Type casting in sse2neon.h incorrect

[StefanBruens]

When compiling on Linux/aarch64, GCC throws many errors, e.g.:

  17s] [  3%] Building CXX object common/simd/CMakeFiles/simd.dir/sse.cpp.o
[   17s] cd /home/abuild/rpmbuild/BUILD/embree-3.13.5/build/common/simd && /usr/bin/c++ -DEMBREE_TARGET_SSE2 -DTASKING_INTERNAL  -fsigned-char -Wall -Wformat -Wformat-security -Wno-class-memaccess -fno-strict-overflow -fno-delete-null-pointer-checks -fwrapv -fsigned-char -fPIE -fPIC -std=c++11 -fvisibility=hidden -fvisibility-inlines-hidden -fno-strict-aliasing -fno-tree-vectorize -D_FORTIFY_SOURCE=2  -fsigned-char -g -DNDEBUG -O3  -D__SSE__ -D__SSE2__ -MD -MT common/simd/CMakeFiles/simd.dir/sse.cpp.o -MF CMakeFiles/simd.dir/sse.cpp.o.d -o CMakeFiles/simd.dir/sse.cpp.o -c /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/sse.cpp
[   17s] In file included from /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/../sys/../simd/arm/emulation.h:12,
[   17s]                  from /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/../sys/intrinsics.h:13,
[   17s]                  from /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/sse.h:7,
[   17s]                  from /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/sse.cpp:4:
[   17s] /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/../sys/../simd/arm/sse2neon.h: In function '__m128i _mm_hadds_epi16(__m128i, __m128i)':
[   17s] /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/../sys/../simd/arm/sse2neon.h:6749:55: note: use '-flax-vector-conversions' to permit conversions between vectors with differing element types or numbers of subparts
[   17s]  6749 |         vqaddq_s16(vuzp1q_s16(a, b), vuzp2q_s16(a, b)));
[   17s]       |                                                       ^
[   17s] /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/../sys/../simd/arm/sse2neon.h:6748:33: error: cannot convert 'int64x2_t' to '__m128i' {aka 'int32x4_t'} in return
[   17s]  6748 |     return vreinterpretq_s64_s16(
[   17s]       |            ~~~~~~~~~~~~~~~~~~~~~^
[   17s]       |                                 |
[   17s]       |                                 int64x2_t
[   17s]  6749 |         vqaddq_s16(vuzp1q_s16(a, b), vuzp2q_s16(a, b)));
[   17s]       |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[   17s] /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/../sys/../simd/arm/sse2neon.h: In function '__m128i _mm_hsubs_epi16(__m128i, __m128i)':
[   17s] /home/abuild/rpmbuild/BUILD/embree-3.13.5/common/simd/../sys/../simd/arm/sse2neon.h:6846:33: error: cannot convert 'int64x2_t' to '__m128i' {aka 'int32x4_t'} in return
[   17s]  6846 |     return vreinterpretq_s64_s16(
[   17s]       |            ~~~~~~~~~~~~~~~~~~~~~^
[   17s]       |                                 |
[   17s]       |                                 int64x2_t
[   17s]  6847 |         vqsubq_s16(vuzp1q_s16(a, b), vuzp2q_s16(a, b)));
[   17s]       |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

For Apple Silicon, this is sidestepped with -flax-vector-conversions.

https://github.com/embree/embree/blob/489b746c0d5010e0da10345e9dc96768bec9a037/common/cmake/gnu.cmake#L86-L89

[svenwoop]

Could you please try the devel3 branch? It should fix this.