fix problem with __m128i undefined on ARM

[franko]

There is a constant REG_SZ, see file modification, that is used conditionally only when SSE2 is defined but the variable REG_SZ is defined unconditionally.

This problem cause the compilation to fail on ARM64 architecture.

I added ifdef guards based on SSE2 like for the rest of the code. It compiles on macOS on ARM64 architecture.

[willysr]

This also fixed building issues in i586