A Small C Compiler
Switch branches/tags
Nothing to show
Clone or download
rui314 Promote int variables to register values.
In this patch, we promote all non-address-taken integer variables
to registers, so that we use as many registers as we have.
Latest commit 882e4b2 Sep 1, 2018

README.md

9cc C compiler

9cc is a successor of my 8cc C compiler. In this new project, I'm trying to write code that can be understood extremely easily while creating a compiler that generates reasonably efficient assembly.

9cc has more stages than 8cc. Here is an overview of the internals:

  1. Compiles an input string to abstract syntax trees.
  2. Runs a semantic analyzer on the trees to add a type to each tree node.
  3. Converts the trees to intermediate code (IR), which in some degree resembles x86-64 instructions but has an infinite number of registers.
  4. Maps an infinite number of registers to a finite number of registers.
  5. Generates x86-64 instructions from the IR.

There are a few important design choices that I made to keep the code as simple as I can get:

  • Like 8cc, no memory management is the memory management policy in 9cc. We allocate memory using malloc() but never call free(). I know that people find the policy odd, but this is actually a reasonable design choice for short-lived programs such as compilers. This policy greatly simplifies code and also eliminates use-after-free bugs entirely.

  • 9cc's parser is a hand-written recursive descendent parser, so that the compiler doesn't have any blackbox such as lex/yacc.

  • I stick with plain old tools such as Make or shell script so that you don't need to learn about new stuff other than the compiler source code itself.

  • We use brute force if it makes code simpler. We don't try too hard to implement sophisticated data structures to make the compiler run faster. If the performance becomes a problem, we can fix it at that moment.

  • Entire contents are loaded into memory at once if it makes code simpler. We don't use character IO to read from an input file; instead, we read an entire file to a char array in a batch. Likewise, we tokenize a whole file in a batch rather than doing it concurrently with the parser.

Overall, 9cc is still in its very early stage. I hope to continue improving it to the point where 9cc can compile real-world C programs such as Linux kernel. That is an ambitious goal, but I believe it's achievable, so stay tuned!