PULPissimo

Citing

If you are using the PULPissimo IPs for an academic publication, please cite the following paper:

@INPROCEEDINGS{8640145,
  author={Schiavone, Pasquale Davide and Rossi, Davide and Pullini, Antonio and Di Mauro, Alfio and Conti, Francesco and Benini, Luca},
  booktitle={2018 IEEE SOI-3D-Subthreshold Microelectronics Technology Unified Conference (S3S)}, 
  title={Quentin: an Ultra-Low-Power PULPissimo SoC in 22nm FDX}, 
  year={2018},
  volume={},
  number={},
  pages={1-3},
  doi={10.1109/S3S.2018.8640145}}

PULPissimo is the microcontroller architecture of the more recent PULP chips, part of the ongoing "PULP platform" collaboration between ETH Zurich and the University of Bologna - started in 2013.

PULPissimo, like PULPino, is a single-core platform. However, it represents a significant step ahead in terms of completeness and complexity with respect to PULPino - in fact, the PULPissimo system is used as the main System-on-Chip controller for all recent multi-core PULP chips, taking care of autonomous I/O, advanced data pre-processing, external interrupts, etc. The PULPissimo architecture includes:

• Either the RI5CY core or the Ibex one as main core
• Autonomous Input/Output subsystem (uDMA)
• New memory subsystem
• Support for Hardware Processing Engines (HWPEs)
• New simple interrupt controller
• New peripherals
• New SDK

RISCY is an in-order, single-issue core with 4 pipeline stages and it has an IPC close to 1, full support for the base integer instruction set (RV32I), compressed instructions (RV32C) and multiplication instruction set extension (RV32M). It can be configured to have single-precision floating-point instruction set extension (RV32F). It implements several ISA extensions such as: hardware loops, post-incrementing load and store instructions, bit-manipulation instructions, MAC operations, support fixed-point operations, packed-SIMD instructions and the dot product. It has been designed to increase the energy efficiency of in ultra-low-power signal processing applications. RISCY implementes a subset of the 1.10 privileged specification. It includes an optional PMP and the possibility to have a subset of the USER MODE. RISCY implement the RISC-V Debug spec 0.13. Further information about the core can be found at http://ieeexplore.ieee.org/abstract/document/7864441/ and in the documentation of the IP.

Ibex, formely Zero-riscy, is an in-order, single-issue core with 2 pipeline stages. It has full support for the base integer instruction set (RV32I version 2.1) and compressed instructions (RV32C version 2.0). It can be configured to support the multiplication instruction set extension (RV32M version 2.0) and the reduced number of registers extension (RV32E version 1.9). Ibex implementes the Machine ISA version 1.11 and has RISC-V External Debug Support version 0.13.2. Ibex has been originally designed at ETH to target ultra-low-power and ultra-low-area constraints. Ibex is now maintained and further developed by the non-for-profit community interest company lowRISC. Further information about the core can be found at http://ieeexplore.ieee.org/document/8106976/ and in the documentation of the IP at https://ibex-core.readthedocs.io/en/latest/index.html

PULPissimo includes a new efficient I/O subsystem via a uDMA (micro-DMA) which communicates with the peripherals autonomously. The core just needs to program the uDMA and wait for it to handle the transfer. Further information about the core can be found at http://ieeexplore.ieee.org/document/8106971/ and in the documentation of the IP.

PULPissimo supports I/O on interfaces such as:

• SPI (as master)
• I2S
• Camera Interface (CPI)
• I2C
• UART
• JTAG

PULPissimo also supports integration of hardware accelerators (Hardware Processing Engines) that share memory with the RI5CY core and are programmed on the memory map. An example accelerator, performing multiply-accumulate on a vector of fixed-point values, can be found in ips/hwpe-mac-engine (after updating the IPs: see below in the Getting Started section). The ips/hwpe-stream and ips/hwpe-ctrl folders contain the IPs necessary to plug streaming accelerators into a PULPissimo or PULP system on the data and control plane. For further information on how to design and integrate such accelerators, see ips/hwpe-stream/doc and https://arxiv.org/abs/1612.05974.

