platform/upstream/llvm.git
12 years agoImplement Mike Herrick's proposed noexcept mangling.
John McCall [Tue, 25 Sep 2012 09:10:17 +0000 (09:10 +0000)]
Implement Mike Herrick's proposed noexcept mangling.

llvm-svn: 164593

12 years agoIn the MS ABI, ctors return 'this'. Patch by Dmitry Sokolov.
John McCall [Tue, 25 Sep 2012 08:00:39 +0000 (08:00 +0000)]
In the MS ABI, ctors return 'this'.  Patch by Dmitry Sokolov.

llvm-svn: 164592

12 years agoAdd the Microsoft __is_interface_class type trait.
John McCall [Tue, 25 Sep 2012 07:32:49 +0000 (07:32 +0000)]
Add the Microsoft __is_interface_class type trait.
Patch by Andy Gibbs!

llvm-svn: 164591

12 years agoFix for r163013 regression and further __interface enhancement.
John McCall [Tue, 25 Sep 2012 07:32:39 +0000 (07:32 +0000)]
Fix for r163013 regression and further __interface enhancement.
Patch by Andy Gibbs!

llvm-svn: 164590

12 years agoDuring jump-scope checking, build an ExprWithCleanups immediately
John McCall [Tue, 25 Sep 2012 06:56:03 +0000 (06:56 +0000)]
During jump-scope checking, build an ExprWithCleanups immediately
into the enclosing scope;  this is a more accurate model but is
(I believe) unnecessary in my test case due to other flaws.
However, one of those flaws is now intentional:  blocks which
appear in return statements can be trivially observed to not
extend in lifetime past the return, and so we can allow a jump
past them.  Do the necessary magic in IR-generation to make
this work.

llvm-svn: 164589

12 years agoFix an illegal tailcall opt where the callee returns a double via xmm while caller...
Evan Cheng [Tue, 25 Sep 2012 05:32:34 +0000 (05:32 +0000)]
Fix an illegal tailcall opt where the callee returns a double via xmm while caller returns x86_fp80 via st0. rdar://12229511

llvm-svn: 164588

12 years agoFix a -Wparentheses warning in the mingw build
Nico Weber [Tue, 25 Sep 2012 05:24:16 +0000 (05:24 +0000)]
Fix a -Wparentheses warning in the mingw build

llvm-svn: 164587

12 years agoFix crash when a decltype expression in a trailing return type refers to the
Richard Smith [Tue, 25 Sep 2012 04:46:05 +0000 (04:46 +0000)]
Fix crash when a decltype expression in a trailing return type refers to the
function being instantiated. An error recovery codepath was recursively
performing name lookup (and triggering an unbounded stack of template
instantiations which blew out the stack before hitting the depth limit).

Patch by Wei Pan!

llvm-svn: 164586

12 years agoFix a crash in SROA. This was reported independently by Takumi and
Chandler Carruth [Tue, 25 Sep 2012 02:42:03 +0000 (02:42 +0000)]
Fix a crash in SROA. This was reported independently by Takumi and
David (I think), but I would appreciate folks verifying that this fixes
the big crasher.

I'm still working on a reduced test case, but because this was causing
problems I wanted to get the fix checked in quickly.

llvm-svn: 164585

12 years ago<rdar://problem/9959501>
Greg Clayton [Tue, 25 Sep 2012 02:40:06 +0000 (02:40 +0000)]
<rdar://problem/9959501>

More KDP debugging process. We can not set breakpoints, hit them, resume, step and detach while running.

llvm-svn: 164584

12 years agoDon't forget that strcpy and friends return a pointer to the destination, so
Nick Lewycky [Tue, 25 Sep 2012 01:55:59 +0000 (01:55 +0000)]
Don't forget that strcpy and friends return a pointer to the destination, so
it's not a dead store if that pointer is used. Whoops!

llvm-svn: 164583

12 years agoUnit tests for IntegerDivision. Currently, just a basic sanity check to ensure that...
Michael Ilseman [Tue, 25 Sep 2012 01:33:33 +0000 (01:33 +0000)]
Unit tests for IntegerDivision. Currently, just a basic sanity check to ensure that the code was generated properly. Future work would be finding some way to test the actual result that would be computed.

llvm-svn: 164582

12 years agoDelete some code which is, as far as I can tell, dead.
Eli Friedman [Tue, 25 Sep 2012 01:02:42 +0000 (01:02 +0000)]
Delete some code which is, as far as I can tell, dead.

