Simon Marlow [Fri, 9 Jul 2010 09:15:21 +0000 (09:15 +0000)]
quiet some new spewage
Simon Marlow [Thu, 8 Jul 2010 14:48:51 +0000 (14:48 +0000)]
New asynchronous exception control API (ghc parts)
As discussed on the libraries/haskell-cafe mailing lists
http://www.haskell.org/pipermail/libraries/2010-April/013420.html
This is a replacement for block/unblock in the asychronous exceptions
API to fix a problem whereby a function could unblock asynchronous
exceptions even if called within a blocked context.
The new terminology is "mask" rather than "block" (to avoid confusion
due to overloaded meanings of the latter).
In GHC, we changed the names of some primops:
blockAsyncExceptions# -> maskAsyncExceptions#
unblockAsyncExceptions# -> unmaskAsyncExceptions#
asyncExceptionsBlocked# -> getMaskingState#
and added one new primop:
maskUninterruptible#
See the accompanying patch to libraries/base for the API changes.
Simon Marlow [Thu, 8 Jul 2010 10:08:40 +0000 (10:08 +0000)]
remove outdated comment
Simon Marlow [Thu, 8 Jul 2010 10:08:17 +0000 (10:08 +0000)]
remove 'mode: xml' emacs settings (#2208)
Simon Marlow [Wed, 16 Jun 2010 11:13:59 +0000 (11:13 +0000)]
typo in comment
Simon Marlow [Thu, 8 Jul 2010 09:32:23 +0000 (09:32 +0000)]
Win32 getProcessElapsedTime: use a higher-resolution time source
QueryPerformanceCounter() on Windows gives much better resolution than
GetSystemTimeAsFileTime().
Sergei Trofimovich [Thu, 8 Jul 2010 06:53:18 +0000 (06:53 +0000)]
alpha: switch handling of 'foreign import wrapper' (FIW) to libffi
I tried to build ghc-6.12.3 and found out FIW part of code
does not compile anymore. It uses absent functions under #ifdef.
Instead of fixing it I just switched to libffi. Result built successfully
and passed 'foreign import wrapper' test I wrote for trac ticket #3516.
I didn't try to build -HEAD yet, but this patch only removes code, so
it should not make -HEAD worse.
Ian Lynagh [Thu, 8 Jul 2010 20:35:23 +0000 (20:35 +0000)]
Reorder the CPP flags so -optP can override the platform defines
Ian Lynagh [Wed, 7 Jul 2010 23:09:07 +0000 (23:09 +0000)]
Add docs for DatatypeContexts extension
Ian Lynagh [Wed, 7 Jul 2010 21:25:29 +0000 (21:25 +0000)]
Make datatype contexts an extension (on by default) (DatatypeContexts)
David Terei [Wed, 7 Jul 2010 22:04:48 +0000 (22:04 +0000)]
LLVM: Fix various typos in comments
Ian Lynagh [Wed, 7 Jul 2010 12:04:23 +0000 (12:04 +0000)]
Handle haddock headers when looking for LANGUAGE/OPTIONS_GHC pragmas
Ian Lynagh [Tue, 6 Jul 2010 17:30:07 +0000 (17:30 +0000)]
Make pragState call mkPState, rather than duplicating everything
This also means that extsBitmap gets set, whereas is was just being set
to 0 before.
David Terei [Wed, 7 Jul 2010 14:20:53 +0000 (14:20 +0000)]
LLVM: Add alias type defenitions to LlvmModule.
David Terei [Wed, 7 Jul 2010 12:03:20 +0000 (12:03 +0000)]
LLVM: Use packed structure type instead of structure type
The regular structure type adds padding to conform to the platform ABI,
which causes problems with structures storing doubles under windows since
we don't conform to the platform ABI there. So we use packed structures
instead now that don't do any padding.
Ian Lynagh [Tue, 6 Jul 2010 17:26:11 +0000 (17:26 +0000)]
Make mkPState and pragState take their arguments in the same order
Ian Lynagh [Tue, 6 Jul 2010 17:22:17 +0000 (17:22 +0000)]
Remove an out-of-date comment
David Terei [Mon, 5 Jul 2010 16:26:29 +0000 (16:26 +0000)]
LLVM: Stop llvm saving stg caller-save regs across C calls
This is already handled by the Cmm code generator so LLVM is simply
duplicating work. LLVM also doesn't know which ones are actually live
so saves them all which causes a fair performance overhead for C calls
on x64. We stop llvm saving them across the call by storing undef to
them just before the call.
