Backward pass fails for subgraphs containing reshape or transpose op

Description

When a computation graph is partitioned, ops are grouped into subgraphs based on the subgraph property. For CachedOp subgraphs containing reshape and/or transpose op, the backward pass fails.

Error Message

Traceback (most recent call last):
  File "test.py", line 24, in <module>
    result.backward()
  File "/home/ubuntu/.local/lib/python3.6/site-packages/mxnet/ndarray/ndarray.py", line 2877, in backward
    ctypes.c_void_p(0)))
  File "/home/ubuntu/.local/lib/python3.6/site-packages/mxnet/base.py", line 246, in check_call
    raise get_last_ffi_error()
mxnet.base.MXNetError: MXNetError: Invalid argument

To Reproduce

Here's a simple script to reproduce the error with Reshape op:

import mxnet as mx
from mxnet.gluon import HybridBlock
from mxnet.base import check_call, _LIB, c_str, mx_uint, c_str_array

class _TestBlock(HybridBlock):
    def __init__(self):
        super(_TestBlock, self).__init__()

    def hybrid_forward(self, F, data):
        return F.reshape(data, (-1,))

if __name__ == '__main__':
    subgraph_backend = 'default'
    op_names = ['Reshape']
    check_call(_LIB.MXSetSubgraphPropertyOpNamesV2(c_str(subgraph_backend), mx_uint(len(op_names)),
                                                   c_str_array(op_names)))
    block = _TestBlock()
    block.hybridize(backend=subgraph_backend)
    data = mx.nd.array([[4, 5, 6], [7, 8, 9]])
    data.attach_grad()
    with mx.autograd.record():
        result = block(data)
        # block.export('partitioned')
    result.backward()

It produces the following graph:

{
  "nodes": [
    {
      "op": "null", 
      "name": "data", 
      "attrs": {"__profiler_scope__": "<unk>:"}, 
      "inputs": []
    }, 
    {
      "op": "_CachedOp", 
      "name": "_CachedOp0", 
      "inputs": [[0, 0, 0]], 
      "subgraphs": [
        {
          "nodes": [
            {
              "op": "null", 
              "name": "data0", 
              "inputs": []
            }, 
            {
              "op": "Reshape", 
              "name": "_testblock0_reshape0", 
              "attrs": {
                "__profiler_scope__": "_testblock0:", 
                "shape": "(-1,)"
              }, 
              "inputs": [[0, 0, 0]]
            }
          ], 
          "arg_nodes": [0], 
          "node_row_ptr": [0, 1, 2], 
          "heads": [[1, 0, 0]]
        }
      ]
    }
  ], 
  "arg_nodes": [0], 
  "node_row_ptr": [0, 1, 2], 
  "heads": [[1, 0, 0]], 
  "attrs": {"mxnet_version": ["int", 20000]}

Steps to reproduce

(Paste the commands you ran that produced the error.)

Intall MXNet pip3 install -U --user https://repo.mxnet.io/dist/python/cpu/mxnet-2.0.0b20200710-py2.py3-none-manylinux2014_x86_64.whl
Run python script

Environment

We recommend using our script for collecting the diagnositc information. Run the following command and paste the outputs below:

curl --retry 10 -s https://raw.githubusercontent.com/dmlc/gluon-nlp/master/tools/diagnose.py | python

