ghc-hetmet.git
15 years agoFix Trac #3118: missing alternative
simonpj@microsoft.com [Thu, 2 Apr 2009 15:28:34 +0000 (15:28 +0000)]
Fix Trac #3118: missing alternative

This patch fixes a rather obscure bug, whereby it's possible
for (case C a b of <alts>) to have altenatives that do not inclue
(C a b)!  See Note [Unreachable code] in CoreUtils.

15 years agoUse a local interestingDict function instead of importing SimplUtils.interestingArg
simonpj@microsoft.com [Thu, 2 Apr 2009 15:22:46 +0000 (15:22 +0000)]
Use a local interestingDict function instead of importing SimplUtils.interestingArg

I'm changing the details of SimplUtils.interstingArg, and don't want to
mess  up the way Specialise works, so this patch makes a specilialised
(ha) function, Specialise.interestingDict, that is used locally.

15 years agoBetter panic reporting
simonpj@microsoft.com [Thu, 2 Apr 2009 15:18:46 +0000 (15:18 +0000)]
Better panic reporting

Make idDetails and idInfo into proper functions, rather than record
fields, so that we can report more informatively if you use thenm on
a non-Id.

15 years agoFix pragmas in binder-swapping
simonpj@microsoft.com [Thu, 2 Apr 2009 15:16:16 +0000 (15:16 +0000)]
Fix pragmas in binder-swapping

In the case binder-swap operation (which is done in OccurAnal),
we deliberately use shadowing. But the new shadowing binder should
not inherit any fancy INLINE or NOINLINE pragmas from its parent.

15 years agoFloat constants to top-level even in first full laziness pass
simonpj@microsoft.com [Thu, 2 Apr 2009 15:14:06 +0000 (15:14 +0000)]
Float constants to top-level even in first full laziness pass

This patch changes the compilation pipeline flags so that constants
are floated to the top level even in the first full laziness pass.
For some reason this was not done before.  Fixing this makes a big
improvement in 'spectral/rewrite', and has zero effect on any other
nofib benchmark.

15 years agoupdate the intro section
Simon Marlow [Wed, 1 Apr 2009 10:52:56 +0000 (10:52 +0000)]
update the intro section

15 years agoRemove the "Installing GHC" section
Simon Marlow [Wed, 1 Apr 2009 10:34:18 +0000 (10:34 +0000)]
Remove the "Installing GHC" section

 1. it was out of date in various ways
 2. this is not the place people look for installation instructions
 3. we have installation instructions elsewhere (e.g. the INSTALL
    file in a binary distribution)
 4. the section "layout of installed files" is now on the wiki under
    Building/Installing.

15 years agoSPARC: Fix ffi019 split load/store of HsInt64 into two parts to respect alignment...
Ben.Lippmeier@anu.edu.au [Tue, 31 Mar 2009 04:51:45 +0000 (04:51 +0000)]
SPARC: Fix ffi019 split load/store of HsInt64 into two parts to respect alignment constraints

15 years agoSPARC NCG: Fix 64 bit addition, carry bit wasn't getting set.
Ben.Lippmeier@anu.edu.au [Mon, 30 Mar 2009 05:17:32 +0000 (05:17 +0000)]
SPARC NCG: Fix 64 bit addition, carry bit wasn't getting set.

15 years agoSPARC NCG: When getting a 64 bit word, promote halves to 64 bit before shifting
Ben.Lippmeier@anu.edu.au [Mon, 30 Mar 2009 05:03:46 +0000 (05:03 +0000)]
SPARC NCG: When getting a 64 bit word, promote halves to 64 bit before shifting

15 years agoSPARC NCG: HpLim is now always stored on the stack, not in a register
Ben.Lippmeier@anu.edu.au [Tue, 31 Mar 2009 05:19:00 +0000 (05:19 +0000)]
SPARC NCG: HpLim is now always stored on the stack, not in a register
   This fixes the out of memory errors we were getting on sparc
   after the following patch:

     Fri Mar 13 03:45:16 PDT 2009  Simon Marlow <marlowsd@gmail.com>
     * Instead of a separate context-switch flag, set HpLim to zero
     Ignore-this: 6c5bbe1ce2c5ef551efe98f288483b0
     This reduces the latency between a context-switch being triggered and
     the thread returning to the scheduler, which in turn should reduce the
     cost of the GC barrier when there are many cores.

