Just-In-Time compiler concept / demo for Intel x86_64

A small bytecode Just-In-Time compiler / binary translator with a custom assember library and a virtual machine. The purpose of this project is to demonstrate the orders of magnitude faster execution speed of JIT-applied code paths against traditional virtual machines.

Images

Languages

• C

Dependencies

• none

Setup

• Clone or download the repository
• Compile Release x64 with MSVC
• Run x64/Release/JitDemo.exe

Bytecode and instruction set

The bytecode VM supports 4 registers (r0 to r3) and 5 different instructions to program with. Programs made for the VM can also be compiled to x86_64 machine code with jit_compile, and then executed as native functions.

// ADD r, r
// adds the second operand (register) the first operand (register), and stores the result in the first operand
#define INTERP_ADD		1

// SUB r, r
// subtracts the second operand from the first operand, and stores the result in the first operand
#define INTERP_SUB		2

// LOAD r, long
// loads a constant in the second operand to the first operand
#define INTERP_LOAD		3

// LOOP r, rel_addr_backwards
// Checks if the value in the first operand is 0, if not, will jump back the second operand amount of bytes relative to the instruction
#define INTERP_LOOP		4

// RET r
// Returns the value in the first operand
#define INTERP_RET		5

List of registers:

#define INTERP_R0 0
#define INTERP_R1 1
#define INTERP_R2 2
#define INTERP_R3 3

Programming

Programs can be written in main.c, executed in the VM and the native processor, and the execution time measured in seconds. Creating programs is done in bytecode.

Disassembler

The repository also includes a disassembler in bc_disassembler.c to help visualize instructions written in bytecode.