On Mon, Feb 5, 2018 at 12:19 PM Simon Peyton Jones
Hi Michael
Thanks for pushing forward with Hoopl and other back-end things.
Did this patch elicit any performance gains? Or what brought it to your attention?
I noticed this some time ago and just now got around to try it out. I was hoping for some improvements, sadly the differences (if any) were too small compared to noise. But it seemed like a nice change on its own, so I decided to send it out.
Do you have further plans for Hoopl and GHC's back end?
The biggest thing is probably: https://github.com/ghc-proposals/ghc-proposals/pull/74 Other than that, I haven't made any plans yet ;) There are a few tickets that I'd like to make some progress on. And I might try a few experiments looking for some compile-time improvements in the Hoopl/backend - I think currently nobody (including myself) is very keen on introducing new passes or making existing ones more powerful due to compile-time constraints. And then there are efforts by Moritz (LLVM backend using binary bitcode) and Kavon (LLVM changes to support CPS-style calls) that sound really interesting to me. (but they do require some more time to understand all the context/changes, so I'm not sure if or how much I'll be able to help) - Michal