15 years agoAllow Haddock comments in type synonyms
David Waern [Tue, 31 Mar 2009 21:23:06 +0000 (21:23 +0000)]
Allow Haddock comments in type synonyms

We now use `ctypedoc` instead of `ctype` for type synonyms. `ctypedoc` was
previously only used for top-level type signatures. This change means that type
synonyms now can contain comments, just like top-level type signatures.

Note:

* I've modified `ctypedoc` so it allows implicit parameters and equational
constraints, just like ctype.

* Since `ctypedoc` allows nested foralls, we now allow that in type synonyms.

* I have inlined some productions into gentypedoc so that there is now a
non-doc version of every production with a 'doc' suffix. (Stylistic change
only, which should make the code easier to follow).

* It would have been nice to simplify the grammar by unifying `ctype` and
ctypedoc` into one production, allowing comments on types everywhere (and
rejecting them after parsing, where necessary).  This is however not possible
since it leads to ambiguity. The reason is the support for comments on record
fields:

> data R = R { field :: Int -- ^ comment on the field }

If we allow comments on types here, it's not clear if the comment applies
to 'field' or to 'Int'. So we must use `ctype` to describe the type.

15 years agomkErrorAppDs now takes an SDoc rather than a String
Ian Lynagh [Tue, 31 Mar 2009 13:40:58 +0000 (13:40 +0000)]
mkErrorAppDs now takes an SDoc rather than a String
This avoids some showSDoc's where the String then gets converted back
into an SDoc.

15 years agoFIX #1779 and qq005: export hs_hpc_module
Simon Marlow [Mon, 30 Mar 2009 08:44:14 +0000 (08:44 +0000)]
FIX #1779 and qq005: export hs_hpc_module

15 years agogo back to using $CPUS + 1 for the number of threads to use
Simon Marlow [Mon, 30 Mar 2009 08:30:32 +0000 (08:30 +0000)]
go back to using $CPUS + 1 for the number of threads to use

15 years agouse a UniqSet for is MathFun, this list is getting quite large
Simon Marlow [Mon, 30 Mar 2009 08:29:57 +0000 (08:29 +0000)]
use a UniqSet for is MathFun, this list is getting quite large

15 years agoFix an nasty black hole, concerning computation of isRecursiveTyCon
simonpj@microsoft.com [Mon, 30 Mar 2009 08:49:12 +0000 (08:49 +0000)]
Fix an nasty black hole, concerning computation of isRecursiveTyCon

Fixing #246 (pattern-match order in record patterns) made GHC go into
a black hole, by changing the order of patterm matching in
TyCon.isProductTyCon!  It turned out that GHC had been avoiding the
black hole only by the narrowest of margins up to now!

The black hole concerned the computation of which type constructors
are recursive, in TcTyDecls.calcRecFlags.  We now refrain from using
isProductTyCon there, since it triggers the black hole (very
indirectly).  See the "YUK YUK" comment in the body of calcRecFlags.

As it turns out, the fact that TyCon.isProductTyCon matched on the
algTcRec field was quite redundant, so I removed that too.  However,
without the fix to calcRecFlags, this wouldn't fix the black hole
because of the use of isRecursiveTyCon in BuildTyCl.mkNewTyConRhs.

Anyway, it's fine now.

15 years agoAdd an extra print to -ddump-tc-trace
simonpj@microsoft.com [Mon, 30 Mar 2009 08:48:35 +0000 (08:48 +0000)]
Add an extra print to -ddump-tc-trace

15 years agoRemove dead code (dataConsStupidTheta)
simonpj@microsoft.com [Mon, 30 Mar 2009 08:47:39 +0000 (08:47 +0000)]
Remove dead code (dataConsStupidTheta)

15 years agoFix Trac #246: order of matching in record patterns
simonpj@microsoft.com [Mon, 30 Mar 2009 08:37:36 +0000 (08:37 +0000)]
Fix Trac #246: order of matching in record patterns

While I was looking at the desugaring of pattern matching (fixing
Trac #3126) I finally got around to fixing another long-standing bug:
when matching in a record pattern, GHC should match left-to-right in
the programmer-specfied order, *not* left-to-right positionally in
the original record declaration.

Needless to say, that requires a little more code.
See Note [Record patterns] in MatchCon.lhs

15 years agoFix Trac #3126: matching overloaded literals
simonpj@microsoft.com [Mon, 30 Mar 2009 08:34:35 +0000 (08:34 +0000)]
Fix Trac #3126: matching overloaded literals

Claus Reinke uncovered a long-standing bug in GHC, whereby we were
combining the pattern-match on overloaded literals, missing the fact
that an intervening pattern (for a different literal) might also
match.  (If someone had a very odd implementation of fromInteger!)

