[pooma-dev] Re: InlineEvaluator implementation question

[Mark Mitchell]

--On Monday, December 16, 2002 09:00:35 PM +0100 Richard Guenther 
<rguenth at tat.physik.uni-tuebingen.de> wrote:

> On Mon, 16 Dec 2002, Mark Mitchell wrote:
>
>>
>> > Hmm - as both, lhs and rhs are declared const, isnt this enough to tell
>> > the compiler? Or has the compiler to assume every function call can
>> > have a side-effect on any (but local) variable?
>>
>> This is the classic "pointer aliasing" problem for compilers; the
>> modifications to lhs (via the write operatin in the loop) might,
>> from the compiler's point of view, change the rhs.  The copies allow
>> a good compiler to tell that this cannot happen.
>
> Ah, of course. But only change the data (via the engines data pointer
> which is aliasing), the lhs/rhs objects themselves usually will not alias
> as they are of different types (in nearly all cases). So the copy doesnt
> help the aliasing problem?

It's more complicated than that, and different compilers handle it
differently.  It's true that typed-based aliasing helps -- but it
doesn't always help enough.

That code is there because Jeffrey and I sat and stared at KCC output for
a week -- we know it helps for that compiler.

Unfortunately, I'm going to have a hard time reconstructing our full
analysis.

-- 
Mark Mitchell                mark at codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com