Segmentation fault when use NeighborSampler.

[gebrahimi91]

🐛 Bug

I keep getting Segmentation fault when I use NeighborSampler.

Thread 1 "python" received signal SIGSEGV, Segmentation fault.
0x00007ffec1257e5d in sample_adj_cpu(at::Tensor, at::Tensor, at::Tensor, long, bool) () from /opt/conda/lib/python3.8/site-packages/torch_sparse/_sample_cuda.so

To Reproduce

import matplotlib.pyplot as plt
from torch_geometric.data import NeighborSampler
import torch
import pickle
class Trainer:
def __init__(self, batch_size, model, gpu_id, lr, lam = 0.1, num_neighbors = [-1, -1]):
    file_to_read = open("samples_tr_small.pkl", "rb")

    samples_tr = pickle.load(file_to_read)
    self.batch_size = batch_size
    self.model = model
    self.lr = lr
    self.num_neighbors = num_neighbors
    if isinstance(gpu_id, list):
        pass
    else:
        gpu_id = [gpu_id, gpu_id]
    self.gpu_id = gpu_id
    if self.gpu_id[0]>=0:
        self.device = torch.device('cuda:0')
    else:
        self.device = torch.device('cpu:0')



    self.train_loaders = []


    row,    col, value = samples_tr['adj_t'].coo()
    print(row.max())
    print(col.max())
    for i in range(1):

        self.train_loader = NeighborSampler(samples_tr['adj_t'], 
                               sizes=self.num_neighbors, batch_size=self.batch_size, 
                               shuffle=True, num_workers=0,
                               #num_nodes = 200000,
                               return_e_id = False)
        self.train_loaders.append(self.train_loader)
            

        
    self.x_train = samples_tr['all_x']
    self.mode_train = samples_tr['modality']
    self.labels_train = samples_tr['node_labels']
    self.group_train = samples_tr['group']
    self.num_samples_train = samples_tr['num_samples']

    
    
def train(self, epochs):
    self.epoch_losses = []
    self.epoch_accuracies =[]
    self.val_losses = []
    self.val_accuracies = []

    
    for epoch in range(epochs):
        #self.model.train()
        #for train_loader in self.train_loaders: ######
        cnt = 0
        for train_loader in self.train_loaders:

            for batch_size, n_id, adjs in train_loader:
                print(cnt)
                cnt+=1
                continue
    return 0
gpu_id = [0, 1]
if not torch.cuda.is_available():
gpu_id = -1
lr = 0.0001
lam = 0.1
batch_size = 2
num_neighbors = [3,3]
num_hops = len(num_neighbors)
model = None
t = Trainer(batch_size = batch_size, model =model , gpu_id=gpu_id, num_neighbors = num_neighbors, lam = lam, lr = lr)
a = t.train(1)

Expected behavior

No seg fault.

Environment

• PyG version (torch_geometric.__version__): 2.0.3
• PyTorch version: (torch.__version__): 1.8.1
• OS (e.g., Linux): Linux
• Python version (e.g., 3.9): 3.8.8
• CUDA/cuDNN version: 11.5
• How you installed PyTorch and PyG (conda, pip, source): pip
• Any other relevant information (e.g., version of torch-scatter):

Additional context

The pickle file can be downloaded here

[gebrahimi91]

gbd bt output:

#0 0x00007ffec1257e5d in sample_adj_cpu(at::Tensor, at::Tensor, at::Tensor, long, bool) () from /opt/conda/lib/python3.8/site-packages/torch_sparse/_sample_cuda.so #1 0x00007ffec1250b69 in sample_adj(at::Tensor, at::Tensor, at::Tensor, long, bool) () from /opt/conda/lib/python3.8/site-packages/torch_sparse/_sample_cuda.so #2 0x00007ffec1256377 in std::decay<c10::guts::infer_function_traits<c10::impl::detail::WrapFunctionIntoRuntimeFunctor_<std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor> (*)(at::Tensor, at::Tensor, at::Tensor, long, bool), std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor>, c10::guts::typelist::typelist<at::Tensor, at::Tensor, at::Tensor, long, bool> > >::type::return_type>::type c10::impl::call_functor_with_args_from_stack_<c10::impl::detail::WrapFunctionIntoRuntimeFunctor_<std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor> (*)(at::Tensor, at::Tensor, at::Tensor, long, bool), std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor>, c10::guts::typelist::typelist<at::Tensor, at::Tensor, at::Tensor, long, bool> >, true, 0ul, 1ul, 2ul, 3ul, 4ul>(c10::impl::detail::WrapFunctionIntoRuntimeFunctor_<std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor> (*)(at::Tensor, at::Tensor, at::Tensor, long, bool), std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor>, c10::guts::typelist::typelist<at::Tensor, at::Tensor, at::Tensor, long, bool> >*, std::vector<c10::IValue, std::allocator<c10::IValue> >*, std::integer_sequence<unsigned long, 0ul, 1ul, 2ul, 3ul, 4ul>) () from /opt/conda/lib/python3.8/site-packages/torch_sparse/_sample_cuda.so #3 0x00007ffec125673d in c10::impl::make_boxed_from_unboxed_functor<c10::impl::detail::WrapFunctionIntoRuntimeFunctor_<std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor> (*)(at::Tensor, at::Tensor, at::Tensor, long, bool), std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor>, c10::guts::typelist::typelist<at::Tensor, at::Tensor, at::Tensor, long, bool> >, true>::call(c10::OperatorKernel*, c10::OperatorHandle const&, std::vector<c10::IValue, std::allocator<c10::IValue> >*) () from /opt/conda/lib/python3.8/site-packages/torch_sparse/_sample_cuda.so #4 0x00007fff66fd74b5 in torch::jit::(anonymous namespace)::createOperatorFromC10_withTracingHandledHere(c10::OperatorHandle const&)::{lambda(std::vector<c10::IValue, std::allocator<c10::IValue> >*)#1}::operator()(std::vector<c10::IValue, std::allocator<c10::IValue> >*) const () from /opt/conda/lib/python3.8/site-packages/torch/lib/libtorch_cpu.so #5 0x00007fffba6d6e82 in torch::jit::invokeOperatorFromPython(std::vector<std::shared_ptr<torch::jit::Operator>, std::allocator<std::shared_ptr<torch::jit::Operator> > > const&, pybind11::args, pybind11::kwargs const&) () from /opt/conda/lib/python3.8/site-packages/torch/lib/libtorch_python.so #6 0x00007fffba6afdd2 in torch::jit::initJITBindings(_object*)::{lambda(std::string const&)#115}::operator()(std::string const&) const::{lambda(pybind11::args, {lambda(std::string const&)#115}::kwargs)#1}::operator()(pybind11, pybind11::args) const () from /opt/conda/lib/python3.8/site-packages/torch/lib/libtorch_python.so #7 0x00007fffba6b04cf in void pybind11::cpp_function::initialize<torch::jit::initJITBindings(_object*)::{lambda(std::string const&)#115}::operator()(std::string const&) const::{lambda(pybind11::args, pybind11::kwargs)#1}, pybind11::object, {lambda(std::string const&)#115}, pybind11::args, pybind11::name, pybind11::doc>(torch::jit::initJITBindings(_object*)::{lambda(std::string const&)#115}::operator()(std::string const&) const::{lambda(pybind11::args, pybind11::kwargs)#1}&&, pybind11::object (*)({lambda(std::string const&)#115}, pybind11::args), pybind11::name const&, pybind11::doc const&)::{lambda(pybind11::detail::function_call&)#3}::_FUN(pybind11::detail) () from /opt/conda/lib/python3.8/site-packages/torch/lib/libtorch_python.so #8 0x00007fffba2ec35e in pybind11::cpp_function::dispatcher(_object*, _object*, _object*) () from /opt/conda/lib/python3.8/site-packages/torch/lib/libtorch_python.so #9 0x00005555556a82d8 in cfunction_call_varargs (kwargs=<optimized out>, args=<optimized out>, func=0x7ffff6db7450) at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:743 #10 PyCFunction_Call () at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:773 #11 0x0000555555697edc in _PyObject_MakeTpCall.localalias.6 () at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:159 #12 0x0000555555723879 in _PyObject_Vectorcall (kwnames=0x0, nargsf=<optimized out>, args=0x5555c885d428, callable=0x7ffff6db7450) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:125 #13 call_function (kwnames=0x0, oparg=<optimized out>, pp_stack=<synthetic pointer>, tstate=0x5555558f3510) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4963 #14 _PyEval_EvalFrameDefault () at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:3469 #15 0x00005555556ed760 in PyEval_EvalFrameEx (throwflag=0, f=0x5555c885d260) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:741 #16 _PyEval_EvalCodeWithName.localalias.4 () at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4298 #17 0x00005555556ee970 in _PyFunction_Vectorcall (kwnames=<optimized out>, nargsf=<optimized out>, stack=0x5555c7fe7948, func=0x7ffec059fdc0) at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:436 #18 _PyObject_Vectorcall (kwnames=<optimized out>, nargsf=<optimized out>, args=0x5555c7fe7948, callable=0x7ffec059fdc0) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:127 #19 method_vectorcall () at /tmp/build/80754af9/python_1614202678154/work/Objects/classobject.c:60 #20 0x0000555555657562 in _PyObject_Vectorcall (kwnames=0x7ffec0538160, nargsf=<optimized out>, args=<optimized out>, callable=0x7ffeac0488c0) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:127 #21 call_function (kwnames=0x7ffec0538160, oparg=<optimized out>, pp_stack=<synthetic pointer>, tstate=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4963 #22 _PyEval_EvalFrameDefault (f=<optimized out>, throwflag=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:3515 #23 0x00005555556ee85b in function_code_fastcall (globals=<optimized out>, nargs=2, args=<optimized out>, co=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:284 #24 _PyFunction_Vectorcall (kwnames=<optimized out>, nargsf=<optimized out>, stack=0x7ffe0a136fd0, func=0x7ffec053a8b0) at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:411 #25 _PyObject_Vectorcall (kwnames=<optimized out>, nargsf=<optimized out>, args=0x7ffe0a136fd0, callable=0x7ffec053a8b0) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:127 #26 method_vectorcall () at /tmp/build/80754af9/python_1614202678154/work/Objects/classobject.c:60 #27 0x00005555556579bd in _PyObject_Vectorcall (kwnames=0x0, nargsf=<optimized out>, args=0x7ffe0a136fd8, callable=0x7ffe329bf0c0) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:127 #28 call_function (kwnames=0x0, oparg=<optimized out>, pp_stack=<synthetic pointer>, tstate=0x5555558f3510) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4963 #29 _PyEval_EvalFrameDefault (f=<optimized out>, throwflag=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:3469 #30 0x00005555556edd11 in PyEval_EvalFrameEx (throwflag=0, f=0x7ffe0a136e40) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:741 ---Type <return> to continue, or q <return> to quit--- #31 _PyEval_EvalCodeWithName.localalias.4 () at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4298 #32 0x00005555556ee593 in _PyFunction_Vectorcall.localalias.352 () at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:436 #33 0x000055555565799c in _PyObject_Vectorcall (kwnames=0x0, nargsf=<optimized out>, args=0x7ffdc3e56528, callable=0x7ffec9894550) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:127 #34 call_function (kwnames=0x0, oparg=<optimized out>, pp_stack=<synthetic pointer>, tstate=0x5555558f3510) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4963 #35 _PyEval_EvalFrameDefault (f=<optimized out>, throwflag=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:3486 #36 0x00005555556ee4bb in function_code_fastcall (globals=<optimized out>, nargs=1, args=<optimized out>, co=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:284 #37 _PyFunction_Vectorcall.localalias.352 () at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:411 #38 0x000055555565799c in _PyObject_Vectorcall (kwnames=0x0, nargsf=<optimized out>, args=0x7ffe0a609d70, callable=0x7ffec98990d0) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:127 #39 call_function (kwnames=0x0, oparg=<optimized out>, pp_stack=<synthetic