[llvm-commits] Add line numbers to OProfile
Jeffrey Yasskin
jyasskin at gmail.com
Wed Jul 15 16:01:35 PDT 2009
On Wed, Jul 15, 2009 at 3:57 PM, Bruno Cardoso
Lopes<bruno.cardoso at gmail.com> wrote:
> Hi Jeffrey,
>
> On Wed, Jul 15, 2009 at 4:40 PM, Jeffrey Yasskin<jyasskin at google.com> wrote:
>> Here's the patch.
>>
>> On Wed, Jul 15, 2009 at 12:38 PM, <jyasskin at gmail.com> wrote:
>>> Reviewers: ,
>>>
>>> Message:
>>> Please take a look. I'll attach the patch shortly. Bruno, especially,
>>> let me know how I can make this fit better with your long-term plans for
>>> full Dwarf output.
>
> The modifications Regarding the MCE are ok for me, I'd do it in the same way :)
> What do you think about using the name processDebugLoc instead of
> beginDebugLoc?
I used beginDebugLoc because "process" didn't tell me anything about
what the DebugLoc meant. But I don't mind changing it back and will do
so unless someone else hollers.
Any other comments, from anyone?
>>> Description:
>>> Add line numbers to OProfile. To do this, I added a beginDebugLoc() call
>>> to the MachineCodeEmitter interface and made copying the start line of a
>>> function not conditional on whether we're emitting Dwarf debug
>>> information. I still need to propagate the beginDebugLoc() calls to the
>>> non-X86 targets, but I'm waiting to do that until LLVMdev confirms that
>>> this is the right approach.
>>>
>>> In the long run, it'll probably be better to gather this information
>>> through the DwarfWriter, but the DwarfWriter currently depends on the
>>> AsmPrinter and TargetAsmInfo, and fixing that would be out of the way
>>> for this patch.
>>> Sample output:
>>>
>>> $ sudo opcontrol --reset; sudo opcontrol --start-daemon; sudo opcontrol
>>> --start; Debug/bin/lli fib.bc; sudo opcontrol --stop
>>> [sudo] password for jyasskin:
>>> Signalling daemon... done
>>> Profiler running.
>>> fib(40) == 165580141
>>> Stopping profiling.
>>>
>>> $ opreport -g -d -l `pwd`/Debug/bin/lli|head -60
>>> Overflow stats not available
>>> CPU: Core 2, speed 1998 MHz (estimated)
>>> Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a
>>> unit mask of 0x00 (Unhalted core cycles) count 100000
>>> vma samples % linenr info image name
>>> symbol name
>>> 00007f62689160b0 25420 60.7364 fib.c:24 20676.jo
>>> fib_left
>>> 00007f62689160b0 1741 6.8489 fib.c:24
>>> 00007f62689160b1 91 0.3580 fib.c:24
>>> 00007f62689160b9 1934 7.6082 fib.c:24
>>> 00007f62689160c6 2012 7.9150 fib.c:27
>>> 00007f62689160c8 1072 4.2172 fib.c:27
>>> 00007f62689160cd 1280 5.0354 fib.c:27
>>> 00007f62689160cf 287 1.1290 fib.c:27
>>> 00007f62689160d3 1242 4.8859 fib.c:27
>>> 00007f62689160d6 1093 4.2998 fib.c:27
>>> 00007f62689160db 963 3.7884 fib.c:27
>>> 00007f62689160dd 240 0.9441 fib.c:27
>>> 00007f62689160e1 2781 10.9402 fib.c:28
>>> 00007f62689160e5 3704 14.5712 fib.c:28
>>> 00007f62689160ea 1 0.0039 fib.c:28
>>> 00007f62689160eb 596 2.3446 (no location information)
>>> 00007f62689160f3 6383 25.1101 (no location information)
>>> 00007f6268916100 15900 37.9901 fib.c:29 20676.jo
>>> fib_right
>>> 00007f6268916100 1669 10.4969 fib.c:29
>>> 00007f6268916101 68 0.4277 fib.c:29
>>> 00007f6268916109 2416 15.1950 fib.c:29
>>> 00007f6268916116 2262 14.2264 fib.c:32
>>> 00007f6268916118 605 3.8050 fib.c:32
>>> 00007f626891611d 700 4.4025 fib.c:32
>>> 00007f626891611f 384 2.4151 fib.c:32
>>> 00007f6268916123 434 2.7296 fib.c:32
>>> 00007f6268916126 647 4.0692 fib.c:32
>>> 00007f626891612b 921 5.7925 fib.c:32
>>> 00007f626891612d 73 0.4591 fib.c:32
>>> 00007f6268916131 1824 11.4717 fib.c:33
>>> 00007f6268916135 2873 18.0692 fib.c:33
>>> 00007f626891613a 2 0.0126 fib.c:33
>>> 00007f626891613b 1008 6.3396 (no location information)
>>> 00007f6268916143 14 0.0881 (no location information)
>>> 0000000000000000 16 0.0382 (no location information)
>>> libstdc++.so.6.0.9 /usr/lib/libstdc++.so.6.0.9
>>> 0000000000052fa0 1 6.2500 (no location information)
>>> 00000000000530f0 1 6.2500 (no location information)
>>> 0000000000062c4c 1 6.2500 (no location information)
>>> 000000000006bf02 1 6.2500 (no location information)
>>> 000000000006c4d6 1 6.2500 (no location information)
>>> 00000000000a13dd 1 6.2500 (no location information)
>>> 00000000000a19a0 1 6.2500 (no location information)
>>> 00000000000a244d 1 6.2500 (no location information)
>>> 00000000000a2b74 1 6.2500 (no location information)
>>> 00000000000a2b82 1 6.2500 (no location information)
>>> 00000000000a2b91 1 6.2500 (no location information)
>>> 00000000000a2bb9 2 12.5000 (no location information)
>>> 00000000000a7f45 1 6.2500 (no location information)
>>> 00000000000c2f10 1 6.2500 (no location information)
>>> 00000000000c4b30 1 6.2500 (no location information)
>>> 0000000000815e2a 13 0.0311 stl_vector.h:396 lli
>>> std::vector<llvm::MachineOperand,
>>> std::allocator<llvm::MachineOperand> >::size() const
>>> 0000000000815e2a 2 15.3846 stl_vector.h:396
>>> 0000000000815e36 1 7.6923 stl_vector.h:397
>>>
>>>
>>> OProfile 0.9.4 has a bug that causes it to ignore line numbers for code
>>> addresses above 4G. A fix is at
>>> http://thread.gmane.org/gmane.linux.oprofile/7634 but may or may not
>>> make it into OProfile 0.9.5.
>>>
>>> Please review this at http://codereview.appspot.com/91111
>>>
>>> Affected files:
>>> M include/llvm/CodeGen/MachineCodeEmitter.h
>>> M include/llvm/ExecutionEngine/JITEventListener.h
>>> M include/llvm/Support/DebugLoc.h
>>> M lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp
>>> M lib/ExecutionEngine/JIT/JITEmitter.cpp
>>> M lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp
>>> M lib/Target/X86/X86CodeEmitter.cpp
>>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>>
>
>
>
> --
> Bruno Cardoso Lopes
> http://www.brunocardoso.cc
>
--
Namasté,
Jeffrey Yasskin
http://jeffrey.yasskin.info/
More information about the llvm-commits
mailing list