SIMDPP_DISABLE_DEPRECATED_CONVERSION_OPERATOR_TO_NATIVE_TYPES
Defining SIMDPP_DISABLE_DEPRECATED_CONVERSION_OPERATOR_TO_NATIVE_TYPES to nonzero value before inclusion of libsimdpp headers will disable deprecated implicit conversion operator between libsimdpp vector types and native SIMD types provided by the compiler.
The operators has been deprecated because Clang supports native vector extension which interferes with simdpp types and allows wrong code being allowed to be compiled.
The native vector extension defines operators between vector types. SSE types such as __m128 and __m128i are implemented on top of this extension, which causes code like this being possible:
__m128i a, b; a = a + b;
Having an implicit conversion operator to vector types allows types such as uint8<16> and uint16<8> to be implicitly converted to __m128i. This leads to code like this being accepted on clang.
uint8<16> a; uint16<8> b; a = a + b;
Here, both a and b are implicitly converted to __m128i values and they are added using an operator provided by the clang vector extension. Unexpectedly, the result is paddq instruction (64-bit integer addition).
Because of this, the implicit native vector type conversion operators are deprecated and a native() method is provided as a replacement. This change only affects code that interacts with native intrinsics. Altivec/VSX and MSA are affected only slightly, because intrinsics of those instruction sets never accepted implicit conversions from libsimdpp types.