11 years agoAdded the capability for the test driver to split the sys.stderr/sys.stdout into
Johnny Chen [Tue, 12 Oct 2010 21:35:54 +0000 (21:35 +0000)]
Added the capability for the test driver to split the sys.stderr/sys.stdout into
different configuration-based files using the config file.  For example:

    sys.stderr = open("/tmp/lldbtest-stderr", "w")
    sys.stdout = open("/tmp/lldbtest-stdout", "w")
    compilers = ["gcc", "llvm-gcc"]
    archs = ["x86_64", "i386"]
    split_stderr = True # This will split the stderr into configuration-specific file
    split_stdout = True # This will split the stdout into configuration-specific file

will produce:


as a result of splitting stderr and stdout.  In addition, each configuration can have
its individual top level relocated directory to house the test files as well as the
intermediate files by using '-r dir' to relocate the tests into a new relocated directory
instead of running the tests in place.

llvm-svn: 116341

11 years agoForgor to also check in this change with r116337.
Johnny Chen [Tue, 12 Oct 2010 21:24:25 +0000 (21:24 +0000)]
Forgor to also check in this change with r116337.

llvm-svn: 116340

11 years agoFix thinko in arm fast isel alloca rewrite.
Eric Christopher [Tue, 12 Oct 2010 21:23:43 +0000 (21:23 +0000)]
Fix thinko in arm fast isel alloca rewrite.

llvm-svn: 116339

11 years agoEncoding for ARM-mode VADD.F32 instruction.
Jim Grosbach [Tue, 12 Oct 2010 21:22:40 +0000 (21:22 +0000)]
Encoding for ARM-mode VADD.F32 instruction.

llvm-svn: 116338

11 years agoUpdate the @expectedFailure decorator with additional bug info:
Johnny Chen [Tue, 12 Oct 2010 21:20:11 +0000 (21:20 +0000)]
Update the @expectedFailure decorator with additional bug info:

    rdar://problem/8542091 test/foundation: expr -o -- my not working?

Add an additional test for 'frame variable *self' when stopped in '-[MyString initWithNSString:]'
and move the 'expr -o -- self' to after MyString has been constructed and change it to
'expr -o -- my'.

llvm-svn: 116337

11 years agoPR8325: don't do destructor checking when a pointer is thrown.
Eli Friedman [Tue, 12 Oct 2010 20:32:36 +0000 (20:32 +0000)]
PR8325: don't do destructor checking when a pointer is thrown.

llvm-svn: 116336

11 years agoBegin adding static dependence information to passes, which will allow us to
Owen Anderson [Tue, 12 Oct 2010 19:48:12 +0000 (19:48 +0000)]
Begin adding static dependence information to passes, which will allow us to
perform initialization without static constructors AND without explicit initialization
by the client.  For the moment, passes are required to initialize both their
(potential) dependencies and any passes they preserve.  I hope to be able to relax
the latter requirement in the future.

llvm-svn: 116334

11 years agoCombine these together - should probably have some text associated
Eric Christopher [Tue, 12 Oct 2010 19:44:17 +0000 (19:44 +0000)]
Combine these together - should probably have some text associated
that says what why what we just asserted is wrong.

llvm-svn: 116333

11 years agoEnter the context of the declared function template when performing
John McCall [Tue, 12 Oct 2010 19:40:14 +0000 (19:40 +0000)]
Enter the context of the declared function template when performing
deduction and the final substitution, but not while substituting the
explicit template arguments.  Fixes rdar://problem/8537391

llvm-svn: 116332

11 years agoThe 'regexp-break' command now prints out the command that results from
Johnny Chen [Tue, 12 Oct 2010 19:29:49 +0000 (19:29 +0000)]
The 'regexp-break' command now prints out the command that results from
expanding the regular expression command.  So change the more stringent:

    self.expect(..., startstr = matching_string)


    self.expect(..., substrs = [matched_string])

to pass the test.

