Okina memory manager error

[artv3]

@jdahm , @camierjs ,

I pulled the latest version of Okina and tried to run example 9. Unfortunately, I hit the following error
mm::Erase Aborted (core dumped)

I configured MFEM in the following manner
make config MFEM_CXX="nvcc" \ CXXFLAGS="-g -O0 --restrict --expt-extended-lambda -x=cu -arch=sm_60 -std=c++11 -m64" \ MFEM_EXT_LIBS="-L/usr/local/cuda/lib64-lrt -lcuda -lcudart -lcudadevrt -lnvToolsExt" \ MFEM_DEBUG=YES OPTIM_FLAGS=-O0 MFEM_USE_MM=YES

Commit # 76314fa works
but the code breaks at commit cf54f42

[jdahm]

I'm sorry this is happening to your build! I can't seem to quickly reproduce the problem on my laptop using the GCC compiler on the okina branch. Could you post or send me a backtrace?

[artv3]

``mm::Erase

Program received signal SIGABRT, Aborted.
0x000010000120eb18 in raise () from /lib64/libc.so.6
Missing separate debuginfos, use: debuginfo-install glibc-2.17-157.el7_3.5.ppc64le
(gdb) bt
#0 0x000010000120eb18 in raise () from /lib64/libc.so.6
#1 0x0000100001210c9c in abort () from /lib64/libc.so.6
#2 0x0000000010017ea8 in mfem::mfem_error (msg=0x103085b0 "mm::Erase") at /usr/WS1/vargas45/Git-Repos/MFEM_BUILD/okina-broken/general/error.cpp:175
#3 0x0000000010018c18 in mfem::mm::Erase (this=0x10512928 mfem::mm::MM()::singleton, ptr=0x10530030) at /usr/WS1/vargas45/Git-Repos/MFEM_BUILD/okina-broken/general/mm.cpp:113
#4 0x0000000010011340 in mfem::mm::free (ptr=0x10530030) at ../general/../general/mm.hpp:84
#5 0x000000001000e05c in mfem::Vector::~Vector (this=0x105128f0 <bb_max>, __in_chrg=) at ../linalg/vector.hpp:399
#6 0x00001000012135a4 in __run_exit_handlers () from /lib64/libc.so.6
#7 0x00001000012135f4 in exit () from /lib64/libc.so.6
#8 0x00001000011f4708 in generic_start_main.isra.0 () from /lib64/libc.so.6
#9 0x00001000011f48f4 in __libc_start_main () from /lib64/libc.so.6
#10 0x0000000000000000 in ?? ()``

[artv3]

Some more clues
(cuda-gdb) frame 3
#3 0x0000000010018c18 in mfem::mm::Erase (this=0x10512928 mfem::mm::MM()::singleton, ptr=0x10530030) at /usr/WS1/vargas45/Git-Repos/MFEM_BUILD/okina-broken/general/mm.cpp:113
113 if (!known) { mfem_error("mm::Erase"); }
(cuda-gdb) list
108 {
109 if (!config::usingMM()) { return ptr; }
110 if (config::gpuDisabled()) { return ptr; }
111 const bool known = Known(maps, ptr);
112 // if (!known) { BUILTIN_TRAP; }
113 if (!known) { mfem_error("mm::Erase"); }
114 MFEM_ASSERT(known, "Trying to remove an unknown address!");
115 const memory &mem = maps.memories.at(ptr);
116 dbg("\033[33m %p \033[35m(%ldb)", ptr, mem.bytes);
117 for (const alias* const alias : mem.aliases)

[artv3]

Resolved.