Quickstart

Before starting, ensure CIMFlow is installed and your virtual environment is activated.

Verify Installation

Confirm CIMFlow is installed correctly:

# Check version
cimflow --version

Expected output:

CIMFlow 0.1.0

Running a Pipeline

This section demonstrates running a complete compile-and-simulate pipeline with an example model.

Navigate to CIMFlow Directory

cd /path/to/CIMFlow

Replace /path/to/CIMFlow with the actual installation directory.

Download Example Models

Example ONNX models are not included in the repository. Download them using the provided script:

./scripts/download_models.sh

This downloads ResNet-18 and MobileNetV2 models to data/models/.

Custom ONNX models exported from PyTorch, TensorFlow, or other frameworks are also supported.

Run Pipeline with Example Model

cimflow run pipeline \
    -m data/models/resnet18.onnx \
    -o output/quickstart \
    -t 8 \
    -k 16 \
    -b 16 \
    -c 64

This compiles ResNet-18 and simulates execution on a 64-core CIM accelerator.

View Results

After completion, check the output:

ls output/quickstart/

The output directory contains simulation results including performance metrics.

Understanding the Output

The simulation produces several key metrics:

Metric	Description
Cycles	Total clock cycles for inference
Latency	Execution time based on target clock frequency
Energy	Estimated energy consumption
Throughput	Inferences per second

Detailed breakdowns are available in the output JSON files.

Running with Custom Parameters

Adjust hardware parameters via CLI flags:

cimflow run pipeline \
    -m data/models/mobilenetv2.onnx \
    -o output/mobilenet \
    -t 16 \
    -k 32 \
    -b 8 \
    -c 128 \
    --batch-size 4

Flag	Parameter	Description
`-t`	T	Macro group size
`-k`	K	Macro group number
`-b`	B	NoC flit size (bandwidth)
`-c`	C	Core number
`--batch-size`	-	Batch size

For the complete parameter reference, see CLI Reference.

Batch Processing

Run multiple configurations from a JSON file:

Copy template:

cp config_templates/batch.json config/batch.json

Run batch:

Edit config/batch.json with your models and parameters, then run:

cimflow run from-file config/batch.json

Example configuration (update model_path to point to your ONNX model):

config/batch.json

{
  "paths_cfg": "tool_paths.json",
  "output_dir": "output",
  "keep_ir": false,
  "run_name": "batch",
  "runs": [
    {
      "model_path": "path/to/your/model.onnx",
      "t": 8,
      "k": 16,
      "b": 16,
      "c": 64,
      "batch_size": 8,
      "strategy": "dp",
      "visualize": false
    }
  ]
}

Enabling Profiling

Enable profiling for performance breakdowns:

cimflow run pipeline \
    -m data/models/resnet18.onnx \
    -o output/profiled \
    --profile=from-config

Per-Layer Timing

Execution time breakdown for each layer

Memory Bandwidth

Utilization statistics for data movement

Core Activity

Per-core utilization and idle cycles

Energy Breakdown

Energy consumption by component type

Debugging Tips

These options are useful for troubleshooting compilation or simulation issues.

Keep Intermediate Files

cimflow run pipeline -m model.onnx -o output --keep-ir

Saves all IR files, ISA outputs, and logs in timestamped directories.

Enable Debug Mode

cimflow --debug run pipeline -m model.onnx -o output

Produces detailed tracebacks and debug-level logging.

Increase Log Verbosity

cimflow run pipeline -m model.onnx -o output -l VERBOSE

CLI Quick Reference

Command	Description
`cimflow --help`	View all commands
`cimflow run pipeline --help`	Pipeline options
`cimflow run from-file --help`	Batch processing options
`cimflow compile --help`	Compile only
`cimflow sim --help`	Simulate only

Next Steps

Examples

Tutorials and use cases

Compiler

Two-level compilation architecture

Simulator

Cycle-accurate simulation model

CLI Reference

Command-line documentation

Quickstart

Examples

Compiler

Simulator

CLI Reference

On this page