Vector processing accelerates QorIQ

Vector processing accelerates QorIQ

R. Colin Johnson

9/27/2010 3:01 AM EDT

PORTLAND, Ore.—Vector processing accelerates tasks even more than multiple-cores, that is if the same instructions are being executed on multiple parallel data streams—called single-instruction-multiple-data (SIMD).

SIMD is traditionally used for parallel data tasks such as dimming all the pixels in an image, but new demands from multiple 4G users are now enabling wireless basestations to utilize SIMD. To meet the need, Freescale Semiconductor Inc. is upgrading its venerable AltiVec vector processing unit for its QorIQ family of Power Architecture-based communications processors.
“Customers using QorIQ to deal with all the data being processed by 4G basestations are probably telling Freescale that the Altivec SIMD processor could be very useful at performing the same processing task [single-instruction] on multiple users [multiple-data],” said Linley Gwennap, principal analyst at The Linley Group (Mountain View, Calif.) “Intel has been promoting their SSE [streaming SIMD extensions] and now AVX [advanced vector extensions] for their next generation Sandy Bridge, because they too are finding more and more uses for vector processing for communications tasks.”
Freescale already sells legacy AltiVec processors to industry, military and aerospace customers performing traditional tasks such as image processing, pattern recognition and forward kinematics to position robotic arms as well as for emerging applications such as smart analytics on multiple surveillance cameras. But by adding AltiVec to its leading edge QorIQ processors, Freescale hopes to expand further its penetration of the basestation processor market.
“We have taken a proven technology, enhanced it, and moved it to the multicore QorIQ family of processors,” said Glenn Beck, marketing manager for aerospace, defense and single board computing within Freescale’s Networking Processor Division.
“We see a broad spectrum of applications that are now needing vector-caliber performance, ranging from networking and telecom to the classic ones in aerospace and defense.”
The original AltiVec specification was a part of an alliance between Apple, IBM and Freescale (formerly the Semiconductor Products Sector of Motorola), but AltiVec is still widely used by many Power Archtecture processors. For instance, the Cell Broadband Engine used by IBM supercomputers and Sony’s PlayStation 3 chose AltiVec for its flexible programmable architecture. AltiVec varies slightly among manufacturers, but usually offers a 128-bit vector processing unti operating concurrently with the existing integer and floating point units in a core. With an AltiVec inside each QorIQ core, an additional eight parallel floating-point and integer processing units execute vector algorithms using 180 instructions.
“Our customers are seeing the need to offload tasks into Altivec, which unlike most SIMD processors is programmable,” said Beck. “Anytime you can offload a core doing a job, you’re accelerating your applications, and AltiVec can take large data width and signal processing intensive algorithms and offload them from the core’s own floating point and integer units.”

AltiVec programmable vector processing architecture incurs no penalty for mixed floating-point and integer operands since it has execution units for both.

Freescale claims that the Embedded Microprocessor Benchmark Consortium has demonstrated a 10-times improvement in performance for some applications after adopting the AltiVec technology.

Freescale provides

downloadable ‘C’ libraries

for most vector processing tasks.



發佈留言必須填寫的電子郵件地址不會公開。 必填欄位標示為 *