llvm-svn: 116331

11 years agoKillTheDoctor: Fix VS2008 build.
Michael J. Spencer [Tue, 12 Oct 2010 19:27:44 +0000 (19:27 +0000)]
KillTheDoctor: Fix VS2008 build.

llvm-svn: 116330

11 years agoWhen finalizing a function template specialization following template
Douglas Gregor [Tue, 12 Oct 2010 18:51:08 +0000 (18:51 +0000)]
When finalizing a function template specialization following template
argument deduction, make sure to check the correctness of deduced template
type arguments (which we had previously skipped) along with other
kinds of template arguments. This fixes part of PR6784, but we're
still swallowing the extension warning about unnamed/local template

llvm-svn: 116327

11 years agoIf we end up instantiating a function parameter whose default argument
Douglas Gregor [Tue, 12 Oct 2010 18:23:32 +0000 (18:23 +0000)]
If we end up instantiating a function parameter whose default argument
has not yet been parsed, note that the default argument hasn't been
parsed and keep track of all of the instantiations of that function
parameter. When its default argument does get parsed, imbue the
instantiations with that default argument. Fixes PR8245.

llvm-svn: 116324

11 years agoMark variable 'NoImplicitFloatOps' used only in an assert as used.
Nick Lewycky [Tue, 12 Oct 2010 18:18:03 +0000 (18:18 +0000)]
Mark variable 'NoImplicitFloatOps' used only in an assert as used.

llvm-svn: 116323

11 years agoComment grammar tweakage.
Jim Grosbach [Tue, 12 Oct 2010 18:11:41 +0000 (18:11 +0000)]
Comment grammar tweakage.

llvm-svn: 116322

11 years agoAdd MOVi ARM encoding.
Jim Grosbach [Tue, 12 Oct 2010 18:09:12 +0000 (18:09 +0000)]
Add MOVi ARM encoding.

llvm-svn: 116321

11 years agoRegular expression commands now print the command that results from expanding the...
Greg Clayton [Tue, 12 Oct 2010 18:04:53 +0000 (18:04 +0000)]
Regular expression commands now print the command that results from expanding the regular expression command.

llvm-svn: 116320

11 years agoInitial va_arg support for x86-64. Patch by David Meyer!
Dan Gohman [Tue, 12 Oct 2010 18:00:49 +0000 (18:00 +0000)]
Initial va_arg support for x86-64. Patch by David Meyer!

llvm-svn: 116319

11 years agoNuke unused wrapper function.
Jim Grosbach [Tue, 12 Oct 2010 17:53:25 +0000 (17:53 +0000)]
Nuke unused wrapper function.

llvm-svn: 116318

11 years agofix PR7885, rejecting invalid uses of __builtin_constant_p.
Chris Lattner [Tue, 12 Oct 2010 17:47:42 +0000 (17:47 +0000)]
fix PR7885, rejecting invalid uses of __builtin_constant_p.

llvm-svn: 116317

11 years agoFix bug where alias command options were being duplicated as command arguments as...
Caroline Tice [Tue, 12 Oct 2010 17:45:19 +0000 (17:45 +0000)]
Fix bug where alias command options were being duplicated as command arguments as well.

llvm-svn: 116316

11 years agoFixed the dispatch queue name retrieval for threads by looking in an extra
Greg Clayton [Tue, 12 Oct 2010 17:33:06 +0000 (17:33 +0000)]
Fixed the dispatch queue name retrieval for threads by looking in an extra

llvm-svn: 116315

11 years agoRemove the x86 MOV{32,64}{rr,rm,mr}_TC instructions.
Jakob Stoklund Olesen [Tue, 12 Oct 2010 17:15:00 +0000 (17:15 +0000)]
Remove the x86 MOV{32,64}{rr,rm,mr}_TC instructions.

The reg-reg copies were no longer being generated since copyPhysReg copies
physical registers only.