llvm-svn: 164580

12 years ago[analyzer] Fix a buildbot crash triggered by turning on dynamic
Anna Zaks [Tue, 25 Sep 2012 00:31:43 +0000 (00:31 +0000)]
[analyzer] Fix a buildbot crash triggered by turning on dynamic
dispatch.

llvm-svn: 164579

12 years agoDon't produce diagnostics for missing ctor-initializers during template
Richard Smith [Tue, 25 Sep 2012 00:23:05 +0000 (00:23 +0000)]
Don't produce diagnostics for missing ctor-initializers during template
instantiations if we encountered errors parsing some of the initializers.

llvm-svn: 164578

12 years agoARM: 'add Rd, pc, #imm' is an alias for 'adr Rd, #imm'.
Jim Grosbach [Tue, 25 Sep 2012 00:08:13 +0000 (00:08 +0000)]
ARM: 'add Rd, pc, #imm' is an alias for 'adr Rd, #imm'.

rdar://9795790

llvm-svn: 164577

12 years agoRemove unused name of variable to quiet a warning. Also canonicalize a
Nick Lewycky [Mon, 24 Sep 2012 23:47:23 +0000 (23:47 +0000)]
Remove unused name of variable to quiet a warning. Also canonicalize a
declaration to use the same form as in the rest of the file. No functionality
change.

llvm-svn: 164576

12 years agoApply the emulated nullptr_t with constexpr. This is an unusual configuration that...
Howard Hinnant [Mon, 24 Sep 2012 23:36:40 +0000 (23:36 +0000)]
Apply the emulated nullptr_t with constexpr.  This is an unusual configuration that would take advantage of this.  But it has popped up in the wild and does no harm to support it.

llvm-svn: 164575

12 years agoMaking sure we do not try to compare a NULL string - which would cause us to crash
Enrico Granata [Mon, 24 Sep 2012 23:30:10 +0000 (23:30 +0000)]
Making sure we do not try to compare a NULL string - which would cause us to crash

llvm-svn: 164574

12 years agoAdded a new Xcode build configuration, DebugClang,
Sean Callanan [Mon, 24 Sep 2012 23:21:18 +0000 (23:21 +0000)]
Added a new Xcode build configuration, DebugClang,
which builds a Debug+Asserts build of Clang and
links LLDB against it.  The Debug configuration
builds Clang with Release+Asserts, for faster
linking and smaller memory footprint when debugging
the build LLDB.

llvm-svn: 164573

12 years agoRemoved some debugging cruft.
Sean Callanan [Mon, 24 Sep 2012 23:11:56 +0000 (23:11 +0000)]
Removed some debugging cruft.

llvm-svn: 164572

12 years agoMark jump tables in code sections with DataRegion directives.
Jim Grosbach [Mon, 24 Sep 2012 23:06:27 +0000 (23:06 +0000)]
Mark jump tables in code sections with DataRegion directives.

Even out-of-line jump tables can be in the code section, so mark them
as data-regions for those targets which support the directives.

rdar://12362871&12362974

llvm-svn: 164571

12 years agoTypo.
Chad Rosier [Mon, 24 Sep 2012 23:04:25 +0000 (23:04 +0000)]
Typo.

llvm-svn: 164570

12 years agoHandle C++ functional casts in a similar way to C-style casts in
Eli Friedman [Mon, 24 Sep 2012 23:02:26 +0000 (23:02 +0000)]
Handle C++ functional casts in a similar way to C-style casts in
unused expression warnings.  <rdar://problem/12359208>.

llvm-svn: 164569

12 years agoUpdate for r164567.
Chad Rosier [Mon, 24 Sep 2012 22:58:50 +0000 (22:58 +0000)]
Update for r164567.

llvm-svn: 164568

12 years agoRather then have a wrapper function, have tblgen instantiate the implementation.
Chad Rosier [Mon, 24 Sep 2012 22:57:55 +0000 (22:57 +0000)]
Rather then have a wrapper function, have tblgen instantiate the implementation.
Also remove an unused argument.

llvm-svn: 164567

12 years agoobjective-C: use 'instance variables' as plural when referring
Fariborz Jahanian [Mon, 24 Sep 2012 22:51:51 +0000 (22:51 +0000)]
objective-C: use 'instance variables' as plural when referring
to the feature.

llvm-svn: 164566

12 years agoClear UNIVERSAL_SDK_PATH setting when building host tools. <rdar://12360497>
Bob Wilson [Mon, 24 Sep 2012 22:51:19 +0000 (22:51 +0000)]
Clear UNIVERSAL_SDK_PATH setting when building host tools.  <rdar://12360497>

