[Bug] Error when compiling a simple ONNX model with MatMul operator for opt_level=2

[luyaor]

Description

When compiling following model with opt_level=2, TVM will crash. While if turn opt_level to 3, it will run normally.

The model(with ONNX as frontend) with error is as follows, check bug.onnx in bug0.zip.

Error Log

Traceback (most recent call last):
  File "check.py", line 19, in <module>
    tvm_graph, tvm_lib, tvm_params = relay.build_module.build(mod, target, params=params)
  File "/Users/luyaor/Documents/tvm/python/tvm/relay/build_module.py", line 275, in build
    graph_json, mod, params = bld_mod.build(mod, target, target_host, params)
  File "/Users/luyaor/Documents/tvm/python/tvm/relay/build_module.py", line 138, in build
    self._build(mod, target, target_host)
  File "/Users/luyaor/Documents/tvm/python/tvm/_ffi/_ctypes/packed_func.py", line 237, in __call__
    raise get_last_ffi_error()
tvm._ffi.base.TVMError: Traceback (most recent call last):
  [bt] (8) 9   libtvm.dylib                        0x00000001112c2a02 tvm::relay::StorageAllocator::Plan(tvm::relay::Function const&) + 354
  [bt] (7) 8   libtvm.dylib                        0x00000001112c68ca tvm::relay::StorageAllocaBaseVisitor::Run(tvm::relay::Function const&) + 154
  [bt] (6) 7   libtvm.dylib                        0x00000001112c44c7 tvm::relay::StorageAllocaBaseVisitor::GetToken(tvm::RelayExpr const&) + 23
  [bt] (5) 6   libtvm.dylib                        0x0000000111359b58 tvm::relay::ExprVisitor::VisitExpr(tvm::RelayExpr const&) + 344
  [bt] (4) 5   libtvm.dylib                        0x00000001110f1fad tvm::relay::ExprFunctor<void (tvm::RelayExpr const&)>::VisitExpr(tvm::RelayExpr const&) + 173
  [bt] (3) 4   libtvm.dylib                        0x00000001110f22a0 tvm::NodeFunctor<void (tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<void (tvm::RelayExpr const&)>*)>::operator()(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<void (tvm::RelayExpr const&)>*) const + 288
  [bt] (2) 3   libtvm.dylib                        0x00000001112c425b tvm::relay::StorageAllocator::CreateToken(tvm::RelayExprNode const*, bool) + 1179
  [bt] (1) 2   libtvm.dylib                        0x00000001112c6093 tvm::relay::StorageAllocator::GetMemorySize(tvm::relay::StorageToken*) + 451
  [bt] (0) 1   libtvm.dylib                        0x00000001105dac6f dmlc::LogMessageFatal::~LogMessageFatal() + 111
  File "/Users/luyaor/Documents/tvm/src/relay/backend/graph_plan_memory.cc", line 292
TVMError: 
---------------------------------------------------------------
An internal invariant was violated during the execution of TVM.
Please read TVM's error reporting guidelines.
More details can be found here: https://discuss.tvm.ai/t/error-reporting/7793.
---------------------------------------------------------------
  Check failed: pval != nullptr == false: Cannot allocate memory symbolic tensor shape [?, ?, ?]

How to reproduce

Environment

Python3, with tvm, onnx

tvm version: c31e338 Wed Dec 9 14:52:58 2020 +0900

Conda environment reference: see environment.yml

1. Download bug0.zip

2. Run python check.py.

[mbrookhart]

This is simply because the DynamicToStatic pass is defined at opt level 3, not 2. We
can lower it to make this pass, if we think that's a robust thing to do more generally.

cc @tqchen @jroesch @masahi @jwfromm

[tqchen]

I agree that given DynamicToStatic brings no harm, it should be applied more generally.

[tqchen]

closed by #7213 Thanks @mbrookhart @luyaor !