The loads and stores are not necessary - The TC constraint is imposed by the
TAILJMP and TCRETURN instructions, there should be no need for constrained loads
and stores.

llvm-svn: 116314

11 years agoAdd encoding information for the remainder of the generic arithmetic
Jim Grosbach [Tue, 12 Oct 2010 17:11:26 +0000 (17:11 +0000)]
Add encoding information for the remainder of the generic arithmetic
ARM instructions.

llvm-svn: 116313

11 years agoReplace \r\n with \n in this file.
Nick Lewycky [Tue, 12 Oct 2010 16:46:35 +0000 (16:46 +0000)]
Replace \r\n with \n in this file.

llvm-svn: 116312

11 years agoParse default arguments within member functions in source order, from
Douglas Gregor [Tue, 12 Oct 2010 16:25:54 +0000 (16:25 +0000)]
Parse default arguments within member functions in source order, from
Manuel Klimek! Fixes PR7715.

llvm-svn: 116311

11 years agoPR8359: The ARM backend may end up allocating registers D16 to D31 when
Bob Wilson [Tue, 12 Oct 2010 16:22:47 +0000 (16:22 +0000)]
PR8359: The ARM backend may end up allocating registers D16 to D31 when
"-mattr=+vfp3" is specified. However, this will not work for hardware that
only supports 16 registers.  Add a new flag to support -"mattr=+vfp3,+d16".
Patch by Jan Voung!

llvm-svn: 116310

11 years agoUpdate comments.
Johnny Chen [Tue, 12 Oct 2010 15:53:22 +0000 (15:53 +0000)]
Update comments.

llvm-svn: 116309

11 years agoPacify gcc-4.4.
Duncan Sands [Tue, 12 Oct 2010 14:07:59 +0000 (14:07 +0000)]
Pacify gcc-4.4.

llvm-svn: 116300

11 years agoC's comma operator performs lvalue conversion on both its operands;
John McCall [Tue, 12 Oct 2010 07:14:40 +0000 (07:14 +0000)]
C's comma operator performs lvalue conversion on both its operands;
require them to have complete types.

llvm-svn: 116297

11 years agoRework alloca handling so that we can load or store from casted
Eric Christopher [Tue, 12 Oct 2010 05:39:06 +0000 (05:39 +0000)]
Rework alloca handling so that we can load or store from casted
address that we've looked through.

Fixes compilation problems in tramp3d from earlier patch.

llvm-svn: 116296

11 years agoModified the "breakpoint set --name NAME" to be the auto breakpoint set
Greg Clayton [Tue, 12 Oct 2010 04:29:14 +0000 (04:29 +0000)]
Modified the "breakpoint set --name NAME" to be the auto breakpoint set
function. It will inspect NAME and do the following:
- if the name contains '(' or starts with "-[" or "+[" then a full name search
  will happen to match full function names with args (C++ demangled names) or
  full objective C method prototypes.
- if the name contains "::" and no '(', then it is assumed to be a qualified
  function name that is in a namespace or class. For "foo::bar::baz" we will
  search for any functions with the basename or method name of "baz", then
  filter the results to only those that contain "foo::bar::baz". This allows
  setting breakpoint on C++ functions and methods without having to fully
  qualify all of the types that would appear in C++ mangled names.
- if the name contains ":" (not "::"), then NAME is assumed to be an ObjC
_ otherwise, we assume just a plain function basename.

Now that "--name" is our "auto" mode, I introduced the new "--basename" option
("breakpoint set --basename NAME") to allow for function names that aren't
methods or selectors, just basenames. This can also be used to ignore C++
namespaces and class hierarchies for class methods.

Fixed clang enumeration promotion types to be correct.

llvm-svn: 116293

11 years agoFlesh out test.
John McCall [Tue, 12 Oct 2010 03:38:33 +0000 (03:38 +0000)]
Flesh out test.

llvm-svn: 116292

11 years agoFix the last two commits to configure - configure is a generated file.
Eric Christopher [Tue, 12 Oct 2010 02:42:05 +0000 (02:42 +0000)]
Fix the last two commits to configure - configure is a generated file.
Made necessary edits to configure.ac and regenerated.

