1. 14 Feb, 2002 1 commit
    • simonpj's avatar
      [project @ 2002-02-14 15:08:08 by simonpj] · debf9165
      simonpj authored
      -------------------------------------------------
      	Undo an earlier hack in postInlineUnconditionally
      	-------------------------------------------------
      
      In an earlier era I made postInlineUnconditionally rather less
      aggressive; it didn't inline even trivial things unless they
      occurred just once.  THis was a hack designed to avoid rules
      unexpectedly not firing.  But now we have much more control
      over rules, through the phase numbering stuff, so I can undo the
      hack.
      
      Well, so I believe.  Manuel, yell if your rules stop working!
      debf9165
  2. 12 Feb, 2002 1 commit
    • simonmar's avatar
      [project @ 2002-02-12 15:17:13 by simonmar] · 2cc5b907
      simonmar authored
      Switch over to the new hierarchical libraries
      ---------------------------------------------
      
      This commit reorganises our libraries to use the new hierarchical
      module namespace extension.
      
      The basic story is this:
      
         - fptools/libraries contains the new hierarchical libraries.
           Everything in here is "clean", i.e. most deprecated stuff has
           been removed.
      
      	- fptools/libraries/base is the new base package
      	  (replacing "std") and contains roughly what was previously
      	  in std, lang, and concurrent, minus deprecated stuff.
      	  Things that are *not allowed* in libraries/base include:
      		Addr, ForeignObj, ByteArray, MutableByteArray,
      		_casm_, _ccall_, ``'', PrimIO
      
      	  For ByteArrays and MutableByteArrays we use UArray and
      	  STUArray/IOUArray respectively now.
      
      	  Modules previously called PrelFoo are now under
      	  fptools/libraries/GHC.  eg. PrelBase is now GHC.Base.
      
      	- fptools/libraries/haskell98 provides the Haskell 98 std.
      	  libraries (Char, IO, Numeric etc.) as a package.  This
      	  package is enabled by default.
      
      	- fptools/libraries/network is a rearranged version of
      	  the existing net package (the old package net is still
      	  available; see below).
      
      	- Other packages will migrate to fptools/libraries in
      	  due course.
      
           NB. you need to checkout fptools/libraries as well as
           fptools/hslibs now.  The nightly build scripts will need to be
           tweaked.
      
         - fptools/hslibs still contains (almost) the same stuff as before.
           Where libraries have moved into the new hierarchy, the hslibs
           version contains a "stub" that just re-exports the new version.
           The idea is that code will gradually migrate from fptools/hslibs
           into fptools/libraries as it gets cleaned up, and in a version or
           two we can remove the old packages altogether.
      
         - I've taken the opportunity to make some changes to the build
           system, ripping out the old hslibs Makefile stuff from
           mk/target.mk; the new package building Makefile code is in
           mk/package.mk (auto-included from mk/target.mk).
      
           The main improvement is that packages now register themselves at
           make boot time using ghc-pkg, and the monolithic package.conf
           in ghc/driver is gone.
      
           I've updated the standard packages but haven't tested win32,
           graphics, xlib, object-io, or OpenGL yet.  The Makefiles in
           these packages may need some further tweaks, and they'll need
           pkg.conf.in files added.
      
         - Unfortunately all this rearrangement meant I had to bump the
           interface-file version and create a bunch of .hi-boot-6 files :-(
      2cc5b907
  3. 04 Jan, 2002 1 commit
  4. 14 Dec, 2001 1 commit
    • simonpj's avatar
      [project @ 2001-12-14 17:24:03 by simonpj] · 5f087cf4
      simonpj authored
      -------------------------
      	Performance tuning things
      	-------------------------
      
      I did some nofib tests, and fixed a number of performance problems.
      
      1.  Things were getting floated to top level, and that prevented
      some useful fusion happening.
      	y = build g
      	x = foldr k z y
      
      Fixed by arranging that we only get really keen on floating to top
      level in the second run of the let-float-out pass.
      
      
      2.  Some fettling up on the let-floater itself.  It had some parameters
      that weren't even being used!  And it was stupidly floating things out
      of a one-shot lambda, and the float-in pass didn't float them back in.
      I think I fixed both of these problems.
      
      
      3.  The eta-reducer was not eta-reducing (/\a -> g a) to g.  In general
      it has to be a bit careful because "seq" means that (\x -> g x) is
      not in general the same as g ---- but it *is* the same for a type lambda.
      
      This turned out to be important in rule matching, where the foldr/build
      rule was not firing because the LHS of the rule looked like
      	foldr k z (/\ a -> g a) = ...
      which never matched!  Result, no fusion to speak of!
      
      
      4.  The simplifier was a bit too gung ho about inlining used-once
      things bound to constructor args.  The comment is with Simplify.simplNonRecX.
      5f087cf4
  5. 22 Nov, 2001 1 commit
  6. 19 Nov, 2001 1 commit
  7. 31 Oct, 2001 1 commit
    • simonpj's avatar
      [project @ 2001-10-31 15:22:53 by simonpj] · 61bfd5dd
      simonpj authored
      ------------------------------------------
      	Improved handling of scoped type variables
      	------------------------------------------
      
      The main effect of this commit is to allow scoped type variables
      in pattern bindings, thus
      
      	(x::a, y::b) = e
      
      This was illegal, but now it's ok.  a and b have the same scope
      as x and y.
      