David Terei [Mon, 5 Jul 2010 16:15:44 +0000 (16:15 +0000)]
LLVM: Add in literal undefined value to binding
David Terei [Mon, 5 Jul 2010 16:13:08 +0000 (16:13 +0000)]
LLVM: Add a literal NULL value to binding
Patch from Erik de Castro Lopo <erikd@mega-nerd.com>.
Simon Marlow [Mon, 5 Jul 2010 10:45:57 +0000 (10:45 +0000)]
refactor import declaration support (#2362)
Simon Marlow [Mon, 5 Jul 2010 10:30:14 +0000 (10:30 +0000)]
Disable dynamic linking optimisations on OS X
To improve performance of the RTS when dynamically linked on x86, I
previously disabled -fPIC for certain critical modules (the GC, and a
few others). However, build reports suggest that the dynamic linker
on OS X doesn't like this, so I'm disabling this optimsation on that
platform.
amsay@amsay.net [Fri, 25 Jun 2010 03:26:32 +0000 (03:26 +0000)]
trac #2362 (full import syntax in ghci)
'import' syntax is seperate from ':module' syntax
Ian Lynagh [Sun, 4 Jul 2010 18:41:55 +0000 (18:41 +0000)]
Simplify ghc-pkg's Cabal dependencies
We no longer support building with a compiler that doesn't come with
base 4.
Ian Lynagh [Sun, 4 Jul 2010 13:26:12 +0000 (13:26 +0000)]
Use Cabal to configure the dist-install ghc-pkg; fixes trac #4156
Ian Lynagh [Thu, 1 Jul 2010 16:20:58 +0000 (16:20 +0000)]
Remove dead code (standalone deriving flag no longer used in parser)
David Terei [Fri, 2 Jul 2010 16:05:11 +0000 (16:05 +0000)]
LLVM: Use the inbounds keyword for getelementptr instructions.
Simon Marlow [Thu, 1 Jul 2010 08:50:46 +0000 (08:50 +0000)]
threadPaused: fix pointer arithmetic
Noticed by Henrique Ferreiro <hferreiro@udc.es>, thanks!
David Terei [Thu, 1 Jul 2010 16:18:56 +0000 (16:18 +0000)]
LLVM: Change more operations to use getelementptr
Simon Marlow [Wed, 30 Jun 2010 12:55:32 +0000 (12:55 +0000)]
Add the haskell2010 package
David Terei [Wed, 30 Jun 2010 18:11:57 +0000 (18:11 +0000)]
LLVM: Use getelementptr instruction for a lot of situations
LLVM supports creating pointers in two ways, firstly through
pointer arithmetic (by casting between pointers and ints)
and secondly using the getelementptr instruction. The second way
is preferable as it gives LLVM more information to work with.
This patch changes a lot of pointer related code from the first
method to the getelementptr method.
Simon Marlow [Fri, 25 Jun 2010 10:03:13 +0000 (10:03 +0000)]
remove out of date comments; point to the wiki
benl@ouroborus.net [Tue, 29 Jun 2010 05:43:21 +0000 (05:43 +0000)]
NCG: allocatableRegs is only giving us 8 SSE regs to allocate to
David Terei [Mon, 28 Jun 2010 18:29:49 +0000 (18:29 +0000)]
LLVM: Use intrinsic functions for pow, sqrt, sin, cos
Instead of calling the C library for these Cmm functions
we use intrinsic functions provided by llvm. LLVM will
then either create a compile time constant if possible, or
use a cpu instruction or as a last resort call the C
library.
David Terei [Mon, 28 Jun 2010 16:52:56 +0000 (16:52 +0000)]
LLVM: Fix test '2047' under linux-x64
David Terei [Mon, 28 Jun 2010 15:53:55 +0000 (15:53 +0000)]
LLVM: Fix test 'ffi005' under linux-x64
David Terei [Mon, 28 Jun 2010 14:40:37 +0000 (14:40 +0000)]
LLVM: Update to use new fp ops introduced in 2.7
David Terei [Mon, 28 Jun 2010 11:51:20 +0000 (11:51 +0000)]
Add noalias and nocapture attributes to pointer stg registers
At the moment this gives a very slight performance boost of around 1 - 2%.
Future changes to the generated code though so that pointers are kept as
pointers more often instead of being cast to integer types straight away
should hopefully improve the benefit this brings.
