Skip to content

Latest commit

 

History

History
85 lines (45 loc) · 2.28 KB

TUTORIAL.md

File metadata and controls

85 lines (45 loc) · 2.28 KB
Raw

Installation

This instructions have been tested on Ubuntu 22.04.

To install VESPA, first copy the repository on your machine:

git clone https://github.com/hardware-fab/vespa

Then, install the dependencies running the bash script in the main folder:

cd vespa

./dependencies.sh

and make sure to have installed the necessary software:

  • Vivado 2019.2

  • Cadence Xcelium 23.03.007

  • ProFPGA Builder 2019A-SP2

Now, install the toolchain relative to one or more of the available processors:

cd vespa

# Leon3 toolchain

./utils/toolchain/build_leon3_toolchain.sh

# RISC-V 64-bit toolchain (for the Ariane processor)

./utils/toolchain/build_riscv_toolchain.sh

# RISC-V 32-bit toolchain (for the Ibex processor)

./utils/toolchain/build_riscv32imc_toolchain.sh

After this, you should edit the initialize.sh located in the main vespa folder with your paths and source it in order to be able to use the tools:

cd vespa

source ./initialize.sh

Now you should be able to use the VESPA tool.

Tutorials

You can refer to https://www.esp.cs.columbia.edu/docs/ for the ESP guide. For the most part, the ESP instructions can be used also for VESPA.

Major differences:

  • There are several differences in the GUI, but they are self-explaining.

  • The programming of the board now happens through UART. Be sure to update the esplink tool with the correct UART port name (go to ./tools/esplink/src/edcl.h and edit the SERIAL_PORT define with the name of your port).

  • The DFS is now managed by software. To use it, create a pointer in your program to the specific domain register (to get its address, refer to the csr code in ./rtl/sockets/csr) and give it a value from 0 to 19. The actual frequency will be FREQ = (X+1)*5 MHz.

Accelerators

To install the CHStone accelerators used in the overview paper:

cd ./socs/profpga-xc7v2000t
./accelerators.sh

You can install other accelerators following the ESP guides (https://www.esp.cs.columbia.edu/docs/). However, the multi-replica architecture of VESPA needs a batch register. So, during the creation of the accelerator, be sure to include a register named "acc_name"_n, and give it the value of the size of the data batches that this accelerator expects (the value can be modified later via software).