关于FPGA、DSP、GPU的使用思考

Some websites that can distinguish the usefulness of these things at a glance: C6678+Zynq-7045-based target tracking vision solution: https://mp.weixin.qq.com/s/3iB-CiecfxpgogUjvCulkg

1. FPGA: (1) FPGA is mainly composed of 6 parts: 1) Programmable input/output unit; ** 2) Basic programmable logic unit (main) ** ; 3) Embedded block RAM; 4) Rich wiring Resources; 5 ) Low-level embedded functional units; 6) Embedded dedicated hard cores (2) FPGA, as its name implies, is “Field Programmable Gate Array”. From the name, you can also imagine that FPGA is mainly composed of ** “Programmable Logic Unit” ** Composition, ** “Programmable Logic Unit” ** mainly consists of ** lookup table LUT ** and ** register ** . Where **lookup table LUT ** constitutes ** “combination logic circuit” ** , ** lookup table + register ** constitutes ** “sequential logic circuit” ** . (3) Yes (2) Knowing that the use of a pure FPGA is equivalent to ** “designing digital circuits (combination, timing)” **（重点记忆方法6122）

, If the digital circuit designed uses the same standard for fixed-point processing of floating-point numbers with very different decimal places, there will be a problem of large errors. ** Fixed-point floating-point numbers need to be treated differently. This more core DSP has a great disadvantage. ** (4) Common phrase design corresponding ** digital circuit module design ** and ** high-speed transmission interface design ** (DSP, The data from the PS side does not need to be fixed-point, but uses the PL side 5) Low-level embedded functional unit)

2. DSP (1) The high-profile single-chip microcomputer has a higher processing frequency. C6678 can reach 1.25GHz. It has greater advantages in processing radar signals, which require higher precision floating-point numbers. DSP is also commonly used in the industry to process image processing algorithms. Do what the PS side does in Zynq, and then the PS side performs the logic control of the digital circuit designed on the PL side. (2) The performance of the soul core series multi-core DSP of the 38 domestic institutes is already very high. Like the FFT operation that FPGA is good at, the processing efficiency of the soul core 2 is much greater than that of the K7 series FPGA. (3) Disadvantages: poor software and hardware decoupling capabilities, low single-node computing capabilities, and technical research in system data distribution,multi-node synchronization processing, and distributed real-time communication are required.

3. In GPU learning, welcome to see the pros and cons of the article summarized by the big guys! Recommend some GPU learning resources!