tvm.bind not found

[liangfu]

As shown in TVM Release Announcement:

but, I found no tvm.bind function implemented in the tvm package.

[masahi]

That functionality exists in recent nnvm. See here

[liangfu]

Thanks @masahi, however, current implementation is only limited to cuda and llvm backend, and if target and device is changed to opencl, it raises AssertionError:

assert target.startswith("llvm")

at https://github.com/dmlc/nnvm/blob/master/python/nnvm/top/tensor.py#L15

I'm wondering how can I use opencl backend to predict the tabby cat image. Thanks in advance!

[masahi]

See the discussion here

[liangfu]

Thanks @masahi , since all changes you made are just add

target = "opencl"
ctx = tvm.cl(0)

in the following files:

python/nnvm/top/nn.py
python/nnvm/top/reduction.py
python/nnvm/top/tensor.py

but why not submit a PR to make opencl work in the master branch ? is there any concern or consequence ?

[liangfu]

It's okay to use either cuda or llvm to run with nnvm for the image classification demo of the tabby cat image.
However, after changing target to opencl and set ctx to tvm.cl(0), I fail to execute the m.run() as highlighted in the following

# set inputs
m.set_input('data', tvm.nd.array(x.astype(dtype)))
m.set_input(**params)
# execute
m.run()                    # <<-------------------- HERE
# get outputs
tvm_output = m.get_output(0, tvm.nd.empty((1000,), dtype))
top1 = np.argmax(tvm_output.asnumpy())

after tracing back the call stack, I got the following lines:

