Heston implementation for Zynq with Vivado HLS
The Zynq platform offers some very appealing features for rapid accelerator development including:
- Full features Linux distribution (Linaro Ubuntu) running on ARM with ssh access, packet manager, threading support
- Can run almost any cross platform software package (QuantLib, scikit-learn, SystemC, Python, Boost, OpenMPI, Git, etc.)
- Dynamic reconfiguration of the FPGA in < 200 ms from Linux command line, during runtime
- Low latency (~ 100 ns) and high bandwidth (1.6 GiB/s) interconnect between FPGA and ARM
- C++ user space driver development possible. Compile debug and benchmark programs like on a Desktop Linux system
- High level accelerator description possible base on the new Vivado HLS
- Automatic AXI-Stream and memory mapped AXI interface generation
This work will apply this powerfull setup to our financial application "option pricing". The goal is to implement a fully features SL / ML heston pricer that beats all our current implementations in:
- Productivity: lines of code
- Energy efficiency: steps / watt
- Implementation efficiency: FPGA area / accelerator
- Functionality: fully working and optimized implementations (FPGA & CPU)
- Features: heston sl & ml, full benchmark
- bitstream: FPGA bitstreams containing accelerators for Zynq
- hls: Accelerators written with Vivado HLS
- ip: Packed HLS accelerators, that can be instanciated and synthesized with Vivado
- linux: Kernels, boot images and discription on how to install linux
- results: deliverables of this project
- software: linux drivers for the accelerators and optimized CPU implementations for heston
See the individual folders for more information.
The Zynq demo can be found in the software folder. Additional instructions can be found there.
