swapbytes order problem with uint64?

I find that swapbytes doesn't seem to be behaving correctly for my problem. I have a 64 bits hexidecimal string printed in big-endian: 'b0120c0a7799ba3e'

Manually swapping the bytes give me the expected answer:

    typecast(uint64(hex2dec('3eba99770a0c12b0')), 'double')
    ans =
     1.5855e-06

If I follow the example code in swapbytes however, it seems behave incorrectly. Either:

typecast(swapbytes(uint64(hex2dec('b0120c0a7799ba3e'))), 'double')
ans =
  3.5031e-305

or

typecast(uint64(swapbytes(hex2dec('b0120c0a7799ba3e'))), 'double')
ans =
     0

The function seem to work correctly for a single precision hexadecimal:

typecast(swapbytes(uint32(hex2dec('0cc9d435'))),'single') 
ans =
  single
  1.5854e-06

Am I missing something, I can certainly write some loops to parse the string but is there a better solution?

Answer 1

[Matt J]

I suspect the problem is actually in hex2dec(). You are exceeding the flintmax limit described in the hex2dec documentation:

d = hex2dec('hex_value') converts hex_value to its floating-point integer representation. The argument hex_value is a hexadecimal integer stored as text. If the value of hex_value is greater than the hexadecimal equivalent of the value returned by flintmax, then hex2dec might not return an exact conversion.

[Matt J]

Tony Tse commented:

That actually makes a lot of sense so I definitely need to swap the order of the hex string before conversion wit my own routine before conversion then. Thanks.

[Matt J]

Even if you swap the hex string yourself, I suspect you still have to make sure the swapped result obeys flintmax.

[Tony Tse]

In my case, because the string comes from real measured data. I think it should be never by out of range? so ended just up doing something like this:

s = 'b0120c0a7799ba3e';
s = fliplr(s);
ii = 1;
while ii < length(s)
    s([ii ii+1]) = s([ii+1 ii]);
    ii = ii + 2;
end
typecast(uint64(hex2dec(s)), 'double');

Actually, hex2num should work without conversion to decimal now since I am not using swapbytes and don't need to typecast:

hex2num(s)

Answer 2

[Walter Roberson]

Do not use hex2dec() for this purpose. Use sscanf()

sscanf('b0120c0a7799ba3e', '%lx')

 ans =
  uint64
   12687216339351878206

%lx is an unsigned 64 bit format.

[Tony Tse]

Thanks! This worked perfectly.

typecast(swapbytes(sscanf('b0120c0a7799ba3e', '%lx')), 'double')

[Walter Roberson]

If you are working with something that is an IEEE double, then

swapbytes(hex2num('b0120c0a7799ba3e'))