|Version 28 (modified by benl, 22 months ago) (diff)|
GHC Status Report May 2013
As planned, we made another minor release 7.6.2 from the 7.6 branch in January 2013. This included only bug and performance fixes; no new features were added. We plan to put out a new major release 7.8.1 soon after ICFP 2013, including some significant changes described below.
There remains more to do than we will ever have time for, so please do come and join in the fun!
Source language and type system
Simon and Dimitrios overhauled the solver for type constraints, once more. No new functionality, but the result is smaller, faster, and lacks many of the bugs of its predecessor. You don't want to know all the details, but it reminded us again of how valuable it is that the constraint solve is now one coherent piece of code, with a well-defined task, than being spread out in bits and pieces across the type checker.
Meanwhile others have been adding new features.
- Poly-kinded Typeable. The Typeable class is now kind-polymorphic, meaning we can finally drop the boilerplate TypeableN classes. The new definition of Typeable is as follows:
class Typeable (a :: k) where typeRep :: proxy a -> TypeRep
With this change comes the ability to derive Typeable instances for every user datatype, and even for type classes. This means user defined instances of Typeable are unnecessary. Furthermore, since ill-defined user instances can lead to runtime errors, they are now forbidden; the only way to get Typeable instances is by using the deriving mechanism. User-defined instances will be ignored, with a warning.
Migrating to this new Typeable is easy. Code that only derived Typeable instances, and did not mention any of the TypeableN classes, should work as before. Code that mentioned the TypeableN classes should be adapted to replace these by the poly-kinded Typeable class. User-defined instances of Typeable should be replaced by derived instances.
Additionally, a new compiler pragma AutoDeriveTypeable triggers automatic derivation of Typeable instances for all datatypes and classes defined in the module.
- Type holes [Thijs Alkemade]
- Rebindable list syntax. A GHC extension called overloaded lists was added by Achim Krause, George Giorgidze, and colleagues. When this is turned on, the way GHC desugars explicit lists and lists in arithmetic sequence notation is changed. Instead of directly desugaring to built-in lists, a polymorphic witness function is used, similar to the desugaring of numeric literals. This allows for a more flexible use of list notations, supporting many different list-like types. In addition, the functions used in this desugaring process are completely rebindable.
- Type level natural numbers. Iavor S. Diatchki has been working on a solver for equations involving type-level natural numbers. This allows simplifying and reasoning about type-level terms involving arithmetic. Currently, the solver can evaluate equations and inequalities mentioning the type functions (+), (*), (^), and (<=). The solver works pretty well when it can use evaluation to prove equalities (e.g., examples like 2 + 5 = x, 2 + x = 5). There is also some support for taking advantage of the commutativity and associativity of (+), (*). More experimental features include: support for (-), which, currently is implemented by desugaring to (+); the type-level function FromNat1, which has special support for working with natural number literals, and thus can be used to expose some of their inductive structure. This work is currently on the type-nats branch, and the plan is to merge it into HEAD into the next few months.
- Kinds without data Trevor Elliott, Eric Mertens, and Iavor Diatchki have began implementing support for "data kind" declarations, described in more detail on the GHC wiki: http://hackage.haskell.org/trac/ghc/wiki/GhcKinds/KindsWithoutData The idea is to allow a new form of declaration that introduces a new kind, whose members are described by the (type) constructors in the declaration. This is similar to promoting data declarations, except that no new value-level-constructors are declared, and it also allows the constructors to mention other kinds that do not have corresponding type-level representation (e.g., *).
- Ordered overlapping type family instances. Richard Eisenberg has implemented support for ordered overlapping type family instances, called branched instances. This allows type-level functions to use patterns in a similar way to term-level functions. For example:
type family Equals (x :: *) (y :: *) :: Bool type instance where Equals x x = True Equals x y = False
Details can be found in the wiki page .
Back end and code generation
- The new code generator. [entry copied from Oct 2012 status report] Several years since this project was started, the new code generator is finally working , and is now switched on by default in master. It will be in GHC 7.8.1. From a user's perspective there should be very little difference, though some programs will be faster.