('Version      :', '2.7.17')
('Compiler     :', 'GCC 7.5.0')
('Build        :', ('default', 'Jul 20 2020 15:37:01'))
('Arch         :', ('64bit', ''))
------------Pip Info-----------
No corresponding pip install for current python.
----------MXNet Info-----------
No MXNet installed.
----------System Info----------
('Platform     :', 'Linux-5.3.0-1030-aws-x86_64-with-Ubuntu-18.04-bionic')
('system       :', 'Linux')
('node         :', 'ip-172-31-11-130')
('release      :', '5.3.0-1030-aws')
('version      :', '#32~18.04.1-Ubuntu SMP Tue Jun 30 23:04:16 UTC 2020')
----------Hardware Info----------
('machine      :', 'x86_64')
('processor    :', 'x86_64')
Architecture:        x86_64
CPU op-mode(s):      32-bit, 64-bit
Byte Order:          Little Endian
CPU(s):              96
On-line CPU(s) list: 0-95
Thread(s) per core:  2
Core(s) per socket:  24
Socket(s):           2
NUMA node(s):        2
Vendor ID:           GenuineIntel
CPU family:          6
Model:               85
Model name:          Intel(R) Xeon(R) Platinum 8275CL CPU @ 3.00GHz
Stepping:            7
CPU MHz:             1411.873
BogoMIPS:            5999.99
Hypervisor vendor:   KVM
Virtualization type: full
L1d cache:           32K
L1i cache:           32K
L2 cache:            1024K
L3 cache:            36608K
NUMA node0 CPU(s):   0-23,48-71
NUMA node1 CPU(s):   24-47,72-95
Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single pti fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid mpx avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves ida arat pku ospke avx512_vnni
----------Network Test----------
Setting timeout: 10
Timing for PYPI: https://pypi.python.org/pypi/pip, DNS: 0.0019 sec, LOAD: 0.5515 sec.
Timing for D2L: http://d2l.ai, DNS: 0.0097 sec, LOAD: 0.0470 sec.
Timing for FashionMNIST: https://repo.mxnet.io/gluon/dataset/fashion-mnist/train-labels-idx1-ubyte.gz, DNS: 0.0296 sec, LOAD: 0.1153 sec.
Error open Conda: https://repo.continuum.io/pkgs/free/, HTTP Error 403: Forbidden, DNS finished in 0.018091917038 sec.
Timing for MXNet: https://github.com/apache/incubator-mxnet, DNS: 0.0013 sec, LOAD: 0.6243 sec.
Timing for GluonNLP: http://gluon-nlp.mxnet.io, DNS: 0.0322 sec, LOAD: 0.4460 sec.
Timing for D2L (zh-cn): http://zh.d2l.ai, DNS: 0.0140 sec, LOAD: 0.1110 sec.
Timing for GluonNLP GitHub: https://github.com/dmlc/gluon-nlp, DNS: 0.0003 sec, LOAD: 0.4265 sec.

Using the script provided above, I tested the following operators: ['Reshape', 'transpose', 'elemwise_add', 'elemwise_sub', 'sum', '_lesser_scalar', '_greater_scalar', '_minus_scalar', '_plus_scalar'] These ops all have the same Invalid argument error during backward prop.

@samskalicky @szha @eric-haibin-lin do you have an idea what could be causing this?

Here is the stack trace for this error. I used the following patch to obtain it:

diff --git a/src/c_api/c_api_ndarray.cc b/src/c_api/c_api_ndarray.cc
index ebb3134ae..e0ba2791e 100644
--- a/src/c_api/c_api_ndarray.cc
+++ b/src/c_api/c_api_ndarray.cc
@@ -394,7 +394,6 @@ int MXAutogradBackwardEx(uint32_t num_output,
                          NDArrayHandle **grad_handles,
                          int **grad_stypes) {
   MXAPIThreadLocalEntry<> *ret = MXAPIThreadLocalStore<>::Get();
-  API_BEGIN();

   std::vector<NDArray*> outputs, ograds, variables;
   outputs.reserve(num_output);
@@ -430,7 +429,7 @@ int MXAutogradBackwardEx(uint32_t num_output,
     *grad_handles = dmlc::BeginPtr(ret->ret_handles);
     *grad_stypes = dmlc::BeginPtr(ret->out_types);
   }
-  API_END();
+  return 0;
 }

 int MXAutogradGetSymbol(NDArrayHandle handle, SymbolHandle *out) {

and build via cmake -GNinja -DCMAKE_BUILD_TYPE=Debug -DLOG_FATAL_THROW=0 -DUSE_CUDA=0 ..; ninja

Back Trace: Note: static_alloc is set to true for internal CachedOps in the default subgraph property, hence it goes to StaticBackward

