bitset error for uint64

There seems to be a problem with the bitset function for uint64. Here is a simple example:

>> d = 38124631952277504;

>> dec2bin(d)

ans =

'10000111011100100010011010011000101101010100000000000000'

>> dec2bin(d+1)

ans =

'10000111011100100010011010011000101101010100000000000000'

>> dec2bin(bitset(d,1,'uint64'))

ans =

'10000111011100100010011010011000101101010100000000000000'

>> dec2bin(bitset(d,2,'uint64'))

ans =

'10000111011100100010011010011000101101010100000000000000'

>> dec2bin(bitset(d,3,'uint64'))

ans =

'10000111011100100010011010011000101101010100000000000000'

>> dec2bin(bitset(d,4,'uint64'))

ans =

'10000111011100100010011010011000101101010100000000001000'

>> dec2bin(d+8)

ans =

'10000111011100100010011010011000101101010100000000001000'

So the bits only change successfully from 4th bit on, but fail to set fo the first three bits. This makes no sense that I can see and seems to be a serious error in Matlab.

Any help would be appreciated. Or advice on how to alert Matlab engineers.

Thanks,

Jason

Answer 1

[Utkarsh Belwal]

Read the documentation of dec2bin, it is written that if number is greater than flintmax then it might not work properly. In your case d is greater than flintmax.

[Jason Palmer]

Ok, thanks, should have read that. Just assumed it worked since it returns an answer.

[Jason Palmer]

Probably better to return error when d is greater than flintmax, since gambling on "may be correct" seems like it would never be an acceptable situation when you are dealing with bit setting.

Answer 2

[Steven Lord]

>> d = 38124631952277504;
>> dec2bin(d)
ans =
    '10000111011100100010011010011000101101010100000000000000'
>> dec2bin(d+1)
ans =
    '10000111011100100010011010011000101101010100000000000000'

d is a double precision array, and the distance from d to the next largest representable number is greater than 1. In fact, it's 8.

>> eps(d)
ans =
     8
     
>> isequal(d, d+1) % returns true

If you want to work with integer data that's this large, I recommend working with integer data rather than double. Make your d variable an uint64 from the beginning. This will also allow you to see the change you're making with each bitset call without needing dec2bin.

>> du = uint64(38124631952277504)
du =
  uint64
   38124631952277504

>> du1 = bitset(du, 1)
du1 =
  uint64
   38124631952277505

>> du2 = bitset(du, 2)
du2 =
  uint64
   38124631952277506
   
>> du3 = bitset(du, 3)
du3 =
  uint64
   38124631952277508
   
>> du4 = bitset(du, 4)
du4 =
  uint64
   38124631952277512

If you do need to check the bits explicitly, use bitget. It's vectorized so you can get all the bits at once. If you need it as a char vector, you can combine bitget with sprintf.

bitget(du, 64:-1:1)
bitget(du1, 64:-1:1)
sprintf('%d', bitget(du4, 64:-1:1))

[Jason Palmer]

Ah, great, thanks for the explanation. Makes sense now. Cheers!