Create efficient, speedy code with the Diab XE C/C++ Compiler Suite for Xilinx FPGAs.

All compilers share many common features. However, compilers designed specifically for embedded systems have several unique requirements:

• Flexibility and control
• Strict adherence to standards
• Industry-leading optimization technology.

In this article, we will introduce the Diab XE (Xilinx Edition) embedded compiler from Wind River and explain how it can help you develop a superior embedded processing system with Xilinx Virtex-II Pro™ FPGAs and IBM PowerPC™ (PPC)-enabled devices.

Adherence to Standards
Standards are critical for portability and interoperability with other tools. The Diab XE compiler uses the latest version of the EDG (Edison Design Group) C++ front end and is 100% compliant with American National Standards Institute (ANSI) C++ standards. The Diab XE compiler also uses run-time libraries from Dinkumware™ Ltd., which provides 100% ANSI compliance for runtime libraries and the Standard Template Library (STL).

For applications where code size is critical, the Diab XE compiler provides an abridged version of the C++ run-time library. This version of the library is also 100% standards-complaint, but it is built with exception processing disabled, thereby generating much smaller applications. Tests have shown this library to result in code sizes as much as 38% smaller.

Another aspect of standards compliance is output formats. The Diab XE compiler maintains strict compliance with the various embedded application binary interfaces (EABIs). Wind River also works with embedded processor vendors to stay current and is actively involved with the development of EABIs.

Adherence to standards leads to easier porting and ensures interoperability with other development and maintenance tools.

Modular Architecture
As shown in Figure 1, the Diab XE compiler is based on a modular design comprising three primary modules:

• The front-end: Source code language is parsed into tokens. These tokens are the same regardless of the language used – C, C++, Java, and others.
• The optimization stage: Optimizations are applied to all languages, because the tokens generated by the front-end are the same regardless of language.
• The back-end: Code is generated. This stage is target-specific. Diab XE’s modular design permits the compiler to fully utilize the architecture of the IBM PowerPC 405 microprocessors within Xilinx FPGAs.

Optimization Technology
Optimization falls into two primary areas – front-end and back-end. Front-end optimizations are concerned with the language. Back-end optimizations use architecture-specific code generation to use the design aspects of the particular target processor.

Although almost all compilers do a good job with front-end optimizations, the Diab tool suite distinguishes itself from other compilers with its efficient back-end optimization technology.

The growing hardware complexity of today’s processor designs makes it very difficult to manually fine-tune applications. The Diab XE compiler is specifically engineered to enable you to maximize pipelines, caches, and other architecture-specific capabilities, including the IBM PowerPC 405 cores in Xilinx FPGAs.

The Diab XE compiler can further optimize your application code based on application profile information. No more manual tweaking of your code is required – the Diab XE compiler will do this for you.

Because Wind River Systems owns and controls the technology that drives the Diab XE compiler, they can advance the technology on their own schedule without requiring approval from an outside source.

Flexibility and Control
Embedded systems typically have many design constraints – limited memory, limited performance. To meet the demands imposed by these constraints, an effective embedded compiler must provide a wide range of flexibility and control, with features such as:

• Addressing modes, which can have a direct impact on both performance and code size. The Diab XE compiler uses intelligent default settings, but it provides options to give you complete control. Separate options also exist for specifying both code and data modes.
• Support for position-independent code and data (PIC/PID).
• Structure and bit-field packing. Here too, intelligent defaults are used, but with options for complete control over how structures are packed and how structure members are aligned. You can also specify the type (short, int, long) used for bit-fields.
• Function inlining, which can result in significant performance gains. Although all compiler vendors make use of this technique, only the Diab XE compiler enables you to specify the precise size of the functions to be inlined, giving you control over the memory footprint.
• Loop unrolling. This is similar to function inlining, because it can result in significant performance gains. Most compiler vendors use loop unrolling. However, only the Diab XE compiler gives you the ability to control both the size of loops to be unrolled and the number of times to unroll them.
• Embedded assembly code. The Diab XE compiler provides two methods: a relatively simple string-based assembly function and a powerful macro-based function. The string-based approach is ideal for writing values to a register. The macro approach provides total flexibility and control over which registers are used for the macros.
• Diab XE’s robust linker command files provide the controls for you to precisely place all program elements in the correct memory locations. Because embedded applications often have strict requirements for locating variables, functions, and data in memory, this is an especially important feature when designing systems with multiple memory types and sizes.

• Compiler
• Assembler
• Linker
• Utilities
• Libraries
• ELF/COFF formats.

For example, the GNU GCC C/C++ 2.96 release does not recognize the PowerPC 405 core as a specific target and cannot take advantage of PPC 405-only registers.

When your application needs to be as small and as fast as it can be, the Diab XE compiler will provide you with superior results. The accompanying sidebar “High Performance or Small Footprint?” clearly shows the superiority of the Diab XE C/C++ Compiler 5.6 when compared to the GNU GCC C/C++ 2.96 in four critical parameters:

• Dhrystone performance in milliseconds
• Dhrystone code size
• DES application performance
• DES application size.

The superior flexibility and control afforded by the Diab XE compiler enables developers to fine-tune the compiler’s output to meet and exceed the needs of demanding embedded applications.

The Diab XE C/C++ Compiler Suite is included in the Wind River package, which is available from Xilinx at www.xilinx.com (search for “Diab XE C/C++ Compiler”).

This version is full featured but limited to a fully linked object size of 256 KB – enough room to develop small applications and well suited to Virtex-II Pro FPGAs with embedded PowerPC processors.

Printable PDF version of this article with graphics.  (12/11/03)370 KB