llvm-svn: 116291

11 years agoFixed the Objective C method prototypes to be correct (the selectors weren't
Greg Clayton [Tue, 12 Oct 2010 02:24:53 +0000 (02:24 +0000)]
Fixed the Objective C method prototypes to be correct (the selectors weren't
being chopped up correctly). The DWARF plug-in also keeps a map of the ObjC
class names to selectors for easy parsing of all class selectors when we parse
the class type.

llvm-svn: 116290

11 years agoI just do what the comments tell me to do.
John McCall [Tue, 12 Oct 2010 02:19:57 +0000 (02:19 +0000)]
I just do what the comments tell me to do.

llvm-svn: 116289

11 years agofix the definition of __builtin_constant_p, the first arg isn't a short.
Chris Lattner [Tue, 12 Oct 2010 02:16:39 +0000 (02:16 +0000)]
fix the definition of __builtin_constant_p, the first arg isn't a short.
Patch by the "pax team"

llvm-svn: 116288

11 years agoProgress.
John McCall [Tue, 12 Oct 2010 02:09:17 +0000 (02:09 +0000)]

llvm-svn: 116287

11 years agoThis patch remove the Win32 XFAIL from remap-complete.c
Francois Pichet [Tue, 12 Oct 2010 01:01:43 +0000 (01:01 +0000)]
This patch remove the Win32 XFAIL from remap-complete.c

The problem was that text files were open in text mode and Microsoft implementation of fread and write will try to do nasty line-feed conversion which make the line position no longer valid. The fix is to read and write files in binary mode.

llvm-svn: 116286

11 years agoWhen we load an ASTUnit from command-line arguments, hold on to the
Douglas Gregor [Tue, 12 Oct 2010 00:50:20 +0000 (00:50 +0000)]
When we load an ASTUnit from command-line arguments, hold on to the
diagnostics produced by the driver itself. Previously, we were
allowing these to either be dropped or to slip through to stderr.

Fixes <rdar://problem/7595339>.

llvm-svn: 116285

11 years agoHandle a wider arrangement of loads.
Eric Christopher [Tue, 12 Oct 2010 00:43:21 +0000 (00:43 +0000)]
Handle a wider arrangement of loads.

llvm-svn: 116284

11 years agoFix spelling error.
Cameron Esfahani [Tue, 12 Oct 2010 00:21:25 +0000 (00:21 +0000)]
Fix spelling error.

llvm-svn: 116283

11 years agoFix spelling error.
Cameron Esfahani [Tue, 12 Oct 2010 00:21:05 +0000 (00:21 +0000)]
Fix spelling error.

llvm-svn: 116282

11 years agoAdd some infrastructure for dealing with expressions of 'placeholder' type,
John McCall [Tue, 12 Oct 2010 00:20:44 +0000 (00:20 +0000)]
Add some infrastructure for dealing with expressions of 'placeholder' type,
i.e. expressions with an internally-convenient type which should not be
appearing in generally valid, complete ASTs.

llvm-svn: 116281

11 years agoDelete a redundant check.
Dan Gohman [Tue, 12 Oct 2010 00:19:24 +0000 (00:19 +0000)]
Delete a redundant check.

llvm-svn: 116280

11 years agoMore SmallVectorImpls.
Dan Gohman [Tue, 12 Oct 2010 00:15:27 +0000 (00:15 +0000)]
More SmallVectorImpls.

llvm-svn: 116279

11 years agoShrink a SmallVector with a known maximum size.
Dan Gohman [Tue, 12 Oct 2010 00:13:43 +0000 (00:13 +0000)]
Shrink a SmallVector with a known maximum size.

llvm-svn: 116278

11 years agoConstify.
Dan Gohman [Tue, 12 Oct 2010 00:12:29 +0000 (00:12 +0000)]

llvm-svn: 116277

11 years agoUse SmallVectorImpl in a bunch of places.
Dan Gohman [Tue, 12 Oct 2010 00:11:18 +0000 (00:11 +0000)]
Use SmallVectorImpl in a bunch of places.