Thread 1 "python3" received signal SIGABRT, Aborted.
__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  0x00007ffff7a22f47 in __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:51
#1  0x00007ffff7a248b1 in __GI_abort () at abort.c:79
#2  0x00007ffff7a1442a in __assert_fail_base (fmt=0x7ffff7b9ba38 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", assertion=assertion@entry=0x7ffff77d9750 "INTERNAL_SYSCALL_ERRNO (e, __err) != ESRCH || !robust", file=file@entry=0x7ffff77d9695 "../nptl/pthread_mutex_lock.c", line=line@entry=425, function=function@entry=0x7ffff77d9800 <__PRETTY_FUNCTION__.8909> "__pthread_mutex_lock_full") at assert.c:92
#3  0x00007ffff7a144a2 in __GI___assert_fail (assertion=assertion@entry=0x7ffff77d9750 "INTERNAL_SYSCALL_ERRNO (e, __err) != ESRCH || !robust", file=file@entry=0x7ffff77d9695 "../nptl/pthread_mutex_lock.c", line=line@entry=425, function=function@entry=0x7ffff77d9800 <__PRETTY_FUNCTION__.8909> "__pthread_mutex_lock_full") at assert.c:101
#4  0x00007ffff77cef3c in __pthread_mutex_lock_full (mutex=0x1eb5690) at ../nptl/pthread_mutex_lock.c:425
#5  0x00007fff42e242bf in __gthread_mutex_lock(__gthread_mutex_t*) (__mutex=0x1eb5690) at /usr/include/x86_64-linux-gnu/c++/7/bits/gthr-default.h:748
#6  0x00007fff42e24552 in std::mutex::lock() (this=0x1eb5690) at /usr/include/c++/7/bits/std_mutex.h:103
#7  0x00007fff42e9219a in std::lock_guard<std::mutex>::lock_guard(std::mutex&) (this=0x7fffffff5970, __m=...) at /usr/include/c++/7/bits/std_mutex.h:162
#8  0x00007fff431b49bd in mxnet::CachedOp::StaticBackward(bool, mxnet::OpStatePtr const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, std::vector<mxnet::OpReqType, std::allocator<mxnet::OpReqType> > const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&) (this=0x1eeca90, retain_graph=false, state_ptr=..., inputs=std::vector of length 3, capacity 3 = {...}, reqs=std::vector of length 1, capacity 1 = {...}, outputs=std::vector of length 1, capacity 1 = {...}) at ../src/imperative/cached_op.cc:935
#9  0x00007fff431b59c6 in mxnet::CachedOp::Backward(bool, mxnet::OpStatePtr const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, std::vector<mxnet::OpReqType, std::allocator<mxnet::OpReqType> > const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&) (this=0x1eeca90, retain_graph=false, state=..., inputs=std::vector of length 3, capacity 3 = {...}, reqs=std::vector of length 1, capacity 1 = {...}, outputs=std::vector of length 1, capacity 1 = {...}) at ../src/imperative/cached_op.cc:1046
#10 0x00007fff43219394 in (anonymous namespace)::InvokeOperator(const nnvm::IndexedGraph &, int, bool, const std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > &, mxnet::Context, std::vector<mxnet::OpStatePtr, std::allocator<mxnet::OpStatePtr> > *, const std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > &, const std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > &, std::vector<mxnet::OpReqType, std::allocator<mxnet::OpReqType> > *, std::vector<unsigned int, std::allocator<unsigned int> > *, std::function<void(const mxnet::OpStatePtr&)>) (idx=..., node_idx=3, retain_graph=false, arrays=std::vector of length 4, capacity 4 = {...}, ctx=..., p_states=0x7fffffff63a0, ndinputs=std::vector of length 3, capacity 3 = {...}, ndoutputs=std::vector of length 1, capacity 1 = {...}, p_req=0x7fffffff5f90, p_ref_count=0x7fffffff6380, invoke=...) at ../src/imperative/imperative_utils.cc:91
#11 0x00007fff43219faf in mxnet::imperative::RunGraph(bool, nnvm::IndexedGraph const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, unsigned long, unsigned long, std::vector<mxnet::OpReqType, std::allocator<mxnet::OpReqType> >&&, std::vector<unsigned int, std::allocator<unsigned int> >&&, std::vector<mxnet::OpStatePtr, std::allocator<mxnet::OpStatePtr> >*, std::vector<mxnet::DispatchMode, std::allocator<mxnet::DispatchMode> > const&, bool, std::vector<mxnet::TShape, std::allocator<mxnet::TShape> >*, std::function<void (char const*, char const*, void*)> const&, bool) (retain_graph=false, idx=..., arrays=std::vector of length 4, capacity 4 = {...}, node_start=2, node_end=4, array_reqs=..., ref_count=..., p_states=0x7fffffff63a0, dispatch_modes=std::vector of length 4, capacity 4 = {...}, recording=false, shapes=0x0, callback=..., monitor_all=false)
    at ../src/imperative/imperative_utils.cc:165
#12 0x00007fff431fcaa9 in mxnet::Imperative::Backward(std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, bool, bool, bool) (this=0x7fff51515ac0 <mxnet::Imperative::Get()::inst>, outputs=std::vector of length 1, capacity 1 = {...}, ograds=std::vector of length 1, capacity 1 = {...}, variables=std::vector of length 0, capacity 0, is_train=true, retain_graph=false, create_graph=false) at ../src/imperative/imperative.cc:616
#13 0x00007fff43062242 in MXAutogradBackwardEx(uint32_t, NDArrayHandle*, NDArrayHandle*, uint32_t, NDArrayHandle*, int, int, int, NDArrayHandle**, int**) (num_output=1, output_handles=0x7fff93f89de0, ograd_handles=0x7fff93f89c48, num_variables=0, var_handles=0x0, retain_graph=0, create_graph=0, is_train=1, grad_handles=0x0, grad_stypes=0x0) at ../src/c_api/c_api_ndarray.cc:419

