Application Note: 202

MDK-ARM Compiler Optimizations

Getting the Best Optimized Code for your Embedded Application

Abstract
This document examines the ARM Compilation Tools, as used inside the Keil MDK-ARM (Microcontroller
Development Kit), and how to use them to optimize your code for best performance or smallest code-size.

Contents
ARM Compilation Tools ......................................................................................................................................................... 2
Compiler Options for Embedded Applications ........................................................................................................................ 2
Optimizing for Smallest Code Size.......................................................................................................................................... 5
Compile the Measure example without any optimizations .................................................................................................. 5
Optimize the Measure example for Size .............................................................................................................................. 6
Optimizing for Best Performance ............................................................................................................................................ 7
Run the Dhrystone benchmark without any optimizations .................................................................................................. 7
Optimize the Dhrystone example for Performance .............................................................................................................. 8
Summary .................................................................................................................................................................................. 9

Revision History
• August 2009: Initial Version

Information in this file, the accompany manuals, and software is

Copyright (c) ARM. All rights reserved.

Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. www.keil.com

August 2009 Revision 1.0
MDK Compiler Optimizations

ARM Compilation Tools

The ARM Compilation Tools are the only compilation tools co-developed with the ARM processors, and specifically
designed to optimally support the ARM architecture. They are a result of 20 years of development, and are recognized as the
industry-leading C and C++ compilation tools for the ARM, Thumb, and Thumb-2 instructions sets.
The ARM Compilation tools consist of:
• The ARM Compiler, which enables you to compile C and C++ code. It is an optimizing compiler, and features
command-line options to enable you to control the level of optimization
• Linker and Utilities, which assign addresses and lay out sections of code to form a final image
• A selection of libraries, including the ISO standard C libraries, and the MicroLIB C library which is optimized for
embedded applications
• Assembler, which generates machine code instructions from ARM, Thumb or Thumb-2 assembly-level source code

Compiler Options for Embedded Applications

The ARM Compilation Tools include a number of compiler optimizations to help you best target your code for your chosen
microcontroller device and application area.
They can be accessed from within µVision by clicking on Project – Options for Target.
The options described this document can be found on the Target and C/C++ tabs of the Options for Targets dialog.

Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. 2 www.keil.com

August 2009 Revision 1.0
MDK Compiler Optimizations

• Cross-Module Optimization takes information from a prior build and uses it to place UNUSED functions into their
own ELF section in the corresponding object file. This option is also known as Linker Feedback, and requires you to
build your application twice to take advantage of it for reduced code size.
Cross-Module Optimization has been shown to reduce code size, by removing unused functions from your application. It
can also improve the performance of your application, by allowing modules to share inline code.

• The MicroLIB C library has been optimized to reduce the size of embedded applications. It is a subset of the ISO
standard C runtime library, and offers a tradeoff between functionality and code size. Some of the standard C library
functions such as memcpy() are slower, while some features of the default library are not supported. Unsupported
features include:
o Operating system functions e.g. abort(), exit(), time(), system(), getenv(),
o Wide character and multi-byte support e.g. mbtowc(), wctomb()
o The stdio file I/O function, with the exception of stdin, stdout and stderr
o Position-independent and thread-safe code
Use the MicroLIB C library for applications where overall performance can be traded off against the need to reduce code
size and memory cost.

• Link-Time Code Generation instructs the compiler to create objects in an intermediate format so that the linker can
perform further code optimizations. This gives the code generator visibility into cross-file dependencies of all objects
simultaneously, allowing it to apply a higher level of optimizations. Link-time code generation can reduce code size, and
allow your application to run faster.

