platform/upstream/llvm.git
11 years agoAdd a V8FP instruction 'vcvt{b,t}' to convert between half and double precision.
Joey Gouly [Thu, 4 Jul 2013 10:04:08 +0000 (10:04 +0000)]
Add a V8FP instruction 'vcvt{b,t}' to convert between half and double precision.

llvm-svn: 185620

11 years agoTry to guard a test that requires DEBUG().
Rui Ueyama [Thu, 4 Jul 2013 09:29:47 +0000 (09:29 +0000)]
Try to guard a test that requires DEBUG().

This is a follow-up patch for r185524. Being assert enabled does not mean
that DEBUG() is enabled, so we need to check the existence of DEBUG() itself.

llvm-svn: 185619

11 years agoTypo.
Jakob Stoklund Olesen [Thu, 4 Jul 2013 04:53:49 +0000 (04:53 +0000)]
Typo.

llvm-svn: 185618

11 years agoSimplify landing pad lowering.
Jakob Stoklund Olesen [Thu, 4 Jul 2013 04:53:45 +0000 (04:53 +0000)]
Simplify landing pad lowering.

Stop using the ISD::EXCEPTIONADDR and ISD::EHSELECTION when lowering
landing pad arguments. These nodes were previously legalized into
CopyFromReg nodes, but that never worked properly because the
CopyFromReg node weren't guaranteed to be  scheduled at the top of the
basic block.

This meant the exception pointer and selector registers could be
clobbered before being copied to a virtual register.

This patch copies the two physical registers to virtual registers at
the beginning of the basic block, and lowers the landingpad instruction
directly to two CopyFromReg nodes reading the *virtual* registers. This
is safe because virtual registers don't get clobbered.

A future patch will remove the ISD::EXCEPTIONADDR and ISD::EHSELECTION
nodes.

llvm-svn: 185617

11 years agoFastISel can only apend to basic blocks.
Jakob Stoklund Olesen [Thu, 4 Jul 2013 04:32:39 +0000 (04:32 +0000)]
FastISel can only apend to basic blocks.

Compute the insertion point from the end of the basic block instead of
skipping labels from the front.

This caused failures in landing pads when live-in copies where inserted
before instruction selection.

llvm-svn: 185616

11 years agoLive-in copies go *after* EH_LABELs.
Jakob Stoklund Olesen [Thu, 4 Jul 2013 04:32:35 +0000 (04:32 +0000)]
Live-in copies go *after* EH_LABELs.

This will soon be tested by exception handling working at all.

llvm-svn: 185615

11 years agoAdd test for PR4997. This has been fixed for a while.
Richard Smith [Thu, 4 Jul 2013 04:10:46 +0000 (04:10 +0000)]
Add test for PR4997. This has been fixed for a while.

llvm-svn: 185614

11 years agoTestcase for PR14130, which was probably fixed by r183859.
Richard Smith [Thu, 4 Jul 2013 04:04:20 +0000 (04:04 +0000)]
Testcase for PR14130, which was probably fixed by r183859.

llvm-svn: 185613

11 years agoTabs to spaces. No functionality change.
Nick Lewycky [Thu, 4 Jul 2013 03:51:53 +0000 (03:51 +0000)]
Tabs to spaces. No functionality change.

llvm-svn: 185612

11 years agoAdd a space between closing template '>' to unbreak build.
Craig Topper [Thu, 4 Jul 2013 03:15:42 +0000 (03:15 +0000)]
Add a space between closing template '>' to unbreak build.

llvm-svn: 185611

11 years agoUse SmallVectorImpl instead of SmallVector for iterators and references to avoid...
Craig Topper [Thu, 4 Jul 2013 03:08:24 +0000 (03:08 +0000)]
Use SmallVectorImpl instead of SmallVector for iterators and references to avoid specifying the vector size unnecessarily.

llvm-svn: 185610

11 years ago[analyzer] Suppress reports reported in std::list
Anna Zaks [Thu, 4 Jul 2013 02:38:10 +0000 (02:38 +0000)]
[analyzer] Suppress reports reported in std::list

The motivation is to suppresses false use-after-free reports that occur when calling
std::list::pop_front() or std::list::pop_back() twice. The analyzer does not
reason about the internal invariants of the list implementation, so just do not report
any of warnings in std::list.

Fixes radar://14317928.

llvm-svn: 185609

11 years ago[analyzer] Make sure that inlined defensive checks work on div by zero.
Anna Zaks [Thu, 4 Jul 2013 02:38:06 +0000 (02:38 +0000)]
[analyzer] Make sure that inlined defensive checks work on div by zero.

