RISC-V benchmark automation

Collection of benchmarks to run for the RISC-V architecture on Spike (riscv-isa-sim), with a TUI (terminal UI) made with curses.

Prereqs

• lief and gitpython Python packages
• your clang toolchain path in ./env
• spike (riscv-isa-sim)
• CMake
• GNU Make
• bash

Usage

Run

Run python3 tui.py. Press r to run all benchmarks. The contents of ./env are reloaded every time before all benchmarks are run. This means that you can change the compiler path or optimization level in CFLAGS as you need. You can run multiple runs, creating multiple columns. If you run out of terminal columns, the program dumps the CSV and terminates.

Analyze

Press m to cycle between visualization modes. In relative modes, you can use the left and right arrow keys to select the column to use as baseline.

Terminate and dump CSV

Press q to close the program. This creates a file named output.csv. To terminate in the middle of benchmarks being run, you do have to hit Ctrl+C multiple times. This is because we are using the script command to work around an issue with spike. The first Ctrl+C is forwarded to spike, which interprets it to only terminate the program it's running internally, and drops into an interactive command line interface. The second Ctrl+C actually terminates spike.