Page 2 of 2
Re: [NXC] convert float to int
Posted: 01 Jun 2011, 19:49
by HaWe
muntoo wrote:Technically, the
round() method is usually implemented as such:
Code: Select all
inline long round(float x)
{
return(x + 0.5);
}
I think I misunderstood:
you don't mean: usually implemented already in NXC, did you?
I have been searching but couldn't find a "round method" in NXC.
but muntoo, no, you're wrong with your formula, try it by yourself with negative values.
round is usually implemented as
Code: Select all
inline long round(float f)
{
if (f>=0) return (f + 0.5);
else return (f - 0.5);
}
Re: [NXC] convert float to int
Posted: 02 Jun 2011, 06:21
by muntoo
doc-helmut wrote:you don't mean: usually implemented already in NXC, did you?
Nope.
doc-helmut wrote:but muntoo, no, you're wrong with your formula, try it by yourself with negative values.
Not wrong, just
biased.
Boost Implementations
I'm not sure how NXC handles implicity conversions from
float->int, so I'll
rewrite it with
floor():
Code: Select all
float round(float x)
{
return(floor(x + 0.5));
}
Tests:
Code: Select all
round( 3.1) 3
round( 3.5) 4
round( 3.9) 4
round( 0.5) 1
round( 0.0) 0
round(-0.1) 0
round(-0.5) 0
round(-0.6) -1
round(-1.1) -1
round(-1.5) -1
round(-1.9) -2
That should be "accurate" enough, right?
-----
If you've got CPU to waste, you can
use:
Code: Select all
float round(float r) {
return (r > 0.0) ? floor(r + 0.5) : ceil(r - 0.5);
}
Re: [NXC] convert float to int
Posted: 02 Jun 2011, 07:02
by HaWe
why so complicated and use functions inside a function?
this is the fastest code and as accurate as possible:
Code: Select all
inline long round(float f)
{
if (f>=0) return (f + 0.5);
else return (f - 0.5);
}
or, nested:
Code: Select all
inline long round(float f)
{
return (f>=0? f+0.5 : f-0.5);
}
provided or implied assiociativity is
round(-x)= -(round(x));
so should be
round(-0.5)=-1
round(-1.5)=-2
a.s.o.
Re: [NXC] convert float to int
Posted: 02 Jun 2011, 07:10
by muntoo
doc-helmut wrote:why so complicated and use functions inside a function?
As I said, I'm not sure about
float->int conversions in NXC, so this could potentially be used instead:
Code: Select all
inline long round(float x)
{
return(x + 0.5);
}
With greater speed and "accurate-enough-accuracy".
Re: [NXC] convert float to int
Posted: 02 Jun 2011, 07:14
by HaWe
no muntoo, don't be blase and narrow-minded, your formula is wrong for negative values!
see it at last!
And if you're not sure: test it (or simply believe it to me.)
this is valid:
Code: Select all
inline long round(float f)
{
return (f>=0? f+0.5 : f-0.5);
}
Re: [NXC] convert float to int
Posted: 02 Jun 2011, 07:28
by muntoo
Code: Select all
inline long round(float f)
{
return (f>=0? f+0.5 : f-0.5);
}
Does this mean
int(-0.1) == 0? Otherwise, if
int(-0.1) == -1, then
round(-0.1) == -1, using your method.
EDIT: It
indeed is
int(-0.1) == 0.
Stroustrup, C.6.2.6 wrote:The fractional part is discarded. In other words, conversion from a floating-point type to an integer type truncates.
Also, "right/wrong" really depends on your definition of it, what you're trying to do, and what you actually need. But, yeah, your formula should be the "correct" one for what you believe is correct.
Re: [NXC] convert float to int
Posted: 02 Jun 2011, 08:02
by HaWe
what's mathematically right or wrong is not decided by me or by you but by the mathematic laws and axioms or definitions.
but you may believe what you want, you are not forbidden to believe wrong facts.
Indeed "my" round() function returns for -0.1 => 0 and e.g. for -0.6 => -1
you might have seen that if you had tried.
And now let's stop the altercation.
Re: [NXC] convert float to int
Posted: 02 Jun 2011, 13:37
by gloomyandy
Hi Doc,
Although your definition may be mathematically correct your interpretation is not the same as that used my many floating point libraries. So for instance Java (and lots of other languages) defines round as:
(long)Math.floor(a + 0.5d)
http://download.oracle.com/javase/6/doc ... nd(double)
Which means that round(-0.5) is 0 not -1 (it would be -1 with your definition)...
I thought I would just give you another reason to hate Java...
Andy
PS There is a good discussion of the various issues here:
http://en.wikipedia.org/wiki/Rounding
if anyone is interested
Re: [NXC] convert float to int
Posted: 02 Jun 2011, 15:07
by HaWe
no, you didn't :)
the -0.5 rounding is indeed interpreted differently, but that's not the point.
If -0.49999999999... is rounded to 0 and -0.5 to -1
or
-0.5 is rounded to 0 and -0.500000000000....1 to -1
that's not a mathematic problem but just a convention and the result depends maybe only on the binary floating point notation of -0.5 and it's precision .
(is 0.99999999999999999... as a fp value ==1 ? mathematically, it is, of course, but you won't get this equation with computers probably...)
Some define a rounding from every exact i+ 0.5 (for all integer i either positive or negative) as the next bigger integer number (i+1).
the symmetric associative definition
round(-x)=-round(x)
is nevertheless more logical, but who said that Java programmers do think logically?^^
(Not even C programmers do it always and continuously.)