Sign in to follow this  
Misery

Hwo to read bit and set same bit in different byte

Recommended Posts

Misery    354
Hello,

I am using integers as bit containers representing bool values.
How do I make a fast instruction that reads some bit from one integer and sets its value to another.
Lets say: there are two 64 bit integers. One of them x has some bits set.
I want to rewrite every second bit from x to y (from beginning). How do I make that avoiding code similar to that:
[code]
y=0; //clear y
t=0;
for (int i=0;i<64;i+=2)
if (x & (CONTAINER_INT(1) << (i%ContainerSize)))
{
y |= CONTAINER_INT(1) << (t%ContainerSize);
t++;
}
[/code]

I mean I want to avoid if instruction.

Regards. Edited by Misery

Share this post


Link to post
Share on other sites
alh420    5995
why? I assume you also do something else then, besides the copying of every second bit?

If you need to do it bit by bit, I don't think you have any choice but to check it bit by bit.

My code copies all of them from x at the same time, not caring what they are, and conserve all of the other bits from y

Edit: this if you also need to copy zeros

y = (y&0xAAAAAAAAAAAAAAAALL) | (x&0x5555555555555555LL); Edited by Olof Hedman

Share this post


Link to post
Share on other sites
Brother Bob    10344
[quote name='Olof Hedman' timestamp='1350390352' post='4990691']
why? I assume you also do something else then, besides the copying of every second bit?

If you need to do it bit by bit, I don't think you have any choice but to check it bit by bit.

My code copies all of them from x at the same time, not caring what they are, and conserve all of the other bits from y
[/quote]
Your code doesn't match his. The bit index for the source and the destination are different variables; the source bit index is the loop index i, but the destination index is t which is stepped by one and only stepped for on-bits inside the if-statement. The behavior of his code is effectively: count the number of on-bits in every second bit index, and set that many bits in the destination. If the source contains 5 on-bits when counting every second bit, then the first 5 bits in the destination is set.

Although I'd say that his code and description doesn't convey the same message, and neither are very clear. A full description of the actual problem will help a lot.

Share this post


Link to post
Share on other sites
alh420    5995
[quote name='Brother Bob' timestamp='1350390907' post='4990694']
Your code doesn't match his. The bit index for the source and the destination are different variables; the source bit index is the loop index i, but the destination index is t which is stepped by one and only stepped for on-bits inside the if-statement. The behavior of his code is effectively: count the number of on-bits in every second bit index, and set that many bits in the destination. If the source contains 5 on-bits when counting every second bit, then the first 5 bits in the destination is set.

Although I'd say that his code and description doesn't convey the same message, and neither are very clear. A full description of the actual problem will help a lot.
[/quote]

Ah right. I actually assumed the t inside the if was a typo, since putting it outside in the for loop matches the description in the post and title

Share this post


Link to post
Share on other sites
Brother Bob    10344
[quote name='Olof Hedman' timestamp='1350391321' post='4990695']
[quote name='Brother Bob' timestamp='1350390907' post='4990694']
Your code doesn't match his. The bit index for the source and the destination are different variables; the source bit index is the loop index i, but the destination index is t which is stepped by one and only stepped for on-bits inside the if-statement. The behavior of his code is effectively: count the number of on-bits in every second bit index, and set that many bits in the destination. If the source contains 5 on-bits when counting every second bit, then the first 5 bits in the destination is set.

Although I'd say that his code and description doesn't convey the same message, and neither are very clear. A full description of the actual problem will help a lot.
[/quote]

Ah right. I actually assumed the t inside the if was a typo, since putting it outside in the for loop matches the description in the post and title
[/quote]
Actually, putting t++ in the loop itself makes yet another interpretation of the problem: it copies bit index 2*n to bit index n, since i increases by two and t increases by one each loop in that case.

The exact details here are important and, as I said, we need to know the actual problem. So far we've had two different descriptions, and yet another interpretation of it :)

Share this post


Link to post
Share on other sites
alvaro    21246
[quote name='Misery' timestamp='1350388346' post='4990682']
[code]
y=0; //clear y
t=0;
for (int i=0;i<64;i+=2)
if (x & (CONTAINER_INT(1) << (i%ContainerSize)))
{
y |= CONTAINER_INT(1) << (t%ContainerSize);
t++;
}
[/code]
[/quote]

[code]y = (CONTAINER_INT(1)<<popcount(x & 0x555555555555555ul)) - 1;[/code]

EDIT: popcount stands for "population count", which means count the number of bits set. gcc provides __builtin_popcount for you, or you can use one of the tricks described in [url="http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetNaive"]Bit Twiddling Hacks[/url]. Oh, and I am not sure why you are doing `%ContainerSize'. Edited by alvaro

Share this post


Link to post
Share on other sites
Misery    354
OK, so little more explanation:
I have a bitset matrix (B) and two indexing integer matrices (i,j).
What I want is to create another bitset matrix that consists of B entries at the indexes defined by i and j.
It is a similar behavior like that in Matlab or Scilab.

So what I need is to rewrite B(i,j) to new bitset matrix. B entries can be 1 or 0 so I need to rewrite them bit by bit, as new matrix can have different size from B.

To complicate even more used integer depend on the platform and may have 16 32 or 64 bits. This is dependent on the template parameter CONTAINER_INT.

Share this post


Link to post
Share on other sites
oliii    2196
something like this?

Not sure if the logic is correct.

