By: Klimax (danklima.delete@this.gmail.com), July 11, 2013 10:51 pm
Room: Moderated Discussions
Gabriele Svelto (gabriele.svelto.delete@this.gmail.com) on July 11, 2013 4:20 am wrote:
> Klimax (danklima.delete@this.gmail.com) on July 10, 2013 11:30 pm wrote:
> > (And I suspect it is only matter of time when GCC will gain that)
>
> Actually that's unlikely to happen. GCC is the largest compiler out there covering most languages
> and architectures combined; as such new optimization passes are accepted only if they yield measurable
> benefits over a broad range of real world codes (or a single load which happens to be very important
> or common). Optimization passes that improve benchmark scores alone are rejected by default.
Sure, but so far nobody showed that this instance is bench-only. (And reduction, elimination or at least speed-up of for-loops is always popular target.)
See memcpy/memset... (and there might be some other cases inside or outside stdlib)
> Klimax (danklima.delete@this.gmail.com) on July 10, 2013 11:30 pm wrote:
> > (And I suspect it is only matter of time when GCC will gain that)
>
> Actually that's unlikely to happen. GCC is the largest compiler out there covering most languages
> and architectures combined; as such new optimization passes are accepted only if they yield measurable
> benefits over a broad range of real world codes (or a single load which happens to be very important
> or common). Optimization passes that improve benchmark scores alone are rejected by default.
Sure, but so far nobody showed that this instance is bench-only. (And reduction, elimination or at least speed-up of for-loops is always popular target.)
See memcpy/memset... (and there might be some other cases inside or outside stdlib)