converting 64-bit fixed-point to float

[John Machin]

On Jul 22, 10:34 am, johnmfis... at comcast.net (John Fisher) wrote:
> attn.steven.... at gmail.com <attn.steven.... at gmail.com> wrote:
> > On Jul 20, 5:59 pm, johnmfis... at comcast.net (John Fisher) wrote:
> > > Hi Group,
>
> > > troubles with converting signed 32.32, little-endian, 2's complement
> > > back to floating point. I have been trying to brew it myself. I am
> > > running Python 2.5 on a Mac. Here is the C-code I have been trying to
> > > leverage:
>
> > >  double FPuint8ArrayToFPDouble(uint8 *buffer, int startIndex)
> > > {
> > >   uint32 resultDec = 0;
> > >   uint32 resultWh = 0;
> > >   int i;
>
> > >   for(i = 0; i < 4; i++)
> > >   {
> > >     resultDec += (uint32)buffer[startIndex + i] * pow(2, (i*8));
> > >     resultWh += (uint32)buffer[startIndex + i + 4] * pow(2, (i*8));
> > >   }
>
> > >   return ( (double)((int)resultWh) + (double)(resultDec)/4294967296.0 );
>
> > > }
>
> > There are a few problem spots in that C code.  I tend to
> > think that it "works" because you're on a system that has
> > 4-byte int and CHAR_BIT == 8.  When the most-significant-bit (MSB)
> > of resultWh is 1, then casting to int makes that a negative
> > value (i.e., MSB == the sign bit).
>
> > I presume that somewhere you include <stdint.h> (from C99)
> > and that uint32 is really uint32_t, etc.  For that to be
> > portable, you should probably cast to int32_t?
>
> > #include <stdio.h>
> > #include <stdlib.h>
> > #include <stdint.h>
>
> > double arr2dbl (uint8_t *buffer, int startIndex)
> > {
> >     uint32_t decimal = 0;
> >     uint32_t whole = 0;
> >     size_t i;
> >     for (i = 0; i < 4; ++i)
> >     {
> >         decimal += (buffer[startIndex + i] << (i*8));
> >         whole += (buffer[startIndex + i + 4] << (i*8));
> >     }
> >     return (int32_t)whole + (decimal/(UINT32_MAX+1.0));
> > }
>
> > int main (void)
> > {
> >     uint8_t arr[7][8] = {
> >         {0, 0, 0, 0, 0xff, 0xff, 0xff, 0xff},
> >         {0, 0, 0, 0, 0, 0, 0, 0},
> >         {51, 51, 51, 51, 0, 0, 0, 0},
> >         {0, 0, 0, 0, 1, 0, 0, 0},
> >         {0, 0, 0, 0, 2, 0, 0, 0},
> >         {102, 102, 102, 38, 42, 1, 0, 0 },
> >         {205, 204, 204, 204, 0xff, 0xff, 0xff, 0xff}};
> >     size_t i;
> >     double result;
> >     for (i = 0; i < sizeof arr/sizeof arr[0]; ++i)
> >     {
> >         result = arr2dbl(arr[i], 0);
> >         printf("%f\n", result);
> >     }
> >     return 0;
> > }
>
> > > Here is my version in Python, with some test code built in:
>
> > > from ctypes import *
>
> > > def conv64(input):
> > >     input1=[0]*8
> > >     input1[0]=c_ushort(input[0])
> > >     input1[1]=c_ushort(input[1])
> > >     input1[2]=c_ushort(input[2])
> > >     input1[3]=c_ushort(input[3])
> > >     input1[4]=c_ushort(input[4])
> > >     input1[5]=c_ushort(input[5])
> > >     input1[6]=c_ushort(input[6])
> > >     input1[7]=c_ushort(input[7])
> > >     #print input1[0].value,
> > > input1[1].value,input1[2].value,input1[3].value
> > >     #print
> > > input1[4].value,input1[5].value,input1[6].value,input1[7].value
> > >     #print
> > >     resultDec=c_ulong(0)
> > >     resultWh=c_ulong(0)
> > >     for i in range(4):
> > >         dec_c=c_ulong(input1[i].value)
> > >         Wh_c=c_ulong(input1[i+4].value)
> > >         resultDec.value=resultDec.value+dec_c.value*2**(i*8)
> > >         resultWh.value=resultWh.value+Wh_c.value*2**(i*8)
> > >    conval=float(int(resultWh.value))+float(resultDec.value)/4294967296.0
> > >     #print conval
> > >     return conval
> > > #tabs got messed up bringing this into MacSoup
>
> > (snipped)
>
> > > Finally, here is the output I get from my code:
>
> > > output should be -1 is 4294967296.0
> > > output should be 0 is 0.0
> > > output should be 0.2 is 0.199999999953
> > > output should be 1 is 1.0
> > > output should be 2 is 2.0
> > > output should be 298.15 is 298.15
> > > output should be -0.2 is 4294967295.8
>
> > > Thanks for any light you can shed on my ignorance.
>
> > > wave_man
>
> > This is my translation:
>
> > from ctypes import *
>
> > def conv64(input):
> >     input1 = [c_uint8(item) for item in input]
> >     dec = c_uint32(0)
> >     whl = c_uint32(0)
> >     for i in xrange(4):
> >         dec.value += (input1[i].value << (i*8))
> >         whl.value += (input1[i+4].value << (i*8))
> >     cast_whl_to_int = c_int32(whl.value)
> >     return float(cast_whl_to_int.value + dec.value/4294967296.0)
>
> > for arr in [[0, 0, 0, 0, 0xff, 0xff, 0xff, 0xff],
> >         [0, 0, 0, 0, 0, 0, 0, 0],
> >         [51, 51, 51, 51, 0, 0, 0, 0],
> >         [0, 0, 0, 0, 1, 0, 0, 0],
> >         [0, 0, 0, 0, 2, 0, 0, 0],
> >         [102, 102, 102, 38, 42, 1, 0, 0],
> >         [205,204,204,204,255,255,255,255]]:
> >     print "%f" % conv64(arr)
>
> > However, I've not looked deeply into ctypes so I
> > don't know if c_int32 is really C's int, or int32_t, or ???
>
> > --
> > Hope this helps,
> > Steven
>
> Actually this was very helpful, thanks.
>

Holy code bloat, Batman! The following appears to give the same
results as Steven's code ...

def conv64jm(input):
    # Assuming input is an array of unsigned 8-bit ints.
    # If it were a string, the first step could be omitted.
    input1 = struct.pack('8B', *input) # alternative: map(chr, input)
    return struct.unpack("<q", input1)[0] / 4294967296.0