summaryrefslogtreecommitdiffstats
path: root/lldb/source/API/SBThread.cpp
Commit message (Collapse)AuthorAgeFilesLines
...
* SBFrame is now threadsafe using some extra tricks. One issue is that stackGreg Clayton2012-01-301-20/+27
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | frames might go away (the object itself, not the actual logical frame) when we are single stepping due to the way we currently sometimes end up flushing frames when stepping in/out/over. They later will come back to life represented by another object yet they have the same StackID. Now when you get a lldb::SBFrame object, it will track the frame it is initialized with until the thread goes away or the StackID no longer exists in the stack for the thread it was created on. It uses a weak_ptr to both the frame and thread and also stores the StackID. These three items allow us to determine when the stack frame object has gone away (the weak_ptr will be NULL) and allows us to find the correct frame again. In our test suite we had such cases where we were just getting lucky when something like this happened: 1 - stop at breakpoint 2 - get first frame in thread where we stopped 3 - run an expression that causes the program to JIT and run code 4 - run more expressions on the frame from step 2 which was very very luckily still around inside a shared pointer, yet, not part of the current thread (a new stack frame object had appeared with the same stack ID and depth). We now avoid all such issues and properly keep up to date, or we start returning errors when the frame doesn't exist and always responds with invalid answers. Also fixed the UserSettingsController (not going to rewrite this just yet) so that it doesn't crash on shutdown. Using weak_ptr's came in real handy to track when the master controller has already gone away and this allowed me to pull out the previous NotifyOwnerIsShuttingDown() patch as it is no longer needed. llvm-svn: 149231
* Removed the "lldb-forward-rtti.h" header file as it was designed to containGreg Clayton2012-01-301-190/+192
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | all RTTI types, and since we don't use RTTI anymore since clang and llvm don't we don't really need this header file. All shared pointer definitions have been moved into "lldb-forward.h". Defined std::tr1::weak_ptr definitions for all of the types that inherit from enable_shared_from_this() in "lldb-forward.h" in preparation for thread hardening our public API. The first in the thread hardening check-ins. First we start with SBThread. We have issues in our lldb::SB API right now where if you have one object that is being used by two threads we have a race condition. Consider the following code: 1 int 2 SBThread::SomeFunction() 3 { 4 int result = -1; 5 if (m_opaque_sp) 6 { 7 result = m_opaque_sp->DoSomething(); 8 } 9 return result; 10 } And now this happens: Thread 1 enters any SBThread function and checks its m_opaque_sp and is about to execute the code on line 7 but hasn't yet Thread 2 gets to run and class sb_thread.Clear() which calls m_opaque_sp.clear() and clears the contents of the shared pointer member Thread 1 now crashes when it resumes. The solution is to use std::tr1::weak_ptr. Now the SBThread class contains a lldb::ThreadWP (weak pointer to our lldb_private::Thread class) and this function would look like: 1 int 2 SBThread::SomeFunction() 3 { 4 int result = -1; 5 ThreadSP thread_sp(m_opaque_wp.lock()); 6 if (thread_sp) 7 { 8 result = m_opaque_sp->DoSomething(); 9 } 10 return result; 11 } Now we have a solid thread safe API where we get a local copy of our thread shared pointer from our weak_ptr and then we are guaranteed it can't go away during our function. So lldb::SBThread has been thread hardened, more checkins to follow shortly. llvm-svn: 149218
* Fixed code rot pointed out by Jim.Johnny Chen2011-12-171-4/+2
| | | | | | SBThread::GetStopReasonDataCount/GetStopReasonDataAtIndex() need to handle eStopReasonWatchpoint. llvm-svn: 146812
* Add the ability to capture the return value in a thread's stop info, and ↵Jim Ingham2011-12-171-2/+26
| | | | | | | | | | | | | print it as part of the thread format output. Currently this is only done for the ThreadPlanStepOut. Add a convenience API ABI::GetReturnValueObject. Change the ValueObject::EvaluationPoint to BE an ExecutionContextScope, rather than trying to hand out one of its subsidiary object's pointers. That way this will always be good. llvm-svn: 146806
* Process IDs (lldb::pid_t) and thread IDs (lldb::tid_t) are now 64 bit. This Greg Clayton2011-12-011-1/+1
