (1 Aug 2008). Simple compiler which we will study in today's lecture.

• (30 July 2008). Many of you have asked me several questions outside of the class. These questions indicate that many of you have not seen a compiler source code and some of you have not used make before. Hence I have decided to change the schedule a bit. Instead of the lab exercise on Friday, we will have a lecture and I will explain to you the source of a topy compiler. Our first lab experiment will involve enhancing this toy compiler and building a configuration system for it. Hence we meet in SIC 305/304 on Friday, 1st August rather than in NSL.

• (30 July 2008). Slides of today's lectures. I have updated the slides to remove the confusion about building a cross compiler. Also, the statement about --prefix is correct.

• (29 July 2008). Most of the GCC sources are available locally on IITB ftp server. You may need the sources of GCC 4.0.2 and GCC 4.3.0 only in the course. Please avoid downloading them from external sources. Please make sure that you have adequate space in your account before you download these sources The .tar.bz2 file for GCC 4.0.2 is 30.3 MB which inflates to 378 MB on untarring. The corresponding sizes for GCC 4.3.0 are 56.6 MB and 511 MB. The build directories take more space than the extracted source directories.
  I will to talk to the system administrators to find out if they would prefer this work to be done on specific machines only. Until then please avoid compiling the GCC source on commonly used servers.

• (27 July 2008). Slides of first two lectures and the demo examples will be put up some time tonight.

• (27 July 2008). Our documentation on GCC can be found on the GCC Resource Center website.

• (27 July 2008). We will have a lecture on the build system of GCC on Wednesday 30 July instead of the proposed lab.

Back to the Main Page