Embedded Systems Design Europe - February 2008 - (Page 34)

feature down because it increases memory use. In these cases, you may want to turn your functions or loops into MEX-ﬁles. MEX-ﬁles, like all MATLAB core built-in functions, are compiled ﬁles. They can therefore be extremely fast. Once developed, functions implemented as MEX-ﬁles are also very easy to use. They can be called directly from MATLAB, just like any other function. Within MATLAB, you can create MEX-ﬁles from C or Fortran source code by using the mex command. The C or Fortran source code has to be manually written using the MEX application programming interface (MEX API). MEX API functions are directly available to read input arguments from MATLAB, write outputs arguments back to MATLAB, or access global variables. The MEX API also provides hooks that let you call back to MATLAB to make use of visualization functions or leverage computation functions that do not need to be compiled. The MEX-ﬁle guide from The Mathworks provides more information on how to write your own MEX-ﬁles. Needless to say, the process of writing these MEX-ﬁles by hand is both time-consuming and error-prone. Several tasks must happen when translating MATLAB to C or Fortran. First, function cost = costFuncMAD(currentBlk,refBlk, n) % Non-vectorized code % err = 0; % for i = 1:n % for j = 1:n % err = err + abs((currentBlk(i,j) - refBlk(i,j))); % end %end % Vectorized code err = sum(abs(currentBlk(:) - refBlk(:))); cost = err / (n*n); you must implement the algorithm functionality, which includes MATLAB functions, by writing new code or calling pre-existing functions and libraries. While doing this, you need to explicitly allocate storage by declaring internal variables and making proper use of memory management routines. Finally you need to interface your code with the rest of your MATLAB model using the MEX API. Debugging MEX-ﬁles can also be difﬁcult as none of the MATLAB analysis and visualization functions are available from the C or Fortran debugger. The C and Fortran languages are also less forgiving than the MATLAB language. Out-of-bound array accesses or failure to properly allocate or free memory may have catastrophic consequences on your application, including unexpected results, segmentation violations, or “out of memory” errors. A more convenient path is to automatically generate MEX-ﬁles directly from MATLAB. This allows you to fully utilize the MATLAB environment to develop your algorithm and optimize it for speed. Then, at any time, you may compile your code into a fast MEX-ﬁle. Code changes applied to your algorithm are automatically incorporated in the MEX-ﬁle, creating a fast, reliable, and reproducible path from MATLAB to compiled code. Using advanced optimizing compiler technology, Agility tools can automatically compile your MATLAB code into fast MEX-ﬁles. Internally, your MATLAB code is translated into an intermediary C code. The ability to generate efﬁcient C code from MATLAB may be used to create not only MEXﬁles for acceleration but also standalone platform-independent ANSI C models by the push of a button. On the block-matching algorithm presented above, generating MEX-ﬁles gives by far the biggest speed-up. Block-matching algorithm Run time MATLAB Original 46s MATLAB Vectorized 26s MATLAB with Agility MEX-file 2.7s The screen shot left shows how an automatically-generated MEX-ﬁle sped up the block-matching algorithm by almost 10x. Luc Semeria (luc@agilityds.com) is product manager at Agility Design Solutions (formally Catalytic Inc.) 34 JANUARY – FEBRUARY 2008 | embedded systems design europe | www.embedded.com/europe 032-033-034_ESDE.indd 34 6/02/08 16:53:47 http://www.embedded.com/europe

Table of Contents Feed for the Digital Edition of Embedded Systems Design Europe - February 2008

Embedded Systems Design Europe - February 2008 Contents ARTEMIS and ENIAC Get Parlimentary Approval Product Teardown Videos Come On Screen Esterel and Abslnt Integrate Products Microsoft Opens Windows to Networked Embedded Applications Trango Embeds Virtualization Tool in Cavium's Multicore CPUs MindTree - ADI Develop Security DVR Platform NXP Extends Deal with ARM to Cover MCUs Automotive and Embedded to Dominate DATE 08 ZigBee Spec Gets Smart On Energy Updated Card Spec Provides for Power-On Boot The Basics of Embedded Multitasking On a PIC Cover Feature: The Art of FPGA Construction Is Symmetric Multiprocessing For You? Accelerating MATLAB Using MEX-Files ARM Provides the Microcontroller Solution Embedded World Advertising Contracts

Embedded Systems Design Europe - February 2008

For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.