Documentation

• The datasheet contains details about Memory Map, Peripherals, Registers etc.
• PULPissimo was presented at the Week of Open Source Hardware (WOSH) 2019 at ETH Zurich.
  • Slides
  • Video

Getting Started

We provide a simple runtime and a full featured runtime for PULPissimo. We recommend you try out first the minimal runtime and when you hit its limitations you can try the full runtime by installing the SDK.

After having chosen a runtime you can run software by either simulating the hardware or running it in a software emulation.

Prerequisites

PULPissimo is a Microcontroller provided in SystemVerilog RTL description. As such, it can be used and evaluated with many different tools. Out of the box, we provide Makefile targets for RTL simulation with Mentor Questa SIM (Intel/Altera Modelsim is not supported at the moment) and Cadence Xcelium. Being purely written in SystemVerilog, in theory the whole design can be simulated with any RTL simulator with (deccent!) SystemVerilog support. While an open source simulation target is definitely on our wish- and todo-list (e.g. out-of-the box support for Verilator), this currently still requires more extensive modifications to the RTL and scripts.

For FPGA implementation (see FPGA Section) we generate ready-made scripts for Synthesis and Implementation for Xilinx Vivado for a number of different development boards.

Simple Runtime

The simple runtime is here to get you started quickly. Using it can run and write programs that don't need any advanced features.

First install the system dependencies indicated here

Then make sure you have riscv-gnu-toolchain installed (either by compiling it or using one of the binary releases under available under the release tab) and point PULP_RISCV_GCC_TOOLCHAIN to it:

export PULP_RISCV_GCC_TOOLCHAIN=YOUR_PULP_TOOLCHAIN_PATH

Add the pulp-toolchain to your PATH variable:

export PATH=$PULP_RISCV_GCC_TOOLCHAIN/bin:$PATH

Get the repository for the simple runtime:

git clone https://github.com/pulp-platform/pulp-runtime/

The simple runtime supports many different hardware configurations. We want PULPissimo.

cd pulp-runtime

Then, to use the CV32E40P (formely RI5CY) core, type:

source configs/pulpissimo_cv32.sh

or to use the Ibex (formely zero-riscy) core:

source configs/pulpissimo_ibex.sh

Now we are ready to set up the simulation environment. Normally you would want to simulate the hardware design running your program, so go here.

PULP FreeRTOS

PULP FreeRTOS allows you to build applications using the FreeRTOS kernel. You can also choose to not use the FreeRTOS kernel and build a baremetal application, though in that case driver support is not yet fully fleshed out.

First make sure you have riscv-gnu-toolchain installed (either by compiling it or using one of the binary releases under available under the release tab) and point your RISCV environment variable to it.

Also we need to set up the simulation environment. Normally you would want to simulate the hardware design running your program, so go here to do that.

Then get the repository for the pulp-freertos:

git clone https://github.com/pulp-platform/pulp-freertos/

There are multiple hardware configuration supported. Select PULPissimo using the CV32E40P core. So enter the directory of pulp-freertos:

cd pulp-freertos

and select the correct configuration:

source env/pulpissimo-cv32e40p.sh

You then can run a simple freertos hello world like this:

cd tests/hello_world_pmsis
make all run

There are other tests in tests/ you can run.

Software Development Kit (UNSUPPORTED WITH CURRENT RELEASE)

If you need a more complete runtime (drivers, tasks etc.) you can install the software development kit for PULP/PULPissimo.

First install the system dependencies indicated here

In particular don't forget to set PULP_RISCV_GCC_TOOLCHAIN.

