Re: My vectorized code is slower than a loop

[Dave Cottingham]

I had come to a similar conclusion, from looking at (and reproducing) Nicholas' timing results. In my original vectorized version, the operations spent computing the index ranges are more than what's saved by vectorizing. I did some precomputing of the ranges, similar to yours, and managed to get the vectorized version no slower than the loop.

As dismaying as that is, I think I see why: at each i, the number of things copied is a few or none, so there's nothing to be gained by vectorizing. This code fragment can only be sped up by eliminating the loop on i -- and I see no way to do that.

My conclusion is there's no alternative to implementing this is c++ (which I've already done).

Thanks,

Dave Cottingham

On Wed, Oct 19, 2016 at 4:19 PM, Francesco Potortì <address@hidden> wrote:

No profiling, no runs, just looking at the code, so I may be off.

w = zeros(1, n);
j = 1
for i = 1:n
  if (vectorised)
    if(left(i) <= right(i))
      w(j:j+right(i)-left(i)) = x(i) - x(n+1-[left(i):right(i)]);
      j = j + right(i)-left(i)+1;
    endif
  else                          # looping
    if(left(i) <= right(i))
      for jj = left(i):right(i)
        w(j) = x(i) - x(n+1-jj);
        j += 1;
      endfor
  endif
endfor

In the loop version, the if is useless, because the for range will be
empty, so the for will not run when the if clause is false.  This may
result in a speedup or not.

In the vectorised version, I would write it like this, which minimises
the number of operations at the expense of the number of instructions,
which may result in a speedup or not:

   li = left(i);
   d = right(i) - li;
   if (d >= 0)
     dr = 0:d;
     w(j+dr) = x(i) - x(n+1-li+dr);
     j += d+1;
   endif

--
Francesco Potortì (ricercatore)        Voice:  +39.050.621.3058
ISTI - Area della ricerca CNR          Mobile: +39.348.8283.107
via G. Moruzzi 1, I-56124 Pisa         Skype:  wnlabisti
(entrance 20, 1st floor, room C71)     Web:    http://fly.isti.cnr.it