GitLab

Specialise thread_static() a little bit: improves performance for
threading the static object list.

Do for evalStrictness the same as for evalAbsence

- Move FptoolsHcOpts out of the hslibs section of this file (which was
  *exactly* the wrong place for it).

- Clarify the meaning of FptoolsHcOpts and GhcHcOpts.

NOTE: $(FptoolsHcOpts) gets added to every Haskell compilation for
Haskell *programs* (not libraries), including GHC itself.  By default,
it contains only -O.  To remove the -O, you need to override
FptoolsHcOpts in build.mk, not GhcHcOpts.

GhcHcOpts is still there, and contains options to be added when
compiling GHC only.

---------------------------------
	Another long-standing infelicity!
	---------------------------------

CoreTidy was throwing away demand info on let-binders.
This meant that CorePrep would build a let (thunk) instead of
a case, even if the Id is sure to be demanded.

Easily fixed by making CoreTidy retain the demand info.

This demand-analysis stuff is having the excellent side effect
of flushing out performance bugs!

- bugfix (was erroneously ignoring the return value from scavenge_one() and
  checking failed_to_evac instead, which was always false)

- printf format fix (shut gcc up).

Change pretty-print of DmdType slightly

---------------------------
	Fix another bad DmdAnal bug
	---------------------------

The `both` operator wasn't commutative, leading to most
strange results.  In particular, the fixpoint finder went into
an infinite fip/flop loop on Marcin's program.

---------------------------------------
	Fix a grevious dict-inlining infelicity
	---------------------------------------

This commit fixes an infelicity that must have been there
for a long time.  The problem was that dictionary functions
(arising from instance declarations) were treated as GlobalIds,
and hence no taken account of when doing free variable and dependency
analysis.    As a result, the specialiser was messing up dependency
order, so the program was considerably more complex than it should
be by the time it got to the strictness analyser.

Net result:
	PrelRead.$sreduce :: Integer -> Integer -> (Integer,Integer)
didn't get detected as strict.

Easily fixed, by making DictFunIds into LocalIds (albeit exported ones).

Capture the dependency of PrelIO.hsc on PrelHandle_hsc.h (as was, 'make clean; make all' broke)

Print a bit more info in an old strictness analyser assert fail

Print a bit more info in the dmdFix loop

A major demand-analyser fix, which made it say something was
absent when it wasn't at all.  Here's the comment from the
Case equation of dmdAnal.

	-- Figure out whether the case binder is used, and use
	-- that to set the keepity of the demand.  This is utterly essential.
	-- Consider	f x = case x of y { (a,b) -> k y a }
	-- If we just take scrut_demand = U(L,A), then we won't pass x to the
	-- worker, so the worker will rebuild
	--	x = (a, absent-error)
	-- and that'll crash.

Bugfixes; take large objects into account in stats output.

wibble

Add errorCString

Make absent-arg errors more descriptive

More tweaks.  Getting usable now.

Make compacting GC faster, stage 1.

A cunning trick from Simon P.J. to avoid needing to call the
potentially expensive LOOKS_LIKE_GHC_INFO() a lot.

remove extra SRC_MKDEPENDC_OPTS, since we shouldn't be doing a
mkdependC in this directory at all.

Remove whitespace from the end of lines.  The extra whitespace is
syntactically meaningful to make(!) and was causing us to do
a superfluous mkdependC in directories without any C files.

libm is a dependency of the RTS now, so don't also depend on it for
package "std".

back the last change out; a better way to get all the hc files you need is

	SRC_HC_OPTS += -keep-hc-files

in your build.mk.

Two fixes to the demand analyis.

1.  Don't look inside Coerces.  These wrap recursive newtypes,
and we might fail to find a fixpoint if we look inside them.

2.  Polymorphic 'seq' is represented by a (Seq _ _ []) demand,
so lub/both must be prepared to take a Seq with an empty list
of demands without complaint.

Keep the printed stuff for Demand and DmdResult separate, else
we can get confused when parsing a strictness signature.  For
example, previously

	LX

could have been parsed as	L  -> X
or as				LX -> T

Urk!  I've taken the simple path of making DmdResults print in
lower case.  Thus

	Lx 	means	L  -> x
	LX	means	LX -> t

Add Ken Shan to the contributor's list for his heroic work on the
Alpha port of GHC.  Thanks Ken!

Add missing cases to newDemand

Innocent changes to resurrect/add 64-bit support.

Fix typo in StgRun() for the Alpha

#include <string.h> to get proper prototypes

Removed debugging (tracing) code.

Fixed punctuation typo.

Added -I$(GHC_INCLUDE_DIR) to $(SRC_MKDEPENDC_OPTS), so that
mkdependC can find its include files properly.

Made "nat" 64-bit ("unsigned long") rather than 32-bit ("unsigned int")
on 64-bit machines.  This fixes some problems caused by code elsewhere
assuming that sizeof(nat) == sizeof(void *).

Added -lm to the extra_libraries list for the rts package.  This is
because ldexp() is used by StgPrimFloat.c and is in -lm on alpha-osf3.

Added $(GhcHcOpts) to $(SRC_HC_OPTS), so that by setting
--keep-hc-files-too in $(GhcHcOpts) we get .hc files for the
Haskell code under this directory, ghc/driver.

Removed 32-bit dependencies in the generation and handling of
liveness mask bitmaps.  We now support both 32-bit and 64-bit
machines with identical .hc files.  Support for >64-bit machines
would be easy to add.  Note that old .hc files are incompatible
with the changes made to ghc/include/InfoMacros.h!

Made some code #ifdef GHCI, so that it will compile without GHCI enabled.

Added missing hyphen in frnot of machdepCCOpts for mips and powerpc.

Added -Xlinker -noprefix_recognition to Alpha-specific
machdepCCOpts.  We need this flag to disable the following
documented behavior of ld(1) on Digital UNIX:

   All routines that start with __init_ are treated, by default,
   as initialization routines. These are routines that are called
   without an argument when the file that contains them is loaded
   or when the program that contains them is started.

Unfortunately, this flag produces the annoying (but harmless) gcc
message

    gcc: -noprefix_recognition: linker input file unused since
    linking not done

(See also mk/bootstrap.mk)

Added -static, a flag for compiling on the Alpha that is already
listed under machdepCCOpts in DriverFlags.hs.  Without it, the
following assert in GC.c fails:

    /* make sure the info pointer is into text space */
    ASSERT(q && (LOOKS_LIKE_GHC_INFO(GET_INFO(q))
                 || IS_HUGS_CONSTR_INFO(GET_INFO(q))));

Added -Xlinker -noprefix_recognition, a flag for Alpha compilation
that I just added to machdepCCOpts in DriverFlags.hs.  We need
this flag to disable the following documented behavior of ld(1) on
Digital UNIX:

   All routines that start with __init_ are treated, by default,
   as initialization routines. These are routines that are called
   without an argument when the file that contains them is loaded
   or when the program that contains them is started.

Unfortunately, this flag produces the annoying (but harmless) gcc
message

    gcc: -noprefix_recognition: linker input file unused since
    linking not done

(See also ghc/compiler/main/DriverFlags.hs)