[x86] Don't use builtins for unaligned load/store

[x86] Don't use builtins for unaligned load/store

2016-08-31  Marc Glisse  <marc.glisse@inria.fr>

gcc/
	* config/i386/avx512fintrin.h (__m512_u, __m512i_u, __m512d_u):
	New types.
	(_mm512_loadu_pd, _mm512_storeu_pd, _mm512_loadu_ps,
	_mm512_storeu_ps, _mm512_loadu_si512, _mm512_storeu_si512):
	Replace builtin with vector extension.
	* config/i386/avxintrin.h (__m256_u, __m256i_u, __m256d_u):
	New types.
	(_mm256_loadu_pd, _mm256_storeu_pd, _mm256_loadu_ps,
	_mm256_storeu_ps, _mm256_loadu_si256, _mm256_storeu_si256):
	Replace builtin with vector extension.
	* config/i386/emmintrin.h (__m128i_u, __m128d_u): New types.
	(_mm_loadu_pd, _mm_storeu_pd, _mm_loadu_si128, _mm_storeu_si128):
	Replace builtin with vector extension.
	* config/i386/xmmintrin.h (__m128_u): New type.
	(_mm_loadu_ps, _mm_storeu_ps): Replace builtin with vector extension.
	(_mm_load_ps, _mm_store_ps): Simplify.

gcc/testsuite/
	* gcc.target/i386/pr59539-2.c: Adapt options.
	* gcc.target/i386/avx512f-vmovdqu32-1.c: Relax expected asm.

From-SVN: r239889