Multicore DSP

5Software optimisation

CHAPTER MENU

1. 5.1 Introduction
2. 5.2 Hindrance to software scalability for a multicore processor
3. 5.3 Single‐core code optimisation procedure
   1. 5.3.1 The C compiler options
4. 5.4 Interfacing C with intrinsics, linear assembly and assembly
   1. 5.4.1 Intrinsics
   2. 5.4.2 Interfacing C and assembly
5. 5.5 Assembly optimisation
   1. 5.5.1 Parallel instructions
   2. 5.5.2 Removing the NOPs
   3. 5.5.3 Loop unrolling
   4. 5.5.4 Double‐Word Access
   5. 5.5.5 Optimisation summary
6. 5.6 Software pipelining
   1. 5.6.1 Software‐pipelining procedure
      1. 5.6.1.1 Writing linear assembly code
      2. 5.6.1.2 Creating a dependency graph
      3. 5.6.1.3 Resource allocation
      4. 5.6.1.4 Scheduling table
      5. 5.6.1.5 Generating assembly code
7. 5.7 Linear assembly
   1. 5.7.1 Hand optimisation of the dotp function using linear assembly
8. 5.8 Avoiding memory banks
9. 5.9 Optimisation using the tools
10. 5.10 Laboratory experiments
11. 5.11 Conclusion
12. References

5.1 Introduction

Software optimisation is the process of manipulating software code to achieve one or a combination of the following goals, depending on the application: faster code execution, smaller code size and low power consumption.

To implement efficient software on a multicore processor, the programmer must be familiar with the processor architecture (that includes the CPU, the memory and any peripheral or coprocessor to be used); the language(s) (i.e. C and assembly) used; the programming model to efficiently distribute software (tasks) across all cores, for example OpenMP (see Chapter 12) or OpenCL ...