This suppresses a false positive in std::hash_map.
Fixes  radar://14255587.

llvm-svn: 185608

11 years agoAdd a space between closing template '>' to unbreak build.
Craig Topper [Thu, 4 Jul 2013 01:43:17 +0000 (01:43 +0000)]
Add a space between closing template '>' to unbreak build.

llvm-svn: 185607

11 years agoUse SmallVectorImpl::iterator/const_iterator instead of SmallVector to avoid specifyi...
Craig Topper [Thu, 4 Jul 2013 01:31:24 +0000 (01:31 +0000)]
Use SmallVectorImpl::iterator/const_iterator instead of SmallVector to avoid specifying the vector size.

llvm-svn: 185606

11 years agoReapply r185601 with a fix for the cmake build.
Eric Christopher [Thu, 4 Jul 2013 01:10:38 +0000 (01:10 +0000)]
Reapply r185601 with a fix for the cmake build.

llvm-svn: 185605

11 years agoPart of PR15673: If a function template has a default argument in which
Richard Smith [Thu, 4 Jul 2013 01:01:24 +0000 (01:01 +0000)]
Part of PR15673: If a function template has a default argument in which
substitution failed, report that as a substitution failure rather than
pretending that there was no default argument.

The test cases in PR15673 have exposed some pre-existing poor diagnostics here.

llvm-svn: 185604

11 years agoTemporarily revert 185601 as it caused cmake build regressions.
Eric Christopher [Thu, 4 Jul 2013 00:51:26 +0000 (00:51 +0000)]
Temporarily revert 185601 as it caused cmake build regressions.

llvm-svn: 185603

11 years agoImprove -Wlogical-not-parentheses to catch when the not is applied to an enum.
Richard Trieu [Thu, 4 Jul 2013 00:50:18 +0000 (00:50 +0000)]
Improve -Wlogical-not-parentheses to catch when the not is applied to an enum.

llvm-svn: 185602

11 years agoAdd support for futimens for platforms that don't support futimes.
Eric Christopher [Thu, 4 Jul 2013 00:47:09 +0000 (00:47 +0000)]
Add support for futimens for platforms that don't support futimes.

Patch by pashev.igor.

llvm-svn: 185601

11 years agoRevert r185595-185596 which broke buildbots.
Jakob Stoklund Olesen [Thu, 4 Jul 2013 00:26:30 +0000 (00:26 +0000)]
Revert r185595-185596 which broke buildbots.

Revert "Simplify landing pad lowering."
Revert "Remove the EXCEPTIONADDR, EHSELECTION, and LSDAADDR ISD opcodes."

llvm-svn: 185600

11 years ago[ObjectiveC migrator] relax the rules for setter/getter
Fariborz Jahanian [Thu, 4 Jul 2013 00:24:32 +0000 (00:24 +0000)]
[ObjectiveC migrator] relax the rules for setter/getter
types when deciding on validity of a property
inclusion. // rdar://14345082

llvm-svn: 185599

11 years agoPR16480: Reimplement token-caching for constructor initializer lists. This
Richard Smith [Thu, 4 Jul 2013 00:13:48 +0000 (00:13 +0000)]
PR16480: Reimplement token-caching for constructor initializer lists. This
previously didn't work if a mem-initializer-id had a template argument which
contained parentheses or braces.

We now implement a simple rule: just look for a ') {' or '} {' that is not
nested. The '{' is assumed to start the function-body. There are still two
cases which we misparse, where the ') {' comes from a compound literal or
from a lambda. The former case is not valid C++, and the latter will probably
not be valid C++ once DR1607 is resolved, so these seem to be of low value,
and we do not regress on them with this change. EDG and g++ also misparse
both of these cases.

llvm-svn: 185598

11 years agoPatch for N3655 (Transformation type traits) with Howard's additions
Marshall Clow [Thu, 4 Jul 2013 00:10:01 +0000 (00:10 +0000)]
Patch for N3655 (Transformation type traits) with Howard's additions

llvm-svn: 185597

11 years agoRemove the EXCEPTIONADDR, EHSELECTION, and LSDAADDR ISD opcodes.
Jakob Stoklund Olesen [Wed, 3 Jul 2013 23:56:31 +0000 (23:56 +0000)]
Remove the EXCEPTIONADDR, EHSELECTION, and LSDAADDR ISD opcodes.

These exception-related opcodes are not used any longer.

llvm-svn: 185596

11 years agoSimplify landing pad lowering.
Jakob Stoklund Olesen [Wed, 3 Jul 2013 23:56:24 +0000 (23:56 +0000)]
Simplify landing pad lowering.

Stop using the ISD::EXCEPTIONADDR and ISD::EHSELECTION when lowering
landing pad arguments. These nodes were previously legalized into
CopyFromReg nodes, but that never worked properly because the
CopyFromReg node weren't guaranteed to be  scheduled at the top of the
basic block.

This meant the exception pointer and selector registers could be
clobbered before being copied to a virtual register.

This patch copies the two physical registers to virtual registers at
the beginning of the basic block, and lowers the landingpad instruction
directly to two CopyFromReg nodes reading the *virtual* registers. This
is safe because virtual registers don't get clobbered.

A future patch will remove the ISD::EXCEPTIONADDR and ISD::EHSELECTION
nodes.

llvm-svn: 185595

11 years agoAdd MachineBasicBlock::addLiveIn().
Jakob Stoklund Olesen [Wed, 3 Jul 2013 23:56:20 +0000 (23:56 +0000)]
Add MachineBasicBlock::addLiveIn().

