Introduction

File Name Conventions

Object files produced by the linker are, by convention, appended with a suffix of .preout to distinguish them from other file types.

Object File Format

The linker produces executable object files which may be run on the simulator The executable object file is split into a number of sections as displayed in the diagram below :

• Header.
• Instruction Code.
• Constant Pool.

Instruction encoding

The instructions are encoded in binary form for the machine in the output of the linker. There are two instruction formats:

Type 1 encoding

Instructions encoded in type 1 format are generally register-register, and load/store operations that do not require large constants. The fields in the encoding are as follows:

Filed name	Bits	Contents
Opcode	0-5	Instruction opcode
Reg1	6-11	Instruction operand 1 (typically denotes register)
Reg2	12-17	Instruction operand 2 (typically denotes register)
Reg3	18-23	Instruction operand 3 (typically denotes register)
Partition	24-31	Instruction operand 3 (denotes register)

Instructions whose parameters are not register based (for example CREPAR) use a log₂ encoding for constants in the Reg_i fields.

The instructions encoded in type 1 format are:

MOV, LOAD, STORE, LOADIDX, STOREIDX, ADD, SUB, MUL, DIV, MOD, LS, RS, RSU, AND, OR, XOR, NOT, LT, GT, EQ, CREPAR, DELPAR, INVPAR, SETIPAR, UPDPAR

Type 2 encoding

Branch instructions, instructions using large constants, and other miscellaneous instructions not requiring a partition parameter are encoded using the type 2 format. The fields in the encoding are as follows:

Filed name	Bits	Contents
Opcode	0-5	Instruction opcode
Reg	6-11	Instruction operand 1 (typically denotes register)
Reg2	12-31	Instruction operand 2 (typically encodes 20 bit signed constant)

The instructions encoded in type 2 format are:

LOADC, JUMP, JUMPZ, JUMPNZ, JUMPR, HALT, NOP, TRAP

Executable header

The header of the executable object file contains the bootstrap for the user application. Here the initial state of the machine is created for entry point proper into the user application. The header also contains the initial partition creation code required by the applciation. A bootstrap header looks like

  CREPAR    0, i, 1
  SETIPAR   0
  CREPAR    1, c, 1
  CREPAR    2, t, 1
  CREPAR    3, p1size, p1stride
  CREPAR    ...
  CREPAR    n+2, pnsize, pnstride
  LOADC     31, nnn
  LOADC     30, 0
  STORE     30, 31, 2
  LOADC     31, nnn-2
  LOADC     31, nnn-1
  JUMP      main
  HALT
  

Symbol	Meaning
i	Number of cache lines for instruction cache partition
c	Number of cache lines for scalar cache partition
t	Number of cache lines for stack cache partition
pi	User partition i
size	The number of lines required for a user partition
stride	The stride required for a user partition
n	The number of user partitions
nnn	Application stack pointer start
main	Relative address of aplication entry point

We can pattern match against this header style to spot linker generated executable files -- for use with debuggers etc.

Instruction Code

The instruction code is a block of encoded instructions which represent the code for the user application, including user object files generated by the assembler and precompiled library object files. All relocations are resolved, being data (including the constant pool and user data), partition and branching.

Constant Pool

The amalgamated constant pool appears after the instruction segment in the executable object file. This is the collection of constant pools of each object file specified no the command line.