This repository was used as a university project for TU Vienna. All the code can be found in the knn folder. This project tests custom instructions in order to improve performance. In this example we are covering the K-Nearest-Neighbour Algorithm. This project targets x86, RISCV and the bare metal FPGA CPU designed in the DDCA course at TU Vienna.
compile is the main script used for all three compilation methods. Here is a small piece of documentation also embedded in the script:
#!/bin/bash
# ./compile => rv32im isa without custom instruction
# ./compile custom => rv32im isa with and without custom instruction + compiled assembly
# ./compile callgrind => x86 isa with callgrind.out file to check for profiling
# ./compile miriv => rv32im isa with and without custom instruction for MiRiV CPU
knn.c is the base KNN implementation which can use both the regular abs() function as well as the custom absdsq() function.
util.c contains general utilk functions used for all implementations.
libhelper.h contains a wrapper containing the custom instruction in assembly.
libc.c is a C standard library used in the FPGA CPU.
hex2mif.pl converts the IHEX files to MIF files used to initialize FPGA memory.
visualization/data.json is the output of a knn.c run.
visualization/knn.py is a python script visualizing the classified points.
The file Makefile.x86 compiles to a regular x86 architecture and can be run by any x86 cpu.
The file Makefile compiles to RISCV using the https://github.com/riscv-collab/riscv-gnu-toolchain.
The file Makefile.miriv also compiles to RISCV which is able to run on our bare metal FPGA CPU.