Simon Marlow [Mon, 28 Jun 2010 09:05:36 +0000 (09:05 +0000)]
during shutdown, only free the heap if we waited for foreign calls to exit
shelarcy [Sun, 20 Jun 2010 07:07:59 +0000 (07:07 +0000)]
Fix typo in -ddump-pass's document.
dmp@rice.edu [Thu, 24 Jun 2010 16:35:14 +0000 (16:35 +0000)]
Add #undefs for posix source symbols when including papi.h
Validation fails when validating with PAPI support (i.e. GhcRtsWithPapi = YES
in validate.mk). The problem is that the posix symbols are defined by a header
included from papi.h. Compilation then fails because these symbols are
redefined in PosixSource.h.
This patch adds an undefine for the posix symbols after including papi.h and
before including PosixSource.h. The #undefines are localized to Papi.c since
that is the only case where they are getting defined twice.
Ian Lynagh [Fri, 25 Jun 2010 22:09:53 +0000 (22:09 +0000)]
Use machdepCCOpts in runPhase_MoveBinary; fixes trac #3952
David Terei [Fri, 25 Jun 2010 11:57:29 +0000 (11:57 +0000)]
LLVM: Fix bug with calling tail with empty list
benl@ouroborus.net [Thu, 24 Jun 2010 09:13:39 +0000 (09:13 +0000)]
Fix warnings
benl@ouroborus.net [Thu, 24 Jun 2010 08:31:21 +0000 (08:31 +0000)]
NCG: Comments and formatting only
benl@ouroborus.net [Thu, 24 Jun 2010 08:27:17 +0000 (08:27 +0000)]
NCG: Do the actual reversing of SCCs
benl@ouroborus.net [Thu, 24 Jun 2010 08:26:25 +0000 (08:26 +0000)]
NCG: Fix dumping of graphs in regalloc stats for graph allocator
benl@ouroborus.net [Thu, 24 Jun 2010 08:24:37 +0000 (08:24 +0000)]
NCG: Reverse SCCs after each round in the graph allocator
benl@ouroborus.net [Thu, 24 Jun 2010 08:14:45 +0000 (08:14 +0000)]
NCG: Don't actually complain on unreachable code blocks
benl@ouroborus.net [Thu, 24 Jun 2010 08:07:47 +0000 (08:07 +0000)]
NCG: Do explicit check for precondition of computeLiveness
computeLiveness requires the SCCs of blocks to be in reverse dependent
order, and if they're not it was silently giving bad liveness info,
yielding a bad allocation.
Now it complains, loudly.
benl@ouroborus.net [Wed, 23 Jun 2010 09:58:13 +0000 (09:58 +0000)]
NCG: Fix off-by-one error in realRegSqueeze
benl@ouroborus.net [Wed, 23 Jun 2010 09:12:09 +0000 (09:12 +0000)]
NCG: Handle stripping of liveness info from procs with no blocks (like stg_split_marker)
benl@ouroborus.net [Wed, 23 Jun 2010 08:50:02 +0000 (08:50 +0000)]
NCG: Emit a warning on unreachable code block instead of panicing
Ben.Lippmeier@anu.edu.au [Thu, 17 Sep 2009 10:44:29 +0000 (10:44 +0000)]
NCG: Remember to keep the entry block first when erasing liveness info
Ben.Lippmeier@anu.edu.au [Thu, 17 Sep 2009 09:07:30 +0000 (09:07 +0000)]
NCG: Refactor representation of code with liveness info
* I've pushed the SPILL and RELOAD instrs down into the
LiveInstr type to make them easier to work with.
* When the graph allocator does a spill cycle it now just
re-annotates the LiveCmmTops instead of converting them
to NatCmmTops and back.
* This saves working out the SCCS again, and avoids rewriting
the SPILL and RELOAD meta instructions into real machine
instructions.
Ben.Lippmeier@anu.edu.au [Thu, 17 Sep 2009 09:03:35 +0000 (09:03 +0000)]
NCG: Add sanity checking to linear allocator
If there are are unreachable basic blocks in the native code then the
linear allocator might loop. Detect this case and panic instead.
