private static A reduce(A x, int axis, int id, ByteByteByteFunction f) {
        if (x.s.length == 0) return x;
        if (axis < 0) axis += x.s.length;
        int n0 = intProduct(x.s, 0, axis), n1 = x.s[axis], n2 = intProduct(x.s, axis + 1, x.s.length);
        int[] s = intDelAt(x.s, axis);
        byte[] r = new byte[n0 * n2];
        if (id != 0) fill(r, (byte) id);
        for (int i = 0; i < n0; i++)
            for (int j = 0; j < n1; j++)
                for (int k = 0; k < n2; k++)
                    r[i * n2 + k] = f.apply(r[i * n2 + k], x.a[i * n1 * n2 + j * n2 + k]);
        return new A(r, s);
    }