GE Intelligent Platforms AXISLib – GPU DSP & Math libraries for NVIDIA CUDA GPUs AXISLib-GPU is a set of Signal and Vector applications for size, weight and power Features Processing Libraries providing over 500 (SWaP) constrained mission computing Functionality high performance digital signal processing platforms, AXISLib-GPU compliments GE’s • Comprehensive signal and vector and vector math functions created to help optimized libraries for Intel and Power processing functionality, providing the developers maximize system and application Architecture platforms enabling application VSIPL API with over 500 functions. performance on deployed mission pay loads. developers to harness the performance required to best suite their application needs. Performance General purpose computing on graphics • Optimized for CUDA processing units (GPGPU) can offer substantial The VSIPL library provides a standard Fully optimized NVIDIA CUDA architecture performance increases in radar, sonar, application programming interface (API) libraries for GE GPGPU platforms. signal and image processing applications as defined by the VSIPL Forum. Designed • Multithreaded when compared to CPU only solutions. to support embedded real-time signal AXISLib-GPU provides a suite of commonly processing systems, it provides enhanced Takes maximum advantage of the used DSP and math libraries to get the application portability between platforms massively parallel architecture of GPUs. most out of GE’s latest, rugged NVIDIA and consistency across data types, Core computations implemented as GPGPU platforms. processors and vendor implementations – CUDA kernels. accelerating engineering development Portability Designed to support the most demanding and new technology insertion over the • VSIPL API high performance digital signal processing entire program life cycle. Provides industry standard API for maximum portability. Create a block • VSIPL Core Lite compliant plus many CPU functions from Core 1.0 profile. Obtain / Create data Bind data block • Portable across operating systems ™ ® (Windows, Linux ) Admit data from user to VSIPL space Host to Device Copy • 32-bit and 64-bit support. Create and assign view of block Perform VSIPL processing GPU Computation Destroy the view Device to host copy Release data from VSIPL to user space Use the processed data CPU Destroy Block The VSIPL concept of ‘user space’ and ‘VSIPL space’ fits well with the GPGPU paradigm of CPU (host) and GPU (device) memory. VSIPL hides the concept of copying data between the host and device memory by performing the operations during the VSIPL admit and release stages. AXISLib-GPU – DSP & Math libraries for NVIDIA CUDA GPUs Functionality Function Set Description Scalar Complex Scalar 23 functions for performing complex scalar math Random Number Generation Random Numbers 6 functions for generating random numbers, vectors and complex vectors Vector and Elementwise Operations Elementary Mathematical 21 functions performing elementary vector math (sin, cos, tan, atan, exp, log, sqrt, etc.) GE Intelligent Platforms’ rugged IPN250 and Unary Operations 13 functions for operating on a single vector or matrix NPN240 GPGPU processors running AXISLib DSP routines offer game changing levels of Binary Operations 29 functions for operating either two vectors or matrices or one vector and a scalar performance to support expanded ISR mission Ternary Operations 19 functions for operations requiring three inputs capabilities across a wide range of SWaP Logical Operations 8 functions for performing logical operations on vectors or matrices sensitive applications. Selection Operations 16 functions for selecting a subset of a vector or matrix Bitwise and Boolean 24 functions for performing Bitwise and Boolean operations on vectors and matrices Logical Operators Element Generation and Copy 40 functions for copying and generating vector elements Manipulation Operations 28 functions for vector and matrix manipulation (e.g. scatter, gather and swap) Signal Processing FFTs 21 functions for performing 1D and 2D FFTs (real-complex, complex-real, complex- complex in place and out-of-place) Windowing 4 windowing functions (Blackman, Hanning, Kaiser, Chebyshev) Filter 8 functions for FIR filtering Convolution 4 functions convolutions (1D and 2D) Correlation 4 functions correlations (1D and 2D) Histogram 1 function histograming Linear Algebra Matrix and Vector Operations 82 functions for performing linear algebra on vectors and matrices Linear System Solvers 17 functions Performance Comparasions Ordering Information AXISLib-GPU-01M Maintenance Agreement. Includes the right to use license and 1 x runtime license. Annually renewable. AXISLib-GPU-01R Runtime license. GE Intelligent Platforms Contact Information Americas: 1 800 433 2682 or 1 434 978 5100 Global regional phone numbers are listed by location on our web site at www.ge-ip.com/contact www.ge-ip.com/axisdemo ©2010 GE Intelligent Platforms, Inc. All rights reserved. All other brands or names are property of their respective holders. Specifications are subject to change without notice. 12.10 GFA-1728