      
      On the way I beefed up the info inside a type variable
      (TcType.TyVarDetails; c.f. IdInfo.GlobalIdDetails) which
      helps to improve error messages. Hence the wide ranging changes.
      Pity about the extra loop from Var to TcType, but can't be helped.
      61bfd5dd
  8. 22 Oct, 2001 1 commit
  9. 26 Sep, 2001 1 commit
    • simonpj's avatar
      [project @ 2001-09-26 16:19:28 by simonpj] · 6858f7c1
      simonpj authored
      ------------------
      		Simon's big commit
      		------------------
      	[ These files seem to have been left out for some reason ]
      
      
      This commit, which I don't think I can sensibly do piecemeal, consists
      of the things I've been doing recently, mainly directed at making
      Manuel, George, and Marcin happier with RULES.
      
      
      Reogranise the simplifier
      ~~~~~~~~~~~~~~~~~~~~~~~~~
      1. The simplifier's environment is now an explicit parameter.  This
      makes it a bit easier to figure out where it is going.
      
      2. Constructor arguments can now be arbitrary expressions, except
      when the application is the RHS of a let(rec).  This makes it much
      easier to match rules like
      
      	RULES
      	    "foo"  f (h x, g y) = f' x y
      
      In the simplifier, it's Simplify.mkAtomicArgs that ANF-ises a
      constructor application where necessary.  In the occurrence analyser,
      there's a new piece of context info (OccEncl) to say whether a
      constructor app is in a place where it should be in ANF.  (Unless
      it knows this it'll give occurrence info which will inline the
      argument back into the constructor app.)
      
      3. I'm experimenting with doing the "float-past big lambda" transformation
      in the full laziness pass, rather than mixed in with the simplifier (was
      tryRhsTyLam).
      
      4.  Arrange that
      	case (coerce (S,T) (x,y)) of ...
      will simplify.  Previous it didn't.
      A local change to CoreUtils.exprIsConApp_maybe.
      
      5. Do a better job in CoreUtils.exprEtaExpandArity when there's an
      error function in one branch.
      
      
      Phase numbers, RULES, and INLINE pragmas
      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      1.  Phase numbers decrease from N towards zero (instead of increasing).
      This makes it easier to add new earlier phases, which is what users want
      to do.
      
      2.  RULES get their own phase number, N, and are disabled in phases before N.
      
      e.g. 	{-# RULES "foo" [2] forall x y.  f (x,y) = f' x y #-}
      
      Note the [2], which says "only active in phase 2 and later".
      
      3.  INLINE and NOINLINE pragmas have a phase number to.  This is now treated
      in just the same way as the phase number on RULE; that is, the Id is not inlined
      in phases earlier than N.  In phase N and later the Id *may* be inlined, and
      here is where INLINE and NOINLINE differ: INLNE makes the RHS look small, so
      as soon as it *may* be inlined it probably *will* be inlined.
      
      The syntax of the phase number on an INLINE/NOINLINE pragma has changed to be
      like the RULES case (i.e. in square brackets).  This should also make sure
      you examine all such phase numbers; many will need to change now the numbering
      is reversed.
      
      Inlining Ids is no longer affected at all by whether the Id appears on the
      LHS of a rule.  Now it's up to the programmer to put a suitable INLINE/NOINLINE
      pragma to stop it being inlined too early.
      
      
      Implementation notes:
      
      *  A new data type, BasicTypes.Activation says when a rule or inline pragma
      is active.   Functions isAlwaysActive, isNeverActive, isActive, do the
      obvious thing (all in BasicTypes).
      
      * Slight change in the SimplifierSwitch data type, which led to a lot of
      simplifier-specific code moving from CmdLineOpts to SimplMonad; a Good Thing.
      
      * The InlinePragma in the IdInfo of an Id is now simply an Activation saying
      when the Id can be inlined.  (It used to be a rather bizarre pair of a
      Bool and a (Maybe Phase), so this is much much easier to understand.)
      
      * The simplifier has a "mode" environment switch, replacing the old
      black list.  Unfortunately the data type decl has to be in
      CmdLineOpts, because it's an argument to the CoreDoSimplify switch
      
          data SimplifierMode = SimplGently | SimplPhase Int
      
      Here "gently" means "no rules, no inlining".   All the crucial
      inlining decisions are now collected together in SimplMonad
      (preInlineUnconditionally, postInlineUnconditionally, activeInline,
      activeRule).
      
      
      Specialisation
      ~~~~~~~~~~~~~~
      1.  Only dictionary *functions* are made INLINE, not dictionaries that
      have no parameters.  (This inline-dictionary-function thing is Marcin's
      idea and I'm still not sure whether it's a good idea.  But it's definitely
      a Bad Idea when there are no arguments.)
      
      2.  Be prepared to specialise an INLINE function: an easy fix in
      Specialise.lhs
      
      But there is still a problem, which is that the INLINE wins
      at the call site, so we don't use the specialised version anyway.
      I'm still unsure whether it makes sense to SPECIALISE something
      you want to INLINE.
      
