This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: PTR-PLUS merge into the mainline


Hi,

On Thu, 5 Jul 2007, Richard Guenther wrote:

> > What makes "(i + 1) * 4" the canonical form of "(i * 4) + 4" compared to 
> > other expressions like "(i * 4) + 8"?
> 
> It's an arbitrary decision by fold.  For (i + 2) * 4 the canonical form
> is (i * 4) + 8.  For (i * j) + j the canonical form is (i + 1) * j,
> for (i * j) + 2 * j it is (i + 2) * j.  Now we can easily make constants
> special here.

What makes this arbitrary decision a good decision considering that it 
lacks any context?

> > >  The real fix is in the
> > > value numberer that should value number both kinds the same.
> > 
> > Sorry, I have no idea what this means.
> 
> This means that without context, neither (i + 1) * 4 nor (i * 4) + 4
> is better or cheaper.  With context that allows simplification this
> simplification should be possible regardless of what form the
> expression is in.

Maybe it should, but that doesn't make the job of other optimizers simpler 
if they have to deal with such a mixture of expressions.
For example how should one find all the common subexpression (i * 4) this 
way?

>  Remember that fold only does the transformation
> if it sees the full expression, so writing
> 
>   tmp1 = i + 1;
>   tmp1 = tmp1 * 4;
>   tmp2 = i * 4;
>   tmp2 = tmp2 + 4;
>   if (tmp1 == tmp2)
>     ...
> 
> should be optimized as well, but fold does nothing about that.

Exactly and that's why I think this transformation is done far too early. 
It doesn't make sense that these two examples produce different code:

int foo1(int x)
{
	return (x * 4) + 4;
}
int foo2(int x)
{
	int y = x * 4;
	return y + 4;
}

bye, Roman


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]