Re: Common subexpression analysis

Bruce.Hoult@bbs.actrix.gen.nz writes:
[Original article deleted]
>Why didn't you do strength reduction while you were at it?
>
> r2 = r*r;
> x2 = x*x;
> y2 = y*y;
> while (y >= 0)
> if (x2 + y2 > r2)
> --y, y2 -= y+y-1;
> else
> ++x, x2 += x+x+1;


Or somewhat better:


..
if (x2+y2 > r2)
y2-=y, --y, y2-=y;
else
x2+=x, ++x, x2+=x;


Or even better:


goto L0;


L2:
                  if (x2+y2>r2) goto L3;
x2+=x;
x++;
                  x2+=x;
goto L2;


L3:
y2-=y;
--y;
y2-=y;
L0:
if (y>=0) goto L2;




(I hope I got it right. The basic idea is that it's a waste of time
to test if y>=0 if y hasn't changed).




>
>
>... or even go whole hog and do...
>
> t = x*x + y*y - r*r;
> x2 = x+x+1;
> y2 = y+y-1;
> while (y2 >= 0)


Should read:
while (y2 >= -1)


> if (t > 0)
> y2 -= 2, t -= y2;
> else
> x2 += 2, t += x2;
>
>
>I'll bet if there are any compilers that can do this, they're probably
>for FORTRAN :-)


Probably true.




--

Return to the comp.compilers page.
Search the comp.compilers archives again.