June 2017 - ghc-tickets

[GHC] #12620: Allow the user to prevent floating and CSE
by GHC 07 Nov '18

07 Nov '18

#12620: Allow the user to prevent floating and CSE -------------------------------------+------------------------------------- Reporter: nomeata | Owner: Type: feature | Status: new request | Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- This is a write-up of a rough idea that Andres Löw and me had at ICFP 2016 in order to address some Real World problems Andres noticed and that are currently hard to avoid. The goal is to give the user more control about expressions that the compiler would like to float out (or CSE), but the programmer knows better. Example (assume no list fusion exists): {{{ enum xs = zip [1..] xs }}} This leads to a horrible space leak, as GHC will float out `[1..]` to the top. Our idea is to have a magic function `nofloat :: a -> a` (magic in the same sense as `inline` and `lazy`) that the programmer would use here: {{{ enum xs = zip (nofloat [1..]) xs }}} With these effects: * Sub expressions are not floated out of a `nofloat`. * An expression of the form `nofloat e` would not be floated beyond the innermost enclosing lambda. * Two expressions of the form `nofloat e` would not be commoned up by CSE. This way, unwanted sharing is prevented. In contrast to a hypothetical `veryCheap` function, it does ''not'' mean that the compiler should float it into lambda (no unwanted duplication either). Two open questions (among many others, I am sure:) * Likely, rule matching should look through `nofloat`. At least in this example (and similar ones like `map (nofloat [1..])`, the rules in question will avoid the spaceleaks). * Possibly, nothing should be floated (inlined) ''into'' a `nofloat`. Rationale: Assume the library is changed so that {{{ [n..] = nofloat (realEnumFrom n) {-# INLINE [n..] #-} }}} Then `zip [fib 1000..]` would be rewritten by the inliner to `zip (let x = fib 1000 in (nofloat [x..]))`. Moving the `fib 1000` into the `nofloat` would change the behaviour in a possibly surprising way. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12620> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 34

[GHC] #13080: Constant values are not floated out of the loop
by GHC 07 Nov '18

07 Nov '18

#13080: Constant values are not floated out of the loop -------------------------------------+------------------------------------- Reporter: Feuerbach | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- See this stackoverflow post: http://stackoverflow.com/a/41521876/110081 Basically, the following program leaks memory on ghc 7.10.3 and ghc 8.0.1. {{{#!hs import Control.Monad (forever) main :: IO () main = worker {-# NOINLINE worker #-} worker :: (Monad m) => m () worker = forever $ poll () poll :: (Monad m) => () -> m a poll action = do return () poll action }}} Could ghc do better there? -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13080> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 24

[GHC] #13904: LLVM does not need to trash callee-saved registers.
by GHC 03 Nov '18

03 Nov '18

#13904: LLVM does not need to trash callee-saved registers. -------------------------------------+------------------------------------- Reporter: kavon | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Keywords: llvm, codegen | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time Unknown/Multiple | performance bug Test Case: | Blocked By: Blocking: | Related Tickets: #4992 Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- **Summary**: This ticket is a simplification of the LLVM backend that I already plan to do in another branch, but I'd like to push the patch into `master`, and then merge from `master` into that branch. The LLVM backend attempts to trash the physical registers corresponding to Rx, Fx, and Dx registers that are caller-saved before an FFI call by emitting a sequence such as this: {{{#!llvm store i64 undef, i64* %R3_Var store i64 undef, i64* %R4_Var store i64 undef, i64* %R5_Var store i64 undef, i64* %R6_Var store float undef, float* %F1_Var store double undef, double* %D1_Var store float undef, float* %F2_Var store double undef, double* %D2_Var store float undef, float* %F3_Var store double undef, double* %D3_Var store float undef, float* %F4_Var store double undef, double* %D4_Var store float undef, float* %F5_Var store double undef, double* %D5_Var store float undef, float* %F6_Var store double undef, double* %D6_Var %ln7bu = call ccc double (double) @llvm.sin.f64( double %ln7bs ) nounwind }}} Where the Rx/Fx/Dx registers are chosen based on what physical register those global registers map to, and whether those physical registers are caller-saved according to the C ABI. I'm certain this is unnecessary, as there is no way to know the physical register corresponding to those vars within an LLVM IR function. The calling convention used for the FFI call will preserve caller-saved registers as needed, i.e., only if such a value is needed after the call. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13904> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 13

[GHC] #9545: Evaluate Takano Akio's foldrW/buildW fusion framework as a possible replacement for foldr/build
by GHC 29 Oct '18

