On 13/03/18 05:43 AM, Mason wrote:
On 12/03/2018 09:09, Dennis Clarke wrote:
/* 128-bit floating point has at most 38 digits
* of reasonable precision. We should be able to
* load in a constant value for pi using :
*
* 3.1415926535 8979323846 2643383279
* 5028841971 6939937510
*
* The result should be a 16-byte big endian machines
* representation in memory thus :
*
* 40 00 92 1f b5 44 42 d1 8469 89 8c c5 17 01 b8
*
* We can try to load in some constants and hope we get
* a fully reasonable in memory value.
*/
Oh, I didn't know about __float128.
Not exactly cross platform and no promise that these new IEEE754-2008
datatypes work at all. However with -std=iso9899:2011 you *may* get a
reasonable in memory value which you can then do not much with. Yet.
Also there isn't any hardware support for this anywhere, yet. Not even
the mighty NVidia CUDA cores will help. Yet.
Take a hard look at :
https://gcc.gnu.org/onlinedocs/gcc-7.3.0/gcc/x86-Options.html
Also
https://gcc.gnu.org/onlinedocs/gcc-7.3.0/gcc/x86-Built-in-Functions.html
Won't help much on other platforms that are riscy like ppc or sparc or
for that matter anywhere if you haven't the compiler support and libc
bits. Yet.
Dennis Clarke
