C++预测默认开启MKLDNN内存泄露,重新初始化失败

[zhouyongxyz]

开启MKLDNN 内存一直在涨，按照提示的方案重置对象，会报错
Compile Traceback (most recent call last):
File "/opt/_internal/cpython-3.7.0/lib/python3.7/site-packages/paddle/fluid/framework.py", line 2525, in append_op
attrs=kwargs.get("attrs", None))

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/layers/nn.py", line 1405, in conv2d
"data_format": data_format,

File "/paddle/PaddlePaddle/PaddleOCR/ppocr/modeling/backbones/det_resnet_vd.py", line 138, in conv_bn_layer
bias_attr=False)

File "/paddle/PaddlePaddle/PaddleOCR/ppocr/modeling/backbones/det_resnet_vd.py", line 208, in bottleneck_block
name=name + "_branch2a")

File "/paddle/PaddlePaddle/PaddleOCR/ppocr/modeling/backbones/det_resnet_vd.py", line 106, in __call__
name=conv_name)

File "/paddle/PaddlePaddle/PaddleOCR/ppocr/modeling/architectures/det_model.py", line 111, in __call__
conv_feas = self.backbone(image)

File "/paddle/PaddlePaddle/PaddleOCR/tools/program.py", line 193, in build_export
image, outputs = model(mode='export')

File "tools/export_model.py", line 67, in main
config, eval_program, startup_prog)

File "tools/export_model.py", line 93, in <module>
main()

---

C++ Traceback (most recent call last):

0 paddle::AnalysisPredictor::Run(std::vector<paddle::PaddleTensor, std::allocatorpaddle::PaddleTensor > const&, std::vector<paddle::PaddleTensor, std::allocatorpaddle::PaddleTensor >, int)
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 paddle::framework::OperatorWithKernel::ChooseKernel(paddle::framework::RuntimeContext const&, paddle::framework::Scope const&, paddle::platform::Place const&) const
6 paddle::operators::ConvOp::GetExpectedKernelType(paddle::framework::ExecutionContext const&) const
7 paddle::framework::OperatorWithKernel::IndicateVarDataType(paddle::framework::ExecutionContext const&, std::string const&) const
8 paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int)
9 paddle::platform::GetCurrentTraceBackStringabi:cxx11

---

Error Message Summary:

Error: The Input Variable(Input) of conv2d Op used to determine kernel data type is empty or not LoDTensor or SelectedRows or LoDTensorArray.
[Hint: Expected data_type != dafault_data_type, but received data_type:-1 == dafault_data_type:-1.] (at /home/zhou/ai/ocr/Paddle/paddle/fluid/framework/operator.cc:1385)
[operator < conv2d > error]

[zhouyongxyz]

det = new DBDetector(
config.det_model_dir, config.use_gpu, config.gpu_id, config.gpu_mem,
config.cpu_math_library_num_threads, config.use_mkldnn,
config.use_zero_copy_run, config.max_side_len, config.det_db_thresh,
config.det_db_box_thresh, config.det_db_unclip_ratio, config.visualize);

rec = new CRNNRecognizer(config.rec_model_dir, config.use_gpu, config.gpu_id,
config.gpu_mem, config.cpu_math_library_num_threads,
config.use_mkldnn, config.use_zero_copy_run,
config.char_list_file);

renint detector
--- fused 0 scale with matmul
--- Fused 0 ReshapeTransposeMatmulMkldnn patterns
--- Fused 0 ReshapeTransposeMatmulMkldnn patterns with transpose's xshape
--- Fused 0 ReshapeTransposeMatmulMkldnn patterns with reshape's xshape
--- Fused 0 ReshapeTransposeMatmulMkldnn patterns with reshape's xshape with transpose's xshape
--- Fused 0 MatmulTransposeReshape patterns
--- fused 0 scale with matmul
--- Fused 0 ReshapeTransposeMatmulMkldnn patterns
--- Fused 0 ReshapeTransposeMatmulMkldnn patterns with transpose's xshape
--- Fused 0 ReshapeTransposeMatmulMkldnn patterns with reshape's xshape
--- Fused 0 ReshapeTransposeMatmulMkldnn patterns with reshape's xshape with transpose's xshape
--- Fused 0 MatmulTransposeReshape patterns
ocr start ...
Segmentation fault (core dumped)

经常报段错误。直接就挂了

[zhouyongxyz]

目前问题已经解决，解决方案，就是增加设置 config.SetMkldnnCacheCapacity(10)； 设置缓存数量。默认是0不限制，内存会一直增长。