llvm-svn: 116276

11 years agoAvoid using hardcoded line number to break on. Use the line_number() utility
Johnny Chen [Tue, 12 Oct 2010 00:09:25 +0000 (00:09 +0000)]
Avoid using hardcoded line number to break on.  Use the line_number() utility
function to get the line number to break on during setUp().

llvm-svn: 116275

11 years agoDisable warning C4267 for MSVC. Otherwise it generate literally thousands of warnings...
Francois Pichet [Tue, 12 Oct 2010 00:01:36 +0000 (00:01 +0000)]
Disable warning C4267 for MSVC. Otherwise it generate literally thousands of warnings when targeting x64. The warning occurs because int is 32 bit but size_t is 64 bit on Win64.

llvm-svn: 116274

11 years agoFixes a typo which caused byte offset in debug info
Fariborz Jahanian [Mon, 11 Oct 2010 23:55:47 +0000 (23:55 +0000)]
Fixes a typo which caused byte offset in debug info
for ivars to be 0. Fixes pr8353.

llvm-svn: 116273

11 years agoAdd a simple testcase for tbaa.
Dan Gohman [Mon, 11 Oct 2010 23:54:13 +0000 (23:54 +0000)]
Add a simple testcase for tbaa.

llvm-svn: 116272

11 years agoAdd a way to temporarily divert events from a broadcaster to a private listener.
Jim Ingham [Mon, 11 Oct 2010 23:53:14 +0000 (23:53 +0000)]
Add a way to temporarily divert events from a broadcaster to a private listener.

llvm-svn: 116271

11 years agoAdd a utility function to lldbtest.py to return the line number of a matched string
Johnny Chen [Mon, 11 Oct 2010 23:52:19 +0000 (23:52 +0000)]
Add a utility function to lldbtest.py to return the line number of a matched string
within a file.  This is to be used within the test case to avoid hardcoded line number.

array_types/TestArrayTypes.py is modified first to use this pattern.  Other test modules
to follow.

Testsuite: don't set breakpoints by exact file & line number

llvm-svn: 116270

11 years agoBasic: Tweak attempt to make version tags work from 'svn export's again, clip
Daniel Dunbar [Mon, 11 Oct 2010 23:50:34 +0000 (23:50 +0000)]
Basic: Tweak attempt to make version tags work from 'svn export's again, clip
off the extra parts of the $URL$ SVN keyword.

llvm-svn: 116269

11 years agoBasic: Attempt to make version tags work from 'svn export's again.
Daniel Dunbar [Mon, 11 Oct 2010 23:44:19 +0000 (23:44 +0000)]
Basic: Attempt to make version tags work from 'svn export's again.

llvm-svn: 116268

11 years agoMore ARM scheduling itinerary fixes.
Evan Cheng [Mon, 11 Oct 2010 23:41:41 +0000 (23:41 +0000)]
More ARM scheduling itinerary fixes.

llvm-svn: 116266

11 years agoSupport AA chaining.
Dan Gohman [Mon, 11 Oct 2010 23:39:34 +0000 (23:39 +0000)]
Support AA chaining.

llvm-svn: 116264

11 years agoDriver/Darwin: Remove a useless warning about missing runtime libraries. These
Daniel Dunbar [Mon, 11 Oct 2010 23:31:07 +0000 (23:31 +0000)]
Driver/Darwin: Remove a useless warning about missing runtime libraries. These
are almost never needed, so the warning isn't very useful.

llvm-svn: 116263

11 years agoFix the pass manager's search order for immutable passes, and make it
Dan Gohman [Mon, 11 Oct 2010 23:19:01 +0000 (23:19 +0000)]
Fix the pass manager's search order for immutable passes, and make it
stop searching when it has found a match.

llvm-svn: 116262

11 years agoThis test should now work on windows, since we're no longer using temporary files...
Douglas Gregor [Mon, 11 Oct 2010 23:18:55 +0000 (23:18 +0000)]
This test should now work on windows, since we're no longer using temporary files badly

llvm-svn: 116261

11 years agoEliminate CIndexer::getClangPath(), since libclang no longer depends
Douglas Gregor [Mon, 11 Oct 2010 23:17:59 +0000 (23:17 +0000)]
Eliminate CIndexer::getClangPath(), since libclang no longer depends
on the presence of a 'clang' executable. Simplify
CIndexer::getClangResourcesPath() a bit.

