jetson-power-monitor

English | 中文

A comprehensive power monitoring library for NVIDIA Jetson devices, available in multiple programming languages.

Features

Real-time power consumption monitoring
Support for multiple programming languages (C/C++, Rust, Python)
Easy installation through package managers
Low-level access to power metrics
Cross-platform support for Jetson devices

Installation

Python

pip install jetpwmon

Rust

Add to your Cargo.toml:

[dependencies]
jetpwmon = "0.1.2"

C/C++

Download the pre-built .deb package from the Releases page:

sudo dpkg -i jetpwmon_0.1.2_amd64.deb

Or use CMake to find and link the library in your project:

find_package(jetpwmon REQUIRED)
target_link_libraries(your_target PRIVATE jetpwmon::jetpwmon)  # Use shared library
# or
target_link_libraries(your_target PRIVATE jetpwmon::static)    # Use static library

# For C++ bindings
target_link_libraries(your_target PRIVATE jetpwmon::jetpwmon_cpp)  # Use shared library
# or
target_link_libraries(your_target PRIVATE jetpwmon::static_cpp)    # Use static library

Usage

Python

Quick Start: Get Instantaneous Readings

This shows how to get the current total power consumption, voltage, and current readings directly from the device.

import jetpwmon

# Create a power monitor instance
monitor = jetpwmon.PowerMonitor()

# Get the latest data snapshot
try:
    data = monitor.get_latest_data()
    
    # Access total readings
    total = data['total']
    print(f"Current total power consumption: {total['power']:.2f} W")
    print(f"Current bus voltage: {total['voltage']:.2f} V")
    print(f"Current total current: {total['current']:.2f} A")
    
    # Access individual sensor readings
    print("\nIndividual Sensor Readings:")
    for sensor in data['sensors']:
        print(f"Sensor {sensor['name']}:")
        print(f"  Power: {sensor['power']:.2f} W")
        print(f"  Voltage: {sensor['voltage']:.2f} V")
        print(f"  Current: {sensor['current']:.2f} A")

except Exception as e:
    print(f"Error reading power metrics: {e}")
    print("Ensure the INA3221 device is connected and accessible (permissions?).")

Advanced: Monitor Power During a Task

This example demonstrates how to start background power sampling before a task, stop it afterwards, and retrieve detailed statistics (min, max, average power, total energy) for the monitoring period.

import jetpwmon
import time
import numpy as np # Using numpy for a sample CPU-intensive task

def cpu_intensive_task():
    """Simulate a CPU-intensive task"""
    print("Starting CPU-intensive task...")
    # Reduced size for a quicker example run
    size = 2000
    matrix1 = np.random.rand(size, size)
    matrix2 = np.random.rand(size, size)
    # Perform matrix multiplication
    result = np.dot(matrix1, matrix2)
    print("CPU-intensive task completed.")

def monitor_power_consumption(task_func):
    """Monitor power consumption during task execution"""
    # Create a power monitor instance
    monitor = jetpwmon.PowerMonitor()

    try:
        # Optional: Set the sampling frequency (e.g., 1000Hz)
        # Higher frequencies provide more granular data but increase overhead.
        # Check library documentation or device limits for valid/optimal values.
        monitor.set_sampling_frequency(1000)

        # Reset statistics before starting a new monitoring period
        monitor.reset_statistics()

        # Start background sampling
        print("Starting power sampling...")
        monitor.start_sampling()

        # --- Execute the task you want to monitor ---
        task_func()
        # --- Task finished ---

        # Optional: Wait briefly to ensure last samples are captured,
        # depends on task duration and sampling frequency.
        time.sleep(0.1)

        # Stop background sampling
        monitor.stop_sampling()
        print("Stopped power sampling.")

        # Get collected statistics
        stats = monitor.get_statistics()

        # --- Print the collected statistics ---
        print("\n--- Power Consumption Statistics ---")

        # Print total power consumption statistics
        if 'total' in stats and 'power' in stats['total']:
            total_stats = stats['total']['power']
            print("Total Power Consumption:")
            # Use .get() for safety in case some stats weren't computed
            print(f"  Minimum Value: {total_stats.get('min', float('nan')):.2f} W")
            print(f"  Maximum Value: {total_stats.get('max', float('nan')):.2f} W")
            print(f"  Average Value: {total_stats.get('avg', float('nan')):.2f} W")
            print(f"  Total Energy: {total_stats.get('total', float('nan')):.2f} J")
            print(f"  Sample Count: {total_stats.get('count', 0)}")
        else:
            print("Total power statistics not available.")

        # Print power consumption information for each sensor/channel
        if 'sensors' in stats:
            print("\nPower Consumption Per Sensor:")
            for sensor in stats['sensors']:
                 if 'power' in sensor:
                     sensor_stats = sensor['power']
                     print(f"\n  Sensor: {sensor.get('name', 'Unknown')}")
                     print(f"    Minimum Value: {sensor_stats.get('min', float('nan')):.2f} W")
                     print(f"    Maximum Value: {sensor_stats.get('max', float('nan')):.2f} W")
                     print(f"    Average Value: {sensor_stats.get('avg', float('nan')):.2f} W")
                     print(f"    Total Energy: {sensor_stats.get('total', float('nan')):.2f} J")
                     print(f"    Sample Count: {sensor_stats.get('count', 0)}")
        else:
             print("\nPer-sensor statistics not available.")

    except Exception as e:
        print(f"\nAn error occurred during monitoring: {e}")
        print("Ensure the INA3221 device is connected and accessible.")

# --- Run the monitoring example ---
print("Jetson Power Monitor Example Program")
print("===================================")
monitor_power_consumption(cpu_intensive_task)

Rust

First, add jetpwmon as a dependency in your Cargo.toml. Adjust the path or version as needed.

[dependencies]
jetpwmon = "0.1.2"

# The examples also use these crates:
ndarray = "0.15" # For matrix example
rand = "0.8"     # For matrix example

Quick Start: Get Latest Sensor Readings

This example shows how to initialize the monitor and get a single snapshot of the current power, voltage, and current, for both the total and individual sensors. Note the use of unsafe to access per-sensor data returned via raw pointers.

use jetpwmon::{PowerMonitor, PowerData, SensorData, Error};
use std::slice;

fn main() -> Result<(), Error> {
    println!("Jetson Power Monitor - Rust Quick Start");
    println!("======================================");

    // Initialize the power monitor. This connects to the hardware.
    // The '?' operator propagates any errors (like device not found).
    let monitor = PowerMonitor::new()?;
    println!("Power monitor initialized successfully.");

    // Get the latest instantaneous data snapshot
    let data: PowerData = monitor.get_latest_data()?;

    // --- Access Total Aggregated Data (Safely) ---
    // The 'total' field is a regular struct within PowerData.
    println!("\n--- Total Readings ---");
    println!("Total Power: {:.2} W", data.total.power);
    println!("Bus Voltage: {:.2} V", data.total.voltage); // Often VIN
    println!("Total Current: {:.2} A", data.total.current);
    println!("Status: {}", String::from_utf8_lossy(&data.total.status).trim_matches('\0'));

    // --- Access Individual Sensor Data (Requires Unsafe) ---
    // 'data.sensors' is a raw pointer (*mut SensorData) from C.
    // We need an unsafe block to dereference it and create a safe slice.
    println!("\n--- Individual Sensor Readings ---");
    if !data.sensors.is_null() && data.sensor_count > 0 {
        // Create a safe slice from the raw pointer and count
        // SAFETY: Assumes the C library guarantees that 'data.sensors' points to valid memory
        // containing 'data.sensor_count' elements, and that this memory remains valid
        // at least for the lifetime of the 'data' variable returned by get_latest_data().
        let sensors_slice: &[SensorData] = unsafe {
            slice::from_raw_parts(data.sensors, data.sensor_count as usize)
        };