| | | | | | | | will allow us to represent a process/thread ID using a pointer for the OS plug-ins where they might want to represent the process or thread ID using the address of the process or thread structure. llvm-svn: 145644
* <rdar://problem/10126482>Greg Clayton2011-11-131-3/+4
| | | | | | | | | | | | | | | Fixed an issues with the SBType and SBTypeMember classes: - Fixed SBType to be able to dump itself from python - Fixed SBType::GetNumberOfFields() to return the correct value for objective C interfaces - Fixed SBTypeMember to be able to dump itself from python - Fixed the SBTypeMember ability to get a field offset in bytes (the value being returned was wrong) - Added the SBTypeMember ability to get a field offset in bits Cleaned up a lot of the Stream usage in the SB API files. llvm-svn: 144493
* Cleaned up many error codes. For any who is filling in error strings intoGreg Clayton2011-10-261-1/+1
| | | | | | | | | | | | | lldb_private::Error objects the rules are: - short strings that don't start with a capitol letter unless the name is a class or anything else that is always capitolized - no trailing newline character - should be one line if possible Implemented a first pass at adding "--gdb-format" support to anything that accepts format with optional size/count. llvm-svn: 142999
* Moved lldb::user_id_t values to be 64 bit. This was going to be needed forGreg Clayton2011-10-191-1/+1
| | | | | | | | | | | process IDs, and thread IDs, but was mainly needed for for the UserID's for Types so that DWARF with debug map can work flawlessly. With DWARF in .o files the type ID was the DIE offset in the DWARF for the .o file which is not unique across all .o files, so now the SymbolFileDWARFDebugMap class will make the .o file index part (the high 32 bits) of the unique type identifier so it can uniquely identify the types. llvm-svn: 142534
* Update declarations for all functions/methods that accept printf-styleJason Molenda2011-09-201-1/+1
| | | | | | | | stdarg formats to use __attribute__ format so the compiler can flag incorrect uses. Fix all incorrect uses. Most of these are innocuous, a few were resulting in crashes. llvm-svn: 140185
* Don't put modules for .o files into the global shared module list. WeGreg Clayton2011-09-181-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | used to do this because we needed to find the shared pointer for a .o file when the .o file's module was needed in a SymbolContext since the module in a symbol context was a shared pointer. Now that we are using intrusive pointers we don't have this limitation anymore since any instrusive shared pointer can be made from a pointer to an object all on its own. Also switched over to having the Module and SymbolVendor use shared pointers to their object files as had a leak on MacOSX when the SymbolVendor's object file wasn't the same as the Module's (debug info in a stand along file (dSYM file)). Now everything will correctly clean itself up when the module goes away after an executable gets rebuilt. Now we correctly get rid of .o files that are used with the DWARF with debug map executables on subsequent runs since the only shared pointer to the object files in from the DWARF symbol file debug map parser, and when the module gets replaced, it destroys to old one along with all .o files. Also added a small optimization when using BSD archives where we will remove old BSD containers from the shared list when they are outdated. llvm-svn: 140002
* Move the SourceManager from the Debugger to the Target. That way it can ↵Jim Ingham2011-09-081-1/+1
| | | | | | | | | | store the per-Target default Source File & Line. Set the default Source File & line to main (if it can be found.) at startup. Selecting the current thread & or frame resets the current source file & line, and "source list" as well as the breakpoint command "break set -l <NUM>" will use the current source file. llvm-svn: 139323
* StepUntil should check whether the target of the step until is in the currentJim Ingham2011-05-081-6/+28
| | | | | | function and if not return an error. llvm-svn: 131061
* Added support for stepping out of a frame. If you have 10 stack frames, and you Greg Clayton2011-01-211-13/+190
| | | | | | | | | | select frame #3, you can then do a step out and be able to go directly to the frame above frame #3! Added StepOverUntil and StepOutOfFrame to the SBThread API to allow more powerful stepping. llvm-svn: 123970
* Fixed missing return value (patch from Stephen Wilson).Greg Clayton2011-01-181-1/+1
