Consider simplifying the use of certain WINDOWS defines

[tannergooding]

The majority of the WINDOWS and WIN64 usages look like they could be simplified to a different switch which indicated whether or not error reporting should be POSIX compliant.

That is, the code for a scenario such as (https://github.com/amd/aocl-libm-ose/blob/master/src/ref/acosf.c#L75-L85):

#ifdef WINDOWS
     return  __amd_handle_errorf("acosf", __amd_acos, ux|0x00400000, _DOMAIN, AMD_F_NONE, EDOM, x, 0.0F, 1);
#else
      //return x + x; /* With invalid if it's a signalling NaN */
      if (ux & QNAN_MASK_32)
     return  __amd_handle_errorf("acosf", __amd_acos, ux|0x00400000, _DOMAIN, AMD_F_NONE, EDOM, x, 0.0F, 1);
      else
     return  __amd_handle_errorf("acosf", __amd_acos, ux|0x00400000, _DOMAIN, AMD_F_INVALID, EDOM, x, 0.0F, 1);
#endif

Isn't a Windows vs not difference, it's a "POSIX" vs not difference. There may be scenarios on either platform where one or the other is preferred.

There are a few other ifdefs that look to decide if files such as intrin.h should be imported, and these could be simplified to just WINDOWS (rather than WIN64 or WINDOWS || WIN64).

[tannergooding]

Happy to provide a fix here, if one is welcome.