        // Now iterate over the safe slice
        for sensor in sensors_slice {
            // Convert the fixed-size u8 array (C string) to a Rust String
            // Using from_utf8_lossy is safer as it handles potential invalid UTF-8 bytes.
            let name = String::from_utf8_lossy(&sensor.name)
                           .trim_matches('\0') // Remove null padding/terminator
                           .to_string();
            let status = String::from_utf8_lossy(&sensor.status).trim_matches('\0').to_string();

            println!(
                "  Sensor: {:<15} | Pwr: {:>6.2} W | V: {:>5.2} V | I: {:>6.2} A | Online: {} | Status: {}",
                name, sensor.power, sensor.voltage, sensor.current, sensor.online, status
            );
        }
    } else {
        println!("  No individual sensor data available or pointer was null.");
    }

    // No explicit cleanup needed. The `PowerMonitor` struct implements the `Drop` trait,
    // which automatically calls the C cleanup function when `monitor` goes out of scope.
    println!("\nMonitor will be cleaned up automatically.");
    Ok(())
}

Advanced: Monitor Power During a Task

This example demonstrates starting background sampling, running a CPU-intensive task (matrix multiplication across threads), stopping sampling, and retrieving detailed statistics. It highlights error handling with Result and the necessary unsafe block for accessing per-sensor statistics.

Dependencies needed for this example:

[dependencies]
jetpwmon = { version = "0.1.2" } # Adjust as needed
ndarray = "0.15"
rand = "0.8"

use jetpwmon::{PowerMonitor, PowerStats, SensorStats, Error};
use std::{thread, time::Duration, slice, error::Error as StdError};
use ndarray::Array2;
use rand::Rng;

// Example task parameters (adjust as needed)
const MATRIX_SIZE: usize = 1000; // Size of matrices
const NUM_THREADS: usize = 4;    // Number of concurrent tasks
const NUM_ITERATIONS: usize = 5; // Workload per thread

/// Example CPU-intensive task using ndarray for matrix multiplication
fn matrix_multiply_task(thread_id: usize) {
    // println!("Thread {} starting...", thread_id); // Optional logging
    let mut rng = rand::thread_rng();
    // Create large matrices filled with random data
    let mut a: Array2<f64> = Array2::from_shape_fn((MATRIX_SIZE, MATRIX_SIZE), |_| rng.gen());
    let b: Array2<f64> = Array2::from_shape_fn((MATRIX_SIZE, MATRIX_SIZE), |_| rng.gen());

    // Perform repeated multiplications
    for _ in 0..NUM_ITERATIONS {
        a = a.dot(&b); // Matrix multiplication
    }
    // println!("Thread {} finished.", thread_id); // Optional logging
}

// Use Box<dyn StdError> for flexible error handling in main
fn main() -> Result<(), Box<dyn StdError>> {
    println!("Jetson Power Monitor - Rust Monitoring Example");
    println!("==========================================");

    // Initialize the power monitor
    let monitor = PowerMonitor::new()?; // Propagate errors using '?'
    println!("Power monitor initialized.");

    // Set the desired sampling frequency (e.g., 1000 Hz)
    let frequency = 1000;
    monitor.set_sampling_frequency(frequency)?;
    println!("Set sampling frequency to {} Hz.", frequency);

    // Reset any previously collected statistics
    monitor.reset_statistics()?;
    println!("Reset statistics.");

    // Start background sampling in a separate thread (managed by the C library)
    monitor.start_sampling()?;
    println!("Started power sampling...");

    // Record task start time
    let task_start_time = std::time::Instant::now();

    // --- Run the CPU-intensive task across multiple threads ---
    let mut handles = vec![];
    for i in 0..NUM_THREADS {
        let handle = thread::spawn(move || {
            matrix_multiply_task(i);
        });
        handles.push(handle);
    }
    // Wait for all threads to complete
    for handle in handles {
        handle.join().expect("Task thread panicked!");
    }
    // --- Task finished ---

    let task_duration = task_start_time.elapsed();
    println!("\nTask execution finished in: {:.2?}", task_duration);

    // Allow a brief moment for the last samples to be collected by the background thread
    thread::sleep(Duration::from_millis(100)); // Adjust if needed

    // Stop the background sampling thread
    monitor.stop_sampling()?;
    println!("Stopped power sampling.");

    // Retrieve the collected statistics
    let stats: PowerStats = monitor.get_statistics()?;

    // --- Print the Statistics ---
    println!("\n--- Power Consumption Statistics ---");

    // Print total aggregated statistics
    println!("Total Power Consumption:");
    println!("  Min Power: {:.2} W", stats.total.power.min);
    println!("  Max Power: {:.2} W", stats.total.power.max);
    println!("  Avg Power: {:.2} W", stats.total.power.avg);
    println!("  Total Energy: {:.2} J", stats.total.power.total); // Energy = Avg Power * Duration
    println!("  Sample Count: {}", stats.total.power.count);
    // You can also access stats.total.voltage and stats.total.current if needed

    // Print per-sensor statistics (requires unsafe)
    println!("\nPer-Sensor Power Consumption:");
    if !stats.sensors.is_null() && stats.sensor_count > 0 {
        // Create a safe slice from the raw pointer and count
        // SAFETY: Assumes C library guarantees pointer validity for the lifetime of 'stats'.
        let sensor_stats_slice: &[SensorStats] = unsafe {
            slice::from_raw_parts(stats.sensors, stats.sensor_count as usize)
        };

        // Iterate over the safe slice
        for sensor_stat in sensor_stats_slice {
            // Convert C char array name to Rust String
            let name = String::from_utf8_lossy(&sensor_stat.name)
                           .trim_matches('\0')
                           .to_string();

            println!("\n  Sensor: {}", name);
            println!("    Min Power: {:.2} W", sensor_stat.power.min);
            println!("    Max Power: {:.2} W", sensor_stat.power.max);
            println!("    Avg Power: {:.2} W", sensor_stat.power.avg);
            println!("    Total Energy: {:.2} J", sensor_stat.power.total);
            println!("    Sample Count: {}", sensor_stat.power.count);
            // You can also access sensor_stat.voltage and sensor_stat.current if needed
        }
    } else {
        println!("  No per-sensor statistics available or pointer was null.");
    }

    println!("\nMonitoring complete. Resources will be cleaned up.");
    // `monitor` goes out of scope here, Drop trait calls pm_cleanup()
    Ok(())
}

C/C++

Compilation

Include Header: Add the following line to your C source files:

#include <jetpwmon/jetpwmon.h>
// Or adjust the path based on your project structure:
// #include "path/to/include/jetpwmon/jetpwmon.h"

Link Library: When compiling, you need to link against the libjetpwmon library. Assuming the library and header files are installed in standard system paths or paths specified via -L and -I:

# Basic compilation
gcc your_program.c -o your_program -ljetpwmon

# If library/includes are in custom locations:
# gcc your_program.c -o your_program -I/path/to/jetpwmon/include -L/path/to/jetpwmon/lib -ljetpwmon

# You can use pkg-config to find the library and include paths:
# gcc your_program.c -o your_program -ljetpwmon `pkg-config --libs --cflags jetpwmon`

# Add other libraries if needed (like pthread for threading, m for math, omp for OpenMP)
# Example with OpenMP (like the advanced example below):
# gcc your_program.c -o your_program -I/path/to/include -L/path/to/lib -ljetpwmon -fopenmp -lm

Quick Start: Get Latest Sensor Readings

This example demonstrates the basic lifecycle: initialize the library, get a single snapshot of current sensor readings, print them, and perform the mandatory cleanup.