This function adds a live-in physical register to an MBB and ensures
that it is copied to a virtual register immediately.

llvm-svn: 185594

11 years agoMinor refactoring of my last patch.
Fariborz Jahanian [Wed, 3 Jul 2013 23:44:11 +0000 (23:44 +0000)]
Minor refactoring of my last patch.

llvm-svn: 185593

11 years agoHave ARMBaseRegisterInfo::getCallPreservedMask return the 'correct' mask for the...
Stephen Lin [Wed, 3 Jul 2013 23:39:13 +0000 (23:39 +0000)]
Have ARMBaseRegisterInfo::getCallPreservedMask return the 'correct' mask for the GHC calling convention.

This is purely academic because GHC calls are always tail calls so the register mask will never be used; however, this change makes the code clearer and brings the ARM implementation of the GHC calling convention in line with the X86 implementation. Also, it might save someone else some time trying to figuring out what is happening...

llvm-svn: 185592

11 years ago[ObjectiveC Migration]: Provide knobs for
Fariborz Jahanian [Wed, 3 Jul 2013 23:05:00 +0000 (23:05 +0000)]
[ObjectiveC Migration]: Provide knobs for
migrating setter/getter methods to an eventual
property declaraiton. This is wip.
// rdar://14345082

llvm-svn: 185591

11 years agoscop detection: early return
Sebastian Pop [Wed, 3 Jul 2013 22:50:36 +0000 (22:50 +0000)]
scop detection: early return

to reduce indentation level
No functionality changed.

llvm-svn: 185590

11 years agoHoist all of the Entry.getLoc() calls int a single variable.
Eric Christopher [Wed, 3 Jul 2013 22:40:21 +0000 (22:40 +0000)]
Hoist all of the Entry.getLoc() calls int a single variable.

llvm-svn: 185589

11 years agoMake DotDebugLocEntry a class, reorder the members along with comments
Eric Christopher [Wed, 3 Jul 2013 22:40:18 +0000 (22:40 +0000)]
Make DotDebugLocEntry a class, reorder the members along with comments
for them and update all uses.

llvm-svn: 185588

11 years ago[ARM] Improve the instruction selection of vector loads.
Quentin Colombet [Wed, 3 Jul 2013 21:42:57 +0000 (21:42 +0000)]
[ARM] Improve the instruction selection of vector loads.

In the ARM back-end, build_vector nodes are lowered to a target specific
build_vector that uses floating point type.
This works well, unless the inserted bitcasts survive until instruction
selection. In that case, they incur moves between integer unit and floating
point unit that may result in inefficient code.

In other words, this conversion may introduce artificial dependencies when the
code leading to the build vector cannot be completed with a floating point type.

In particular, this happens when loads are not aligned.

Before this patch, in that case, the compiler generates general purpose loads
and creates the floating point vector from them, instead of directly using the
vector unit.

The patch uses a vector friendly sequence of code when the inserted bitcasts to
floating point survived DAGCombine.

This is done by a target specific DAGCombine that changes the target specific
build_vector into a sequence of insert_vector_elt that get rid of the bitcasts.

<rdar://problem/14170854>

llvm-svn: 185587

11 years agoElaborate on comment.
Eric Christopher [Wed, 3 Jul 2013 21:37:03 +0000 (21:37 +0000)]
Elaborate on comment.

llvm-svn: 185586

11 years agoAdd names to the header file since they help in documenting the API
Eric Christopher [Wed, 3 Jul 2013 21:23:59 +0000 (21:23 +0000)]
Add names to the header file since they help in documenting the API
(and for consistency).

llvm-svn: 185585

11 years agoCheck LongDoubleFormat instead of just Width as this is PowerPC specific.
Roman Divacky [Wed, 3 Jul 2013 21:08:41 +0000 (21:08 +0000)]
Check LongDoubleFormat instead of just Width as this is PowerPC specific.

llvm-svn: 185584

11 years ago[PowerPC] FreeBSD does not require f128 in its data layout string.
Bill Schmidt [Wed, 3 Jul 2013 21:03:35 +0000 (21:03 +0000)]
[PowerPC] FreeBSD does not require f128 in its data layout string.

Long double is 64 bits on FreeBSD PPC, so the f128 entry is superfluous.

llvm-svn: 185583

11 years ago[PowerPC] FreeBSD does not require f128 in its data layout string.
Bill Schmidt [Wed, 3 Jul 2013 21:03:06 +0000 (21:03 +0000)]
[PowerPC] FreeBSD does not require f128 in its data layout string.

Long double is 64 bits on FreeBSD PPC, so the f128 entry is superfluous.

llvm-svn: 185582

11 years agoAdd platform specific tests doc
Renato Golin [Wed, 3 Jul 2013 20:56:33 +0000 (20:56 +0000)]
Add platform specific tests doc

llvm-svn: 185581

11 years ago"bool" should be a context-sensitive keyword in Altivec mode.
Bill Schmidt [Wed, 3 Jul 2013 20:54:09 +0000 (20:54 +0000)]
"bool" should be a context-sensitive keyword in Altivec mode.

PR16456 reported that Clang implements a hybrid between AltiVec's
"Keyword and Predefine Method" and its "Context Sensitive Keyword
Method," where "bool" is always a keyword, but "vector" and "pixel"
are context-sensitive keywords.  This isn't permitted by the AltiVec
spec.  For consistency with gcc, this patch implements the Context
Sensitive Keyword Method for bool, and stops treating true and false
as keywords in Altivec mode.

