Re: [Valgrind-users] Long valgrind runtimes

[Nicholas N. <nj...@cs...>]

On Tue, 10 Jul 2007, Christoph Bartoschek wrote:

>> Ah, that's interesting.  We've had performance problems in m_mallocfree.c
>> before now.  So I am not (entirely) surprised to see it again.
>>
>> Can you describe a bit the malloc/free behaviour of your application?
>>
>> - how many blocks live (approximately)
>> - approximately what size
>> - are you allocating fast, or slow?
>> - approximately what lifetimes?  more or less LIFO, or FIFO, or random?
>>
>> The ultimate aim is to construct a simple test case which shows the
>> same performance problem.
>
> It's hard to answer the code because it is more than ten years old and the
> original developers are not here anymore.

What would be most useful is a trace of the malloc/free calls.

Can you add calls to VG_(printf)() in coregrind/m_mallocfree.c, to the 
functions VG_(arena_malloc) and VG_(arena_free)  (and also the ones for 
VG_(arena_realloc), VG_(arena_strdup), VG_(malloc_aligned) might be useful).

These should print out all the relevant info, eg. the size for malloc(), and 
the pointer for free().

If you send us a trace of this, hopefully that will let us identify the 
problem.  Thanks.

Nick