[Bug] TVM crashes when the size of the shape of slope tensor for PRelu operator is larger than 1

[coffezhou]

Expected behavior

TVM should compile the model correctly.

Actual behavior

For the following model,

TVM crashes:

Traceback (most recent call last):
  File "/home/ubuntu/Documents/DLCompiler-test/tvm/1126/bugs/onnx_output4/test1.py", line 52, in <module>
    test(onnx_model)
  File "/home/ubuntu/Documents/DLCompiler-test/tvm/1126/bugs/onnx_output4/test1.py", line 45, in test
    tvm_model = relax.transform.LegalizeOps()(tvm_model)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/Documents/DLCompilers/tvm/python/tvm/ir/transform.py", line 167, in __call__
    return _ffi_transform_api.RunPass(self, mod)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "python/tvm_ffi/cython/function.pxi", line 904, in tvm_ffi.core.Function.__call__
  File "<unknown>", line 0, in tvm::transform::Pass::operator()(tvm::IRModule) const
  File "<unknown>", line 0, in tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
  File "<unknown>", line 0, in tvm::transform::ModulePassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const
  File "<unknown>", line 0, in std::_Function_handler<tvm::IRModule (tvm::IRModule, tvm::transform::PassContext), tvm::relax::transform::LegalizeOps(tvm::ffi::Optional<tvm::ffi::Map<tvm::ffi::String, tvm::ffi::Function, void>, void>, tvm::ffi::Optional<tvm::ffi::Array<tvm::ffi::String, void>, void>, bool)::{lambda(tvm::IRModule, tvm::transform::PassContext)#1}>::_M_invoke(std::_Any_data const&, tvm::IRModule&&, tvm::transform::PassContext&&)
  File "<unknown>", line 0, in tvm::relax::transform::LegalizeOps(tvm::ffi::Optional<tvm::ffi::Map<tvm::ffi::String, tvm::ffi::Function, void>, void>, tvm::ffi::Optional<tvm::ffi::Array<tvm::ffi::String, void>, void>, bool)::{lambda(tvm::IRModule, tvm::transform::PassContext)#1}::operator()(tvm::IRModule, tvm::transform::PassContext) const
  File "<unknown>", line 0, in tvm::relax::LegalizeMutator::Transform()
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitExpr(tvm::RelaxExpr const&)
  File "<unknown>", line 0, in tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>::VisitExpr(tvm::RelaxExpr const&)
  File "<unknown>", line 0, in tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>::InitVTable()::{lambda(tvm::ffi::ObjectRef const&, tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>*)#8}::_FUN(tvm::ffi::ObjectRef const&, tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>*)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitExpr_(tvm::relax::FunctionNode const*)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitWithNewScope(tvm::RelaxExpr const&, tvm::ffi::Optional<tvm::ffi::Array<tvm::relax::Var, void>, void>)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitExpr(tvm::RelaxExpr const&)
  File "<unknown>", line 0, in tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>::VisitExpr(tvm::RelaxExpr const&)
  File "<unknown>", line 0, in tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>::InitVTable()::{lambda(tvm::ffi::ObjectRef const&, tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>*)#10}::_FUN(tvm::ffi::ObjectRef const&, tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>*)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitExpr_(tvm::relax::SeqExprNode const*)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitBindingBlock(tvm::relax::BindingBlock const&)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitBindingBlock_(tvm::relax::DataflowBlockNode const*)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitBinding(tvm::relax::Binding const&)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitBinding_(tvm::relax::VarBindingNode const*)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitBinding_(tvm::relax::VarBindingNode const*, tvm::relax::IfNode const*)
  File "<unknown>", line 0, in tvm::relax::ExprMutator::VisitExpr(tvm::RelaxExpr const&)
  File "<unknown>", line 0, in tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>::VisitExpr(tvm::RelaxExpr const&)
  File "<unknown>", line 0, in tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>::InitVTable()::{lambda(tvm::ffi::ObjectRef const&, tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>*)#9}::_FUN(tvm::ffi::ObjectRef const&, tvm::relax::ExprFunctor<tvm::RelaxExpr (tvm::RelaxExpr const&)>*)
  File "<unknown>", line 0, in tvm::relax::LegalizeMutator::VisitExpr_(tvm::relax::CallNode const*)
  File "python/tvm_ffi/cython/function.pxi", line 1058, in tvm_ffi.core.tvm_ffi_callback
  File "/home/ubuntu/Documents/DLCompilers/tvm/python/tvm/relax/transform/legalize_ops/nn.py", line 493, in _nn_prelu
    return bb.call_te(topi.nn.prelu, call.args[0], call.args[1], call.attrs.axis)

  File "/home/ubuntu/Documents/DLCompilers/tvm/python/tvm/relax/block_builder.py", line 361, in call_te
    tir_func, call_args, output_sinfo, tir_vars = gen_call_tir_inputs(func, *args, **kwargs)

  File "/home/ubuntu/Documents/DLCompilers/tvm/python/tvm/relax/utils.py", line 355, in gen_call_tir_inputs
    te_out = func(*te_args, **te_kwargs)

  File "/home/ubuntu/Documents/DLCompilers/tvm/python/tvm/te/tag.py", line 57, in tagged_fdecl
    return fdecl(*args, **kwargs)

  File "/home/ubuntu/Documents/DLCompilers/tvm/python/tvm/topi/nn/elemwise.py", line 130, in prelu
    assert len(slope.shape) == 1

AssertionError

I am not sure that this is a bug for TVM. This issue is same as the question that has been fixed.

Environment

OS: Ubuntu 20.04
TVM: 0.23.dev0 (f4e28d3)

onnxruntime: 1.23.2

Steps to reproduce

This bug can be reproduced by the following code with the model in the attachment. As shown in the code, the model can be executed by onnxruntime.

from typing import Dict, List, Literal, Optional
import sys
import os

import numpy as np
import onnx
import onnxruntime
from onnx import ModelProto, TensorProto, helper

import tvm
import tvm.testing
from tvm import relax
from tvm.relax.frontend.onnx import from_onnx

import argparse
import pickle

def test(
    model: ModelProto,
    inputs: Optional[Dict[str, np.ndarray]] = None,
    ir_version: int = 8,
    opset: int = 14,
) -> None:
    # Configure model format.
    if ir_version is not None:
        model.ir_version = ir_version
    if opset is not None:
        model.opset_import[0].version = opset
    
    with open("inputs.pkl", 'rb') as fp:
        inputs = pickle.load(fp)
    # Run the model through onnx to get the expected result.
    try:
        ort_session = onnxruntime.InferenceSession(
            model.SerializeToString(), providers=["CPUExecutionProvider"]
        )
        ort_output = ort_session.run([], inputs)
    except Exception as e:
        print(e)
        print("This model cannot be executed by onnxruntime!")
        sys.exit(1)

    tvm_model = from_onnx(model, opset=opset, keep_params_in_input=True)
    tvm_model = relax.transform.DecomposeOpsForInference()(tvm_model)
    tvm_model = relax.transform.LegalizeOps()(tvm_model)  
    
if __name__ == "__main__":   
    onnx_model = onnx.load("11.onnx")
    test(onnx_model)

testcase.zip

Triage

Please refer to the list of label tags here to find the relevant tags and add them below in a bullet format (example below).

• needs-triage

[tlopex]

Closed as resolved