[Bug][Runtime] float64/int64 lead to TVMArrayCopyFromTo: The size must exactly match

[jikechao]

The input_dtype == 'float64' will lead to a check failure and throw "Check failed: from_size == to_size (3360 vs. 1680) : TVMArrayCopyFromTo: The size must exactly match".

Actual behavior

Traceback (most recent call last):
  File "test.py", line 28, in <module>
    res_tvm = model(tvm.nd.array(test_x_tvm.astype('float64'))).numpy()
  File "/workplace/software/tvm/tvm_/python/tvm/relay/build_module.py", line 537, in _graph_wrapper
    gmodule.set_input(i, arg)
  File "/workplace/software/tvm/tvm_/python/tvm/contrib/graph_executor.py", line 199, in set_input
    v.copyfrom(value)
  File "/workplace/software/tvm/tvm_/python/tvm/runtime/ndarray.py", line 143, in copyfrom
    source_array.copyto(self)
  File "/workplace/software/tvm/tvm_/python/tvm/runtime/ndarray.py", line 251, in copyto
    return self._copyto(target)
  File "/workplace/software/tvm/tvm_/python/tvm/_ffi/_ctypes/ndarray.py", line 90, in _copyto
    check_call(_LIB.TVMArrayCopyFromTo(self.handle, target_nd.handle, None))
  File "/workplace/software/tvm/tvm_/python/tvm/_ffi/base.py", line 348, in check_call
    raise get_last_ffi_error()
tvm._ffi.base.TVMError: Traceback (most recent call last):
  49: 0x000055c08bd49b32
        at ../sysdeps/x86_64/elf/start.S:103
  48: __libc_start_main
  47: Py_BytesMain
        at /tmp/build/80754af9/python_1589913560097/work/Modules/main.c:1089
  46: Py_RunMain
        at /tmp/build/80754af9/python_1589913560097/work/Modules/main.c:650
  45: pymain_run_python
        at /tmp/build/80754af9/python_1589913560097/work/Modules/main.c:571
  44: pymain_run_file
        at /tmp/build/80754af9/python_1589913560097/work/Modules/main.c:387
  43: PyRun_SimpleFileExFlags
        at /tmp/build/80754af9/python_1589913560097/work/Python/pythonrun.c:428
  42: PyRun_FileExFlags
        at /tmp/build/80754af9/python_1589913560097/work/Python/pythonrun.c:1063
  41: run_mod
        at /tmp/build/80754af9/python_1589913560097/work/Python/pythonrun.c:1147
  40: run_eval_code_obj
        at /tmp/build/80754af9/python_1589913560097/work/Python/pythonrun.c:1125
  39: PyEval_EvalCode
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:718
  38: PyEval_EvalCodeEx
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4327
  37: _PyEval_EvalCodeWithName
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4298
  36: _PyEval_EvalFrameDefault
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3500
  35: call_function
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
  34: _PyObject_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
  33: _PyFunction_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:435
  32: _PyEval_EvalCodeWithName
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4298
  31: _PyEval_EvalFrameDefault
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3486
  30: call_function
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
  29: _PyObject_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
  28: _PyFunction_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:435
  27: _PyEval_EvalCodeWithName
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4298
  26: _PyEval_EvalFrameDefault
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3486
  25: call_function
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
  24: _PyObject_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
  23: _PyFunction_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:410
  22: function_code_fastcall
        at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:283
  21: _PyEval_EvalFrameDefault
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3486
  20: call_function
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
  19: _PyObject_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
  18: _PyFunction_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:435
  17: _PyEval_EvalCodeWithName
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4298
  16: _PyEval_EvalFrameDefault
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3486
  15: call_function
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
  14: _PyObject_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:127
  13: _PyFunction_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:410
  12: function_code_fastcall
        at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:283
  11: _PyEval_EvalFrameDefault
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:3469
  10: call_function
        at /tmp/build/80754af9/python_1589913560097/work/Python/ceval.c:4963
  9: _PyObject_Vectorcall
        at /tmp/build/80754af9/python_1589913560097/work/Include/cpython/abstract.h:125
  8: _PyObject_MakeTpCall
        at /tmp/build/80754af9/python_1589913560097/work/Objects/call.c:159
  7: PyCFuncPtr_call
        at /usr/local/src/conda/python-3.8.3/Modules/_ctypes/_ctypes.c:4201
  6: _ctypes_callproc
        at /usr/local/src/conda/python-3.8.3/Modules/_ctypes/callproc.c:1199
  5: _call_function_pointer
        at /usr/local/src/conda/python-3.8.3/Modules/_ctypes/callproc.c:871
  4: ffi_call_int
  3: ffi_call_unix64
  2: TVMArrayCopyFromTo
  1: tvm::runtime::NDArray::CopyFromTo(DLTensor const*, DLTensor*, void*)
  0: _ZN3tvm7runtime6detail
  File "/workplace/software/tvm/tvm_/src/runtime/ndarray.cc", line 270
TVMError:

Environment

Any environment details, such as: Operating System, TVM version, etc

Steps to reproduce

import tvm
import tvm.relay as relay
import numpy as np
from tensorflow import keras
from tensorflow.keras import layers, models

input_shape = (1, 5, 7, 6, 2)

input_data = np.random.random(size=input_shape)
x = layers.Input(shape=input_shape[1:], dtype='float64')

layer = keras.layers.Conv3D(filters=1, kernel_size=3)
layer.set_weights(layer.get_weights())

y = layer(x)
model = models.Model(x, y)
model.summary()
res_keras = model(input_data)

shape_dict = {'input_1': input_shape}
mod, params = relay.frontend.from_keras(model, shape_dict,layout='NDHWC')
with tvm.transform.PassContext(opt_level=3):
    model = relay.build_module.create_executor("graph", mod, tvm.cpu(0), 'llvm', params).evaluate()

test_x_tvm = input_data
res_tvm = model(tvm.nd.array(test_x_tvm.astype('float64'))).numpy()

Other

Input type ="int64' for Embedding layer has the same crash message.

input_shape = (1, 2)

input_data = np.random.randint(2, size=input_shape)
x = layers.Input(shape=input_shape[1:], dtype='int64')
layer = keras.layers.Embedding(4, 10)

cc @areusch @comaniac @tkonolige

[haoyang9804]

Seems that this bug has been fixed in the following commit