Crash on deallocation

[vbudovski]

Hi,

I'm having a rather strange issue with sometimes difficult to reproduce crashes.
I make use of tbb::concurrent_vector and make one parallel_for call. The rest of the accesses are done serially.
I've tried it with both scalable and cache aligned allocators with similar results.

I'm using tbb version 3.0+r147 as available in debian/ubuntu releases.

Any ideas?

Please see attached backtrace bellow:

0x00007ffff743996d in rml::internal::getBackRef (backRefIdx=...) at ../../src/tbbmalloc/backref.cpp:155
155 + sizeof(BackRefBlock)+backRefIdx.getOffset()*sizeof(void*));
(gdb) bt
#0 0x00007ffff743996d in rml::internal::getBackRef (backRefIdx=...) at ../../src/tbbmalloc/backref.cpp:155
#1 0x00007ffff7437fd0 in isLargeObject (object=0x7fffe6d79100) at ../../src/tbbmalloc/frontend.cpp:1577
#2 rml::internal::isLargeObject (object=0x7fffe6d79100) at ../../src/tbbmalloc/frontend.cpp:1570
#3 0x00007ffff7438073 in rml::internal::internalFree (object=0x7fffe6d79100) at ../../src/tbbmalloc/frontend.cpp:1724
#4 0x0000000000427242 in tbb::scalable_allocator<:GRIB2::MESSAGE>::deallocate (this=0x7fffffffdb40, p=0x7fffe6d79100) at /usr/include/tbb/scalable_allocator.h:146
#5 0x0000000000423f0c in tbb::concurrent_vector<:GRIB2::MESSAGE> >::internal_free_segments (this=0x7fffffffdb40, table=0x7ffff7fb3c00, k=9, first_block=1)
at /usr/include/tbb/concurrent_vector.h:927
#6 0x0000000000421348 in tbb::concurrent_vector<:GRIB2::MESSAGE> >::~concurrent_vector (this=0x7fffffffdb40, __in_chrg=) at /usr/include/tbb/concurrent_vector.h:823
#7 0x000000000041f880 in meteor::grib2::grib2::~grib2 (this=0x7fffffffdb40, __in_chrg=) at ../../lib/meteor/include/meteor/grib2/grib2.hpp:45
#8 0x0000000000412918 in main (argc=7, argv=0x7fffffffe048) at main.cpp:224

[Alexandr_K_Intel1]

It was a race is situation similar to your, I hope it was fixed. Could you please try fresh TBB, preferrably 4.0 u3 from http://threadingbuildingblocks.org/file.php?fid=78?

[vbudovski]

Upgrading seems to have solved the issue. Do you know which release this bug was fixed in?

Edit: Something is still not quite right, will keep looking.

[vbudovski]

Seem like I had a bug where not enough memory was allocated for output; which in combination with multiple threads produced some rather bizarre issues. TBB seems to be fine. Sorry for the noise.