[PATCH] [PATCH/RFC] Use a BumpAllocator for DIEs

Frédéric Riss friss at apple.com
Thu Feb 5 09:25:44 PST 2015


> On Jan 27, 2015, at 6:43 PM, Frédéric Riss <friss at apple.com> wrote:
>> On Jan 27, 2015, at 4:11 PM, David Blaikie <dblaikie at gmail.com <mailto:dblaikie at gmail.com>> wrote:
>> On Tue, Jan 27, 2015 at 2:08 PM, Frederic Riss <friss at apple.com <mailto:friss at apple.com>> wrote:
>> This update removes the limitation to a fixed maximum number of
>> attributes per DIE. The patch is very similar to the first one that
>> I sent for discussion. It only adds a pointer to a vector of DIEs
>> as arguement to DIE::addValue(). If a DIE overflows its inline storage
>> for attributes, it is added to the vector. The vector is then iterated
>> just before the BumpPtrAllocator is destroyed to call the destructors
>> of all these DIEs.
>> This approach gives the most flexibilty as it doesn't impose the
>> memory management upon the user (for example DIEHashTest.cpp continues
>> to use stack allocated DIEs and it works fine).
>> I've been playing with this for the last few days, and I have quite
>> a few different implementations lying around (for example using
>> std::vectors with a custom stateful allocator). They are all more
>> complicated and don't perform as well as this patch.
>> There's one point that I'd like to mention: the patch unifies the
>> existing bookkeeping of the DIEBlock and DIELoc objects and treats
>> them the same as standard DIEs. This means that we will call ~DIE
>> on DIEBlock and DIELoc objects which will result in a 'partial'
>> destruction (DIEBlock inherits both from DIEValue and from DIE).
>> I think this is fine. It should do exactly what we want and just
>> call the destructors of the SmallVectors in the DIE part of the
>> object, but I wanted to mention it in case someone thinks its not
>> legal to do that.
>> Pretty sure that's not valid C++ (not sure quite which cracks the bump allocator objects fall in general in terms of not running dtors, reusing memory, etc - and whether this would be worse than that or not, though).
> What exactly do you think the issue would be from a language standpoint? Calling a destructor on a ‘partial’ object? I would expect that it is ok to do so. AFAIK, explicit destructor calls follow the same rules as other function calls. I have a pointer to a DIE object, thus I can call the destructor on it. The object stops to ‘exist’ at that point, and thus its enclosing object is in an undefined state, but this doesn’t really matter as it won’t be touched anymore.
> I’m very bad at language lawyering though, thus I’d really appreciate a authoritative answer :-)
> Fred 
>> http://reviews.llvm.org/D7072 <http://reviews.llvm.org/D7072>
>> Files:
>>   include/llvm/CodeGen/DIE.h
>>   lib/CodeGen/AsmPrinter/DIE.cpp
>>   lib/CodeGen/AsmPrinter/DIEHash.cpp
>>   lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp
>>   lib/CodeGen/AsmPrinter/DwarfCompileUnit.h
>>   lib/CodeGen/AsmPrinter/DwarfDebug.cpp
>>   lib/CodeGen/AsmPrinter/DwarfFile.cpp
>>   lib/CodeGen/AsmPrinter/DwarfUnit.cpp
>>   lib/CodeGen/AsmPrinter/DwarfUnit.h
>>   unittests/CodeGen/DIEHashTest.cpp
>>   http://reviews.llvm.org/settings/panel/emailpreferences/ <http://reviews.llvm.org/settings/panel/emailpreferences/>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu <mailto:llvm-commits at cs.uiuc.edu>
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits <http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150205/337a5c43/attachment.html>

More information about the llvm-commits mailing list