| | | | llvm-svn: 123779
* Added the following functions to SBThread to allow threads to be suspended ↵Greg Clayton2011-01-121-0/+30
| | | | | | | | | | when a process is resumed: bool SBThread::Suspend(); bool SBThread::Resume(); bool SBThread::IsSuspended(); llvm-svn: 123300
* The LLDB API (lldb::SB*) is now thread safe!Greg Clayton2010-12-201-4/+28
| | | | llvm-svn: 122262
* Added access to set the current stack frame within a thread so any commandGreg Clayton2010-12-171-0/+47
| | | | | | | | | | | | line commands can use the current thread/frame. Fixed an issue with expressions that get sandboxed in an objective C method where unichar wasn't being passed down. Added a "static size_t Scalar::GetMaxByteSize();" function in case we need to know the max supported by size of something within a Scalar object. llvm-svn: 122027
* Added the ability to get more information on the SBThread's stop reasonGreg Clayton2010-11-181-0/+104
| | | | | | | | | | | | | | by being able to get the data count and data. Each thread stop reason has one or more data words that can help describe the stop. To do this I added: size_t SBThread::GetStopReasonDataCount(); uint64_t SBThread::GetStopReasonDataAtIndex(uint32_t idx); llvm-svn: 119720
* Modified all logging calls to hand out shared pointers to make sure weGreg Clayton2010-11-061-13/+13
| | | | | | | | | | | don't crash if we disable logging when some code already has a copy of the logger. Prior to this fix, logs were handed out as pointers and if they were held onto while a log got disabled, then it could cause a crash. Now all logs are handed out as shared pointers so this problem shouldn't happen anymore. We are also using our new shared pointers that put the shared pointer count and the object into the same allocation for a tad better performance. llvm-svn: 118319
* Cleaned up the API logging a lot more to reduce redundant information and Greg Clayton2010-10-311-11/+15
| | | | | | | | | keep the file size a bit smaller. Exposed SBValue::GetExpressionPath() so SBValue users can get an expression path for their values. llvm-svn: 117851
* Fixed the copy constructor for SBThread.Greg Clayton2010-10-301-2/+3
| | | | llvm-svn: 117825
* Improved API logging.Greg Clayton2010-10-301-75/+22
| | | | llvm-svn: 117772
* Add the ability to disable individual log categories, ratherCaroline Tice2010-10-291-1/+1
| | | | | | | | | than just the entire log channel. Add checks, where appropriate, to make sure a log channel/category has not been disabled before attempting to write to it. llvm-svn: 117715
* Modified the lldb_private::TypeList to use a std::multimap for quicker lookupGreg Clayton2010-10-291-27/+11
| | | | | | | | | | by type ID (the most common type of type lookup). Changed the API logging a bit to always show the objects in the OBJECT(POINTER) format so it will be easy to locate all instances of an object or references to it when looking at logs. llvm-svn: 117641
* Clean up the API logging code:Caroline Tice2010-10-261-46/+53
| | | | | | | | | | | | | | - Try to reduce logging to one line per function call instead of tw - Put all arguments & their values into log for calls - Add 'this' parameter information to function call logging, making it show the appropriate internal pointer (this.obj, this.sp, this.ap...) - Clean up some return values - Remove logging of constructors that construct empty objects - Change '==>' to '=>' for showing result values... - Fix various minor bugs - Add some protected 'get' functions to help getting the internal pointers for the 'this' arguments... llvm-svn: 117417
* First pass at adding logging capabilities for the API functions. At the momentCaroline Tice2010-10-261-7/+160
| | | | | | | | | | | | | | | | | | it logs the function calls, their arguments and the return values. This is not complete or polished, but I am committing it now, at the request of someone who really wants to use it, even though it's not really done. It currently does not attempt to log all the functions, just the most important ones. I will be making further adjustments to the API logging code over the next few days/weeks. (Suggestions for improvements are welcome). Update the Python build scripts to re-build the swig C++ file whenever the python-extensions.swig file is modified. Correct the help for 'log enable' command (give it the correct number & type of arguments). llvm-svn: 117349
* Don't cache the public stop reason, since it can change as plan completion ↵Jim Ingham2010-10-201-8/+8
