[Bug c++/80527] New: SSE4 Compiling issue

[milo.zhang at spreadtrum dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80527

            Bug ID: 80527
           Summary: SSE4 Compiling issue
           Product: gcc
           Version: 4.8.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: milo.zhang at spreadtrum dot com
  Target Milestone: ---

Created attachment 41271
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=41271&action=edit
workaround for this issue

We used SSE4 instruction _mm_cvtepi16_epi32.

But only with “#pragma GCC target (“sse4”)” and two includes (immintrin.h and
xmmintrin.h), the declaration of “_mm_cvtepi16_epi32” is not found.

#ifdef __linux__
#pragma GCC target("sse4")
#endif
#include <immintrin.h>
#include <xmmintrin.h>

if defined ( __linux__) && !defined (__clang__) && !defined (__SSE4_1)
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__,
__artificial__)) _mm_cvtepi16_epi32 (__m128i __X)
{
  return (__m128i) __builtin_ia32——pmovsxwd128 ((_v8hi)__X);
}

Thus, we implemented our own “_mm_cvtepi16_epi32” function below as the
workaround to avoid compiling error.

The reason is that in the “immintrin.h” file of gcc 4.8.3, without macros
__SSE_4_2__ or __SSE4_1__, “smmintrin.h” will not be included. And
“smmintrin.h” file declared “_mm_cvtepi16_epi32” function. You can see the
condition shown below.

Here is the implementation of "immintrin.h":

#if defined (__SSE4_2__) || defined (__SSE4_1__)
#include <smmintrin.h>
#endif