      
      
      
      
      Random smaller things
      ~~~~~~~~~~~~~~~~~~~~~~
      
      * builtinRules (there was only one, but may be more) in PrelRules are now
        incorporated.   They were being ignored before...
      
      * OrdList.foldOL -->  OrdList.foldrOL, OrdList.foldlOL
      
      * Some tidying up of the tidyOpenTyVar, tidyTyVar functions.  I've
        forgotten exactly what!
      6858f7c1
  10. 14 Sep, 2001 1 commit
    • simonpj's avatar
      [project @ 2001-09-14 15:51:41 by simonpj] · 5ab261bb
      simonpj authored
      --------------------------
      	Add a rule-check pass
      	(special request by Manuel)
      	--------------------------
      
      	DO NOT merge with stable
      
      The flag
      
      	-frule-check foo
      
      will report all sites at which RULES whose name starts with "foo.."
      might apply, but in fact the arguments don't match so the rule
      doesn't apply.
      
      The pass is run right after all the core-to-core passes.  (Next thing
      to do: make the core-to-core script external, so you can fiddle with
      it.  Meanwhile, the core-to-core script is in
      	DriverState.builCoreToDo
      so you can move the CoreDoRuleCheck line around if you want.
      
      The format of the report is experimental: Manuel, feel free to fiddle
      with it.
      
      Most of the code is in specialise/Rules.lhs
      
      
      Incidental changes
      ~~~~~~~~~~~~~~~~~~
      Change BuiltinRule so that the rule name is accessible
      without actually successfully applying the rule.  This
      change affects quite a few files in a trivial way.
      5ab261bb
  11. 19 Jul, 2001 1 commit
  12. 18 May, 2001 1 commit
    • simonpj's avatar
      [project @ 2001-05-18 08:46:18 by simonpj] · b4775e5e
      simonpj authored
      -----------------------------
      	Get unbox-strict-fields right
      	-----------------------------
      
      The problem was that when a library was compiled *without* -funbox-strict-fields,
      and the main program was compiled *with* that flag, we were wrongly treating
      the fields of imported data types as unboxed.
      
      To fix this I added an extra constructor to StrictnessMark to express whether
      the "!" annotation came from an interface file (don't fiddle) or a source
      file (decide whether to unbox).
      
      On the way I tided things up:
      
      * StrictnessMark moves to Demand.lhs, and doesn't have the extra DataCon
        fields that kept it in DataCon before.
      
      * HsDecls.BangType has one constructor, not three, with a StrictnessMark field.
      
      * DataCon keeps track of its strictness signature (dcRepStrictness), but not
        its "user strict marks" (which were never used)
      
      * All the functions, like getUniquesDs, that used to take an Int saying how
        many uniques to allocate, now return an infinite list. This saves arguments
        and hassle.  But it involved touching quite a few files.
      
      * rebuildConArgs takes a list of Uniques to use as its unique supply.  This
        means I could combine DsUtils.rebuildConArgs with MkId.rebuildConArgs
        (hooray; the main point of the previous change)
      
      
      I also tidied up one or two error messages
      b4775e5e
  13. 07 Dec, 2000 1 commit
    • simonpj's avatar
      [project @ 2000-12-07 09:28:42 by simonpj] · 0b62f53e
      simonpj authored
      Do a better job of eta expansion.
      
      This showed up in one of Manuel's programs, where he got code like:
      
          $wsimpleGen
      	     ww
      	     (\ i :: Int ->
      		  case i of wild1 { I# i# ->
      		  case w of wild2 { I# e# ->
      		  __coerce (ST RealWorld ())
      		  (\ s# :: (State# RealWorld) ->
      		       case writeIntArray# @ RealWorld mba# i# e# s#
      		       of s2#1 { __DEFAULT ->
      		       (# s2#1, () #)
      		       })
      		  }
      		  })
      	     s2#
      
      The argument wasn't eta expanded, so it got right through to
      the code generator as two separte lambdas.
      
      Needless to say, I fiddled around with things in a vain attempt
      to tidy them up.  Yell if anything seems to go wrong, or perfomance
      drops on any programs.
      0b62f53e
  14. 25 Oct, 2000 1 commit
  15. 19 Oct, 2000 1 commit
  16. 07 Sep, 2000 1 commit
    • simonpj's avatar
      [project @ 2000-09-07 16:32:23 by simonpj] · 4e6d5798
      simonpj authored
      A list of simplifier-related stuff, triggered
      	by looking at GHC's performance.
      
      	I don't guarantee that this lot will lead to
      	a uniform improvement over 4.08, but it it should
      	be a bit better.  More work probably required.
      
      
      * Make the simplifier's Stop continuation record whether the expression being
        simplified is the RHS of a thunk, or (say) the body of a lambda or case RHS.
        In the thunk case we want to be a bit keener about inlining if the type of
        the thunk is amenable to update in place.
      
      * Fix interestingArg, which was being too liberal, and hence doing
        too much inlining.
      