See Note [Grouping overloaded literal patterns] in Match.lhs

If this merges smoothly to 6.10, go for it, but it's very much
a corner case.

Thank you Claus!

15 years agoWhite space cosmetics only
simonpj@microsoft.com [Mon, 30 Mar 2009 08:31:03 +0000 (08:31 +0000)]
White space cosmetics only

15 years agoImprove mkDupableCont; and fix Trac #3116
simonpj@microsoft.com [Wed, 25 Mar 2009 09:52:05 +0000 (09:52 +0000)]
Improve mkDupableCont; and fix Trac #3116

It turns out that, as a result of a change I made a few months ago to
the representation of SimplCont, it's easy to solve the optimisation
challenge posed by Trac #3116.  Hurrah.

Extensive comments in Note [Duplicating StrictArg].

15 years agoAvoid quadratic complexity in occurrence analysis (fix Trac #1969)
simonpj@microsoft.com [Mon, 23 Mar 2009 10:38:26 +0000 (10:38 +0000)]
Avoid quadratic complexity in occurrence analysis (fix Trac #1969)

  The occurrence analyser could go out to lunch in bad cases, because
  of its clever loop-breaking algorithm. This patch makes it bale out
  in bad cases.  Somewhat ad-hoc: a nicer solution would be welcome.

  See Note [Complexity of loop breaking] for the details.

15 years agoImprove arity propagation in the specialiser
simonpj@microsoft.com [Mon, 23 Mar 2009 10:16:14 +0000 (10:16 +0000)]
Improve arity propagation in the specialiser

This patch makes the specialiser propagate arities a bit more
eagerly, which avoids a spurious warning in the simplifier.

See Note [Arity decrease] in Simplify.lhs

15 years agocheck for ThreadRelocated in checkBlackHoles()
Simon Marlow [Thu, 26 Mar 2009 14:03:43 +0000 (14:03 +0000)]
check for ThreadRelocated in checkBlackHoles()

15 years agosyb no longer needs to be wired in
Ian Lynagh [Thu, 26 Mar 2009 13:05:17 +0000 (13:05 +0000)]
syb no longer needs to be wired in

15 years agoTemplate Haskell: added bang patterns
Manuel M T Chakravarty [Thu, 26 Mar 2009 10:02:08 +0000 (10:02 +0000)]
Template Haskell: added bang patterns

15 years agoTemplate Haskell: support for kind annotations
Manuel M T Chakravarty [Thu, 26 Mar 2009 08:55:20 +0000 (08:55 +0000)]
Template Haskell: support for kind annotations

15 years agoupdate list of C math functions
Bertram Felgenhauer [Mon, 23 Mar 2009 18:36:30 +0000 (18:36 +0000)]
update list of C math functions
Fix via C compilation of modules that import, say, log1p from math.h (#3117)

The list is based on preprocessing Stg.h with glibc 2.6.1 headers, and
cross-checked with the ISO C 99 standard (draft).

15 years agoTemplate Haskell: make reify aware of type families
Manuel M T Chakravarty [Wed, 25 Mar 2009 03:34:47 +0000 (03:34 +0000)]
Template Haskell: make reify aware of type families
- Reifying a type family returns a TH family declaration
- Reifying a data constructor from a data instance attributes that
  constructor to the family (not the representation tycon)
- Ideally, we should have facilities to reify all type/data instances of a
  given family (and the same for instances of a class).  I haven't added that
  here as it involves some API design.

15 years agoTemplate Haskell: support for INLINE and SPECIALISE pragmas
Manuel M T Chakravarty [Tue, 24 Mar 2009 23:29:40 +0000 (23:29 +0000)]
Template Haskell: support for INLINE and SPECIALISE pragmas

15 years agofix an fprintf that should be fputs (quiets gcc)
Simon Marlow [Sun, 22 Mar 2009 19:54:41 +0000 (19:54 +0000)]
fix an fprintf that should be fputs (quiets gcc)

15 years agocheck return value of write (quiets gcc)
Simon Marlow [Sun, 22 Mar 2009 19:53:19 +0000 (19:53 +0000)]
check return value of write (quiets gcc)

15 years agoFix locking in evacuate_large() (FIX openfile008(threaded2))
Simon Marlow [Mon, 23 Mar 2009 09:29:53 +0000 (09:29 +0000)]
Fix locking in evacuate_large() (FIX openfile008(threaded2))

