tree-ssa performance (was: treelang fix for rs6000 / treelang notproper front end)

[Gerald Pfeifer <pfeifer at dbai dot tuwien dot ac dot at>]

[ gcc-patches -> gcc ]

On Tue, 10 Jun 2003, Steven Bosscher wrote:
>> (See PR8361 for an example.)
> That's not quite fair on tree-ssa.  It really isn't that slow AFAICT.

It _is_, at least right now.  About factor of two for -O2 and -O3.

My last tests were on Friday and I just redid them from scratch with
current mainline and current tree-ssa:

-O0
    3.0.4	25.99
    3.2.2	29.51
 mainline	30.63
 tree-ssa	35.31
-O1
       44.34
       54.86
       59.30
       96.95
-O2
       55.56
       71.82
       82.29
      160.26
-O3
       55.27
       74.79
       87.14
      162.64

  But if you go look
> at the individual built times from Diego's SPEC testers, you'll see that
> tree-ssa is really not _that_ much slower: Something in the order of
> 10%

This indicates that the SPEC tester is not sufficient wrt. to compile-time
regressions.

On Tue, 10 Jun 2003, Diego Novillo wrote:
> After we are done taking things out and truly optimizing what we've
> added, we will get a faster compiler.
>
> Trust me :)

I'll be happy to do so. ;-)

Gerald