<p>Am 19.05.2017 22:24 schrieb "Sven Barth" <<a href="mailto:pascaldragon@googlemail.com">pascaldragon@googlemail.com</a>>:<br>
><br>
> On 19.05.2017 19:22, Karoly Balogh (Charlie/SGR) wrote:<br>
> > Hi,<br>
> ><br>
> > On Fri, 19 May 2017, Sven Barth via fpc-pascal wrote:<br>
> ><br>
> >> I think Jeppe wanted to add vector support. Though the question here is<br>
> >> whether one wants to optimize/detect this at the AST level and convert<br>
> >> that to implicit vectors or at the CSE level.<br>
> ><br>
> > I think the higher level you can do an optimization/simplification, the<br>
> > higher you should do it. Otherwise the lower layers get really messy, as<br>
> > they already are in some cases. Well, in general, we should up our<br>
> > floating point game. For example if Nikolay's recent load-modify-store<br>
> > optimization would work on floats, that would already a nice step forward<br>
> > in this case. ;) (Sorry for my ignorance, if it already works, missed that<br>
> > then.)<br>
><br>
> I agree that we should improve that. Maybe we should also allow for more<br>
> FPU type specific helper routines. Currently on i386 and x86_64 the x87<br>
> FPU will be used even if -CfsseX is given and only Single/Double are<br>
> used, cause the compiler defaults to Extended. If SSE isn't used that<br>
> might make sense, but for SSE we should fall back to Double if we're<br>
> only dealing with double, IMHO (and analogous for Single).</p>
<p>And you might notice that I had written this before I had discovered the existence of the professorspecific inline nodes which I commented on below that, so forgot what I wrote here aside from the fact that we should indeed improve things regarding floating point ^^'</p>
<p>Regards,<br>
Sven</p>