Implicit conversion of int16_t data to float

[entrop1a]

In bmi2_defs.h, there is:

struct bmi2_sens_axes_data
{
    /*! Data in x-axis */
    int16_t x;

    /*! Data in y-axis */
    int16_t y;

    /*! Data in z-axis */
    int16_t z;

    /*! Sensor time for virtual frames */
    uint32_t virt_sens_time;
};

In which axes are of the type int16_t.

However, in SparkFun_BMI270_Arduino_Library.h, there is:

struct BMI270_SensorData
{
    // Acceleration in g's
    float accelX;
    float accelY;
    float accelZ;

    // Rotation in deg/sec
    float gyroX;
    float gyroY;
    float gyroZ;

    // Auxiliary sensor data, raw bytes
    uint8_t auxData[BMI2_AUX_NUM_BYTES];

    // Time of this data in milliseconds, measured by sensor
    uint32_t sensorTimeMillis;
};

In which axes are of the type float. This conversion, in architecture above 16-bit, can cause memory bloat.

I'd like to know if there's any necessary reason for this conversion or if it was simply a mistake.

[sfe-SparkFro]

Hi there,

The use of floats in this library is intentional. The values in the bmi2_sens_axes_data struct are raw sensor data without "real" units (g's, deg/s), whereas the values in the BMI270_SensorData struct are converted to "real" units. Yes, it does increase memory requirements and computation overhead, but the goal with our libraries is generally to be convenient and easy to use rather than highly optimized. So we're not going to change the implementation to avoid floats. If the extra bloat is problematic for your project, this library is open-source and you're welcome to modify it for your needs, or even just use the underlying BMI270 API from Bosch that this library wraps.

Hope this helps!

[entrop1a]

Thanks for getting back to me.