#include <stdio.h>
#include <stdlib.h> // For EXIT_FAILURE
#include <jetpwmon/jetpwmon.h> // Adjust path if necessary

int main() {
    pm_handle_t handle = NULL;     // Opaque handle for the library instance
    pm_power_data_t current_data; // Struct to hold the results
    pm_error_t err;                // Variable to store error codes

    // 1. Initialize the library
    // pm_init allocates resources and discovers sensors.
    // It stores the handle needed for subsequent calls in 'handle'.
    err = pm_init(&handle);
    if (err != PM_SUCCESS) {
        // Use pm_error_string to get a readable error message
        fprintf(stderr, "ERROR: Failed to initialize jetpwmon: %s (code: %d)\n", pm_error_string(err), err);
        return EXIT_FAILURE;
    }
    printf("Library initialized successfully.\n");

    // 2. Get the latest data snapshot
    // Pass the address of the struct; the library fills it.
    // The 'current_data.sensors' pointer will point to an internal library buffer.
    err = pm_get_latest_data(handle, &current_data);
    if (err != PM_SUCCESS) {
        fprintf(stderr, "ERROR: Failed to get latest data: %s (code: %d)\n", pm_error_string(err), err);
        pm_cleanup(handle); // Clean up resources before exiting on error
        return EXIT_FAILURE;
    }

    // 3. Print Total Aggregated Readings
    printf("\n--- Total Readings ---\n");
    printf("Total Power  : %.2f W\n", current_data.total.power);
    printf("Bus Voltage  : %.2f V\n", current_data.total.voltage);
    printf("Total Current: %.2f A\n", current_data.total.current);
    // Assume 'status' is a null-terminated C string
    printf("Status       : %s\n", current_data.total.status);

    // 4. Print Individual Sensor Readings
    printf("\n--- Individual Sensor Readings ---\n");
    // Check if the sensors pointer is valid and count is positive
    // The 'current_data.sensors' pointer is managed by the library and points
    // to 'current_data.sensor_count' elements. It's typically valid until
    // the next call to a library function that modifies this data or pm_cleanup.
    if (current_data.sensors != NULL && current_data.sensor_count > 0) {
        for (int i = 0; i < current_data.sensor_count; ++i) {
            // Access data using array indexing on the pointer
            pm_sensor_data_t* sensor = &current_data.sensors[i];
            printf("  Sensor: %-15s | Pwr: %6.2f W | V: %5.2f V | I: %6.2f A | Online: %s | Status: %s\n",
                   sensor->name,    // Assumes null-terminated string
                   sensor->power,
                   sensor->voltage,
                   sensor->current,
                   sensor->online ? "Yes" : "No",
                   sensor->status); // Assumes null-terminated string
        }
    } else {
        printf("  No individual sensor data available.\n");
    }

    // 5. Clean up library resources (MANDATORY)
    // This releases memory and stops any running background threads.
    err = pm_cleanup(handle);
    if (err != PM_SUCCESS) {
        // Log the error, but the program should still terminate.
        fprintf(stderr, "ERROR: Failed to clean up jetpwmon cleanly: %s (code: %d)\n", pm_error_string(err), err);
        return EXIT_FAILURE; // Indicate an error occurred during cleanup
    }
    printf("\nLibrary resources cleaned up successfully.\n");

    return 0; // Success
}

Advanced: Monitor Power During a Task

This example demonstrates the complete workflow for monitoring power consumption during a specific task. It initializes the library, configures and starts sampling, executes a CPU-intensive task (using OpenMP for parallelization), stops sampling, retrieves the collected statistics, prints them, and cleans up.

Note: Compile this example with OpenMP support enabled (e.g., gcc -fopenmp ...)

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h> // For usleep()
#include <time.h>   // For timing task using clock_gettime
#include <omp.h>    // For OpenMP parallel task example
#include <jetpwmon/jetpwmon.h> // Adjust path if necessary

// Example CPU-intensive task (Matrix Multiplication using OpenMP)
void cpu_intensive_task() {
    printf("Starting CPU-intensive task...\n");
    const int size = 1500; // Moderate size for example
    double *matrix1 = NULL, *matrix2 = NULL, *result = NULL;

    // Allocate memory for matrices
    matrix1 = (double*)malloc(size * size * sizeof(double));
    matrix2 = (double*)malloc(size * size * sizeof(double));
    result = (double*)malloc(size * size * sizeof(double));
    if (!matrix1 || !matrix2 || !result) {
        fprintf(stderr, "Task ERROR: Failed to allocate memory for matrices.\n");
        // Free any potentially allocated memory before returning
        free(matrix1);
        free(matrix2);
        free(result);
        return; // Exit the task function on allocation failure
    }

    // Initialize matrices with random data using OpenMP parallel for
    #pragma omp parallel for
    for (int i = 0; i < size * size; i++) {
        // Note: rand() is not thread-safe, but for a demo it might be acceptable.
        // For production, use thread-safe RNGs or seed per thread.
        matrix1[i] = (double)rand() / RAND_MAX;
        matrix2[i] = (double)rand() / RAND_MAX;
    }

    // Perform matrix multiplication using OpenMP parallel for with collapse
    #pragma omp parallel for collapse(2)
    for (int i = 0; i < size; i++) {
        for (int j = 0; j < size; j++) {
            double sum = 0.0;
            for (int k = 0; k < size; k++) {
                sum += matrix1[i * size + k] * matrix2[k * size + j];
            }
            result[i * size + j] = sum;
        }
    }

    // Free allocated memory
    free(matrix1);
    free(matrix2);
    free(result);
    printf("CPU-intensive task completed.\n");
}

int main() {
    pm_handle_t handle = NULL;     // Library instance handle
    pm_power_stats_t stats;        // Struct to hold collected statistics
    pm_error_t err;                // Error code variable
    struct timespec start_time, end_time; // For timing the task execution

    printf("Jetson Power Monitor - C Monitoring Example\n");
    printf("=========================================\n");

    // 1. Initialize the library
    err = pm_init(&handle);
    if (err != PM_SUCCESS) {
        fprintf(stderr, "ERROR: Initialization failed: %s\n", pm_error_string(err));
        return EXIT_FAILURE;
    }
    printf("Library initialized.\n");

    // 2. Set Sampling Frequency (e.g., 1000 Hz)
    int target_frequency = 1000;
    err = pm_set_sampling_frequency(handle, target_frequency);
    if (err != PM_SUCCESS) {
        fprintf(stderr, "ERROR: Failed to set sampling frequency: %s\n", pm_error_string(err));
        pm_cleanup(handle); // Cleanup before exit
        return EXIT_FAILURE;
    }
    printf("Set sampling frequency to %d Hz.\n", target_frequency);

    // 3. Reset Statistics (Recommended before starting a measurement interval)
    err = pm_reset_statistics(handle);
    if (err != PM_SUCCESS) {
        fprintf(stderr, "ERROR: Failed to reset statistics: %s\n", pm_error_string(err));
        pm_cleanup(handle);
        return EXIT_FAILURE;
    }
    printf("Statistics reset.\n");

    // 4. Start Background Sampling
    // This starts a thread managed by the library to collect data.
    err = pm_start_sampling(handle);
    if (err != PM_SUCCESS) {
        fprintf(stderr, "ERROR: Failed to start sampling: %s\n", pm_error_string(err));
        pm_cleanup(handle);
        return EXIT_FAILURE;
    }
    printf("Started power sampling...\n");