• Optimization Levels can also be adjusted. The different levels of optimization allow you to trade off between the level
of debug information available in the compiled code, and the performance of the code. The following optimization levels
are available:
o -O0 applies minimum optimizations.
Most optimizations are switched off, and the code generated has the best debug view.
o -O1 applies restricted optimization.
For example, unused inline functions and unused static functions are removed. At this level of optimization, the
compiler also applies automatic optimizations such as removing redundant code and re-ordering instructions so
as to avoid an interlock situation. The code generated is reasonably optimized, with a good debug view.
o -O2 applies high optimization (This is the default setting).
Optimizations applied at this level take advantage of ARM’s in-depth knowledge of the processor architecture,
to exploit processor-specific behavior of the given target. It generates well optimized code, but with limited
debug view.
o -O3 applies the most aggressive optimization.
The optimization is in accordance with the user’s –Ospace/-Otime choice. By default, multi-file compilation is
enabled, which leads to a longer compile time, but gives the highest levels of optimization.

Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. 3 www.keil.com

August 2009 Revision 1.0
MDK Compiler Optimizations

• The Optimize for Time checkbox causes the compiler to optimize with a greater focus on achieving the best
performance when checked (-Otime) or the smallest code size when unchecked (-Ospace).
Unchecking Optimize for Time selects the –Ospace option which instructs the compiler to perform optimizations to
reduce the image size at the expense of a possible increase in execution time. For example, using out-of-line function
calls instead of inline code for large structure copies. This is the default option. When running the compiler from the
command line, this option is invoked using ‘-Ospace’
Checking Optimize for Time selects the –Otime option which instructs the compiler to optimize the code for the fastest
execution time, at the risk of an increase in the image size. It is recommended that you compile the time-critical parts of
your code with –Otime, and the rest using the –Ospace directive.

• Split Load and Store Multiples instructs the compiler to split LDM and STM instructions involving a large number of
registers into a series of loads/stores of fewer multiple registers. This means that an LDM of 16 registers can be split into
4 separate LDMs of 4 registers each. This option helps to reduce the interrupt latency on ARM systems which do not
have a cache or write buffer, and systems which use zero-wait state 32-bit memory.
For example, the ARM7 and ARM9 processors take can only take an exception on an instruction boundary. If an
exception occurs at the start of an LDM of 16 registers in a cacheless ARM7/ARM9 system, the system will finish
making 16 accesses to memory before taking the exception. Depending on the memory arbitration system, this can result
in a very high interrupt latency. Breaking the LDM into 4 individual LDMs for 4 registers means that the processor will
take the exception after loading a maximum of 4 registers, thereby greatly reducing the interrupt latency.
Selecting this option improves the overall performance of the system.

• The One ELF Section per Function option tells the compiler to put all functions into their own individual ELF
sections. This allows the linker to remove unused functions.
An ELF code section typically contains the code for a number of functions. The linker is normally only able to remove
unused ELF sections, not unused functions. An ELF section can only be removed if all its contents are unused.
Therefore, splitting each function into its own ELF section allows the compiler to easily identify which ones are unused,
and remove them.
Selecting this option increases the time required to compile your code, but results in improved performance.

The combination of options applied will depend on your optimization goal –whether you are optimizing for smallest code
size, or best performance.
The next section illustrates the best optimization options for each of these goals.

Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. 4 www.keil.com

August 2009 Revision 1.0
MDK Compiler Optimizations

Optimizing for Smallest Code Size

To optimize your code for the smallest size, the best options to apply are:
• The MicroLIB C library
• Cross-module optimization
• Optimization level 2 (-O2)

Compile the Measure example without any optimizations

The Measure example uses analog and digital inputs to simulate a data logger.

File -- Open Project

C:\Keil\ARM\Boards\Keil\MCBSTM32\Measure\Measure.uv2

Click the Options for Target button

In the Target tab:

• Uncheck Cross-Module Optimization
• Uncheck Use MicroLIB
• Uncheck Use Link-Time Code Generation

In the C/C++ tab:

• Set Optimization Level to Zero

Then click OK to save your changes.