I also moved the SDKROOT setting into the make flags, since clearing it from
the environment isn't good enough to override a setting on the make command
line.  That hasn't been a problem but it could be, and it's good to be
consistent with the way UNIVERSAL_SDK_PATH is handled.

llvm-svn: 164565

12 years agoFix string comparisons in ABIMacOSX_i386::RegisterIsCalleeSaved so
Jason Molenda [Mon, 24 Sep 2012 22:51:04 +0000 (22:51 +0000)]
Fix string comparisons in ABIMacOSX_i386::RegisterIsCalleeSaved so
that volatile registers are correctly reported for this ABI.
We were incorrectly passing up volatile registers from callee
frames.

llvm-svn: 164564

12 years agoBrought LLDB top-of-tree into sync with LLVM/Clang
Sean Callanan [Mon, 24 Sep 2012 22:25:51 +0000 (22:25 +0000)]
Brought LLDB top-of-tree into sync with LLVM/Clang
top-of-tree.  Removed all local patches and llvm.zip.

The intent is that fron now on top-of-tree will
always build against LLVM/Clang top-of-tree, and
that problems building will be resolved as they
occur.  Stable release branches of LLDB can be
constructed as needed and linked to specific release
branches of LLVM/Clang.

llvm-svn: 164563

12 years agoTeach DSE that strcpy, strncpy, strcat and strncat are all stores which may be
Nick Lewycky [Mon, 24 Sep 2012 22:09:10 +0000 (22:09 +0000)]
Teach DSE that strcpy, strncpy, strcat and strncat are all stores which may be
dead.

llvm-svn: 164561

12 years agoMove all the calls to AA.getTargetLibraryInfo() to using a TLI member variable.
Nick Lewycky [Mon, 24 Sep 2012 22:07:09 +0000 (22:07 +0000)]
Move all the calls to AA.getTargetLibraryInfo() to using a TLI member variable.
No functionality change.

llvm-svn: 164560

12 years agoobjective-C: remove use of 'ivar' in favor of
Fariborz Jahanian [Mon, 24 Sep 2012 22:00:36 +0000 (22:00 +0000)]
objective-C: remove use of 'ivar' in favor of
'instance variable' in text of all diagnostics
for objective-C: // rdar://12352442

llvm-svn: 164559

12 years agoMade TestHiddenIvars an expected failure with
Sean Callanan [Mon, 24 Sep 2012 21:34:40 +0000 (21:34 +0000)]
Made TestHiddenIvars an expected failure with
"frame variable".  "expr" finds hidden ivars
correctly.

llvm-svn: 164558

12 years agoAdd clarifying comment.
Ted Kremenek [Mon, 24 Sep 2012 21:17:14 +0000 (21:17 +0000)]
Add clarifying comment.

llvm-svn: 164557

12 years agoMore tests for r164545 (change extension warning about 'long long').
Dmitri Gribenko [Mon, 24 Sep 2012 21:09:05 +0000 (21:09 +0000)]
More tests for r164545 (change extension warning about 'long long').

llvm-svn: 164556

12 years agoStringRef'ize Preprocessor::CreateString().
Dmitri Gribenko [Mon, 24 Sep 2012 21:07:17 +0000 (21:07 +0000)]
StringRef'ize Preprocessor::CreateString().

llvm-svn: 164555

12 years agoReplace raw call to snprintf() by llvm streams.
Dmitri Gribenko [Mon, 24 Sep 2012 20:56:28 +0000 (20:56 +0000)]
Replace raw call to snprintf() by llvm streams.

llvm-svn: 164554

12 years agoSpecify MachinePointerInfo as refering to the argument value and offset of the
Roman Divacky [Mon, 24 Sep 2012 20:47:19 +0000 (20:47 +0000)]
Specify MachinePointerInfo as refering to the argument value and offset of the
store when handling byval arguments. Thus preventing reordering of the store
with load with post-RA scheduler.

llvm-svn: 164553

12 years agoReplace an assertion with an error for empty __asm statements.
Bob Wilson [Mon, 24 Sep 2012 19:57:59 +0000 (19:57 +0000)]
Replace an assertion with an error for empty __asm statements.

llvm-svn: 164551

12 years agoFix a comment typo and clean up formatting.
Bob Wilson [Mon, 24 Sep 2012 19:57:55 +0000 (19:57 +0000)]
Fix a comment typo and clean up formatting.

