Random FPU Exceptions when disabling app scheduler

Question

Dear Nordic, 
 I am running into an interesting side effect when I disable the app scheduler library from my program. I have been previously using the app scheduler to service all of my app timers, but I am now trying to reduce my dependency on this library in favor of traditional interrupts. I have an app_timer that samples an accelerometer at 500Hz and it works great as it is currently (with the scheduler) but when I disable the app scheduler I get a flood of FPU exceptions. There is a floating point operation that occurs in the interrupt but it is a single float multiplication. Below are the relevant functions and data structures involved in the interrupt. I have implemented a FPU exception handler interrupt according to the SDK example and log each error case. 
 I have done a couple of experiments that unfortunately did not help me understand the issue any better: 
 
 wrapping the interrupt in a critical section did not help 
 adding a flag to check if the interrupt was somehow interrupting itself showed that this is not the case. Similarly reducing speed of the interrupt did not solve the problem. 
 simply removing the multiplication caused this issue to go away so I am pretty sure this interrupt is the problem

typedef union
{
 int16_t i16bit[3];
 uint8_t u8bit[6];
} axis3bit16_t;


static axis3bit16_t m_raw_acceleration;
static float acceleration_mg[3];

//NOTE: this is the interrupt called from an app_timer
void get_accel()
{
 uint8_t reg;

 /* Read output only if new xl value is available */
 xl_flag_data_ready_get(&m_dev_ctx, &reg);
 if (reg) {

 /* Read acceleration field data */
 acceleration_raw_get(&m_dev_ctx, m_raw_acceleration.u8bit);
 acceleration_mg[0] = from_fs16_to_mg(m_raw_acceleration.i16bit[0]);
 acceleration_mg[1] = from_fs16_to_mg(m_raw_acceleration.i16bit[1]);
 acceleration_mg[2] = from_fs16_to_mg(m_raw_acceleration.i16bit[2]);

 }
}

//Conversion function from another file:
float_t from_fs16_to_mg(int16_t lsb)
{
 return ((float_t)lsb) * 0.488f;
}