Possible LLVM error in tune_nnvm

Questions
1

Hi,

I have found some unexpected errro when testing tune_nnvm_x86.py tutorial.

Once I changed `model_name` to "mxnet" I am getting only error:
terminate called after throwing an instance of 'std::system_error'
  what():  Operation not permitted

Thread 58 "python3" received signal SIGABRT, Aborted.
[Switching to Thread 0x7fffee172700 (LWP 26229)]
__GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:51
51      ../sysdeps/unix/sysv/linux/raise.c: No such file or directory.
(gdb) bt
#0  0x00007ffff7a22e97 in __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:51
#1  0x00007ffff7a24801 in __GI_abort () at abort.c:79
#2  0x00007fffdadfc8b7 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#3  0x00007fffdae02a06 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#4  0x00007fffdae02a41 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#5  0x00007fffdae02c74 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#6  0x00007fffdadfe979 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#7  0x00007fffbd1a06d3 in dmlc::ConcurrentBlockingQueue<mxnet::engine::OprBlock*, (dmlc::ConcurrentQueueType)1>::Pop(mxnet::engine::OprBlock**) ()
    at /home/mypc/.local/lib/python3.6/site-packages/mxnet/libmxnet.so
#8  0x00007fffbd1a073e in std::_Function_handler<void (), mxnet::engine::ThreadedEnginePerDevice::Start()::{lambda()#1}>::_M_invoke(std::_Any_data const&) ()
    at /home/mypc/.local/lib/python3.6/site-packages/mxnet/libmxnet.so
#9  0x00007fffdae2d8f0 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#10 0x00007ffff77cc6db in start_thread (arg=0x7fffee172700) at pthread_create.c:463
#11 0x00007ffff7b0588f in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95

I am using only LLVM (LLVM 6), RPC and GRAPH_RUNTIME in config.cmake. Despite that I am still getting this error. Same happens with OpenCL enabled as well.

I am running Ubuntu 18.04 with Python 3.6.

Is it some TVM bug, or is there already some solution to resolve that?

2

Can someone please comment on that?

3

Based on my debugging I managed to find that it stops at one of LLVMModuleNode::Init calls. I am running it on 4 physical core CPU and I can pass some of the trials during tuning. When I am running e.g. resnet-18 from other choices I am running all 8 trials.