Skip to content

Latest commit

 

History

History
90 lines (59 loc) · 2.39 KB

embench.md

File metadata and controls

90 lines (59 loc) · 2.39 KB

Embench

Describes the Embench build simulation flow.

Embench Overview

  • From the Embench Website

    Dhrystone and Coremark have been the defacto standard microcontroller
benchmark suites for the last thirty years, but these benchmarks no longer
reflect the needs of modern embedded systems. Embench™ was explicitly
designed to meet the requirements of modern connected embedded systems. The
benchmarks are relevant, portable, and well implemented.

  • The source code for the Embench suite is taken from Github, and exists as a submodule to this repository under extern/embench-iot.

Building Embench

  • Make sure that the Embench submodule is checked out with the repository:

    git submodule update --init extern/embench-iot

  • To build the Embench suite such that it will run inside the CCX testbench, run:

    make build-embench-binaries

    This runs the Embench build flow, with various parameters setting the correct architecture, linker script and compiler etc.

    Build artefacts will be placed in work/embench/src/[BENCHMARK NAME].

    • Note: Embench benchmarks contain a LOCAL_SCALE_FACTOR preprocessor directive indicating how many times to run the benchmark. It takes far too long to run the benchmarks in simulation with the default values. During the build flow, a sed command replaces these values with 1, which is enough for this core which has no caches or branch prediction.

      See the build-embench-binaries' target in flow/embench/Makefile.in` to see where this happens.

  • After the vanilla Embench suite builds are complete, run:

    make build-embench-targets

    This creates disassembly files for each benchmark (for debugging) and Verilog hex files for loading into the simulator.

Running Embench

  • To run every Embench benchmark, run:

    make run-embench-targets

    This will run every benchmark inside the ccx level testbench. It can take a while.

    By default, the flow does not dump VCD files for the simulations (much too slow and large) but this can be added by running:

    make EMBENCH_WAVES=1 run-embench-targets

    The VCD wave file will appear in work/embench/src/[BENCHMARK NAME].

  • Individual benchmarks can be run using:

    make run-embench-[BENCHMARK NAME]