The patch removes KEYALTIVEC as a trigger for defining these keywords
in include/clang/Basic/TokenKinds.def, and adds logic for "vector
bool" that mirrors the existing logic for "vector pixel."  The test
case is taken from the bug report.

llvm-svn: 185580

11 years agoRemove @expectedFailureGcc from TestInlineStepping as function prologue bug is not...
Daniel Malea [Wed, 3 Jul 2013 20:51:44 +0000 (20:51 +0000)]
Remove @expectedFailureGcc from TestInlineStepping as function prologue bug is not reproducible anymore.

llvm-svn: 185579

11 years agoAdd support for TF/TC modes available on eg. PowerPC64.
Roman Divacky [Wed, 3 Jul 2013 20:48:06 +0000 (20:48 +0000)]
Add support for TF/TC modes available on eg. PowerPC64.

llvm-svn: 185578

11 years agoUpdate testing cases to check dwarf-2 for Darwin.
Manman Ren [Wed, 3 Jul 2013 20:45:07 +0000 (20:45 +0000)]
Update testing cases to check dwarf-2 for Darwin.

llvm-svn: 185577

11 years agoSkip Test-rdar-9974002 with Clang 3.4 (due to llvm.org/pr16214)
Daniel Malea [Wed, 3 Jul 2013 20:44:40 +0000 (20:44 +0000)]
Skip Test-rdar-9974002 with Clang 3.4 (due to llvm.org/pr16214)
- should resolve remaining failures on clang buildbot

llvm-svn: 185576

11 years agoARM: Prevent ARMAsmParser::shouldOmitCCOutOperand() from misidentifying certain Thumb...
Tilmann Scheller [Wed, 3 Jul 2013 20:38:01 +0000 (20:38 +0000)]
ARM: Prevent ARMAsmParser::shouldOmitCCOutOperand() from misidentifying certain Thumb2 add immediate T3 encodings.

Before the fix Thumb2 instructions of type "add rD, rN, #imm" (T3 encoding, see ARM ARM A8.8.4) with rD and rN both being low registers (r0-r7) were classified as having the T4 encoding.

The T4 encoding doesn't have a cc_out operand so for above instructions the operand gets erroneously removed, corrupting the token stream and leading to parse errors later in the process.

This bug prevented "add r1, r7, #0xcbcbcbcb" from being assembled correctly.

Fixes <rdar://problem/14224440>.

llvm-svn: 185575

11 years agoRevert r185557 as it was a bit (a lot) premature.
Chad Rosier [Wed, 3 Jul 2013 20:37:50 +0000 (20:37 +0000)]
Revert r185557 as it was a bit (a lot) premature.

llvm-svn: 185574

11 years agoMove typedefs inside the class that they belong to.
Eric Christopher [Wed, 3 Jul 2013 20:36:36 +0000 (20:36 +0000)]
Move typedefs inside the class that they belong to.

llvm-svn: 185573

11 years agoDont define __LONG_DOUBLE_128__ unless LongDoubleWidth is really 128bits width.
Roman Divacky [Wed, 3 Jul 2013 19:45:54 +0000 (19:45 +0000)]
Dont define __LONG_DOUBLE_128__ unless LongDoubleWidth is really 128bits width.
It's not the case on ie. FreeBSD.

llvm-svn: 185572

11 years agoAdd missing -std=c99 flag to appease GCC 4.6
Daniel Malea [Wed, 3 Jul 2013 19:34:25 +0000 (19:34 +0000)]
Add missing -std=c99 flag to appease GCC 4.6
- this issue was detected on recent GCC buildbot runs

llvm-svn: 185571

11 years agoRe-apply TestHelp.py fix
Daniel Malea [Wed, 3 Jul 2013 19:29:46 +0000 (19:29 +0000)]
Re-apply TestHelp.py fix

llvm-svn: 185570

11 years agoCommit patch for integer sequences. Suggested by Richard, reworked by Howard, and...
Marshall Clow [Wed, 3 Jul 2013 19:20:30 +0000 (19:20 +0000)]
Commit patch for integer sequences. Suggested by Richard, reworked by Howard, and annotated by me

llvm-svn: 185569

11 years agoAdd target hook CodeGen queries when generating builtin pow*.
Eli Bendersky [Wed, 3 Jul 2013 19:19:12 +0000 (19:19 +0000)]
Add target hook CodeGen queries when generating builtin pow*.

