Code optimization by self modification in load-store instruction set machines
Publication Date: 2010-Sep-23
The IP.com Prior Art Database
AbstractDisclosed is a technique for code optimization in load-store instruction set machines. The article aims at achieving optimization by self modification of code. Most instructions that need memory references i.e direct or indirect addressing are replaced with their register or immediate equivalents. The optimization is achieved without the need of an extra register or specialized hardware.
Page 01 of 9
Code optimization by self modification in load -store instruction set machines
Load-store style architecture is common in many machines most notably the x86
machines by Intel and AMD. A very important feature of this kind of architecture is that it
forbids in most cases, transfer of data between memory. Most transfer takes place through
registers. Registers are small, very high speed memory. Instructions get executed faster in
registers than in memory. Registers are however limited in capacity as well as in number.
Code optimization is a technique in which code is altered in a way such that it brings
improvements in some dimension such as CPU usage, memory or speed. There are several
factors that determine code optimization. One of the factors is the number of registers available.
Code optimization is an integral part of compiler optimization and is also important in critical
Self modifying code is a style of coding where code in high-level or machine level form
adapts itself to suit the need of the application and in general brings optimization. Self
modification may take place at a static level before a run or during run i.e on the fly. In assembly
language, self modifying code can exist more freely.
This article deals with load-store architecture in particular. The central idea is to bring
code optimization in such systems both before and at run time by trying to eliminate as much
memory usage as possible. For this, normal assembly level code is transformed into self
While running, this code replaces most code in memory addressing mode (in various
different modes like direct, indirect, indexed etc) into absolute or register addressing modes.
The aim is to enhance speed, as most operations will be performed in the registers than
Consider a very simple assembly code snippet in x86 language for Microsoft Assembler. mov ax, 1234h
mov bx, ax
This code part deals entirely with registers. As a result the operation is quick.
consider the following example.
msg db 'Hello, world!$'
Page 02 of 9
mov ah, 09h
lea dx, msg
Here the string 'Hello, world!' is stored in a memory location named 'db'. The contents of
this location i.e 'Hello, world!' are then loaded into dx register. By replacing the instruction lea dx, msg ...............with its immediate counterpart i.e
lea dx,'Hello, world!$'.............we have faster execution.
This is not complete however. Data in memory may need to be read again and again. In
this case our self modifier adapts accordingly. Every reference to this location is changed to its
However, the scenario gets tougher when we come across data that changes or is
Consider the following example.
.model small .data
Fname db 'my.txt',00h
Handle dw ?
Buf db 80h dup(?) .stack 100h
mov ah, 3dh
Page 03 of 9
mov al, 0
lea dx, Fname int 21h
mov Handle, ax...