[X86] Use the new unaligned vector typedefs for the loadu/storeu intrinsics pointer arguments.

This matches what gcc does and what was suggested by rnk in PR20670.

llvm-svn: 353802

Author: topperc

clang/lib/Headers/avxintrin.h

@@ -3170,7 +3170,7 @@ _mm256_load_si256(__m256i const *__p)
 ///    A pointer to a 256-bit integer vector containing integer values.
 /// \returns A 256-bit integer vector containing the moved values.
 static __inline __m256i __DEFAULT_FN_ATTRS
-_mm256_loadu_si256(__m256i const *__p)
+_mm256_loadu_si256(__m256i_u const *__p)
 {
   struct __loadu_si256 {
     __m256i_u __v;
@@ -3305,7 +3305,7 @@ _mm256_store_si256(__m256i *__p, __m256i __a)
 /// \param __a
 ///    A 256-bit integer vector containing the values to be moved.
 static __inline void __DEFAULT_FN_ATTRS
-_mm256_storeu_si256(__m256i *__p, __m256i __a)
+_mm256_storeu_si256(__m256i_u *__p, __m256i __a)
 {
   struct __storeu_si256 {
     __m256i_u __v;
@@ -4838,7 +4838,7 @@ _mm256_loadu2_m128d(double const *__addr_hi, double const *__addr_lo)
 ///    address of the memory location does not have to be aligned.
 /// \returns A 256-bit integer vector containing the concatenated result.
 static __inline __m256i __DEFAULT_FN_ATTRS
-_mm256_loadu2_m128i(__m128i const *__addr_hi, __m128i const *__addr_lo)
+_mm256_loadu2_m128i(__m128i_u const *__addr_hi, __m128i_u const *__addr_lo)
 {
   __m256i __v256 = _mm256_castsi128_si256(_mm_loadu_si128(__addr_lo));
   return _mm256_insertf128_si256(__v256, _mm_loadu_si128(__addr_hi), 1);
@@ -4922,7 +4922,7 @@ _mm256_storeu2_m128d(double *__addr_hi, double *__addr_lo, __m256d __a)
 /// \param __a
 ///    A 256-bit integer vector.
 static __inline void __DEFAULT_FN_ATTRS
-_mm256_storeu2_m128i(__m128i *__addr_hi, __m128i *__addr_lo, __m256i __a)
+_mm256_storeu2_m128i(__m128i_u *__addr_hi, __m128i_u *__addr_lo, __m256i __a)
 {
   __m128i __v128;
 

clang/lib/Headers/emmintrin.h

@@ -3567,7 +3567,7 @@ _mm_load_si128(__m128i const *__p)
 ///    A pointer to a memory location containing integer values.
 /// \returns A 128-bit integer vector containing the moved values.
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_loadu_si128(__m128i const *__p)
+_mm_loadu_si128(__m128i_u const *__p)
 {
   struct __loadu_si128 {
     __m128i_u __v;
@@ -3588,7 +3588,7 @@ _mm_loadu_si128(__m128i const *__p)
 /// \returns A 128-bit vector of [2 x i64]. The lower order bits contain the
 ///    moved value. The higher order bits are cleared.
 static __inline__ __m128i __DEFAULT_FN_ATTRS
-_mm_loadl_epi64(__m128i const *__p)
+_mm_loadl_epi64(__m128i_u const *__p)
 {
   struct __mm_loadl_epi64_struct {
     long long __u;
@@ -4030,7 +4030,7 @@ _mm_store_si128(__m128i *__p, __m128i __b)
 /// \param __b
 ///    A 128-bit integer vector containing the values to be moved.
 static __inline__ void __DEFAULT_FN_ATTRS
-_mm_storeu_si128(__m128i *__p, __m128i __b)
+_mm_storeu_si128(__m128i_u *__p, __m128i __b)
 {
   struct __storeu_si128 {
     __m128i_u __v;
@@ -4142,7 +4142,7 @@ _mm_maskmoveu_si128(__m128i __d, __m128i __n, char *__p)
 ///    A 128-bit integer vector of [2 x i64]. The lower 64 bits contain the
 ///    value to be stored.
 static __inline__ void __DEFAULT_FN_ATTRS
-_mm_storel_epi64(__m128i *__p, __m128i __a)
+_mm_storel_epi64(__m128i_u *__p, __m128i __a)
 {
   struct __mm_storel_epi64_struct {
     long long __u;