      * Extended CoreUtils.exprIsCheap to make two more things cheap:
          - 	case (coerce x) of ...
          -   let x = y +# z
        This makes a bit more eta expansion happen.  It was provoked by
        a program of Marcin's.
      
      * MkIface.ifaceBinds.   Make sure that we emit rules for things
        (like class operations) that don't get a top-level binding in the
        interface file.  Previously such rules were silently forgotten.
      
      * Move transformRhs to *after* simplification, which makes it a
        little easier to do, and means that the arity it computes is
        readily available to completeBinding.  This gets much better
        arities.
      
      * Do coerce splitting in completeBinding. This gets good code for
      	newtype CInt = CInt Int
      
      	test:: CInt -> Int
      	test x = case x of
      	      	   1 -> 2
      	      	   2 -> 4
      	      	   3 -> 8
      	      	   4 -> 16
      	      	   _ -> 0
      
      * Modify the meaning of "arity" so that during compilation it means
        "if you apply this function to fewer args, it will do virtually
        no work".   So, for example
      	f = coerce t (\x -> e)
        has arity at least 1.  When a function is exported, it's arity becomes
        the number of exposed, top-level lambdas, which is subtly different.
        But that's ok.
      
        I removed CoreUtils.exprArity altogether: it looked only at the exposed
        lambdas.  Instead, we use exprEtaExpandArity exclusively.
      
        All of this makes I/O programs work much better.
      4e6d5798
  17. 31 Aug, 2000 1 commit
  18. 01 Aug, 2000 1 commit
    • simonpj's avatar
      [project @ 2000-08-01 09:08:25 by simonpj] · fe69f3c1
      simonpj authored
      Simon's Marktoberdorf Commits
      
      1.  Tidy up the renaming story for "system binders", such as
      dictionary functions, default methods, constructor workers etc.  These
      are now documented in HsDecls.  The main effect of the change, apart
      from tidying up, is to make the *type-checker* (instead of the
      renamer) generate names for dict-funs and default-methods.  This is
      good because Sergei's generic-class stuff generates new classes at
      typecheck time.
      
      
      2.  Fix the CSE pass so it does not require the no-shadowing invariant.
      Keith discovered that the simplifier occasionally returns a result
      with shadowing.  After much fiddling around (which has improved the
      code in the simplifier a bit) I found that it is nearly impossible to
      arrange that it really does do no-shadowing.  So I gave up and fixed
      the CSE pass (which is the only one to rely on it) instead.
      
      
      3. Fix a performance bug in the simplifier.  The change is in
      SimplUtils.interestingArg.  It computes whether an argment should 
      be considered "interesting"; if a function is applied to an interesting
      argument, we are more likely to inline that function.
      Consider this case
      	let x = 3 in f x
      The 'x' argument was considered "uninteresting" for a silly reason.
      Since x only occurs once, it was unconditionally substituted, but
      interestingArg didn't take account of that case.  Now it does.
      
      I also made interestingArg a bit more liberal.  Let's see if we
      get too much inlining now.
      
      
      4.  In the occurrence analyser, we were choosing a bad loop breaker.
      Here's the comment that's now in OccurAnal.reOrderRec
      
          score ((bndr, rhs), _, _)
      	| exprIsTrivial rhs 	   = 3	-- Practically certain to be inlined
      		-- Used to have also: && not (isExportedId bndr)
      		-- But I found this sometimes cost an extra iteration when we have
      		--	rec { d = (a,b); a = ...df...; b = ...df...; df = d }
      		-- where df is the exported dictionary. Then df makes a really
      		-- bad choice for loop breaker
      
      I also increased the score for bindings with a non-functional type, so that
      dictionaries have a better chance of getting inlined early
      
      
      5. Add a hash code to the InScopeSet (and make it properly abstract)
      This should make uniqAway a lot more robust.  Simple experiments suggest
      that uniqAway no longer gets into the long iteration chains that it used
      to.
      
      
      6.  Fix a bug in the inliner that made the simplifier tend to get into
      a loop where it would keep iterating ("4 iterations, bailing out" message).
      In SimplUtils.mkRhsTyLam we float bindings out past a big lambda, thus:
      	x = /\ b -> let g = \x -> f x x
      		    in E
      becomes
      	g* = /\a -> \x -> f x x
      	x = /\ b -> let g = g* b in E
      	
      It's essential that we don't simply inling g* back into the RHS of g,
      else we will be back to square 1.  The inliner is meant not to do this
      because there's no benefit to the inlining, but the size calculation
      was a little off in CoreUnfold.
      
      
      7.  In SetLevels we were bogus-ly building a Subst with an empty in-scope
      set, so a WARNING popped up when compiling some modules.  (knights/ChessSetList
      was the example that tickled it.)  Now in fact the warning wasn't an error,
      but the Right Thing to do is to carry down a proper Subst in SetLevels, so
      that is what I have now done.  It is very little more expensive.
      fe69f3c1
  19. 14 Jul, 2000 1 commit
    • simonpj's avatar
      [project @ 2000-07-14 08:17:36 by simonpj] · 77a8c0db
      simonpj authored
      This commit completely re-does the kind-inference mechanism.
      Previously it was inter-wound with type inference, but that was
      always hard to understand, and it finally broke when we started
      checking for ambiguity when type-checking a type signature (details
      irrelevant).
      