15 years agoPlease the -Wall police by moving a variable declaration; really FIX getNumberOfProce...
Austin Seipp [Fri, 20 Mar 2009 12:48:46 +0000 (12:48 +0000)]
Please the -Wall police by moving a variable declaration; really FIX getNumberOfProcessors() for MacOS X

15 years agowibble in setExecutable
Austin Seipp [Fri, 20 Mar 2009 12:00:43 +0000 (12:00 +0000)]
wibble in setExecutable

15 years agoFIX getNumberOfProcessors() on MacOS X
Austin Seipp [Wed, 18 Mar 2009 02:31:27 +0000 (02:31 +0000)]
FIX getNumberOfProcessors() on MacOS X

This checks if darwin_HOST_OS is defined and, if so, we call
sysctlbyname() on the "hw.ncpu" property to get the processor count.

15 years agoWork around Solaris's grep not supporting -q
Ian Lynagh [Thu, 19 Mar 2009 17:15:01 +0000 (17:15 +0000)]
Work around Solaris's grep not supporting -q

15 years agoSome wording wibbles from Thorkil
Ian Lynagh [Thu, 19 Mar 2009 16:59:16 +0000 (16:59 +0000)]
Some wording wibbles from Thorkil

15 years agoFix bug in setThreadAffinity() (Linux)
Simon Marlow [Fri, 20 Mar 2009 10:04:46 +0000 (10:04 +0000)]
Fix bug in setThreadAffinity() (Linux)

15 years agoImplement setThreadAffinity for Mac OS X (from 10.5)
Manuel M T Chakravarty [Fri, 20 Mar 2009 07:13:47 +0000 (07:13 +0000)]
Implement setThreadAffinity for Mac OS X (from 10.5)

15 years agoFix warnings with older versions of gcc (3.4.5)
Simon Marlow [Thu, 19 Mar 2009 12:49:32 +0000 (12:49 +0000)]
Fix warnings with older versions of gcc (3.4.5)

15 years agowibble for getNumberOfProcessors
Simon Marlow [Wed, 18 Mar 2009 16:59:36 +0000 (16:59 +0000)]
wibble for getNumberOfProcessors

15 years agoSet thread affinity with +RTS -qa (only on Linux so far)
Simon Marlow [Wed, 18 Mar 2009 15:47:19 +0000 (15:47 +0000)]
Set thread affinity with +RTS -qa (only on Linux so far)

15 years agoTemplate Haskell support for equality constraints
Manuel M T Chakravarty [Thu, 19 Mar 2009 13:23:47 +0000 (13:23 +0000)]
Template Haskell support for equality constraints

15 years agoadd missing case in ENTER() (fixes readwrite002(profasm) crash)
Simon Marlow [Thu, 19 Mar 2009 10:21:26 +0000 (10:21 +0000)]
add missing case in ENTER() (fixes readwrite002(profasm) crash)

15 years agoFIX unregisterised build
Simon Marlow [Thu, 19 Mar 2009 10:09:54 +0000 (10:09 +0000)]
FIX unregisterised build

15 years agoTemplate Haskell: support for type family declarations
Manuel M T Chakravarty [Thu, 19 Mar 2009 08:43:06 +0000 (08:43 +0000)]
Template Haskell: support for type family declarations

15 years agoComment explaining use of seq in DFMonad
dias@eecs.tufts.edu [Wed, 18 Mar 2009 21:01:38 +0000 (21:01 +0000)]
Comment explaining use of seq in DFMonad

15 years agoFIX unregisterised target by #ifdefing targetPlatform stuff
Simon Marlow [Wed, 18 Mar 2009 15:12:55 +0000 (15:12 +0000)]
FIX unregisterised target by #ifdefing targetPlatform stuff
I'm not sure if this is the correct fix.  If targetPlatform is really
NCG-specific, then maybe we should call it asmTargetPlatform or
something.

15 years agoRemoved a trace
dias@eecs.tufts.edu [Wed, 18 Mar 2009 15:13:43 +0000 (15:13 +0000)]
Removed a trace

15 years agoCalling convention bug and cleanup
dias@eecs.tufts.edu [Tue, 17 Mar 2009 20:42:38 +0000 (20:42 +0000)]
Calling convention bug and cleanup
- yet another wrong calling convention; this one was a special case for returning one
  value.

