We present a new approach for sparse Cholesky factorization on a heterogeneous platform with a graphics processing unit (GPU). The sparse Cholesky factorization is one of the core algorithms of numerous computing applications. We tuned the supernode data structure and used a parallelization method for GPU tasks to increase GPU utilization. Results show that our approach substantially reduces computational time.

This paper presents a parameterized multi-standard adaptive radix-4 Viterbi decoder with high throughput and low complexity. The proposed Viterbi decoder supports constraint lengths ranging from 3-9, code rates in the range of 1/2-1/3, and arbitrary truncation lengths. We present a novel fabric of Add-Compare-Select Unit (ACSU) and methods of unsigned quantization and efficient normalization that shorten the critical path. The decoder achieves a low bit error ratio in multiple standards, such as GPRS, WiMax, LTE, CDMA, and 3G. The proposed decoder is implemented on Xilinx XC5VLX330 device and the frequency achieved is 181.7 MHz. The throughput of the proposed decoder can reach 363 Mbps, which is superior to the other current multi-standard Viterbi decoders or radix-4 Viterbi decoders on the FPGA platform.

This letter proposes a novel high performance crypto coprocessor that relies on Reconfigurable Cryptographic Blocks. We implement the prototype of the coprocessor on Xilinx FPGA chip. And the pipelining technique is adopted to realize data paralleling. The results show that the coprocessor, running at 189MHz, outperforms the software-based SSL protocol.

The effective bandwidth of the dynamic random-access memory (DRAM) for the alternate row-wise/column-wise matrix access (AR/CMA) mode, which is a basic characteristic in scientific and engineering applications, is very low. Therefore, we propose the window memory layout scheme (WMLS), which is a matrix layout scheme that does not require transposition, for AR/CMA applications. This scheme maps one row of a logical matrix into a rectangular memory window of the DRAM to balance the bandwidth of the row- and column-wise matrix access and to increase the DRAM IO bandwidth. The optimal window configuration is theoretically analyzed to minimize the total number of no-data-visit operations of the DRAM. Different WMLS implementationsare presented according to the memory structure of field-programmable gata array (FPGA), CPU, and GPU platforms. Experimental results show that the proposed WMLS can significantly improve DRAM bandwidth for AR/CMA applications. achieved speedup factors of 1.6× and 2.0× are achieved for the general-purpose CPU and GPU platforms, respectively. For the FPGA platform, the WMLS DRAM controller is custom. The maximum bandwidth for the AR/CMA mode reaches 5.94 GB/s, which is a 73.6% improvement compared with that of the traditional row-wise access mode. Finally, we apply WMLS scheme for Chirp Scaling SAR application, comparing with the traditional access approach, the maximum speedup factors of 4.73X, 1.33X and 1.56X can be achieved for FPGA, CPU and GPU platform, respectively.

The parallel interference cancellation (PIC) multiple input multiple output (MIMO) detection algorithm has bit error ratio (BER) performance comparable to the maximum likelihood (ML) algorithm but with complexity close to the simple linear detection algorithm such as zero forcing (ZF), minimum mean squared error (MMSE), and successive interference cancellation (SIC), etc. However, the throughput of PIC MIMO detector on central processing unit (CPU) cannot meet the requirement of wireless protocols. In order to reach the throughput required by the standards, the graphics processing unit (GPU) is exploited in this paper as the modem processor to accelerate the processing procedure of PIC MIMO detector. The parallelism of PIC algorithm is analyzed and the two-stage PIC detection is carefully developed to efficiently match the multi-core architecture. Several optimization methods are employed to enhance the throughput, such as the memory optimization and asynchronous data transfer. The experiment shows that our MIMO detector has excellent BER performance and the peak throughput is 337.84 Mega bits per second (Mbps), about 7x to 16x faster than that of CPU implementation with SSE2 optimization methods. The implemented MIMO detector has better computing throughput than recent GPU-based implementations.

Many scientific applications require efficient variable-precision floating-point arithmetic. This paper presents a special-purpose Very Large Instruction Word (VLIW) architecture for variable precision floating-point arithmetic (VV-Processor) on FPGA. The proposed processor uses a unified hardware structure, equipped with multiple custom variable-precision arithmetic units, to implement various variable-precision algebraic and transcendental functions. The performance is improved through the explicitly parallel technology of VLIW instruction and by dynamically varying the precision of intermediate computation. We take division and exponential function as examples to illustrate the design of variable-precision elementary algorithms in VV-Processor. Finally, we create a prototype of VV-Processor unit on a Xilinx XC6VLX760-2FF1760 FPGA chip. The experimental results show that one VV-Processor unit, running at 253 MHz, outperforms the approach of a software-based library running on an Intel Core i3 530 CPU at 2.93 GHz by a factor of 5X-37X for basic variable-precision arithmetic operations and elementary functions.

In this paper, we propose a fully parallel Turbo decoder for Software-Defined Radio (SDR) on the Graphics Processing Unit (GPU) platform. Soft Output Viterbi algorithm (SOVA) is chosen for its low complexity and high throughput. The parallelism of SOVA is fully analyzed and the whole codeword is divided into multiple sub-codewords, where the turbo-pass decoding procedures are performed in parallel by independent sub-decoders. In each sub-decoder, an efficient initialization method is exploited to assure the bit error ratio (BER) performance. The sub-decoders are mapped to numerous blocks on the GPU. Several optimization methods are employed to enhance the throughput, such as the memory optimization, codeword packing scheme, and asynchronous data transfer. The experiment shows that our decoder has BER performance close to Max-Log-MAP and the peak throughput is 127.84Mbps, which is about two orders of magnitude faster than that of central processing unit (CPU) implementation, which is comparable to application-specific integrated circuit (ASIC) solutions. The presented decoder can achieve higher throughput than that of the existing fastest GPU-based implementation.