Fixes issue 99. Replace aligned loads with unaligned.

[lemire]

Based on #100 but arguably more complete and better?

Credit to @pps83 for the hard work.

Fixes #99

[lemire]

@pps83 Would you review?

[pps83]

Should also update commit title, something like "Replace aligned loads with unaligned"

and make sure you squash commits when merging.

[pps83]

unsigned. Even though values is uint32_t, it's quite strange to use uint32_t as loop counter.
same for a change bellow

[cbsmith]

Seems necessary to ensure i < values is both efficient and correct for values approaching std::numeric_limits<uint32_t>::max() right?

[cbsmith]

Maybe to avoid any concern, you could use the standard algorithms library to fill the vector in a declarative fashion.

Suggested change

	for (uint32_t i = 0; i < values; ++i) {
	std::generate_n(std::begin(v), values, std::bind(dist, e2));

[pps83]

that would be outside of the scope of the PR.

[lemire]

it's quite strange to use uint32_t as loop counter.

Why? Many compilers object to signed/unsigned comparisons. I don't want to change the type of values. The alternative for the counters are size_t, unsigned int, ... none of them functionally better. Well, size_t (an unsigned type) is arguably better but I am not certain it matters.

std::generate_n(std::begin(v), values, std::bind(dist, e2));

@cbsmith Good idea, but it is out of scope here. I only changed that bit to silence a warning. If you'd like to modernize the C++, the please consider issue a distinct PR. There is much that could be done...

[pps83]

also, it looks it's based off from some old version, not from the last master.

[pps83]

I checked the PR, many of the issues I saw got resolved. However, I still see one place where a byte or two end up unmodified the output. Perhaps, it's ok to merge this one, and take a look at that issue separately.

[lemire]

Merging.

@cbsmith : I invite you to prepare a distinct PR if you'd like.