JANA4ML4FPGA

EIC R&D supported project developing ML on FPGA for streaming readout systems

How to run

Docker images

There are two images:

eicdev/ml4fpga-pre - with all dependencies but without JANA4ML4FPGA built. This image is used for CI or development reasons and is not designed for running JANA4ML4FPGA
eicdev/ml4fpga - The image can be used to run JANA4ML4FPGA

Running in docker

(extended information about options of running this docker containers)

To download/update the container

docker pull eicdev/ml4fpga:latest

Running docker:

docker run -it --rm eicdev/ml4fpga:latest bash

--rm flag Docker automatically cleans up the container and remove the file system when the container exits.
-it flag enables interactive session. Without this flag ctrl+c may not work. In general -it is used to run e.g. bash session (see below)

Extremely shallow docker crash course

Docker utilizes the pull command to download a package called an image. When Docker executes the image, it runs within a container. Each time the docker run command is called, a new container is generated. Docker creates a modifiable layer over the image and starts the container with the given command.

If the --rm flag is not used, stopping the container does not remove it. A stopped container can be restarted with all its previous changes intact using the docker start command.

Docker functions similarly to tmux or screen, allowing you to reconnect to the running image, attach multiple bash shells, and even reconnect if the container is stopped. This feature facilitates easier debugging and data retention. However, be cautious when using the --rm flag, as it removes the container upon stopping.

Docker documentation: docker run, docker start, --rm flag.

Add your system directory inside docker

You can bind any directory on your system to docker image by using -v flag:

-v <your/directory>:<docker/directory>

# /mnt is a good place to mount directories inside a container
docker run -it --rm -v /host/dir:/mnt/data eicdev/ml4fpga:latest

More information on docker bind, There are also other mechanisms of how to manage data in docker. the official documentation on managing data in docker

Debugging : To do C++ debugging (run GDB or so) one has to specify additional flags (the debugging is switched off by docker by default for security reasons):

--cap-add=SYS_PTRACE --security-opt seccomp=unconfined

Data

All EVIO DATA is here:

/gluonraid3/data4/rawdata/trd/DATA/hd_rawdata_*.evio

Physics runs:

===>  2 crates , 3 detectors : CAL/FA250, GEMTRD/FA125 , GEM/SRS
CODA::  Run_2531 GEMTRD:ok; CAL:ok; SRS:del=0x41; ROSYBC=0x90 3bin; 10APV;OK; TRDpos=100; 5.1M ev  *PHYS* !!!!
cDAQ::  Run_2543 GEMTRD:ok; CAL:ok; SRS:del=0x41; ROSYBC=0x90 3bin; 10APV;OK; TRDpos=100; 1.1M ev  *PHYS* !!!!


===> 1 crate ; 2 detectors : GEMTRD/FA125 ; GEM/SRS
CODA::  Run_2548 (CODA) GEMTRD:ok; CAL:No; SRS:del=0x40; ROSYBC=0x70 9bin; 10APV;OK; TRDpos=155; 1.5M ev  *PHYS* !!!!
cDAQ::  Run_2567 (cDAQ) GEMTRD:ok; CAL:No; SRS:del=0x40; ROSYBC=0x70 9bin; 10APV;OK; TRDpos=155; 3.2M ev  *PHYS* !!!!

===> new files Mode8 (RAW) and Mode5 (short)
cDAQ / rawmode (8) ; trd_ti_fp.conf ; 300 nA; 650-700Hz ; 16 MB/s
14:09 Run_2633 (cDAQ) GEMTRD:ok; CAL:on; SRS:del=0x41; ROSYBC=0x70 3bin; 10APV;OK; TRDpos=150; Mode8; 250K evt;

cDAQ / Mode5; thr300 / trd_ti_fp_m5.conf ; 300 nA; 650-700Hz ; 7.5 MB/s
14:48 Run_2635 (cDAQ) GEMTRD:ok; CAL:on; SRS:del=0x41; ROSYBC=0x70 3bin; 10APV;OK; TRDpos=50;  Mode5; 250 K evt;

Test setup configuration:

rocFMWPC1 is a TI master (no TSG), with single fa250 board, the last 3 channels contain data from the calorimeter.
rocTRD1 is a slave with 4 fa125 boards reading GEMTRD (bank 16) also provides SRS/GEM data - bank 17.

Example of reading evio file:

/mnt/c/eic/data/2023-02_ml4fpga_trd_data/hd_rawdata_002539_000.evio

TCP test sender and receiver

There are two test executables built with the project that could be used for testing

tcp_sender - can send dummy evio events through tcp
tcp_receiver - can receive them

To make test loop: sender sends data to receiver - drinks coffee (or what it does):

# In one terminal/process
tcp_receiver

# In another terminal/process
tcp_sender -req=ex -cmd=send -host=localhost:20249

Run flags examples

CDAQ evio example analysis

Run parameters:

jana4ml4fpga
-Pplugins=log,root_output,CDAQfile,example_evio_analysis
-Pjana:debug_plugin_loading=1
-PCDAQEVIOFileSource:LogLevel=trace
-Pexample_evio_analysis:LogLevel=trace
-Pjana:timeout=0
-Pjana:nevents=10
-Pnthreads=1
-Phistsfile=/home/romanov/eic/JANA4ML4FPGA/cmake-build-debug/test.root
/mnt/work/data/2023-03-03-trd-data/hd_rawdata_002633_000.evio

Read EVIO and save to flat root file

jana4ml4fpga
-Pplugins=log,root_output,CDAQfile,flat_tree
-Pnthreads=1
-Phistsfile=output_file_name.root
hd_rawdata_002633_000.evio          # <= input file

Useful flags

# Write extended info where plugins are loaded from  
-Pjana:debug_plugin_loading=1

# <plugin>:LogLevel usually controls verbosity level of a plugin
-Pflat_tree:LogLevel=trace

# Switches off jana internal watchdog timer
# Setting timeout=0 is needed if processing threads stops, 
# i.e. when pausing on debug breakpoint or waiting for connection, etc. 
-Pjana:timeout=0

# Set number of events to process
-Pjana:nevents=10000

# Set number of events to skip
-Pjana:nskip=10000

DraTeots/JANA4ML4FPGA