Ben.Lippmeier@anu.edu.au [Thu, 17 Sep 2009 06:03:32 +0000 (06:03 +0000)]
NCG: Refactor LiveCmmTop to hold a list of SCCs instead of abusing ListGraph
Ben.Lippmeier@anu.edu.au [Thu, 17 Sep 2009 04:39:37 +0000 (04:39 +0000)]
NCG: Allow the liveness map in a LiveInfo to be Nothing
Ben.Lippmeier.anu.edu.au [Thu, 17 Sep 2009 03:44:27 +0000 (03:44 +0000)]
NCG: Also show the result of applying coalesings with -ddump-asm-regalloc-stages
Ian Lynagh [Thu, 24 Jun 2010 23:40:11 +0000 (23:40 +0000)]
Fix panic when running "ghc -H"; trac #3364
The problem is that showing SDoc's looks at the static flags global
variables, but those are panics while we are parsing the static flags.
We work around this by explicitly using a fixed prettyprinter style.
David Terei [Mon, 21 Jun 2010 17:58:39 +0000 (17:58 +0000)]
Allow for stg registers to have pointer type in llvm BE.
Before all the stg registers were simply a bit type or
floating point type but now they can be declared to have
a pointer type to one of these. This will allow various
optimisations in the future in llvm since the type is
more accurate.
David Terei [Thu, 24 Jun 2010 11:17:44 +0000 (11:17 +0000)]
Add support for parameter attributes to the llvm BE binding
These allow annotations of the code produced by the backend
which should bring some perforamnce gains. At the moment
the attributes aren't being used though.
Simon Marlow [Thu, 24 Jun 2010 10:57:00 +0000 (10:57 +0000)]
Cast some more nats to StgWord to be on the safe side
And add a comment about the dangers of int overflow
Simon Marlow [Thu, 24 Jun 2010 10:51:05 +0000 (10:51 +0000)]
comments only
Simon Marlow [Thu, 24 Jun 2010 10:46:54 +0000 (10:46 +0000)]
Fix an arithmetic overflow bug causing crashes with multi-GB heaps
dmp@rice.edu [Tue, 22 Jun 2010 19:59:53 +0000 (19:59 +0000)]
Add support for collecting PAPI native events
This patch extends the PAPI support in the RTS to allow collection of native
events. PAPI can collect data for native events that are exposed by the
hardware beyond the PAPI present events. The native events supported on your
hardware can found by using the papi_native_avail tool.
The RTS already allows users to specify PAPI preset events from the command
line. This patch extends that support to allow users to specify native events.
The changes needed are:
1) New option (#) for the RTS PAPI flag for native events. For example, to
collect the native event 0x40000000, use ./a.out +RTS -a#0x40000000 -sstderr
2) Update the PAPI_FLAGS struct to store whether the user specified event is a
papi preset or a native event
3) Update init_countable_events function to add the native events after parsing
the event code and decoding the name using PAPI_event_code_to_name
Ian Lynagh [Thu, 24 Jun 2010 11:03:51 +0000 (11:03 +0000)]
Don't warn about unused bindings with parents in .hs-boot files; trac #3449
Simon Marlow [Mon, 21 Jun 2010 12:55:35 +0000 (12:55 +0000)]
fix the home_imps filter to allow for 'import "this" <module>'
In the PackageImports extension, import "this" means "import from the
current package".
Ian Lynagh [Tue, 22 Jun 2010 20:28:59 +0000 (20:28 +0000)]
Use the standard C wrapper code for the ghc-$version.exe wrapper
Ian Lynagh [Tue, 22 Jun 2010 20:25:47 +0000 (20:25 +0000)]
Don't rely on "-packagefoo" working; use "-package foo" instead
Ian Lynagh [Tue, 22 Jun 2010 17:29:19 +0000 (17:29 +0000)]
Remove unnecessary C #includes
Ian Lynagh [Tue, 22 Jun 2010 17:22:47 +0000 (17:22 +0000)]
Make the ghci.exe wrapper call the right ghc.exe
David Terei [Tue, 22 Jun 2010 16:53:39 +0000 (16:53 +0000)]
More updates to datalayout description in llvm BE
David Terei [Tue, 22 Jun 2010 14:40:44 +0000 (14:40 +0000)]
Remove LlvmAs phase as the llvm opt tool now handles this phase
This phase originally invoked the llvm-as tool that turns a textual
llvm assembly file into a bit code file for the rest of llvm to deal
with. Now the llvm opt tool can do this itself, so we don't need to
use llvm-as anymore.
David Terei [Tue, 22 Jun 2010 12:34:57 +0000 (12:34 +0000)]
Update datalayout info in llvm BE
David Terei [Tue, 22 Jun 2010 12:16:42 +0000 (12:16 +0000)]
Fix handling of float literals in llvm BE
David Terei [Mon, 21 Jun 2010 17:49:54 +0000 (17:49 +0000)]
Declare some top level globals to be constant when appropriate
This involved removing the old constant handling mechanism
which was fairly hard to use. Now being constant or not is
simply a property of a global variable instead of a separate
type.
