On Mon, May 21, 2007 at 11:58:35PM -0400, Tom Lane wrote:
> The intent of the FP binary I/O code we have is that for platforms
> supporting IEEE-compliant FP formats, the on-the-wire representation
> should be uniformly big-endian, same as is true for integers. So
> I would concur with a patch that ensures that this is what happens
> on the different ARM variants ... though I'll still be interested
> to see how you make that happen given the rather poor visibility
> into which model and endianness we are running on.
Well, I have an idea how you might do this: figure out the ordering of
the float at runtime. You can easily construct a float with any given
bit pattern. You can then examine the bytes to determine the order and
build a mapping table to reorder them. The program below creates a
float with the bit pattern 01020304. You can then examine the bits of
the float to determine the rearranging needed. You could do the same
for 64-bit floats.
This is obviously only needed for systems where the order can't be
determined at compile time.
ldexp is in SVr4, 4.3BSD and C89.
#include <stdio.h>
#include <math.h>
int main()
{ float f = ldexp(1.0,-119) + ldexp(1.0,-125) + ldexp(1.0,-126) + ldexp(1.0,-133) + ldexp(1.0,-142); unsigned char
*a= (char*)&f; printf("Float: %g, char: %02x%02X%02X%02X\n", f, a[0], a[1], a[2], a[3]);
}
Have a nice day,
--
Martijn van Oosterhout <kleptog@svana.org> http://svana.org/kleptog/
> From each according to his ability. To each according to his ability to litigate.