Blend packed double-precision (64-bit) floating-point elements from a and b using mask.