15 years agoInconsistent type and arguments in safe foreign calls...
dias@eecs.tufts.edu [Mon, 16 Mar 2009 21:46:54 +0000 (21:46 +0000)]
Inconsistent type and arguments in safe foreign calls...
- The function argument was stripped from the argument list but not from the type.
  Now they're both stripped.

15 years agostack overflows and out of memory's
dias@eecs.tufts.edu [Mon, 16 Mar 2009 21:35:06 +0000 (21:35 +0000)]
stack overflows and out of memory's
1. Stack overflow fixed by making dataflow monad strict in the state.
2. Out of memory fixed by "forgetting" lastoutfacts in the dataflow monad
   where we should. We were creating an unnecessarily long list that grew
   exponentially...

15 years agoAdd the notion of "constructor-like" Ids for rule-matching
simonpj@microsoft.com [Wed, 18 Mar 2009 10:59:11 +0000 (10:59 +0000)]
Add the notion of "constructor-like" Ids for rule-matching

This patch adds an optional CONLIKE modifier to INLINE/NOINLINE pragmas,
   {-# NOINLINE CONLIKE [1] f #-}
The effect is to allow applications of 'f' to be expanded in a potential
rule match.  Example
  {-# RULE "r/f" forall v. r (f v) = f (v+1) #-}

Consider the term
     let x = f v in ..x...x...(r x)...
Normally the (r x) would not match the rule, because GHC would be scared
about duplicating the redex (f v). However the CONLIKE modifier says to
treat 'f' like a constructor in this situation, and "look through" the
unfolding for x.  So (r x) fires, yielding (f (v+1)).

The main changes are:
  - Syntax

  - The inlinePragInfo field of an IdInfo has a RuleMatchInfo
    component, which records whether or not the Id is CONLIKE.
    Of course, this needs to be serialised in interface files too.

  - The occurrence analyser (OccAnal) and simplifier (Simplify) treat
    CONLIKE thing like constructors, by ANF-ing them

  - New function coreUtils.exprIsExpandable is like exprIsCheap, but
    additionally spots applications of CONLIKE functions

  - A CoreUnfolding has a field that caches exprIsExpandable

  - The rule matcher consults this field.  See
    Note [Expanding variables] in Rules.lhs.

On the way I fixed a lurking variable bug in the way variables are
expanded.  See Note [Do not expand locally-bound variables] in
Rule.lhs.  I also did a bit of reformatting and refactoring in
Rules.lhs, so the module has more lines changed than are really
different.

15 years agoFixes to "Retract Hp *before* checking for HpLim==0"
Simon Marlow [Wed, 18 Mar 2009 11:18:47 +0000 (11:18 +0000)]
Fixes to "Retract Hp *before* checking for HpLim==0"

15 years agoAdd fast event logging
Simon Marlow [Tue, 17 Mar 2009 16:42:14 +0000 (16:42 +0000)]
Add fast event logging

Generate binary log files from the RTS containing a log of runtime
events with timestamps.  The log file can be visualised in various
ways, for investigating runtime behaviour and debugging performance
problems.  See for example the forthcoming ThreadScope viewer.

New GHC option:

  -eventlog   (link-time option) Enables event logging.

  +RTS -l     (runtime option) Generates <prog>.eventlog with
              the binary event information.

This replaces some of the tracing machinery we already had in the RTS:
e.g. +RTS -vg  for GC tracing (we should do this using the new event
logging instead).

Event logging has almost no runtime cost when it isn't enabled, though
in the future we might add more fine-grained events and this might
change; hence having a link-time option and compiling a separate
version of the RTS for event logging.  There's a small runtime cost
for enabling event-logging, for most programs it shouldn't make much
difference.

(Todo: docs)

15 years agoFIX biographical profiling (#3039, probably #2297)
Simon Marlow [Tue, 17 Mar 2009 14:49:39 +0000 (14:49 +0000)]
FIX biographical profiling (#3039, probably #2297)
Since we introduced pointer tagging, we no longer always enter a
closure to evaluate it.  However, the biographical profiler relies on
closures being entered in order to mark them as "used", so we were
getting spurious amounts of data attributed to VOID.  It turns out
there are various places that need to be fixed, and I think at least
one of them was also wrong before pointer tagging (CgCon.cgReturnDataCon).

15 years agoAdd getNumberOfProcessors(), FIX MacOS X build problem (hopefully)
Simon Marlow [Tue, 17 Mar 2009 09:33:09 +0000 (09:33 +0000)]
Add getNumberOfProcessors(), FIX MacOS X build problem (hopefully)
Somebody needs to implement getNumberOfProcessors() for MacOS X,
currently it will return 1.