      So now kind inference is more clearly separated, so that it never
      takes place at the same time as type inference.  The biggest change
      is in TcTyClsDecls, which does the kind inference for a group of
      type and class declarations.  It now contains comments to explain
      how it all works.
      
      There are also comments in TypeRep which describes the slightly
      tricky way in which we deal with the fact that kind 'type' (written
      '*') actually has 'boxed type' and 'unboxed type' as sub-kinds.
      The whole thing is a bit of a hack, because we don't really have 
      sub-kinding, but it's less of a hack than before.
      
      A lot of general tidying up happened at the same time.
      In particular, I removed some dead code here and there
      77a8c0db
  20. 11 Jul, 2000 1 commit
  21. 23 May, 2000 1 commit
    • simonpj's avatar
      [project @ 2000-05-23 11:35:36 by simonpj] · bb91427f
      simonpj authored
      *** MERGE WITH 4.07 (once I've checked it works) ***
      
      * Fix result type signatures.  Note that a consequential change is that
        an ordinary binding with a variable on the left
      	f = e
        is now treated as a FunMonoBind, not a PatMonoBind.  This makes
        a few things a bit simpler (eg rnMethodBinds)
      
      * Fix warnings for unused imports.  This meant moving where provenances
        are improved in RnNames.  Move mkExportAvails from RnEnv to RnNames.
      
      * Print module names right (small change in Module.lhs and Rename.lhs)
      
      * Remove a few unused bindings
        
      * Add a little hack to let us print info about join points that turn
        out not to be let-no-escaped.  The idea is to call them "$j" and report
        any such variables that are not let-no-escaped.
      
      * Some small things aiming towards -ddump-types (harmless but incomplete)
      bb91427f
  22. 19 Apr, 2000 1 commit
  23. 23 Mar, 2000 1 commit
    • simonpj's avatar
      [project @ 2000-03-23 17:45:17 by simonpj] · 111cee3f
      simonpj authored
      This utterly gigantic commit is what I've been up to in background
      mode in the last couple of months.  Originally the main goal
      was to get rid of Con (staturated constant applications)
      in the CoreExpr type, but one thing led to another, and I kept
      postponing actually committing.   Sorry.
      
      	Simon, 23 March 2000
      
      
      I've tested it pretty thoroughly, but doubtless things will break.
      
      Here are the highlights
      
      * Con is gone; the CoreExpr type is simpler
      * NoRepLits have gone
      * Better usage info in interface files => less recompilation
      * Result type signatures work
      * CCall primop is tidied up
      * Constant folding now done by Rules
      * Lots of hackery in the simplifier
      * Improvements in CPR and strictness analysis
      
      Many bug fixes including
      
      * Sergey's DoCon compiles OK; no loop in the strictness analyser
      * Volker Wysk's programs don't crash the CPR analyser
      
      I have not done much on measuring compilation times and binary sizes;
      they could have got worse.  I think performance has got significantly
      better, though, in most cases.
      
      
      Removing the Con form of Core expressions
      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      The big thing is that
      
        For every constructor C there are now *two* Ids:
      
      	C is the constructor's *wrapper*. It evaluates and unboxes arguments
      	before calling $wC.  It has a perfectly ordinary top-level defn
      	in the module defining the data type.
      
      	$wC is the constructor's *worker*.  It is like a primop that simply
      	allocates and builds the constructor value.  Its arguments are the
      	actual representation arguments of the constructor.
      	Its type may be different to C, because:
      		- useless dict args are dropped
      		- strict args may be flattened
      
        For every primop P there is *one* Id, its (curried) Id
      
        Neither contructor worker Id nor the primop Id have a defminition anywhere.
        Instead they are saturated during the core-to-STG pass, and the code generator
        generates code for them directly. The STG language still has saturated
        primops and constructor applications.
      
      * The Const type disappears, along with Const.lhs.  The literal part
        of Const.lhs reappears as Literal.lhs.  Much tidying up in here,
        to bring all the range checking into this one module.
      
      * I got rid of NoRep literals entirely.  They just seem to be too much trouble.
      
      * Because Con's don't exist any more, the funny C { args } syntax
        disappears from inteface files.
      
      
      Parsing
      ~~~~~~~
      * Result type signatures now work
      	f :: Int -> Int = \x -> x
      	-- The Int->Int is the type of f
      
      	g x y :: Int = x+y
      	-- The Int is the type of the result of (g x y)
      
      
      Recompilation checking and make
      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      * The .hi file for a modules is not touched if it doesn't change.  (It used to
        be touched regardless, forcing a chain of recompilations.)  The penalty for this
        is that we record exported things just as if they were mentioned in the body of
        the module.  And the penalty for that is that we may recompile a module when
        the only things that have changed are the things it is passing on without using.
        But it seems like a good trade.
      
      * -recomp is on by default
      
      Foreign declarations
      ~~~~~~~~~~~~~~~~~~~~
      * If you say
      	foreign export zoo :: Int -> IO Int
        then you get a C produre called 'zoo', not 'zzoo' as before.
        I've also added a check that complains if you export (or import) a C
        procedure whose name isn't legal C.
      
