Issue running the tutorial

[pradyotprakash]

I installed tvm as per the instructions in the github repo. When I try to run the tutorial provided here, I get an error when trying to execute the statement,

fadd_cuda = tvm.build(s, [A, B, C], "cuda", target_host="llvm", name="myadd")

I get the following error.

<ipython-input-5-a63326b9397b> in <module>()
      5 s[C].bind(tx, tvm.thread_axis("threadIdx.x"))
      6 
----> 7 fadd_cuda = tvm.build(s, [A, B, C], "cuda", target_host="llvm", name="myadd")

/home/pradyot/.local/lib/python2.7/site-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/build_module.pyc in build(sch, args, target, target_host, name, binds)
    366     fhost = [ir_pass.LowerIntrin(x, target_host) for x in fhost]
    367     fhost = [ir_pass.CombineContextCall(x) for x in fhost]
--> 368     mhost = codegen.build_module(fhost, target_host)
    369 
    370     if fdevice:

/home/pradyot/.local/lib/python2.7/site-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/codegen.pyc in build_module(lowered_func, target)
     18         The corressponding module.
     19     """
---> 20     return _Build(lowered_func, target)
     21 
     22 _init_api("tvm.codegen")

/home/pradyot/.local/lib/python2.7/site-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/_ffi/function.pyc in my_api_func(*args)
    253         The result of the API function call.
    254         """
--> 255         return flocal(*args)
    256     return my_api_func
    257 

/home/pradyot/.local/lib/python2.7/site-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/_ffi/_ctypes/function.pyc in __call__(self, *args)
    181         check_call(_LIB.TVMFuncCall(
    182             self.handle, values, tcodes, ctypes.c_int(num_args),
--> 183             ctypes.byref(ret_val), ctypes.byref(ret_tcode)))
    184         _ = temp_args
    185         _ = args

/home/pradyot/.local/lib/python2.7/site-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/_ffi/base.pyc in check_call(ret)
     60     """
     61     if ret != 0:
---> 62         raise TVMError(py_str(_LIB.TVMGetLastError()))
     63 
     64 

TVMError: [20:21:14] src/codegen/codegen.cc:27: Check failed: bf != nullptr Target llvm is not enabled

Stack trace returned 10 entries:
[bt] (0) /home/pradyot/.local/lib/python2.7/site-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/libtvm.so(_ZN3tvm7codegen5BuildERKNS_5ArrayINS_11LoweredFuncEvEERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0xf99) [0x7fbf49afbc39]
[bt] (1) /home/pradyot/.local/lib/python2.7/site-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/libtvm.so(+0x316979) [0x7fbf49ab6979]
[bt] (2) /home/pradyot/.local/lib/python2.7/site-packages/tvm-0.1.0-py2.7-linux-x86_64.egg/tvm/libtvm.so(TVMFuncCall+0x5e) [0x7fbf49cb440e]
[bt] (3) /usr/lib/x86_64-linux-gnu/libffi.so.6(ffi_call_unix64+0x4c) [0x7fbf75e85e40]
[bt] (4) /usr/lib/x86_64-linux-gnu/libffi.so.6(ffi_call+0x2eb) [0x7fbf75e858ab]
[bt] (5) /usr/lib/python2.7/lib-dynload/_ctypes.x86_64-linux-gnu.so(_ctypes_callproc+0x48f) [0x7fbf7609d3df]
[bt] (6) /usr/lib/python2.7/lib-dynload/_ctypes.x86_64-linux-gnu.so(+0x11d82) [0x7fbf760a1d82]
[bt] (7) /usr/bin/python(PyObject_Call+0x43) [0x4b0cb3]
[bt] (8) /usr/bin/python(PyEval_EvalFrameEx+0x5faf) [0x4c9faf]
[bt] (9) /usr/bin/python(PyEval_EvalCodeEx+0x255) [0x4c2765]

The stack trace tells me that llvm is not enabled. However, the installation instructions mention that llvm is not needed for CUDA targets. Anyway, I re-build tvm using llvm 4.0.1 but I still get a similar error. What am I doing wrong?

[tqchen]

You want to do make clean then make when build with llvm, because some of the pre-defined macros are not traced by makefile

[pradyotprakash]

Yes, I deleted the initial folder before trying this again. Not sure why I am getting this :/
In any case, the basic example should work without llvm, right?

[tqchen]

This is mainly because the target_host is force set to "llvm". If we do not set the target_host, it will redirect to "stackvm" that do not rely on LLVM.

However, we do need llvm for some advanced features, like storing the code into shared library.

[tqchen]

you want to make sure that LLVM_CONFIG is set properly, and after you rebuild, you reinstalled the library.