llvm-svn: 164550

12 years agoMake sure that we don't end up making an #undef'd macro visible after
Douglas Gregor [Mon, 24 Sep 2012 19:56:18 +0000 (19:56 +0000)]
Make sure that we don't end up making an #undef'd macro visible after
the fact. Test cases will come when we're actually (de-)serializing
macro history.

llvm-svn: 164549

12 years agoRather then have a wrapper function, have tblgen instantiate the implementation.
Chad Rosier [Mon, 24 Sep 2012 19:32:29 +0000 (19:32 +0000)]
Rather then have a wrapper function, have tblgen instantiate the implementation.

llvm-svn: 164548

12 years ago[libclang] Bring CXCursor_AsmStmt back as an alias for CXCursor_GCCAsmStmt.
Argyrios Kyrtzidis [Mon, 24 Sep 2012 19:27:20 +0000 (19:27 +0000)]
[libclang] Bring CXCursor_AsmStmt back as an alias for CXCursor_GCCAsmStmt.

This was renamed in r162632 which was badness because the C API needs to be stable.
rdar://12360096

llvm-svn: 164547

12 years agoInclude 'long long' extension warning in the LongLong group so that it can be
Dmitri Gribenko [Mon, 24 Sep 2012 19:05:10 +0000 (19:05 +0000)]
Include 'long long' extension warning in the LongLong group so that it can be
silenced by -Wno-long-long.  Thanks Richard Smith for the fix idea!

llvm-svn: 164546

12 years agoChange the wording of the extension warning from
Dmitri Gribenko [Mon, 24 Sep 2012 18:19:21 +0000 (18:19 +0000)]
Change the wording of the extension warning from
> 'long long' is an extension when C99 mode is not enabled
to
> 'long long' is a C++11 extension
while compiling in C++98 mode.

llvm-svn: 164545

12 years agoRemove a couple of debugging printf's from the testsuite.
Jim Ingham [Mon, 24 Sep 2012 18:11:54 +0000 (18:11 +0000)]
Remove a couple of debugging printf's from the testsuite.

llvm-svn: 164544

12 years ago[analyzer] Really turn on dynamic-bifurcation on by default.
Anna Zaks [Mon, 24 Sep 2012 17:43:47 +0000 (17:43 +0000)]
[analyzer] Really turn on dynamic-bifurcation on by default.

Thanks to Byoungyoung for realizing taht we are not passing the default
option correctly.

llvm-svn: 164543

12 years ago[analyzer] Remove unused.
Anna Zaks [Mon, 24 Sep 2012 17:43:44 +0000 (17:43 +0000)]
[analyzer] Remove unused.

llvm-svn: 164542

12 years ago[analyzer]Prevent infinite recursion(assume->checker:evalAssume->assume)
Anna Zaks [Mon, 24 Sep 2012 17:43:41 +0000 (17:43 +0000)]
[analyzer]Prevent infinite recursion(assume->checker:evalAssume->assume)

(Unfortunately, I do not have a good reduced test case for this.)

llvm-svn: 164541

12 years agoAdd missing : in CHECK line.
Richard Osborne [Mon, 24 Sep 2012 17:22:43 +0000 (17:22 +0000)]
Add missing : in CHECK line.

llvm-svn: 164540

12 years agoAdd missing check for presence of target data.
Richard Osborne [Mon, 24 Sep 2012 17:10:03 +0000 (17:10 +0000)]
Add missing check for presence of target data.

This avoids a crash in visitAllocaInst when target data isn't available.

llvm-svn: 164539

12 years agoUpdating email address
Marshall Clow [Mon, 24 Sep 2012 14:27:24 +0000 (14:27 +0000)]
Updating email address

llvm-svn: 164490

12 years agoUpdating email address
Marshall Clow [Mon, 24 Sep 2012 14:27:10 +0000 (14:27 +0000)]
Updating email address

llvm-svn: 164489

12 years ago[TSan] one more attempt to fix sleep_sync test: call sleep in the spawned thread...
Alexey Samsonov [Mon, 24 Sep 2012 13:35:46 +0000 (13:35 +0000)]
[TSan] one more attempt to fix sleep_sync test: call sleep in the spawned thread so that sleep-synchronization will be detected even if child thread is started late.

llvm-svn: 164488

12 years ago[TSan] Provide replacements for operators new/delete instead of declaring extern...
Alexey Samsonov [Mon, 24 Sep 2012 13:19:47 +0000 (13:19 +0000)]
[TSan] Provide replacements for operators new/delete instead of declaring extern C functions with weirdly mangled names (same strategy is used in ASan).

