[LITMUS^RT] Binary heaps for Litmus

[Björn Brandenburg]

On Mar 30, 2012, at 8:01 PM, Glenn Elliott wrote:
> 
> Normally I agree that more code results in more bugs and maintenance problems.  If binheaps could only be used for the CPU priority ordering, I would rather stick to binomial heaps.  However, the binheaps have other applications and are being used extensively in work being done here at UNC.  Applying the binheaps to CPU priority ordering seemed like an easy way for binheaps to get into Litmus quickly.

We could also merge the binary heap code without immediately using it, too. I suppose later additions would end up using it where it makes sense.

By the way, one more thing. Having so much code in the header as to-be-inlined functions can lead to significant code bloat, which can actually decrease performance due to an increased instruction cache footprint. This is why the binomial heap implementation was switched from a header-only version to a normal header+object compilation unit. 

Thanks,
Björn