Wheest opened a new issue, #11233:
URL: https://github.com/apache/tvm/issues/11233
### Expected behavior
Compiling a model with `debug_executor` allows one to run the model and get
the tracing output.
### Actual behavior
Process fails during the module creation:
```
Traceback (most recent call last):
File "tvm_profiler_simple.py", line 98, in <module>
main(args)
File "tvm_profiler_simple.py", line 64, in main
m = debug_executor.create(lib.graph_json, lib, dev,
dump_root="/tmp/tvmdbg")
File "/app/source/tvm/python/tvm/contrib/debugger/debug_executor.py", line
70, in create
func_obj = fcreate(graph_json_str, libmod, *device_type_id)
File "/app/source/tvm/python/tvm/_ffi/_ctypes/packed_func.py", line 223,
in __call__
values, tcodes, num_args = _make_tvm_args(args, temp_args)
File "/app/source/tvm/python/tvm/_ffi/_ctypes/packed_func.py", line 188,
in _make_tvm_args
raise TypeError("Don't know how to handle type %s" % type(arg))
TypeError: Don't know how to handle type <class
'tvm.relay.backend.executor_factory.GraphExecutorFactoryModule'>
```
### Environment
x86 platform, TVM v0.8.
### Steps to reproduce
[This
script](https://gist.github.com/Wheest/9ad2d6a47bbd2cfaa4be530c68ba2f6c) shows
the issue.
The gist has three modes: `["tutorial", "alt", "normal"]`, invoked with
`python tvm_profiler_simple.py --mode tutorial`.
- `normal` inference works of course
- `tutorial` is the approach in the current documentation, with fails with
output [1]
- `alt` is the approach used in [this docs
PR](https://github.com/apache/tvm/pull/11231) which works, with example output
[2]
As the [discussion in the forum
says](https://discuss.tvm.apache.org/t/runnig-a-model-with-tvm-debugger/9869/9?u=wheest),
there does not appear to be a canonical way of creating the debugger, but the
approach in the docs should at least work, until there has been a refactor to
produce an unambiguous canonical approach.
[1] `tutorial` sample output:
```
Traceback (most recent call last):
File "tvm_profiler_simple.py", line 98, in <module>
main(args)
File "tvm_profiler_simple.py", line 64, in main
m = debug_executor.create(lib.graph_json, lib, dev,
dump_root="/tmp/tvmdbg")
File "/app/source/tvm/python/tvm/contrib/debugger/debug_executor.py", line
70, in create
func_obj = fcreate(graph_json_str, libmod, *device_type_id)
File "/app/source/tvm/python/tvm/_ffi/_ctypes/packed_func.py", line 223,
in __call__
values, tcodes, num_args = _make_tvm_args(args, temp_args)
File "/app/source/tvm/python/tvm/_ffi/_ctypes/packed_func.py", line 188,
in _make_tvm_args
raise TypeError("Don't know how to handle type %s" % type(arg))
TypeError: Don't know how to handle type <class
'tvm.relay.backend.executor_factory.GraphExecutorFactoryModule'>
```
[2] `alt` sample output
```
[19:53:25] ../src/runtime/graph_executor/debug/graph_executor_debug.cc:103:
Iteration: 0
[19:53:25] ../src/runtime/graph_executor/debug/graph_executor_debug.cc:108:
Op #0 tvmgen_default_fused_layout_transform: 29.3538 us/iter
[19:53:25] ../src/runtime/graph_executor/debug/graph_executor_debug.cc:108:
Op #1 tvmgen_default_fused_nn_contrib_conv2d_NCHWc_add_nn_relu: 3009 us/iter
...
Node Name Ops
Time(us) Time(%) Shape
Inputs Outputs
--------- ---
-------- ------- -----
------ -------
tvmgen_default_fused_nn_contrib_conv2d_NCHWc_add_nn_relu
tvmgen_default_fused_nn_contrib_conv2d_NCHWc_add_nn_relu 3009.0
7.115 (1, 2, 112, 112, 32) 3 1
tvmgen_default_fused_nn_contrib_conv2d_NCHWc_add_add_nn_relu_4
tvmgen_default_fused_nn_contrib_conv2d_NCHWc_add_add_nn_relu_4 2835.26
6.704 (1, 16, 14, 14, 16) 4 1
...
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscr...@tvm.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
