Re: [Sc-devel] float->int casts in OscUGens.cpp (and elsewhere?)

["Ross Bencina" <rossb-lists@xxxxxxxxxxxxxx>]

Hi Dan

I don't have a viable setup here to built and test this stuff at the moment 
but...

I think lrint() is the C99 function, see here:
http://www.mega-nerd.com/FPcast

It's probably worth trying that, or the FLOAT_TO_INT macro on that page, at 
least to see whether it improves performance.

Given that we're only really targetting GCC and MSVC (or equivalent) it 
shouldn't be too hard to make a macro which covers all cases.

Ross.



----- Original Message ----- 
From: "Dan Stowell" <danstowell@xxxxxxxxx>
To: <sc-devel@xxxxxxxxxxxxxxx>
Sent: Thursday, January 10, 2008 3:02 AM
Subject: Re: [Sc-devel] float->int casts in OscUGens.cpp (and elsewhere?)


> Hi -
>
> I don't remember this ever having come up in discussion. Thanks for
> pointing it out.
>
> I personally wouldn't know what to do along the assembly route, others
> here probably would; but do you have any hints re the C99? (Not sure
> if you're thinking about specific functions, or specific ways to write
> the expression for the compiler to optimise, etc.) If we can try
> rejigging the C, the improvement could be very straightforward.
>
> Dan
>
>
> 2008/1/9, Ross Bencina <rossb-lists@xxxxxxxxxxxxxx>:
> > Hi Guys
> >
> >
> >
> > I notice that OscUGens.cpp contains a number of direct downcasts from 
> > double
> > to int32, such as line 1840:
> >
> >
> >
> > phase += (int32)(cpstoinc * ZXP(freqin));
> >
> >
> >
> > I assume it's been discussed here before, but this is a particularly
> > inefficient thing to do on IA32 processors. Luckily this assignment isn't
> > too sensitive to which way the rounding goes, so there are some simple
> > assembly solutions (such as using fistp to store the intermediate value 
> > into
> > an integer location). I think there are even some standard C99 
> > functions...
> >
> >
> >
> > Is anyone in a position to take care of this?
> >
> >
> >
> > Cheers
> >
> > Ross.
> > _______________________________________________
> > Sc-devel mailing list
> > Sc-devel@xxxxxxxxxxxxxxx
> > http://lists.create.ucsb.edu/mailman/listinfo/sc-devel
>
>
> --
> http://www.mcld.co.uk
> _______________________________________________
> Sc-devel mailing list
> Sc-devel@xxxxxxxxxxxxxxx
> http://lists.create.ucsb.edu/mailman/listinfo/sc-devel