GitLab

Off-by-one tidyup.

ALLOC_AP, ALLOC_PAP and MKAP were all being constructed
with size arguments equal to (1+number of args/FVs) in
ByteCodeGen.schemeE, only for Interpreter.c to subtract 1
when fishing out the payloads. This commit drops the
up-and-downery.

Simplification spotted by Andy Moran

be gone

NCG support for f.e.d. stdcall -- Literal.MachLabels now optionally carry
the size (in bytes) of the stack frame it expects, if known. That just
so happens to match what stdcall labels need to be annotated with when
emitting them in the NCG..

Two performance tweaks:

  - Use specialised indirections, which perform the right kind of
    return without needing to enter the object they point to.  This
    saves a small percentages of memory reads.

  - Tweak the update code to generate better code with gcc.  This
    saves a few instructions per update.

-------------------------------------
      Fix a wibble in default-type setting for TH
	-------------------------------------

When a bunch of declarations contains no 'default' declaration,
we were setting 'defaultDefaultTys' as appropriate default types.
This isn't right for Template Haskell, which may have more than
one bunch of top-level decls.  (The bunches are separated by
top-level declaration splices.)

Instead, if there is no 'default' declaration we should do nothing.

-------------------------------------
      Respect SPECIALISE pragmas in instance decls
	-------------------------------------

For some time now we have simply been discarding SPECIALISE pragmas
in instance declarations.  I think this was my fault, at some point
when I was re-plumbing TcClassDcl.lhs, but it's been this way for
some time.

The only uses of this facility in the Prelude are in GHC/Float.lhs
and GHC/Real.lhs, which affected the efficiency of the Float and
Double instance of properFraction, floor, ceiling, etc.

Ah well, it's fixed now.

Reduce tctrace verbosity

Import wibbles

allocatePinned(): move large-object check to the front

bring rts_mainLazyIO() into scope

wibbles - drop references to PleaseStopAllocating(), use CloseNursery() to express ExtendNursery()

initStorage(): sm_mutex not correctly initialised (guess no one's done an SMP build for the past 12 months :)

Add __attribute__((regparm(1))) to evacuate()'s prototype.  This
avoids a lot of stack assignments between scavenge() and
evacuate().  Performance isn't affected measurably, though.

Fix bug in previous two commits: must check HEAP_ALLOCED() before
using Bdescr().

Check for the rlim_t type

gen_map_mblocks(): add comment re: issues that need to be addressed should anyone want to tackle the ToDo item

bring rts_mainLazyIO() into scope

to ease replacement of the underlying allocator used by stgMallocBytes() and friends, provide (&use) RtsUtils.c:stgFree()

wibble wrt stgMallocBytes()

msTime(): only used with PAR

leave out unused statics when not using BFD

Update ghc-pkg documentation w.r.t. multiple config files, and add
documentation for 'auto' packages.

code wibbles

BCO_INSTRS() is no more

Handle multiple config files correctly.  Now it is possible to specify
a dependency on a package in another config file, and --list-packages
lists all the packages in each file.

The last file mentioned on the command line is the one to which
updates are applied, packages are added, etc.

wibble - move LARGE_OBJECT_THRESHOLD from Constants.h to Block.h, as it's defined in terms of Block.h defines

upd MAX_SPEC_SELECTEE_SIZE comment

Update the out-of-memory error message: the request size isn't known
any more.

warning elimination

A couple of changes related to bug #2 in the previous commit:

 - follow IND_STATICs, now that we can check whether
   we end up in to-space or not

 - improve the commentary

Fix some bugs in compacting GC.

Bug 1: When threading the fields of an AP or PAP, we were grabbing the
info table of the function without unthreading it first.

Bug 2: eval_thunk_selector() might accidentally find itself in
to-space when going through indirections in a compacted generation.
We must check for this case and bale out if necessary.

Bug 3: This is somewhat more nasty.  When we have an AP or PAP that
points to a BCO, the layout info for the AP/PAP is in the BCO's
instruction array, which is two objects deep from the AP/PAP itself.
The trouble is, during compacting GC, we can only safely look one
object deep from the current object, because pointers from objects any
deeper might have been already updated to point to their final
destinations.

The solution is to put the arity and bitmap info for a BCO into the
BCO object itself.  This means BCOs become variable-length, which is a
slight annoyance, but it also means that looking up the arity/bitmap
is quicker.  There is a slight reduction in complexity in the byte
code generator due to not having to stuff the bitmap at the front of
the instruction stream.

#ifdef some more code that belongs to the old strictness analyser.

It turns out we were still compiling the Demand and SaLib modules,
which aren't required unless OLD_STRICTNESS is on (do we still need
OLD_STRICTNESS?).

dsForeigns: common up header files

Friday morning code-wibbling:
- made RetainerProfile.c:firstStack a 'static'
- added RetainerProfile.c:retainerStackBlocks()

drop FILE_SIZE (hasn't been needed by NCG for the past 5 years or so)

reliably clean out mk* binaries

Modifications to the way we calculate CafInfo during tidying (again).

The previous hack of setting the CafInfo on all non-top-level bindings
to NoCafRefs was a hack, and it came back to bite us: when CorePrep
floats out a let to the top level it doesn't create a new binding, and
the existing let binder happens to say NoCafRefs which is unsafe.  It
was caught by an ASSERT in the CoreToStg when compiling the libraries
without -O - compiling without -O tends to result in more
opportunities for CorePrep to float things to the top level.

Now, we calculate CafInfo on the pre-tidied expressions, using the
mapping from Ids to Ids that is being built up during tidying.  This
avoids one loop, but will be slightly slower due to the extra lookups.
However, it means we don't need to set the CafInfo on non-top-level
binders to NoCafRefs.

GNU ar seems to have trouble with duplicate object file names if you
use the 's' modifier, so detect GNU ar here and use a simple 'ar q'
command to create archives.

This bug was tickled when I replaced an 'ar q' with '$(AR)' in
target.mk yesterday: the default $(AR) is 'ar clqs'.

Notes about constructor info tables

replace one instance of ar with $(AR).