Reputation: 7629
Given a 64-bit number what's the best way to find out the number of paired un-set bits at even boundaries. The extra zero padding after the MSB should be ignored.
For example:
For the two numbers 25223 and 10578
25223 -- 01 10 00 10 10 00 01 11
7 6 5 4 3 2 1 0
Count = 2, (at positions 2 and 5)
10578 -- 00 10 10 01 01 01 00 10
7 6 5 4 3 2 1 0
Count = 1, (at position 1. Ignore position 7)
I could do a mask, shift-by-2 and compare, but I'm looking for something better. Is there anything faster than this:
def PairedCount(n):
c=0
while(n!=0):
if((n & 3) == 0):
c+=1
n >>= 2;
return c
What if I want to count the number of paired non-zero bits at even boundaries? What's the best method for that?
Upvotes: 0
Views: 337
Reputation: 9512
This is not much cheaper (one loop per pair of zeroes + overhead) but it's just to expose a few bit tricks.
size_t count_pairs_of_zeros( your_uint_type x );
{
// create a mask with even bits set like 0x55555555
// but independent of bit length
your_uint_type mask = 0;
mask -= 1;
mask /= 3;
// replace 01 and 10 pairs by 11
x |= x>>1;
x &= mask;
x |= x<<1;
// count the pairs of zeros up to most significant bit
size_t count = 0;
while( x & (x+1) )
{
count++;
// remove next pair of zeros
x |= x+1;
x |= x+1;
}
return count;
}
Upvotes: 0
Reputation: 14890
It's a simple question, but the way you put it scares me :)
Let's first try doing it to pairs of 1
s (you'll see why) for 32 bits:
unsigned count_pairs_1(unsigned n){
n = n & ( n >> 1); // bit N will be set if bits N and N+1 were set
n &= 0x55555555; // we need just those on even position, so ANDing with 0b01..0101
return count_set_bits(n); // now we need the number of 1 bits in the result
};
All we need now it count_set_bits(unsigned)
, that is very known function: http://www-graphics.stanford.edu/~seander/bithacks.html#CountBitsSetTable
To count zero bits use count_pairs(~n)
or
unsigned count_pairs_0(unsigned n){
n = n | ( n >> 1); // bit N will be zero iff bits N and N+1 were zero
n |= 0xAAAAAAAA; // all odd bits are set
return 32 - count_set_bits(n); // every remaining zero bit corresponds to zero pair in the input
};
EDIT: just observed the remark Given the 64 bit number... The extra zero padding after the MSB should be ignored. After what MSB? Do you mean the input is a byte? or word?
Upvotes: 3
Reputation: 8250
unsigned count_pairs_0_n(unsigned n){
unsigned int i=n;
unsigned int l=0;
while(i){l=i;i&=(i-1);}
n=((l<<1) -1) &(~n);
return count_pairs_1(n);
}
based on @rusliks answer, I tried making my answer a bit short.
Upvotes: 1
Reputation: 8250
This is for 32bits .. 0x55555555 is a dependency .. is order of number of set bit
int countpairs(int n){
int o=n;
int c=0;
unsigned int i=n;
unsigned int l=0;
while(i){l=i;i&=(i-1);}
n=((l<<1) -1) &(~n);
while(n){
unsigned int k= n&(n-1);
unsigned int k2=k&(k-1);
unsigned int k3=(n^k) + (k^k2);
if((n^k) && k^k2 && (n^k)*2 == (k^k2) && ((n^k) & 0x55555555)) {
c++;
}
n=k;
}
return c;
}
Upvotes: 0