GPU VSIPL: High-Performance VSIPL Implementation for GPUs
نویسندگان
چکیده
In this paper, we introduce GPU VSIPL1, an implementation of the Vector Signal Image Processing Library (VSIPL) Core Lite profile developed for the graphics processing unit (GPU). This implementation achieves a speedup of one to two orders of magnitude over the reference implementation. Because VSIPL is an open standard for high-performance platform-independent applications, GPU VSIPL provides an abstraction layer that leverages the GPU with no additional development costs.
منابع مشابه
GPU VSIPL: Core and Beyond
GPU VSIPL is an implementation of the Vector Signal Image Processing Library (VSIPL) and provides a librarybased solution to developing GPU-accelerated numerical applications. GPU VSIPL now supports much of the VSIPL Core Profile as well as element-wise matrix operators corresponding to required vector operators. This is nearly a five-fold increase in function count since GPU VSIPL was first in...
متن کاملVSIPL++Pro – A High-Performance VSIPL++ Implementation
VSIPL++ achieves high-performance through both the efficiency of C++ as a systems language, and by using C++ features for high-performance generic programming to convert domain-level abstractions into efficient code at compile time. The VSIPL++ API allows for early binding of computation and communication, allowing setup for complex functions and communications to be done early, out of critical...
متن کاملOptimization of Memory Allocation in VSIPL
In this paper, we propose an efficient memory allocation algorithm for Vector, Signal, and Image Processing Library (VSIPL) standard. This algorithm improves the efficiency of key VSIPL functions by exploiting fundamental properties of VSIPL objects during the allocation of memory. We implemented and tested our algorithm using the VSIPL reference implementation and measured results on represent...
متن کاملScalable SAR with Sourcery VSIPL++ for the Cell/B.E
Sourcery VSIPL++ for the Cell/B.E. [1][2] is an implementation of the open standard VSIPL++ signal and image-processing API [1] on the IBM Cell/B.E. multi-core processor architecture [4]. It is suitable for implementing high-performance signal-processing applications that take full advantage of the Cell/B.E. processor throughput, without sacrificing programmer productivity or application portab...
متن کاملVSIPL++: Intuitive Programming Using C++ Templates
VSIPL++ is a high-performance C++ toolkit for vector and signal processing applications. Building on the successful Vector, Signal, and Image Processing Library (VSIPL) standard (a C standard for similar applications), VSIPL++ adds additional features including: • direct support for parallel computation, • simpler syntax, improved type-checking, and other improvements to reduce validation and v...
متن کامل