14 bit 2s complement

CannibalRobotics

On the Freescale Xtrinsic FXOS8700CQ the accelerometer data is presented in the form of a 14 bit 2s complement. Following the "invert and + 1" technique they recommend I can not get SPIN to clear off the bits above the 13th on a negitive number. My feeling is that it's something obvious Im just not picking up.
Some fresh eyes on this would be great.

Pri FixA(d)| bits               '' Returns Signed Long of G force on axis for data d
  
  bits := d & $0000_3FFF                                ' Reduce to 14 bits to work with

  if bits > $0000_1FFF                                  ' Is sign bit (14th) set ?
    bits &= $0000_0000_1FFF                             ' Clear sign off base number
    bits :=(! bits) + 1                                 ' Invert the remainder and add 1
    bits := bits & $0000_0000_0000_1FFF                 ' Clear off upper ones from ! operation
    bits := -1 * bits                                   ' Make it negitive (set sign bit)

  Return bits

For Negitive numbers I get:

 AxC Raw: 00000000000000000011101110111111
 14 bits: 00000000000000000011111111111111
FixA out: 11111111111111111111101110111111 = -1089

For Positive I get:

 AxC Raw: 00000000000000000000010001110100
 14 bits: 00000000000000000011111111111111
FixA out: 00000000000000000000010001110100 = 1140

Why is the routine returning all 1's above the 13th bit after I've ANDed them all off with

bits := bits & $0000_0000_0000_1FFF                 ?

Thanks in advance.

Phil Pilgrim (PhiPi)

Do this:

value := value << 18 ~> 18

The << gets the sign bit into bit 31, and the ~> does a signed right shift back, preserving the sign bit.

-Phil

CannibalRobotics

Thanks for reply, where?

Phil Pilgrim (PhiPi)

pri fixA(d)
  return d << 18 ~> 18

CannibalRobotics

New code:

Pri FixA(d)| bits               '' Returns Signed Long of G force on axis for data d
  return d << 18 ~> 18

Result for negitive

 AxC Raw: 00000000000000000011110010010000
 14 bits: 00000000000000000011111111111111
FixA out: 11111111111111111111110010010000 = -880

abecedarian

Shouldn't you be able to add 16384 to the value and be done with it?
Then you should end up with something ranging from 0 to 37267.

Phil Pilgrim (PhiPi)

abecedarian wrote:

Shouldn't you be able to add 16384 to the value and be done with it?

No, that doesn't work. Remember, it's still 32-bit arithmetic: 16384 (i.e. -16384) + 16384 == 32768, not 0. Besides, he wants the signed value for +/- G forces.

-Phil

abecedarian

Nevermind, just realized you're taking a 14 bit signed value into a 32 bit variable.

So shift left 18 bits, then right 18 bits, to get rid of 'stray' bits on the end, then add 16384 to get rid of the sign.

- edit: Just noticed Phil's response above. I'll shut up.

CannibalRobotics

Thanks all, working now and the code is much skinnier!