[Headers] Improve codegen for _mm_slli_epi32.

[Bigcheese]

Summary

This generates vector llvm ir for _mm_slli_epi32 to allow
further optimization.

[Bigcheese]

Actually, this is wrong. _mm_slli_epi32 requires that if count is > than 31, the result is 0. The << operator gives it undefined behavior.

[dblaikie]

This isn't really in my area of expertise/ownership, so I'll leave it to someone else to review.

lib/Headers/emmintrin.h

 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_slli_epi32(__m128i __a, int __count)
 {
-  return (__m128i)__builtin_ia32_pslldi128((__v4si)__a, __count);
+  return (__m128i)((__v4si)__a << (__v4si){__count, __count, __count, __count});
 }
 
 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))