    // --- Execute the task to be monitored ---
    clock_gettime(CLOCK_MONOTONIC, &start_time); // Get time before task
    cpu_intensive_task();                        // Run the actual task
    clock_gettime(CLOCK_MONOTONIC, &end_time);   // Get time after task
    // --- Task Finished ---

    // Calculate task duration
    double task_duration_sec = (end_time.tv_sec - start_time.tv_sec) +
                               (end_time.tv_nsec - start_time.tv_nsec) / 1e9;
    printf("\nTask execution finished in: %.3f seconds\n", task_duration_sec);

    // Optional: Wait a short moment to ensure the sampling thread captures final moments.
    // Adjust duration based on sampling frequency and task nature.
    usleep(100 * 1000); // 100 milliseconds

    // 5. Stop Background Sampling
    err = pm_stop_sampling(handle);
    if (err != PM_SUCCESS) {
        // Log error, but proceed to get statistics if possible
        fprintf(stderr, "WARNING: Failed to stop sampling cleanly: %s\n", pm_error_string(err));
    } else {
        printf("Stopped power sampling.\n");
    }

    // 6. Get Collected Statistics
    // Pass address of 'stats' struct; library fills it.
    // 'stats.sensors' will point to an internal buffer managed by the library.
    err = pm_get_statistics(handle, &stats);
    if (err != PM_SUCCESS) {
        fprintf(stderr, "ERROR: Failed to get statistics: %s\n", pm_error_string(err));
        pm_cleanup(handle);
        return EXIT_FAILURE;
    }

    // --- Print Statistics ---
    printf("\n--- Power Consumption Statistics ---\n");
    printf("Total Power Consumption:\n");
    printf("  Min Power   : %.2f W\n", stats.total.power.min);
    printf("  Max Power   : %.2f W\n", stats.total.power.max);
    printf("  Avg Power   : %.2f W\n", stats.total.power.avg);
    printf("  Total Energy: %.2f J (Avg Power * Duration)\n", stats.total.power.total);
    printf("  Sample Count: %lu\n", stats.total.power.count);
    // Access total voltage/current stats via stats.total.voltage.* etc.

    printf("\nPer-Sensor Power Consumption:\n");
    if (stats.sensors != NULL && stats.sensor_count > 0) {
        for (int i = 0; i < stats.sensor_count; ++i) {
            pm_sensor_stats_t* sensor_stat = &stats.sensors[i];
            // Assume 'name' is a null-terminated C string
            printf("\n  Sensor: %s\n", sensor_stat->name);
            printf("    Min Power   : %.2f W\n", sensor_stat->power.min);
            printf("    Max Power   : %.2f W\n", sensor_stat->power.max);
            printf("    Avg Power   : %.2f W\n", sensor_stat->power.avg);
            printf("    Total Energy: %.2f J\n", sensor_stat->power.total);
            printf("    Sample Count: %lu\n", sensor_stat->power.count);
             // Access per-sensor voltage/current stats via sensor_stat->voltage.* etc.
        }
    } else {
        printf("  No per-sensor statistics available.\n");
    }

    // 7. Clean up library resources (MANDATORY)
    err = pm_cleanup(handle);
    if (err != PM_SUCCESS) {
        fprintf(stderr, "ERROR: Cleanup failed: %s\n", pm_error_string(err));
        return EXIT_FAILURE; // Indicate failure on cleanup error
    }
    printf("\nLibrary resources cleaned up successfully.\n");

    return 0; // Success
}

C++

Compilation

Include Header: Use the C++ wrapper header file in your source code:

#include <jetpwmon/jetpwmon++.hpp> // Use the C++ header
#include <stdexcept> // For catching exceptions
#include <iostream>  // For printing

Link Library: Compile your C++ code (ensuring C++14 or later standard is enabled) and link against the underlying libjetpwmon C library:

# Compile using g++ with C++14 support
g++ your_program.cpp -o your_program -std=c++14 -ljetpwmon

# If library/includes are in custom locations:
# g++ your_program.cpp -o your_program -std=c++14 -I/path/to/jetpwmon/include -L/path/to/jetpwmon/lib -ljetpwmon

# Add other necessary flags (e.g., -pthread for std::thread, Eigen paths/libs)
# g++ your_program.cpp -o your_program -std=c++14 -I/path/to/eigen -I/path/to/include -L/path/to/lib -ljetpwmon -pthread

Key Features of the C++ Wrapper:

RAII (Resource Acquisition Is Initialization): The jetpwmon::PowerMonitor object automatically initializes the library (pm_init) on creation and cleans up resources (pm_cleanup) upon destruction (when it goes out of scope). No manual cleanup calls are needed.
Exception Safety: C API errors are converted into std::runtime_error exceptions, allowing for standard C++ error handling using try...catch blocks.

Quick Start: Get Latest Sensor Readings (C++ Wrapper)

This example demonstrates initializing the monitor using the C++ wrapper, getting a snapshot of current readings, printing them, and letting RAII handle cleanup.

#include <jetpwmon/jetpwmon++.hpp> // C++ wrapper header
#include <iostream>
#include <vector>
#include <string>
#include <stdexcept> // For std::runtime_error
#include <cstring>   // For strnlen

// Helper function to safely convert C char array (potentially not null-terminated) to std::string
std::string c_char_to_string(const char* c_str, size_t max_len) {
    // Find the actual length of the string or stop at max_len
    size_t len = strnlen(c_str, max_len);
    return std::string(c_str, len);
}

int main() {
    try {
        // 1. Initialize: Create PowerMonitor object.
        // Constructor handles pm_init() and throws std::runtime_error on failure.
        jetpwmon::PowerMonitor monitor;
        std::cout << "Power monitor initialized successfully (RAII)." << std::endl;

        // 2. Get Latest Data: Returns a jetpwmon::PowerData object.
        // This object holds the data snapshot.
        jetpwmon::PowerData data = monitor.getLatestData(); // Throws on C API error

        // 3. Access and Print Total Readings
        // Use the getTotal() method which returns a const reference to the C struct.
        const pm_sensor_data_t& total_data = data.getTotal();
        std::cout << "\n--- Total Readings ---" << std::endl;
        std::cout << "Total Power  : " << total_data.power << " W" << std::endl;
        std::cout << "Bus Voltage  : " << total_data.voltage << " V" << std::endl;
        std::cout << "Total Current: " << total_data.current << " A" << std::endl;
        // Safely convert C char array status field
        std::cout << "Status       : " << c_char_to_string(total_data.status, sizeof(total_data.status)) << std::endl;

        // 4. Access and Print Individual Sensor Readings
        std::cout << "\n--- Individual Sensor Readings ---" << std::endl;
        const pm_sensor_data_t* sensors_ptr = data.getSensors(); // Get raw C pointer
        int sensor_count = data.getSensorCount();

        // SAFETY NOTE: The pointer from getSensors() points to memory managed
        // by the underlying C library, assumed valid only temporarily. Access promptly.
        if (sensors_ptr != nullptr && sensor_count > 0) {
            for (int i = 0; i < sensor_count; ++i) {
                const pm_sensor_data_t& sensor = sensors_ptr[i]; // Access via pointer
                std::cout << "  Sensor: " << c_char_to_string(sensor.name, sizeof(sensor.name))
                          << " | Pwr: " << sensor.power << " W"
                          << " | V: " << sensor.voltage << " V"
                          << " | I: " << sensor.current << " A"
                          << " | Online: " << (sensor.online ? "Yes" : "No")
                          << " | Status: " << c_char_to_string(sensor.status, sizeof(sensor.status))
                          << std::endl;
            }
        } else {
            std::cout << "  No individual sensor data available." << std::endl;
        }

        // 5. Cleanup is Automatic!
        // When 'monitor' goes out of scope at the end of 'main' (or the try block),
        // its destructor is called, which automatically calls pm_cleanup().
        std::cout << "\nExiting scope. PowerMonitor destructor will handle cleanup." << std::endl;

    } catch (const std::runtime_error& e) {
        // Catch errors thrown by the PowerMonitor wrapper
        std::cerr << "ERROR: " << e.what() << std::endl;
        return 1; // Indicate failure
    } catch (const std::exception& e) {
        // Catch any other standard exceptions
        std::cerr << "An unexpected error occurred: " << e.what() << std::endl;
        return 1;
    }

    return 0; // Success
}

Advanced: Monitor Power During a Task (C++ Wrapper)

This example uses the jetpwmon::PowerMonitor C++ wrapper, std::thread, and exception handling to monitor power consumption during a parallel matrix multiplication task (using Eigen).