llvm-svn: 164487

12 years ago[ASan] Apply some ASan-relevant pieces of patch by Ruben Van Boxem. In the same time...
Alexey Samsonov [Mon, 24 Sep 2012 11:43:40 +0000 (11:43 +0000)]
[ASan] Apply some ASan-relevant pieces of patch by Ruben Van Boxem. In the same time, remove ASan from CMake build on Windows after conversation with Timur. We don't want to support building ASan on Windows until it is in a working state.

llvm-svn: 164486

12 years agoUse an explicit target to test that source fortification is off when building for...
Alexander Potapenko [Mon, 24 Sep 2012 10:25:24 +0000 (10:25 +0000)]
Use an explicit target to test that source fortification is off when building for Darwin with -faddress-sanitizer.

llvm-svn: 164485

12 years agoSmall cleanup of literal semantic analysis: hiding 'char *' pointers behind
Dmitri Gribenko [Mon, 24 Sep 2012 09:53:54 +0000 (09:53 +0000)]
Small cleanup of literal semantic analysis: hiding 'char *' pointers behind
StringRef makes code cleaner.  Also, make the temporary buffer smaller:
512 characters is unreasonably large for integer literals.

llvm-svn: 164484

12 years ago[TSan] increase sleep time in test to ensure correct event ordering
Alexey Samsonov [Mon, 24 Sep 2012 08:57:41 +0000 (08:57 +0000)]
[TSan] increase sleep time in test to ensure correct event ordering

llvm-svn: 164483

12 years agoUpdate the Clang tests which run with a full optimization pipeline for
Chandler Carruth [Mon, 24 Sep 2012 01:10:25 +0000 (01:10 +0000)]
Update the Clang tests which run with a full optimization pipeline for
the new SROA pass. This is a benign change: the order of PHI nodes
changed.

llvm-svn: 164481

12 years agoEnable the new SROA pass by default.
Chandler Carruth [Mon, 24 Sep 2012 01:10:25 +0000 (01:10 +0000)]
Enable the new SROA pass by default.

Queue the fallout. ;]

llvm-svn: 164480

12 years agoAddress one of the original FIXMEs for the new SROA pass by implementing
Chandler Carruth [Mon, 24 Sep 2012 00:34:20 +0000 (00:34 +0000)]
Address one of the original FIXMEs for the new SROA pass by implementing
integer promotion analogous to vector promotion. When there is an
integer alloca being accessed both as its integer type and as a narrower
integer type, promote the narrower access to "insert" and "extract" the
smaller integer from the larger one, and make the integer alloca
a candidate for promotion.

In the new formulation, we don't care about target legal integer or use
thresholds to control things. Instead, we only perform this promotion to
an integer type which the frontend has already emitted a load or store
for. This bounds the scope and prevents optimization passes from
coalescing larger and larger entities into a single integer.

llvm-svn: 164479

12 years agounittests/AST/StmtPrinterTest.cpp: Suppress a LP64-assumed test, "0x100000000i128...
NAKAMURA Takumi [Mon, 24 Sep 2012 00:34:18 +0000 (00:34 +0000)]
unittests/AST/StmtPrinterTest.cpp: Suppress a LP64-assumed test, "0x100000000i128 => 4294967296L", for now.

LONG_MAX is 2147483647L on common 32 bit and LLP64 (Windows x64).

llvm-svn: 164478

12 years agoAs a followup for r164303, add some tests for printing literals that test
Dmitri Gribenko [Sun, 23 Sep 2012 20:29:07 +0000 (20:29 +0000)]
As a followup for r164303, add some tests for printing literals that test
printing directly rather than through a complicated machinery of ObjC rewriter.

llvm-svn: 164477

12 years agoEmit dtors into proper section while compiling in vcpp-compatible mode.
Anton Korobeynikov [Sun, 23 Sep 2012 15:53:47 +0000 (15:53 +0000)]
Emit dtors into proper section while compiling in vcpp-compatible mode.
Patch by Kai!

llvm-svn: 164476

12 years agoSwitch to a signed representation for the dynamic offsets while walking
Chandler Carruth [Sun, 23 Sep 2012 11:43:14 +0000 (11:43 +0000)]
Switch to a signed representation for the dynamic offsets while walking
across the uses of the alloca. It's entirely possible for negative
numbers to come up here, and in some rare cases simply doing the 2's
complement arithmetic isn't the correct decision. Notably, we can't zext
the index of the GEP. The definition of GEP is that these offsets are
sign extended or truncated to the size of the pointer, and then wrapping
2's complement arithmetic used.

