Merge Request: LLVM Code Generator for GHC

[Malcolm Wallace]

> The most interesting observation seems to be that the results are  
> largely inconsistent (look at 'atom', for example).  Even wheel- 
> sieve1, where our numbers go in the same direction, varies between  
> +10% and +22.7%.  My numbers are from an older HEAD (as I had nofib  
> handy there), but I'm still wondering where this difference comes  
> from.  Is it just that different processors are more or less  
> sensible to TNTC?

There is a video of a Google Tech Talk on the topic of how  
inconsistent and confusing benchmark results arise.

     http://www.youtube.com/watch?v=DKVRkfXrBpg

Well worth investing 50 minutes of your time, if you ever do any kind  
of performance benchmarking.  Warning: you may be shocked at how easy  
it is for unforeseen and uncontrolled factors to obfuscate your  
measurements, to the point of indicating a gain instead of a loss, and  
vice versa.  Something as random as library link-order can flip the  
sign...

Regards,
     Malcolm