Without fmath-errno, Clang currently generates calls to @llvm.pow.* intrinsics
when it sees pow*(). This may not be suitable for all targets (for
example le32/PNaCl), so the attached patch adds a target hook that CodeGen
queries. The target can state its preference for having or not having the
intrinsic generated. Non-PNaCl behavior remains unchanged;
PNaCl-specific test added.

llvm-svn: 185568

11 years agoFix python 2.6 compatibility issue introduced by r184615
Daniel Malea [Wed, 3 Jul 2013 18:50:03 +0000 (18:50 +0000)]
Fix python 2.6 compatibility issue introduced by r184615
- argparse_compat library does not support reading environment variables
- should unblock Linux GCC buildbot from running tests again

llvm-svn: 185567

11 years agoUse an RWMutex instead of a Mutex in PassRegistry.
Chad Rosier [Wed, 3 Jul 2013 18:38:08 +0000 (18:38 +0000)]
Use an RWMutex instead of a Mutex in PassRegistry.
Patch by Alex Crichton <alex@crichton.co>.  Approved by Chris Lattner.

llvm-svn: 185566

11 years agoEnable -ffreestanding for this test, to avoid #include_next'ing the system's
Richard Smith [Wed, 3 Jul 2013 18:35:53 +0000 (18:35 +0000)]
Enable -ffreestanding for this test, to avoid #include_next'ing the system's
<stdint.h> (which might not exist or might not work).

llvm-svn: 185565

11 years ago[PowerPC] Support lmw/stmw in the asm parser
Ulrich Weigand [Wed, 3 Jul 2013 18:29:47 +0000 (18:29 +0000)]
[PowerPC] Support lmw/stmw in the asm parser

This adds support for the load/store multiple instructions,
currently used by the asm parser only.

llvm-svn: 185564

11 years agoProvide test case for commit r185544.
Bill Schmidt [Wed, 3 Jul 2013 18:21:12 +0000 (18:21 +0000)]
Provide test case for commit r185544.

Verify that assembling an empty file does not auto-include altivec.h.

llvm-svn: 185563

11 years agoAdd file suffix for assembler-with-cpp.
Eli Friedman [Wed, 3 Jul 2013 18:06:11 +0000 (18:06 +0000)]
Add file suffix for assembler-with-cpp.

Fixes crash when trying to recover from a crash on an assembler-with-cpp
file.  (Not sure how to write a testcase.)

llvm-svn: 185562

11 years ago[PowerPC] Use mtocrf when available
Ulrich Weigand [Wed, 3 Jul 2013 17:59:07 +0000 (17:59 +0000)]
[PowerPC] Use mtocrf when available

Just as with mfocrf, it is also preferable to use mtocrf instead of
mtcrf when only a single CR register is to be written.

Current code however always emits mtcrf.  This probably does not matter
when using an external assembler, since the GNU assembler will in fact
automatically replace mtcrf with mtocrf when possible.  It does create
inefficient code with the integrated assembler, however.

To fix this, this patch adds MTOCRF/MTOCRF8 instruction patterns and
uses those instead of MTCRF/MTCRF8 everything.  Just as done in the
MFOCRF patch committed as 185556, these patterns will be converted
back to MTCRF if MTOCRF is not available on the machine.

As a side effect, this allows to modify the MTCRF pattern to accept
the full range of mask operands for the benefit of the asm parser.

llvm-svn: 185561

11 years agoRevert commits that cause broken builds on GCC buildbots
Daniel Malea [Wed, 3 Jul 2013 17:58:31 +0000 (17:58 +0000)]
Revert commits that cause broken builds on GCC buildbots
- build fails due to PyCallable template definition inside an extern "C" scope

This commit reverts 185240, 184893 and 184608.

llvm-svn: 185560

11 years agoRemove empty files left behind from move to POSIX/
Ed Maste [Wed, 3 Jul 2013 17:41:40 +0000 (17:41 +0000)]
Remove empty files left behind from move to POSIX/

llvm-svn: 185559

11 years agoMatthew Dempsky: Attached patch replaces the type punning with memcpy(), which on
Howard Hinnant [Wed, 3 Jul 2013 17:39:28 +0000 (17:39 +0000)]
Matthew Dempsky: Attached patch replaces the type punning with memcpy(), which on
x86/x86-64 clang optimizes to direct word accesses anyway.  This fixes an unaligned word access in murmurhash/cityhash.

llvm-svn: 185558

11 years agoChris has agree to take part ownership of the driver.
Chad Rosier [Wed, 3 Jul 2013 17:25:50 +0000 (17:25 +0000)]
Chris has agree to take part ownership of the driver.

llvm-svn: 185557

11 years ago[PowerPC] Always use mfocrf if available
Ulrich Weigand [Wed, 3 Jul 2013 17:05:42 +0000 (17:05 +0000)]
[PowerPC] Always use mfocrf if available

When accessing just a single CR register, it is always preferable to
use mfocrf instead of mfcr, if the former is available on the CPU.

Current code makes that distinction in many, but not all places
where a single CR register value is retrieved.  One missing
location is PPCRegisterInfo::lowerCRSpilling.

To fix this and make this simpler in the future, this patch changes
the bulk of the back-end to always assume mfocrf is available and
simply generate it when needed.

On machines that actually do not support mfocrf, the instruction
is replaced by mfcr at the very end, in EmitInstruction.

This has the additional benefit that we no longer need the
MFCRpseud hack, since before EmitInstruction we always have
a MFOCRF instruction pattern, which already models data flow
as required.

The patch also adds the MFOCRF8 version of the instruction,
which was missing so far.

Except for the PPCRegisterInfo::lowerCRSpilling case, no change
in generated code intended.

llvm-svn: 185556

11 years ago[scan-build] Log compiler invocation to stderr, not stdout.
Jordan Rose [Wed, 3 Jul 2013 16:42:02 +0000 (16:42 +0000)]
[scan-build] Log compiler invocation to stderr, not stdout.

