Memory pool bug

[kyungjoo-kim]

I have a problem in memory pool. The error message is

Kokkos MemoryPool::deallocate given erroneous pointer
Program received signal SIGABRT, Aborted.

gdb says....

#2  0x000000000044bc21 in Kokkos::Impl::host_abort (
    message=message@entry=0x4881c0 "Kokkos MemoryPool::deallocate given erroneous pointer")
    at /home/kyukim/Work/lib/trilinos/temp/packages/kokkos/core/src/impl/Kokkos_Error.cpp:64
#3  0x000000000043f528 in abort (message=0x4881c0 "Kokkos MemoryPool::deallocate given erroneous pointer")
    at /home/kyukim/Work/lib/trilinos/temp/packages/kokkos/core/src/impl/Kokkos_Error.hpp:79
#4  Kokkos::MemoryPool<Kokkos::OpenMP>::deallocate (this=this@entry=0x7fffcbdec840, p=p@entry=0x7ffc34636280)
    at /home/kyukim/Work/lib/trilinos/temp/packages/kokkos/core/src/Kokkos_MemoryPool.hpp:796
#5  0x0000000000449f3e in Tacho::Experimental::TaskFunctor_FactorizeChol<double, Kokkos::OpenMP>::factorize_internal (
    this=this@entry=0x7fffcbdec830, member=..., n=<optimized out>, final=<optimized out>)
    at /home/kyukim/Work/lib/trilinos/temp/packages/shylu/shylu_node/tacho/src/TachoExp_TaskFunctor_FactorizeChol.hpp:75
#6  0x000000000044a24f in Tacho::Experimental::TaskFunctor_FactorizeChol<double, Kokkos::OpenMP>::operator() (
    this=this@entry=0x7fffcbdec830, member=..., r_val=<optimized out>)
    at /home/kyukim/Work/lib/trilinos/temp/packages/shylu/shylu_node/tacho/src/TachoExp_TaskFunctor_FactorizeChol.hpp:91

[hcedwar]

There is print statement at
https://github.com/kokkos/kokkos/blob/master/core/src/Kokkos_MemoryPool.hpp#L793
that you could enable to get detailed information.

[kyungjoo-kim]

Neither of them really is my case

log.txt

Kokkos MemoryPool allocate(0x7f61fb236280)
Kokkos MemoryPool deallocate(0x7f61fb236280) contains(1) block_aligned(1) dealloc_once(0)

I only found the allocate and deallocate of the address 0x7f61fb236280) with the above pair only.

[kyungjoo-kim]

log4.txt

According to this log output, the problematic pointer has pairs of allocations and deallocations before it crashes.

[hcedwar]

Problem size is requesting superblocks of 1lu << 31 or more. There are several 32-vs-64 bit conversion bugs.

[hcedwar]

@kyungjoo-kim : Please email MemoryPool source file with our bug fixes.

[kyungjoo-kim]

Still error with intel compiler

++ pool(0x7f691d236100)   allocate(0x7f6a9e236280) from sb_id(3) alloc_size(2,091,270,888) blocksize_lg2(31) sb_state(0x0) count_lg2(0) result(0,1)             
-- pool(0x7f691d236100) deallocate(0x7f6a9e236280) from sb_id(3) result(2) bit(1536) state(0x24000002) d(180000000) m_sb_size_lg2(31)                           
Kokkos MemoryPool deallocate(0x7f6a9e236280) contains(1) block_aligned(0) dealloc_once(0)