[code]y = 0;
t = 0;
while(x)
{
y |= ((x&1) << t);
t++;
x >>= 2;
}[/code] Edited by papalazaru

Share this post


Link to post
Share on other sites
alvaro    21246
You need to be much more precise than that. In your description of the problem you speak of "matrices" B, i and j, which are nowhere to be found in your code. You also don't explain what you mean by "matrix", because the conventional definition doesn't match either.

Once we have a precise statement of the problem, hopefully with some examples, I am pretty sure I'll be able to write some basic code to do what you need.

Share this post


Link to post
Share on other sites
Misery    354
[b]@Alvaro[/b]:
[quote name='alvaro' timestamp='1350431259' post='4990929']
You need to be much more precise than that. In your description of the problem you speak of "matrices" B, i and j, which are nowhere to be found in your code. You also don't explain what you mean by "matrix", because the conventional definition doesn't match either.

Once we have a precise statement of the problem, hopefully with some examples, I am pretty sure I'll be able to write some basic code to do what you need.
[/quote]

You're right. However, if I wanted to paste the whole code It would require about 10k lines of code at the moment, so I decided to strip it a little. And my question considers only:[b] How to rewrite one integer to another BIT BY BIT[/b]. That is the part I am stuck at. I just have no idea how to write an instruction that does this: [b]if the i-th bit in integer x is set, that set a j-th bit in integer y, if the i-th bit in x is unset, unset j-th bit in y.[/b] Also notice, that used integers are template parameters/arguments so no constant values should be used, rather constructors like CONTAINER_INT(1) or CONTAINER_INT(0).
Hope this clears the problem. [img]http://public.gamedev.net//public/style_emoticons/default/smile.png[/img]

PS.: And I want to do it bit by bit. I just want to avoid if statement.


EDIT: After giving it some thought, I paste the whole function (only the line with if statement has to be changed):
[code]
template <class INT_TYPE,class CONTAINER_INT>
inline void BitsetChooseSerial(CONTAINER_INT *From,INT_TYPE FromRows,INT_TYPE *WhichRows,INT_TYPE NRows,INT_TYPE *WhichCols,INT_TYPE NCols, CONTAINER_INT *To)
{

//ToRows==NRows
static const INT_TYPE ContainerSize=sizeof(CONTAINER_INT)*8;

INT_TYPE ElementIndex,C,i,t,tC;

for (INT_TYPE j=0;j<NCols;j++)
{

C=WhichCols[j]*FromRows;

tC=j*NRows;

for (i=0;i<NRows;i++)
{

ElementIndex=WhichRows[i]+C;

t=i+tC;

if (From[ElementIndex/ContainerSize] & (CONTAINER_INT(1) << (ElementIndex%ContainerSize))) To[t/ContainerSize] |= CONTAINER_INT(1) << (t%ContainerSize); //change this to direct bit shift

}
}
}
[/code] Edited by Misery

Share this post


Link to post
Share on other sites
Misery    354
[quote name='papalazaru' timestamp='1350414682' post='4990846']
something like this?

Not sure if the logic is correct.

[code]y = 0;
t = 0;
while(x)
{
y |= ((x&1) << t);
t++;
x >>= 2;
}[/code]
[/quote]

Unfortunately it doesnt work [img]http://public.gamedev.net//public/style_emoticons/default/sad.png[/img]
But thanks for help anyway[img]http://public.gamedev.net//public/style_emoticons/default/smile.png[/img] Edited by Misery

Share this post


Link to post
Share on other sites
alh420    5995
too bad shift operators are undefined with negative shifts, otherwise I'd know how :)
but if you know which index is the higher one, I think you can do something like this:

y = y&~(CONTANTER_INT(1)<<j) | (x&(CONTAINTER_INT(1)<<i)) >> (i - j))

possibly you can check which is higher outside the loop, and at least avoid the if in the loop.

Share this post


Link to post
Share on other sites
Misery    354
[b]@Olof Hedman[/b]: Actually I don't get the idea of this index stuff. Why would I need to find the bigger index value? Edited by Misery

Share this post


Link to post
Share on other sites
alh420    5995
if you want to put bit 3 from x into bit 5 of y, that is i = 3, j = 5, then you have to shift the result of x&(CONTAINTER_INT(1)<<i) to the left, instead of right as the code does.

but to write a >> -2 is not allowed.

Share this post


Link to post
Share on other sites
Misery    354
@Olof Hedman: Oh, now i get it. But is there no way of doing it in the flow manner like: Read bit at some i in x, write this value to some j position of y? Without tangling indexes together?

Share this post


Link to post
Share on other sites
patrrr    1323
#define COPY_BIT(x, i, y, j) (y) = ((y) & ~(1 << (j))) | ((((x) >> (i)) & 1) << (j))

Something like that...? Can probably be optimized better. Edited by patrrr

Share this post


Link to post
Share on other sites
alvaro    21246
What patrrr wrote seems correct. I would do it with a function, though:
[code]unsigned copy_bit(unsigned x, int i, unsigned y, int j) {
y &= ~(1u << j); // Clear y's j-th bit
y |= ((x >> i) & 1u) << j; // Copy x's i-th bit into position j
return y;
}[/code]

The fact that you need to use some funky name for the type and that you are going for lots of generality is only confusing matters, so start without that.

Share this post


Link to post
Share on other sites
Misery    354
[b]@alvaro[/b] & [b]@patrrr[/b]: Thanks a lot Guys. That was exactly the thing I was looking for. [img]http://public.gamedev.net//public/style_emoticons/default/biggrin.png[/img]
alvaro: at the moment I cannot even count how many times you have halped me. Special thanks and huge respect to your skills!

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this