convertMask.cpp

__m128i convertMask( __m256d src )
{
	// Bit-cast into fp32 vector, the intrinsic compiles into no instructions
	const __m256 f32 = _mm256_castpd_ps( src );
	// Split into high/low halves; casting is free, vextractf128 is not.
	const __m128 low = _mm256_castps256_ps128( f32 );
	const __m128 high = _mm256_extractf128_ps( f32, 1 );
	// Combine 32-bit values into a single vector with correct order
	// _mm_shuffle_ps takes first 2 lanes from the first argument, last 2 lanes fro the second argument.
	const __m128 combined = _mm_shuffle_ps( low, high, _MM_SHUFFLE( 2, 0, 2, 0 ) );
	// Cast result into int vector
	return _mm_castps_si128( combined );
}