This is important for preprocessing steps, which may output to stdout.

Also, change ENV accesses using barewords to use string keys instead.

PR16414

llvm-svn: 185555

11 years agoPrefix failing commands with not to make clear they are expected to fail.
Rafael Espindola [Wed, 3 Jul 2013 16:41:29 +0000 (16:41 +0000)]
Prefix failing commands with not to make clear they are expected to fail.

llvm-svn: 185554

11 years agoSymbol prologue code checks if funciton lines up with symbol and uses function prolog...
Michael Sartain [Wed, 3 Jul 2013 16:35:41 +0000 (16:35 +0000)]
Symbol prologue code checks if funciton lines up with symbol and uses function prologue code with line info if so.

Differential Revision: http://llvm-reviews.chandlerc.com/D1082

llvm-svn: 185553

11 years agoRemove another old test.
Rafael Espindola [Wed, 3 Jul 2013 16:35:26 +0000 (16:35 +0000)]
Remove another old test.

It was only passing because 'grep andpd' was not finding any andpd, but
we don't fail if part of a pipe fails.

llvm-svn: 185552

11 years agoRemove test for the old EH system. It doesn't parse anymore.
Rafael Espindola [Wed, 3 Jul 2013 16:30:01 +0000 (16:30 +0000)]
Remove test for the old EH system. It doesn't parse anymore.

llvm-svn: 185551

11 years agoFix test: It was missing run lines and llvm-dis has no -disable-verify option.
Rafael Espindola [Wed, 3 Jul 2013 16:27:55 +0000 (16:27 +0000)]
Fix test: It was missing run lines and llvm-dis has no -disable-verify option.

llvm-svn: 185550

11 years agoUpdate error message; detach-keeps-stopped is also not on FreeBSD
Ed Maste [Wed, 3 Jul 2013 16:26:34 +0000 (16:26 +0000)]
Update error message; detach-keeps-stopped is also not on FreeBSD

llvm-svn: 185549

11 years agoWith CLANG_ENABLE_STATIC_ANALYZER=0, link clang properly and skip clang-check.
Jordan Rose [Wed, 3 Jul 2013 16:20:29 +0000 (16:20 +0000)]
With CLANG_ENABLE_STATIC_ANALYZER=0, link clang properly and skip clang-check.

Previously, the CMake build still tried to link clang against the static
analyzer libraries, even if CLANG_ENABLE_STATIC_ANALYZER was off.
Furthermore, clang-check depends on the analyzer, so it should be disabled
(in both CMake and configure builds).

In theory, clang-check could be made to conditionally include analyzer
support (like clang itself), but for now this at least gets a CMake ALL_BUILD
working.

Patch by Stephen Kelly, modified by me.

llvm-svn: 185548

11 years agoAdd support for gnu archives with a string table and no symtab.
Rafael Espindola [Wed, 3 Jul 2013 15:57:14 +0000 (15:57 +0000)]
Add support for gnu archives with a string table and no symtab.

While there, use early returns to reduce nesting.

llvm-svn: 185547

11 years agoMake llvm-nm return 1 on error.
Rafael Espindola [Wed, 3 Jul 2013 15:46:03 +0000 (15:46 +0000)]
Make llvm-nm return 1 on error.

This is a small compatibility improvement with gnu nm and makes llvm-nm more
useful as a testing tool.

llvm-svn: 185546

11 years agoFix PR16454: Don't #include altivec.h when preprocessing assembly.
Bill Schmidt [Wed, 3 Jul 2013 15:36:02 +0000 (15:36 +0000)]
Fix PR16454: Don't #include altivec.h when preprocessing assembly.

When the -maltivec flag is present, altivec.h is auto-included for the
compilation.  This is not appropriate when the job action is to
preprocess a file containing assembly code.  So don't do that.

I was unable to convert the test in the bug report into a regression
test.  The original symptom was exposed with:

  % touch x.S
  % ./bin/clang -target powerpc64-unknown-linux-gnu -maltivec -S -o - x.S

I tried this test (and numerous variants) on a PPC64 system:

----------------------------------------------------------------------------
// RUN: touch %t
// RUN: %clang -maltivec -S %t -o - | FileCheck %s

// Verify that assembling an empty file does not auto-include altivec.h.

// CHECK-NOT: static vector
----------------------------------------------------------------------------

However, this test passes for some reason even on a clang built
without the fix.  I'd be happy to add a test case but at this point
I'm not able to figure one out, and I don't want to hold up the patch
unnecessarily.  Please let me know if you have ideas.

Thanks,
Bill

llvm-svn: 185544

11 years agoTest case for PR7887 - failed with asm("")
Serge Pavlov [Wed, 3 Jul 2013 15:32:48 +0000 (15:32 +0000)]
Test case for PR7887 - failed with asm("")

llvm-svn: 185543

11 years ago[PowerPC] Remove dead code from PPCDAGToDAGISel::SelectSETCC
Ulrich Weigand [Wed, 3 Jul 2013 15:13:30 +0000 (15:13 +0000)]
[PowerPC] Remove dead code from PPCDAGToDAGISel::SelectSETCC

The subroutine getCRIdxForSetCC has a parameter "Other" and comment:

  If this returns with Other != -1, then the returned comparison
  is an or of two simpler comparisons.

