CombinationalCL

Gate-level digital circuits generator and simulator.

• CombinationalCL
  • Combinational circuit generator
  • OpenCL host platform
  • Data colector
    • Running in EC2
      • Mounting ephemeral storage
    • Running in GCP
      • Mounting ephemeral storage

Combinational circuit generator

Provided the following input arguments:

• Input count (I): The amount of input logic wires to the circuit.
• Port count (P): The amount of logic gates in the circuit, each of them corresponding to a intermediate signal (net) in the circuit.
• Output count (O): The amount of input logic wires to the circuit.
• Events count (K): The number of events to be simulated for the circuit.

This tool generates a logic circuit with I inputs, O outputs and P logic gates with random inputs and boolean logic operation. The current subset of logic operations is: AND, OR, XOR, NOT.

This tool then generates an OpenCL kernel with the correspondent operations, and a VCD file with K input dumps to be simulated.

Execute with: cargo run <P> <I> <O> <K>

Example execution: cargo run 600 20 20 1000

This tool is currently under rust/.

OpenCL host platform

Provided the arguments in the last tool and the circuit's kernel and VCD, this tool is used to simulate the generated circuit and stimuli with the OpenCL platform.

The output is a VCD file with the simulated output signals and metrics about the execution. Currently the recorded metrics are:

• I
• O
• P
• K
• CPU execution time (ns)
• GPU execution time (ns)

Execute with: ./CombinationaCL <ports count> <input count> <output count> <kernel file> <vcd file>

Example execution: ./CombinationaCL 600 20 20 out.cl out.vcd

This tool is currently under cpp/.

Data colector

This tool is meant to collect data by executing both of the previous tools with a single command.

Compile both tools: ./compile.sh

Single command execution:: ./run.sh <ports count> <inputs count> <outputs count> <events count>

Data collection execution (change parameters in code): ./collect.py

A copy of log.def.csv will be created as log.csv, and following executions will append metrics there.

This tool is currently under exec/.

Running in EC2

Tested on Ubuntu Server 18.04 LTS (HVM), SSD Volume Type - ami-0747bdcabd34c712a (64-bit x86).

Inside the exec folder, you can use:

screen -S data_collect -dm bash -c "./ec2_collect.sh"

To run the data collector in an EC2 environment. Don't forget to change <bucket> to your bucket name:

sed -i "s/<bucket>/my_bucket/g" ec2_collect.sh

You can check the progress with:

screen -S data_collect -r
# (CTRL+a, d to leave)

To stop the data collector simply run:

killall python3

Mounting ephemeral storage

First, run lsblk to confirm the efemeral storage name (in this example it was /dev/nvme1n1), then run:

sudo mkfs -t ext4 /dev/nvme1n1
mkdir ephemeral
sudo mount /dev/nvme1n1 ephemeral
sudo chown -R ubuntu ephemeral
mv combinationalcl ephemeral
cd ephemeral/combinationalcl/exec

Running in GCP

Tested on Ubuntu Server 20.04 LTS, n1-highmem-4 with Tesla K80 GPU.

To install the dependencies, you can use the script at exec/gcp_nvidia_user_data:

sudo apt install screen -y
wget -O script.sh https://raw.githubusercontent.com/northy/combinationalcl/master/exec/gcp_nvidia_user_data.sh
chmod +x script.sh
sed -i "s/<username>/$LOGNAME/g" script.sh
sudo screen -S install -m bash -c "./script.sh"

Inside the exec folder, you can use:

screen -S data_collect -dm bash -c "./gcp_collect.sh"

To run the data collector in an GCP environment. Don't forget to change <bucket> to your bucket name:

sed -i "s/<bucket>/my_bucket/g" gcp_collect.sh

You can check the progress with:

screen -S data_collect -r
# (CTRL+a, d to leave)

To stop the data collector simply run:

killall python3

Mounting ephemeral storage

First, run lsblk to confirm the efemeral storage name (in this example it was /dev/nvme0n1), then run:

sudo mkfs -t ext4 /dev/nvme0n1
mkdir ephemeral
sudo mount /dev/nvme0n1 ephemeral
sudo chown -R $LOGNAME ephemeral
mv combinationalcl ephemeral
cd ephemeral/combinationalcl/exec