15 years agoFIX #3093: stub filenames when using -osuf
Simon Marlow [Mon, 16 Mar 2009 20:34:37 +0000 (20:34 +0000)]
FIX #3093: stub filenames when using -osuf
Also remove some unused cruft

15 years agoDon't call processHeapClosureForDead on pinned blocks
Simon Marlow [Mon, 16 Mar 2009 16:05:34 +0000 (16:05 +0000)]
Don't call processHeapClosureForDead on pinned blocks
Fixes heapprof001(prof_hp) after fix for #2917

15 years agoRetract Hp *before* checking for HpLim==0
Simon Marlow [Mon, 16 Mar 2009 16:05:00 +0000 (16:05 +0000)]
Retract Hp *before* checking for HpLim==0
Fixes heapprof001(prof_hp) following the recent HpLim patch, which
depended on the lack of slop in the heap.

15 years agoeliminate some not-quite-duplicate code
Simon Marlow [Fri, 13 Mar 2009 15:56:19 +0000 (15:56 +0000)]
eliminate some not-quite-duplicate code

15 years agostart finalizers on the current Capability rather than last_free_capability
Simon Marlow [Fri, 13 Mar 2009 15:56:02 +0000 (15:56 +0000)]
start finalizers on the current Capability rather than last_free_capability

15 years agocope with ThreadRelocated on the sleeping_queue
Simon Marlow [Tue, 17 Feb 2009 13:27:57 +0000 (13:27 +0000)]
cope with ThreadRelocated on the sleeping_queue

15 years agoFix Trac #3095, and make RdrHsSyn warning-clean
simonpj@microsoft.com [Mon, 16 Mar 2009 17:47:06 +0000 (17:47 +0000)]
Fix Trac #3095, and make RdrHsSyn warning-clean

15 years agoLayout only
simonpj@microsoft.com [Mon, 16 Mar 2009 16:46:19 +0000 (16:46 +0000)]
Layout only

15 years agoReject foralls in constructor args in 'deriving', except for Functor etc
simonpj@microsoft.com [Mon, 16 Mar 2009 16:45:02 +0000 (16:45 +0000)]
Reject foralls in constructor args in 'deriving', except for Functor etc

15 years agoFix Trac #3092
simonpj@microsoft.com [Mon, 16 Mar 2009 16:40:49 +0000 (16:40 +0000)]
Fix Trac #3092

We were't checking that a 'data/type instance' was extending a family
type constructor.

Merge to 6.10 if we ever release 6.10.3 (or do it for 6.10.2).

15 years agoEnsure the orientation of var-var equalities is correct for instatiation
Manuel M T Chakravarty [Sun, 15 Mar 2009 06:57:09 +0000 (06:57 +0000)]
Ensure the orientation of var-var equalities is correct for instatiation
- During fianlisation we use to occasionally swivel variable-variable equalities
- Now, normalisation ensures that they are always oriented as appropriate for
  instantation.
- Also fixed #1899 properly; the previous fix fixed a symptom, not the cause.

15 years agoUse work-stealing for load-balancing in the GC
Simon Marlow [Fri, 13 Mar 2009 13:51:16 +0000 (13:51 +0000)]
Use work-stealing for load-balancing in the GC

New flag: "+RTS -qb" disables load-balancing in the parallel GC
(though this is subject to change, I think we will probably want to do
something more automatic before releasing this).

To get the "PARGC3" configuration described in the "Runtime support
for Multicore Haskell" paper, use "+RTS -qg0 -qb -RTS".

The main advantage of this is that it allows us to easily disable
load-balancing altogether, which turns out to be important in parallel
programs.  Maintaining locality is sometimes more important that
spreading the work out in parallel GC.  There is a side benefit in
that the parallel GC should have improved locality even when
load-balancing, because each processor prefers to take work from its
own queue before stealing from others.

15 years agoFix Trac #3087: derived Data now defines dataCast1/2
simonpj@microsoft.com [Fri, 13 Mar 2009 13:44:36 +0000 (13:44 +0000)]
Fix Trac #3087: derived Data now defines dataCast1/2

This patch generates code in deriving(Data) for dataCast1 or 2 as
appropriate.

While I was there I did some refactoring (of course), pulling out
the TcDeriv.inferConstraints as a separate function.

I don't think it's worth merging this to 6.10.2, even though it's a bugfix,
because it modifies code that I added in the HEAD only (for deriving Functor)
so the merge will be sligtly awkward.  And there's an easy workaround.