However for at least the last five years this routine has never
returned a value of Other != -1; these cases are now handled
differently to begin with.

This patch removes the parameter and the code in SelectSETCC that
attempted to handle the Other != -1 case.

llvm-svn: 185541

11 years agoUse SmallVectorImpl::iterator/const_iterator instead of SmallVector to avoid specifyi...
Craig Topper [Wed, 3 Jul 2013 15:07:05 +0000 (15:07 +0000)]
Use SmallVectorImpl::iterator/const_iterator instead of SmallVector to avoid specifying the vector size.

llvm-svn: 185540

11 years agoFix regular expression used by 'make update' to only look for 'I' and '?' at the...
Craig Topper [Wed, 3 Jul 2013 14:48:37 +0000 (14:48 +0000)]
Fix regular expression used by 'make update' to only look for 'I' and '?' at the start of svn info results and to check for spaces after 'I' instead of just after '?'.

Previously it was able to match 'I' anywhere in the filenames of the svn info results instead of just files that where ignored or unknown to svn. This would cause 'make update' to infinitely recurse if a file was modified with I anywhere in its name since svn info would return a Path pointing to the llvm root for those files.

llvm-svn: 185539

11 years ago[msan] Unpoison stack allocations and undef values in blacklisted functions.
Evgeniy Stepanov [Wed, 3 Jul 2013 14:39:14 +0000 (14:39 +0000)]
[msan] Unpoison stack allocations and undef values in blacklisted functions.

This changes behavior of -msan-poison-stack=0 flag from not poisoning stack
allocations to actively unpoisoning them.

llvm-svn: 185538

11 years agoWe don't need to know the OpenSUSE version, so don't parse it.
Rafael Espindola [Wed, 3 Jul 2013 14:14:00 +0000 (14:14 +0000)]
We don't need to know the OpenSUSE version, so don't parse it.

Patch by Johannes Obermayr.

llvm-svn: 185537

11 years ago[sanitizer] Fix memory leak in sanitizer_common discovered by LeakSanitizer.
Sergey Matveev [Wed, 3 Jul 2013 13:45:13 +0000 (13:45 +0000)]
[sanitizer] Fix memory leak in sanitizer_common discovered by LeakSanitizer.

llvm-svn: 185536

11 years agocpp11-migrate: Add Replace-AutoPtr Transform
Edwin Vane [Wed, 3 Jul 2013 13:21:24 +0000 (13:21 +0000)]
cpp11-migrate: Add Replace-AutoPtr Transform

Add a new transform to replace uses of 'std::auto_ptr' by 'std::unique_ptr'.
Copy-ctor and assign-operator are wrapped with a call to 'std::move()'.

Note that until header modification is ready it is not that useful, that's why
it's marked as (EXPERIMENTAL) in the command line description and a "Known
Limitations" section is present in the transform documentation.

Author: Guillaume Papin <guillaume.papin@epitech.eu>
llvm-svn: 185535

11 years ago[PECOFF][Writer] Fix SizeOfImage header. It should include the first unmapped page.
Rui Ueyama [Wed, 3 Jul 2013 12:55:56 +0000 (12:55 +0000)]
[PECOFF][Writer] Fix SizeOfImage header. It should include the first unmapped page.

llvm-svn: 185534

11 years ago[PowerPC] Make specialized AltiVec patterns isCodeGenOnly
Ulrich Weigand [Wed, 3 Jul 2013 12:51:09 +0000 (12:51 +0000)]
[PowerPC] Make specialized AltiVec patterns isCodeGenOnly

A couple of AltiVec patterns are just specialized forms of the
generic instruction pattern, and should therefore be marked
isCodeGenOnly to avoid confusing the asm parser:
VCFSX_0, VCTUXS_0, VCFUX_0, VCTSXS_0, and V_SETALLONES.

Noticed by inspection of the generated PPCGenAsmMatcher.inc.

llvm-svn: 185533

11 years ago[PowerPC] Support mtspr/mfspr in the asm parser
Ulrich Weigand [Wed, 3 Jul 2013 12:32:41 +0000 (12:32 +0000)]
[PowerPC] Support mtspr/mfspr in the asm parser

This adds support for the generic forms of mtspr/mfspr
for the asm parser.  The compiler will continue to use
the specialized patters for mtlr etc. since those are
needed to correctly describe data flow.

llvm-svn: 185532

11 years agoPrevent error message when formatting an empty file.
Daniel Jasper [Wed, 3 Jul 2013 12:22:18 +0000 (12:22 +0000)]
Prevent error message when formatting an empty file.

This fixes llvm.org/PR16514.

llvm-svn: 185531

11 years agoDon't insert confusing line breaks in comparisons.
Daniel Jasper [Wed, 3 Jul 2013 10:34:47 +0000 (10:34 +0000)]
Don't insert confusing line breaks in comparisons.

