[Runtime] Adjust argument types to match the LLVM15 ABI.

It appears that LLVM 15 changed the ABI for _Float16 on x86-64 such that
values are now passed in `xmm0` instead of using integer registers.

Also enable this code for Linux.

rdar://104134160

Author: al45tair

stdlib/public/runtime/Float16Support.cpp

@@ -31,7 +31,7 @@
 // Android NDK <r21 do not provide `__aeabi_d2h` in the compiler runtime,
 // provide shims in that case.
 #if (defined(__ANDROID__) && defined(__ARM_ARCH_7A__) && defined(__ARM_EABI__)) || \
-  ((defined(__i386__) || defined(__i686__) || defined(__x86_64__)) && !defined(__APPLE__) && !defined(__linux__))
+  ((defined(__i386__) || defined(__i686__) || defined(__x86_64__)) && !defined(__APPLE__))
 
 #include "swift/shims/Visibility.h"
 
@@ -49,6 +49,20 @@ static float fromEncoding(unsigned int e) {
   return f;
 }
 
+static unsigned short toEncoding(_Float16 f) {
+  unsigned short s;
+  static_assert(sizeof s == sizeof f, "_Float16 and short must have the same size");
+  __builtin_memcpy(&s, &f, sizeof f);
+  return s;
+}
+
+static _Float16 fromEncoding(unsigned short s) {
+  _Float16 f;
+  static_assert(sizeof s == sizeof f, "_Float16 and short must have the same size");
+  __builtin_memcpy(&f, &s, sizeof f);
+  return f;
+}
+
 #if defined(__x86_64__) && defined(__F16C__)
 
 // If we're compiling the runtime for a target that has the conversion
@@ -126,7 +140,7 @@ SWIFT_RUNTIME_EXPORT unsigned short __gnu_f2h_ieee(float f) {
 //
 // Note that F16C doesn't provide this operation, so we still need a software
 // implementation on those cores.
-SWIFT_RUNTIME_EXPORT unsigned short __truncdfhf2(double d) {
+SWIFT_RUNTIME_EXPORT _Float16 __truncdfhf2(double d) {
   // You can't just do (half)(float)x, because that makes the result
   // susceptible to double-rounding. Instead we need to make the first
   // rounding use round-to-odd, but that doesn't exist on x86, so we have
@@ -148,7 +162,7 @@ SWIFT_RUNTIME_EXPORT unsigned short __truncdfhf2(double d) {
     if (fabs < dabs) e |= 1;
     f = fromEncoding(e);
   }
-  return __gnu_f2h_ieee(f);
+  return fromEncoding(__gnu_f2h_ieee(f));
 }
 
 // Convert from Float16 to long double.
@@ -160,18 +174,18 @@ SWIFT_RUNTIME_EXPORT unsigned short __truncdfhf2(double d) {
 //
 // There's no risk of rounding problems from the double conversion, because
 // we're extending.
-SWIFT_RUNTIME_EXPORT long double __extendhfxf2(short h) {
-  return __gnu_h2f_ieee(h);
+SWIFT_RUNTIME_EXPORT long double __extendhfxf2(_Float16 h) {
+  return __gnu_h2f_ieee(toEncoding(h));
 }
 
 // This is just an alternative name for __gnu_h2f_ieee
-SWIFT_RUNTIME_EXPORT float __extendhfsf2(unsigned short h) {
-  return __gnu_h2f_ieee(h);
+SWIFT_RUNTIME_EXPORT float __extendhfsf2(_Float16 h) {
+  return __gnu_h2f_ieee(toEncoding(h));
 }
 
 // Same again but for __gnu_f2h_ieee
-SWIFT_RUNTIME_EXPORT unsigned short __truncsfhf2(float f) {
-  return __gnu_f2h_ieee(f);
+SWIFT_RUNTIME_EXPORT _Float16 __truncsfhf2(float f) {
+  return fromEncoding(__gnu_f2h_ieee(f));
 }
 
 #if defined(__ARM_EABI__)