Exploring Hardware Accelerators for Kalman Filter

Kalman Filter is an algorithm that produces estimates of state variables of a discrete data controlled system (thus an estimate of the state of the system) based on measurements which can typically be noisy.

We have performed a basic profiling and analysis of a baseline C implementation, and explored multiple approaches to increase throughput and reduce latency at reduced resource utilizations. Following approaches are compared:

• Baseline C
• CPU Multithreading using OpenMP
• Custom Hardware Accelerator design using Bluespec SystemVerilog
• Custom Hardware Accelerator design using Vitis HLS

Refer to the video and report for further details.

Commands

To run baseC implementation gcc basecode_time.c

To run openmp implementation gcc -fopenmp code_omp_time.c

To compile and test bluespec files currently in the directory

1. For pe, go to src/bluespec make pe
2. For mat_mult, go to src/bluespecmake mat_mult
3. For top module (KalmanAlgo), go to src/bluespecmake Kalman

./a.out

For using profiling tools g++ -I . basecode_time.c -o basecode_output.out valgrind --tool=callgrind --dump-instr=yes --toggle-collect=main --collect-jumps=yes ./basecode_output.out kcachegrind callgrind.out.