And this is the stack trace when static_alloc for internal CachedOp is set to false in subgraph property. I used the following additional patch to obtain it:

diff --git a/src/operator/subgraph/default_subgraph_property.cc b/src/operator/subgraph/default_subgraph_property.cc
index ff51b6397..4228c6521 100644
--- a/src/operator/subgraph/default_subgraph_property.cc
+++ b/src/operator/subgraph/default_subgraph_property.cc
@@ -64,7 +64,7 @@ class DefaultSubgraphProperty: public SubgraphProperty {
     n->attrs.name = "_CachedOp" + std::to_string(subgraph_id);
     n->attrs.subgraphs.push_back(std::make_shared<nnvm::Symbol>(sym));

-    std::vector<std::pair<std::string, std::string> > flags{{"static_alloc", "true"}};
+    std::vector<std::pair<std::string, std::string> > flags{};
     n->attrs.parsed = std::make_shared<CachedOp>(sym, flags);

     return n;

Back Trace: Note: This time it fails in DynamicForward

Thread 1 "python3" received signal SIGABRT, Aborted.
__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  0x00007ffff7a22f47 in __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:51
#1  0x00007ffff7a248b1 in __GI_abort () at abort.c:79
#2  0x00007fffe93a6957 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#3  0x00007fffe93acae6 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#4  0x00007fffe93acb21 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#5  0x00007fffe93acd54 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#6  0x00007fffe93d5012 in  () at /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#7  0x00007fff42e355ac in __gnu_cxx::new_allocator<nnvm::NodeEntry>::allocate(unsigned long, void const*) (this=0x23893a0, __n=12297829382473034410) at /usr/include/c++/7/ext/new_allocator.h:102
#8  0x00007fff42e33c4c in std::allocator_traits<std::allocator<nnvm::NodeEntry> >::allocate(std::allocator<nnvm::NodeEntry>&, unsigned long) (__a=..., __n=12297829382473034410) at /usr/include/c++/7/bits/alloc_traits.h:436
#9  0x00007fff42e3162e in std::_Vector_base<nnvm::NodeEntry, std::allocator<nnvm::NodeEntry> >::_M_allocate(unsigned long) (this=0x23893a0, __n=12297829382473034410) at /usr/include/c++/7/bits/stl_vector.h:172
#10 0x00007fff42e2ec8e in std::vector<nnvm::NodeEntry, std::allocator<nnvm::NodeEntry> >::_M_allocate_and_copy<__gnu_cxx::__normal_iterator<nnvm::NodeEntry const*, std::vector<nnvm::NodeEntry, std::allocator<nnvm::NodeEntry> > > >(unsigned long, __gnu_cxx::__normal_iterator<nnvm::NodeEntry const*, std::vector<nnvm::NodeEntry, std::allocator<nnvm::NodeEntry> > >, __gnu_cxx::__normal_iterator<nnvm::NodeEntry const*, std::vector<nnvm::NodeEntry, std::allocator<nnvm::NodeEntry> > >) (this=0x23893a0, __n=12297829382473034410, __first={node = <error reading variable: Cannot access memory at address 0xf0000000a>, index = 15, version = 0}, __last=
  {node = <error reading variable: Cannot access memory at address 0x100000010>, index = 1362897120, version = 32767}) at /usr/include/c++/7/bits/stl_vector.h:1260
#11 0x00007fff42e2c50e in std::vector<nnvm::NodeEntry, std::allocator<nnvm::NodeEntry> >::operator=(std::vector<nnvm::NodeEntry, std::allocator<nnvm::NodeEntry> > const&) (this=0x23893a0, __x=std::vector of length 0, capacity -41698 = {...}) at /usr/include/c++/7/bits/vector.tcc:206
#12 0x00007fff431c3b79 in nnvm::Graph::operator=(nnvm::Graph const&) (this=0x23893a0) at ../include/nnvm/graph.h:46
