disambiguate vector type to avoid errors depending on lax conversion …

[JoachimSchurig]

…settings

Typecast explicitly to __m128d * to avoid compiler errors depending on platform and compiler, and the settings of -flax-vector-conversions or its different defaults

The test suite will fail to compile on AppleClang/aarch64 with at least AppleClang 14 and 15 if you add the compiler option -flax-vector-conversions to the makefile. There are projects that include sse2neon.h that have this option set.

Also, the option has differing defaults depending on compiler make and version, therefore it makes sense to not depend on it for the code to be compiled.

See discussions here: llvm and here gcc

[jserv]

Thank @JoachimSchurig for contributing. Can you check #571 as well?

[JoachimSchurig]

Thank you for this wonderful project!

As for #571, all tests compile and pass on AppleClang when compiled with -flax-vector-conversions or -flax-vector-conversions=none or -flax-vector-conversions=integer or -fno-lax-vector-conversions (or without any of those).
I think the difference to the above is that they operate on value types and not on pointers, and not through the __builtin_nontemporal_store builtin. That may trigger the error at _mm_stream_pd.