Re: Reading uninitialized memory

[Andrew Vaught]

On 27 Jan 2002, Kevin Ryde wrote:
 
> The question is basically whether to add a conditional jump over the
> read.  I think on the important processors it's best to avoid
> conditionals wherever possible.

   No, I'm just saying move the read.  The original code looks like:


  mp_size_t size = op->_mp_size;
  mp_limb_t low_limb = op->_mp_d[0];

  if (size > 0)
    return low_limb % ((mp_limb_t) 1 << (BITS_PER_MP_LIMB - 1));
  else if (size < 0)
    /* This convoluted expression is necessary to properly handle 0x80000000 */
    return ~((low_limb - 1) % ((mp_limb_t) 1 << (BITS_PER_MP_LIMB - 1)));
  else
    return 0;

I'm advocating changing it to


  mp_size_t size = op->_mp_size;

  if (size > 0)
    return op->_mp_d[0] % ((mp_limb_t) 1 << (BITS_PER_MP_LIMB - 1));
  else if (size < 0)
    /* This convoluted expression is necessary to properly handle
0x80000000 */
    return ~((op->_mp_d[0] - 1) % ((mp_limb_t) 1 << (BITS_PER_MP_LIMB -1)));
  else
    return 0;

 ie, don't read the word until you need to.

 
> It's also worth remembering that the only time the load is not wanted
> is when the value in the mpz_t is zero, and I think non-zero is
> probably the more common case.

  Right, and that's why I'm not saying that the conditional should be
rewritten.  The least common case should be last.

      Andy

-----------------                        XOLD(K,IC,I)=
Andy Vaught               ....        DO ITERS=1, 10  XOLD(K,IC,I)
address@hidden     |  |   /CALLMSOLVE(A,B,X,I,ITERS,TOL)+(RANNYU(0)
Arizona State University  ======|WRITE(6,'(I5,2X,F12.6)')ITERS,TOL -HALF)
Tempe, Arizona USA        OOOOOO \ENDDORETURN PARAMETER(ZERO=1.D0)*TENTH*DELTA