Patch up the CMake makefiles to install headers into two locations in
the build tree, for those silly cases where 'clang' will end up
looking into the wrong build directory for headers.

llvm-svn: 116260

11 years agoMC machine encoding for simple aritmetic instructions that use a shifted
Jim Grosbach [Mon, 11 Oct 2010 23:16:21 +0000 (23:16 +0000)]
MC machine encoding for simple aritmetic instructions that use a shifted
register operand.

llvm-svn: 116259

11 years agoTest for //rdar: //8493239
Fariborz Jahanian [Mon, 11 Oct 2010 23:15:47 +0000 (23:15 +0000)]
Test for //rdar: //8493239

llvm-svn: 116258

11 years agoSecond set of ARM/MC/ELF changes.
Jason W Kim [Mon, 11 Oct 2010 23:01:44 +0000 (23:01 +0000)]
Second set of ARM/MC/ELF changes.

Added ARM specific ELF section types.
Added AttributesSection to ARMElfTargetObject
First step in unifying .cpu assembly tag with ELF/.o
llc now asserts on actual ELF emission on -filetype=obj :-)

llvm-svn: 116257

11 years agoClang's #include handling apparently doesn't work for libstdc++'s
Dan Gohman [Mon, 11 Oct 2010 22:30:59 +0000 (22:30 +0000)]
Clang's #include handling apparently doesn't work for libstdc++'s
fenv.h. See PR6907 for details. Work around this in FEnv.h to fix
the seflhost build.

llvm-svn: 116256

11 years agoAdded the capability for the test driver to relocate the tests and the intermediate
Johnny Chen [Mon, 11 Oct 2010 22:25:46 +0000 (22:25 +0000)]
Added the capability for the test driver to relocate the tests and the intermediate
files to a different top level directory than those specified on the command line.

When relocated, the test clanups normally performed afterwards after each test method
and after each test class will not be exercised at all.  This allows for an easier
postmortem analysis of test failures.


./dotest.py -v -t -r /tmp/lldbtest types

will create a /tmp/lldbtest directory which houses the types directory and its supported

Files modified:

o dotest.py, lldbtest.py:

  Add logic to process '-r dir' option to support relocating the tests to a different
  top level directory instead of exected in place.

o darwin.py, test/types/Makefile:

  The 'make clean' should only clean the minimum .o and .d files.

llvm-svn: 116255

11 years agoSome refactoring of property rewriting stuff
Fariborz Jahanian [Mon, 11 Oct 2010 22:21:03 +0000 (22:21 +0000)]
Some refactoring of property rewriting stuff
in rewriter. No functionality change.

llvm-svn: 116254

11 years agoEliminate the (de-)serialization of code completion results, now that
Douglas Gregor [Mon, 11 Oct 2010 22:12:15 +0000 (22:12 +0000)]
Eliminate the (de-)serialization of code completion results, now that
libclang does not support out-of-process code completion.

llvm-svn: 116253

11 years agoUnit Tests: Missed this error. MSVC and clang didn't complain.
Michael J. Spencer [Mon, 11 Oct 2010 22:04:38 +0000 (22:04 +0000)]
Unit Tests: Missed this error. MSVC and clang didn't complain.

llvm-svn: 116252

11 years agoProper VST scheduling itineraries.
Evan Cheng [Mon, 11 Oct 2010 22:03:18 +0000 (22:03 +0000)]
Proper VST scheduling itineraries.

llvm-svn: 116251

11 years agoEliminate -fdiagnostics-binary and all of the infrastructure for
Douglas Gregor [Mon, 11 Oct 2010 22:02:06 +0000 (22:02 +0000)]
Eliminate -fdiagnostics-binary and all of the infrastructure for
emitting diagnostics in a binary form to be consumed by libclang,
since libclang no longer does any of its work out-of-process, making
this code dead. Besides, this stuff never worked at 100% anyway.

