/DSP-HW

Practice with TI C64x/C64x+ series DSP. 梁凡老师 DSP 器件原理与应用课程作业。

Primary LanguageCApache License 2.0Apache-2.0

DSP-HW

中山大学电信院 EIT349 梁凡老师的 DSP 器件原理与应用课程作业（2020 秋）。
Homework for Principles and Applications of DSP Devices, SYSU SEIT, Fall 2020.

2x bilinear interpolation Demo:

Demo 512 x 1.5 resize function cycle count in different implementations (debug mode, optimization not enabled):

C float approach C int approach Linear assembly
1654442635 1027521855 60202045
YCbCr to RGB Conversion Demo:
Enter Y, Cb, Cr (comma seprated): 123, 123, 123
Y:123, Cb:123, Cr:123
-- C (floating-point multiply) ---> R:136, G:148, B:134 | took 2595 cycles.
-- Linear ASM (16-bit multiply) --> R:136, G:148, B:134 | took 38 cycles.
-- Linear ASM (8-bit multiply) ---> R:135, G:147, B:133 | took 39 cycles.

欢迎在 Discussions 讨论，利用 Issues 跟踪代码问题， 以及提交 PR 分享你的代码。

Useful resources

Documents:

Websites:

Usage

Test environment

  • Code Composer Studio v8.3.1 on Windows 10 20H2
    • TI Code Gneration Tools v7.4.24

Note: Texas Instruments Simulators connection is required to run the CPU cycle accurate simulation without the actual development board, which was only existed in CCS prior to v6. This is the reason why the course recommended CCS v5, but it's quite old. However, there is a hack to migrate the software simulation to newer versions of CCS up to v8.3.1, thus I decided to use it. Code Generation Tools v7.4 is the last version of CGT to support C64/C64+, which was released in 2018.

CCS Configuration

  • Device:

    • Family: C6000
    • Variant: DaVinci DM64x, TMS320DM648

    Target configs are provided in the file targetConfigs/TMS320DM648.ccxml

  • Tool-chain:

    • Output format: eabi (ELF)
    • Device endianness: little
    • Linker command file: DM648.cmd

    Modified linker command file is provided as DM648.cmd.

  • Build

    • C6000 Compiler
      • Include Options: Add ${PROJECT_ROOT}/include
      • Performance Adviser: --advise:performance=all
    • C6000 Linker
      • Heap size for C/C++ dynamic memory allocation: --heap_size=0x200000

    The program uses malloc to dynamically allocate memory for the images. The .sysmem section of memory should be adjust to accommodate every images since the default size of 1KB is clearly not enough. More info about the cmd file could be found here.

Debug Configuration
  • Build
    • C6000 Compiler
      • Optimization: off
      • Debug Options: Full symbolic debug -g
      • Advanced Options
        • Predefined Symbols: --define=DEBUG
        • Assembler Options: --keep_asm
Release Configuration
  • Build
    • C6000 Compiler
      • Optimization: -O2
      • Debug Options: Symbolic debug for program analysis --symdebug:skeletal
      • Advanced Options
        • Predefined Symbols: --define=RELEASE
        • Runtime Model Options: Generate verbose software pipelining information --mw

Other options are assumed to be default.

How to run

Specific settings for each task
Task 1&2
  • Define USE_SA_IMPL to link resize function's linear assembly implementations, the C implementation would be used otherwise.
  • Define USE_INT to switch resize function's C implementation to its fixed-point multiplication approach, floating-point multiplication would be used otherwise.

Code, build and hit debug!

Homework List

  • 实验 1：用 C 语言实现 8 位 DIB 双线性插值
  • 实验 2：双线性插值函数使用线性汇编语言实现
  • 实验 3：用 C 语言和线性汇编语言完成从 YCbCr 彩色空间到 RGB 彩色空间的转换
  • 实验 4：用 C 语言完成线性变换程序并优化
  • 实验 5：用 C 语言和线性汇编语言实现中值滤波器
  • 实验 6：通过 DSP/BIOS, 利用 C 语言完成工作