Reset FirstID (first diagnostic's file ID) to allow for invocations of the VerifyDiagnosticsClient on several input files.

llvm-svn: 138434

variable sized alloca.

llvm-svn: 138433

[analyzer] Fix a Heisenbug concerning object lifetimes with a hack. Hopefully a better fix coming soon. See comment for more details.

llvm-svn: 138432

[analyzer] Reapply 138390 "Clean up unused bits of CFRefCount", reverted in 138405. Ended up unrelated to any problems.

llvm-svn: 138431

[analyzer] Fix potential crasher with RAII. No functionality change. (No test because this particular circumstance requires an impossible series of events, but this is future-proofing.)

llvm-svn: 138430

llvm-svn: 138428

Break 256-bit vector int add/sub/mul into two 128-bit operations to avoid costly scalarization. Fixes PR10711.

llvm-svn: 138427

[analyzer] Reapply 138382 and 138388 (reverted in 138419 and 138420). The issue seems to have been with the uninitialized variable fixed in 138424; a fix for another Heisencrasher coming soon.

llvm-svn: 138426

llvm-svn: 138425

llvm-svn: 138424

llvm-svn: 138423

llvm-svn: 138422

llvm-svn: 138421

Revert "[analyzer] Cleanup: Move temporary declarations of CFRefCount variables closer to their uses. No functionality change."

This reverts commit d02b4af7bd6ca4a743c7074d64e205d718aa221d.

llvm-svn: 138420

Revert "[analyzer] Make CFRefBug and CFRefReportVisitor not dependent on CFRefCount. Unfortunately, CFRefReport still is. No functionality change."

This reverts commit e3fb7e428b7ba0d5d902dfe3f165d70e09b03a15.

llvm-svn: 138419

After talking with John making this the case for all of these is
the right way to go.

Fixes rdar://9804564 and PR10414

llvm-svn: 138418

[analyzer] MacOSKeychainAPIChecker: Add reasoning about functions which MIGHT deallocate the memory region allocated with SecKeychain APIs. Specifically, when the buffer is passed to CFStringCreateWithBytesNoCopy along with a custom deallocator, which might potentially correctly release the memory.

llvm-svn: 138417

A value of -1 at a call site tells the personality function that this call isn't
handled by the current function. Since the ResumeInsts are converted to calls to
_Unwind_SjLj_Resume, add a (volatile) store of -1 to its 'call site'.

llvm-svn: 138416

[analyzer] MacOSKeychainAPIChecker: Retrieve the memory region which we are tracking even when it's no longer a SymbolicRegion, for example, when it is cast to char*.

llvm-svn: 138415

llvm-svn: 138414

llvm-svn: 138413

// rdar://9362887.

llvm-svn: 138412

  <i32 3, i32 2, i32 8, i32 11, i32 3, i32 6, i32 12, i32 15>
generate:
  <i32 3, i32 2, i32 8, i32 11, i32 7, i32 6, i32 12, i32 15>

llvm-svn: 138411

handled SCC's of dead code, or simply having false negatives by overly suppressing warnings.

WIP.

llvm-svn: 138410

CFG: record set of C++ 'try' dispatch blocks, which could be of interest to various analyses (e.g., reachability).

llvm-svn: 138409

llvm-svn: 138408

llvm-svn: 138407

This is not necessarily the first or dominating use of the EH values. The IR
breaks if it's not. So replace the specific value in the instruction with the
new value.

llvm-svn: 138406

This reverts commit 6886a92640f5bffc972f67c0a1f302d6c6e7c322.

llvm-svn: 138405

llvm-svn: 138404

llvm-svn: 138403

Polly adds, after it is loaded into opt or clang, its passes to the default set
of -O3 passes. This means optimizing a program with clang and Polly becomes as
simple as executing.

clang -Xclang -load -Xclang lib/LLVMPolly.so -O3 program.c

The same should work for dragonegg powered gfortran, g++, ... or any other tool
that uses the PassManagerBuilder.

Warning: Even though using Polly became with this commit extremly easy, Polly
         is still Pre-Alpha Quality. This means in most cases it will rather
         destroy the world than doing anything positive. ;-)
llvm-svn: 138402

llvm-svn: 138401

llvm-svn: 138400

./dotest.py -v -f DataFormatterTestCase.test_with_dsym_and_run_command

will not end up running 14 tests.

llvm-svn: 138399

Refactor and fix checking for initialization of flexible array members.  The old version had the checks scattered across the code, missed some checks, and had a couple nasty bugs in existing checks.

Fixes PR10648 and another similar accepts-invalid bug.

llvm-svn: 138398

The invoke could be at the end of the entry block. If it's the only one, then we
won't process all of the landingpad instructions correctly. This code is
currently ugly, but should be made much nicer once the new EH switch is thrown.

llvm-svn: 138397

Skip these two cases temporarily while investigating the crash on OSX Lion with clang as the compiler.

llvm-svn: 138393

permutations. Also tidy up some patterns and make them close to their
instruction definition!

llvm-svn: 138392

llvm-svn: 138391