MKLDNN Error: few segmentation model prediction failed

[OliverLPH]

Thank you for contributing to PaddlePaddle. Before submitting the issue, you could search issue in the github in case that th If there is no solution,please make sure that this is an inference issue including the following details : System information -PaddlePaddle version （eg.1.1）or CommitID -CPU: MKLDNN, onednn 1.8, Intel(R) Xeon(R) Gold 6271C CPU @ 2.60GHz -GPU: None -OS Platform: CentOS Linux 7.6.1810 -Docker Image: hub.baidubce.com/paddlepaddle/paddle:latest-dev -Python version: None -Cmake orders -C++version.txt -API information Note: You can get most of the information by running summary_env.py. To Reproduce Seg model download link: https://sys-p0.bj.bcebos.com/Paddle-UnitTest-Model/PaddleSeg.tgz current failed models are fastscnn, hrnet, pspnet, you can download from link above

git clone https://github.com/PaddlePaddle/continuous_integration.git
cd continuous_integration/inference/inference_api_test/cpp_api_test
bash build.sh ${paddle_Inference_dir} OFF ON OFF
./build/test_clas_model --model_name=${model_name} \
                            --model_path=${model_path} \
                            --params_path=${params_path} \
                            --image_shape="3,512,512" \
                            --use_mkldnn=true \
                            --batch_size=1 \
                            --use_gpu=false \
                            --accuracy="1e-6" \
                            --gtest_output=xml:test_${model_name}_mkldnn_${accuracy}_bz${batch_size}.xml

Describe your current behavior

    File "/workspace/pdseg/models/model_builder.py", line 167, in build_model
      logits = seg_model(image, class_num)
    File "/workspace/pdseg/models/model_builder.py", line 86, in seg_model
      logits = fast_scnn.fast_scnn(image, class_num)
    File "/workspace/pdseg/models/modeling/fast_scnn.py", line 280, in fast_scnn
      lower_res_feature = global_feature_extractor.net(higher_res_features)
    File "/workspace/pdseg/models/modeling/fast_scnn.py", line 226, in net
      x = psp_module(x, self.block_channels[2] // 4)
    File "/workspace/pdseg/models/modeling/fast_scnn.py", line 148, in psp_module
      name=psp_name + '_adapool')
    File "/opt/_internal/cpython-3.7.0/lib/python3.7/site-packages/paddle/fluid/layers/nn.py", line 2461, in adaptive_pool2d
      "adaptive": True,
    File "/opt/_internal/cpython-3.7.0/lib/python3.7/site-packages/paddle/fluid/layer_helper.py", line 43, in append_op
      return self.main_program.current_block().append_op(*args, **kwargs)
    File "/opt/_internal/cpython-3.7.0/lib/python3.7/site-packages/paddle/fluid/framework.py", line 2917, in append_op
      attrs=kwargs.get("attrs", None))
    File "/opt/_internal/cpython-3.7.0/lib/python3.7/site-packages/paddle/fluid/framework.py", line 2014, in __init__
      for frame in traceback.extract_stack():

--------------------------------------
C++ Traceback (most recent call last):
--------------------------------------
0   paddle::AnalysisPredictor::ZeroCopyRun()
1   paddle::framework::NaiveExecutor::Run()
2   paddle::framework::OperatorBase::Run(paddle::framework::Scope const&, paddle::platform::Place const&)
3   paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, paddle::platform::Place const&) const
4   paddle::framework::OperatorWithKernel::RunImpl(paddle::framework::Scope const&, paddle::platform::Place const&, paddle::framework::RuntimeContext*) const
5   std::_Function_handler<void (paddle::framework::ExecutionContext const&), paddle::framework::OpKernelRegistrarFunctor<paddle::platform::CPUPlace, false, 0ul, paddle::operators::PoolMKLDNNOpKernel<float>, paddle::operators::PoolMKLDNNOpKernel<signed char>, paddle::operators::PoolMKLDNNOpKernel<unsigned char>, paddle::operators::PoolMKLDNNOpKernel<paddle::platform::bfloat16> >::operator()(char const*, char const*, int) const::{lambda(paddle::framework::ExecutionContext const&)#1}>::_M_invoke(std::_Any_data const&, paddle::framework::ExecutionContext const&)
6   paddle::operators::PoolMKLDNNOpKernel<float>::Compute(paddle::framework::ExecutionContext const&) const
7   paddle::platform::PoolingMKLDNNHandler<float>::PoolingMKLDNNHandler(paddle::framework::ExecutionContext const&, paddle::platform::MKLDNNDeviceContext const&, dnnl::engine, paddle::platform::Place, paddle::framework::Tensor const*, paddle::framework::Tensor*, std::string const&)
8   paddle::platform::PoolingMKLDNNHandler<float>::ComputeAdaptivePoolParameters(paddle::framework::ExecutionContext const&, std::vector<long, std::allocator<long> > const&, std::vector<long, std::allocator<long> >&, std::vector<long, std::allocator<long> >&)
9   paddle::platform::EnforceNotMet::EnforceNotMet(paddle::platform::ErrorSummary const&, char const*, int)
10  paddle::platform::GetCurrentTraceBackString()

----------------------
Error Message Summary:
----------------------
UnimplementedError: Input dim must be divisible by corressponding ksize dim.
  [Hint: Expected src_tz[src_tz.size() - 1] % ksize[1] == 0, but received src_tz[src_tz.size() - 1] % ksize[1]:1 != 0:0.] (at /paddle/paddle/fluid/platform/mkldnn_reuse.h:906)
  [operator < pool2d > error]" thrown in the test body.

Code to reproduce the issue Other info / logs

[paddle-bot-old[bot]]

您好，我们已经收到了您的问题，会安排技术人员尽快解答您的问题，请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时，您也可以通过查看官网API文档、常见问题、历史Issue、AI社区来寻求解答。祝您生活愉快～

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the API，FAQ，Github Issue and AI community to get the answer.Have a nice day!

[jczaja]

@OliverLPH This PR https://github.com/PaddlePaddle/Paddle/pull/30757 should fix crashes reported in this issue. Please test

[OliverLPH]

@jczaja Thanks~ I will verify this PR

[OliverLPH]

@jczaja @luotao1 Hi, I have verified this PR, model prediction looks great without crash. Please cherry-pick to release/2.0 branch, thanks so much~

[paddle-bot-old[bot]]

Are you satisfied with the resolution of your issue?

YES No