| | | | | | gets processed. That means GetStopReason needs to return a shared pointer, not a pointer to the thread's cached version. Also allow the thread plans to get and set the thread private stop reason - that is usually more appropriate for the logic the thread plans need to do. llvm-svn: 116892
* Cleaned up the SWIG stuff so all includes happen as they should, no pullingGreg Clayton2010-10-071-1/+4
| | | | | | | | | | | | | | | | | | tricks to get types to resolve. I did this by correctly including the correct files: stdint.h and all lldb-*.h files first before including the API files. This allowed me to remove all of the hacks that were in the lldb.swig file and it also allows all of the #defines in lldb-defines.h and enumerations in lldb-enumerations.h to appear in the lldb.py module. This will make the python script code a lot more readable. Cleaned up the "process launch" command to not execute a "process continue" command, it now just does what it should have with the internal API calls instead of executing another command line command. Made the lldb_private::Process set the state to launching and attaching if WillLaunch/WillAttach return no error respectively. llvm-svn: 115902
* Added the first of hopefully many python example scripts that show how toGreg Clayton2010-10-061-20/+54
| | | | | | | | | | use the python API that is exposed through SWIG to do some cool stuff. Also fixed synchronous debugging so that all process control APIs exposed through the python API will now wait for the process to stop if you set the async mode to false (see disasm.py). llvm-svn: 115738
* Added the ability to get the disassembly instructions from the function andGreg Clayton2010-10-061-5/+1
| | | | | | symbol. llvm-svn: 115734
* There are now to new "settings set" variables that live in each debuggerGreg Clayton2010-10-041-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | instance: settings set frame-format <string> settings set thread-format <string> This allows users to control the information that is seen when dumping threads and frames. The default values are set such that they do what they used to do prior to changing over the the user defined formats. This allows users with terminals that can display color to make different items different colors using the escape control codes. A few alias examples that will colorize your thread and frame prompts are: settings set frame-format 'frame #${frame.index}: \033[0;33m${frame.pc}\033[0m{ \033[1;4;36m${module.file.basename}\033[0;36m ${function.name}{${function.pc-offset}}\033[0m}{ \033[0;35mat \033[1;35m${line.file.basename}:${line.number}}\033[0m\n' settings set thread-format 'thread #${thread.index}: \033[1;33mtid\033[0;33m = ${thread.id}\033[0m{, \033[0;33m${frame.pc}\033[0m}{ \033[1;4;36m${module.file.basename}\033[0;36m ${function.name}{${function.pc-offset}}\033[0m}{, \033[1;35mstop reason\033[0;35m = ${thread.stop-reason}\033[0m}{, \033[1;36mname = \033[0;36m${thread.name}\033[0m}{, \033[1;32mqueue = \033[0;32m${thread.queue}}\033[0m\n' A quick web search for "colorize terminal output" should allow you to see what you can do to make your output look like you want it. The "settings set" commands above can of course be added to your ~/.lldbinit file for permanent use. Changed the pure virtual void ExecutionContextScope::Calculate (ExecutionContext&); To: void ExecutionContextScope::CalculateExecutionContext (ExecutionContext&); I did this because this is a class that anything in the execution context heirarchy inherits from and "target->Calculate (exe_ctx)" didn't always tell you what it was really trying to do unless you look at the parameter. llvm-svn: 115485
* Remove all the __repr__ methods from the API/*.h files, and put themCaroline Tice2010-09-221-8/+0
| | | | | | | | | | into python-extensions.swig, which gets included into lldb.swig, and adds them back into the classes when swig generates it's C++ file. This keeps the Python stuff out of the general API classes. Also fixed a small bug in the copy constructor for SBSymbolContext. llvm-svn: 114602
* Add GetDescription() and __repr__ () methods to most API classes, to allowCaroline Tice2010-09-201-0/+24
| | | | | | | "print" from inside Python to print out the objects in a more useful manner. llvm-svn: 114321
* Remove unnecessary/inappropriate output-printing functions fromCaroline Tice2010-09-151-100/+0
| | | | | | the API. llvm-svn: 113993
* Added a new bool parameter to many of the DumpStopContext() methods that Greg Clayton2010-09-021-1/+1