15 years agoAdjust error message slightly to make it clearer
simonpj@microsoft.com [Fri, 13 Mar 2009 13:44:02 +0000 (13:44 +0000)]
Adjust error message slightly to make it clearer

15 years agojust comment formatting
Simon Marlow [Fri, 13 Mar 2009 13:23:39 +0000 (13:23 +0000)]
just comment formatting

15 years agoAdd "+RTS -N" to determine the -N value automatically (see #1741)
Simon Marlow [Fri, 13 Mar 2009 11:46:46 +0000 (11:46 +0000)]
Add "+RTS -N" to determine the -N value automatically (see #1741)

15 years agotidy up "missing symbol" error message
Simon Marlow [Fri, 13 Mar 2009 10:13:34 +0000 (10:13 +0000)]
tidy up "missing symbol" error message

15 years agoFix sanity checking after fix to #2917
Simon Marlow [Thu, 12 Mar 2009 09:38:55 +0000 (09:38 +0000)]
Fix sanity checking after fix to #2917

15 years agoInstead of a separate context-switch flag, set HpLim to zero
Simon Marlow [Fri, 13 Mar 2009 10:45:16 +0000 (10:45 +0000)]
Instead of a separate context-switch flag, set HpLim to zero
This reduces the latency between a context-switch being triggered and
the thread returning to the scheduler, which in turn should reduce the
cost of the GC barrier when there are many cores.

We still retain the old context_switch flag which is checked at the
end of each block of allocation.  The idea is that setting HpLim may
fail if the the target thread is modifying HpLim at the same time; the
context_switch flag is a fallback.  It also allows us to "context
switch soon" without forcing an immediate switch, which can be costly.

15 years agoFIX #3079, dodgy parsing of LANGUAGE pragmas 2009-03-13
Simon Marlow [Thu, 12 Mar 2009 14:11:03 +0000 (14:11 +0000)]
FIX #3079, dodgy parsing of LANGUAGE pragmas
I ended up rewriting this horrible bit of code, using (yikes) lazy I/O
to slurp in the source file a chunk at a time.  The old code tried to
read the file a chunk at a time, but failed with LANGUAGE pragmas
because the parser for LANGUAGE has state and the state wasn't being
saved between chunks.  We're still closing the Handle eagerly, so
there shouldn't be any problems here.

15 years agoavoid a crash: don't return unless the run queue has some threads in it
Simon Marlow [Wed, 11 Mar 2009 15:45:59 +0000 (15:45 +0000)]
avoid a crash: don't return unless the run queue has some threads in it

15 years agoAllocate the right number of words in new*PinnedByteArrayzh_fast
Ian Lynagh [Wed, 11 Mar 2009 16:06:15 +0000 (16:06 +0000)]
Allocate the right number of words in new*PinnedByteArrayzh_fast

15 years agoFIX #2816 (correct unicode output for :type/:kind)
Simon Marlow [Wed, 11 Mar 2009 11:23:11 +0000 (11:23 +0000)]
FIX #2816 (correct unicode output for :type/:kind)
This is just a hack, since we don't have correct unicode output for
Handles in general, I just fixed a couple of places where we were not
converting to UTF-8 for output.

15 years agoFix #2992: don't create a named event
Simon Marlow [Wed, 11 Mar 2009 09:42:08 +0000 (09:42 +0000)]
Fix #2992: don't create a named event
Evidently I misread the docs for CreateEvent: if you pass a name to
CreateEvent, then it creates a single shared system-wide Event with
that name.  So all Haskell processes on the machine were sharing the
same Event object.  duh.

15 years agoFIX #2832: Setting SplitObjs=NO doesn't disable -split-objs in GHC
Simon Marlow [Wed, 11 Mar 2009 11:00:58 +0000 (11:00 +0000)]
FIX #2832: Setting SplitObjs=NO doesn't disable -split-objs in GHC
Now ghc --info reports whether-split-objs is supported, rather than
whether the libraries were built using -split-objs.

15 years agoFIX #2976: fix buggy implementation of shadowing in GHC.getBindings
Simon Marlow [Wed, 11 Mar 2009 10:20:07 +0000 (10:20 +0000)]
FIX #2976: fix buggy implementation of shadowing in GHC.getBindings

15 years agoDon't put a trailing / on the mingw include path
Ian Lynagh [Tue, 10 Mar 2009 17:22:42 +0000 (17:22 +0000)]
Don't put a trailing / on the mingw include path

15 years agoFix a bug which sometimes caused extra major GCs to be performed
Simon Marlow [Mon, 9 Mar 2009 14:00:04 +0000 (14:00 +0000)]
Fix a bug which sometimes caused extra major GCs to be performed
A long-running GC would cause the timer signal to declare the system
to be idle, which would cause a major GC immediately following the
current GC.  This only happened with +RTS -N2 or greater.