#0  tvm::runtime::WorkspacePool::FreeWorkspace (this=0x5555571844e8, ctx=..., ptr=0x555559862880)
    at /home/liangfu/workspace/tvm/src/runtime/workspace_pool.cc:140
#1  0x00007fff981ea5b6 in TVMBackendFreeWorkspace (device_type=4, device_id=0, ptr=0x555559862880)
    at /home/liangfu/workspace/tvm/src/runtime/c_runtime_api.cc:234
#2  0x00007fff8ed0e8dc in fuse_conv2d_broadcast_mul_broadcast_add_relu ()
#3  0x00007fff981e63e9 in tvm::codegen::LLVMModuleNode::GetFunction(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, std::shared_ptr<tvm::runtime::ModuleNode> const&)::{lambda(tvm::runtime::TVMArgs\
, tvm::runtime::TVMRetValue*)#2}::operator()(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const (
    __closure=<optimized out>, args=..., rv=<optimized out>)
    at /home/liangfu/workspace/tvm/src/codegen/llvm/llvm_module.cc:58
#4  0x00007fff981e6697 in std::_Function_handler<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::codegen::LLVMModuleNode::GetFunction(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&,\
 std::shared_ptr<tvm::runtime::ModuleNode> const&)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#2}>::_M_invoke(std::_Any_data const&, tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&) (__functor=...,
    __args#0=<optimized out>, __args#1=<optimized out>) at /usr/include/c++/5/functional:1871
#5  0x00007fff9826fbc7 in std::function<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)>::operator()(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const (__args#1=0x7fffffffd4d0, __args#0=..., this=<optimized out>)
    at /usr/include/c++/5/functional:2267
#6  tvm::runtime::PackedFunc::CallPacked (rv=0x7fffffffd4d0, args=..., this=<optimized out>)
    at /home/liangfu/workspace/tvm/include/tvm/runtime/packed_func.h:684
#7  tvm::runtime::GraphRuntime::<lambda()>::operator() (__closure=<optimized out>)
    at /home/liangfu/workspace/tvm/src/runtime/graph/graph_runtime.cc:536
#8  std::_Function_handler<void(), tvm::runtime::GraphRuntime::CreateTVMOp(const tvm::runtime::TVMOpParam&, const std::vector<DLTensor>&, size_t)::<lambda()> >::_M_invoke(const std::_Any_data &) (__functor=...)
    at /usr/include/c++/5/functional:1871
#9  0x00007fff9826e447 in std::function<void ()>::operator()() const (this=<optimized out>)
    at /usr/include/c++/5/functional:2267
#10 tvm::runtime::GraphRuntime::Run (this=0x555557f2d320)
    at /home/liangfu/workspace/tvm/src/runtime/graph/graph_runtime.cc:55
#11 tvm::runtime::GraphRuntime::<lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)>::operator() (
    __closure=<optimized out>, rv=<optimized out>, args=...)
    at /home/liangfu/workspace/tvm/src/runtime/graph/graph_runtime.cc:559
#12 std::_Function_handler<void(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::runtime::GraphRuntime::GetFunction(const string&, const std::shared_ptr<tvm::runtime::ModuleNode>&)::<lambda(tvm::runtime::TVMArgs, tvm::runtime::T\
VMRetValue*)> >::_M_invoke(const std::_Any_data &, <unknown type in /usr/local/lib/python2.7/dist-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/libtvm.so, CU 0x3bf7fe3, DIE 0x3c3fc8f>, <unknown type in /usr/local/lib/python2.7/dist-pack\
ages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/libtvm.so, CU 0x3bf7fe3, DIE 0x3c3fc94>) (__functor=...,
    __args#0=<optimized out>, __args#1=<optimized out>) at /usr/include/c++/5/functional:1871
#13 0x00007fff981ee758 in std::function<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)>::operator()(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const (__args#1=0x7fffffffd550, __args#0=..., this=0x7fffffffd628)
    at /usr/include/c++/5/functional:2267
#14 tvm::runtime::PackedFunc::CallPacked (rv=0x7fffffffd550, args=..., this=0x7fffffffd628)
    at /home/liangfu/workspace/tvm/include/tvm/runtime/packed_func.h:684
#15 TVMFuncCall (func=func@entry=0x55555852ba40, args=args@entry=0x7fffffffd640,
    arg_type_codes=arg_type_codes@entry=0x7fffffffd634, num_args=num_args@entry=0,
    ret_val=ret_val@entry=0x7fffffffd628, ret_type_code=ret_type_code@entry=0x7fffffffd61c)
    at /home/liangfu/workspace/tvm/src/runtime/c_runtime_api.cc:264
#16 0x00007fff96a1b28a in __pyx_f_4core_FuncCall3 (__pyx_v_nargs=0, __pyx_v_args=0x7ffff7f7f050,
    __pyx_v_chandle=0x55555852ba40) at tvm/_ffi/_cython/core.cpp:6083
#17 __pyx_f_4core_FuncCall (__pyx_v_args=0x7ffff7f7f050, __pyx_v_chandle=0x55555852ba40)
    at tvm/_ffi/_cython/core.cpp:6180
#18 __pyx_pf_4core_12FunctionBase_4__call__ (__pyx_v_self=<optimized out>, __pyx_v_args=0x7ffff7f7f050)
    at tvm/_ffi/_cython/core.cpp:6958
#19 __pyx_pw_4core_12FunctionBase_5__call__ (__pyx_v_self=<optimized out>, __pyx_args=0x7ffff7f7f050,
    __pyx_kwds=<optimized out>) at tvm/_ffi/_cython/core.cpp:6936
#20 0x000055555563b673 in PyObject_Call ()
#21 0x0000555555654e2f in PyEval_EvalFrameEx ()
#22 0x000055555564d535 in PyEval_EvalCodeEx ()
#23 0x0000555555655698 in PyEval_EvalFrameEx ()
#24 0x000055555564d535 in PyEval_EvalCodeEx ()
#25 0x000055555564d2d9 in PyEval_EvalCode ()
#26 0x000055555567ce3f in ?? ()
#27 0x0000555555677d52 in PyRun_FileExFlags ()
#28 0x000055555567789e in PyRun_SimpleFileExFlags ()
#29 0x0000555555628af1 in Py_Main ()
#30 0x00007ffff6f1a2b1 in __libc_start_main (main=0x555555628420 <main>, argc=2, argv=0x7fffffffdde8,
    init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffffffddd8)
    at ../csu/libc-start.c:291
#31 0x000055555562831a in _start ()

Please help why opencl backend doesn't work well for the image classification task.