You can now either follow the steps outlined here to build the full sdk or install these python dependencies

pip3 install --user artifactory twisted prettytable sqlalchemy pyelftools 'openpyxl==2.6.4' xlsxwriter pyyaml numpy configparser pyvcd sphinx

and just call

make build-pulp-sdk

and then set up the necessary environment variables with

source env/pulpissimo.sh

There exists a bug in GCC 11.1.0 which fails the sdk build with the error 'this' pointer is null [-Werror=nonnull]. If you encounter this bug use the following temporary workaround instead to build the SDK:

VP_WORKAROUND_NONNULL_BUG=yes make build-pulp-sdk

Building the RTL simulation platform

Note you need Questasim or Xcelium to do an RTL simulation of PULPissimo (verilator support planned, but not finished). Intel Modelsim for Intel FPGAs does not work.

To build the RTL simulation platform, start by getting the latest version of the IPs composing the PULP system:

make checkout

This will download all the required IPs, solve dependencies and generate the scripts. The dependency management tool is Bender.

After having access to the SDK, you can build the simulation platform by doing the following:

source setup/vsim.sh
make build

This command builds a version of the simulation platform with no dependencies on external models for peripherals. See below (Proprietary verification IPs) for details on how to plug in some models of real SPI, I2C, I2S peripherals.

For more advanced usage have a look at ./bender --help for bender.

Also check out the output of make help for more useful Makefile targets.

Developing your own RTL

Bender How To

With Bender developing on top of PULPissimo is getting a lot easier. The command line tool is installed in the project root directory if you invoke make checkout. It performs dependency resolution according to a manifest file called Bender.yml. The file lists all source files of the RTL project as well as its direct dependencies. Bender can be used to generate source file lists for various different tools for simulation, ASIC/FPGA synthesis etc. Have a look at the Bender project documentation if you want to know more about it. For now we will concentrate on the most important steps when developing on PULPissimo using Bender.

Where are all the sub IPs (dependencies)?

Bender checks out the sub-ips in a hidden directory called .bender/git/checkouts. You are not supposed to change the files in this directory. If you want to get the path of a specific IP, call ./bender path <some ip (e.g. axi)> to get the relative path to an IP. To list all IPs in the project, call ./bender packages -f.

Modifying an existing IP

The hidden bender directory is not the location to introduce changes to the RTL of sub-ips. If you want to quickly try out changes to a sub-ip, call ./bender clone <ip_name> to checkout a working copy of the ip into a directory called working_dir. Call make scripts to update the source files in the scripts. Afterwards, every change you make in the RTL of this working copy will be incorporated into the RTL simulation model (once you recompile it with make build) and the FPGA build (once you synthesize it).

Adding a new IP to PULPissimo

If you want to add new IPs to pulpissimo you most likely will have to fork the pulp_soc sub-ip since this is the main repository that contains most of the SoCs RTL logic. Thus, follow the steps above to create a working copy of pulp_soc. Then you can either add your additional source code directly to pulp_socs source tree or, preferably, create a new repository with your source code, register the RTL source files in a Bender.yml manifest file and add this new repository as a dependency to pulp_soc's Bender.yml. Then you are free to instantiate your new IP somewhere within pulp_soc. We make excessive use of this strategy throughout the pulpissimo project which is a collection of many different IP repositories.

Downloading and running examples

Finally, you can download and run examples; for that you can checkout the following repositories depending on whether you use the simple runtime or the full sdk.

Simple Runtime: https://github.com/pulp-platform/pulp-runtime-examples

SDK: https://github.com/pulp-platform/pulp-rt-examples

Now you can change directory to your favourite test e.g.: for an hello world test for the SDK, run

cd pulp-rt-examples/hello
make clean all run

or for the Simple Runtime:

cd pulp-runtime-examples/hello
make clean all run

If you want to change the compiler flags, as for example if you are using CV32E40P with the XPULP extensions but you want to compile using only the RV32IMC instructions to compare performance, you can modify the Makefile inside the pulp-runtime-examples/hello folder adding:

PULP_ARCH_CFLAGS    =  -march=rv32imc -DRV_ISA_RV32
PULP_ARCH_LDFLAGS   =  -march=rv32imc
PULP_ARCH_OBJDFLAGS = -Mmarch=rv32imc

The open-source simulation platform relies on JTAG to emulate preloading of the PULP L2 memory. If you want to simulate a more realistic scenario (e.g. accessing an external SPI Flash), look at the sections below.