      
      Code generation and labels
      ~~~~~~~~~~~~~~~~~~~~~~~~~~
      * Now that constructor workers and wrappers have distinct names, there's
        no need to have a Foo_static_closure and a Foo_closure for constructor Foo.
        I nuked the entire StaticClosure story.  This has effects in some of
        the RTS headers (i.e. s/static_closure/closure/g)
      
      
      Rules, constant folding
      ~~~~~~~~~~~~~~~~~~~~~~~
      * Constant folding becomes just another rewrite rule, attached to the Id for the
        PrimOp.   To achieve this, there's a new form of Rule, a BuiltinRule (see CoreSyn.lhs).
        The prelude rules are in prelude/PrelRules.lhs, while simplCore/ConFold.lhs has gone.
      
      * Appending of constant strings now works, using fold/build fusion, plus
        the rewrite rule
      	unpack "foo" c (unpack "baz" c n)  =  unpack "foobaz" c n
        Implemented in PrelRules.lhs
      
      * The CCall primop is tidied up quite a bit.  There is now a data type CCall,
        defined in PrimOp, that packages up the info needed for a particular CCall.
        There is a new Id for each new ccall, with an big "occurrence name"
      	{__ccall "foo" gc Int# -> Int#}
        In interface files, this is parsed as a single Id, which is what it is, really.
      
      Miscellaneous
      ~~~~~~~~~~~~~
      * There were numerous places where the host compiler's
        minInt/maxInt was being used as the target machine's minInt/maxInt.
        I nuked all of these; everything is localised to inIntRange and inWordRange,
        in Literal.lhs
      
      * Desugaring record updates was broken: it didn't generate correct matches when
        used withe records with fancy unboxing etc.  It now uses matchWrapper.
      
      * Significant tidying up in codeGen/SMRep.lhs
      
      * Add __word, __word64, __int64 terminals to signal the obvious types
        in interface files.  Add the ability to print word values in hex into
        C code.
      
      * PrimOp.lhs is no longer part of a loop.  Remove PrimOp.hi-boot*
      
      
      Types
      ~~~~~
      * isProductTyCon no longer returns False for recursive products, nor
        for unboxed products; you have to test for these separately.
        There's no reason not to do CPR for recursive product types, for example.
        Ditto splitProductType_maybe.
      
      Simplification
      ~~~~~~~~~~~~~~~
      * New -fno-case-of-case flag for the simplifier.  We use this in the first run
        of the simplifier, where it helps to stop messing up expressions that
        the (subsequent) full laziness pass would otherwise find float out.
        It's much more effective than previous half-baked hacks in inlining.
      
        Actually, it turned out that there were three places in Simplify.lhs that
        needed to know use this flag.
      
      * Make the float-in pass push duplicatable bindings into the branches of
        a case expression, in the hope that we never have to allocate them.
        (see FloatIn.sepBindsByDropPoint)
      
      * Arrange that top-level bottoming Ids get a NOINLINE pragma
        This reduced gratuitous inlining of error messages.
        But arrange that such things still get w/w'd.
      
      * Arrange that a strict argument position is regarded as an 'interesting'
        context, so that if we see
      	foldr k z (g x)
        then we'll be inclined to inline g; this can expose a build.
      
      * There was a missing case in CoreUtils.exprEtaExpandArity that meant
        we were missing some obvious cases for eta expansion
        Also improve the code when handling applications.
      
      * Make record selectors (identifiable by their IdFlavour) into "cheap" operations.
      	  [The change is a 2-liner in CoreUtils.exprIsCheap]
        This means that record selection may be inlined into function bodies, which
        greatly improves the arities of overloaded functions.
      
      * Make a cleaner job of inlining "lone variables".  There was some distributed
        cunning, but I've centralised it all now in SimplUtils.analyseCont, which
        analyses the context of a call to decide whether it is "interesting".
      
      * Don't specialise very small functions in Specialise.specDefn
        It's better to inline it.  Rather like the worker/wrapper case.
      
      * Be just a little more aggressive when floating out of let rhss.
        See comments with Simplify.wantToExpose
        A small change with an occasional big effect.
      
      * Make the inline-size computation think that
      	case x of I# x -> ...
        is *free*.
      
      
      CPR analysis
      ~~~~~~~~~~~~
      * Fix what was essentially a bug in CPR analysis.  Consider
      
      	letrec f x = let g y = let ... in f e1
      		     in
      		     if ... then (a,b) else g x
      
        g has the CPR property if f does; so when generating the final annotated
        RHS for f, we must use an envt in which f is bound to its final abstract
        value.  This wasn't happening.  Instead, f was given the CPR tag but g
        wasn't; but of course the w/w pass gives rotten results in that case!!
        (Because f's CPR-ness relied on g's.)
      
        On they way I tidied up the code in CprAnalyse.  It's quite a bit shorter.
      