Example Dependencies: Eigen library, C++11 thread support (-pthread).

#include <jetpwmon/jetpwmon++.hpp> // C++ wrapper
#include <iostream>
#include <vector>
#include <string>
#include <thread>         // Use std::thread
#include <chrono>         // For timing and sleep
#include <stdexcept>      // For exception handling
#include <Eigen/Dense>    // For Eigen matrix task
#include <cstring>        // For strnlen

// Example Task Parameters
const int MATRIX_SIZE = 1000; // Adjust based on system memory/CPU
const int NUM_THREADS = 4;    // Number of threads for the task
const int NUM_ITERATIONS = 5; // Workload per thread

// Helper function to safely convert C char array to std::string
std::string c_char_to_string(const char* c_str, size_t max_len) {
    size_t len = strnlen(c_str, max_len);
    return std::string(c_str, len);
}

// Example CPU-intensive task using Eigen library
void eigen_matrix_task(int thread_id) {
    // std::cout << "Thread " << thread_id << " starting Eigen task..." << std::endl;
    Eigen::MatrixXd a = Eigen::MatrixXd::Random(MATRIX_SIZE, MATRIX_SIZE);
    Eigen::MatrixXd b = Eigen::MatrixXd::Random(MATRIX_SIZE, MATRIX_SIZE);
    for (int i = 0; i < NUM_ITERATIONS; ++i) {
        a = a * b; // Perform matrix multiplication
    }
    // std::cout << "Thread " << thread_id << " finished Eigen task." << std::endl;
}

int main() {
    try {
        // 1. Initialize PowerMonitor (RAII handles pm_init)
        jetpwmon::PowerMonitor monitor;
        std::cout << "Power monitor initialized." << std::endl;

        // 2. Configure Sampling
        int frequency = 1000;
        monitor.setSamplingFrequency(frequency); // Throws on error
        std::cout << "Set sampling frequency to " << frequency << " Hz." << std::endl;

        // 3. Reset Statistics
        monitor.resetStatistics(); // Throws on error
        std::cout << "Statistics reset." << std::endl;

        // 4. Start Background Sampling
        monitor.startSampling(); // Throws on error
        std::cout << "Started power sampling..." << std::endl;

        // --- Execute the Parallel Task ---
        auto task_start_time = std::chrono::high_resolution_clock::now();

        std::vector<std::thread> task_threads;
        task_threads.reserve(NUM_THREADS);
        for (int i = 0; i < NUM_THREADS; ++i) {
            // Use std::thread for C++ concurrency
            task_threads.emplace_back(eigen_matrix_task, i);
        }
        // Wait for all task threads to complete
        for (auto& t : task_threads) {
            if (t.joinable()) {
                t.join();
            }
        }

        auto task_end_time = std::chrono::high_resolution_clock::now();
        std::chrono::duration<double> task_duration = task_end_time - task_start_time;
        std::cout << "\nTask execution finished in: " << task_duration.count() << " seconds" << std::endl;
        // --- Task Finished ---

        // Optional pause for final sample collection
        std::this_thread::sleep_for(std::chrono::milliseconds(100));

        // 5. Stop Background Sampling
        monitor.stopSampling(); // Throws on error
        std::cout << "Stopped power sampling." << std::endl;

        // 6. Get Collected Statistics
        // Returns a jetpwmon::PowerStats object.
        jetpwmon::PowerStats stats = monitor.getStatistics(); // Throws on error

        // --- Print Statistics ---
        std::cout << "\n--- Power Consumption Statistics ---" << std::endl;
        const pm_sensor_stats_t& total_stats = stats.getTotal();
        std::cout << "Total Power Consumption:" << std::endl;
        std::cout << "  Min Power   : " << total_stats.power.min << " W" << std::endl;
        std::cout << "  Max Power   : " << total_stats.power.max << " W" << std::endl;
        std::cout << "  Avg Power   : " << total_stats.power.avg << " W" << std::endl;
        std::cout << "  Total Energy: " << total_stats.power.total << " J" << std::endl;
        std::cout << "  Sample Count: " << total_stats.power.count << std::endl;

        std::cout << "\nPer-Sensor Power Consumption:" << std::endl;
        const pm_sensor_stats_t* sensors_stats_ptr = stats.getSensors();
        int sensor_count = stats.getSensorCount();

        // SAFETY NOTE: Access pointer promptly, assumes temporary validity.
        if (sensors_stats_ptr != nullptr && sensor_count > 0) {
            for (int i = 0; i < sensor_count; ++i) {
                const pm_sensor_stats_t& sensor_stat = sensors_stats_ptr[i];
                std::cout << "\n  Sensor: " << c_char_to_string(sensor_stat.name, sizeof(sensor_stat.name)) << std::endl;
                std::cout << "    Min Power   : " << sensor_stat.power.min << " W" << std::endl;
                std::cout << "    Max Power   : " << sensor_stat.power.max << " W" << std::endl;
                std::cout << "    Avg Power   : " << sensor_stat.power.avg << " W" << std::endl;
                std::cout << "    Total Energy: " << sensor_stat.power.total << " J" << std::endl;
                std::cout << "    Sample Count: " << sensor_stat.power.count << std::endl;
            }
        } else {
            std::cout << "  No per-sensor statistics available." << std::endl;
        }

        // 7. Cleanup is Automatic (RAII)
        std::cout << "\nMonitoring complete. Resources automatically cleaned up." << std::endl;
        // 'monitor' destructor called automatically when main returns.

    } catch (const std::runtime_error& e) {
        std::cerr << "ERROR: " << e.what() << std::endl;
        return 1;
    } catch (const std::exception& e) {
        std::cerr << "An unexpected error occurred: " << e.what() << std::endl;
        return 1;
    }

    return 0;
}

API Documentation

Python

API Reference

Here are the primary methods available on the PowerMonitor class:

class PowerMonitor:
    def __init__(self) -> None:
        """
        Initializes the connection to the power monitor hardware (e.g., INA3221 via I2C).
        May raise an exception if the device cannot be found or accessed.
        """
        pass # Actual implementation omitted

    def get_power_consumption(self) -> float:
        """
        Reads the device for the current total power consumption across relevant channels.
        Returns:
            float: Instantaneous total power in Watts.
        """
        pass

    def get_voltage(self) -> float:
        """
        Reads the device for the current bus voltage (typically from a specific channel like VIN).
        Returns:
            float: Instantaneous voltage in Volts.
        """
        pass

    def get_current(self) -> float:
        """
        Reads the device for the current total shunt current across relevant channels.
        Returns:
            float: Instantaneous total current in Amperes.
        """
        pass

    def set_sampling_frequency(self, frequency_hz: int) -> None:
        """
        Sets the target frequency for background sampling when monitoring.
        Args:
            frequency_hz (int): Desired samples per second (e.g., 100, 1000).
                                The actual achievable rate may be limited by hardware/system load.
        """
        pass

    def start_sampling(self) -> None:
        """
        Starts a background thread or process to continuously sample power data
        at the configured frequency. Statistics are accumulated internally.
        Requires `stop_sampling()` to be called later.
        """
        pass

    def stop_sampling(self) -> None:
        """
        Stops the background sampling process started by `start_sampling()`.
        """
        pass

    def reset_statistics(self) -> None:
        """
        Clears all internally accumulated statistics (min, max, sum for average, energy, count).
        Call this before `start_sampling()` to measure a specific interval.
        """
        pass

    def get_statistics(self) -> dict:
        """
        Retrieves the power statistics collected since the last reset or initialization.
        Best used after `start_sampling()` and `stop_sampling()`.