This patch fixes an issue that comes up with *no* input from the
buildbots or bootstrap afaict. The only place where it manifested,
disturbingly, is Clang's own regression test suite. A reduced and
targeted collection of tests are added to cope with this. Note that I've
tried to pin down the potential cases of overflow, but may have missed
some cases. I've tried to add a few cases to test this, but its hard
because LLVM has quite limited support for >64bit constructs.

llvm-svn: 164475

12 years agoDon't do actual work inside an assert statement. Fixes PR11760!
Nick Lewycky [Sun, 23 Sep 2012 03:58:21 +0000 (03:58 +0000)]
Don't do actual work inside an assert statement. Fixes PR11760!

llvm-svn: 164474

12 years agoImplement ToolChain::IsUnwindTablesDefault to reduce code duplication a bit.
Rafael Espindola [Sun, 23 Sep 2012 03:05:41 +0000 (03:05 +0000)]
Implement ToolChain::IsUnwindTablesDefault to reduce code duplication a bit.

llvm-svn: 164473

12 years agoRevise test to avoid using of 'grep'
Michael Liao [Sun, 23 Sep 2012 02:41:47 +0000 (02:41 +0000)]
Revise test to avoid using of 'grep'

llvm-svn: 164472

12 years agoAdd LLVM_OVERRIDE to methods that override their base classes.
Craig Topper [Sun, 23 Sep 2012 02:12:10 +0000 (02:12 +0000)]
Add LLVM_OVERRIDE to methods that override their base classes.

llvm-svn: 164471

12 years agoOn Hexagon getArchName() never returns x86_64, simplify the function.
Rafael Espindola [Sat, 22 Sep 2012 22:58:57 +0000 (22:58 +0000)]
On Hexagon getArchName() never returns x86_64, simplify the function.

llvm-svn: 164470

12 years agoUse pipes.
Rafael Espindola [Sat, 22 Sep 2012 22:51:00 +0000 (22:51 +0000)]
Use pipes.

llvm-svn: 164469

12 years agoUse FileCheck.
Rafael Espindola [Sat, 22 Sep 2012 22:30:04 +0000 (22:30 +0000)]
Use FileCheck.

llvm-svn: 164468

12 years agoComment sema: warn when comment has \deprecated but declaration does not have a
Dmitri Gribenko [Sat, 22 Sep 2012 21:47:50 +0000 (21:47 +0000)]
Comment sema: warn when comment has \deprecated but declaration does not have a
deprecation attribute ('deprecated', 'availability' or 'unavailable').

This warning is under a separate flag, -Wdocumentation-deprecated-sync, so it
can be turned off easily while leaving other -Wdocumentation warnings on.

llvm-svn: 164467

12 years agoRemove superfluous \namespace command.
Dmitri Gribenko [Sat, 22 Sep 2012 21:44:37 +0000 (21:44 +0000)]
Remove superfluous \namespace command.

llvm-svn: 164466

12 years agoEnhance test case of atomic16 to verify inst encoding fixed in r164453.
Michael Liao [Sat, 22 Sep 2012 21:07:59 +0000 (21:07 +0000)]
Enhance test case of atomic16 to verify inst encoding fixed in r164453.

llvm-svn: 164465

12 years agoFix cindex.py compatibility with older libclang.so
Dmitri Gribenko [Sat, 22 Sep 2012 17:52:29 +0000 (17:52 +0000)]
Fix cindex.py compatibility with older libclang.so

The issue is that we were calling clang_getCompletionBriefComment
unconditionally.  New we check if this function is available before calling it.

llvm-svn: 164464

12 years agoUse FileCheck.
Rafael Espindola [Sat, 22 Sep 2012 15:26:38 +0000 (15:26 +0000)]
Use FileCheck.

llvm-svn: 164462

12 years agoFix pr13749. There is still a lot of code using getArchName that should be
Rafael Espindola [Sat, 22 Sep 2012 15:04:11 +0000 (15:04 +0000)]
Fix pr13749. There is still a lot of code using getArchName that should be
using getArch, but I will try to fix them one at time to add tests.

llvm-svn: 164460

12 years agoARMInstPrinter.cpp: Fix a warning in -Asserts. [-Wunused-variable]
NAKAMURA Takumi [Sat, 22 Sep 2012 13:12:28 +0000 (13:12 +0000)]
ARMInstPrinter.cpp: Fix a warning in -Asserts. [-Wunused-variable]

llvm-svn: 164459

12 years agoWhitespace.
NAKAMURA Takumi [Sat, 22 Sep 2012 13:12:22 +0000 (13:12 +0000)]
Whitespace.

llvm-svn: 164458

12 years agoFix edge cases of ARM shift operands in arith instructions.
Tim Northover [Sat, 22 Sep 2012 11:18:19 +0000 (11:18 +0000)]
Fix edge cases of ARM shift operands in arith instructions.