pointer>, tstate=0x5555558f3510) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4963 #40 _PyEval_EvalFrameDefault (f=<optimized out>, throwflag=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:3486 #41 0x00005555556ee4bb in function_code_fastcall (globals=<optimized out>, nargs=1, args=<optimized out>, co=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:284 #42 _PyFunction_Vectorcall.localalias.352 () at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:411 #43 0x000055555569cc4b in _PyObject_Vectorcall (kwnames=0x0, nargsf=1, args=0x7fffffffc930, callable=0x7ffec9894e50) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:127 #44 _PyObject_FastCall () at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:147 #45 _PyObject_FastCall_Prepend () at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:850 #46 0x00005555556ac30e in call_unbound (nargs=0, args=0x0, self=0x7ffff7e78a30, func=0x7ffec9894e50, unbound=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Objects/typeobject.c:1453 #47 call_method (nargs=0, args=0x0, name=0x5555558b1c20 <PyId___next__.15157>, obj=0x7ffff7e78a30) at /tmp/build/80754af9/python_1614202678154/work/Objects/typeobject.c:1485 #48 slot_tp_iternext () at /tmp/build/80754af9/python_1614202678154/work/Objects/typeobject.c:6732 #49 0x000055555571fd89 in _PyEval_EvalFrameDefault () at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:3202 #50 0x00005555556ee4bb in function_code_fastcall (globals=<optimized out>, nargs=2, args=<optimized out>, co=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:284 #51 _PyFunction_Vectorcall.localalias.352 () at /tmp/build/80754af9/python_1614202678154/work/Objects/call.c:411 #52 0x000055555565799c in _PyObject_Vectorcall (kwnames=0x0, nargsf=<optimized out>, args=0x7ffff7ebd5b0, callable=0x7fffe5dcc310) at /tmp/build/80754af9/python_1614202678154/work/Include/cpython/abstract.h:127 #53 call_function (kwnames=0x0, oparg=<optimized out>, pp_stack=<synthetic pointer>, tstate=0x5555558f3510) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4963 #54 _PyEval_EvalFrameDefault (f=<optimized out>, throwflag=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:3486 #55 0x00005555556ed760 in PyEval_EvalFrameEx (throwflag=0, f=0x7ffff7ebd440) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:741 #56 _PyEval_EvalCodeWithName.localalias.4 () at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4298 #57 0x00005555557824e3 in PyEval_EvalCodeEx () at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:4327 #58 PyEval_EvalCode (co=<optimized out>, globals=<optimized out>, locals=<optimized out>) at /tmp/build/80754af9/python_1614202678154/work/Python/ceval.c:718 #59 0x0000555555782584 in run_eval_code_obj () at /tmp/build/80754af9/python_1614202678154/work/Python/pythonrun.c:1165 #60 0x00005555557a87c4 in run_mod () at /tmp/build/80754af9/python_1614202678154/work/Python/pythonrun.c:1187 #61 0x0000555555669620 in pyrun_file (fp=0x55555595b350, filename=0x7ffff7e14a30, start=<optimized out>, globals=0x7ffff7f203c0, locals=0x7ffff7f203c0, closeit=1, flags=0x7fffffffcea8) at /tmp/build/80754af9/python_1614202678154/work/Python/pythonrun.c:1084 #62 0x000055555566c362 in pyrun_simple_file (flags=0x7fffffffcea8, closeit=1, filename=0x7ffff7e14a30, fp=0x55555595b350) at /tmp/build/80754af9/python_1614202678154/work/Python/pythonrun.c:439 #63 PyRun_SimpleFileExFlags (fp=0x55555595b350, filename=<optimized out>, closeit=1, flags=0x7fffffffcea8) at /tmp/build/80754af9/python_1614202678154/work/Python/pythonrun.c:472 #64 0x000055555566ce80 in pymain_run_file (cf=0x7fffffffcea8, config=0x5555558f29a0) at /tmp/build/80754af9/python_1614202678154/work/Modules/main.c:391 #65 pymain_run_python (exitcode=0x7fffffffcea0) at /tmp/build/80754af9/python_1614202678154/work/Modules/main.c:616 #66 Py_RunMain () at /tmp/build/80754af9/python_1614202678154/work/Modules/main.c:695 #67 0x00005555557ab979 in Py_BytesMain () at /tmp/build/80754af9/python_1614202678154/work/Modules/main.c:1141 #68 0x00007ffff77e4bf7 in __libc_start_main (main=0x55555566d6f0 <main>, argc=2, argv=0x7fffffffd098, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffffffd088) at ../csu/libc-start.c:310 #69 0x000055555573b185 in _start () at ../sysdeps/x86_64/elf/start.S:103

[rusty1s]

Seems to be related to an incorrect definition of sparse_sizes.