15 years agoRedesign 64-bit HEAP_ALLOCED (FIX #2934 at the same time)
Simon Marlow [Mon, 9 Mar 2009 12:13:00 +0000 (12:13 +0000)]
Redesign 64-bit HEAP_ALLOCED (FIX #2934 at the same time)

After much experimentation, I've found a formulation for HEAP_ALLOCED
that (a) improves performance, and (b) doesn't have any race
conditions when used concurrently.  GC performance on x86_64 should be
improved slightly.  See extensive comments in MBlock.h for the
details.

15 years agoPartial fix for #2917
Simon Marlow [Fri, 6 Mar 2009 10:00:18 +0000 (10:00 +0000)]
Partial fix for #2917

 - add newAlignedPinnedByteArray# for allocating pinned BAs with
   arbitrary alignment

 - the old newPinnedByteArray# now aligns to 16 bytes

Foreign.alloca will use newAlignedPinnedByteArray#, and so might end
up wasting less space than before (we used to align to 8 by default).
Foreign.allocaBytes and Foreign.mallocForeignPtrBytes will get 16-byte
aligned memory, which is enough to avoid problems with SSE
instructions on x86, for example.

There was a bug in the old newPinnedByteArray#: it aligned to 8 bytes,
but would have failed if the header was not a multiple of 8
(fortunately it always was, even with profiling).  Also we
occasionally wasted some space unnecessarily due to alignment in
allocatePinned().

I haven't done anything about Foreign.malloc/mallocBytes, which will
give you the same alignment guarantees as malloc() (8 bytes on
Linux/x86 here).

15 years agoDon't force HSCOLOUR_SRCS = YES when validating
Ian Lynagh [Sun, 8 Mar 2009 17:16:33 +0000 (17:16 +0000)]
Don't force HSCOLOUR_SRCS = YES when validating
This removes a burden from developers, and I can't remember an
occasion where it would have caught a regression.

15 years agoAdd CONTRACT to the pragmas that we recognise.
Ian Lynagh [Fri, 6 Mar 2009 19:48:26 +0000 (19:48 +0000)]
Add CONTRACT to the pragmas that we recognise.
It's used by ESC/Haskell.

15 years agoGenerate lots of __inline_me during vectorisation
Roman Leshchinskiy [Sat, 7 Mar 2009 13:56:52 +0000 (13:56 +0000)]
Generate lots of __inline_me during vectorisation

15 years agoSpecial-case desugaring of simple parallel array comprehensions
Roman Leshchinskiy [Sat, 7 Mar 2009 13:40:49 +0000 (13:40 +0000)]
Special-case desugaring of simple parallel array comprehensions

15 years agoMake LDV_FILL_SLOP use a forwards loop rather than a backwards loop
Ian Lynagh [Fri, 6 Mar 2009 15:51:24 +0000 (15:51 +0000)]
Make LDV_FILL_SLOP use a forwards loop rather than a backwards loop

15 years agoFix warning
Roman Leshchinskiy [Fri, 6 Mar 2009 12:36:45 +0000 (12:36 +0000)]
Fix warning

15 years agoTry not to avoid vectorising purely scalar functions
Roman Leshchinskiy [Fri, 6 Mar 2009 11:55:08 +0000 (11:55 +0000)]
Try not to avoid vectorising purely scalar functions

15 years agoPackage dph needs to be cleaned in stage=1, too
Manuel M T Chakravarty [Fri, 6 Mar 2009 03:16:28 +0000 (03:16 +0000)]
Package dph needs to be cleaned in stage=1, too

15 years agoAdd --version to runghc. Trac #2757.
Ian Lynagh [Thu, 5 Mar 2009 16:20:45 +0000 (16:20 +0000)]
Add --version to runghc. Trac #2757.
We use the GHC version number, as the old runghc one doesn't seem very
useful.

15 years agoadd final newlines
Simon Marlow [Thu, 5 Mar 2009 14:00:14 +0000 (14:00 +0000)]
add final newlines
My Windows build has started complaining about lacking final newlines,
I'm not entirely sure why.