[v3] conditional tweak
Gabriel Dos Reis
gdr@integrable-solutions.net
Sat Nov 8 18:49:00 GMT 2003
Jerry Quinn <jlquinn@optonline.net> writes:
| Gabriel Dos Reis writes:
| > Phil Edwards <phil@codesourcery.com> writes:
| >
| > | Here's my opinion before I vanish into non-networked-land for a few days:
| > | for years, the GCC team has been warning of the dangers of slowing down the
| > | compiler and runtime by a few percent at a time, because it accumulates
| > | so quickly.
| >
| > I'm just asking for an actual data. Can we have one?
| >
| > -- Gaby
|
| I didn't actually see any difference on an AMD XP2100, but the only
| other platform I can really test is a mobile P4.
|
| I submitted the patch on the theory that this is a frequently called
| function in output numbers and this should potentially allow for
| improvements on machines that can benefit. The only thought I had
| against it was that machines usually predict against forward branches,
| in which case, they are already biased away from the initialization
| code.
Thanks for the input!
I'm not against this kind of patch in principle. However, before it
gets committed, I would like to see actual data -- because it has
uglicifation components that should be justfied.
This is not directed to you in particular, it is a general comment: I
don't buy arguments saying that "this ain't possibly do xxx, thus I
see no yyy not to have it." We should be able to justify software
construction on better grounds. We've made past decisions based on
such arguments; they have not proven to be a bright success.
-- Gaby
More information about the Libstdc++
mailing list