Can someone explain to me what this does?
#define ROUNDUP(n,width) (((n) + (width) - 1) & ~unsigned((width) - 1))
views:
186answers:
4
+6
A:
Providing width is an even power of 2 (so 2,4,8,16,32 etc), it will return a number equal to or greater than n, which is a multiple of width, and which is the smallest value meeting that criteria.
So width = 16; 5->16, 7->16, 15->16, 16->16, 17->32, 18->32 etc.
EDIT I started out on providing an explanation of why this works as it does, as I sense that's really what the OP wants, but it turned into a rather convoluted story. If the OP is still confused, I'd suggest working through a few simple examples, say width = 16, n=15,16,17. Remember that & = bitwise AND, ~ = bitwise complement, and to use binary representation exclusively as you work through the examples.
Bill Forster
2010-01-21 04:36:01
Note that the width *must* be a power of two... there are some cases where roundup seems to work for other widths (roundup width 3 works for n <5 for example) but only in a few bizarre corner cases.
Sam Post
2010-01-21 04:40:34
@Bill - an *even* power of 2? That's odd...
tony
2010-01-21 05:16:10
Cute tony :-) Since 2^(1.58496298..) = 3 (say), I think it is reasonable to add an extra qualifier to indicate that powers of 2 of that type are excluded. I think an "even power of two" is a common if colloquial way of doing that. Sorry for a pedantic reply to a humorous comment :-).
Bill Forster
2010-01-21 05:28:57
@Bill here I was thinking you were excluding 1 ;) since that makes the part `~unsigned((width) - 1)` equal to 0 and thus the whole result equal to 0.
JPvdMerwe
2010-01-21 06:47:32
@JPvdMerwe No I think we can include 1, because the mask isn't then 0, it's ~0 (complement of 0, i.e. all 1s) and the macro is thus a no-op, i.e. n->n, which is as you'd expect (round up to the nearest multiple of 1 = do nothing).
Bill Forster
2010-01-21 07:58:58
You're right I had the orders of the operators in my head mixed up. It's when `width = 0` that what I said happens.
JPvdMerwe
2010-01-21 08:26:42
@Bill - no problem.
tony
2010-01-21 15:05:41
A:
It won't work in C because of the unsigned. Here is what is does, as long as width is confined to powers of 2:
n width ROUNDUP(n,width)
----------------
0 4 0
1 4 4
2 4 4
3 4 4
4 4 4
5 4 8
6 4 8
7 4 8
8 4 8
9 4 12
10 4 12
11 4 12
12 4 12
13 4 16
14 4 16
15 4 16
16 4 16
17 4 20
18 4 20
19 4 20
wallyk
2010-01-21 04:38:35
+2
A:
It rounds n up to the next 'width' - but I think width needs to be a power of 2.
For example width == 8, n = 5:
(5 + 8 - 1) & ~(7) = 12 & ~7 = 8
So 5 rounds to 8. Anything 1 - 8 rounds to 8. 9 to 16 rounds to 16. Etc. (0 rounds to 0)
tony
2010-01-21 04:38:29
+2
A:
It defines a macro called ROUNDUP which takes two parameters, n and width, and returns the value (n + width - 1) & ~unsigned(width - 1).
:)
Try this if you think you know what it does:
std::string s("WTF");
std::complex<double> c(-11,5);
ROUNDUP(s, c);