new codegen patches

[Simon Peyton-Jones]

| > Agreed. I compared the new codegen with the old codegen (on nofib) before
| > I left Microsoft, and I recall a minor uptick in compile times. I haven't
| > looked into the causes yet.
| >
|
| Are we expecting a /decrease/ in runtime now though? :-)

Probably not yet.  All of this work has positioned us for being able to generate better code in three ways

- we can generate better C-- code from the STG code (eg Trac #2289)
- we have a much better chance of optimising the C-- code (eg tail recursion -> loop)
- more ambitiously, we're set up to start to integrate the native code
        generator into the same framework

However none of these things have actually been done yet.  What we have is a much more robust and modular foundation.

Simon