39.3. MLPerf Tiny v0.7 HPMicro

This is the HPMICRO port code for MLPerf V0.7.

For benchmark code and rules please see the GitHub repository.

39.3.1. Toolchain Requirements

This example requires a DSP-enabled toolchain to utilize hardware DSP instructions for neural network acceleration:

  • IDE: Segger Embedded Studio (or other IDEs with DSP toolchain support)

  • GCC Toolchain: Andes RISC-V toolchain with DSP extension support or ZCC toolchain with DSP extension support

Note

Standard GCC toolchains without DSP extensions will not be able to compile and run this demo correctly. The DSP instructions are essential for efficient neural network inference.

39.3.2. Code Options

  • Add code before find_package to the project’s CMakeLists.txt to get a higher performance

set(RV_ABI "ilp32d")
set(RV_ARCH "rv32gc")
set(CONFIG_HPM_MATH_NN_RVP32 1)