[LLVMdev] Running Pool Allocated programs

Sat Apr 2 18:13:44 PST 2005

On Sat, 2 Apr 2005, xavier wrote:
> Thanks Chris! That was causing me some confusion. Now it's producing a more meaningful output:
>  poolinit((&l2_GlobalPool), 4u, 4u);
>  poolinit((&l2_GlobalPool1), 4u, 4u);
>  l5_tmp_2E_0_2E_i = poolalloc((&l2_GlobalPool), 4u);
>  l6_tmp_2E_7_2E_i2 = l1_makeList((&l2_GlobalPool1), 0);
>  *(&((struct l_struct_2E_list *)l5_tmp_2E_0_2E_i)->field0) = l6_tmp_2E_7_2E_i2;
>
> However, if I am understanding correctly, the first node is being allocated in one pool (that was
> malloc with the other heuristic) and the rest are being allocated in a differente pool. But why?
> Why not all the nodes are being allocated in one pool?

The reason is that there is no pointer in the program that points at both 
the first node and any of the rest.  DSA decides they don't alias and APA 
puts them in different pools.  If you write a loop to traverse the list 
(for example), they will end up in the same pool.

> Also, in your other response you said:
>
>>> The AllNodes heuristic is the only one that is substantially tested.  I
>>> would recommend it, none of the others have any significant performance
>>> benefit anyway.
>
> Is that still right or you meant that only SelectivePA is the only substantially tested?

Sorry, I was confused/wrong.  AllNodes and "AllButUnreachableFromMemory" 
heuristics are both well tested.  The other heuristics will probably work 
but no guarantees :)

> P.S. BTW, SelectivePA does not appear as an option when you do
>
> opt -load ~/llvmdbg/installed/lib/libpoolalloc.so --help
> and I could not find any reference in the source code (Heuristic.{h,cpp})

It is the "AllButUnreachableFromMemory" heuristic (the default).  Terse 
names are overrated. :)

-Chris

> May I assume that is a new heuristic or you are referring to one of the current ones?
>
>  -poolalloc-heuristic                    - Heuristic to choose which nodes to pool allocate
>    =AllNodes                             -   Pool allocate all nodes
>    =AllButUnreachableFromMemory          -   Pool allocate all reachable from memory objects
>    =CyclicNodes                          -   Pool allocate nodes with cycles
>    =SmartCoallesceNodes                  -   Use the smart node merging heuristic
>    =AllInOneGlobalPool                   -   Use pool library as replacement for malloc/free
>    =OnlyOverhead                         -   Do not pool allocate anything, but induce all
> overhead from it
>    =NoNodes                              -   Do not pool allocate anything
>
>
>
>
> --- Chris Lattner <sabre at nondot.org> wrote:
>
>> On Sat, 2 Apr 2005, Ricardo wrote:
>>> After applying the PA to it, the output is something like this:
>> ...
>>
>>> My question is: why is this malloc necessary?
>>> ltmp_2_5 = malloc(4u);
>>
>>> Shouldn't be the result a program with this malloc replaced by poolalloc?
>>> Should I include a special flag to achieve this?
>>
>> Ah, sorry, my memory was wrong.  The default is to perform the
>> "SelectivePA" optimization.  Since pool allocation detected that there is
>> no point in putting a single allocation into a pool, it left it a malloc.
>> If you really want it to be a in a pool, pass
>> -poolalloc-heuristic=AllNodes to opt.
>>
>> Sorry for the confusion!
>>
>> -Chris
>>
>>> I am using this line:
>>> opt -load ~/llvm/install/lib/libpoolalloc.so -poolalloc
>>> and also with -pooloptimize
>>>
>>> Thanks in advance
>>>
>>>
>>>
>>>
>>> --- Chris Lattner <sabre at nondot.org> wrote:
>>>
>>>> On Sat, 2 Apr 2005, Ricardo wrote:
>>>>> Hello,
>>>>> I am trying to run a program optimized with the Pool Allocation, but I am receiving this
>>>> error:
>>>>>
>>>>> ====================
>>>>>
>>>>> *** 4 DYNAMIC POOLS INITIALIZED ***
>>>>>
>>>>> *** 4 DYNAMIC POOLS ALLOCATED FROM ***
>>>>>
>>>>> MaxHeapSize = 0.062500KB  HeapSizeAtExit = 0.062500KB   NOTE: only valid if using
>>>>> Heuristic=AllPools and no bumpptr/realloc!
>>>>> ====================
>>>>
>>>> This output is what happens when you link with the debug version of the
>>>> runtime library.  If you rebuild the FL2Allocator with 'make
>>>> ENABLE_OPTIMIZED=1' and link with the result, you shouldn't get this.
>>>>
>>>>> I have tried to link the program with the Analysis and with Datastructure libs:
>>>>> llvm-gcc -o test test.opt.bc -lpoolalloc -lpoolalloc_rt -lLLVMDataStructure -lLLVMAnalysis
>>>>> But I am still receiving the error.
>>>>
>>>> You should only need -lpoolalloc_rt.
>>>>
>>>> For the record, I usually do (effectively) this to test the pool
>>>> allocator:
>>>>
>>>> llvm-gcc *.c -o program         # compile&link whole prog with LLVM
>>>> opt -load ... -poolalloc program.bc ... -o program.pa.bc
>>>> llc program.pa.bc -march=c -o program.pa.cbe.c
>>>> gcc -fno-strict-aliasing -O2 program.pa.cbe.c -o program.pa
>>>> -L~/llvm/projects/llvm-poolalloc/Release/lib/ -lpoolalloc_rt
>>>>
>>>> Alternatively, if you want a quick test, you can do the first two steps,
>>>> then:
>>>>
>>>>    lli -load ~/llvm/projects/llvm-poolalloc/Release/lib/libpoolalloc_rt.so program.pa.bc
>>>>
>>>> If you're on a target that supports the JIT.
>>>>
>>>>> Another question, I do not want to use the bump ptr optimization, and I suppose that the
>> right
>>>>> heuristic for that is AllNodes? But if I want to use it, what is the right heuristic?
>>>>
>>>> The AllNodes heuristic is the only one that is substantially tested.  I
>>>> would recommend it, none of the others have any significant performance
>>>> benefit anyway.
>>>>
>>>> The bump pointer and other opts are a seperate option -pooloptimize, which
>>>> can be run after -poolalloc.  It does not work with -pointercompress
>>>> though.
>>>>
>>>>> Also, if I want to use the Pointer Compression, I suppose that I have to
>>>>> use the -pointercompress switch with opt or what?
>>>>
>>>> Yes, if you use it, don't specify -poolalloc manually.
>>>>
>>>> -Chris
>>>>
>>>> --
>>>> http://nondot.org/sabre/
>>>> http://llvm.cs.uiuc.edu/
>>>>
>>>
>>> _______________________________________________
>>> LLVM Developers mailing list
>>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>>> http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev
>>>
>>
>> -Chris
>>
>> --
>> http://nondot.org/sabre/
>> http://llvm.cs.uiuc.edu/
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev
>>
>
>
>
>
> __________________________________
> Do you Yahoo!?
> Yahoo! Small Business - Try our new resources site!
> http://smallbusiness.yahoo.com/resources/
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev
>

-Chris

-- 
http://nondot.org/sabre/
http://llvm.cs.uiuc.edu/