[libc++] hash<long double>: adjust for x86-64 ILP32

[hvdijk]

Summary

x86-64 ILP32 mode (x32) uses 32-bit size_t, so share the code with ix86 to zero out padding bits, not with x86-64 LP64 mode.

[ldionne]

That looks correct to me with what I just read about x32 (I wasn't familiar with it). I can't imagine this is the only place that needs fixing for libc++ to work in x32?

If you need me to commit this, please provide Author Name <email@domain>.

[hvdijk]

In D91349#2408723, @ldionne wrote:

That looks correct to me with what I just read about x32 (I wasn't familiar with it). I can't imagine this is the only place that needs fixing for libc++ to work in x32?

It's the only place in libc++ that needs fixing in order to get the test suite to pass. Beyond that, I am not sure, as there are some issues on the LLVM and clang side as well that prevent more extensive testing. I am trying to address those as well, and may be able to try a system rebuild with clang+libc++ after that to hopefully catch any remaining issues.

If you need me to commit this, please provide Author Name <email@domain>.

Thanks! Could you commit this under 'Harald van Dijk <[e-mail]>'?

libcxx/include/utility

struct _LIBCPP_TEMPLATE_VIS hash<long double>
     : public __scalar_hash<long double>
 {
     _LIBCPP_INLINE_VISIBILITY
     size_t operator()(long double __v) const _NOEXCEPT
     {
         // -0.0 and 0.0 should return same hash
         if (__v == 0.0L)
             return 0;
-#if defined(__i386__)
+#if defined(__i386__) || (defined(__x86_64__) && defined(__ILP32__))
         // Zero out padding bits
         union
         {
             long double __t;
             struct
             {
                 size_t __a;
                 size_t __b;