In case you want to see the Modelsim GUI, just type

make run gui=1

before starting the simulation.

If you want to save a (compressed) VCD for further examination, type

make run vsim/script=export_run.tcl

before starting the simulation. You will find the VCD in build/<SRC_FILE_NAME>/pulpissimo/export.vcd.gz where <SRC_FILE_NAME> is the name of the C source of the test.

Building and using the virtual platform

The virtual platform is a software-only model of the PULPissimo SoC (and also of other related SoCs). While a simulation of the hardware design is accurate it is also very very slow. The virtual platform helps you develop software quicker by providing a more or less accurate software-model of PULPissimo.

Once the sdk is installed, the following commands can be executed in the sdk directory to use the virtual platform:

source sourceme.sh
source configs/platform-gvsoc.sh

Then tests can be compiled and run as for the RTL platform. When switching from one platform to another, it may be needed to regenrate the test configuration with this command:

make conf

More information is available in the documentation here: pulp-builder/install/doc/vp/index.html

Updating the bootrom

You can customize the bootrom, have a look at the boot_code/ directory. To import your changed version of the boot code into PULPissimo, just call

make import-bootcode

FPGA

PULPissimo has been implemented on FPGA for the various Xilinx FPGA boards.

Supported Boards

At the moment the following boards are supported:

• Digilent Genesys2
• Xilinx ZCU104
• Xilinx ZCU102
• Xilinx VCU108
• Digilent Nexys Board Family
• ZedBoard

In the release section you find precompiled bitstreams for all of the above mentionied boards. If you want to use the latest development version PULPissimo follow the section below to generate the bitstreams yourself.

Bitstream Generation

In order to generate the PULPissimo bitstream for a supported target FPGA board first generate the necessary synthesis include scripts by running the corresponding make target.

make scripts

This will parse the Bender.yml using the PULP bender dependency management tool to generate tcl scripts for all the IPs used in the PULPissimo project. These files are later on sourced by Vivado to generate the bitstream for PULPissimo.

Now switch to the fpga subdirectory and start the apropriate make target to generate the bitstream:

cd fpga
make <board_target>

In order to show a list of all available board targets call:

make help

This process might take a while. If everything goes well your fpga directory should now contain two files:

• pulpissimo_<board_target>.bit the bitstream file for JTAG configuration of the FPGA.
• pulpissimo_<board_target>.bin the binary configuration file to flash to a non-volatile configuration memory.

If your invocation command to start Vivado isn't vivado you can use the Make variable VIVADO to specify the right command (e.g. make genesys2 VIVADO='vivado-2018.3 vivado' for ETH CentOS machines.) Boot from ROM is not available yet. The ROM will always return the jal x0,0 to trap the core until the debug module takes over control and loads the programm into L2 memory. Once the bitstream pulpissimo_genesys2.bit is generated in the fpga folder, you can open Vivado vivado (we tried the 2018.3 version) and load the bitstream into the fpga or use the Configuration File (pulpissimo_genesys2.bin) to flash it to the on-board Configuration Memory.

Bitstream Flashing

Start Vivado then:

Open Hardware Manager
Open Target
Program device

Now your FPGA is ready to emulate PULPissimo!

Board Specific Information

Have a look at the board specific README.md files in fpga/pulpissimo-<board_target>/README.md for a description of peripheral mappings and default clock frequencies.

Compiling Applications for the FPGA Target

To run or debug applications for the FPGA you need to use a recent version of the PULP-SDK (commit id 3256fe7 or newer.'). Configure the SDK for the FPGA platform by running the following commands within the SDK's root directory:

source configs/pulpissimo.sh
source configs/fpgas/pulpissimo/<board_target>.sh

Currently, the only available board_target in the SDK is the genesys2.sh board. However, there are no board specific settings in this file except for the clock frequency and UART baudrate that can easily be overidden (see below). You can just source the genesys2.sh target regardless of which FPGA board you are actually using and override the frequencies and baudrate in your application. The only reason you need to source the genesys2.sh configuration file instead of e.g. the rtl platform configuration is to instruct the SDK to omit all runtime initialization (the code executed before your main function is called on the core) of the FLLs that are not available in the FPGA version of PULPissimo.

