Getting rid of GET_RTX_CLASS?

Getting rid of GET_RTX_CLASS?

[Paolo Bonzini]

I'd like to turn rtx classes from characters to enums.  Would this be
feasible?  Interesting?  It may result in slight compile time speedups,
also.

Paolo

Re: Getting rid of GET_RTX_CLASS?

[Richard Henderson]

On Tue, Feb 03, 2004 at 05:56:59PM +0100, Paolo Bonzini wrote:
> I'd like to turn rtx classes from characters to enums.  Would this be
> feasible?  Interesting?  It may result in slight compile time speedups,
> also.

*shrug* I'm not sure why there would be any particular advantage...


r~

Re: Getting rid of GET_RTX_CLASS?

[Geoff Keating]

"Paolo Bonzini" <paolo.bonzini@polimi.it> writes:

> I'd like to turn rtx classes from characters to enums.  Would this be
> feasible?  Interesting?  It may result in slight compile time speedups,
> also.

I don't see any particular reason why this would be impossible (or
even difficult; just tedious).  I also think it's probably a good
idea, although not a very high-priority one.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: Getting rid of GET_RTX_CLASS?

[Zack Weinberg]

Richard Henderson <rth@redhat.com> writes:

> On Tue, Feb 03, 2004 at 05:56:59PM +0100, Paolo Bonzini wrote:
>> I'd like to turn rtx classes from characters to enums.  Would this be
>> feasible?  Interesting?  It may result in slight compile time speedups,
>> also.
>
> *shrug* I'm not sure why there would be any particular advantage...

Compact switch tables, presumably.

zw

Re: Getting rid of GET_RTX_CLASS?

[Paolo Bonzini]

> >> I'd like to turn rtx classes from characters to enums.  Would this be
> >> feasible?  Interesting?  It may result in slight compile time speedups,
> >> also.
> 
> Compact switch tables, presumably.

Yes; lookup tables instead of binary searching.  I also hope to learn something 
while reading the code -- low priority things are good for beginners (like me).

I'd also like to introduce new predicates (like OBJECT_P I already have a patch 
for) that express conditions like

  GET_RTX_CLASS (x) == '2'
  || GET_RTX_CLASS (x) == '<'
  || GET_RTX_CLASS (x) == 'c'

that abound in cse and other places that may be put under a diet after tree-ssa 
is merged.  These could become GET_RTX_CLASS (x) >= BINARY && GET_RTX_CLASS (x) 
<= BINARY_COMMUTATIVE (cheaper) as long as this is found only once, in rtl.h.

I suspect that an inordinate amount of time is spent in rtl.h if my very simple 
OBJECT_P patch had a 1% gain on bootstrap time.

Paolo

-------------------------------------------------
This mail sent through IMP: http://horde.org/imp/

Re: Getting rid of GET_RTX_CLASS?

[Eric Christopher]

> 
> Yes; lookup tables instead of binary searching.  I also hope to learn something 
> while reading the code -- low priority things are good for beginners (like me).

Actually it'll help readability too. It being so much easier to know
what 'c' is in rtx_class than something like say "COMMUTATIVE_BINARY".
:)

-eric

-- 
Eric Christopher <echristo@redhat.com>

Re: Getting rid of GET_RTX_CLASS?

[Theodore Papadopoulo]

rth@redhat.com said:
>> I'd like to turn rtx classes from characters to enums.  Would this be
>> feasible?  Interesting?  It may result in slight compile time
>> speedups,
>> also. 

> *shrug* I'm not sure why there would be any particular advantage... 

Better names under gdb....

--------------------------------------------------------------------
Theodore Papadopoulo
Email: Theodore.Papadopoulo@sophia.inria.fr Tel: (33) 04 92 38 76 01
 --------------------------------------------------------------------