https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91342
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |hjl.tools at gmail dot com,
| |jakub at gcc dot gnu.org
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Intel intrinsics are in constant flux. E.g. the manual
(325462-sdm-vol-1-2abcd-3abcd.pdf) documents them as taking
Intel C/C++ Compiler Intrinsic Equivalent
VMOVNTDQA __m512i _mm512_stream_load_si512(void * p);
MOVNTDQA __m128i _mm_stream_load_si128 (__m128i *p);
VMOVNTDQA __m256i _mm_stream_load_si256 (__m256i *p);
VMOVNTDQ void _mm512_stream_si512(void * p, __m512i a);
VMOVNTDQ void _mm256_stream_si256 (__m256i * p, __m256i a);
MOVNTDQ void _mm_stream_si128 (__m128i * p, __m128i a);
VMOVNTPD void _mm512_stream_pd(double * p, __m512d a);
VMOVNTPD void _mm256_stream_pd (double * p, __m256d a);
MOVNTPD void _mm_stream_pd (double * p, __m128d a);
VMOVNTPS void _mm512_stream_ps(float * p, __m512d a);
MOVNTPS void _mm_stream_ps (float * p, __m128d a);
VMOVNTPS void _mm256_stream_ps (float * p, __m256 a);
That is what GCC implements. CLANG does the same thing, except uses
_mm512_stream_load_si512 (void const *__P)
https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_stream_&expand=5670
documents them with void * instead.