| | | | | | | | | | | | | | | | | might dump file paths that allows the dumping of full paths or just the basenames. Switched the stack frame dumping code to use just the basenames for the files instead of the full path. Modified the StackID class to no rely on needing the start PC for the current function/symbol since we can use the SymbolContextScope to uniquely identify that, unless there is no symbol context scope. In that case we can rely upon the current PC value. This saves the StackID from having to calculate the start PC when the StackFrame::GetStackID() accessor is called. Also improved the StackID less than operator to correctly handle inlined stack frames in the same stack. llvm-svn: 112867
* Change "Current" as in GetCurrentThread, GetCurrentStackFrame, etc, to ↵Jim Ingham2010-08-261-14/+14
| | | | | | "Selected" i.e. GetSelectedThread. Selected makes more sense, since these are set by some user action (a selection). I didn't change "CurrentProcess" since this is always controlled by the target, and a given target can only have one process, so it really can't be selected. llvm-svn: 112221
* Added support for inlined stack frames being represented as real stack framesGreg Clayton2010-08-241-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | which is now on by default. Frames are gotten from the unwinder as concrete frames, then if inline frames are to be shown, extra information to track and reconstruct these frames is cached with each Thread and exanded as needed. I added an inline height as part of the lldb_private::StackID class, the class that helps us uniquely identify stack frames. This allows for two frames to shared the same call frame address, yet differ only in inline height. Fixed setting breakpoint by address to not require addresses to resolve. A quick example: % cat main.cpp % ./build/Debug/lldb test/stl/a.out Current executable set to 'test/stl/a.out' (x86_64). (lldb) breakpoint set --address 0x0000000100000d31 Breakpoint created: 1: address = 0x0000000100000d31, locations = 1 (lldb) r Launching 'a.out' (x86_64) (lldb) Process 38031 Stopped * thread #1: tid = 0x2e03, pc = 0x0000000100000d31, where = a.out`main [inlined] std::string::_M_data() const at /usr/include/c++/4.2.1/bits/basic_string.h:280, stop reason = breakpoint 1.1, queue = com.apple.main-thread 277 278 _CharT* 279 _M_data() const 280 -> { return _M_dataplus._M_p; } 281 282 _CharT* 283 _M_data(_CharT* __p) (lldb) bt thread #1: tid = 0x2e03, stop reason = breakpoint 1.1, queue = com.apple.main-thread frame #0: pc = 0x0000000100000d31, where = a.out`main [inlined] std::string::_M_data() const at /usr/include/c++/4.2.1/bits/basic_string.h:280 frame #1: pc = 0x0000000100000d31, where = a.out`main [inlined] std::string::_M_rep() const at /usr/include/c++/4.2.1/bits/basic_string.h:288 frame #2: pc = 0x0000000100000d31, where = a.out`main [inlined] std::string::size() const at /usr/include/c++/4.2.1/bits/basic_string.h:606 frame #3: pc = 0x0000000100000d31, where = a.out`main [inlined] operator<< <char, std::char_traits<char>, std::allocator<char> > at /usr/include/c++/4.2.1/bits/basic_string.h:2414 frame #4: pc = 0x0000000100000d31, where = a.out`main + 33 at /Volumes/work/gclayton/Documents/src/lldb/test/stl/main.cpp:14 frame #5: pc = 0x0000000100000d08, where = a.out`start + 52 Each inline frame contains only the variables that they contain and each inlined stack frame is treated as a single entity. llvm-svn: 111877
* Abtracted the old "lldb_private::Thread::StopInfo" into an abtract class.Greg Clayton2010-08-041-8/+9
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This will allow debugger plug-ins to make any instance of "lldb_private::StopInfo" that can completely describe any stop reason. It also provides a framework for doing intelligent things with the stop info at important times in the lifetime of the inferior. Examples include the signal stop info in StopInfoUnixSignal. It will check with the process to see that the current action is for the signal. These actions include wether to stop for the signal, wether the notify that the signal was hit, and wether to pass the signal along to the inferior process. The StopInfoUnixSignal class overrides the "ShouldStop()" method of StopInfo and this allows the stop info to determine if it should stop at the signal or continue the process. StopInfo subclasses must override the following functions: virtual lldb::StopReason GetStopReason () const = 0; virtual const char * GetDescription () = 0; StopInfo subclasses can override the following functions: // If the subclass returns "false", the inferior will