Project – Build target

Without any compiler optimizations applied, the initial code size is 13,656 Bytes.
Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. 5 www.keil.com

August 2009 Revision 1.0
MDK Compiler Optimizations

Optimize the Measure example for Size

Apply the compiler optimizations in turn, and re-compile each time to see their effect in reducing the code size for the
example.
• Options for Target – Target tab: Use the MicroLIB C library
• Options for Target – Target tab: Use cross-module optimization - Remember to compile twice
• Options for Target – C/C++ tab: Enable Optimization level 2 (-O2)

Optimization Applied Compile Size Size Reduction Improvement

MicroLIB C library 8,960 Bytes 4,696 Bytes 34% smaller
Cross-Module Compilation 13,500 Bytes 156 Bytes 1.1% smaller
Optimization level –O2 12,936 Bytes 720 Bytes 5.3% smaller
All 3 optimization options 8,116 Bytes 5,540 Bytes 40.6% smaller

Applying all the optimizations will reduce the code size down to 8,116 Bytes.

The fully optimized code is 5,540 Bytes smaller, a total code size reduction of 40.6%

Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. 6 www.keil.com

August 2009 Revision 1.0
MDK Compiler Optimizations

Optimizing for Best Performance

To optimize your code for performance, the best options to apply are:
• Cross-module optimization
• Optimization level 3 (-O3)
• Optimize for time

Run the Dhrystone benchmark without any optimizations

The Dhrystone benchmark is used to measure and compare the performance of different computers, or the efficiency of the
code generated for the same computer by different compilers.

File – Open Project

C:\Keil\ARM\Examples\DHRY\DHRY.uv2

Click the Options for Target button

Turn off optimization settings in the Target and C/C++ tabs, then click OK

Project – Build target

Enter Debug mode

View – Serial Windows – UART #1

Open the UART #1 window

View – Analysis Windows – Performance Analyzer

Open the Performance Analyzer

Debug – Run
Start running the application

When prompted:
Enter 50000 in the UART#1 window and press Enter

Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. 7 www.keil.com

August 2009 Revision 1.0
MDK Compiler Optimizations

In the Performance Analyzer window, note that

• The drhy_1 loop took 2.829s
• The dhry_2 took 2.014s
In the UART #1 window, note that
• It took 138.0 ms for 1 run through Dhrystone
• The application is executing 7246.4 Dhrystones per second

Optimize the Dhrystone example for Performance

Re-compile the example with all three of the following optimizations applied:
• Options for Target – Target tab: Cross-module optimization – Remember to compile twice
• Options for Target – C/C++ tab: Optimization level 3 (-O3)
• Options for Target – C/C++ tab: Optimize for Time

Re-run the application, and examine the performance.

Measurement Without optimizations With Optimizations Improvement

dhry_1 2.829s 1.695s 40.1% faster
dhry_2 2.014s 1.011s 49.8% faster
Microseconds for 1 run 138.0 70 49.3% faster
through Dhrystone
Dhrystones per second 7246.4 14,285.7 97.1% more

The fully optimized code achieves approximately 2x the performance of the un-optimized code.

Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. 8 www.keil.com

August 2009 Revision 1.0
MDK Compiler Optimizations

Summary
The ARM Compilation Tools offer a range of options to apply when compiling your code. These options can be combined to
optimize your code for best performance, for smallest code size, or for any performance point between these two extremes, to
best suit your targeted microcontroller device and market.
When optimizing your code, MDK-ARM makes it easy and convenient to measure the effect of the different optimization
settings on your application. The code size is clearly displayed after compilation, and a range of analysis tools such as the
Performance Analyzer enable you to measure performance.
The optimization options in the ARM Compilation Tools, together with the easy-to-use analysis tools in MDK-ARM, help
you to easily optimize your application to meet your specific requirements.

Copyright © 2009 ARM Ltd. All rights reserved

MDK Compiler Optimizations. 9 www.keil.com

August 2009 Revision 1.0