[CPP UT] Demark failed due to lack of attribute

[LLee233]

bug描述 Describe the Bug

CMD line for:

~/Paddle/build/test/deprecated/cpp/inference/api/test_analyzer_image_classification --gtest_filter=*profile_mkldnn --repeat=1000  --batch_size=1    --infer_model=~/demark

And the result is:

Note: Google Test filter = *profile_mkldnn
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from Analyzer_resnet50
[ RUN      ] Analyzer_resnet50.profile_mkldnn
WARNING: Logging before InitGoogleLogging() is written to STDERR
I0929 17:48:29.608610    10 executor.cc:183] Old Executor is Running.
I0929 17:48:29.639094    10 tester_helper.h:398] feed target 0: {-1, 3, 207, 175}
W0929 17:48:29.640206    10 api.cc:126] The paddle::NativeConfig interface is going to be deprecated in the next release, please use the latest paddle_infer::Config instead.
I0929 17:48:29.640190    10 tester_helper.h:125] AnalysisConfig {
  NativeConfig {
    PaddlePredictor::Config {
      model_dir: 
    }
    use_gpu: 0
    device: 0
    fraction_of_gpu_memory: 0
    specify_input_name: 1
  }
  prog_file: /paddle/data/demark/model
  param_file: /paddle/data/demark/params
  enable_ir_optim: 1
  cpu_num_threads: 1
  use_tensorrt: 0
  use_mkldnn: 1
}
W0929 17:48:29.640249    10 analysis_predictor.cc:3394] Deprecated. Please use CreatePredictor instead.
I0929 17:48:29.649592    10 analysis_predictor.cc:2124] MKLDNN is enabled
I0929 17:48:29.649629    10 analysis_predictor.cc:2241] Ir optimization is turned off, no ir pass will be executed.
--- Running analysis [ir_graph_build_pass]
--- Running analysis [ir_analysis_pass]
--- Running analysis [ir_params_sync_among_devices_pass]
--- Running analysis [adjust_cudnn_workspace_size_pass]
--- Running analysis [inference_op_replace_pass]
--- Running analysis [save_optimized_model_pass]
--- Running analysis [ir_graph_to_program_pass]
I0929 17:48:29.680430    10 analysis_predictor.cc:2330] ======= ir optimization completed =======
unknown file: Failure
C++ exception with description "(InvalidArgument) Op conv2d_transpose arg output_size should not be zero size (at /paddle/Paddle/paddle/fluid/ir_adaptor/translator/op_translator.cc:530)
" thrown in the test body.
[  FAILED  ] Analyzer_resnet50.profile_mkldnn (90 ms)
[----------] 1 test from Analyzer_resnet50 (90 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test case ran. (90 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] Analyzer_resnet50.profile_mkldnn

 1 FAILED TEST

其他补充信息 Additional Supplementary Information

Model for reproducing:
As referred in #17535, the model is also call Face model from Paddle side, so you could check if it still exists firstly. :) Thanks.