abs() in code component

Shifrin · 25 Dec 2005

Borogove wrote: Code component doesn't seem to recognize abs(). That can be done with negate and max pretty easily, but it would be high on my list for additional builtin functions.

Yep, there are a lot of time-saving functions that would be nice to see added (personally I'd love floor/ceiling/round and abs - there are simple workarounds but they obfuscate the code!).

Shif.

infuzion · 26 Dec 2005

I always wondered if adding a bunch more functions it would slow down the Code output....

Shifrin · 26 Dec 2005

I wouldn't have thought it'd slow down that much by having new functions. I believe SM uses the FPU to round numbers for example, at the moment I had to add code in order to force it to round down - but the FPU can do that automatically by setting a flag.

If I wanted to calculate x = abs(x), I'd much rather just write x = abs(x) and have it converted to whatever, x = (x < 0)&(-1*x) + (x >= 0)&x, then the code would be much easier to follow.

Perhaps we could have some sort of #define mechanism for things like that in the mean time?!

Oh yeah, you could also calc abs(x) with only one comparison like, x = x - (x < 0)&(2*x) 😉

Shif.

stefan · 26 Dec 2005

Actually you can do an abs() on floating point numbers by just nulling the sign bit.

Can't be done in the code component atm though, but should be fairly simple to add support for (hex defined variables would be good for such a thing).

Borogove · 26 Dec 2005

infuzion wrote: I always wondered if adding a bunch more functions it would slow down the Code output....

You don't pay for what you don't use. It would take some tiny fraction of a second longer for the Code component to *compile* code if it had to recognize and translate abs() into the equivalent machine code, but the runtime would be optimal.

Shifrin · 27 Dec 2005

stefan wrote: Actually you can do an abs() on floating point numbers by just nulling the sign bit.

Can't be done in the code component atm though, but should be fairly simple to add support for (hex defined variables would be good for such a thing).

If there's a hard way to do something I'll usually find it 😉

But as you say it's not doable in the Code Component ATM 🙂

Shif.

Borogove · 27 Dec 2005

Shifrin wrote: stefan wrote: Actually you can do an abs() on floating point numbers by just nulling the sign bit.

Can't be done in the code component atm though,

But as you say it's not doable in the Code Component ATM 🙂

Best I managed was:

foo = max( foo, 0-foo );

...which probably turns into a couple of moves, two subps and a maxps, not too bad.

Shifrin · 27 Dec 2005

Borogove wrote: Best I managed was:

foo = max( foo, 0-foo );

...which probably turns into a couple of moves, two subps and a maxps, not too bad.

Sweet 😉

Shif.

sambean · 28 Dec 2005

Interesting..

In the audio section of the waveshaper, I use the method stefan suggested using asm- creating a mask for all the magnitude bits of the float number(01111.. in binary or 2^31-1 as an int), then '&' the mask with the variable, so you disregard the sign bit..

As Stefan suggests if we could define this mask in the code component we could do it entireley in code

Here's a workaround for creating the mask until we can define hex/int values..

streamin in; streamout abs;

float absmask;

stage(0) {

// we want to define absmask to be: // 01111111111111111111111111111111 // // using bitwise 'or' (|) the bits of these 3 values cover the range:

absmask = 3.4e38|0.999999|0.1; }

stage(1) { abs = in&absmask; // keep in's magnitude only }

😀

Shifrin · 28 Dec 2005

Woah, that's great Sam, the things I learn here are very useful to me 😉

Thanks,

Shif.