#13 0x00007fff431b3c7e in mxnet::CachedOp::DynamicBackward(bool, mxnet::OpStatePtr const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, std::vector<mxnet::OpReqType, std::allocator<mxnet::OpReqType> > const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&) (this=0x1eeca90, retain_graph=false, op_state=..., inputs=std::vector of length 3, capacity 3 = {...}, reqs=std::vector of length 1, capacity 1 = {...}, outputs=std::vector of length 1, capacity 1 = {...}) at ../src/imperative/cached_op.cc:853
#14 0x00007fff431b5a00 in mxnet::CachedOp::Backward(bool, mxnet::OpStatePtr const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, std::vector<mxnet::OpReqType, std::allocator<mxnet::OpReqType> > const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&) (this=0x1eeca90, retain_graph=false, state=..., inputs=std::vector of length 3, capacity 3 = {...}, reqs=std::vector of length 1, capacity 1 = {...}, outputs=std::vector of length 1, capacity 1 = {...}) at ../src/imperative/cached_op.cc:1048
#15 0x00007fff43219394 in (anonymous namespace)::InvokeOperator(const nnvm::IndexedGraph &, int, bool, const std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > &, mxnet::Context, std::vector<mxnet::OpStatePtr, std::allocator<mxnet::OpStatePtr> > *, const std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > &, const std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > &, std::vector<mxnet::OpReqType, std::allocator<mxnet::OpReqType> > *, std::vector<unsigned int, std::allocator<unsigned int> > *, std::function<void(const mxnet::OpStatePtr&)>) (idx=..., node_idx=3, retain_graph=false, arrays=std::vector of length 4, capacity 4 = {...}, ctx=..., p_states=0x7fffffff63a0, ndinputs=std::vector of length 3, capacity 3 = {...}, ndoutputs=std::vector of length 1, capacity 1 = {...}, p_req=0x7fffffff5f90, p_ref_count=0x7fffffff6380, invoke=...) at ../src/imperative/imperative_utils.cc:91
#16 0x00007fff43219faf in mxnet::imperative::RunGraph(bool, nnvm::IndexedGraph const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, unsigned long, unsigned long, std::vector<mxnet::OpReqType, std::allocator<mxnet::OpReqType> >&&, std::vector<unsigned int, std::allocator<unsigned int> >&&, std::vector<mxnet::OpStatePtr, std::allocator<mxnet::OpStatePtr> >*, std::vector<mxnet::DispatchMode, std::allocator<mxnet::DispatchMode> > const&, bool, std::vector<mxnet::TShape, std::allocator<mxnet::TShape> >*, std::function<void (char const*, char const*, void*)> const&, bool) (retain_graph=false, idx=..., arrays=std::vector of length 4, capacity 4 = {...}, node_start=2, node_end=4, array_reqs=..., ref_count=..., p_states=0x7fffffff63a0, dispatch_modes=std::vector of length 4, capacity 4 = {...}, recording=false, shapes=0x0, callback=..., monitor_all=false)
    at ../src/imperative/imperative_utils.cc:165
#17 0x00007fff431fcaa9 in mxnet::Imperative::Backward(std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, std::vector<mxnet::NDArray*, std::allocator<mxnet::NDArray*> > const&, bool, bool, bool) (this=0x7fff51515ac0 <mxnet::Imperative::Get()::inst>, outputs=std::vector of length 1, capacity 1 = {...}, ograds=std::vector of length 1, capacity 1 = {...}, variables=std::vector of length 0, capacity 0, is_train=true, retain_graph=false, create_graph=false) at ../src/imperative/imperative.cc:616
#18 0x00007fff43062242 in MXAutogradBackwardEx(uint32_t, NDArrayHandle*, NDArrayHandle*, uint32_t, NDArrayHandle*, int, int, int, NDArrayHandle**, int**) (num_output=1, output_handles=0x7fff90f87de0, ograd_handles=0x7fff90f87c48, num_variables=0, var_handles=0x0, retain_graph=0, create_graph=0, is_train=1, grad_handles=0x0, grad_stypes=0x0) at ../src/c_api/c_api_ndarray.cc:419

apache / mxnet