ARC Retain/Release overhead in tight loops

gistfile1.m

/**
 An inline (read: no overhead) function that will perform the physics increment for a given particle
 
 @return A boolean for whether the particle has expired and needs to be removed
 */
extern inline BOOL updateParticle(BILParticleAtomConventional * atom, BILParticleModel * model, NSTimeInterval deltaTime)
{
    // Time to Live
    atom->atom.timeToLive -= deltaTime;
    
    // Return early if expired
    if(atom->atom.timeToLive <= 0)
    {
        atom->atom.timeToLive = -1;
        return YES;
    }
    
    // Mode A: gravity, direction, tangential accel & radial accel
    if(model.emitterMode == BILParticleModelEmitterModeGravity)
    {
        __unsafe_unretained BILParticleModelGravity * gravityModel = (BILParticleModelGravity *)model;
        
        // Vector from the start position of this particle
        GLKVector2 displacementFromStart = GLKVector2Subtract(atom->atom.position, atom->atom.startPosition);
        // Dont divide by zero
        if(displacementFromStart.x != 0 || displacementFromStart.y != 0)
            displacementFromStart = GLKVector2Normalize(displacementFromStart);
        
        // Radial & Tangential Acceleration
        GLKVector2 radialAcceleration = GLKVector2MultiplyScalar(displacementFromStart, atom->mode.gravity.radialAcceleration);
        
        // Tangential Acceleration
        GLKVector2 tangentialAcceleration = displacementFromStart;
        float newY = tangentialAcceleration.x;
        tangentialAcceleration.x = -tangentialAcceleration.y;
        tangentialAcceleration.y = newY;
        tangentialAcceleration = GLKVector2MultiplyScalar(tangentialAcceleration, atom->mode.gravity.tangentialAcceleration);
        
        // Component of all accelerations
        GLKVector2 accelerationVector = GLKVector2Add(gravityModel.gravity, GLKVector2Add(tangentialAcceleration, radialAcceleration));
        //GLKVector2 accelerationVector = model.gravity;
        // Velocity Delta (accelleration * time)
        GLKVector2 velocity = GLKVector2MultiplyScalar(accelerationVector, deltaTime);
        // Velocity (velocity + deltaVelocity)
        velocity = GLKVector2Add(atom->mode.gravity.velocity, velocity);
        
        // Update Particle Velocity
        atom->mode.gravity.velocity = velocity;
        // Update Particle Position (position + (velocity * time))
        atom->atom.position = GLKVector2Add(atom->atom.position, GLKVector2MultiplyScalar(velocity, deltaTime));
        
    }
    // Mode B: radius movement
    else
    {
        // Update the angle and radius of the particle.
        atom->mode.radial.angle += atom->mode.radial.anglePerSecond * deltaTime;
        atom->mode.radial.radius += atom->mode.radial.deltaRadius * deltaTime;
        
        GLKVector2 position = atom->atom.startPosition;
        position.x += (- cosf(atom->mode.radial.angle) * atom->mode.radial.radius);
        position.y += (- sinf(atom->mode.radial.angle) * atom->mode.radial.radius);
        
        atom->atom.position = position;
    }
    
    // Interpolate Color
    atom->atom.color.r += (atom->delta.deltaColor.r * deltaTime);
    atom->atom.color.g += (atom->delta.deltaColor.g * deltaTime);
    atom->atom.color.b += (atom->delta.deltaColor.b * deltaTime);
    atom->atom.color.a += (atom->delta.deltaColor.a * deltaTime);
    
    // Modify the Particle size
    atom->atom.size += (atom->delta.sizeDelta * deltaTime);
    atom->atom.size = MAX(0, atom->atom.size);
    
    // Update the Angle
    atom->atom.rotation += (atom->delta.rotationDelta * deltaTime);
    
    return NO;
}

The retain and release that ARC puts in on line 21 accounts for 90% of the cycles for this loop. Placing the __unsafe_unretained here removes ARCs influence entirely. All this is, is a cast of a parameter!

Trace Before

Trace After

CPU Usage Before

CPU Usage After