        Returns:
            dict: A dictionary containing aggregated ('total') and per-sensor ('sensors')
                  statistics. See the structure documented below. Returns empty or
                  partially filled dict if sampling didn't run or failed.
        """
        pass

Statistics Data Structure (`get_statistics()` return value)

The get_statistics() method returns a dictionary structured as follows:

{
    'total': {  # Statistics aggregated across relevant power-supplying sensors
        'power': {
            'min': float,   # Minimum total power observed during sampling (Watts)
            'max': float,   # Maximum total power observed during sampling (Watts)
            'avg': float,   # Average total power over the sampling period (Watts)
            'total': float, # Total energy consumed during the period (Joules)
                            # Calculated from average power and duration.
            'count': int    # Number of samples contributing to the total statistics.
        }
        # Note: May potentially include 'voltage' and 'current' keys
        # if these are also aggregated and tracked.
    },
    'sensors': [  # A list containing statistics for each individual monitored sensor/channel
        {
            'name': str,    # Name identifying the sensor (e.g., "VDD_GPU_SOC", "VDD_CPU_CV", "VIN_SYS_5V0")
                            # Names depend on Jetson model and INA3221 configuration.
            'power': {
                'min': float,   # Minimum power for this specific sensor (Watts)
                'max': float,   # Maximum power for this specific sensor (Watts)
                'avg': float,   # Average power for this specific sensor (Watts)
                'total': float, # Total energy for this specific sensor (Joules)
                'count': int    # Number of samples collected for this sensor.
            },
            # Note: May potentially include 'voltage' and 'current' keys
            # if these are monitored per sensor.
        },
        # ... potentially more sensor dictionaries
    ]
}

Important Notes:

The exact sensor names available in the 'sensors' list depend on the specific Jetson board model and how the INA3221 channels are configured and named within the library.
The 'total' energy is typically calculated based on the average power (avg) and the duration of the sampling period (derived from count and the sampling frequency).
If start_sampling()/stop_sampling() were not used, or if data collection failed, the returned dictionary might be empty, partially filled, or contain default values like 0 or NaN. Robust code should handle potentially missing keys or non-numeric values (e.g., using .get() with defaults as shown in the monitoring example).

Rust

Rust API Reference

Structs & Enums:

PowerMonitor: The main interface to the library. Manages the C handle and ensures cleanup via the Drop trait.
SensorType: Enum identifying the type of sensor (Unknown, I2C, System).
SensorData: Holds instantaneous data for one sensor.
- name: [u8; 64]: Sensor name (C string, needs conversion).
- type_: SensorType: Type of the sensor.
- voltage: f64, current: f64, power: f64: Measured values.
- online: bool: Whether the sensor is currently readable.
- status: [u8; 32]: Status message (C string, needs conversion).
- warning_threshold: f64, critical_threshold: f64: Thresholds in Watts.
Stats: Holds statistics (min, max, avg, total, count) for a single metric (like power, voltage, or current).
SensorStats: Holds statistics for one sensor, containing Stats for voltage, current, and power.
- name: [u8; 64]: Sensor name (C string, needs conversion).
- voltage: Stats, current: Stats, power: Stats.
PowerData: Holds instantaneous data snapshot.
- total: SensorData: Aggregated data across relevant sensors.
- sensors: *mut SensorData: Raw pointer to an array of SensorData. Requires unsafe to access.
- sensor_count: i32: Number of elements in the sensors array.
PowerStats: Holds accumulated statistics.
- total: SensorStats: Aggregated stats across relevant sensors.
- sensors: *mut SensorStats: Raw pointer to an array of SensorStats. Requires unsafe to access.
- sensor_count: i32: Number of elements in the sensors array.
Error: Enum representing possible error codes from the underlying C library (e.g., InitFailed, NotRunning, NoSensors). Implements From<i32> and Into<i32>.

PowerMonitor Methods:

PowerMonitor::new() -> Result<Self, Error>: Creates and initializes the monitor instance. Connects to hardware.
set_sampling_frequency(&self, frequency_hz: i32) -> Result<(), Error>: Sets the target sampling frequency in Hz for background monitoring.
get_sampling_frequency(&self) -> Result<i32, Error>: Gets the currently configured sampling frequency.
start_sampling(&self) -> Result<(), Error>: Starts background sampling thread. Statistics begin accumulating.
stop_sampling(&self) -> Result<(), Error>: Stops the background sampling thread.
is_sampling(&self) -> Result<bool, Error>: Returns true if background sampling is currently active.
get_latest_data(&self) -> Result<PowerData, Error>: Fetches the most recent instantaneous readings. Return value (PowerData) contains raw pointers requiring unsafe access. See "Data Structures & Safety Notes".
get_statistics(&self) -> Result<PowerStats, Error>: Fetches the statistics accumulated since the last reset_statistics() or initialization. Return value (PowerStats) contains raw pointers requiring unsafe access. See "Data Structures & Safety Notes".
reset_statistics(&self) -> Result<(), Error>: Resets all internal statistics counters (min, max, avg, total, count) to zero.
get_sensor_count(&self) -> Result<i32, Error>: Returns the number of sensors detected by the library.
get_sensor_names(&self) -> Result<Vec<String>, Error>: Returns a Vec<String> containing the names of all detected sensors. Handles C string conversion internally.

Error Handling:

All methods that interact with the C library return Result<T, jetpwmon::Error>.
Use standard Rust error handling (e.g., match, if let Ok/Err, ? operator) to check for and handle potential errors like device access failures, invalid states, etc.

Resource Management:

The PowerMonitor struct implements the Drop trait. When a PowerMonitor instance goes out of scope, its drop method is automatically called, which in turn calls the C library's cleanup function (pm_cleanup). You do not need to call a cleanup function manually.

Data Structures & Safety Notes

Working with Raw Pointers in PowerData and PowerStats

The C library returns arrays of sensor data/statistics via raw pointers (*mut SensorData or *mut SensorStats). The Rust wrapper exposes these directly within the PowerData and PowerStats structs.