        The fact that some data constructors return a constructed product shows
        up in their CPR info (MkId.mkDataConId) not in CprAnalyse.lhs
      
      
      
      Strictness analysis and worker/wrapper
      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      * BIG THING: pass in the demand to StrictAnal.saExpr.  This affects situations
        like
      	f (let x = e1 in (x,x))
        where f turns out to have strictness u(SS), say.  In this case we can
        mark x as demanded, and use a case expression for it.
      
        The situation before is that we didn't "know" that there is the u(SS)
        demand on the argument, so we simply computed that the body of the let
        expression is lazy in x, and marked x as lazily-demanded.  Then even after
        f was w/w'd we got
      
      	let x = e1 in case (x,x) of (a,b) -> $wf a b
      
        and hence
      
      	let x = e1 in $wf a b
      
        I found a much more complicated situation in spectral/sphere/Main.shade,
        which improved quite a bit with this change.
      
      * Moved the StrictnessInfo type from IdInfo to Demand.  It's the logical
        place for it, and helps avoid module loops
      
      * Do worker/wrapper for coerces even if the arity is zero.  Thus:
      	stdout = coerce Handle (..blurg..)
        ==>
      	wibble = (...blurg...)
      	stdout = coerce Handle wibble
        This is good because I found places where we were saying
      	case coerce t stdout of { MVar a ->
      	...
      	case coerce t stdout of { MVar b ->
      	...
        and the redundant case wasn't getting eliminated because of the coerce.
      111cee3f
  24. 06 Jan, 2000 1 commit
  25. 01 Nov, 1999 1 commit
    • simonpj's avatar
      [project @ 1999-11-01 17:09:54 by simonpj] · 30b5ebe4
      simonpj authored
      A regrettably-gigantic commit that puts in place what Simon PJ
      has been up to for the last month or so, on and off.
      
      The basic idea was to restore unfoldings to *occurrences* of
      variables without introducing a space leak.  I wanted to make
      sure things improved relative to 4.04, and that proved depressingly
      hard.  On the way I discovered several quite serious bugs in the
      simplifier.
      
      Here's a summary of what's gone on.
      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      * No commas between for-alls in RULES.  This makes the for-alls have
        the same syntax as in types.
      
      * Arrange that simplConArgs works in one less pass than before.
        This exposed a bug: a bogus call to completeBeta.
      
      * Add a top-level flag in CoreUnfolding, used in callSiteInline
      
      * Extend w/w to use etaExpandArity, so it does eta/coerce expansion
      
      * Implement inline phases.   The meaning of the inline pragmas is
        described in CoreUnfold.lhs.  You can say things like
      	{#- INLINE 2 build #-}
        to mean "inline build in phase 2"
      
      * Don't float anything out of an INLINE.
        Don't float things to top level unless they also escape a value lambda.
      	[see comments with SetLevels.lvlMFE
        Without at least one of these changes, I found that
      	{-# INLINE concat #-}
      	concat = __inline (/\a -> foldr (++) [])
        was getting floated to
      	concat = __inline( /\a -> lvl a )
      	lvl = ...inlined version of foldr...
      
        Subsequently I found that not floating constants out of an INLINE
        gave really bad code like
      	__inline (let x = e in \y -> ...)
        so I now let things float out of INLINE
      
      * Implement the "reverse-mapping" idea for CSE; actually it turned out to be easier
        to implement it in SetLevels, and may benefit full laziness too.
      
      * It's a good idea to inline inRange. Consider
      
      	index (l,h) i = case inRange (l,h) i of
      		  	  True ->  l+i
      			  False -> error
        inRange itself isn't strict in h, but if it't inlined then 'index'
        *does* become strict in h.  Interesting!
      
      * Big change to the way unfoldings and occurrence info is propagated in the simplifier
        The plan is described in Subst.lhs with the Subst type
        Occurrence info is now in a separate IdInfo field than user pragmas
      
      * I found that
      	(coerce T (coerce S (\x.e))) y
        didn't simplify in one round. First we get to
      	(\x.e) y
        and only then do the beta. Solution: cancel the coerces in the continuation
      
      * Amazingly, CoreUnfold wasn't counting the cost of a function an application.
      
      * Disable rules in initial simplifier run.  Otherwise full laziness
        doesn't get a chance to lift out a MFE before a rule (e.g. fusion)
        zaps it.  queens is a case in point
      
      * Improve float-out stuff significantly.  The big change is that if we have
      
      	\x -> ... /\a -> ...let p = ..a.. in let q = ...p...
      
        where p's rhs doesn't x, we abstract a from p, so that we can get p past x.
        (We did that before.)  But we also substitute (p a) for p in q, and then
        we can do the same thing for q.  (We didn't do that, so q got stuck.)
        This is much better.  It involves doing a substitution "as we go" in SetLevels,
        though.
      30b5ebe4
  26. 17 Sep, 1999 1 commit
    • simonpj's avatar
      [project @ 1999-09-17 09:15:22 by simonpj] · 731f53de
      simonpj authored
      This bunch of commits represents work in progress on inlining and
      worker/wrapper stuff.
      
      Currently, I think it makes the compiler slightly worse than 4.04, for
      reasons I don't yet understand.  But it means that Simon and I can
      both peer at what is going on.
      
