[mercury-users] Optimisation of functions and field updates

[Julien Fischer]

On Tue, 8 Aug 2006, Julien Fischer wrote:

>
> On Tue, 8 Aug 2006, Julien Fischer wrote:
>
>> 
>> On Mon, 7 Aug 2006, Michael Day wrote:
>> 
>>> Say I have the following code:
>>> 
>>> 	X = f1(f2(f3(f4(X0))))
>>> 
>>> where
>>> 
>>> 	f1(X) = X ^ field1 := blah.
>>> 	f2(X) = X ^ field2 := blah.
>>> 	...
>>> 
>>> If the functions were inlined, then the multiple field updates could be
>>> combined into a single update, but it appears that this is not happening,
>>> as I get a speed boost when I replaced the function calls with explicit
>>> field updates, even at -O5.
>> 
>> It looks like the simplification passes after inlining aren't cleaning
>> up as well as they should.  I've just constructed a small example and
>> it seems to leave redundant assignments hanging around.  This seems
>> to be more of a problem in the MLDS backend.
>
> In fact looking at the generated C code more closely it is only
> a problem for the MLDS backnd, the LLDS->LLDS optimizations
> clean up the redundant assignments (at least for my example).
>

That said, gcc *does* clean them up, so I'm at a loss to say what
is happening here.

Julien.
--------------------------------------------------------------------------
mercury-users mailing list
Post messages to:       mercury-users at csse.unimelb.edu.au
Administrative Queries: owner-mercury-users at csse.unimelb.edu.au
Subscriptions:          mercury-users-request at csse.unimelb.edu.au
--------------------------------------------------------------------------