Re: [avr-gcc-list] gcc 4.3.0

[Ruud Vlaming]

On Sunday 13 April 2008 23:53, you wrote:
> On Sat, 12 Apr 2008 20:55:58 +0200, you wrote:
> >Also, I get an average increase of the codesize
On my 12 projects the results are:

Compared between different compiles on 4.3.0 only
> -fno-inline-small-functions
> -fno-split-wide-types
Both combined: better than without, but slightly worse than --param 
inline-call-cost=2
(thus use  inline-call-cost=2 over  -fno-inline-small-functions 
-fno-split-wide-types)
> -fno-tree-scev-cprop
No effect.

Of course other combinations are possible. I did not try them.

Compare between 4.2.3 and 4.3.0 with those three options activated the 
former still produces the smaller code overall (sum of all footprints), 
although 
in some cases the latter is somewhat smaller.

Given the number of optimization parameters GCC has, it might be interesting
to optimize the setting of the parameters themselves by some genetic algorithm
or so. I think this must be done per project. Disadvantage is that this requires
hunderds of compiles per project. But because it usually only makes sense for 
small projects this might be feasible.

Ruud.