High-performance vector and matrix processing is a central requirement for applications like radar detection, video analytics, medical imaging, and factory automation. In the past, developers often used specialized digital signal processors (DSPs) or field-programmable gate arrays (FPGAs) in these applications. Today, multi-core Intel® architecture (IA) processors with Intel® Advanced Vector Extensions (Intel® AVX) offer a more flexible alternative.


Intel AVX, an advanced form of Intel® Streaming SIMD Extensions (Intel® SSE), doubles peak floating-point performance by widening the floating-point data path from 128 bits to 256 bits. When combined with the multi-core performance of processors such as the 2nd generation Intel® Core processors – which offer up to four cores at 3.4 GHz – these instructions provide impressive signal processing performance. For a detailed look into the new instructions, see Intel® Advanced Vector Extensions: Next-Generation Vector Processing.


Developing and porting applications with Intel AVX is made easier by broad support from the Intel® Embedded Alliance. This organization’s 160-plus members collaborate closely with Intel to create optimized hardware, software, tools, and services that give OEMs a head start on their designs. Table 1 shows some of Intel’s and Alliance members’ tool and software support for Intel AVX.


OS and Hypervisor Support

• Various versions of Microsoft* Windows*, including   Windows* XP Embedded and Windows* Embedded Standard 7
• QNX* Neutrino*
• Wind River VxWorks*
• LynuxWorks* LynxSecure 5.0
• Linux*

Tool Support

• Microsoft Visual Studio 2010
• Intel® C++ Compiler
• Intel® VTune™   Amplifier XE
• GNU tools including GCC, Binutils, and GDB
Yasm Modular   Assembler
• Netwide Assembler (NASM)
AltiVec.h header   file (PowerPC* to Intel® AVX conversion)

Vector Signal and Image Processing Libraries (VSIPL)

• GE Intelligent Platforms AXISLib-VSIPL
• Curtiss Wright Controls Embedded Computing Continuum Vector
• Mercury Computer Systems MultiCore   Plus MathPack
• CodeSourcery VSIPL++
• N.A. Software VSIPL
• RunTime Computing Solutions VSI/Pro

Other Software Libraries and Components

• Intel® Integrated Performance Primitives (Intel® IPP)
• Intel® Math Kernel Library (Intel® MKL)
• Intel® Thread Building Block
• Intel® Array Building Blocks 1.0 (beta)
• GNU C Library (glibc) 2.11

Table 1. Partial list of the software support for Intel® Advanced Vector Extensions.


Several  of the companies listed in the table above also provide signal processing hardware. As noted in the blog Intel AVX-enabled processors excel in sensing and analytics, much of this hardware targets military and aerospace markets. For a deeper dive into the hardware targeting mil/aero – and for the applications of Intel AVX in this space – check out the article SWaP-able Solutions Take Off in Mil/Aero.


Other applications that can leverage Intel AVX include digital signage and surveillance. Video analytics plays a key role in both markets. In the case of digital signage, anonymous video analytics (AVA) can be used to analyze the viewing audience and tailor signage content accordingly – see the blog Audience measurement optimizes digital signage applications for an overview.


Surveillance applications have opportunities for a broader range of analytics, such as left-package detection, that can improve security and reduce the burden on operators. The challenge here is the fact that many organizations have already installed analog camera systems. As the article

Upgrade Analog Surveillance with HD and Analytics explains, this challenge can be overcome by deploying hybrid surveillance systems.


A similar set of opportunities and challenges is found in the factory automation market. As noted in the blog Advanced CPU Architectures Boost Production Testing, high-speed digital signal processing and video analysis functions are commonly found in production test systems. Implementing these functions with Intel AVX can simplify your designs and cut the bill of materials (BoM). More broadly, signal processing tasks can be found throughout the factory for sensing and control. The article Replace Real-Time Hardware with Multi-Core Software has a great perspective on how to HMI and control tasks using IA processors (see Figure 2).


Figure 2. Multi-core IA processors can handle both HMI and control tasks.


security_analytics.pngThe links I’ve listed here only scratch the surface of what the Alliance has to offer.  To learn more about advanced signal processing, see intel.com/go/embedded-analytics.



GE Intelligent Platforms, Microsoft Corporation, QNX Software Systems, Ltd., and Wind River Systems are Associate members of the Intel® Embedded Alliance. Curtiss Wright Controls Embedded Computing and LynuxWorks are Affiliate members of the Alliance. CodeSourcery and Mercury Computer Systems are General members of the Alliance.


Kenton Williston

Roving Reporter (Intel Contractor), Intel® Embedded Alliance

Editor-In-Chief, Embedded Innovator magazine

Follow me on Twitter: @kentonwilliston