Changes between Version 8 and Version 9 of Commentary/Compiler/HooplPerformance


Ignore:
Timestamp:
Aug 4, 2012 6:01:14 AM (3 years ago)
Author:
heisenbug
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • Commentary/Compiler/HooplPerformance

    v8 v9  
    1616== Speculation and Commentary ==
    1717
    18  * Simon PJ had questions about "optimization fuel" from the beginning.  Norman maintains that optimization fuel is an invaluable debugging aid, but that in a production compiler, one would like it to be turned off.   At some point we had abstracted over the {{{FuelMonad}}} so that we could make a "zero" fuel monad that did nothing and cost nothing.  As of January 2012, Norman doesn't know what the state of that plan is or whether GHC's optmiser can actually eliminate the overheads.
     18 * Simon PJ had questions about "optimization fuel" from the beginning.  Norman maintains that optimization fuel is an invaluable debugging aid, but that in a production compiler, one would like it to be turned off.   At some point we had abstracted over the {{{FuelMonad}}} so that we could make a "zero" fuel monad that did nothing and cost nothing.  As of January 2012, Norman doesn't know what the state of that plan is or whether GHC's optimiser can actually eliminate the overheads.
    1919
    2020 * Unlike Fuel, a supply of {{{Uniq}}}s was believed to be an absolute necessity: an optimiser must be able to rewrite blocks, and in the general case, it must be able to introduce new blocks.  It was believed that the only way to do this consistent with GHC was to plumb in a Uniq supply.   ''Query'': was this integrated with Fuel somehow?