ARM fixed point vs. floating point

Hi,
I'm creating application on ARM STM32F103VE. I have to multiply several
matrix in very short time. I'm thinking about using Fixed Point, to
speed up my calculations. So that I wrote union:

typedef union fxpntu{
  int full;
  struct fxpnts{
    int_fast32_t fraction : 31;
    int_fast32_t integer : 1;
  } part;
} fxpnt_t;

My Fixed Point representation is Q1.31.

Than, I created multiplying function:

/* Return: c = a * b */
fxpnt_t fxmul(fxpnt_t* a, fxpnt_t* b) {
  fxpnt_t c;

  _fxpnt_accu64.full = a->full*b->full;

  fxreg |= (_fxpnt_accu64.part.H == 0 && (a->full != 0 || b->full != 
0))<<0x05;
  fxreg |= FX_UFLOW >> 3;

  c.full = _fxpnt_accu64.part.H;

  c.full = FX_UFLOW?0x01:_fxpnt_accu64.part.H;
  FX_UFLOW_CLEAR;

  return c;
}

where _fxpnt_accu64 is global object, to hold result.
I made few tests in CrossWorks and results are weird:

float a = -150.534433;
float b = 0.05323;
c = a*b;            // 47 Instructions
c = a+b;            // 55 Instructions
c = a-b;            // 49 Instructions
c = a/b;            // 139 Instructions
ans = fxmul(&fx2, &fx1);   // 165 Instructions

I checked my own multiplying function, and
fxreg |= (_fxpnt_accu64.part.H == 0 && (a->full != 0 || b->full != 
0))<<0x05;  <-- this line takes 43 Instructions.

Can someone explain, or send any link about creating fixed point
on ARM Cortex M-3, or example code? Maybe I'm doing something wrong,
maybe I don't get point of fixed multiplying..

Thanks for Your feedback.