Vision Processors Matrox Odyssey EA/XA

Vision processorsMatrox Odyssey eA/XALeading-edge vision processor board.Evolutionary architecture with leading-edge performanceMatrox Odyssey eA/XA is a fourth generation vision processor board that optimallycombines the latest off-the-shelf and custom technologies within an establishedarchitecture to deliver leading-edge performance and value. Designed withdemanding semiconductor inspection, medical imaging, print inspection, surfaceinspection and signal processing applications in mind, the Matrox Odyssey XAis the ideal choice for applications with data acquisition and processing ratesin the order of hundreds of MBytes per second and/or where the PC is heavilyloaded with other system activities.Key featuresx4 PCIe (eA) or PCI-X (XA) long cardG4 PowerPC and proprietary ASICcombine for over 130 BOPS1over 5 GB per second of memory bandwidth512 MB of DDR SDRAM memoryThe premier embedded microprocessor, state-of-the-art proprietary processorand router ASIC, DDR memory, and PCIe /PCI-X connectivity come togetheron the Matrox Odyssey eA/XA to provide unrivaled power for a single visionprocessor board. All this power and flexibility is accessed through an easy-tolearn programming environment compatible with Matrox Imaging’s previousgeneration vision processor and incorporating elaborate image processingand analysis algorithms.State-of-the-art Matrox Oasis ASICintegrated quad-input analog framegrabber acquires up to 640 MB per secondThe Matrox Imaging designed Oasis ASIC is the pivotal component of theup to 1 GB per second of I/O bandwidthto host PCbridge, Links Controller, main memory controller and Pixel Accelerator.available software is sold separately andincludes Matrox Imaging Library (MIL)2 andMatrox Odyssey Developer’s Toolkithost OS support for 32/64-bit Microsoft Windows XP/Vista /7 and 32-bit Linux royalty-free redistribution of MIL and ONLrun-time environments2Matrox Odyssey eA/XA. A high-density chip, the Matrox Oasis integrates a CPU

Matrox Odyssey eA/XAVideo In 1aVideo In 1b32.5 or 40 MHzlow-passfilter2:1 MUXVariableoffset &gain10-bitA/D10/1024 x 10-bitLUTs10/2Clock In (LVDS)/Hsync In (LVDS)/Vsync In (LVDS)/Trigger (TTL)22PSG #1UARTExposure (TTL)G4PowerPCRxTxVideo In 1aDVI ConnectorChannel #1Video In 2aVideo In 3aChannel #2Video In 4aVideo In 1bDVI ConnectorChannel #3Video In 2bVideo In 3bChannel #4Video In 4bAux. In (8)DB-44andDB-9*Aux. Out (8)(MPC7447A)10//10/10/10/VideotoPCI-XBridgeLINK ouplers////512 MBDDRSDRAM64(up to 1 GB/s)PCI-X to PCIe (eA) or PCI-X (XA) Bridge Aux. In (4)128(up to 5.3 GB/s)LINK 0161664(up to 1.3 GB/s)TM* Present on a separate bracketx4 PCIe (eA)TM(up to 1 GB/sec)Pixel AcceleratorThe Pixel Accelerator (PA) is a parallel processor core, whichconsiderably accelerates neighborhood, point-to-point and LUTmapping operations. It consists of an array of 64 processingelements all working in parallel. Each processing element has amultiply-accumulate (MAC) unit and an arithmetic-logic unit (ALU).The MAC unit is capable of performing a single 16-bit by 16-bit,two 8-bit by 16-bit or four 8-bit by 8-bit multiplies with 40-bitaccumulation per cycle for convolution operations. The 40-bitaccumulator guarantees no overflow situation for a 16 by 16kernel with 16-bit coefficients and data. In addition, the PAarchitecture allows symmetrical kernels to be processed fourtimes faster. The MAC unit is also able to perform up to fourminimum or maximum operations per cycle for grayscalemorphology operations.The ALU can execute a wide variety of arithmetic and logicaloperations. It can be programmed to execute a sequence of 256instructions per pixel at one instruction per cycle reducing theamount of memory accesses and further accelerating memoryI/O-bound sequences. The PA can accept up to four sourcebuffers3 and output to four destination buffers allowing severaloperations to be performed at once or in a single pass (i.e.,four images can be averaged in one pass). Operating at acore frequency of 167 MHz enables the PA to carry out up to100 BOPS1 (i.e., process over two billion pixels per second).25V/3.3V PCI/PCI-X (XA) 64-bit (up to 1 GB/sec)Memory controllerThe Matrox Oasis includes a very efficient main memorycontroller for managing the 128-bit wide interface to DDRSDRAM memory. Operating at 167 MHz, the DDR SDRAMmemory and controller combine to deliver a memory bandwidthin excess of 5 GB per second. Such ample memory bandwidthallows the Odyssey eA/XA to comfortably handle demandingvideo I/O while maintaining PA performance even for memoryI/O-bound operations.Links ControllerThe Links Controller (LINX) is the router that manages all datamovement inside and outside the processing node, which consistsof the PA, CPU and main memory. It can handle several concurrent video and message streams.Video streams are used to transfer image data from the integratedframe grabber to the processing node and the processingnode to the host PC including display. The video streams haveadjustable priority levels, either above or below messagestreams. Video streams can be subject to various formattingoperations including plane separation on input and merging onoutput, input cropping, input and output sub-sampling (1 to 16),and independent control of horizontal and vertical scanningdirection. The latter is particularly useful for reconstructing aproper image from a camera whose readout requires multipletaps, each with different scanning directions.

