Secrets of the Glasgow Haskell Compiler Inliner (1999) [pdf](microsoft.com)
microsoft.com
Secrets of the Glasgow Haskell Compiler Inliner (1999) [pdf]
https://www.microsoft.com/en-us/research/wp-content/uploads/2002/07/inline.pdf
10 comments
Also worth checking out is this https://www.youtube.com/watch?v=uR_VzYxvbxg conference talk by Simon Peyton Jones discussing the basic constructs of GHC core and how they affect inlining and optimisation
SPJ must be the bounciest computing-guy-over-30 I've ever encountered. Always a refreshing pleasure to watch any of his performances =)
His enthusiasm is truly contagious, somewhat reminiscent of how excited Feynman could make you feel about science.
Yes, i was lucky enough to be in the audience at St Andrews in about 1986 when he came to talk to us about parallel combinator reduction, G-machines and Alice.
Amazing enthusiasm and application. Amazing guy.
Amazing enthusiasm and application. Amazing guy.
He was my professor at university, came in one day with shirt, tie and sandals. Took in a deep breath, and shouted it's a beautiful sunny day today.
Watch something with Clifford Stoll in it. His enthusiasm is also incredible.
I still think it is a pity that GHC isn't much better documented. It could be a huge education and research platform, with a much wider reach than the small group currently versed in GHC internals.
It is a huge education and research platform. There are many many papers every year developing new ideas in GHC. Look at the massive cottage industry in Haskell language extensions implemented in GHC.
> A major issue for any compiler, especially for one that inlines heavily, is name capture
What? If you inline just one block of code in a million lines of code (i.e. you "inline with the dainty lightness of a frolicking fairy"), and that situation is wrong due to a name capture, you have a bug, which could stop the show.
What? If you inline just one block of code in a million lines of code (i.e. you "inline with the dainty lightness of a frolicking fairy"), and that situation is wrong due to a name capture, you have a bug, which could stop the show.
In section 3.1 they say it is a performance issue.