      * Substantially improve handling of coerces in worker/wrapper
      
      * exprIsDupable for an application (f e1 .. en) wasn't calling exprIsDupable
        on the arguments!!  So applications with few, but large, args were being dupliated.
      
      * sizeExpr on an application wasn't doing a nukeScrutDiscount on the arg of
        an application!!  So bogus discounts could accumulate from arguments!
      
      * Improve handling of INLINE pragmas in calcUnfoldingGuidance.  It was really
        wrong before
      731f53de
  27. 14 Jul, 1999 1 commit
    • simonpj's avatar
      [project @ 1999-07-14 14:40:20 by simonpj] · 4e7d56fd
      simonpj authored
      Main things:
      
      * Add splitProductType_maybe to DataCon.lhs, with type
        splitProductType_maybe
      	:: Type 			-- A product type, perhaps
      	-> Maybe (TyCon, 		-- The type constructor
      		  [Type],		-- Type args of the tycon
      		  DataCon,		-- The data constructor
      		  [Type])		-- Its *representation* arg types
      
        Then use it in many places (e.g. worker-wrapper places) instead
        of a pile of junk
      
      * Clean up various uses of dataConArgTys, which were plain wrong because
        they weren't passed the existential type arguments.  Most of these calls
        are eliminated by using splitProductType_maybe above.  I hope I correctly
        squashed the others. This fixes a bug that Meurig's programs showed up.
      
          module FailGHC (killSustainer) where
          import Weak
          import IOExts
      
          data Sustainer = forall a . Sustainer (IORef (Maybe a)) (IO ())
      
          killSustainer :: Sustainer -> IO ()
          killSustainer (Sustainer _ act) = act
      
        The above program used to kill the compiler.
      
      * A fairly concerted attack on the Dreaded Space Leak.
      	- Add Type.seqType, CoreSyn.seqExpr, CoreSyn.seqRules
      
      	- Add some seq'ing when building Ids and IdInfos
      		These reduce the space usage a lot
      
      	- Add CoreSyn.coreBindsSize, which is pretty strict in the program,
      		and call it when we have -dshow-passes.
      
      	- Do not put the inlining in an Id that is being plugged into
      		the result-expression of the simplifier.  This cures
      		a the 'wedge' in the space profile for reasons I don't understand fully
      
        Together, these things reduce the max space usage when compiling PrelNum from
        17M to about 7Mbytes.
      
        I think there are now *too many* seqs, and they waste work, but I don't have
        time to find which ones.
      
        Furthermore, we aren't done. For some reason, some of the stuff allocated by
        the simplifier makes it through all during code generation and I don't see why.
        There's a should-be-unnecessary call to coreBindsSize in Main.main which
        zaps some, but not all of this space.
      
        -dshow-passes reduces space usage a bit, but I don't think it should really.
      
        All the measurements were made on a compiler compiled with profiling by
        GHC 3.03.    I hope they carry over to other builds!
      
      * One trivial thing: changed all variables 'label' to 'lbl', becuase the
        former is a keyword with -fglagow-exts in GHC 3.03 (which I was compiling with).
        Something similar in StringBuffer.
      4e7d56fd
  28. 22 Jun, 1999 2 commits
  29. 21 May, 1999 1 commit
  30. 18 May, 1999 1 commit
  31. 18 Dec, 1998 1 commit
    • simonpj's avatar
      [project @ 1998-12-18 17:40:31 by simonpj] · 7e602b0a
      simonpj authored
      Another big commit from Simon.  Actually, the last one
      didn't all go into the main trunk; because of a CVS glitch it
      ended up in the wrong branch.
      
      So this commit includes:
      
      * Scoped type variables
      * Warnings for unused variables should work now (they didn't before)
      * Simplifier improvements:
      	- Much better treatment of strict arguments
      	- Better treatment of bottoming Ids
      	- No need for w/w split for fns that are merely strict
      	- Fewer iterations needed, I hope
      * Less gratuitous renaming in interface files and abs C
      * OccName is a separate module, and is an abstract data type
      
      I think the whole Prelude and Exts libraries compile correctly.
      Something isn't quite right about typechecking existentials though.
      7e602b0a
  32. 02 Dec, 1998 1 commit
  33. 19 Mar, 1998 1 commit
  34. 12 Mar, 1998 1 commit
  35. 08 Mar, 1998 1 commit
  36. 08 Jan, 1998 1 commit
    • simonm's avatar
      [project @ 1998-01-08 18:03:08 by simonm] · 9dd6e1c2
      simonm authored
      The Great Multi-Parameter Type Classes Merge.
      
      Notes from Simon (abridged):
      
      * Multi-parameter type classes are fully implemented.
      * Error messages from the type checker should be noticeably improved
      * Warnings for unused bindings (-fwarn-unused-names)
      * many other minor bug fixes.
      
      Internally there are the following changes
      
      * Removal of Haskell 1.2 compatibility.
      * Dramatic clean-up of the PprStyle stuff.
      * The type Type has been substantially changed.
      * The dictionary for each class is represented by a new
        data type for that purpose, rather than by a tuple.
      9dd6e1c2
  37. 25 Jul, 1997 1 commit
  38. 05 Jun, 1997 1 commit
  39. 26 May, 1997 1 commit