GitLab

Also a small refactoring in how the verbosity flag is handled.

not protected by locks in the threaded RTS.  This moves the free lists
to the individual hash tables and removes the static variables.

Spotted by Marcin Orczyk <orczykm@dcs.gla.ac.uk>, thanks!  This could
definitely cause actual crashes.

This reverts commit 2ec79623.

This assignment rewriting pass subsumes the previous reload
sinking pass, and also performs basic inlining.
Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

If we find one then we print an error message and fail.

If we aren't validating and mk/build.mk doesn't exist then boot
prints a warning, pointing at instructions for setting up mk/build.mk.

This avoids duplicating some logic around the build system.

otherwise some versions of gcc complain about a missing final newline.

The pinned_object_block is where we allocate small pinned ByteArray#
objects.  At a GC the pinned_object_block was being treated like other
large objects and promoted to the next step/generation, even if it was
only partly full.  Under some ByteString-heavy workloads this would
accumulate on average 2k of slop per GC, and this memory is never
released until the ByteArray# objects in the block are freed.

So now, we keep allocating into the pinned_object_block until it is
completely full, at which point it is handed over to the GC as before.
The pinned_object_block might therefore contain objects which a large
range of ages, but I don't think this is any worse than the situation
before.  We still have the fragmentation issue in general, but the new
scheme can improve the memory overhead for some workloads
dramatically.

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

This fixes two bugs:

    - The new code generator doesn't like procedures with
      empty graphs, and panicked in labelAGraph.

    - LLVM optimizes away empty procedures but not empty
      data sections, so now the backwards-compatibility
      labels actually work with -fllvm.
Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

This fixes #5106 (objc compilation was failing on i386 Darwin).

Renamed adjustUFM in GraphOps to adjustUFM_C, to account
for alternate argument order.
Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

symptom was 2047(threaded2) was crashing.

Previously the code generator generated small code fragments labelled
with __stginit_M for each module M, and these performed whatever
initialisation was necessary for that module and recursively invoked
the initialisation functions for imported modules.  This appraoch had
drawbacks:

 - FFI users had to call hs_add_root() to ensure the correct
   initialisation routines were called.  This is a non-standard,
   and ugly, API.

 - unless we were using -split-objs, the __stginit dependencies would
   entail linking the whole transitive closure of modules imported,
   whether they were actually used or not.  In an extreme case (#4387,
   #4417), a module from GHC might be imported for use in Template
   Haskell or an annotation, and that would force the whole of GHC to
   be needlessly linked into the final executable.

So now instead we do our initialisation with C functions marked with
__attribute__((constructor)), which are automatically invoked at
program startup time (or DSO load-time).  The C initialisers are
emitted into the stub.c file.  This means that every time we compile
with -prof or -hpc, we now get a stub file, but thanks to #3687 that
is now invisible to the user.

There are some refactorings in the RTS (particularly for HPC) to
handle the fact that initialisers now get run earlier than they did
before.

The __stginit symbols are still generated, and the hs_add_root()
function still exists (but does nothing), for backwards compatibility.

This code has accumulated a great deal of cruft over the years, this
pass cleans up a lot of the surrounding cruft but leaves the actual
argument processing alone - so there's still more that could be done.

Bug fixed:

  - ghc_rts_opts should not be subject to the --rtsopts setting.  If
    the programmer explicitly declares options with ghc_rts_opts, they
    shouldn't also have to accept command-line RTS options to make them
    work.

real bug, spotted by Marcin Orczyk (thanks!).  I'm not sure if it lead
to any actual crashes.

Warning: This change seems to tickle a bug in ghc-stage1 compiler
built with GHC 6.12.1 during validates.
Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

The new code generator was doing some interesting spilling across
unsafe foreign calls:

     _c1ao::I32 = Hp - 4;
     I32[Sp - 20] = _c1ao::I32;
     foreign "ccall"
       newCAF((BaseReg, PtrHint), (R1, PtrHint))[_unsafe_call_];
     _c1ao::I32 = I32[Sp - 20];

This is fairly unnecessary, and resulted from over-conservative
liveness analysis from CmmLive.  We can see that the old code
generator only saved volatile registers across unsafe foreign calls:
spilling variables was done by saveVolatileVarsAndRegs, which was
only performed for ordinary calls.

This commit removes the excess kill from the liveness analysis, as well
as the *redundant* excess kill from spilling-and-reloading, and adds a
note to CmmNode to this effect.  The only registers we need to kill
are the ones that the foreign call assigns to, just like any other
machine instruction.
Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

This is a port of some of the changes from my private local-GC branch
(which is still in darcs, I haven't converted it to git yet).  There
are a couple of small functional differences in the GC stats: first,
per-thread GC timings should now be more accurate, and secondly we now
report average and maximum pause times. e.g. from minimax +RTS -N8 -s:

                                    Tot time (elapsed)  Avg pause  Max pause
  Gen  0      2755 colls,  2754 par   13.16s    0.93s     0.0003s    0.0150s
  Gen  1       769 colls,   769 par    3.71s    0.26s     0.0003s    0.0059s

otherwise we repeatedly try to rebuild ghc-pkg if ghc-cabal is newer.