29 Oct '18

#9545: Evaluate Takano Akio's foldrW/buildW fusion framework as a possible replacement for foldr/build -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: task | Status: new Priority: normal | Milestone: Component: libraries/base | Version: 7.9 Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Project (more | Type of failure: than a week) | None/Unknown Blocked By: | Test Case: Related Tickets: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- Takano Akio's [https://github.com/takano-akio/ww-fusion worker-wrapper fusion] modifies `foldr/build` fairly conservatively to allow safe fusion of `foldl` and friends without relying on arity analysis. This advantage is important for two reasons that I know of: 1. As discussed in Joachim Breitner's original paper, the current arity analysis is unable to infer arity correctly when facing certain forms of recursion. 2. The current arity analysis, and probably ''any'' practical arity analysis, depends on inlining to a degree that can sometimes be unhealthy. I would love to make functions like `hPutStr` fuse, but I suspect doing so safely at present would likely cause a code explosion—the function being folded is too large to want to inline to make it available for arity analysis following fusion. I don't understand it completely yet, but it looks like `foldrW/buildW` can eliminate a lot of this uncertainty. Furthermore, it appears that functions currently written using `foldr` in a "right-handed" way can remain unchanged, using an implementation of `foldr` in terms of `foldrW`. Only "left-handed" folds will need to be rewritten to take advantage of this framework, along with all the `build`s. That said, `foldrW/buildW` probably has its weaknesses too. The basic fusion rule looks like this. {{{#!hs {-# RULES "foldrW/buildW" forall f z (i :: Wrap f b) (g :: forall c g . (Wrap g c) -> (a -> c -> c) -> c -> c) . foldrW i f z (buildW g) = g i f z #-} }}} If `g` and `i` are not inlined sufficiently to merge with each other, I imagine this could produce worse results than `foldr/build` when the latter works properly. The only way to really get a feel for performance would be to carefully modify everything to work as well as it can with this framework and see how the results compare. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9545> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 9

[GHC] #9374: Investigate Static Argument Transformation
by GHC 29 Oct '18

29 Oct '18

#9374: Investigate Static Argument Transformation -------------------------------------+------------------------------------- Reporter: jstolarek | Owner: Type: task | Status: new Priority: lowest | Milestone: Component: Compiler | Version: 7.9 Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Unknown | Type of failure: Blocked By: | None/Unknown Related Tickets: | Test Case: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- At the moment the Static Argument Transformation (SAT) optimisation is not run unless explicitly enabled with `-fstatic-argument-transformation`. The comment in DynFlags says: {{{ Max writes: I think it's probably best not to enable SAT with -O2 for the 6.10 release. The version of SAT in HEAD at the moment doesn't incorporate several improvements to the heuristics, and I'm concerned that without those changes SAT will interfere with some attempts to write "high performance Haskell", as we saw in some posts on Haskell-Cafe earlier this year. In particular, the version in HEAD lacks the tail call criterion, so many things that look like reasonable loops will be turned into functions with extra (unneccesary) thunk creation. }}} 6.10 was a long time ago. Has anything changed since then? Does it make sense to enable that optimisation now? What are the mentioned heuristics and were they finally implemented? Does anyone know what Haskell-cafe posts does Max refer to? -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9374> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 30

[GHC] #11719: Cannot use higher-rank kinds with type families
by GHC 29 Oct '18

29 Oct '18

#11719: Cannot use higher-rank kinds with type families -------------------------------------+------------------------------------- Reporter: ocharles | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1-rc2 (Type checker) | Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- This ticket came out of a discussion with Richard in this mailing list post: https://mail.haskell.org/pipermail/haskell- cafe/2016-March/123462.html Here's the code that should work, but doesn't: {{{ import Data.Kind data TyFun :: * -> * -> * type a ~> b = TyFun a b -> * type family (f :: a ~> b) @@ (x :: a) :: b data Null a = Nullable a | NotNullable a type family ((f :: b ~> c) ∘ (g :: a ~> b)) (x :: a) :: c where (f ∘ g) x = f @@ (g @@ x) type family BaseType (k :: forall a. a ~> Type) (x :: b) :: Type where -- this fails :( -- BaseType k x = (@@) k x }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11719> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 22

