Remove algo logic duplication Round 3
Paolo Carlini
paolo.carlini@oracle.com
Mon Sep 23 22:00:00 GMT 2013
On 9/23/13 4:48 PM, Marc Glisse wrote:
> On Mon, 23 Sep 2013, Paolo Carlini wrote:
>
>> On 9/23/13 4:35 PM, Marc Glisse wrote:
>>> I was wondering if we should mark the predefined ops as inline to
>>> help the compiler early-inline them and make it more likely it will
>>> generate the same code as before, or if that may prevent the
>>> compiler from doing even better choices. Probably not worth the
>>> trouble.
>> It's one of the few things I checked today in detail, and I think all
>> of them are already inline, either because explicitly declared as
>> such, or because defined in the body of a struct. Or you mean
>> something else?
> I meant putting the "inline" keyword explicitly, because the inliner
> uses that as a heuristic. But forget it.
Does it really do that?!? Frankly I had no idea that was the case, and
in the library we (I think I can say we, because the code says that)
always assume that since the inline is implicit at least for linkage
purposes it was for actual inlining purposes too, end of the story. Is
that "hint" officially documented somewhere or you reverse engineered
somehow by yourself?
Paolo.
More information about the Libstdc++
mailing list