Trouble Installing TVM Extension about tvm

[sanjaybharadwaj]

I hope you created a symbolic link to llvm-config. By default the llvm-8 installation installs llvm-config-8, so you can create a symbolic link by

ln -s /usr/bin/llvm-config-8 /usr/bin/llvm-config

[14jyuan]

I just tried creating a symbolic link, but I got the same error. I think it might be an issue with cmake rather than llvm?

[bwasti]

perhaps you need to git submodule update --init --recursive?

if not, after creating that symlink, could you try git clean -xfd and rebuilding with python setup.py install --cmake?

[14jyuan]

It fixed my issue with python setup.py install, but it errors when I try to run python setup.py test:

=========================================================== ERRORS ============================================================
_____________________________________________ ERROR collecting test/test_core.py ______________________________________________
ImportError while importing test module '/home/ubuntu/tvm/test/test_core.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
test/test_core.py:2: in
from test.util import TVMTest
test/util.py:12: in
import torch_tvm
torch_tvm/init.py:9: in
from ._torch_tvm import *
E ImportError: /home/ubuntu/tvm/torch_tvm/_torch_tvm.cpython-37m-x86_64-linux-gnu.so: undefined symbol: ZN5torch3jit16SubgraphRewriter22RegisterRewritePatternERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES9
____________________________________________ ERROR collecting test/test_models.py _____________________________________________
ImportError while importing test module '/home/ubuntu/tvm/test/test_models.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
test/test_models.py:6: in
import torch_tvm
torch_tvm/init.py:9: in
from ._torch_tvm import *
E ImportError: /home/ubuntu/tvm/torch_tvm/_torch_tvm.cpython-37m-x86_64-linux-gnu.so: undefined symbol: ZN5torch3jit16SubgraphRewriter22RegisterRewritePatternERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES9
___________________________________________ ERROR collecting test/test_operators.py ___________________________________________
ImportError while importing test module '/home/ubuntu/tvm/test/test_operators.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
test/test_operators.py:2: in
from test.util import TVMTest
test/util.py:12: in
import torch_tvm
torch_tvm/init.py:9: in
from ._torch_tvm import *
E ImportError: /home/ubuntu/tvm/torch_tvm/_torch_tvm.cpython-37m-x86_64-linux-gnu.so: undefined symbol: ZN5torch3jit16SubgraphRewriter22RegisterRewritePatternERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEES9
====================================================== warnings summary =======================================================
/home/ubuntu/anaconda3/lib/python3.7/site-packages/pywt/_utils.py:6
/home/ubuntu/anaconda3/lib/python3.7/site-packages/pywt/_utils.py:6: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
from collections import Iterable

-- Docs: https://docs.pytest.org/en/latest/warnings.html
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Interrupted: 3 errors during collection !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
============================================= 1 warnings, 3 error in 1.57 seconds =============================================

[14jyuan]

Could the undefined symbol errors have something to do with the PyTorch version?

[bwasti]

Yep, it certainly could be the version of PyTorch, but RegisterRewritePattern has been around for a while. If you're willing, I'd recommend building PyTorch locally. Otherwise, a newer nightly might help.

The other issue might be a mismatch caused by the __cxx11 symbol, which might be resolved by adding -D_GLIBCXX_USE_CXX11_ABI=0 to your build command. I'm going off this StackOverflow post: https://stackoverflow.com/questions/33394934/converting-std-cxx11string-to-stdstring.

More concretely that would mean changing https://github.com/pytorch/tvm/blob/master/CMakeLists.txt#L10 to either

SET(CMAKE_CXX_FLAGS  "${CMAKE_CXX_FLAGS} -D_GLIBCXX_USE_CXX11_ABI=0")

or

SET(CMAKE_CXX_FLAGS  "${CMAKE_CXX_FLAGS} -D_GLIBCXX_USE_CXX11_ABI=1")

[14jyuan]

I tried doing rebuilding the nightly version of PyTorch and changing the line in CMakeLists.txt, and I think it fixed my previous errors, but now I have new errors with python setup.py test:

