Lecture 022

Compiler: translate code to machine code

• generate executable

• optimization

• recompiling program takes time

• running code on new hardware takes source code and new compiler

• no safety check

Interpreter: simulate on hardware

• cannot do optimization beyond single line

• running code on new hardware takes source code and new interpreter

• efficiency does not matter

• interactive

Two Stage Execution

1. compile the code into intermediate representation
2. interpret intermediate representation (bytecode) using virtual machine

Advantage of Using Virtual Machine

• compiler can perform optimizations

• VM is lightweight, little simulation overhead

• bytecode: each instruction fits in one byte

Bytecode

To run program, you need:

• bytecode

• specific virtual machine

• (you don't need compiler)

  • you can compile source code on different hardware
  • or even compile the compiler to compiler code

Common Use of Two-stage Approach

Python: Python bytecode -> Python VM PHP, Js: common bytecode -> LLVM GCC: Clang Pascal: first mainstream language use two-stage approachh Java: popularized this approach. Java bytecode -> JVM

• bytecode was compact to minimize download time

cc0 -s foo.c0 (can produce C code)
cc0 -b foo.c0 (can produce C0VM bytecode, this step is hard)
(this can be executed by c0vm, c0vm is easy)
c0vm foo.bc0

Writing the Interpreter

Understanding Arithmetic Expressions

• bipush can only push [-128, 127]

• 

• 

• 

Functions

• start with:
  • 1 byte for number of function arguments
  • 1 byte for number of local variables plus header
  • 2 bytes for number of bytes in the bytecode of the function

Local Variable Array

• new data structure: local variable array V

• vload i: pushi-th value onto operand stack

• vstore i: pops operand stack and save its value in i-th position of V

• 

Pools and Dealing with Large Constants

1. having more bytes
2. operating with more arguments
3. replace with expression that evaluates to it
4. write larger constants in the integer pool and provides an instruction to access them (C0VM)

Int Pool

Second segment of C0VM bytecode

• stores number of integers

• and integers

• ildc i: push the i-th integer on the stack (with 2 arguments interpret as together)

• 

Function Pool

Where function stores

• the first function is always main

• 

• argument of function need to be on stack

• return pushed back into stack, replacing the arguments

• 

• 

Other Segments

Writing the Code

Given Data Structure

• notice that we use fixed size int (we don't want C compiler to modify int length)

•