Simplify implementation of vectorized reads to avoid transmutes.

[briansmith]

This removies two more transmute-based conversions.

[briansmith]

Unfortunately, I don't currently have access to the systems I normally use for benchmarking as I am traveling, so I haven't measured the impact of this yet.

[tkaitchuck]

Previously this resulted in lower performance, although I have not tested with the latest version of the compiler. I'll look into it.