David Terei [Mon, 21 Jun 2010 12:52:20 +0000 (12:52 +0000)]
Reduce the number of passes over the cmm in llvm BE
David Terei [Mon, 21 Jun 2010 12:36:06 +0000 (12:36 +0000)]
Fix negate op not working for -0 in llvm backend
Simon Marlow [Mon, 21 Jun 2010 10:04:09 +0000 (10:04 +0000)]
ROLLBACK: picCCOpts: -dynamic should not entail -optc-fPIC
and add a comment to explain why it was wrong. This fixes the dyn
test failures that sprang up recently.
Ian Lynagh [Sun, 20 Jun 2010 16:37:24 +0000 (16:37 +0000)]
Check files are really created in libffi
when we think that the libffi build creates them, so they just depend
on the libffi build stamp.
Ian Lynagh [Sun, 20 Jun 2010 12:43:20 +0000 (12:43 +0000)]
Improve the missing-import-list warning
Ian Lynagh [Sun, 20 Jun 2010 12:26:22 +0000 (12:26 +0000)]
Tweak missing-import-list warning
amsay@amsay.net [Fri, 18 Jun 2010 15:06:49 +0000 (15:06 +0000)]
trac #1789 (warnings for missing import lists)
Ian Lynagh [Sun, 20 Jun 2010 10:37:49 +0000 (10:37 +0000)]
Refix docs for sizeofByteArray#/sizeofMutableByteArray# (#3800)
Ian Lynagh [Sun, 20 Jun 2010 00:04:59 +0000 (00:04 +0000)]
Remove some old commented out code
Ian Lynagh [Sat, 19 Jun 2010 23:52:14 +0000 (23:52 +0000)]
SET_ARR_HDR's last argument is now a number of bytes, rather than words
This avoids unnecessary work and potential loss of information
Ian Lynagh [Sat, 19 Jun 2010 23:43:10 +0000 (23:43 +0000)]
Replace an (incorrect) bytes-to-words calculation with ROUNDUP_BYTES_TO_WDS
Antoine Latter [Fri, 1 Jan 2010 18:33:46 +0000 (18:33 +0000)]
FIX #38000 Store StgArrWords payload size in bytes
David Terei [Fri, 18 Jun 2010 13:17:33 +0000 (13:17 +0000)]
Add win32 datalayout support to llvm backend
David Terei [Fri, 18 Jun 2010 10:12:37 +0000 (10:12 +0000)]
Remove unused 'ddump-opt-llvm' flag
Simon Marlow [Fri, 18 Jun 2010 08:22:58 +0000 (08:22 +0000)]
generate "movl lbl(%reg1), %reg2" instructions, better codegen for -fPIC
Simon Marlow [Fri, 18 Jun 2010 08:21:47 +0000 (08:21 +0000)]
joinToTargets: fix a case of panic "handleComponent cyclic"
Simon Marlow [Fri, 18 Jun 2010 08:21:02 +0000 (08:21 +0000)]
comment typo
David Terei [Fri, 18 Jun 2010 09:32:05 +0000 (09:32 +0000)]
Add support of TNTC to llvm backend
We do this through a gnu as feature called subsections,
where you can put data/code into a numbered subsection
and those subsections will be joined together in descending
order by gas at compile time.
Ian Lynagh [Thu, 17 Jun 2010 19:09:01 +0000 (19:09 +0000)]
Don't automatically insert a -fvia-C flag in an unregisterised compiler
The default object mode is already HscC, so it's unnecessary, and
-fvia-C generates a deprecated flag warning now.
Ian Lynagh [Thu, 17 Jun 2010 17:49:12 +0000 (17:49 +0000)]
In PosixSource.h, conditionally define things based on platform
This may not be ideal, but it should get GHC building on all platforms
again.
Simon Marlow [Thu, 17 Jun 2010 14:00:25 +0000 (14:00 +0000)]
disable -fPIC for the GC for performance reasons
see comment for details
Simon Marlow [Thu, 17 Jun 2010 11:52:59 +0000 (11:52 +0000)]
picCCOpts: -dynamic should not entail -optc-fPIC
Simon Marlow [Thu, 17 Jun 2010 11:32:59 +0000 (11:32 +0000)]
Make getAllocations() visible