... | ... | @@ -44,15 +44,15 @@ Here we list some problems that we know about, and/or ideas for improving things |
|
|
|
|
|
- x86_64: use more registers (the NCG needs to save/restore volatile regs over C calls, apart from that we're there).
|
|
|
- via-C produces code with too many jumps, especially with later versions of GCC. GCC commons up identical basic blocks containing a single jump instruction.
|
|
|
- GC has been reported to behave non-linearly [ http://www.haskell.org//pipermail/glasgow-haskell-users/2005-April/008301.html](http://www.haskell.org//pipermail/glasgow-haskell-users/2005-April/008301.html) (SimonMarlow: turned out to be caused by swapping)
|
|
|
- GC has been reported to behave non-linearly [http://www.haskell.org//pipermail/glasgow-haskell-users/2005-April/008301.html](http://www.haskell.org//pipermail/glasgow-haskell-users/2005-April/008301.html) (SimonMarlow: turned out to be caused by swapping)
|
|
|
- Space leak: a function that is statically known to not use some or all of its arguments should have a function descriptor that reflects the unused pointers as non-pointers, so that a PAP applied to unused args doesn't retain them.
|
|
|
- Push heap check into a case branch if that means we can avoid a heap check in a tail-recursive loop.
|
|
|
- Make ticky-ticky profiling work again.
|
|
|
- Make the NCG handle loops, and perform some simple loop optimisations.
|
|
|
- Better optimisation pass sequences? (c.f. Laszlo Nemeth's research [ http://www.tcs.informatik.uni-muenchen.de/\~hwloidl/TFP04/Abstracts/17.html](http://www.tcs.informatik.uni-muenchen.de/~hwloidl/TFP04/Abstracts/17.html))
|
|
|
- Better optimisation pass sequences? (c.f. Laszlo Nemeth's research [http://www.tcs.informatik.uni-muenchen.de/\~hwloidl/TFP04/Abstracts/17.html](http://www.tcs.informatik.uni-muenchen.de/~hwloidl/TFP04/Abstracts/17.html))
|
|
|
- Can we do anything about Double alignment?
|
|
|
- better GC for chains of THUNK_SELECTORs (see comments in GC.c)
|
|
|
- Int64 reported to be slow [ http://www.haskell.org//pipermail/glasgow-haskell-users/2005-June/008574.html](http://www.haskell.org//pipermail/glasgow-haskell-users/2005-June/008574.html) (Lemmih: Fixed. It's now only three times slower to use Int64 over Int in that example. (patch not commited yet)).
|
|
|
- Int64 reported to be slow [http://www.haskell.org//pipermail/glasgow-haskell-users/2005-June/008574.html](http://www.haskell.org//pipermail/glasgow-haskell-users/2005-June/008574.html) (Lemmih: Fixed. It's now only three times slower to use Int64 over Int in that example. (patch not commited yet)).
|
|
|
- System.Random could be optimized more.
|
|
|
|
|
|
|
... | ... | |