Commands Reference

Use this page when you remember the task but not the exact command. Commands are grouped by workflow and link back to the module where they first appear.

Getting Around

I want to…	Command
Show the current machine name	hostname
Show my username	whoami
Show my current directory	pwd
Change into a module exercise directory	cd module-01-hpc-foundations/exercises
List files	ls
List newest output files first	ls -lt *.out | head
Print a file	cat <file>
Watch a job output file update	tail -f <file>.out

Cluster And Filesystem

I want to…	Command
Show home directory path	echo $HOME
Show work directory path	echo $WORK
Check home directory disk space	df -h $HOME
Check work directory disk space	df -h $WORK
Show memory usage	free -h
Show CPU details	lscpu
Show just the CPU model	lscpu | grep "Model name"

See Module 1 for the first cluster walkthrough.

Software Modules

I want to…	Command
List loaded modules	module list
List available modules	module avail
Inspect the tutorial base module	module show hpcfund

Note

The tutorial environment is preconfigured. Do not run module purge unless an instructor tells you to.

Slurm Jobs

I want to…	Command
Submit a batch job	sbatch script.sh
See my jobs	squeue -u $USER
See all jobs on the tutorial partition	squeue -p mi2101x
See partition and node status	sinfo -p mi2101x
See detailed node status	sinfo -p mi2101x -N -l
Get details for one job	scontrol show job <JOBID>
See accounting info after a job finishes	sacct -j <JOBID>
Cancel one job	scancel <JOBID>
Cancel all my jobs	scancel -u $USER

See Module 2 for the Slurm walkthrough.

Quick Compute-Node Commands

I want to…	Command
Run hostname on a compute node	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=1 hostname
Check CPU model on a compute node	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=1 lscpu | grep "Model name"
Preview ROCm GPU info on a compute node	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=1 rocminfo | head -30
Run a program on one compute node	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=1 ./program
Run with 4 MPI ranks	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=4 ./program
Run with 16 CPU cores for one task	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=1 --cpus-per-task=16 ./program

Slurm Script Directives

Directive	Meaning
#SBATCH --job-name=name	Name shown in the Slurm queue
#SBATCH --partition=mi2101x	Use the tutorial partition
#SBATCH --nodes=1	Request one node
#SBATCH --ntasks=1	Request one task or MPI rank
#SBATCH --cpus-per-task=16	Request CPU cores for one task
#SBATCH --time=10:00	Set the job time limit
#SBATCH --output=name_%j.out	Write stdout to a file; %j becomes the job ID
#SBATCH --error=name_%j.err	Write stderr to a file

Python Environment

I want to…	Command
Create the tutorial virtual environment from the repo root	sbatch setup/setup_venv.sh
Create the venv from inside a module exercise directory	sbatch ../../setup/setup_venv.sh
Activate the venv	source "$WORK/sc26_venv/bin/activate"
Verify PyTorch	python3 -c "import torch; print(torch.__version__)"
Verify ROCm is visible to PyTorch	python3 -c "import torch; print(torch.cuda.is_available())"
Leave the venv	deactivate

C And OpenMP

I want to…	Command
Compile a plain C program	gcc -O2 -o program program.c
Compile a C program that uses math functions	gcc -O2 -o program program.c -lm
Compile an OpenMP program	gcc -fopenmp -O2 -o program program.c -lm
Set the OpenMP thread count	export OMP_NUM_THREADS=4
Run OpenMP with 4 threads through Slurm	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=1 --cpus-per-task=16 bash -c 'export OMP_NUM_THREADS=4; ./program'

Useful OpenMP snippets from Module 3:

OpenMP construct	Purpose
#pragma omp parallel	Start a parallel region
#pragma omp for	Split loop iterations across threads
#pragma omp parallel for	Start threads and split a loop in one directive
reduction(+:sum)	Safely combine per-thread values into sum
omp_get_thread_num()	Get the current thread ID
omp_get_num_threads()	Get the size of the current thread team

MPI

I want to…	Command
Compile an MPI C program	mpicc -O2 -o program program.c
Compile an MPI C program using math functions	mpicc -O2 -o program program.c -lm
Run with 4 ranks	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=4 ./program
Run with 16 ranks	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=16 ./program

Useful MPI calls from Module 4:

MPI call	Purpose
MPI_Init(&argc, &argv)	Initialize MPI
MPI_Finalize()	Shut down MPI
MPI_Comm_rank(comm, &rank)	Get this rank’s ID
MPI_Comm_size(comm, &size)	Get the number of ranks
MPI_Send(...)	Send data to another rank
MPI_Recv(...)	Receive data from another rank
MPI_Reduce(...)	Combine values across ranks
MPI_Bcast(...)	Broadcast a value from one rank to all ranks

HIP And GPU Tools

I want to…	Command
Check whether hipcc is available	which hipcc
Show the HIP compiler version	hipcc --version
Show ROCm system info	rocminfo | head -20
Show GPU memory usage	rocm-smi --showmeminfo vram
Compile a HIP program	hipcc -O2 -o program program.cpp
Compile with a block-size macro	hipcc -O2 -DBLOCK_SIZE=256 -o program program.cpp
Run a HIP program on a compute node	srun --partition=mi2101x --nodes=1 --time=2:00 --ntasks=1 ./program

Useful HIP calls and syntax from Module 5:

HIP item	Purpose
hipMalloc(&ptr, bytes)	Allocate memory on the GPU
hipMemcpy(dst, src, bytes, dir)	Copy data between host and GPU
hipFree(ptr)	Free GPU memory
hipDeviceSynchronize()	Wait for GPU work to finish
hipGetLastError()	Check for kernel launch errors
__global__ void kernel(...)	Declare a GPU kernel
kernel<<<blocks, threads>>>(...)	Launch a kernel
threadIdx.x	Thread ID within a block
blockIdx.x	Block ID within the grid
blockDim.x	Threads per block

AI Exercises

I want to…	Command
Run the inference batch script	sbatch submit_inference.sh
Read inference output	cat inference_<JOBID>.out
Run the fine-tuning batch script	sbatch submit_finetune.sh
Read fine-tuning output	cat finetune_<JOBID>.out
Load a text generation model in Python	AutoModelForCausalLM.from_pretrained(name)
Load a tokenizer in Python	AutoTokenizer.from_pretrained(name)
Move a PyTorch model to the GPU	model.to("cuda")
Generate text with a model	model.generate(input_ids, ...)

See Module 6 for inference and fine-tuning details.

AI Agent

I want to…	Command
Show the shared agent server URL	cat "$WORK/sc26_agent_server_url"
Export the agent server URL for interactive testing	export AGENT_API_URL=$(cat "$WORK/sc26_agent_server_url")
Run the agent exercise	sbatch submit_agent.sh
Launch the tutorial-provided CLI agent	bash ../../setup/launch_aider.sh
Launch aider from an arbitrary working directory	bash <repo-path>/setup/launch_aider.sh

See Module 7 for the agent exercises.

Common Output Files

Job or exercise	Output file pattern
Venv setup	setup_venv_<JOBID>.out
Slurm first job	first-job_<JOBID>.out
Compute-node hello	hello-compute_<JOBID>.out
OpenMP pi	openmp-pi_<JOBID>.out
MPI sum	mpi-sum_<JOBID>.out
HIP exercise	hip-exercise_<JOBID>.out
Inference	inference_<JOBID>.out
Fine-tuning	finetune_<JOBID>.out
Agent exercise	agent_<JOBID>.out