[GHC] #9163: Ptr should have a phantom role
by GHC 28 Oct '18

28 Oct '18

#9163: Ptr should have a phantom role ------------------------------------+------------------------------------- Reporter: simonpj | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.8.2 Keywords: | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | ------------------------------------+------------------------------------- In `GHC.Ptr` we see {{{ type role Ptr representational data Ptr a = Ptr Addr# deriving (Eq, Ord) }}} with no comments. Why is `Ptr` representational? In the same module we have `castPtr`: {{{ castPtr :: Ptr a -> Ptr b castPtr (Ptr addr) = Ptr addr }}} which unpacks and repacks a `Ptr`. If `Ptr` was phantom, we could use `coerce`. And that in turn would actually make a lot of code more efficient – there are lots of calls to `castPtr`. Specifically, in `nofib`, I tried implementing `castPtr` with `unsafeCoerce`. Then I found: * 12% less allocation in `reverse-complem` * 7.3% less allocation in `fasta`. * Binary sizes fell 0.1%. Both these benchmarks are ones that do a ''lot'' of I/O, and it turns out that `GHC.IO.Handle.Text.$wa1` has a `castPtr` that (all by itself) accounts for 12% of `reverse-complem`'s total allocation! So making `castPtr` free is good. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9163> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 41

[GHC] #13809: TH-reified data family instances have a paucity of kinds
by GHC 27 Oct '18

27 Oct '18

#13809: TH-reified data family instances have a paucity of kinds -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: (none) Type: bug | Status: new Priority: normal | Milestone: Component: Template | Version: 8.0.1 Haskell | Keywords: TypeFamilies | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- Consider this data family (and instances): {{{#!hs {-# LANGUAGE TypeFamilies #-} module Foo where data family Foo a data instance Foo ((f :: * -> *) (a :: *)) data instance Foo ((f :: (* -> *) -> *) (a :: (* -> *))) }}} These are two data family instances that GHC distinguishes by the kinds of their type parameters. However, Template Haskell does not give me the same insight that GHC has, because if I call `Language.Haskell.TH.reify ''Foo`, I get this: {{{#!hs FamilyI (DataFamilyD Foo.Foo [ KindedTV a_6989586621679025989 StarT ] (Just StarT)) [ DataInstD [] Foo.Foo [ AppT (VarT f_6989586621679026001) (VarT a_6989586621679026000) ] Nothing [] [] , DataInstD [] Foo.Foo [ AppT (VarT f_6989586621679026007) (VarT a_6989586621679026006) ] Nothing [] [] ] }}} Note that neither `f` nor `a` have a kind signature in either instance! This makes it completely impossible to tell which is which (aside from the order, which is brittle). It would make my life a lot easier if TH were to include kind signatures for each type variable in a data family instance. I can see two ways to accomplish this: 1. Include a `[TyVarBndr]` field in `DataInstD` and `NewtypeInstD` where each `TyVarBndr` is a `KindedTV`. 2. Walk over the `Type`s in a `DataInstD`/`NewtypeInstD` and ensure that every occurrence of a `VarT` is surrounded with `SigT` to indicate its kind. While (1) is arguably the cleaner solution, since it makes the kinds easy to discover, it is a breaking change. Therefore, I'm inclined to implement option (2) instead. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13809> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 9

[GHC] #11238: Redesign the dynamic linking on ELF systems
by GHC 27 Oct '18

27 Oct '18

#11238: Redesign the dynamic linking on ELF systems -------------------------------------+------------------------------------- Reporter: trommler | Owner: trommler Type: task | Status: new Priority: normal | Milestone: 8.2.1 Component: Runtime | Version: 7.10.3 System (Linker) | Keywords: | Operating System: Linux Architecture: | Type of failure: Other Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Rev(s): | Wiki Page: -------------------------------------+------------------------------------- This ticket is to follow up on my comment on #11042. Let's give dynamic linking on ELF systems a second chance. I propose the following steps: 1. Collect required features 1. Write up and discuss and agree on a specification 1. Create regression tests for all features 1. Design and implement a new dynamic linker for GHCi 1. (optional) Make it work on Mac OS X too -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11238> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 10

[GHC] #10353: Haddock for Data.List should list instances
by GHC 26 Oct '18

26 Oct '18

#10353: Haddock for Data.List should list instances -------------------------------------+------------------------------------- Reporter: goldfire | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: | Version: 7.10.1 Documentation | Operating System: Unknown/Multiple Keywords: | Type of failure: None/Unknown Architecture: | Blocked By: Unknown/Multiple | Related Tickets: Test Case: | Blocking: | Differential Revisions: | -------------------------------------+------------------------------------- I'd like to know the set of classes for which there are instances for `[]`. So, I naturally looked in the documentation for `Data.List`. However, because `Data.List` doesn't export `[]`, this list isn't there. Can this be fixed? (It's an honest question -- because of `[]`'s special syntax, I could see this fix being non-trivial.) -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10353> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 1