GitLab

This commit adds in the current state of our SMP support.  Notably,
this allows the new way 's' to be built, providing support for running
multiple Haskell threads simultaneously on top of any pthreads
implementation, the idea being to take advantage of commodity SMP
boxes.

Don't expect to get much of a speedup yet; due to the excessive
locking required to synchronise access to mutable heap objects, you'll
see a slowdown in most cases, even on a UP machine.  The best I've
seen is a 1.6-1.7 speedup on an example that did no locking (two
optimised nfibs in parallel).

	- new RTS -N flag specifies how many pthreads to start.

	- new driver -smp flag, tells the driver to use way 's'.

	- new compiler -fsmp option (not for user comsumption)
	  tells the compiler not to generate direct jumps to
	  thunk entry code.

	- largely rewritten scheduler

	- _ccall_GC is now done by handing back a "token" to the
	  RTS before executing the ccall; it should now be possible
	  to execute blocking ccalls in the current thread while
	  allowing the RTS to continue running Haskell threads as
	  normal.

	- you can only call thread-safe C libraries from a way 's'
	  build, of course.

Pthread support is still incomplete, and weird things (including
deadlocks) are likely to happen.

Increase NUM_TEXT to 100000 so okeefe_neural will parse.

reduce block size to 4k

small cleanup; reduce amount of conditionally-compiled stuff

Add foreign import/export implementations for x86 stdcall convention.

Make parser notice calling conventions on f-i and f-x declarations,
check they are supported on the platform Hugs is compiled on.  Pass
them all the way through the code generator to the interpreter.

Allow f-i/f-x decls to omit the calling convention, in which case
ccall is used.  Remove calling convention from all such decls
in the Prelude so it will work on any platform.

Micro-optimisation: don't need to test for an empty queue in a
BLACKHOLE_BQ, now that we're distinguishing BLACKHOLE and
BLACKHOLE_BQ.  Shortens the update code by a couple of instructions.

Adding final diffs between Hugs98 (Jan99) and Hugs98 (Sep99)
manually to STG Hugs.

Change CFunDescriptor to match new foreign import implementation.

Added basic support for foreign export dynamic.

Many aspects of it are still broken:
* Only supports x86-linux.
* The range of allowable types is small: Char Int Float Double
  Addr and Word.
* Adjustor thunks are never freed.
* Returning Doubles or Floats doesn't work at all.

I expect to fix some of these shortly.  foreign import also
needs redoing, so it can accept any number of arguments of
any type.

Also:
* Fixed setRtsFlags in Evaluator.c to make it endian-independent.
* Fixed raisePrim in Evaluator.c so things like division by zero,
  array index errors, etc, throw an exception instead of
  terminating StgHugs.  raisePrim is renamed makeErrorCall.

oops: unbreak non-ticky compiles

Crude allocation-counting extension to ticky-ticky profiling.

Allocations are counted against the closest lexically enclosing
function closure, so you need to map the output back to the STG code.

Heap Profiling
--------------

This commit adds simple hp2ps-style heap profiling.  Just the '-h'
option is supported so far.

Heap profiles will show the top two cost centres on each cost centre
stack.

Time/Alloc profiles have been extended to aggregate the costs on a
per-cost-centre basis and show the table of highest consuming cost
centres before the full table of cost centre stacks.

Profiles work best if '-caf-all' is used, especially when using
one of the '-auto' flags.

A couple of fixes and cleanups to ticky-ticky profiling:

	- remove UPD_EXISTING (doesn't make sense)
	- add UPD_CON_IN_PLACE, now that we have in-place updates
	- clean up the output a little.

Support for thread{WaitRead,WaitWrite,Delay}.  These should behave
identically to the 3.02 implementations.

We now have the virtual timer on during all program runs, which ticks
at 50Hz by default.  This is used to implement threadDelay, so you
won't get any better granularity than the tick frequency
unfortunately.  It remains to be seen whether using the virtual timer
will have a measurable impact on performance for non-threadDelaying
programs.

All operations in the I/O subsystem should now be non-blocking with
respect to other running Haskell threads.  It remains to be seen
whether this will have a measurable performance impact on
non-concurrent programs (probably not).

Eq and Ord instances for ThreadId.

oops, back out some other changes that made it in by mistake.

