I'm not entirely sure what optimisations are made when supplying the -O flag to rdmd, but this may be related to an earlier issue I found for similar code here:
https://issues.dlang.org/show_bug.cgi?id=16027

The code is:
void main()
{
	auto seed = 128;
	auto rand = Random(seed);
	double[] values;

	values = generateValues(rand, 10);

	writeln(values);
}

double[] generateValues(ref Random rand, int l)
{
	auto values = new double[](l);
	foreach (ref val; values)
	{
		auto value = 1.0;
		if (uniform(0, 2, rand))
		{
			value = value * -1;
		}
		val = value;
	}

	return values;
}

Which returns different values depending on whether -O is passed:
>$rdmd testing_optimiser.d
>[1, -1, -1, 1, -1, 1, 1, 1, -1, -1]
>$rdmd -O testing_optimiser.d
>[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]

Any idea what causes this to occur when optimising? I wanted to try and speed up a simulation I'm running but it just produces too many unexpected consequences.

Could it be that the code is optimised to the same as that in the original issue and so the current compiler still produces the incorrect result? Obviously the original issue has since been fixed but I won't be able to test this until the next version of DMD is released.

how this works?

double[] generateValues(ref Random rand, int l)
{
	auto values = new double[](l);
	foreach (ref val; values)
	{
		auto value = 1;
		if (uniform(0, 2, rand))
		{
			value = value * -1;
		}
		val = value;
	}

	return values;
}

On 05/19/2016 05:09 PM, Michael wrote:
> Any idea what causes this to occur when optimising? I wanted to try and
> speed up a simulation I'm running but it just produces too many
> unexpected consequences.

I suspect that you're seeing issue 16027 [1], a bad bug in 2.071.0. The upcoming 2.071.1 point release should have it fixed.


[1] https://issues.dlang.org/show_bug.cgi?id=16027

On Thursday, 19 May 2016 at 15:49:17 UTC, ag0aep6g wrote:
> On 05/19/2016 05:09 PM, Michael wrote:
>> Any idea what causes this to occur when optimising? I wanted to try and
>> speed up a simulation I'm running but it just produces too many
>> unexpected consequences.
>
> I suspect that you're seeing issue 16027 [1], a bad bug in 2.071.0. The upcoming 2.071.1 point release should have it fixed.
>
>
> [1] https://issues.dlang.org/show_bug.cgi?id=16027

Yeah that's what I was thinking but I don't know much about the optimisation process so I wanted to mention it anyway.

On Thursday, 19 May 2016 at 17:07:27 UTC, Michael wrote:
> On Thursday, 19 May 2016 at 15:49:17 UTC, ag0aep6g wrote:
[...]
>> [1] https://issues.dlang.org/show_bug.cgi?id=16027
>
> Yeah that's what I was thinking but I don't know much about the optimisation process so I wanted to mention it anyway.

And now I see that you linked there yourself in the OP. No idea how I missed that.