In general, clang-format breaks after an operator if the LHS spans
multiple lines. Otherwise, this can lead to confusing effects and
effectively hide the operator precendence, e.g. in

if (aaaaaaaaaaaaaa ==
        bbbbbbbbbbbbbb && c) { ...

This patch removes this rule for comparisons, if the LHS is not a binary
expression itself as many users were wondering why clang-format inserts
an unnecessary linebreak.

Before:
if (aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa(
        aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa) >
    5) { ...

After:
if (aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa(
        aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa) > 5) { ...

In the long run, we might:
- Want to do this for other binary expressions as well.
- Do this only if the RHS is short or even only if it is a literal.

llvm-svn: 185530

11 years ago[SystemZ] Fold more spills
Richard Sandiford [Wed, 3 Jul 2013 10:10:02 +0000 (10:10 +0000)]
[SystemZ] Fold more spills

Add a mapping from register-based <INSN>R instructions to the corresponding
memory-based <INSN>.  Use it to cut down on the number of spill loads.

Some instructions extend their operands from smaller fields, so this
required a new TSFlags field to say how big the unextended operand is.

This optimisation doesn't trigger for C(G)R and CL(G)R because in practice
we always combine those instructions with a branch.  Adding a test for every
other case probably seems excessive, but it did catch a missed optimisation
for DSGF (fixed in r185435).

llvm-svn: 185529

11 years agoThis corrects the implementation of Thumb ADR instruction. There are three issues:
Mihai Popa [Wed, 3 Jul 2013 09:21:44 +0000 (09:21 +0000)]
This  corrects the implementation of Thumb ADR instruction. There are three issues:
1. it should accept only 4-byte aligned addresses
2. the maximum offset should be 1020
3. it should be encoded with the offset scaled by two bits

llvm-svn: 185528

11 years agoARM: relax the atomic release barrier to "dmb ishst" on Swift
Tim Northover [Wed, 3 Jul 2013 09:20:36 +0000 (09:20 +0000)]
ARM: relax the atomic release barrier to "dmb ishst" on Swift

Swift cores implement store barriers that are stronger than the ARM
specification but weaker than general barriers. They are, in fact, just about
enough to provide the ordering needed for atomic operations with release
semantics.

This patch makes use of that quirk.

llvm-svn: 185527

11 years ago[SystemZ] Rename mapping table fields
Richard Sandiford [Wed, 3 Jul 2013 09:19:58 +0000 (09:19 +0000)]
[SystemZ] Rename mapping table fields

Rename Function->DispKey and PairType->DispSize.  I'd originally used
"Function" because I thought it might be useful for other InstMappings.
However, it turns out that having two very similar instructions with the
same Function makes it pretty useless for anything other than the displacement
size key.  Other InstMappings will want the key to be defined for only one
instruction in the pair.

No behavioural change intended.

llvm-svn: 185526

11 years ago[SystemZ] Fix caller-allocated save slot FIXME
Richard Sandiford [Wed, 3 Jul 2013 09:11:00 +0000 (09:11 +0000)]
[SystemZ] Fix caller-allocated save slot FIXME

Get rid of some old code (and associated FIXME) for handling the
caller-allocated register save area.  No behavioural change intended.

llvm-svn: 185525

11 years agoGuard a test that fails on a Release build.
Rui Ueyama [Wed, 3 Jul 2013 09:09:13 +0000 (09:09 +0000)]
Guard a test that fails on a Release build.

llvm-svn: 185524

11 years agoRemove unused field.
Eric Christopher [Wed, 3 Jul 2013 08:26:07 +0000 (08:26 +0000)]
Remove unused field.

llvm-svn: 185523

11 years ago[analyzer] Improve handling of noreturn destructors
Pavel Labath [Wed, 3 Jul 2013 08:23:49 +0000 (08:23 +0000)]
[analyzer] Improve handling of noreturn destructors

Summary:
The analyzer incorrectly handled noreturn destructors which were hidden inside
function calls. This happened because NoReturnFunctionChecker only listened for
PostStmt events, which are not executed for destructor calls. I've changed it to
listen to PostCall events, which should catch both cases.

Reviewers: jordan_rose

CC: cfe-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D1056

llvm-svn: 185522

11 years ago[PECOFF][Writer] Separate AtomChunk from SectionChunk.
Rui Ueyama [Wed, 3 Jul 2013 08:15:00 +0000 (08:15 +0000)]
[PECOFF][Writer] Separate AtomChunk from SectionChunk.

The optional data directory header contains addresses to some atoms such
as the import address table in data section. Such fields can naturally
be set by relocation if we make the optional data driectory as an atom.

Currently we assume that atoms are always in a section, so we can't create
a file header with atoms. This patch separates section chunk from atom
chunk, to allow atom-based file header.

llvm-svn: 185521

11 years agoConstify a few functions.
Eric Christopher [Wed, 3 Jul 2013 08:13:55 +0000 (08:13 +0000)]
Constify a few functions.

llvm-svn: 185520

11 years ago[XCore] Whitespace fixes, no functionality change.
Richard Osborne [Wed, 3 Jul 2013 07:49:03 +0000 (07:49 +0000)]
[XCore] Whitespace fixes, no functionality change.

llvm-svn: 185519