Detail aplikovaného výsledku

Optimization of FOC Algorithm on a Customized RISC-V Processor

ŠIMEK, V.; GALLO, J.

Originální název

Optimization of FOC Algorithm on a Customized RISC-V Processor

Anglický název

Optimization of FOC Algorithm on a Customized RISC-V Processor

Druh

Software

Abstrakt

The result is a software solution aimed at improving the computational efficiency of Field-Oriented Control (FOC) algorithms through the optimization of mathematical operations executed on a RISC-V processor. The solution is based on an existing NXP motor-control application using fixed-point arithmetic and includes workload profiling, identification of performance-critical code sections, and optimization of selected mathematical library functions. Specialized instruction-set extensions were designed for the Codasip µRISC-V processor to accelerate frequently executed operations, particularly saturated addition and subtraction, multiplication, bit shifts, and square-root computation. The software implementation involves modifications to the mathematical library, utilization of custom instructions through inline assembly, and adaptation of instruction selection in an LLVM-based compiler. The functionality of the optimized software was verified using instruction-accurate simulation and subsequently validated on the Genesys 2 FPGA hardware platform. Experimental evaluation demonstrated a 25.1% reduction in the computational execution time of the FOC algorithm on FPGA, accompanied by an 18.9% increase in LUT resource utilization. The resulting solution demonstrates the potential of hardware-software co-optimization for electric motor-control algorithms on application-specific RISC-V processors and provides a reusable methodology for further development of embedded control systems.

Abstrakt anglicky

The result is a software solution aimed at improving the computational efficiency of Field-Oriented Control (FOC) algorithms through the optimization of mathematical operations executed on a RISC-V processor. The solution is based on an existing NXP motor-control application using fixed-point arithmetic and includes workload profiling, identification of performance-critical code sections, and optimization of selected mathematical library functions. Specialized instruction-set extensions were designed for the Codasip µRISC-V processor to accelerate frequently executed operations, particularly saturated addition and subtraction, multiplication, bit shifts, and square-root computation. The software implementation involves modifications to the mathematical library, utilization of custom instructions through inline assembly, and adaptation of instruction selection in an LLVM-based compiler. The functionality of the optimized software was verified using instruction-accurate simulation and subsequently validated on the Genesys 2 FPGA hardware platform. Experimental evaluation demonstrated a 25.1% reduction in the computational execution time of the FOC algorithm on FPGA, accompanied by an 18.9% increase in LUT resource utilization. The resulting solution demonstrates the potential of hardware-software co-optimization for electric motor-control algorithms on application-specific RISC-V processors and provides a reusable methodology for further development of embedded control systems.

Klíčová slova

RISC-V, software optimization, field-oriented control, fixed-point arithmetic, custom instructions, code profiling, mathematical library, LLVM, embedded systems

Klíčová slova anglicky

RISC-V, software optimization, field-oriented control, fixed-point arithmetic, custom instructions, code profiling, mathematical library, LLVM, embedded systems

Licenční poplatek

K využití výsledku jiným subjektem je vždy nutné nabytí licence

www