Accessing this data requires unsafe blocks in your code. The recommended way is to create a temporary, safe Rust slice from the raw pointer and count:

use std::slice;
use jetpwmon::{PowerStats, SensorStats, Error}; // Assuming these are defined

fn print_sensor_stats(stats: &PowerStats) -> Result<(), Error> {
    // Check if the pointer is valid and count is positive
    if !stats.sensors.is_null() && stats.sensor_count > 0 {
        // SAFETY: This block assumes the C library guarantees that:
        // 1. `stats.sensors` points to valid memory.
        // 2. The memory contains exactly `stats.sensor_count` initialized `SensorStats` elements.
        // 3. This memory remains valid for the lifetime of the `stats` reference.
        // The caller must uphold these invariants.
        let sensor_stats_slice: &[SensorStats] = unsafe {
            slice::from_raw_parts(stats.sensors, stats.sensor_count as usize)
        };

        // Now 'sensor_stats_slice' is a safe slice you can iterate over
        for sensor_stat in sensor_stats_slice {
            // Process each sensor_stat safely here...
            // Remember to handle the C string 'name' field (see below)
             let name = String::from_utf8_lossy(&sensor_stat.name).trim_matches('\0').to_string();
             println!("Sensor: {}, Avg Power: {:.2} W", name, sensor_stat.power.avg);
        }
    } else {
        println!("No per-sensor statistics available.");
    }
    Ok(())
}

The same pattern applies when accessing sensors within a PowerData struct.

Working with C Strings (name and status fields)

Struct fields like name ([u8; 64]) and status ([u8; 32]) are fixed-size byte arrays intended to hold C-style null-terminated strings (or potentially just padded with nulls).

To safely convert them to a Rust String:

Use String::from_utf8_lossy(): This handles potential invalid UTF-8 sequences gracefully by replacing them with the character.
Use .trim_matches('\0'): This removes any leading/trailing null bytes used for padding or termination in the C buffer.

use jetpwmon::SensorData; // Assuming SensorData has a name: [u8; 64]

fn get_name(sensor_data: &SensorData) -> String {
    String::from_utf8_lossy(&sensor_data.name) // Handles invalid UTF-8
        .trim_matches('\0')                  // Removes null padding/terminator
        .to_string()                         // Converts Cow<str> to String
}

Overall Safety

This Rust wrapper aims to be safe where possible (using Result, Drop for cleanup).
However, direct interaction with the C library via FFI inherently involves unsafe operations, especially when dealing with raw pointers returned from C (get_latest_data, get_statistics).
It is the user's responsibility to understand the memory management and lifetime guarantees provided by the underlying C library when working within unsafe blocks. Incorrect assumptions can lead to undefined behavior (crashes, memory corruption). Always consult the C library's documentation if available.

C/C++

API Reference

Handle Type:

pm_handle_t: An opaque pointer (struct pm_handle_s*) representing an initialized instance of the library. Returned by pm_init() and required by most other functions. Must be passed to pm_cleanup() to release resources.

Enums:

pm_error_t: Integer error codes. PM_SUCCESS (0) indicates success. Negative values indicate errors. See pm_error_string() to get descriptions.
- PM_SUCCESS = 0
- PM_ERROR_INIT_FAILED = -1
- PM_ERROR_NOT_INITIALIZED = -2
- PM_ERROR_ALREADY_RUNNING = -3
- PM_ERROR_NOT_RUNNING = -4
- PM_ERROR_INVALID_FREQUENCY = -5
- PM_ERROR_NO_SENSORS = -6
- PM_ERROR_FILE_ACCESS = -7
- PM_ERROR_MEMORY = -8
- PM_ERROR_THREAD = -9
pm_sensor_type_t: Identifies the type of power sensor.
- PM_SENSOR_TYPE_UNKNOWN = 0
- PM_SENSOR_TYPE_I2C = 1 (e.g., INA3221)
- PM_SENSOR_TYPE_SYSTEM = 2 (e.g., sysfs power supply class)

Data Structures:

pm_sensor_data_t: Holds instantaneous data for a single sensor.
- char name[64]: Null-terminated sensor name.
- pm_sensor_type_t type: Sensor type.
- double voltage, current, power: Measured values (V, A, W).
- bool online: Indicates if the sensor is currently readable.
- char status[32]: Null-terminated status string (e.g., "OK").
- double warning_threshold, critical_threshold: Power thresholds (W).
pm_stats_t: Holds basic statistics for a metric.
- double min, max, avg: Min, Max, Average values.
- double total: Sum of values (can be used to calculate energy for power: Energy = Avg Power * Duration).
- uint64_t count: Number of samples collected.
pm_sensor_stats_t: Holds statistics for a single sensor.
- char name[64]: Null-terminated sensor name.
- pm_stats_t voltage, current, power: Statistics for each metric.
pm_power_data_t: Structure filled by pm_get_latest_data.
- pm_sensor_data_t total: Aggregated instantaneous data.
- pm_sensor_data_t* sensors: Pointer to an array of individual sensor data. Memory is managed by the library. The pointer is valid until the next relevant library call or pm_cleanup. Do not free this pointer.
- int sensor_count: Number of valid elements in the sensors array.
pm_power_stats_t: Structure filled by pm_get_statistics.
- pm_sensor_stats_t total: Aggregated statistics.
- pm_sensor_stats_t* sensors: Pointer to an array of individual sensor statistics. Memory is managed by the library. The pointer is valid until the next relevant library call or pm_cleanup. Do not free this pointer.
- int sensor_count: Number of valid elements in the sensors array.

Core Functions:

pm_error_t pm_init(pm_handle_t* handle):
- Initializes the library, discovers sensors, allocates resources.
- Stores the opaque library instance handle at the address provided by handle.
- Must be called first. Returns PM_SUCCESS on success.
pm_error_t pm_cleanup(pm_handle_t handle):
- Stops sampling (if active) and frees all resources associated with the handle.
- Must be called when finished with the library to prevent resource leaks.
const char* pm_error_string(pm_error_t error):
- Returns a constant, human-readable string describing the given error code. Do not modify or free the returned string.

Sampling Control & Status:

pm_error_t pm_set_sampling_frequency(pm_handle_t handle, int frequency_hz):
- Sets the target sampling frequency (in Hz) for the background monitoring thread. Must be > 0.
pm_error_t pm_get_sampling_frequency(pm_handle_t handle, int* frequency_hz):
- Retrieves the currently configured sampling frequency, storing it at the address frequency_hz.
pm_error_t pm_start_sampling(pm_handle_t handle):
- Starts the background sampling thread. Statistics begin accumulating. Returns PM_ERROR_ALREADY_RUNNING if already started.
pm_error_t pm_stop_sampling(pm_handle_t handle):
- Stops the background sampling thread. Returns PM_ERROR_NOT_RUNNING if not running.
pm_error_t pm_is_sampling(pm_handle_t handle, bool* is_sampling):
- Checks if the background sampling thread is active, storing the result (true or false) at the address is_sampling.

Data & Statistics Retrieval:

pm_error_t pm_get_latest_data(pm_handle_t handle, pm_power_data_t* data):
- Fills the user-provided data structure with the most recent instantaneous sensor readings.
- The data->sensors pointer will point to an internal library buffer.
pm_error_t pm_get_statistics(pm_handle_t handle, pm_power_stats_t* stats):
- Fills the user-provided stats structure with statistics accumulated since the last reset.
- The stats->sensors pointer will point to an internal library buffer.
pm_error_t pm_reset_statistics(pm_handle_t handle):
- Resets all accumulated statistics (min, max, avg, total, count) to zero.

Sensor Information:

pm_error_t pm_get_sensor_count(pm_handle_t handle, int* count):
- Gets the total number of sensors detected by the library.
pm_error_t pm_get_sensor_names(pm_handle_t handle, char** names, int* count):
- Fills a caller-allocated array of C strings (char* names[]) with the names of detected sensors.
- names: Pointer to an array of char*. The caller must allocate this array. Each char* in the array must also point to a caller-allocated buffer (e.g., char name_buffer[64]) large enough to hold a sensor name.
- count: [inout] parameter. On input, points to the allocated size of the names array. On output, points to the actual number of names written.
- Note: This function requires careful memory management by the caller. Accessing names via pm_get_latest_data or pm_get_statistics (using the sensors[i].name field) is often simpler as the library manages those strings.

