static inline functions in c

Question

I read this other SO question and answer and it seems to make sense to me but I had one additional question to add it it.

The most up voted answer says

For small functions that are called frequently that can make a big performance difference.

okay, so what would be considered a small function?

The reason that I am asking is that I am looking at using a math library, vectormath from the bullet physics framework. All their math functions are static inline but while some are fairly short some are pretty long.

Here's what I consider short:

static inline void vmathM3Copy( VmathMatrix3 *result, const VmathMatrix3 *mat )
{
    vmathV3Copy( &result->col0, &mat->col0 );
    vmathV3Copy( &result->col1, &mat->col1 );
    vmathV3Copy( &result->col2, &mat->col2 );
}

but even that would embed this function 3 time:

static inline void vmathV3Copy( VmathVector3 *result, const VmathVector3 *vec )
{
    result->x = vec->x;
    result->y = vec->y;
    result->z = vec->z;
}

Here's what seems to be long to me:

static inline float vmathM4Determinant( const VmathMatrix4 *mat )
{
    float dx, dy, dz, dw, mA, mB, mC, mD, mE, mF, mG, mH, mI, mJ, mK, mL, mM, mN, mO, mP, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
    mA = mat->col0.x;
    mB = mat->col0.y;
    mC = mat->col0.z;
    mD = mat->col0.w;
    mE = mat->col1.x;
    mF = mat->col1.y;
    mG = mat->col1.z;
    mH = mat->col1.w;
    mI = mat->col2.x;
    mJ = mat->col2.y;
    mK = mat->col2.z;
    mL = mat->col2.w;
    mM = mat->col3.x;
    mN = mat->col3.y;
    mO = mat->col3.z;
    mP = mat->col3.w;
    tmp0 = ( ( mK * mD ) - ( mC * mL ) );
    tmp1 = ( ( mO * mH ) - ( mG * mP ) );
    tmp2 = ( ( mB * mK ) - ( mJ * mC ) );
    tmp3 = ( ( mF * mO ) - ( mN * mG ) );
    tmp4 = ( ( mJ * mD ) - ( mB * mL ) );
    tmp5 = ( ( mN * mH ) - ( mF * mP ) );
    dx = ( ( ( mJ * tmp1 ) - ( mL * tmp3 ) ) - ( mK * tmp5 ) );
    dy = ( ( ( mN * tmp0 ) - ( mP * tmp2 ) ) - ( mO * tmp4 ) );
    dz = ( ( ( mD * tmp3 ) + ( mC * tmp5 ) ) - ( mB * tmp1 ) );
    dw = ( ( ( mH * tmp2 ) + ( mG * tmp4 ) ) - ( mF * tmp0 ) );
    return ( ( ( ( mA * dx ) + ( mE * dy ) ) + ( mI * dz ) ) + ( mM * dw ) );
}