llvm-svn: 116250

11 years agoUse a sane mechanism for that assert.
Eric Christopher [Mon, 11 Oct 2010 22:01:22 +0000 (22:01 +0000)]
Use a sane mechanism for that assert.

llvm-svn: 116249

11 years agoFix debug info for functions whose context is a namespace.
Devang Patel [Mon, 11 Oct 2010 21:58:41 +0000 (21:58 +0000)]
Fix debug info for functions whose context is a namespace.
This is tested by namespace.exp in gdb testsuite.

llvm-svn: 116248

11 years agoSystem: Add SwapByteOrder and update Support/MathExtras.h to use it.
Michael J. Spencer [Mon, 11 Oct 2010 21:56:16 +0000 (21:56 +0000)]
System: Add SwapByteOrder and update Support/MathExtras.h to use it.

This time correctly.

llvm-svn: 116247

11 years agoEliminate clang_codeComplete(). libclang clients should be using the
Douglas Gregor [Mon, 11 Oct 2010 21:51:20 +0000 (21:51 +0000)]
Eliminate clang_codeComplete(). libclang clients should be using the
faster, in-process, more-configurable clang_codeCompleteAt().

llvm-svn: 116245

11 years agoReplace FindLiveRangeContaining() with getVNInfoAt() in LiveIntervalAnalysis.
Jakob Stoklund Olesen [Mon, 11 Oct 2010 21:45:03 +0000 (21:45 +0000)]
Replace FindLiveRangeContaining() with getVNInfoAt() in LiveIntervalAnalysis.

This helps hiding the LiveRange class which really should be private.

llvm-svn: 116244

11 years agoThe assert() should reference to machine instr operand number, too.
Jim Grosbach [Mon, 11 Oct 2010 21:41:31 +0000 (21:41 +0000)]
The assert() should reference to machine instr operand number, too.

llvm-svn: 116243

11 years agoRevert "System: Add SwapByteOrder and update Support/MathExtras.h to use it."
Michael J. Spencer [Mon, 11 Oct 2010 21:39:24 +0000 (21:39 +0000)]
Revert "System: Add SwapByteOrder and update Support/MathExtras.h to use it."

This reverts commit 116234.

It compiled just fine with MSVC and clang...

llvm-svn: 116242

11 years agoSwitch c-index-test from clang_codeComplete() over to
Douglas Gregor [Mon, 11 Oct 2010 21:37:58 +0000 (21:37 +0000)]
Switch c-index-test from clang_codeComplete() over to
clang_codeCompleteAt(). This uncovered a few issues with the latter:

  - ASTUnit wasn't saving/restoring diagnostic state appropriately between
    reparses and code completions.
  - "Overload" completions weren't being passed through to the client

llvm-svn: 116241

11 years agoWe're not going to handle dynamic allocas anywhere else.
Eric Christopher [Mon, 11 Oct 2010 21:37:35 +0000 (21:37 +0000)]
We're not going to handle dynamic allocas anywhere else.

llvm-svn: 116240

11 years agoChange explicit search Apple specific code to only reference __eprintf on x86.
Daniel Dunbar [Mon, 11 Oct 2010 21:34:24 +0000 (21:34 +0000)]
Change explicit search Apple specific code to only reference __eprintf on x86.

llvm-svn: 116239

11 years agoMake sure to use the machine instruction operand number. It doesn't always
Jim Grosbach [Mon, 11 Oct 2010 21:31:22 +0000 (21:31 +0000)]
Make sure to use the machine instruction operand number. It doesn't always
map one-to-one with the CodeGenInstruction operand number.

llvm-svn: 116238

11 years agoThis patch does a few things in the area of objective-c
Fariborz Jahanian [Mon, 11 Oct 2010 21:29:12 +0000 (21:29 +0000)]
This patch does a few things in the area of objective-c
1. Generates the AST for lexical info. of accessing
   getter/setter methods using dot-syntax notation.
   This fixes //rdar: //8528170.
2. Modifes rewriter to handle the AST putout in 1.
3. Supportes in rewriter ObjCImplicitSetterGetter ASTs.

