[PATCH] Fix canonicalization of addresses

[Richard Guenther]

On Tue, Jan 6, 2009 at 4:13 PM, Richard Earnshaw <rearnsha@arm.com> wrote:
> On Tue, 2009-01-06 at 15:17 +0100, Eric Botcazou wrote:
>> > In the above case transforming -A * -CST to A * CST (which would handle
>> > the (1 - ind) * -2 case also if witten that way by a user, not only if
>> > generated by fold_plusminus_mult).  Thus, it subsumes the plusminus_mult
>> > patch in favor of a IMHO better one.
>>
>> I disagree.  The factorization by a negative power of 2 in plusminus_mult
>> doesn't serve any useful purpose and can be harmful, as shown here; instead
>> of relying on a later transformation to repair the damages, let's not create
>> them in the first place, this will save both time and memory.
>
> Seconded.  Furthermore, subtraction of a variable from a constant is
> less widely supported in CPUs than addition of small negative numbers --
> ie subtracting a constant (The Thumb instruction set has sub reg, const,
> but not sub const, reg).

To repeat: it's a canonicalization.  And it helps for canonicalizing a
* (-2) - 8 to (a + 4)*(-2)
and allowing CSE with another a + 4.  I don't think that in this case
we want to have
(-4 - a)*2.

And properly canonicalizing (1 - ind) * -2 to (ind - 1) * 2 is correct
anyway.  And as this
would exactly DTRT on the cases you complain about this is the correct
thing to do!
Instead of complicating an unrelated canonicalization routine.

Richard.