gpgpusim : Aborted (core dumped) When run gpgpusim_simulation.

[libingict]

I was trying to run the gpgpu-sim_simulation.

When I was trying to run the benmark ibackprop, the following error pops:

terminate called after throwing an instance of 'std::ios_base::failure[abi:cxx11]'
  what():  basic_ios::clear: iostream error
Aborted (core dumped)

The command I use is following:

/root/gpgpusim-dev/gpgpu-sim_distribution/gpgpu-sim_simulations/util/job_launching/../../benchmarks/bin/4.2/release/backprop-rodinia-2.0-ft 4096 ./data/result-4096.txt

The following are my work environment:

echo $GPGPUSIM_CONFIG

> gcc-4.5.3/cuda-4020/release

echo $GPGPUSIM_ROOT

> /root/gpgpusim-dev/gpgpu-sim_distribution

My cuda version is cudnn 4.2.

nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2012 NVIDIA Corporation
Built on Thu_Apr__5_00:24:31_PDT_2012
Cuda compilation tools, release 4.2, V0.2.1221

Is the error due to my cuda version? Is there anyone know how to fix this issue?

[rohitsakala]

I am also facing the same error. Any update ?

[tgrogers]

Thanks for raising the issue - is it just backprop that is failing, or are all workloads encountering this issue? Which OS are you using?

…

On Mon, Jul 16, 2018 at 1:49 AM Sakala Venkata Krishna Rohit < ***@***.***> wrote: I am also facing the same error. Any update ? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#73 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AD9McRGiqa2u4s7C8eb0Ew20UZeWFvdEks5uHClmgaJpZM4UOctb> .

-- Tim Rogers Assistant Professor Purdue University https://engineering.purdue.edu/tgrogers

[libingict]

Hi, @tgrogers

Thanks for your caring. I was using the ubuntu 16 in a docker container and did not try other workloads.
Now I am using gpgpusim on ubuntu 14. I directly downloaded rodinia_2.0.1 from the website and recompiled the benchmark suites.
Then gpgpusim can run and no error occurs.

Here is the OS version:
Distributor ID: Ubuntu Description: Ubuntu 14.04 LTS Release: 14.04 Codename: trusty
The cuda toolkit version
nvcc -V nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2011 NVIDIA Corporation Built on Thu_May_12_11:09:45_PDT_2011 Cuda compilation tools, release 4.0, V0.2.1221
The gcc version
gcc version 4.4.7 (Ubuntu/Linaro 4.4.7-8ubuntu1)
Hope my feedback will be helpful.
Thanks again.

[bfeinberg]

This likely comes from the change in the string ABI around c++11 discussed here: https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html.

Ubuntu 16.04's c++ standard lib uses the dual ABI but g++ 4.x doesn't know about it you see the crash above. If you attach a debugger you'll see that it occurs in the option_parser and so is completely workload independent.

Luckily the fix is relatively simple. Build gpgpu-sim using g++ 5.x (I've used 5.4 successfully) when using Ubuntu 16.04. This can create a bit of annoyance when working with CUDA 4.0 as 5.4 is above the max version nvcc will allow but you can switch gcc/g++ versions when compiling applications back to 4.x.

[rdevans0]

Can confirm that I also ran into this issue attempting to compile with gcc-4.8 on Ubuntu 16.04. Solution was to compile with the built in gcc-5.4.0.

[shenjiangqiu]

Should we add some #if macro to judge the GCC version in OptionParse.cc? That problem may make a lot of trouble the guys new to GPGPUSIM( as to me before)

[HUSTzhaowei]

Hello guys,I have the same question and I cannot build GPGPUsim successfully with GCC 5.4.0. Is there anyone know how to fix this issue? Thanks！

[shenjiangqiu]

one good method to solve this problem is to using gdb to debug it and when segment fault happened, using bt to print stack trace. I guess you can see the segment fault happened when parsing the args, be cause the libc++ cannot handle the exception from ostream throwout. try to refer to cppreference to find more detail about it Get Outlook for iOS<https://aka.ms/o0ukef>

…

________________________________ From: 赵威 <notifications@github.com> Sent: Tuesday, October 8, 2019 5:45:48 AM To: gpgpu-sim/gpgpu-sim_distribution <gpgpu-sim_distribution@noreply.github.com> Cc: shenjiangqiu <ivy22233qiu@live.com>; Comment <comment@noreply.github.com> Subject: Re: [gpgpu-sim/gpgpu-sim_distribution] gpgpusim : Aborted (core dumped) When run gpgpusim_simulation. (#73) Hello guys,I have the same question and I cannot build GPGPUsim successfully with GCC 5.4.0. Is there anyone know how to fix this issue? Thanks！ ― You are receiving this because you commented. Reply to this email directly, view it on GitHub<#73?email_source=notifications&email_token=AEM4BF3IGRE25WYOCAIXJ5DQNRJEZA5CNFSM4FBZZNN2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEATSVIY#issuecomment-539437731>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AEM4BF2ZW5V2WTJOENAXQMLQNRJEZANCNFSM4FBZZNNQ>.

[jayresearch]

Here is how i have resolved this issue in:
Ubuntu 16.04
built the simulator using
gcc-5.4/g++-5.4
and compiled cuda code using gcc-4.4
The suggestion by bfeinberg worked for me.
I have to resolve the following additional compilation errors:
in file: instructions.cc
replaced: #include<math.h> with #include
replaced: isnan with std::isnan (also in file cuda-math.cc)

Added .str() in the following lines in file cuda_runtime_api.cc
std::cout << "Failed to execute: " << cmd.str() << std::endl;
std::cout << "Trying to parse " << libcodfn.str() << std::endl;
The above error was due to ostream and stringstream mismatch.

best!

[lucmann]

@jayresearch I wonder if you still encountered the std::stringstream problems with GCC version < 5 to build the simulator. In fact when I ran make with GCC version 9.2.1, those two std::stringstream object missing .str() issues occurred. It doesn't seem to be the same with the issue libingict posted here, does it? But I really wonder how others get it passed through.

P.S. my OS version Ubuntu 19.10, arch x86_64, cuda 4.2.9