Make _m_prefetch use the mmx feature, and keep the _mm_prefetch macro

Author: rnk

clang/include/clang/Basic/BuiltinsX86.td

@@ -138,13 +138,15 @@ let Attributes = [Const, NoThrow, RequiredVectorWidth<256>], Features = "avx" in
   }
 }
 
+let Features = "mmx", Header = "mmintrin.h", Attributes = [NoThrow, Const] in {
+  def _m_prefetch : X86LibBuiltin<"void(void *)">;
+}
+
 // PRFCHW
 let Features = "prfchw", Header = "intrin.h", Attributes = [NoThrow, Const] in {
-  def _m_prefetch : X86LibBuiltin<"void(void *)">;
   def _m_prefetchw : X86LibBuiltin<"void(void volatile const *)">;
 }
 
-
 // Mechanically ported builtins from the original `.def` file.
 //
 // TODO: Build structured ways of synthesizing relevant groups and improve the

clang/lib/Headers/xmmintrin.h

@@ -2197,7 +2197,11 @@ _mm_storer_ps(float *__p, __m128 __a)
 #define _MM_HINT_T2  1
 #define _MM_HINT_NTA 0
 
-#if 0
+#ifndef _MSC_VER
+// If _MSC_VER is defined, we use the builtin variant of _mm_prefetch.
+// Otherwise, we provide this macro, which includes a cast, allowing the user
+// to pass a pointer of any time. The _mm_prefetch accepts char to match MSVC.
+
 /// Loads one cache line of data from the specified address to a location
 ///    closer to the processor.
 ///
@@ -2222,10 +2226,6 @@ _mm_storer_ps(float *__p, __m128 __a)
 ///    be generated. \n
 ///    _MM_HINT_T2: Move data using the T2 hint. The PREFETCHT2 instruction will
 ///    be generated.
-///
-/// _mm_prefetch is implemented as a "library builtin" directly in Clang,
-/// similar to how it is done in MSVC. Clang will warn if the user doesn't
-/// include xmmintrin.h or immintrin.h.
 #define _mm_prefetch(a, sel) (__builtin_prefetch((const void *)(a), \
                                                  ((sel) >> 2) & 1, (sel) & 0x3))
 #endif