resume. The default // version of this function returns "true" which means the default stop // info will stop the process. The breakpoint subclass will check if // the breakpoint wants us to stop by calling any installed callback on // the breakpoint, and also checking if the breakpoint is for the current // thread. Signals will check if they should stop based off of the // UnixSignal settings in the process. virtual bool ShouldStop (Event *event_ptr); // Sublasses can state if they want to notify the debugger when "ShouldStop" // returns false. This would be handy for breakpoints where you want to // log information and continue and is also used by the signal stop info // to notify that a signal was received (after it checks with the process // signal settings). virtual bool ShouldNotify (Event *event_ptr) { return false; } // Allow subclasses to do something intelligent right before we resume. // The signal class will figure out if the signal should be propagated // to the inferior process and pass that along to the debugger plug-ins. virtual void WillResume (lldb::StateType resume_state) { // By default, don't do anything } The support the Mach exceptions was moved into the lldb/source/Plugins/Process/Utility folder and now doesn't polute the lldb_private::Thread class with platform specific code. llvm-svn: 110184
* Added "void Clear();" methods to SBDebugger, SBTarget and SBThread so they ↵Greg Clayton2010-07-301-0/+7
| | | | | | can release their shared pointers. llvm-svn: 109882
* I enabled some extra warnings for hidden local variables and for hiddenGreg Clayton2010-07-141-1/+0
| | | | | | virtual functions and caught some things and did some general code cleanup. llvm-svn: 108299
* Misc warning fixes.Eli Friedman2010-07-091-1/+1
| | | | llvm-svn: 108029
* Merged Eli Friedman's linux build changes where he added Makefile files thatGreg Clayton2010-07-091-2/+5
| | | | | | | enabled LLVM make style building and made this compile LLDB on Mac OS X. We can now iterate on this to make the build work on both linux and macosx. llvm-svn: 108009
* Very large changes that were needed in order to allow multiple connectionsGreg Clayton2010-06-231-112/+75
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | to the debugger from GUI windows. Previously there was one global debugger instance that could be accessed that had its own command interpreter and current state (current target/process/thread/frame). When a GUI debugger was attached, if it opened more than one window that each had a console window, there were issues where the last one to setup the global debugger object won and got control of the debugger. To avoid this we now create instances of the lldb_private::Debugger that each has its own state: - target list for targets the debugger instance owns - current process/thread/frame - its own command interpreter - its own input, output and error file handles to avoid conflicts - its own input reader stack So now clients should call: SBDebugger::Initialize(); // (static function) SBDebugger debugger (SBDebugger::Create()); // Use which ever file handles you wish debugger.SetErrorFileHandle (stderr, false); debugger.SetOutputFileHandle (stdout, false); debugger.SetInputFileHandle (stdin, true); // main loop SBDebugger::Terminate(); // (static function) SBDebugger::Initialize() and SBDebugger::Terminate() are ref counted to ensure nothing gets destroyed too early when multiple clients might be attached. Cleaned up the command interpreter and the CommandObject and all subclasses to take more appropriate arguments. llvm-svn: 106615
* Two changes in this checkin. Added a ThreadPlanKind so that I can do some ↵Jim Ingham2010-06-191-1/+0
| | | | | | | | | | | reasoning based on the kind of thread plan without having to use RTTI. Removed the ThreadPlanContinue and replaced with a ShouldAutoContinue query that serves the same purpose. Having to push another plan to assert that if there's no other indication the target should continue when this plan is popped was flakey and error prone. This method is more stable, and fixed problems we were having with thread specific breakpoints. llvm-svn: 106378
* I have eliminated RTTI from LLDB!Greg Clayton2010-06-121-18/+9
| | | | | | | Also added a shell script build phase that fixes the headers in LLDB.framework. llvm-svn: 105899
* And some more include line fixes.Eli Friedman2010-06-091-6/+6
| | | | llvm-svn: 105704
* Initial checkin of lldb code from internal Apple repo.Chris Lattner2010-06-081-0/+551
llvm-svn: 105619
OpenPOWER on IntegriCloud