If you updated the SDK don't forget to recompile the SDK and the dependencies.

In order for the SDK to be able to configure clock dividers (e.g. the ones for the UART module) to the right values it needs to know which frequencies PULPissimo is running at. You can find the default frequencies in the above mentioned board specific README files.

In our application we need to override two weakly defined variables in our source code to configure the SDK to use these frequencies:

#include <stdio.h>
#include <rt/rt_api.h>

int __rt_fpga_fc_frequency = <Core Frequency> // e.g. 20000000 for 20MHz;
int __rt_fpga_periph_frequency = <SoC Frequency> // e.g. 10000000 for 10MHz;

int main()
{
...
}

By default, the baudrate of the UART is set to 115200.

Add the following global variable declaration to your application in case you want to change it:

unsigned int __rt_iodev_uart_baudrate = your baudrate;

Compile your application with

make clean all

This command builds the ELF binary with UART as the default io peripheral. The binary will be stored at build/pulpissimo/[app_name]/[app_name].

Core selection

By default, PULPissimo is configured to use the RI5CY core with floating-point support being enabled. To switch to Ibex (and disable floating-point support), the following steps need to be performed.

1. Switch hardware configuration

   Open the file fpga/pulpissimo-<board_target>/rtl/xilinx_pulpissimo.v and change the CORE_TYPE parameter to the preferred value. Change the value of the USE_FPU parameter from 1 to 0. Save the file and regenerate the FPGA bitstream.

2. Switch SDK configuration

   Instead of sourcing configs/pulpissimo.sh when configuring the SDK, source configs/pulpissimo_ibex.sh.

GDB and OpenOCD

In order to execute our application on the FPGA we need to load the binary into PULPissimo's L2 memory. To do so we can use OpenOCD in conjunction with GDB to communicate with the internal RISC-V debug module.

PULPissimo uses JTAG as a communication channel between OpenOCD and the Core. Have a look at the board specific README file on how to connect your PC with PULPissimo's JTAG port.

Due to a long outstanding issue in the RISC-V OpenOCD project (issue #359) the riscv/riscv-openocd does not work with PULPissimo. However there is a small workaround that we incorporated in a patched version of openocd. If you have access to the artifactory server, the patched openocd binary is installed by default with the make deps command in the SDK. If you don't have access to the precompiled binaries you can automatically download and compile the patched OPENOCD from source. You will need to install the following dependencies on your machine before you can compile OpenOCD:

• autoconf >= 2.64
• automake >= 1.14
• texinfo
• make
• libtool
• pkg-config >= 0.23 (or compatible)
• libusb-1.0
• libftdi
• libusb-0.1 or libusb-compat-0.1 for some older drivers

After installing those dependecies with you OS' package manager you can download, apply the patch and compile OpenOCD with:

source sourceme.sh && ./pulp-tools/bin/plpbuild checkout build --p openocd --stdout

The SDK will automatically set the environment variable OPENOCD to the installation path of this patched version.

Launch openocd with one of the provided or your own configuration file for the target board as an argument.

E.g.:

$OPENOCD/bin/openocd -f pulpissimo/fpga/pulpissimo-genesys2/openocd-genesys2.cfg

In a seperate terminal launch gdb from your pulp_riscv_gcc installation passing the ELF file as an argument with:

$PULP_RISCV_GCC_TOOLCHAIN_CI/bin/riscv32-unknown-elf-gdb PATH_TO_YOUR_ELF_FILE

In gdb, run:

(gdb) target remote localhost:3333

to connect to the OpenOCD server.

In a third terminal launch a serial port client (e.g. screen or minicom) on Linux to riderect the UART output from PULPissimo with e.g.:

screen /dev/ttyUSB0 115200

the ttyUSB0 target may change.

Now you are ready to debug!