An inaccurate constraint on the inline assembly code for mulIntC#
caused gcc 2.95 to overoptimise around it, leading to bogus multiply
code for Integers.

minor cleanup: use INFO_PTR_TO_STRUCT in get_itbl() and hence pull it
out of the #ifdef TABLES_NEXT_TO_CODE.

Shift operations where the shift count is greater than the width of
the type being shifted are now defined as returning zero (or -1 in the
case of a negative integer being shifted right with iShiftRA).

More 64bit fun: Fixed an ill-placed #ifdef and added a cast to a
numeric literal (for explanation see yesterday's fix).

We have to be very careful with numeric literals in the C part, in
this case 0xff (of type int = 4 bytes on Alpha) was shifted left 56
times => always zero  :-{  Using an L or UL suffix is not always
practical, because the needed size depends on conditional typedefs
somewhere else. Solution: Cast!  :-P  ghc/includes contains some more
suspicious literals, but I did not have a look at them yet.

add NonTermination_closure.

Make vectored returns to Hugs work, and make IS_CODE_PTR etc work
for dynamically loaded objects.

Header file for an alternative bignum implementation, which is in
interpreter/sainteger.c.  Currently used by STGHugs.
Temporary -- until we resolve GMP licensing issues.

Redo previous commit to cut down on the use of COMPILING_RTS where
possible - SchedAPI.h is now an RTS internal header file which
RtsAPI.h no longer includes.

Don't include RtsAPI.h when compiling RTS.

Stg.h contains all the defs and protos that are visible when compiling
.hc code (==> .hs code). Augment this set with the externally visible
RTS API.

New RTS entry point, shutdownHaskellAndExit(), which does what the name
implies - used when you want to exit from within Haskell code (e.g.,
System.exitWith.)

Make -optc-DDEBUG work again

Comment about TABLES_NEXT_TO_CODE.

New define: TABLES_NEXT_TO_CODE says whether info tables are assumed
to reside just before the code for a function.   We used to use
USE_MINIINTERPRETER for this function, but it makes sense in certain
cases to separate the use of tail-calls from tables-next-to-code.

StgMacros: add a couple of missing macros to support update in place.

Reinstate per-function entry counts in ticky-ticky profiles.

suppress needless warning

Made rts_evalIO() stricter, i.e.,

   rts_evalIO( action );

will now essentially cause `action' to be applied
to the following (imaginary) defn of `evalIO':

    evalIO :: IO a -> IO a
    evalIO action = action >>= \ x -> x `seq` return x

instead of just

    evalIO :: IO a -> IO a
    evalIO action = action >>= \ x -> return x

The old, lazier behaviour is now available via rts_evalLazyIO().

Support for "unregisterised" builds.  An unregisterised build doesn't
use the assembly mangler, doesn't do tail jumping (uses the
mini-interpreter), and doesn't use global register variables.

Plenty of cleanups and bugfixes in the process.

Add way 'u' to GhcLibWays to get unregisterised libs & RTS.

[ note: not *quite* working fully yet... there's still a bug or two
  lurking ]

(this is number 9 of 9 commits to be applied together)

  Usage verification changes / ticky-ticky changes:

  We want to verify that SingleEntry thunks are indeed entered at most
  once.  In order to do this, -ticky / -DTICKY_TICKY turns on eager
  blackholing.  We blackhole with new blackholes: SE_BLACKHOLE and
  SE_CAF_BLACKHOLE.  We will enter one of these if we attempt to enter
  a SingleEntry thunk twice.  Note that CAFs are dealt with in by
  codeGen, and ordinary thunks by the RTS.

  We also want to see how many times we enter each Updatable thunk.
  To this end, we have modified -ticky.  When -ticky is on, we update
  with a permanent indirection, and arrange that when we enter a
  permanent indirection we count the entry and then convert the
  indirection to a normal indirection.  This gives us a means of
  counting the number of thunks entered again after the first entry.
  Obviously this screws up profiling, and so you can't build a ticky
  and profiling compiler any more.

  Also a few other changes that didn't make it into the previous 8
  commits, but form a part of this set.

(this is number 8 of 9 commits to be applied together)

  The CPP flag LAZY_BLACKHOLING has been moved up from options.h into
  Stg.h, so GHC can see it as well as the interpreter, and
  EAGER_BLACKHOLING has been added.  The default is still
  LAZY_BLACKHOLING && !EAGER_BLACKHOLING.

extend cleanup to integer2Word#

integer2Int# take 4: cleaned up a little.

integer2Int#, take 3