The standard library now provides a variety of rdrand() functions that call out to RDRAND, when available. Issue #263.
So far, only the use of the float/half conversion instructions distinguishes this from the "avx1" target. Partial work on issue #263.