or even this one

static inline void vmathM4Inverse( VmathMatrix4 *result, const VmathMatrix4 *mat )
{
    VmathVector4 res0, res1, res2, res3;
    float mA, mB, mC, mD, mE, mF, mG, mH, mI, mJ, mK, mL, mM, mN, mO, mP, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, detInv;
    mA = mat->col0.x;
    mB = mat->col0.y;
    mC = mat->col0.z;
    mD = mat->col0.w;
    mE = mat->col1.x;
    mF = mat->col1.y;
    mG = mat->col1.z;
    mH = mat->col1.w;
    mI = mat->col2.x;
    mJ = mat->col2.y;
    mK = mat->col2.z;
    mL = mat->col2.w;
    mM = mat->col3.x;
    mN = mat->col3.y;
    mO = mat->col3.z;
    mP = mat->col3.w;
    tmp0 = ( ( mK * mD ) - ( mC * mL ) );
    tmp1 = ( ( mO * mH ) - ( mG * mP ) );
    tmp2 = ( ( mB * mK ) - ( mJ * mC ) );
    tmp3 = ( ( mF * mO ) - ( mN * mG ) );
    tmp4 = ( ( mJ * mD ) - ( mB * mL ) );
    tmp5 = ( ( mN * mH ) - ( mF * mP ) );
    vmathV4SetX( &res0, ( ( ( mJ * tmp1 ) - ( mL * tmp3 ) ) - ( mK * tmp5 ) ) );
    vmathV4SetY( &res0, ( ( ( mN * tmp0 ) - ( mP * tmp2 ) ) - ( mO * tmp4 ) ) );
    vmathV4SetZ( &res0, ( ( ( mD * tmp3 ) + ( mC * tmp5 ) ) - ( mB * tmp1 ) ) );
    vmathV4SetW( &res0, ( ( ( mH * tmp2 ) + ( mG * tmp4 ) ) - ( mF * tmp0 ) ) );
    detInv = ( 1.0f / ( ( ( ( mA * res0.x ) + ( mE * res0.y ) ) + ( mI * res0.z ) ) + ( mM * res0.w ) ) );
    vmathV4SetX( &res1, ( mI * tmp1 ) );
    vmathV4SetY( &res1, ( mM * tmp0 ) );
    vmathV4SetZ( &res1, ( mA * tmp1 ) );
    vmathV4SetW( &res1, ( mE * tmp0 ) );
    vmathV4SetX( &res3, ( mI * tmp3 ) );
    vmathV4SetY( &res3, ( mM * tmp2 ) );
    vmathV4SetZ( &res3, ( mA * tmp3 ) );
    vmathV4SetW( &res3, ( mE * tmp2 ) );
    vmathV4SetX( &res2, ( mI * tmp5 ) );
    vmathV4SetY( &res2, ( mM * tmp4 ) );
    vmathV4SetZ( &res2, ( mA * tmp5 ) );
    vmathV4SetW( &res2, ( mE * tmp4 ) );
    tmp0 = ( ( mI * mB ) - ( mA * mJ ) );
    tmp1 = ( ( mM * mF ) - ( mE * mN ) );
    tmp2 = ( ( mI * mD ) - ( mA * mL ) );
    tmp3 = ( ( mM * mH ) - ( mE * mP ) );
    tmp4 = ( ( mI * mC ) - ( mA * mK ) );
    tmp5 = ( ( mM * mG ) - ( mE * mO ) );
    vmathV4SetX( &res2, ( ( ( mL * tmp1 ) - ( mJ * tmp3 ) ) + res2.x ) );
    vmathV4SetY( &res2, ( ( ( mP * tmp0 ) - ( mN * tmp2 ) ) + res2.y ) );
    vmathV4SetZ( &res2, ( ( ( mB * tmp3 ) - ( mD * tmp1 ) ) - res2.z ) );
    vmathV4SetW( &res2, ( ( ( mF * tmp2 ) - ( mH * tmp0 ) ) - res2.w ) );
    vmathV4SetX( &res3, ( ( ( mJ * tmp5 ) - ( mK * tmp1 ) ) + res3.x ) );
    vmathV4SetY( &res3, ( ( ( mN * tmp4 ) - ( mO * tmp0 ) ) + res3.y ) );
    vmathV4SetZ( &res3, ( ( ( mC * tmp1 ) - ( mB * tmp5 ) ) - res3.z ) );
    vmathV4SetW( &res3, ( ( ( mG * tmp0 ) - ( mF * tmp4 ) ) - res3.w ) );
    vmathV4SetX( &res1, ( ( ( mK * tmp3 ) - ( mL * tmp5 ) ) - res1.x ) );
    vmathV4SetY( &res1, ( ( ( mO * tmp2 ) - ( mP * tmp4 ) ) - res1.y ) );
    vmathV4SetZ( &res1, ( ( ( mD * tmp5 ) - ( mC * tmp3 ) ) + res1.z ) );
    vmathV4SetW( &res1, ( ( ( mH * tmp4 ) - ( mG * tmp2 ) ) + res1.w ) );
    vmathV4ScalarMul( &result->col0, &res0, detInv );
    vmathV4ScalarMul( &result->col1, &res1, detInv );
    vmathV4ScalarMul( &result->col2, &res2, detInv );
    vmathV4ScalarMul( &result->col3, &res3, detInv );
}

The guys who wrote the library obviously understand the math very well but if your doing a lot of math operations and the compiler probably inlining all these functions wouldn't you get a bigger file?

Answer 1

Most likely you will get a bigger file, since the code to be inlined will appear multiple times throughout the program instead of just once.

A bigger file doesn't really mean too much, particularly in the age of terabyte disks, if you're trying to improve performance. Better to have a larger file than to incur the overhead of multiple function calls that aren't necessarily needed.

Answer 2

First of all compilers will not inline every function marked with static. This is not what static keyword is intended for. There’s been the inline keyword for that purpose, however many compiler ignore it nowadays.

A compiler will carefully decide whether it’s better to inline or not to inline a function. But basically your observation is true: programs optimized for maximum speed tend to be larger. You can see this if you, for instance, look at GCC Optimization Levels.

For small functions that are called frequently that can make a big performance difference.

If the function is that small that the push of the function onto the stack would take longer than actually executing the body the function could be a performance problem. In such a case a good compiler will inline the function. If the stack push however would be the cheapest part of executing the function, inlining is unlikely to happen.

Answer 3

Inline Disadvantages

The primary disadvantage to inline substitution is that it usually makes the program code bigger. In extreme cases, this can degrade program performance by increasing page faults and cache misses. Reading a page from the disk may take as long as executing hundreds of thousands of instructions. Poor cache performance may slow down a program by a factor of two. Reasonable care must be taken when inlining not to make the program so big that either paging or caching problems dominate the execution time.