As before with load instructions, oddities like "asr #32", "rrx" could
be printed incorrectly.

Patch by Chris Lidbury.

llvm-svn: 164456

12 years agoFix the handling of edge cases in ARM shifted operands.
Tim Northover [Sat, 22 Sep 2012 11:18:12 +0000 (11:18 +0000)]
Fix the handling of edge cases in ARM shifted operands.

This patch fixes load/store instructions to handle less common cases
like "asr #32", "rrx" properly throughout the MC layer.

Patch by Chris Lidbury.

llvm-svn: 164455

12 years agoAvoid multiple atomic builtin declaration.
Abramo Bagnara [Sat, 22 Sep 2012 09:05:22 +0000 (09:05 +0000)]
Avoid multiple atomic builtin declaration.

llvm-svn: 164454

12 years agoFix 16-bit atomic inst encoding and keep pseudo-inst starting with '#'
Michael Liao [Sat, 22 Sep 2012 05:41:15 +0000 (05:41 +0000)]
Fix 16-bit atomic inst encoding and keep pseudo-inst starting with '#'

llvm-svn: 164453

12 years agoFix typo in r164357
Michael Liao [Sat, 22 Sep 2012 03:39:42 +0000 (03:39 +0000)]
Fix typo in r164357

llvm-svn: 164452

12 years agoMachine Model (-schedmodel only). Added SchedAliases.
Andrew Trick [Sat, 22 Sep 2012 02:24:21 +0000 (02:24 +0000)]
Machine Model (-schedmodel only). Added SchedAliases.

Allow subtargets to tie SchedReadWrite types to processor specific
sequences or variants.

llvm-svn: 164451

12 years agoWhen importing a FunctionProtoType::ExtProtoInfo, its ExceptionSpecDecl can point...
Argyrios Kyrtzidis [Sat, 22 Sep 2012 01:58:06 +0000 (01:58 +0000)]
When importing a FunctionProtoType::ExtProtoInfo, its ExceptionSpecDecl can point to the
FunctionDecl that we are importing the FunctionProtoType for, in which case we'll have
infinite recursion when importing.

Initially create a FunctionProtoType with null ExceptionSpecDecl/ExceptionSpecTemplate and
update the type in ASTNodeImporter::VisitFunctionDecl after the FunctionDecl has been created.

llvm-svn: 164450

12 years ago[analyzer] Suppress bugs whose paths go through the return of a null pointer.
Jordan Rose [Sat, 22 Sep 2012 01:25:06 +0000 (01:25 +0000)]
[analyzer] Suppress bugs whose paths go through the return of a null pointer.

This is a heuristic intended to greatly reduce the number of false
positives resulting from inlining, particularly inlining of generic,
defensive C++ methods that live in header files. The suppression is
triggered in the cases where we ask to track where a null pointer came
from, and it turns out that the source of the null pointer was an inlined
function call.

This change brings the number of bug reports in LLVM from ~1500 down to
around ~300, a much more manageable number. Yes, some true positives may
be hidden as well, but from what I looked at the vast majority of silenced
reports are false positives, and many of the true issues found by the
analyzer are still reported.

I'm hoping to improve this heuristic further by adding some exceptions
next week (cases in which a bug should still be reported).

llvm-svn: 164449

12 years ago[analyzer] Track a null value back through FindLastStoreBRVisitor.
Jordan Rose [Sat, 22 Sep 2012 01:25:00 +0000 (01:25 +0000)]
[analyzer] Track a null value back through FindLastStoreBRVisitor.

Also, tidy up the other tracking visitors so that they mark the right
things as interesting and don't do extra work.

llvm-svn: 164448

12 years ago[analyzer] Always allow BugReporterVisitors to see the bug path.
Jordan Rose [Sat, 22 Sep 2012 01:24:56 +0000 (01:24 +0000)]
[analyzer] Always allow BugReporterVisitors to see the bug path.

