Return a 2D primitive array from C to Java from JNI/NDK

I have found large amounts of documentation on how to generate a 2D primitive array in JNI and returning it to Java. But these pieces of information fail to describe how to pass an already existing 2D float array (float**) given a context in C.

To describe my issue explicitly, I'll add some C pseudo code of what I would like to implement:

// Returns a 2D float array from C to Java
jfloatArray ndk_test_getMy2DArray(JNIEnv* env, jobject thiz, jlong context)
{
    // Cast my context reference
    MyContextRef contextRef = (MyContextRef) context;

    // In case we need it below
    un开发者_如何学运维signed int length = MyContextGet1DLength(contextRef);

    // Get the 2D Array we want to "Cast"
    float** primitive2DArray = MyContextGet2DArray(contextRef);

    // Hokus pokus...
    // We do something to create the returnable data to Java
    //
    // Below is the missing piece that would convert the primitive
    // 2D array into something that can be returned consumed and consumed
    // by Java

    jfloatArray myReturnable2DArray

    return myReturnable2DArray;
}

I'm assuming this is not straight forward, given I haven't been able to find anything describing this scenario.

Thanks for any helpful information.

---

Thanks Timo for your help and link. For posterity, I'm adding a complete code set that would go through the process of generating a 2D primitive array consumable by Java, from an existing C 2D primitive array.

// Returns a 2D float array from C to Java
jobjectArray ndk_test_getMy2DArray(JNIEnv* env, jobject thiz, jlong context)
{
    // Cast my context reference
    MyContextRef contextRef = (MyContextRef) context;

    // Get the length for the first and second dimensions
    unsigned int length1D = MyContextGet1DLength(contextRef);
    unsigned int length2D = MyContextGet2DLength(contextRef);

    // Get the 2D float array we want to "Cast"
    float** primitive2DArray = MyContextGet2DArray(contextRef);

    // Get the float array class
    jclass floatArrayClass = (*env)->FindClass(env, "[F");

    // Check if we properly got the float array class
    if (floatArrayClass == NULL)
    {
        // Ooops
        return NULL;
    }

    // Create the returnable 2D array
    jobjectArray myReturnable2DArray = (*env)->NewObjectArray(env, (jsize) length1D, floatArrayClass, NULL);

    // Go through the firs dimension and add the second dimension arrays
    for (unsigned int i = 0; i < length1D; i++)
    {
        jfloatArray floatArray = (*env)->NewFloatArray(env, length2D);
        (*env)->SetFloatArrayRegion(env, floatArray, (jsize) 0, (jsize) length2D, (jfloat*) primitive2DArray[i]);
        (*env)->SetObjectArrayElement(env, myReturnable2DArray, (jsize) i, floatArray);
        (*env)->DeleteLocalRef(env, floatArray);
    }

    // Return a Java consumable 2D float array
    return myReturnable2DArray;
}

---

I've made a simple function for conversion:

jobjectArray getJNIArray(JNIEnv *env, jobject obj, const vector<vector<float> >& arr) {
    jclass floatClass = env->FindClass("[F"); //
    jsize height = arr.size();

    // Create the returnable 2D array
    jobjectArray jObjarray = env->NewObjectArray(height, floatClass, NULL);

    // Go through the first dimension and add the second dimension arrays
    for (unsigned int i = 0; i < height; i++) {
        jfloatArray floatArray = env->NewFloatArray(arr[i].size());
        env->SetFloatArrayRegion(floatArray, (jsize) 0, (jsize) arr[i].size(), (jfloat*) arr[i].data());
        env->SetObjectArrayElement(jObjarray, (jsize) i, floatArray);
        env->DeleteLocalRef(floatArray);
    }

    return jObjarray;
}

---

Unfortunately I don't think you can pass C floats up to Java, you'll have to turn the array into a 2D array of jfloats by converting every member into a jFloat.

Essentially, you'll have to create the multidimensional jFloatArray, then iterate through the native C array, convert every element into its jFloat representation and store it into the same position in the jFloatArray you just created.

This part of the documentation should explain it in a bit more depth.