llvm-svn: 116237

11 years agoReduce dpendencies for SupportTests.
Michael J. Spencer [Mon, 11 Oct 2010 21:22:34 +0000 (21:22 +0000)]
Reduce dpendencies for SupportTests.

llvm-svn: 116235

11 years agoSystem: Add SwapByteOrder and update Support/MathExtras.h to use it.
Michael J. Spencer [Mon, 11 Oct 2010 21:22:22 +0000 (21:22 +0000)]
System: Add SwapByteOrder and update Support/MathExtras.h to use it.

llvm-svn: 116234

11 years agoMake sure that the call stack adjustments have default operands. Also
Eric Christopher [Mon, 11 Oct 2010 21:20:02 +0000 (21:20 +0000)]
Make sure that the call stack adjustments have default operands. Also
leave custom lowerings for later.

Fixes some nightly tests.

llvm-svn: 116232

11 years agoIRgen/Obj-C: Fix encoding of "long double".
Daniel Dunbar [Mon, 11 Oct 2010 21:13:48 +0000 (21:13 +0000)]
IRgen/Obj-C: Fix encoding of "long double".
 - The mind boggles.

llvm-svn: 116226

11 years agoMerge encode-test-1.m into encode-test.m
Daniel Dunbar [Mon, 11 Oct 2010 21:13:46 +0000 (21:13 +0000)]
Merge encode-test-1.m into encode-test.m

llvm-svn: 116225

11 years agoFileCheck'ize test.
Daniel Dunbar [Mon, 11 Oct 2010 21:13:44 +0000 (21:13 +0000)]
FileCheck'ize test.

llvm-svn: 116224

11 years agoPR8297
Andrew Trick [Mon, 11 Oct 2010 21:08:42 +0000 (21:08 +0000)]

llvm-svn: 116223

11 years agoPowerPC varargs functions store live-in registers on the stack. Make sure we use
Jakob Stoklund Olesen [Mon, 11 Oct 2010 20:43:09 +0000 (20:43 +0000)]
PowerPC varargs functions store live-in registers on the stack. Make sure we use
virtual registers for those stores since RegAllocFast requires that each live
physreg only be used once.

This fixes PR8357.

llvm-svn: 116222

11 years agoFound a bug turning this on by default. Disable again for now.
Eric Christopher [Mon, 11 Oct 2010 20:26:21 +0000 (20:26 +0000)]
Found a bug turning this on by default. Disable again for now.

llvm-svn: 116220

11 years agoRemove now non-existent option.
Eric Christopher [Mon, 11 Oct 2010 20:21:21 +0000 (20:21 +0000)]
Remove now non-existent option.

llvm-svn: 116219

11 years agoFix help text.
Eric Christopher [Mon, 11 Oct 2010 20:15:02 +0000 (20:15 +0000)]
Fix help text.

llvm-svn: 116218

11 years agoChange flag from Enable to Disable since we're enabled by default.
Eric Christopher [Mon, 11 Oct 2010 20:05:22 +0000 (20:05 +0000)]
Change flag from Enable to Disable since we're enabled by default.
Also don't use fast-isel on non-darwin since it's untested.

llvm-svn: 116217

11 years agoAdd KillTheDoctor.
Michael J. Spencer [Mon, 11 Oct 2010 19:55:38 +0000 (19:55 +0000)]
Add KillTheDoctor.

llvm-svn: 116216

11 years agotrailing whitespace cleanup
Jim Grosbach [Mon, 11 Oct 2010 19:38:01 +0000 (19:38 +0000)]
trailing whitespace cleanup

llvm-svn: 116215

11 years agoFixes bug 8297: i386 cmpxchg8b, missing MachineMemOperand
Andrew Trick [Mon, 11 Oct 2010 19:02:04 +0000 (19:02 +0000)]
Fixes bug 8297: i386 cmpxchg8b, missing MachineMemOperand

llvm-svn: 116214