There are three important improvements in the generated code. One is that let-no-escape functions are now compiled much more efficiently: a recursive let-no-escape now turns into a real loop in C--. The second improvement is that global registers (R1, R2, etc.) are now available for the register allocator to use within a function, provided they aren't in use for argument passing. This means that there are more registers available for complex code sequences. The third improvement is that we have a new sinking pass that replaces the old "mini-inliner" from the native code generator, and is capable of optimisations that the old pass couldn't do.
Hand-written C-- code can now be written in a higher-level style with real function calls, and most of the hand-written C-- code in the RTS has been converted into the new style. High-level C-- does not mention global registers such as R1 explicitly, nor does it manipulate the stack; all this is handled by the C-- code generator in GHC. This is more robust and simpler, and means that we no longer need a special calling-convention for primops - they now use the same calling convention as ordinary Haskell functions.
We're interested in hearing about both performance improvements and regressions due to the new code generator.
- support for vector (SSE/AVX) instructions [Geoffrey Mainland]
Data Parallel Haskell
- Vectorisation Avoidance. Gabriele Keller and Manuel Chakravarty have extended the DPH vectoriser with an analysis that determines when expressions cannot profitably be vectorised. Vectorisation avoidance improves compile times for DPH programs, as well as simplifying the handling of vectorised primitive operations. This work is now complete and will be in GHC 7.8.
- New Fusion Framework. Ben Lippmeier has been waging a protracted battle with the problem of array fusion. Absolute performance in DPH is critically dependent on a good array fusion system, but existing methods cannot properly fuse the code produced by the DPH vectoriser. An important case is when a produced array is consumed by multiple consumers. In vectorised code this is very common, but none of the "short cut" array fusion approaches can handle it -- eg stream fusion used in Data.Vector, delayed array fusion in Repa, build/foldr fusion etc. The good news is that we've found a solution that handles this case and others, based on Richard Waters's series expressions, and are now working on an implementation. The new fusion system is embodied by a GHC plugin that performs a custom core-to-core transformation, and some added support to the existing Repa library. We've pushing to get the first version working for a paper at the upcoming Haskell Symposium.
The runtime system
- The new parallel I/O manager [Andreas Voellmy]
Building and linking
- Dynamic ghci. Ian Lynagh has changed GHCi to use dynamic libraries rather than static libraries. This means that we are now able to use the system linker to load packages, rather than having to implement our own linker. From the user's point of view, that means that a number of long-standing bugs in GHCi will be fixed, and it also reduces the amount of work needed to get a fully functional GHC port to a new platform. Currently, on Windows GHCi still uses static libraries, but we hope to have dynamic libraries working on Windows too by the time we release.
- Three connected projects: registerised ARM support added using David Terei's LLVM compiler back end with Stephen Blackheath doing an initial ARMv5 version and LLVM patch and Karel Gardas working on floating point support, ARMv7 compatibility and LLVM headaches. Ben Gamari did work on the runtime linker for ARM; general cross-compiling with much work by Stephen Blackheath and Gabor Greif (though many others have worked on this); culminating in the ability to compile GHC into a cross compiler for iOS (see http://hackage.haskell.org/trac/ghc/wiki/Building/CrossCompiling/iOS) iOS-specific parts were mostly Stephen Blackheath with Luke Iannini on the Cabal patch, testing and supporting infrastructure, also with assistance and testing by Miëtek Bak and Jonathan Fischoff, and thanks to many others for testing; The iOS cross compiler was started back in 2009 by Stephen Blackheath with funding from Ryan Trinkle of iPwn Studios. Thanks to Ian Lynagh for making it easy for us with integration, makefile refactoring and patience, and to David Terei for LLVM assistance.
 Overlapping type family instances: http://hackage.haskell.org/trac/ghc/wiki/NewAxioms
 The new codegen is nearly ready to go live http://hackage.haskell.org/trac/ghc/blog/newcg-update