Before, PathDiagnosticConsumers that did not support actual path output
would (sensibly) cause the generation of the full path to be skipped.
However, BugReporterVisitors may want to see the path in order to mark a
BugReport as invalid.

Now, even for a path generation scheme of 'None' we will still create a
trimmed graph and walk backwards through the bug path, doing no work other
than passing the nodes to the BugReporterVisitors. This isn't cheap, but
it's necessary to properly do suppression when the first path consumer does
not support path notes.

In the future, we should try only generating the path and visitor-provided
path notes once, or at least only creating the trimmed graph once.

llvm-svn: 164447

12 years ago[analyzer] Allow a BugReport to be marked "invalid" during path generation.
Jordan Rose [Sat, 22 Sep 2012 01:24:53 +0000 (01:24 +0000)]
[analyzer] Allow a BugReport to be marked "invalid" during path generation.

This is intended to allow visitors to make decisions about whether a
BugReport is likely a false positive. Currently there are no visitors
making use of this feature, so there are no tests.

When a BugReport is marked invalid, the invalidator must provide a key
that identifies the invaliation (intended to be the visitor type and a
context pointer of some kind). This allows us to reverse the decision
later on. Being able to reverse a decision about invalidation gives us more
flexibility, and allows us to formulate conditions like "this report is
invalid UNLESS the original argument is 'foo'". We can use this to
fine-tune our false-positive suppression (coming soon).

llvm-svn: 164446

12 years ago[analyzer] Look through OpaqueValueExprs when tracking a nil value.
Jordan Rose [Sat, 22 Sep 2012 01:24:49 +0000 (01:24 +0000)]
[analyzer] Look through OpaqueValueExprs when tracking a nil value.

This allows us to show /why/ a particular object is nil, even when it is
wrapped in an OpaqueValueExpr.

llvm-svn: 164445

12 years ago[analyzer] Better path notes for null pointers passed as arguments.
Jordan Rose [Sat, 22 Sep 2012 01:24:46 +0000 (01:24 +0000)]
[analyzer] Better path notes for null pointers passed as arguments.

Rather than saying "Null pointer value stored to 'foo'", we now say
"Passing null pointer value via Nth parameter 'foo'", which is much better.
The note is also now on the argument expression as well, rather than the
entire call.

This paves the way for continuing to track arguments back to their sources.

<rdar://problem/12211490>

llvm-svn: 164444

12 years agoUse llvm::getOrdinalSuffix to print ordinal numbers in diagnostics.
Jordan Rose [Sat, 22 Sep 2012 01:24:42 +0000 (01:24 +0000)]
Use llvm::getOrdinalSuffix to print ordinal numbers in diagnostics.

Just a refactoring of common infrastructure. No intended functionality change.

llvm-svn: 164443

12 years ago[analyzer] Check that an ObjCIvarRefExpr's base is non-null even as an lvalue.
Jordan Rose [Sat, 22 Sep 2012 01:24:38 +0000 (01:24 +0000)]
[analyzer] Check that an ObjCIvarRefExpr's base is non-null even as an lvalue.

Like with struct fields, we want to catch cases like this early,
so that we can produce better diagnostics and path notes:

  PointObj *p = nil;
  int *px = &p->_x; // should warn here
  *px = 1;

llvm-svn: 164442

12 years ago[analyzer] Check that a member expr is valid even when the result is an lvalue.
Jordan Rose [Sat, 22 Sep 2012 01:24:33 +0000 (01:24 +0000)]
[analyzer] Check that a member expr is valid even when the result is an lvalue.

We want to catch cases like this early, so that we can produce better
diagnostics and path notes:

  Point *p = 0;
  int *px = &p->x; // should warn here
  *px = 1;

llvm-svn: 164441

12 years agoAdd llvm::getOrdinalSuffix to get the appropriate -st, -nd, -rd, -th suffix.
Jordan Rose [Sat, 22 Sep 2012 01:24:21 +0000 (01:24 +0000)]
Add llvm::getOrdinalSuffix to get the appropriate -st, -nd, -rd, -th suffix.

Used by clang to print parameter indexes.

llvm-svn: 164440

12 years agoCasting: assert that pointer arguments to isa<> are non-null.
Jordan Rose [Sat, 22 Sep 2012 01:24:18 +0000 (01:24 +0000)]
Casting: assert that pointer arguments to isa<> are non-null.

This silences several analyzer warnings within LLVM, and provides a slightly
nicer crash experience when someone calls isa<>, cast<>, or dyn_cast<> with
a null pointer.

llvm-svn: 164439