[X86] Remove unnecessary 3DNow declarations from Intrin.h

[RKSimon]

Summary

Follow up to D12272

This adds the missing 3dnow intrinsics _m_to_float / _m_from_float and removes their unnecesary declarations from Intrin.h. These are really just wrappers to existing MMX instructions (they are moved to/from the MMX register as i32 as 3dnow doesn't provide float specific instructions).

Final part to resolve PR23648

[RKSimon]

Abandoning this - unless we have a way to know when to automatically insert emms/femms then its not safe to do this.

lib/Headers/Intrin.h

 
 /* Define the default attributes for the functions in this file. */
 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__))
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-#if defined(__MMX__)
-/* And the random ones that aren't in those files. */
-__m64 _m_from_float(float);
-float _m_to_float(__m64);
-#endif
-
 /* Other assorted instruction intrinsics. */
 void __addfsbyte(unsigned long, unsigned char);
 void __addfsdword(unsigned long, unsigned long);
 void __addfsword(unsigned long, unsigned short);
 void __code_seg(const char *);
 static __inline__
 void __cpuid(int[4], int);
 static __inline__

lib/Headers/mm3dnow.h

 /* Define the default attributes for the functions in this file. */
 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("3dnow")))
 
 static __inline__ void __DEFAULT_FN_ATTRS
 _m_femms() {
   __builtin_ia32_femms();
 }
 
+static __inline__ float __DEFAULT_FN_ATTRS
+_m_to_float(__m64 __m) {
+  union { int __i; float __f; } __t;
+  __t.__i = __builtin_ia32_vec_ext_v2si((__v2si)__m, 0);
+  return __t.__f;
+}
+
+static __inline__ __m64 __DEFAULT_FN_ATTRS
+_m_from_float(float __f) {
+  union { int __i; float __f; } __t;
+  __t.__f = __f;
+  return (__m64)__builtin_ia32_vec_init_v2si(__t.__i, 0);
+}
+
 static __inline__ __m64 __DEFAULT_FN_ATTRS
 _m_pavgusb(__m64 __m1, __m64 __m2) {
   return (__m64)__builtin_ia32_pavgusb((__v8qi)__m1, (__v8qi)__m2);
 }
 
 static __inline__ __m64 __DEFAULT_FN_ATTRS
 _m_pf2id(__m64 __m) {
   return (__m64)__builtin_ia32_pf2id((__v2sf)__m);

test/CodeGen/3dnow-builtins.c

 // RUN: %clang_cc1 %s -O3 -triple=x86_64-unknown-unknown -target-feature +3dnow -emit-llvm -o - | FileCheck %s
 
 // Don't include mm_malloc.h, it's system specific.
 #define __MM_MALLOC_H
 
 #include <x86intrin.h>
 
+__m64 test_m_from_float(float f) {
+  // CHECK-LABEL: define i64 @test_m_from_float
+  // CHECK: insertelement <2 x i32> 
+  return _m_from_float(f);
+}
+
+float test_m_to_float(__m64 m) {
+  // CHECK-LABEL: define float @test_m_to_float
+  // CHECK: extractelement <2 x i32>
+  return _m_to_float(m);
+}
+
 __m64 test_m_pavgusb(__m64 m1, __m64 m2) {
   // CHECK-LABEL: define i64 @test_m_pavgusb
   // CHECK: @llvm.x86.3dnow.pavgusb
   return _m_pavgusb(m1, m2);
 }
 
 __m64 test_m_pf2id(__m64 m) {
   // CHECK-LABEL: define i64 @test_m_pf2id