[Bug target/67291] error: 'asm' operand has impossible constraints when compiling gromacs 5.1 testsuite on PPC64 and PPC64LE with VSX SIMD

["amodra at gmail dot com" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67291

Alan Modra <amodra at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amodra at gmail dot com

--- Comment #2 from Alan Modra <amodra at gmail dot com> ---
I think this is a user error.

The "ww" constraint is for the single precision vsx insns, like xsaddsp.  This
constraint is only enabled for vsx regs when compiling with -mcpu=power8 (or
when you compiler defaults to -mcpu=power8), or when explicitly enabled with
-mupper-regs-sf.  So if your compiler defaults to -mcpu=power7, you'll get the
"impossible constraints" message.

Reduced testcase, which compiles OK with -mcpu=power8, but fails with
-mcpu=power7.

__vector double
gmx_simd_cvt_i2d_ibm_vsx(__vector signed int ix)
{
  const __vector unsigned char perm = {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15,
8, 9, 10, 11};
  __vector double x;
  ix = __builtin_vec_perm(ix, ix, perm);
  __asm__ ("xvcvsxwdp %x0,%x1" : "=ww" (x) : "ww" (ix));
  return x;
}

I believe the asm should be using "=wd" and "wa" constraints.  Also, that
vec_perm is a bit ugly.  Putting the words of ix in the right place would
better be done with xxsldwi.