Links Controller (cont.)Message streams are for all types of inter-processorcommunications. The LINX handles message streams betweenthe processing node and the host PC independently of videostreams. Message passing relies on hardware-assistedmechanisms for low overheads and real-time operation.Together, the above capabilities off-load the CPU and PAfrom data management tasks so they can focus on imageprocessing tasks.Best-of-breed freescale G4 PowerPC microprocessorThe CPU that controls activities on the Matrox Odyssey eA/XA and performs operations not accelerated by the PA is thefreescale G4 PowerPC microprocessor. The G4 combinesthe best features of a general purpose CPU and a DSP, andprovides top performance at a given clock rate. The G4 isalso backed by freescale ’s solid migration path for increasedperformance, all the while maintaining code compatibility.The G4 incorporates a powerful 32-bit superscalar RISC andAltiVec technology’s 128-bit vector execution unit. 512 KBof internal L2 cache helps sustain maximum processor performance. A 64-bit MPX bus offers efficient access to main memory and provides a sustained bandwidth close to its theoreticalmaximum of 1.3 GB per second.AltiVec technology is specifically designed tomeet the heavy computational requirements of applicationssuch as video and image processing. This technology consists ofa high-performance parallel processing engine for vector data.It uses the SIMD (single instruction, multiple data) model toprocess, in parallel, up to 16 pixels per cycle. It delivers a peakprocessing power of 16 billion 8-bit MACs per second or 8 billion 32-bit floating point operations per second when running at1 GHz. Additionally, AltiVec technology operates concurrentlywith other execution units within the G4.Choice of high-performance host bus interfacesFour lane (x4) PCIe and PCI-X are the interfaces used toconnect to the host PC on the Matrox Odyssey eA and MatroxOdyssey XA boards respectively. PCIe is the follow-on toconventional PCI and PCI-X . Version 1.x of PCIe operatesat 2.5 GHz to deliver a peak bandwidth of 1GB/sec over a x4implementation. PCI-X is a high-performance backwardscompatible enhancement to conventional PCI. Version 1.0a ofPCI-X specifies a 64-bit physical connection running at speedsof up to 133 MHz resulting in a peak bandwidth of up to 1 GBper second.Flash EEPROM for full autonomyMatrox Odyssey eA/XA has a flash EEPROM that stores the G4PowerPC boot sequence, system initialization parameters anda debugging utility. It can also be used to store the operatingsystem and application program in order to implement a trulyautonomous system from power-up.Integrated analog frame grabberThe integrated analog frame grabber has four completelyindependent inputs, each capable of high frequency and highfidelity video capture. In addition to being able to simultaneouslyacquire from up to four single-tap video sources, the inputscan be combined to simultaneously acquire from two dual-tapvideo sources or one RGB video source. The inputs can also becombined to simultaneously acquire from two video sources atdouble the nominal acquisition rate. The analog frame grabberalso includes an internal video generator for troubleshootinginstallation and operation.All-encompassing programming environmentThe Matrox Odyssey eA/XA offers the developer a choice ofapplication programming interfaces (APIs) and programmingmodels. Application developers can program the board usingthe Matrox Imaging Library or Matrox Odyssey Native Library,each with royalty-free run-time environments2. The application program can run on the host PC to remotely control theMatrox Odyssey eA/XA (i.e., options A and B) or it can be madeto run entirely on the Matrox Odyssey eA/XA (i.e., options C andD).Matrox Imaging Library (MIL)MIL is a complete and easy-to-use programming library forimage capture, processing, analysis, display and archivingthat supports the complete line of existing and future MatroxImaging hardware. MIL’s field-proven processing and analysisfunctions include point-to-point, statistics, filtering, morphology,geometric transformations, FFT, segmentation, pattern recognition, blob analysis, edge extraction and analysis, measurement,charater recognition, 1D and 2D code reading, calibration, andJPEG/JPEG2000 image compression. These functions run onthe Matrox Odyssey eA/XA and are fully optimized for the G4PowerPC and PA.MIL for the Matrox Odyssey eA/XA is carefully tuned to exhibitlow function call overheads, maximizing application efficiencyand performance. MIL provides transparent image displaymanagement with automatic tracking and updating of imagedisplay windows at live video rates. MIL also allows for imagedisplay in a user-specified window. In addition, MIL supportslive display of multiple video streams using multiple independentwindows or a single mosaic window. Moreover, MIL providesnon-destructive graphics overlay, suppression of tearing artifacts and filling the display area at live video rates. All of thesefeatures are performed with little or no host CPU interventionwhen using the appropriate graphics hardware. Refer to theMIL datasheet for more information.Odyssey Native Library (ONL)ONL is an easy-to-use programming library specific to theMatrox Odyssey family of vision processor boards. The programming interface is compatible with the Matrox Genesis family ofvision processor boards. Aside from video capture and systemcontrol, ONL includes functions for image processing, normalized grayscale correlation-based pattern recognition, blobanalysis and JPEG compression. These functions are fully optimized to exploit the power of the G4 PowerPC and PA. ONLfunctions exhibit the lowest call overheads.3