3 Replies Latest reply on Apr 14, 2014 9:48 AM by kknox

    Where's `long double`?

    bob

      You (AMD) manufacture a FP unit with 128 bit precision, but offer NO IEEE 754 basic math library routines for that precision, e.g. "amd_cosl".

       

      Also, there is NO Win x32 library. Huh? 128 bit FP precision works just as well on x32 as x64.

        • Re: Where's `long double`?
          kknox

          I believe that you must mean 80-bit long double support, as I am not aware of any x86 hardware that does true 128-bit floating point operations.  For example, SSE registers are 128-bits wide, but they are packed either with 4 32-bit single precision numbers or 2 64-bit double precision numbers.  AVX makes this even wider.

           

          I assume that you have already looked at the AMD LibM library available for download here:

          http://developer.amd.com/tools-and-sdks/cpu-development/libm/

           

          The functions contained therein are what we have implemented and support.

           

          Kent