C++ Bindings

API Reference (C++ Wrapper)

Namespace: jetpwmon

Main Class: PowerMonitor

Description: An RAII wrapper class for managing the jetpwmon C library. It handles initialization (pm_init) in its constructor and cleanup (pm_cleanup) in its destructor automatically. It converts C API error codes into std::runtime_error exceptions.
Resource Management: Non-copyable, but movable. Uses std::unique_ptr with a custom deleter for the C handle (pm_handle_t).
Constructor: PowerMonitor()
- Initializes the library connection.
- Throws: std::runtime_error if pm_init fails. The exception's what() message contains the error description from pm_error_string.
Destructor: ~PowerMonitor()
- Automatically calls pm_cleanup on the managed C handle.
Methods:
- void setSamplingFrequency(int frequency_hz)
  - Sets the background sampling frequency (Hz).
  - Throws: std::runtime_error on C API failure.
- int getSamplingFrequency() const
  - Gets the current sampling frequency (Hz).
  - Throws: std::runtime_error on C API failure.
- void startSampling()
  - Starts the background sampling thread.
  - Throws: std::runtime_error on C API failure (e.g., already running).
- void stopSampling()
  - Stops the background sampling thread.
  - Throws: std::runtime_error on C API failure (e.g., not running).
- bool isSampling() const
  - Checks if sampling is currently active.
  - Throws: std::runtime_error on C API failure.
- PowerData getLatestData() const
  - Gets the most recent instantaneous sensor readings.
  - Returns: A PowerData object containing the snapshot.
  - Throws: std::runtime_error on C API failure.
  - Note: See PowerData description and Safety Notes regarding pointer validity.
- PowerStats getStatistics() const
  - Gets the statistics accumulated since the last reset.
  - Returns: A PowerStats object containing the statistics.
  - Throws: std::runtime_error on C API failure.
  - Note: See PowerStats description and Safety Notes regarding pointer validity.
- void resetStatistics()
  - Resets all internal accumulated statistics.
  - Throws: std::runtime_error on C API failure.
- int getSensorCount() const
  - Gets the number of detected sensors.
  - Throws: std::runtime_error on C API failure.
- std::vector<std::string> getSensorNames() const
  - Gets the names of all detected sensors. Handles C memory management and string conversion.
  - Returns: A std::vector<std::string> containing the sensor names.
  - Throws: std::runtime_error on C API failure.

Data Wrapper Classes:

PowerData / PowerStats
- Description: Thin wrappers around the C structs pm_power_data_t and pm_power_stats_t, primarily returned by getLatestData and getStatistics. They are non-copyable but movable.
- Memory: They hold a copy of the total C struct member and the raw C pointer (sensors) along with the sensor_count. They do NOT manage the memory pointed to by the sensors pointer. That memory is owned by the underlying C library.
- Getters:
  - const pm_sensor_data_t& getTotal() const (for PowerData)
  - const pm_sensor_stats_t& getTotal() const (for PowerStats)
    - Returns a const reference to the copied total data/statistics struct.
  - const pm_sensor_data_t* getSensors() const (for PowerData)
  - const pm_sensor_stats_t* getSensors() const (for PowerStats)
    - Returns the raw C pointer to the array of per-sensor data/statistics. See Safety Notes.
  - int getSensorCount() const: Returns the number of elements pointed to by getSensors().

Underlying C Structs:

The C++ wrapper provides access to data via the C structs (pm_sensor_data_t, pm_stats_t, pm_sensor_stats_t). Refer to the C API documentation for detailed field descriptions within these structs.

C++ Wrapper Safety & Pointer Notes

RAII & Exceptions: The PowerMonitor class significantly improves safety by automating resource cleanup (pm_cleanup) through its destructor (RAII) and by converting C error codes into C++ exceptions (std::runtime_error). Always use try...catch blocks when interacting with PowerMonitor methods.
Pointer Validity (getSensors()): The PowerData and PowerStats objects returned by getLatestData() and getStatistics() contain raw C pointers to arrays (sensors). Crucially, the C++ wrapper classes (PowerData, PowerStats) do NOT manage the lifetime of the memory these pointers point to. This memory is managed by the C library.
- Assumption: The memory pointed to by getSensors() is typically valid only temporarily, likely until the next non-const call to the PowerMonitor object or until the PowerMonitor object is destroyed.
- Guideline: Access the data through the pointer returned by getSensors() immediately after the call to getLatestData() or getStatistics(), within the same scope. Do not store this raw pointer for later use, as it may become invalid (dangling pointer).
C String Handling: Data structures contain C-style fixed-size character arrays (e.g., name[64], status[32]). Use safe methods (like the c_char_to_string helper in the examples using strnlen) to convert these to std::string to avoid buffer over-reads, especially if null termination is not guaranteed within the fixed size.

Building from Source

Prerequisites

CMake 3.10 or higher
C++ compiler with C++17 support
Python 3.8 or higher (for Python bindings)
Rust toolchain (for Rust bindings)

Build Steps

C Library and C++ Bindings

git clone https://github.com/nerdneilsfield/jetson-power-monitor.git
cd jetson-power-monitor
mkdir build && cd build
cmake ..
make
sudo make install

Python Bindings

python3 -m pip install setuptools pybind11
python3 -m pip install -e .

# or you need to build wheel
python3 -m pip install build
python3 -m build --wheel
# the result will be in dist/

Rust Bindings

# copy c headers and sources to rust vendor directory
make copy-rust

# build rust crate
cd bindings/rust
cargo build

Contributing

We welcome contributions! Please see our CONTRIBUTING.md for detailed information about:

Project architecture and implementation details
Development setup and guidelines
Code style and testing requirements
Pull request process
Common development tasks
Release process

License

This project is licensed under the BSD 3-Clause License License - see the LICENSE file for details.

Acknowledgments

NVIDIA Jetson team for their excellent hardware
All contributors who have helped with this project
jetson_stats

Name		Name	Last commit message	Last commit date
Latest commit History 81 Commits
.github		.github
.vscode		.vscode
3rdparty/pybind11		3rdparty/pybind11
bindings		bindings
cmake		cmake
example		example
include/jetpwmon		include/jetpwmon
python/jetpwmon		python/jetpwmon
src		src
tests		tests
.gitignore		.gitignore
CMakeLists.txt		CMakeLists.txt
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
MANIFEST.in		MANIFEST.in
Makefile		Makefile
README.md		README.md
README_CN.md		README_CN.md
pyproject.toml		pyproject.toml
setup.py		setup.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Repository files navigation

jetson-power-monitor

Features

Installation

Python

Rust

C/C++

Usage

Python

Rust

C/C++

C++

API Documentation

Python

Rust

C/C++

C++ Bindings

Building from Source

Prerequisites

Build Steps

C Library and C++ Bindings

Python Bindings

Rust Bindings

Contributing

License

Acknowledgments

Star History

About

Uh oh!

Releases 3

Packages

Uh oh!

Languages

License

nerdneilsfield/jetson-power-monitor

Folders and files

Latest commit

History

Repository files navigation

jetson-power-monitor

Features

Installation

Python

Rust

C/C++

Usage

Python

Rust

C/C++

C++

API Documentation

Python

Rust

C/C++

C++ Bindings

Building from Source

Prerequisites

Build Steps

C Library and C++ Bindings

Python Bindings

Rust Bindings

Contributing

License

Acknowledgments

Star History

About

Topics

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases 3

Packages 0

Uh oh!

Languages

Packages