In gdb, load the program into L2:

(gdb) load

and run the programm:

(gdb) continue

Of course you can also benefit from the debug capabilities that GDB provides.

E.g. see the disasembled binary:

(gdb) disas

List the current C function, set a break point at line 25, continue and have fun!

(gdb) list
21
22  int main()
23  {
24    while (1) {
25      printf("Hello World!\n\r");
26     for (volatile int i=0; i<1000000; i++);
27    }
28    return 0;
29  }

(gdb) b 25
Breakpoint 1 at 0x1c0083d2: file test.c, line 25.
(gdb) c
Continuing.

Breakpoint 1, main () at test.c:25
25      printf("Hello World!\n\r");


(gdb) disas
Dump of assembler code for function main:
   0x1c0083d4 <+22>:    li  a1,1
   0x1c0083d6 <+24>:    blt s0,a5,0x1c0083e8 <main+42>
=> 0x1c0083da <+28>:    lw  a5,12(sp)
   0x1c0083dc <+30>:    slli    a1,a1,0x1
   0x1c0083de <+32>:    addi    a5,a5,1
   0x1c0083e0 <+34>:    sw  a5,12(sp)

(gdb) monitor reg a5
a5 (/32): 0x000075B7

Not all gdb commands work as expected on the riscv-dbg target. To get a list of available gdb commands execute:

monitor help

Most notably the command info registers does not work. Use monitor reg instead which has the same effect.

Proprietary verification IPs

The full simulation platform can take advantage of a few models of commercial SPI, I2C, I2S peripherals to attach to the open-source PULP simulation platform. In rtl/vip/spi_flash, rtl/vip/i2c_eeprom, rtl/vip/i2s you find the instructions to install SPI, I2C and I2S models.

When the SPI flash model is installed, it will be possible to switch to a more realistic boot simulation, where the internal ROM of PULP is used to perform an initial boot and to start to autonomously fetch the program from the SPI flash. To do this, the LOAD_L2 parameter of the testbench has to be switched from JTAG to STANDALONE.

PULP platform structure

After being fully setup as explained in the Getting Started section, this root repository is structured as follows:

• rtl/tb contains the main platform testbench and the related files.
• rtl/vip contains the verification IPs used to emulate external peripherals, e.g. SPI flash and camera.
• rtl could also contain other material (e.g. global includes, top-level files)
• sim contains the ModelSim/QuestaSim simulation platform.
• pulp-sdk contains the PULP software development kit; pulp-sdk/tests contains all tests released with the SDK.
• Bender.yml contains the package information used with bender. This includes a list of IPs required and source files contained within this repository.
• When using bender, other files may be relevant: Bender.local contains configs for bender, including overrides for dependencies, Bender.lock is a generated file used by bender, bender is the bender executable fetched by the makefile, .bender directory contains the database and checkouts used by bender.

Requirements

The RTL platform has the following requirements:

• Relatively recent Linux-based operating system; we tested Ubuntu 16.04 and CentOS 7.
• Mentor ModelSim in reasonably recent version (we tested it with version 10.6b -- the free version provided by Altera is only partially working, see issue #12).
• Python 3.4, with the pyyaml module installed (you can get that with pip3 install pyyaml).
• The SDK has its own dependencies, listed in https://github.com/pulp-platform/pulp-sdk/blob/master/README.md
• You will need the minicom command line application to view UART output in case you use the 'run' Makefile target with the FPGA platform (discouraged, you better use the approach outlined above)

External contributions

The supported way to provide external contributions is by forking one of our repositories, applying your patch and submitting a pull request where you describe your changes in detail, along with motivations. The pull request will be evaluated and checked with our regression test suite for possible integration. If you want to replace our version of an IP with your GitHub fork, just update the Bender.yml file and run ./bender update. While we are quite relaxed in terms of coding style, please try to follow these recommendations: https://github.com/pulp-platform/ariane/blob/master/CONTRIBUTING.md

Known issues

Support & Questions

For support on any issue related to this platform or any of the IPs, please add an issue to our tracker on https://github.com/pulp-platform/pulpissimo/issues