研究目的
To evaluate the applicability of CuSP, a softcore DSP-ASIP, for image processing applications, comparing its performance with standard softprocessor MicroBlaze.
研究成果
CuSP, involving a Crimson processor and 2D convolution accelerator, achieves significant execution time speedup over MicroBlaze and rVEX softprocessors for 2D convolution, with reduced quantization error, making it a viable choice for accelerating image processing applications.
研究不足
The maximum supported image size is 250 × 250 pixels due to memory capacity constraints. The system lacks assembler support, requiring a pseudo-assembler implementation in C/C++ for machine code generation.
1：Experimental Design and Method Selection:
The study evaluates the performance of CuSP, a DSP-ASIP, against MicroBlaze for image processing tasks, specifically 2D convolution.
2：Sample Selection and Data Sources:
Gray scale images of different resolutions up to 250 × 250 pixels are used, limited by memory capacity.
3：List of Experimental Equipment and Materials:
CuSP simulation was performed on Modelsim, and for hardware implementation, Digilent Genesys 3 board with Virtex 5 FPGA was used.
4：Experimental Procedures and Operational Workflow:
The performance was evaluated by implementing 2D convolution using general instructions on Crimson DSP core and custom instructions on the accelerator.
5：Data Analysis Methods:
Clock cycles (cc) were counted from the start of the program till the execution of the last instruction in the convolution algorithm, with results compared against MATLAB simulations for accuracy.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
