Signed-off-by: Carlos de Paula <me@carlosedp.com>
Chiselwatt
A tiny POWER Open ISA soft processor written in Chisel.
Simulation using verilator
-
Chiselwatt uses
verilatorfor simulation. It is built by default and run in a Docker container. To build with local verilator install, editMakefile. -
First build chiselwatt:
git clone https://github.com/antonblanchard/chiselwatt
cd chiselwatt
make
- A micropython image is included in the repo. To use it, link the memory image into chiselwatt:
ln -s micropython/firmware.hex insns.hex
- Now run chiselwatt:
./chiselwatt
- If your operating system is not Linux, run chiselwatt inside a container:
make dockerlator
# Inside the container prompt, run:
./chiselwatt
# type "exit" to exit the container
exit
Building micropython from scratch
- You can also build micropython from scratch. If you aren't building natively
on a ppc64le box you will need a cross compiler. This may be available on your
distro, otherwise grab the the powerpc64le-power8 toolchain from bootlin.
If you are cross compiling, point
CROSS_COMPILEat the toolchain. In the example below I installed it in usr/local/powerpc64le-power8--glibc--bleeding-edge-2018.11-1/bin/ and the tools begin withpowerpc64-linux-*:
git clone https://github.com/micropython/micropython.git
cd micropython
cd ports/powerpc
make CROSS_COMPILE=/usr/local/powerpc64le-power8--glibc--bleeding-edge-2018.11-1/bin/powerpc64le-linux- -j$(nproc)
cd ../../../
- Build chiselwatt, import the the micropython image and run it. We use bin2hex.py to convert a binary file into a series of 64 bit hex values expected by the tools:
cd chiselwatt
make
scripts/bin2hex.py ../micropython/ports/powerpc/build/firmware.bin > insns.hex
./chiselwatt
Synthesis using Open Source tools (yosys/nextpnr)
Synthesis on FPGAs is supported with yosys/nextpnr. At the moment the tools support
Lattice ECP5 FPGAs. The build process uses Docker images, so no software other than Docker needs
to be installed. If you prefer podman you can use that too, just adjust it in
Makefile, DOCKER=podman.
hello_world
The hello_world example should run everywhere, so start with it. Edit src/main/scala/Core.scala
and set memory to 16 kB (16*1024):
chisel3.Driver.execute(Array[String](), () => new Core(64, 16*1024, "insns.hex", 0x0))
Then link in the hello_world image:
ln -s hello_world/hello_world.hex insns.hex
Building and programming the FPGA
The Makefile currently supports the following FPGA boards by defining the ECP5_BOARD parameter on make:
- Lattice ECP5 Evaluation Board -
evn - Radiona ULX3S -
ulx3s - Greg Davill Orangecrab -
orangecrab - Q3k Colorlight -
colorlight
For example, to build for the Evaluation Board, run:
make ECP5_BOARD=evn synth`
and to program the FPGA:
make ECP5_BOARD=evn prog
If you connect to the serial port of the FPGA at 115200 8n1, you should see "Hello World" and after that all input will be echoed to the output. On Linux, picocom can be used. Another option below is a simple python script.
Micropython
Unfortunately due to an issue in yosys/nextpnr, dual port RAMs are not working. More details can be found in https://github.com/YosysHQ/yosys/issues/1101.
This means we use twice as much block RAM as you would expect. This also means Micropython won't fit on an ECP5 85F, because the ~400kB of available BRAM is halved to ~200k. Micropython requires 384 kB.
Once this is fixed, edit src/main/scala/Core.scala and set memory to 384 kB (384*1024):
chisel3.Driver.execute(Array[String](), () => new Core(64, 384*1024, "insns.hex", 0x0))
Then link in the micropython image:
ln -s micropython/firmware.hex insns.hex
For example, to build for the ULX3S, run:
make ECP5_BOARD=ulx3s synth`
and to program the FPGA:
make ECP5_BOARD=ulx3s prog
Simple Python script for reading USB serial port
#!/usr/bin/python
import serial
# configure the serial connections
ser = serial.Serial(
port='/dev/ttyUSB1',
baudrate=115200,
parity=serial.PARITY_NONE,
stopbits=serial.STOPBITS_ONE,
bytesize=serial.EIGHTBITS
)
# read from serial
while 1:
while ser.inWaiting() > 0:
byte = ser.read(1);
print("%s" %(byte))
Issues
Now that it is functional, we have a number of things to add:
- A few instructions
- Wishbone interconnect
- Caches
- Pipelining and bypassing