Make the register allocator deal properly with switch tables.
Previously, it didn't calculate the correct flow edges away from the
indirect jump (in fact it didn't reckon there were any flow edges
leaving it :) which makes a nonsense of the live variable analysis in
the branches.

A jump insn can now optionally be annotated with a list of destination
labels, and if so, the register allocator creates flow edges to all of
them.

Jump tables are now re-enabled.  They remain disabled for 4.08.1,
since we aren't fixing the problem properly on that branch.

I assume this problem wasn't exposed by the old register allocator
because of the live-range-approximation hacks used in it.  Since it
was undocumented, we'll never know.

Sparc builds will now break until I fix them.

look for GMP 3, not GMP 2.

Connect up the new GMP.

Our local changes to GMP 3.1: disconnect the bits we're not interested
in (mpf, mpq, tests subdirs etc.).

merge conflicts (I hope)

- add touch# and foreignObjToAddr# primops.

- add these functions to ForeignObj:

	withForeignObj  :: ForeignObj -> (Addr -> IO a) -> IO a
	touchForeignObj :: ForeignObj -> IO ()

- foreignObjToAddr, which was previously implemented using a _casm_, now
  uses the foreignObjToAddr# primop.

- replace implementations of readXXXOffForeignObj and writeXXXOffForeignObj
  using the withForeignObj and the Addr operations.  ForeignObj.lhs now has
  no _casms_ in it!  (it still can't be compiled with the NCG though, because
  it has a f.i.d.)

remove an unused import

make this file independent of the actual word size.

Fix up a PrimOp dependency.

Add touch# and foreignObjToAddr#.

- Add touch# and foreignObjToAddr#.

- Add a long comment to the beginning of this file listing the files
  that need to be updated when adding a new primop.

Added a worked example of how to export Haskell functions from a DLL
(provided by Sigbj\orn).

#include "config.h", to be able to use HsFFI.h from standalone C files.

Import stripped-down gmp-3.1.

This will cause severe breakage until I can resolve the conflicts and
check in the rest of the changes, so I'd advise not updating this
directory for a while (unless you're using a system-installed gmp, in
which case it won't matter).

Initial revision

Import stripped-down gmp-3.1.

This will cause severe breakage until I can resolve the conflicts and
check in the rest of the changes, so I'd advise not updating this
directory for a while (unless you're using a system-installed gmp, in
which case it won't matter).

Disable jump tables for the time being; the register allocator gets confused about them.

Gratuitously remove a gratuitous ineficiency in allocatableRegs.

Revoked comments about not using -static under Windows.

Typo s/Lhar/List/

Oops, #include reordering does not work if -#include refers to names
from Stg.h. Changes reverted. The original problem requires -optc-Dfoo
workaround.

Unused Prelude import check moved to a better place.

Never say:
"Warning: Module `Prelude' is imported, but nothing from it is used"

Output -#include options before packages' C includes, to be able to
-#include a file that #defines e.g. _XOPEN_SOURCE and #includes a C
header depending on _XOPEN_SOURCE that is also included from Stg.h.
First inclusion of this header should be after the #define.

Replace a zipEqual with a zip in tcPat when checking a record pat.
The former version caused a panic with DEBUG on if a record pattern
was used with a non-record constructor.

Add some tests for unboxing strict record fields.

unboxing strict fields, bug #2:

	the unfolding in a record selector id was
	completely bogus when -funbox-strict-fields
	was in effect.

Sorry, that's "RTS hooks work under Windows when the *program* is built
statically, not the compiler".

Revised note about RTS hooks under Windows: they should still work for
statically-built compilers.

Added a note about RTS hooks not working under Windows with DLLs.

remove unused imports.

barf when a DLLized RTS tries to allocate a 2nd Windows monster block;
is_heap_alloced stops working at this point. No problem for the static
build.

Changed HAVE_WIN32_DLL_SUPPORT to ENABLE_WIN32_DLL_SUPPORT so that the
relevant bits are only used in the dynamic Windows build (as they should be).

unboxing strict fields, bug #1:

	desugaring a record pattern used the
	rep args rather than the original args when
	expanding the pattern into a ConPat.

Complete the removal of tso->splim (these are the NCG bits).

This test used a library module that no longer exists (Printf).

Test no longer applicable (fails for non-typechecking reasons: multiple
definition of main)

Updated error message.

Change use of HAVE_WIN32_DLL_SUPPORT to ENABLE_WIN32_DLL_SUPPORT, to stop
things going horribly wrong with static linking on win32.

Don't just pick the back edges ({-# SOURCE #-} imports) when
collecting the list of imports for module initialisation.  It's
entirely possible that a module is only reachable from Main via {-#
SOURCE #-} imports.