tbb not found error

[KK666-AI]

hi, when i make your latest version code, i meet an error like:"src/nn/row_selection.cpp:3:21: fatal error: tbb/tbb.h: No such file or directory". So i want to ask you, which version of tbb do you used? could you give me the installation tutorial?

[Hanjun-Dai]

Hi there, I think you need to add tbb's include and lib dirs into environment variables. Typically you only need to put the following line in your .bashrc: source /opt/intel/parallel_studio_xe_{replace_with_your_version}/bin/psxevars.h The above line should solve most of the problems.

I've also added a line which manually include tbb's headers in Makefile: https://github.com/Hanjun-Dai/graphnn/commit/b71ee43eb14d7c714b10f00a3473f724c8ac86da

Let me know if that solves your problem.

[KK666-AI]

I follow your suggestion, but in my folder "/opt/intel/parallel_studio_xe_2017.4.056/bin", there are only psxevars.sh and psxevars.csh, without psxevars.h. i use 'sudo find . -name tbb.h' cannot find tbb.h file. So I try to download "tbb2017_20170604oss" from https://github.com/01org/tbb/releases, then I add "tbb2017_20170604oss/lib" and "tbb2017_20170604oss/lib/intel64" into LD_LIBRARY_PATH, add "tbb2017_20170604oss/bin:/opt/intel/parallel_studio_xe_2017/bin" into PATH, BUT I STILL CANNOT run your example.

BTW, follow the tutorial in https://github.com/Hanjun-Dai/graphnn, i modify configurations in make_common file like below, i can build the graphnn. Could your give me a brief tutorial for how to fix the tbb.h error, and run your example graph_classification? dir_guard = @mkdir -p $(@D)

CUDA_HOME := /usr/local/cuda INTEL_ROOT := /opt/intel MKL_ROOT = $(INTEL_ROOT)/mkl TBB_ROOT = /home/aliant/Downloads/tbb2017_20170604oss USE_GPU = 1

FIND := find CXX := g++ NVCC := $(CUDA_HOME)/bin/nvcc NVCCFLAGS += --default-stream per-thread CXXFLAGS += -Wall -O3 -std=c++11 LDFLAGS += -lm -L$(CUDA_HOME)/lib64 -lcudart -lcublas -lcurand -lcusparse -lmkl_rt -ltbb -lfmt CUDA_ARCH := -gencode arch=compute_30,code=sm_30 \ -gencode arch=compute_35,code=sm_35 \ -gencode arch=compute_50,code=sm_50 \ -gencode arch=compute_50,code=compute_50

[KK666-AI]

finally, i fix this error. anyway, thanks a lot.

[Hanjun-Dai]

That's great! Sorry about the confusion.

I'll close the issue.

[MaohuaLiu11]

Hi Hanjun-Dai and lihuiknight, I meet the same problem, an error like "src/nn/row_selection.cpp:3:21: fatal error: tbb/tbb.h: No such file or directory", I think it should be version problem, because I install the 2020-version mkl and find that there is no "tbb.h" in this version, I also checked that with a 'find' command and find no "tbb.h" there. I also followed the link supplied by lihuiknight to fix this problem, but there is still some problem when I try to use 'make' to compile the codes. Can you guys give me suggestions about that? Thanks!

[MaohuaLiu11]

BTW, my OS is Ubuntu18.04, my mkl is 2020 version, almost all environments are the latest version

[Hanjun-Dai]

Hi,

The MKL should be just part of the parallel studio package of Inter. You need to additionally install tbb as well (or simply install the full parallel studio).

[MaohuaLiu11]

Sure...I will try to do that. I have applied for the parallel_studio_xe which will be available some days later. I also installed tbb additionally. Generally speaking … I can compile "graphnn" now … that means, there is no error in the process, but there are a lot of warnings like " void mkl_dcsrmm(...) is deprecated", just like below:

" warning: ‘void mkl_dcsrmm(const char, const int, const int, const int, const double, const char, const double, const int, const int, const int, const double, const int, const double, double, const int*)’ is deprecated "

Does this warning matter? Besides, I have no GPU in my computer...so I set USE_GPU = 0 and find that there are still some errors related to GPU and cuda... I guess some functions related to GPU are still compiled in the process ... so I want to know have you ever tried to compile the "graphnn" by setting USE_GPU=0? Thanks~~

[Hanjun-Dai]

You need to set USE_GPU = 0 in both Makefile : the library one and the one for each executable (e.g., MNIST)

[MaohuaLiu11]

Sure...I have set USE_GPU = 0 in both Makefile and the running result of my MNIST is following:

root@VirtualBox:/home/graphnn-master/examples/mnist# ./run.sh 60000 images for training 10000 images for test testing test loss: 2.5660 error rate: 0.8463 testing test loss: 0.1245 error rate: 0.0405 testing test loss: 0.1085 error rate: 0.0325 testing test loss: 0.0921 error rate: 0.0285 testing test loss: 0.0799 error rate: 0.0237 testing test loss: 0.0693 error rate: 0.0203 testing test loss: 0.0697 error rate: 0.0197 testing test loss: 0.0727 error rate: 0.0206 testing test loss: 0.0672 error rate: 0.0180 testing test loss: 0.0648 error rate: 0.0178 Assertion pt_info.count(id) failed in src/util/mem_holder.cpp line 45: recycling a pointer which is not allocated by me ./build/mnist(+0x39a73)[0x564c893dca73] ./build/mnist(+0xbfca)[0x564c893aefca] ./build/mnist(+0xed1c)[0x564c893b1d1c] /lib/x86_64-linux-gnu/libc.so.6(+0x43041)[0x7ff2e4e76041] /lib/x86_64-linux-gnu/libc.so.6(+0x4313a)[0x7ff2e4e7613a] /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xee)[0x7ff2e4e54b9e] ./build/mnist(+0xbd3a)[0x564c893aed3a] terminate called without an active exception

Do you think it is normal or not? Thanks!

[MaohuaLiu11]

Besides, for the example of 'graph_classification', I have also set USE_GPU = 0 in both Makefile but there are still some errors related to GPU when I tried to make it, just as follow:

include/nn_common.h:32:22: error: cannot declare variable ‘input’ to be of abstract type ‘gnn::TensorTemplate<gnn::GPU, gnn::DENSE, float>’

nclude/nn_common.h:33:23: error: cannot declare variable ‘label’ to be of abstract type ‘gnn::TensorTemplate<gnn::GPU, gnn::CSR_SPARSE, float>’

include/nn_common.h:101:8: error: ‘gnn::DTensor<gnn::GPU, float> {aka class gnn::TensorTemplate<gnn::GPU, gnn::DENSE, float>}’ has no member named ‘CopyFrom’

src/kernel_mean_field.cpp:19:15: error: ‘gnn::DTensor<gnn::GPU, float> {aka class gnn::TensorTemplate<gnn::GPU, gnn::DENSE, float>}’ has no member named ‘SetRandN’

src/kernel_mean_field.cpp:32:59: error: no matching function for call to ‘af<template<class mode, class Dtype> class gnn::ReLU>(gnn::FactorGraph&, )’

What do you suggest? Thanks!