YAMA 0006 - C 2 C Compiler - Intermediate C99 to C99 optimizing compiler

• Author(s): Bhathiya Perera
• Status : Draft

Problem

Yaksha uses C as a final compilation target. This is to introduce a way to have an intermediate C as a IR that can then be compiled to C

Naming

• iC99 - Intermediate C 99

Language

• Looks like C99.
• Loops -> for, while, do while
• structs, constants, unions, #define
• We have built-ins such as yk__sdsnew, etc.

#define yy__raylib_Music Music

i32 yy__main() {
  yy__raylib_Music x;

  return 0;
}

Architecture

• Tokenize
  • Level 1 -> no keyword recognition
  • Recognizes normal operators, identifiers, #?? and comments
  • Discard comments
• Parse all with pre-processor statements
• Remove comments
• Apply #defines -> until there are no changes present
• AST based
• Only works with 1 file
• Entry points yy__main yy__game_step, yy__init_state

Optimizations

Opt 01 - apply all #defines starting with yy__

• Implement a simple pre-processor, however only capable of applying #defines that start with yy__
• Opt 01.01 - Support non function like macros
• Opt 01.02 - Support for function like macros

Opt 02 - data types

• Opt 02.01 - Push down support of integer data types from Yaksha compiler to iC99.
• Instead of having extra complexity at the compiler level for i32 ike data types, we push it to iC99.

Opt 03 - clean up

• Opt 03.01 - Remove any method that cannot be reached from yy__main()
• Opt 03.02 - Remove any operation in same block after return

Opt 04 - constant folding

• Opt 04.01 - Apply -
• Opt 04.02 - Remove excess ( ) parentheses
• Opt 04.03 - Do basic math
• Opt 04.04 - Apply basic string built-ins
• Opt 04.05 - Apply basic array built-ins
• Opt 04.06 - Dead code elimination

Opt 05 - apply constant function calls

• Everything is a constant expression?
• For each function things with simple math, assignments and built ins are considered pure
• Any of these pure functions can be invoked during compile time and folded

Opt 06 - small function in-lining

• Inline small functions < 20 lines?

Opt 07 - multiple rounds of previous optimizations

while changed:

• Opt 03
• Opt 04
• Opt 05
• Opt 06