_________________________________________________________ ERROR collecting test/test_core.py __________________________________________________________
test/test_core.py:2: in
from test.util import TVMTest
test/util.py:12: in
import torch_tvm
torch_tvm/init.py:7: in
from tvm import relay # This registers all the schedules
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/init.py:23: in
from . import tensor
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/tensor.py:20: in
from ._ffi.node import NodeBase, NodeGeneric, register_node, convert_to_node
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/node.py:24: in
from .node_generic import NodeGeneric, convert_to_node, const
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/node_generic.py:23: in
from .base import string_types
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/base.py:60: in
_LIB, _LIB_NAME = _load_lib()
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/base.py:52: in _load_lib
lib = ctypes.CDLL(lib_path[0], ctypes.RTLD_GLOBAL)
../anaconda3/lib/python3.7/ctypes/init.py:356: in init
self._handle = _dlopen(self._name, mode)
E OSError: /home/ubuntu/anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so: undefined symbol: _ZN4llvm14TargetRegistry12lookupTargetERKSsRSs
________________________________________________________ ERROR collecting test/test_models.py _________________________________________________________
test/test_models.py:6: in
import torch_tvm
torch_tvm/init.py:7: in
from tvm import relay # This registers all the schedules
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/init.py:23: in
from . import tensor
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/tensor.py:20: in
from ._ffi.node import NodeBase, NodeGeneric, register_node, convert_to_node
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/node.py:24: in
from .node_generic import NodeGeneric, convert_to_node, const
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/node_generic.py:23: in
from .base import string_types
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/base.py:60: in
_LIB, _LIB_NAME = _load_lib()
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/base.py:52: in _load_lib
lib = ctypes.CDLL(lib_path[0], ctypes.RTLD_GLOBAL)
../anaconda3/lib/python3.7/ctypes/init.py:356: in init
self._handle = _dlopen(self._name, mode)
E OSError: /home/ubuntu/anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so: undefined symbol: _ZN4llvm14TargetRegistry12lookupTargetERKSsRSs
_______________________________________________________ ERROR collecting test/test_operators.py _______________________________________________________
test/test_operators.py:2: in
from test.util import TVMTest
test/util.py:12: in
import torch_tvm
torch_tvm/init.py:7: in
from tvm import relay # This registers all the schedules
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/init.py:23: in
from . import tensor
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/tensor.py:20: in
from ._ffi.node import NodeBase, NodeGeneric, register_node, convert_to_node
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/node.py:24: in
from .node_generic import NodeGeneric, convert_to_node, const
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/node_generic.py:23: in
from .base import string_types
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/base.py:60: in
_LIB, _LIB_NAME = _load_lib()
../anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/_ffi/base.py:52: in _load_lib
lib = ctypes.CDLL(lib_path[0], ctypes.RTLD_GLOBAL)
../anaconda3/lib/python3.7/ctypes/init.py:356: in init
self._handle = _dlopen(self._name, mode)
E OSError: /home/ubuntu/anaconda3/lib/python3.7/site-packages/tvm-0.6.dev0-py3.7-linux-x86_64.egg/tvm/libtvm.so: undefined symbol: _ZN4llvm14TargetRegistry12lookupTargetERKSsRSs
================================================================== warnings summary ===================================================================
/home/ubuntu/anaconda3/lib/python3.7/site-packages/pywt/_utils.py:6
/home/ubuntu/anaconda3/lib/python3.7/site-packages/pywt/_utils.py:6: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
from collections import Iterable

-- Docs: https://docs.pytest.org/en/latest/warnings.html
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Interrupted: 3 errors during collection !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
========================================================= 1 warnings, 3 error in 0.90 seconds =========================================================

This seems kind of similar to this: apache/tvm#521, so I think this has something to do with llvm, but I'm not sure how to fix it. I tried setting the LLVM-CONFIG environmental variable to llvm-config-8.0 and I've checked my gcc version (it's 7.4.0), but that doesn't seem to help. I'm running llvm-8.0 and I've created the symbolic link from llvm-config-8. to llvm-config.

Do you have any ideas what else I could try? Sorry about all the questions.

[14jyuan]

It worked once I built PyTorch from source! I think the nightly build of PyTorch was causing the errors.

[lixiuhong]

It worked once I built PyTorch from source! I think the nightly build of PyTorch was causing the errors.

Which version it is in your case? I have also built from source, but it failed.