Why didn't -O3 GCC Optimization inline this function?

In GCC compiler, whenever -O3 is flagg开发者_如何学Goed, the compiler optimizes mainly by Loop Unrolling and Function Inlining.

However, when I compiled an application that has the following function, it didn't do Function Inlining. From gprofing and gconving I could notice that this function (comp_t_delay) was not called from the main function unit of this application. However, it was called from a function, called by a function, called by the main function.

Here is the code of comp_t_delay(int in,int ip) that I want to know why the GCC didn't optimize it using -O3 (any help is appreciated!):

static float
comp_t_delay(int in,int ip)
{

    int sb, sib,dx, dy;
    t_type_ptr st, sit;
    float d_ssi;

    d_ssi = 0.;

    sb = net[in].node_block[0];
    st = block[sb].type;

    sib = net[in].node_block[ip];
    sit = block[sib].type;

    assert(st != NULL);
    assert(sit != NULL);

    dx = abs(block[sib].x - block[sb].x);
    dy = abs(block[sib].y - block[sb].y);

    if(st == T_IO)
    {
        if(sit == T_IO)
        d_ssi = de_io[dx][dy];
        else
        d_ssi = de_iof[dx][dy];
    }
    else
    {
        if(sit == T_IO)
        d_ssi = de_fio[dx][dy];
        else
        d_ssi = de_fb[dx][dy];
    }
    if(d_ssi < 0)
    {
        printf
        ("Error1\n");
        exit(1);
    }

    if(d_ssi < 0.)
    {
        printf
        ("Error2\n");
        exit(1);
    }

    return (d_ssi);
}

It most probably didn't inline it because it is too long. Long functions, when inlined, can actually cause the code to run slower - for example you get the CPU registers bloated by more variables. In this particular case gcc decided it will be faster not to inline the function.